[PATCH v2 00/15] lib/lz4: stop forking upstream LZ4, vendor it instead

[PATCH v2 00/15] lib/lz4: stop forking upstream LZ4, vendor it instead

[Michal Wilczynski]

The in-kernel LZ4 is a fork. The decompressor was last synced with
upstream v1.8.3 in 2018 and the compressor with v1.7.3 in 2017, both by
hand. Upstream has made 488 commits against lib/ since, and the gap is
maintained one cherry-pick at a time.

That has left real bugs in place for example the forked
LZ4_decompress_fast() has no bounds checks on its input, so corrupted
input reads past its end and before the output buffer.

This series vendors the upstream sources unmodified and adapts them at
build time, so a re-sync becomes a directory copy:

  upstream/lz4.c, upstream/lz4.h    verbatim upstream sources
  upstream/lz4hc.c, upstream/lz4hc.h
  lz4_deps.h                        the build environment
  lz4_kernel_api.h                  hands the LZ4 names back
  freestanding/                     ISO C headers -nostdinc drops
  lz4_compress.c                    the kernel's entry points
  lz4_decompress.c
  lz4hc_compress.c

Patches 1-3 apply to the fork as it stands: patch 1 moves the pre-boot
decompressor off LZ4_decompress_fast(), patch 2 removes it and its
variants, and patch 3 stops zram's lz4hc backend resetting its stream
for every page. Patch 4 imports the sources. Patches 5-7 backport the
upstream fixes the kernel needs on top of v1.10.0. Patch 8 adds the
build environment. Patch 9 puts the freestanding headers on the pre-boot
decompressor path. Patches 10, 11 and 13 convert the compressor, HC
compressor and decompressor in turn; patch 12 widens the in-place margin
erofs relies on for the new decoder. Patch 14 drops the last duplicated
prototype list. Patch 15 gives lib/lz4 a MAINTAINERS entry, which it has
never had.

This is meant to go through Andrew's tree; patch 9 needs acks from the
boot maintainers.

The exported API changes in four ways.

The three stream types become incomplete. <linux/lz4.h> spelled out their
layout for the fork's benefit, upstream headers define the same types and
lib/decompress_unlz4.c ends up with both in one translation unit. Callers
allocate LZ4_MEM_COMPRESS, LZ4HC_MEM_COMPRESS or LZ4_MEM_DECOMPRESS bytes
instead of sizeof() every caller but zram already did. Each conversion
patch makes its own type incomplete so no step leaves the tree broken.

Sixteen macros go with them: the hash geometry (LZ4_HASHLOG,
LZ4_HASHTABLESIZE, LZ4_HASH_SIZE_U32, LZ4HC_DICTIONARY_LOGSIZE, LZ4HC_MAXD,
LZ4HC_MAXD_MASK, LZ4HC_HASH_LOG, LZ4HC_HASHTABLESIZE, LZ4HC_HASH_MASK) and
the sizes derived from it (LZ4_MEMORY_USAGE, LZ4_STREAMSIZE{,_U64},
LZ4_STREAMHCSIZE{,_SIZET}, LZ4_STREAMDECODESIZE{,_U64}). They existed to
lay out structures the fork defined here none has an in tree user outside
lib/lz4. LZ4_compressBound() goes too a static inline whose only caller
was lib/decompress_unlz4.c, now using the LZ4_COMPRESSBOUND() macro.

LZ4_MEM_DECOMPRESS and LZ4HC_CLAMP_CLEVEL are new.

LZ4_decompress_fast(), LZ4_decompress_fast_continue() and
LZ4_decompress_fast_usingDict() go. Upstream has deprecated them since
v1.9, and their last in-tree user, the pre-boot decompressor, now uses
LZ4_decompress_safe_partial().

There are some behaviour changes. All of them come from upstream rather
than from this series and I have kept them rather than patching them
out: a local fix on top of vendored sources is exactly the divergence
this series exists to remove.

 - LZ4HC levels 10 and up are clamped to 9. Upstream routes them to its
   optimal parser, whose ~64K opt[] does not fit a kernel stack. The
   fork had no optimal parser at all - level only scaled search depth -
   so its levels 10-16 beat its own level 9 by 0.01%. Clamping rather
   than rejecting keeps existing f2fs and zram settings working.

 - LZ4_decompress_safe_partial() with a zero-length target returns 0
   rather than -1, matching upstream since 725cb0aafdf7. No in-tree
   caller depends on it.

 - The compressors require wrkmem aligned to 8 bytes. Upstream checks
   this and LZ4_initStream() returns NULL, which upstream's own callers
   then dereference or ignore. The entry points check the alignment
   themselves, warn once and fail the call. Every in-tree caller uses
   kmalloc() or vmalloc() and is already aligned; <linux/lz4.h> now
   documents the requirement.

The layout follows lib/zstd, which keeps upstream's sources in a
subdirectory and the kernel glue on top. zstd has an upstream import
tool and an upstream zstd_deps.h; LZ4 has neither, so the kernel glue
here lives outside the vendored files.

The import is the v1.10.0 release tag. It does not build under
-Wmissing-prototypes without upstream commit 5ef1f16929b5, which patch 5
backports. Patch 6 backports upstream's arm64 fast decode loop and patch
7 its match length limit fix; those are the only deviations from the
tag. The rest of PR #1753 fixes code added after v1.10.0, so it will
come with the next release.

checkpatch reports some 3500 errors against patch 4: upstream's coding
style, and no SPDX tags, in files kept verbatim.

Measured in the kernel: three kernels differing only in lib/lz4, booted
one core under KVM, each reading 1.3GB of kernel source from an LZ4
erofs image with the page cache dropped before every pass. For v2 this
was remeasured with more runs on an idle machine: median of 140 passes.
The middle column is upstream v1.10.0 with LZ4_FAST_DEC_LOOP=0,
to separate the version bump from the decode loop upstream added in
v1.9.0.

                        erofs read, MB/s
  erofs cluster    today   v1.10.0   + fast loop    net
             4K     2486      2549          2665    +7%
            64K     3013      3104          3225    +7%

The same three decoders at the library level, each block decompressed
on its own the way zram treats a page. x86_64, the fork's and upstream's
sources built with the kernel's code generation flags (so no SSE), all
decoding the same compressed input, median of 21 interleaved runs. The
first corpus is upstream's tests/datagen with a fixed seed, so it is
byte-identical anywhere; the second is a tar of kernel sources:

                                 decompression, MB/s
  corpus           block  ratio    today   v1.10.0  + fast loop   net
  datagen -s0 -P60   4K  1.09x     4584      4535         5097   +11%
                    16K  1.34x     3960      3978         4828   +22%
                    64K  1.90x     4797      4802         5598   +17%
                   256K  1.92x     6749      6559         7600   +13%
  kernel source tar  4K  2.29x     3719      3832         4219   +13%
                    16K  2.57x     3778      3880         4203   +11%
                    64K  2.80x     3792      3886         4179   +10%
                   256K  2.95x     4165      4280         4750   +14%

Compression is not uniformly faster: lz4hc on small blocks of poorly
compressible data is up to 15% slower. Patches 10 and 11 have the
numbers.

The cost is size: lib/lz4 grows from 45K to 86K of text on x86_64,
lz4hc most (12K to 39K). That is upstream code the fork never had.

---
Changes in v2:
- Drop the RFC prefix
- New patches 1-2: move the pre-boot decompressor to
  LZ4_decompress_safe_partial() and remove LZ4_decompress_fast() and its
  variants from the fork, ahead of the import (Sergey)
- New patch 3: zram's lz4hc backend resets its stream once, not per page
- New patches 6-7: backport upstream's arm64 fast decode loop (Sergey)
  and its match length limit fix
- New patch 12: widen LZ4_DECOMPRESS_INPLACE_MARGIN for the fast decode
  loop
- Compressor entry points check wrkmem alignment, warn and fail (Sergey)
- Patch 13 names the upstream counterpart of each retired local patch
  (Sergey)
- MAINTAINERS: also cover lib/decompress_unlz4.c and
  include/linux/decompress/unlz4.h (Eric)
- Remeasure erofs reads with more runs on an idle machine
- Link to v1: https://lore.kernel.org/r/20260925-lz4-vendor-upstream-v1-0-1c7ffbe21c4b@samsung.com

---
Michal Wilczynski (15):
      lib/decompress_unlz4: use LZ4_decompress_safe_partial() before boot
      lib/lz4: remove LZ4_decompress_fast() and its variants
      zram: lz4hc: reset the dictionary stream once, not per page
      lib/lz4: import upstream LZ4 sources verbatim
      lib/lz4: backport upstream's -Wmissing-prototypes fix
      lib/lz4: backport upstream's aarch64 fast decode loop enablement
      lib/lz4: backport upstream's match length limit fix
      lib/lz4: add the build environment for the vendored sources
      arch: boot: put the LZ4 freestanding headers on the decompressor path
      lib/lz4: switch the compressor to the vendored sources
      lib/lz4: switch the HC compressor to the vendored sources
      lib/lz4: widen LZ4_DECOMPRESS_INPLACE_MARGIN for the fast decode loop
      lib/lz4: switch the decompressor to the vendored sources
      lib/lz4: replace lz4_kernel_api.h's prototypes with <linux/lz4.h>
      MAINTAINERS: add an entry for the LZ4 compression library

 Documentation/filesystems/f2fs.rst   |    2 +
 MAINTAINERS                          |   10 +
 arch/arm/boot/compressed/Makefile    |    3 +
 arch/mips/boot/compressed/Makefile   |    4 +
 arch/parisc/boot/compressed/Makefile |    3 +
 arch/s390/boot/Makefile              |    3 +
 arch/x86/boot/compressed/Makefile    |    3 +
 drivers/block/zram/backend_lz4.c     |    8 +-
 drivers/block/zram/backend_lz4hc.c   |    8 +-
 include/linux/lz4.h                  |  253 +--
 lib/decompress_unlz4.c               |   17 +-
 lib/lz4/Makefile                     |   16 +-
 lib/lz4/freestanding/limits.h        |   14 +
 lib/lz4/freestanding/stddef.h        |   15 +
 lib/lz4/freestanding/stdint.h        |   14 +
 lib/lz4/freestanding/string.h        |   14 +
 lib/lz4/lz4_compress.c               |  946 +-----------
 lib/lz4/lz4_decompress.c             |  724 +--------
 lib/lz4/lz4_deps.h                   |  118 ++
 lib/lz4/lz4_kernel_api.h             |   34 +
 lib/lz4/lz4defs.h                    |  247 ---
 lib/lz4/lz4hc_compress.c             |  792 +---------
 lib/lz4/upstream/lz4.c               | 2824 ++++++++++++++++++++++++++++++++++
 lib/lz4/upstream/lz4.h               |  884 +++++++++++
 lib/lz4/upstream/lz4hc.c             | 2192 ++++++++++++++++++++++++++
 lib/lz4/upstream/lz4hc.h             |  414 +++++
 26 files changed, 6813 insertions(+), 2749 deletions(-)
---
base-commit: 93f51579e7df248780214094418f205253383cc5
change-id: 20260904-lz4-vendor-upstream-4bc2202d0ffc

Best regards,
-- 
Michal Wilczynski <m.wilczynski@samsung.com>

[f2fs-dev] [PATCH v2 00/15] lib/lz4: stop forking upstream LZ4, vendor it instead

[Michal Wilczynski]

The in-kernel LZ4 is a fork. The decompressor was last synced with
upstream v1.8.3 in 2018 and the compressor with v1.7.3 in 2017, both by
hand. Upstream has made 488 commits against lib/ since, and the gap is
maintained one cherry-pick at a time.

That has left real bugs in place for example the forked
LZ4_decompress_fast() has no bounds checks on its input, so corrupted
input reads past its end and before the output buffer.

This series vendors the upstream sources unmodified and adapts them at
build time, so a re-sync becomes a directory copy:

  upstream/lz4.c, upstream/lz4.h    verbatim upstream sources
  upstream/lz4hc.c, upstream/lz4hc.h
  lz4_deps.h                        the build environment
  lz4_kernel_api.h                  hands the LZ4 names back
  freestanding/                     ISO C headers -nostdinc drops
  lz4_compress.c                    the kernel's entry points
  lz4_decompress.c
  lz4hc_compress.c

Patches 1-3 apply to the fork as it stands: patch 1 moves the pre-boot
decompressor off LZ4_decompress_fast(), patch 2 removes it and its
variants, and patch 3 stops zram's lz4hc backend resetting its stream
for every page. Patch 4 imports the sources. Patches 5-7 backport the
upstream fixes the kernel needs on top of v1.10.0. Patch 8 adds the
build environment. Patch 9 puts the freestanding headers on the pre-boot
decompressor path. Patches 10, 11 and 13 convert the compressor, HC
compressor and decompressor in turn; patch 12 widens the in-place margin
erofs relies on for the new decoder. Patch 14 drops the last duplicated
prototype list. Patch 15 gives lib/lz4 a MAINTAINERS entry, which it has
never had.

This is meant to go through Andrew's tree; patch 9 needs acks from the
boot maintainers.

The exported API changes in four ways.

The three stream types become incomplete. <linux/lz4.h> spelled out their
layout for the fork's benefit, upstream headers define the same types and
lib/decompress_unlz4.c ends up with both in one translation unit. Callers
allocate LZ4_MEM_COMPRESS, LZ4HC_MEM_COMPRESS or LZ4_MEM_DECOMPRESS bytes
instead of sizeof() every caller but zram already did. Each conversion
patch makes its own type incomplete so no step leaves the tree broken.

Sixteen macros go with them: the hash geometry (LZ4_HASHLOG,
LZ4_HASHTABLESIZE, LZ4_HASH_SIZE_U32, LZ4HC_DICTIONARY_LOGSIZE, LZ4HC_MAXD,
LZ4HC_MAXD_MASK, LZ4HC_HASH_LOG, LZ4HC_HASHTABLESIZE, LZ4HC_HASH_MASK) and
the sizes derived from it (LZ4_MEMORY_USAGE, LZ4_STREAMSIZE{,_U64},
LZ4_STREAMHCSIZE{,_SIZET}, LZ4_STREAMDECODESIZE{,_U64}). They existed to
lay out structures the fork defined here none has an in tree user outside
lib/lz4. LZ4_compressBound() goes too a static inline whose only caller
was lib/decompress_unlz4.c, now using the LZ4_COMPRESSBOUND() macro.

LZ4_MEM_DECOMPRESS and LZ4HC_CLAMP_CLEVEL are new.

LZ4_decompress_fast(), LZ4_decompress_fast_continue() and
LZ4_decompress_fast_usingDict() go. Upstream has deprecated them since
v1.9, and their last in-tree user, the pre-boot decompressor, now uses
LZ4_decompress_safe_partial().

There are some behaviour changes. All of them come from upstream rather
than from this series and I have kept them rather than patching them
out: a local fix on top of vendored sources is exactly the divergence
this series exists to remove.

 - LZ4HC levels 10 and up are clamped to 9. Upstream routes them to its
   optimal parser, whose ~64K opt[] does not fit a kernel stack. The
   fork had no optimal parser at all - level only scaled search depth -
   so its levels 10-16 beat its own level 9 by 0.01%. Clamping rather
   than rejecting keeps existing f2fs and zram settings working.

 - LZ4_decompress_safe_partial() with a zero-length target returns 0
   rather than -1, matching upstream since 725cb0aafdf7. No in-tree
   caller depends on it.

 - The compressors require wrkmem aligned to 8 bytes. Upstream checks
   this and LZ4_initStream() returns NULL, which upstream's own callers
   then dereference or ignore. The entry points check the alignment
   themselves, warn once and fail the call. Every in-tree caller uses
   kmalloc() or vmalloc() and is already aligned; <linux/lz4.h> now
   documents the requirement.

The layout follows lib/zstd, which keeps upstream's sources in a
subdirectory and the kernel glue on top. zstd has an upstream import
tool and an upstream zstd_deps.h; LZ4 has neither, so the kernel glue
here lives outside the vendored files.

The import is the v1.10.0 release tag. It does not build under
-Wmissing-prototypes without upstream commit 5ef1f16929b5, which patch 5
backports. Patch 6 backports upstream's arm64 fast decode loop and patch
7 its match length limit fix; those are the only deviations from the
tag. The rest of PR #1753 fixes code added after v1.10.0, so it will
come with the next release.

checkpatch reports some 3500 errors against patch 4: upstream's coding
style, and no SPDX tags, in files kept verbatim.

Measured in the kernel: three kernels differing only in lib/lz4, booted
one core under KVM, each reading 1.3GB of kernel source from an LZ4
erofs image with the page cache dropped before every pass. For v2 this
was remeasured with more runs on an idle machine: median of 140 passes.
The middle column is upstream v1.10.0 with LZ4_FAST_DEC_LOOP=0,
to separate the version bump from the decode loop upstream added in
v1.9.0.

                        erofs read, MB/s
  erofs cluster    today   v1.10.0   + fast loop    net
             4K     2486      2549          2665    +7%
            64K     3013      3104          3225    +7%

The same three decoders at the library level, each block decompressed
on its own the way zram treats a page. x86_64, the fork's and upstream's
sources built with the kernel's code generation flags (so no SSE), all
decoding the same compressed input, median of 21 interleaved runs. The
first corpus is upstream's tests/datagen with a fixed seed, so it is
byte-identical anywhere; the second is a tar of kernel sources:

                                 decompression, MB/s
  corpus           block  ratio    today   v1.10.0  + fast loop   net
  datagen -s0 -P60   4K  1.09x     4584      4535         5097   +11%
                    16K  1.34x     3960      3978         4828   +22%
                    64K  1.90x     4797      4802         5598   +17%
                   256K  1.92x     6749      6559         7600   +13%
  kernel source tar  4K  2.29x     3719      3832         4219   +13%
                    16K  2.57x     3778      3880         4203   +11%
                    64K  2.80x     3792      3886         4179   +10%
                   256K  2.95x     4165      4280         4750   +14%

Compression is not uniformly faster: lz4hc on small blocks of poorly
compressible data is up to 15% slower. Patches 10 and 11 have the
numbers.

The cost is size: lib/lz4 grows from 45K to 86K of text on x86_64,
lz4hc most (12K to 39K). That is upstream code the fork never had.

---
Changes in v2:
- Drop the RFC prefix
- New patches 1-2: move the pre-boot decompressor to
  LZ4_decompress_safe_partial() and remove LZ4_decompress_fast() and its
  variants from the fork, ahead of the import (Sergey)
- New patch 3: zram's lz4hc backend resets its stream once, not per page
- New patches 6-7: backport upstream's arm64 fast decode loop (Sergey)
  and its match length limit fix
- New patch 12: widen LZ4_DECOMPRESS_INPLACE_MARGIN for the fast decode
  loop
- Compressor entry points check wrkmem alignment, warn and fail (Sergey)
- Patch 13 names the upstream counterpart of each retired local patch
  (Sergey)
- MAINTAINERS: also cover lib/decompress_unlz4.c and
  include/linux/decompress/unlz4.h (Eric)
- Remeasure erofs reads with more runs on an idle machine
- Link to v1: https://lore.kernel.org/r/20260925-lz4-vendor-upstream-v1-0-1c7ffbe21c4b@samsung.com

---
Michal Wilczynski (15):
      lib/decompress_unlz4: use LZ4_decompress_safe_partial() before boot
      lib/lz4: remove LZ4_decompress_fast() and its variants
      zram: lz4hc: reset the dictionary stream once, not per page
      lib/lz4: import upstream LZ4 sources verbatim
      lib/lz4: backport upstream's -Wmissing-prototypes fix
      lib/lz4: backport upstream's aarch64 fast decode loop enablement
      lib/lz4: backport upstream's match length limit fix
      lib/lz4: add the build environment for the vendored sources
      arch: boot: put the LZ4 freestanding headers on the decompressor path
      lib/lz4: switch the compressor to the vendored sources
      lib/lz4: switch the HC compressor to the vendored sources
      lib/lz4: widen LZ4_DECOMPRESS_INPLACE_MARGIN for the fast decode loop
      lib/lz4: switch the decompressor to the vendored sources
      lib/lz4: replace lz4_kernel_api.h's prototypes with <linux/lz4.h>
      MAINTAINERS: add an entry for the LZ4 compression library

 Documentation/filesystems/f2fs.rst   |    2 +
 MAINTAINERS                          |   10 +
 arch/arm/boot/compressed/Makefile    |    3 +
 arch/mips/boot/compressed/Makefile   |    4 +
 arch/parisc/boot/compressed/Makefile |    3 +
 arch/s390/boot/Makefile              |    3 +
 arch/x86/boot/compressed/Makefile    |    3 +
 drivers/block/zram/backend_lz4.c     |    8 +-
 drivers/block/zram/backend_lz4hc.c   |    8 +-
 include/linux/lz4.h                  |  253 +--
 lib/decompress_unlz4.c               |   17 +-
 lib/lz4/Makefile                     |   16 +-
 lib/lz4/freestanding/limits.h        |   14 +
 lib/lz4/freestanding/stddef.h        |   15 +
 lib/lz4/freestanding/stdint.h        |   14 +
 lib/lz4/freestanding/string.h        |   14 +
 lib/lz4/lz4_compress.c               |  946 +-----------
 lib/lz4/lz4_decompress.c             |  724 +--------
 lib/lz4/lz4_deps.h                   |  118 ++
 lib/lz4/lz4_kernel_api.h             |   34 +
 lib/lz4/lz4defs.h                    |  247 ---
 lib/lz4/lz4hc_compress.c             |  792 +---------
 lib/lz4/upstream/lz4.c               | 2824 ++++++++++++++++++++++++++++++++++
 lib/lz4/upstream/lz4.h               |  884 +++++++++++
 lib/lz4/upstream/lz4hc.c             | 2192 ++++++++++++++++++++++++++
 lib/lz4/upstream/lz4hc.h             |  414 +++++
 26 files changed, 6813 insertions(+), 2749 deletions(-)
---
base-commit: 93f51579e7df248780214094418f205253383cc5
change-id: 20260904-lz4-vendor-upstream-4bc2202d0ffc

Best regards,
-- 
Michal Wilczynski <m.wilczynski@samsung.com>



_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

[PATCH v2 01/15] lib/decompress_unlz4: use LZ4_decompress_safe_partial() before boot

[Michal Wilczynski]

The pre-boot decompressor is the only in-tree user of
LZ4_decompress_fast(), which upstream has deprecated since v1.9. It
never checks the input against the chunk length in the header.

Decode with LZ4_decompress_safe_partial() instead, bounded by the chunk
length and by the expected output length. Count every chunk against the
size the image ends with, and fail a chunk that decodes short or comes
after that size is reached.

x86 decompresses in place; the margin arch/x86/boot/header.S reserves,
(output >> 8) + 128K, covers the safe decoder's 32-byte overshoot.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 lib/decompress_unlz4.c | 13 ++++++++-----
 1 file changed, 8 insertions(+), 5 deletions(-)

diff --git a/lib/decompress_unlz4.c b/lib/decompress_unlz4.c
index c0dbb3cea915eb91d8f4ac26c38c4785bbf19770..dc20c22e087900e21da49f9fcb267e6b511e1072 100644
--- a/lib/decompress_unlz4.c
+++ b/lib/decompress_unlz4.c
@@ -151,14 +151,17 @@ STATIC inline int INIT unlz4(u8 *input, long in_len,
 			}
 		}
 #ifdef PREBOOT
-		if (out_len >= uncomp_chunksize) {
+		if (out_len >= uncomp_chunksize)
 			dest_len = uncomp_chunksize;
-			out_len -= dest_len;
-		} else
+		else
 			dest_len = out_len;
+		out_len -= dest_len;
 
-		ret = LZ4_decompress_fast(inp, outp, dest_len);
-		chunksize = ret;
+		ret = LZ4_decompress_safe_partial(inp, outp, chunksize,
+						  dest_len, dest_len);
+		/* A chunk once the image's stated size is reached is corrupt. */
+		if (!dest_len || ret != (int)dest_len)
+			ret = -1;
 #else
 		dest_len = uncomp_chunksize;
 

-- 
2.34.1

[f2fs-dev] [PATCH v2 01/15] lib/decompress_unlz4: use LZ4_decompress_safe_partial() before boot

[Michal Wilczynski]

The pre-boot decompressor is the only in-tree user of
LZ4_decompress_fast(), which upstream has deprecated since v1.9. It
never checks the input against the chunk length in the header.

Decode with LZ4_decompress_safe_partial() instead, bounded by the chunk
length and by the expected output length. Count every chunk against the
size the image ends with, and fail a chunk that decodes short or comes
after that size is reached.

x86 decompresses in place; the margin arch/x86/boot/header.S reserves,
(output >> 8) + 128K, covers the safe decoder's 32-byte overshoot.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 lib/decompress_unlz4.c | 13 ++++++++-----
 1 file changed, 8 insertions(+), 5 deletions(-)

diff --git a/lib/decompress_unlz4.c b/lib/decompress_unlz4.c
index c0dbb3cea915eb91d8f4ac26c38c4785bbf19770..dc20c22e087900e21da49f9fcb267e6b511e1072 100644
--- a/lib/decompress_unlz4.c
+++ b/lib/decompress_unlz4.c
@@ -151,14 +151,17 @@ STATIC inline int INIT unlz4(u8 *input, long in_len,
 			}
 		}
 #ifdef PREBOOT
-		if (out_len >= uncomp_chunksize) {
+		if (out_len >= uncomp_chunksize)
 			dest_len = uncomp_chunksize;
-			out_len -= dest_len;
-		} else
+		else
 			dest_len = out_len;
+		out_len -= dest_len;
 
-		ret = LZ4_decompress_fast(inp, outp, dest_len);
-		chunksize = ret;
+		ret = LZ4_decompress_safe_partial(inp, outp, chunksize,
+						  dest_len, dest_len);
+		/* A chunk once the image's stated size is reached is corrupt. */
+		if (!dest_len || ret != (int)dest_len)
+			ret = -1;
 #else
 		dest_len = uncomp_chunksize;
 

-- 
2.34.1



_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

Re: [PATCH v2 01/15] lib/decompress_unlz4: use LZ4_decompress_safe_partial() before boot

[sashiko-bot]

Thank you for your contribution! Sashiko AI review found 3 potential
issues to consider:
- [Medium] lib/decompress_unlz4: unvalidated chunksize passed to decoder
  in unlz4()
- [Low] lib/decompress_unlz4: partial decode skips full-block checks in
  unlz4()
- [Low] lib/decompress_unlz4: unlz4() does not check out_len == 0 at
  stream end
--

commit 8a11d492456e8edabb6dad4a7ecb828134c99572
Author: Michal Wilczynski <m.wilczynski@samsung.com>

lib/decompress_unlz4: use LZ4_decompress_safe_partial() before boot

This commit replaces LZ4_decompress_fast() with
LZ4_decompress_safe_partial() in the pre-boot unlz4() path and checks that
each chunk produces the expected output size.

> diff --git a/lib/decompress_unlz4.c b/lib/decompress_unlz4.c
> index c0dbb3cea915e..dc20c22e08790 100644
> --- a/lib/decompress_unlz4.c
> +++ b/lib/decompress_unlz4.c
> @@ -151,14 +151,17 @@ STATIC inline int INIT unlz4(u8 *input, long in_len,
>  			}
>  		}
>  #ifdef PREBOOT
> -		if (out_len >= uncomp_chunksize) {
> +		if (out_len >= uncomp_chunksize)
>  			dest_len = uncomp_chunksize;
> -			out_len -= dest_len;
> -		} else
> +		else
>  			dest_len = out_len;
> +		out_len -= dest_len;
>  
> -		ret = LZ4_decompress_fast(inp, outp, dest_len);
> -		chunksize = ret;
> +		ret = LZ4_decompress_safe_partial(inp, outp, chunksize,
> +						  dest_len, dest_len);

[Severity: Medium]
Can a truncated or corrupted chunk header cause LZ4_decompress_safe_partial()
to read past the end of the input buffer?

Previously, LZ4_decompress_fast() ignored chunksize and overwrote it with the
consumed byte count before the !fill path checked size -= chunksize at the
bottom of the loop:

lib/decompress_unlz4.c:unlz4() {
	...
		if (!fill) {
			inp += 4;
			size -= 4;
		} else {
	...
		if (!fill) {
			size -= chunksize;

			if (size == 0)
				break;
			else if (size < 0) {
				error("data corrupted");
				goto exit_2;
			}
			inp += chunksize;
		}
}

Passing chunksize from get_unaligned_le32(inp) directly to
LZ4_decompress_safe_partial() before checking it against size lets the
decoder read past input + in_len when chunksize > size, or pass a negative
compressedSize when chunksize >= 0x80000000.

> +		/* A chunk once the image's stated size is reached is corrupt. */
> +		if (!dest_len || ret != (int)dest_len)
> +			ret = -1;

[Severity: Low]
Should this call LZ4_decompress_safe(inp, outp, chunksize, dest_len) instead
of LZ4_decompress_safe_partial()?

In LZ4_decompress_generic(), partial_decode mode clamps literal runs and
matches that would exceed dest_len and stops as soon as dest_len bytes are
written without verifying that all chunksize input bytes were consumed.  A
corrupted chunk that expands beyond dest_len or leaves trailing bytes inside
chunksize will still return dest_len and pass this check.

[Severity: Low]
Can a truncated image that ends on a chunk boundary still return success when
out_len has not reached zero?

This check rejects chunks that decode short or arrive after out_len reaches
zero, but if the input stream ends early between chunks (or has chunksize ==
0), the loop jumps to exit_3 and returns 0 with out_len > 0.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261106-lz4-vendor-upstream-v2-0-ee04c3f0d7dc@samsung.com?part=1

[PATCH v2 02/15] lib/lz4: remove LZ4_decompress_fast() and its variants

[Michal Wilczynski]

LZ4_decompress_fast(), LZ4_decompress_fast_continue() and
LZ4_decompress_fast_usingDict() have no users left. Upstream has
deprecated all three since v1.9: they take only the decompressed size,
so a corrupt or malicious stream has nothing to stop the decoder from
reading past the end of its input.

Remove them, their exports and their declarations in <linux/lz4.h>,
along with the two helpers only they used. Doing it before lib/lz4 moves
to upstream LZ4 keeps them out of the glue that move adds.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 include/linux/lz4.h      | 81 ----------------------------------------------
 lib/lz4/lz4_decompress.c | 84 ------------------------------------------------
 2 files changed, 165 deletions(-)

diff --git a/include/linux/lz4.h b/include/linux/lz4.h
index ad6042a718b5428792b795db7a8d4ad44c24a985..0df69e2b1ad523c291e3a96ab5595df9239f7252 100644
--- a/include/linux/lz4.h
+++ b/include/linux/lz4.h
@@ -249,27 +249,6 @@ int LZ4_compress_destSize(const char *source, char *dest, int *sourceSizePtr,
  *	Decompression Functions
  **************************************************************************/
 
-/**
- * LZ4_decompress_fast() - Decompresses data from 'source' into 'dest'
- * @source: source address of the compressed data
- * @dest: output buffer address of the uncompressed data
- *	which must be already allocated with 'originalSize' bytes
- * @originalSize: is the original and therefore uncompressed size
- *
- * Decompresses data from 'source' into 'dest'.
- * This function fully respect memory boundaries for properly formed
- * compressed data.
- * It is a bit faster than LZ4_decompress_safe().
- * However, it does not provide any protection against intentionally
- * modified data stream (malicious input).
- * Use this function in trusted environment only
- * (data to decode comes from a trusted source).
- *
- * Return: number of bytes read from the source buffer
- *	or a negative result if decompression fails.
- */
-int LZ4_decompress_fast(const char *source, char *dest, int originalSize);
-
 /**
  * LZ4_decompress_safe() - Decompression protected against buffer overflow
  * @source: source address of the compressed data
@@ -561,44 +540,6 @@ int LZ4_decompress_safe_continue(LZ4_streamDecode_t *LZ4_streamDecode,
 	const char *source, char *dest, int compressedSize,
 	int maxDecompressedSize);
 
-/**
- * LZ4_decompress_fast_continue() - Decompress blocks in streaming mode
- * @LZ4_streamDecode: the 'LZ4_streamDecode_t' structure
- * @source: source address of the compressed data
- * @dest: output buffer address of the uncompressed data
- *	which must be already allocated with 'originalSize' bytes
- * @originalSize: is the original and therefore uncompressed size
- *
- * This decoding function allows decompression of multiple blocks
- * in "streaming" mode.
- * Previously decoded blocks *must* remain available at the memory position
- * where they were decoded (up to 64 KB)
- * In the case of a ring buffers, decoding buffer must be either :
- *    - Exactly same size as encoding buffer, with same update rule
- *      (block boundaries at same positions) In which case,
- *      the decoding & encoding ring buffer can have any size,
- *      including very small ones ( < 64 KB).
- *    - Larger than encoding buffer, by a minimum of maxBlockSize more bytes.
- *      maxBlockSize is implementation dependent.
- *      It's the maximum size you intend to compress into a single block.
- *      In which case, encoding and decoding buffers do not need
- *      to be synchronized, and encoding ring buffer can have any size,
- *      including small ones ( < 64 KB).
- *    - _At least_ 64 KB + 8 bytes + maxBlockSize.
- *      In which case, encoding and decoding buffers do not need to be
- *      synchronized, and encoding ring buffer can have any size,
- *      including larger than decoding buffer. W
- * Whenever these conditions are not possible, save the last 64KB of decoded
- * data into a safe buffer, and indicate where it is saved
- * using LZ4_setStreamDecode()
- *
- * Return: number of bytes decompressed into destination buffer
- *	(necessarily <= maxDecompressedSize)
- *	or a negative result in case of error
- */
-int LZ4_decompress_fast_continue(LZ4_streamDecode_t *LZ4_streamDecode,
-	const char *source, char *dest, int originalSize);
-
 /**
  * LZ4_decompress_safe_usingDict() - Same as LZ4_setStreamDecode()
  *	followed by LZ4_decompress_safe_continue()
@@ -623,28 +564,6 @@ int LZ4_decompress_safe_usingDict(const char *source, char *dest,
 	int compressedSize, int maxDecompressedSize, const char *dictStart,
 	int dictSize);
 
-/**
- * LZ4_decompress_fast_usingDict() - Same as LZ4_setStreamDecode()
- *	followed by LZ4_decompress_fast_continue()
- * @source: source address of the compressed data
- * @dest: output buffer address of the uncompressed data
- *	which must be already allocated with 'originalSize' bytes
- * @originalSize: is the original and therefore uncompressed size
- * @dictStart: pointer to the start of the dictionary in memory
- * @dictSize: size of dictionary
- *
- * This decoding function works the same as
- * a combination of LZ4_setStreamDecode() followed by
- * LZ4_decompress_fast_continue()
- * It is stand-alone, and doesn't need an LZ4_streamDecode_t structure.
- *
- * Return: number of bytes decompressed into destination buffer
- *	(necessarily <= maxDecompressedSize)
- *	or a negative result in case of error
- */
-int LZ4_decompress_fast_usingDict(const char *source, char *dest,
-	int originalSize, const char *dictStart, int dictSize);
-
 #define LZ4_DECOMPRESS_INPLACE_MARGIN(compressedSize)          (((compressedSize) >> 8) + 32)
 
 #ifndef LZ4_DISTANCE_MAX	/* history window size; can be user-defined at compile time */
diff --git a/lib/lz4/lz4_decompress.c b/lib/lz4/lz4_decompress.c
index 3a2cd9acada4a09ef70a344090d48167c46088a1..f89b0e547f2156990c16770f69f89e0d8b46156d 100644
--- a/lib/lz4/lz4_decompress.c
+++ b/lib/lz4/lz4_decompress.c
@@ -474,14 +474,6 @@ int LZ4_decompress_safe_partial(const char *src, char *dst,
 				      noDict, (BYTE *)dst, NULL, 0);
 }
 
-int LZ4_decompress_fast(const char *source, char *dest, int originalSize)
-{
-	return LZ4_decompress_generic(source, dest, 0, originalSize,
-				      endOnOutputSize, decode_full_block,
-				      withPrefix64k,
-				      (BYTE *)dest - 64 * KB, NULL, 0);
-}
-
 /* ===== Instantiate a few more decoding cases, used more than once. ===== */
 
 static int LZ4_decompress_safe_withPrefix64k(const char *source, char *dest,
@@ -517,17 +509,6 @@ static int LZ4_decompress_safe_forceExtDict(const char *source, char *dest,
 				      (const BYTE *)dictStart, dictSize);
 }
 
-static int LZ4_decompress_fast_extDict(const char *source, char *dest,
-				       int originalSize,
-				       const void *dictStart, size_t dictSize)
-{
-	return LZ4_decompress_generic(source, dest,
-				      0, originalSize,
-				      endOnOutputSize, decode_full_block,
-				      usingExtDict, (BYTE *)dest,
-				      (const BYTE *)dictStart, dictSize);
-}
-
 /*
  * The "double dictionary" mode, for use with e.g. ring buffers: the first part
  * of the dictionary is passed as prefix, and the second via dictStart + dictSize.
@@ -546,18 +527,6 @@ int LZ4_decompress_safe_doubleDict(const char *source, char *dest,
 				      (const BYTE *)dictStart, dictSize);
 }
 
-static FORCE_INLINE
-int LZ4_decompress_fast_doubleDict(const char *source, char *dest,
-				   int originalSize, size_t prefixSize,
-				   const void *dictStart, size_t dictSize)
-{
-	return LZ4_decompress_generic(source, dest,
-				      0, originalSize,
-				      endOnOutputSize, decode_full_block,
-				      usingExtDict, (BYTE *)dest - prefixSize,
-				      (const BYTE *)dictStart, dictSize);
-}
-
 /* ===== streaming decompression functions ===== */
 
 int LZ4_setStreamDecode(LZ4_streamDecode_t *LZ4_streamDecode,
@@ -636,45 +605,6 @@ int LZ4_decompress_safe_continue(LZ4_streamDecode_t *LZ4_streamDecode,
 	return result;
 }
 
-int LZ4_decompress_fast_continue(LZ4_streamDecode_t *LZ4_streamDecode,
-	const char *source, char *dest, int originalSize)
-{
-	LZ4_streamDecode_t_internal *lz4sd = &LZ4_streamDecode->internal_donotuse;
-	int result;
-
-	if (lz4sd->prefixSize == 0) {
-		assert(lz4sd->extDictSize == 0);
-		result = LZ4_decompress_fast(source, dest, originalSize);
-		if (result <= 0)
-			return result;
-		lz4sd->prefixSize = originalSize;
-		lz4sd->prefixEnd = (BYTE *)dest + originalSize;
-	} else if (lz4sd->prefixEnd == (BYTE *)dest) {
-		if (lz4sd->prefixSize >= 64 * KB - 1 ||
-		    lz4sd->extDictSize == 0)
-			result = LZ4_decompress_fast(source, dest,
-						     originalSize);
-		else
-			result = LZ4_decompress_fast_doubleDict(source, dest,
-				originalSize, lz4sd->prefixSize,
-				lz4sd->externalDict, lz4sd->extDictSize);
-		if (result <= 0)
-			return result;
-		lz4sd->prefixSize += originalSize;
-		lz4sd->prefixEnd  += originalSize;
-	} else {
-		lz4sd->extDictSize = lz4sd->prefixSize;
-		lz4sd->externalDict = lz4sd->prefixEnd - lz4sd->extDictSize;
-		result = LZ4_decompress_fast_extDict(source, dest,
-			originalSize, lz4sd->externalDict, lz4sd->extDictSize);
-		if (result <= 0)
-			return result;
-		lz4sd->prefixSize = originalSize;
-		lz4sd->prefixEnd = (BYTE *)dest + originalSize;
-	}
-	return result;
-}
-
 int LZ4_decompress_safe_usingDict(const char *source, char *dest,
 				  int compressedSize, int maxOutputSize,
 				  const char *dictStart, int dictSize)
@@ -693,26 +623,12 @@ int LZ4_decompress_safe_usingDict(const char *source, char *dest,
 		compressedSize, maxOutputSize, dictStart, dictSize);
 }
 
-int LZ4_decompress_fast_usingDict(const char *source, char *dest,
-				  int originalSize,
-				  const char *dictStart, int dictSize)
-{
-	if (dictSize == 0 || dictStart + dictSize == dest)
-		return LZ4_decompress_fast(source, dest, originalSize);
-
-	return LZ4_decompress_fast_extDict(source, dest, originalSize,
-		dictStart, dictSize);
-}
-
 #ifndef STATIC
 EXPORT_SYMBOL(LZ4_decompress_safe);
 EXPORT_SYMBOL(LZ4_decompress_safe_partial);
-EXPORT_SYMBOL(LZ4_decompress_fast);
 EXPORT_SYMBOL(LZ4_setStreamDecode);
 EXPORT_SYMBOL(LZ4_decompress_safe_continue);
-EXPORT_SYMBOL(LZ4_decompress_fast_continue);
 EXPORT_SYMBOL(LZ4_decompress_safe_usingDict);
-EXPORT_SYMBOL(LZ4_decompress_fast_usingDict);
 
 MODULE_LICENSE("Dual BSD/GPL");
 MODULE_DESCRIPTION("LZ4 decompressor");

-- 
2.34.1

[f2fs-dev] [PATCH v2 02/15] lib/lz4: remove LZ4_decompress_fast() and its variants

[Michal Wilczynski]

LZ4_decompress_fast(), LZ4_decompress_fast_continue() and
LZ4_decompress_fast_usingDict() have no users left. Upstream has
deprecated all three since v1.9: they take only the decompressed size,
so a corrupt or malicious stream has nothing to stop the decoder from
reading past the end of its input.

Remove them, their exports and their declarations in <linux/lz4.h>,
along with the two helpers only they used. Doing it before lib/lz4 moves
to upstream LZ4 keeps them out of the glue that move adds.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 include/linux/lz4.h      | 81 ----------------------------------------------
 lib/lz4/lz4_decompress.c | 84 ------------------------------------------------
 2 files changed, 165 deletions(-)

diff --git a/include/linux/lz4.h b/include/linux/lz4.h
index ad6042a718b5428792b795db7a8d4ad44c24a985..0df69e2b1ad523c291e3a96ab5595df9239f7252 100644
--- a/include/linux/lz4.h
+++ b/include/linux/lz4.h
@@ -249,27 +249,6 @@ int LZ4_compress_destSize(const char *source, char *dest, int *sourceSizePtr,
  *	Decompression Functions
  **************************************************************************/
 
-/**
- * LZ4_decompress_fast() - Decompresses data from 'source' into 'dest'
- * @source: source address of the compressed data
- * @dest: output buffer address of the uncompressed data
- *	which must be already allocated with 'originalSize' bytes
- * @originalSize: is the original and therefore uncompressed size
- *
- * Decompresses data from 'source' into 'dest'.
- * This function fully respect memory boundaries for properly formed
- * compressed data.
- * It is a bit faster than LZ4_decompress_safe().
- * However, it does not provide any protection against intentionally
- * modified data stream (malicious input).
- * Use this function in trusted environment only
- * (data to decode comes from a trusted source).
- *
- * Return: number of bytes read from the source buffer
- *	or a negative result if decompression fails.
- */
-int LZ4_decompress_fast(const char *source, char *dest, int originalSize);
-
 /**
  * LZ4_decompress_safe() - Decompression protected against buffer overflow
  * @source: source address of the compressed data
@@ -561,44 +540,6 @@ int LZ4_decompress_safe_continue(LZ4_streamDecode_t *LZ4_streamDecode,
 	const char *source, char *dest, int compressedSize,
 	int maxDecompressedSize);
 
-/**
- * LZ4_decompress_fast_continue() - Decompress blocks in streaming mode
- * @LZ4_streamDecode: the 'LZ4_streamDecode_t' structure
- * @source: source address of the compressed data
- * @dest: output buffer address of the uncompressed data
- *	which must be already allocated with 'originalSize' bytes
- * @originalSize: is the original and therefore uncompressed size
- *
- * This decoding function allows decompression of multiple blocks
- * in "streaming" mode.
- * Previously decoded blocks *must* remain available at the memory position
- * where they were decoded (up to 64 KB)
- * In the case of a ring buffers, decoding buffer must be either :
- *    - Exactly same size as encoding buffer, with same update rule
- *      (block boundaries at same positions) In which case,
- *      the decoding & encoding ring buffer can have any size,
- *      including very small ones ( < 64 KB).
- *    - Larger than encoding buffer, by a minimum of maxBlockSize more bytes.
- *      maxBlockSize is implementation dependent.
- *      It's the maximum size you intend to compress into a single block.
- *      In which case, encoding and decoding buffers do not need
- *      to be synchronized, and encoding ring buffer can have any size,
- *      including small ones ( < 64 KB).
- *    - _At least_ 64 KB + 8 bytes + maxBlockSize.
- *      In which case, encoding and decoding buffers do not need to be
- *      synchronized, and encoding ring buffer can have any size,
- *      including larger than decoding buffer. W
- * Whenever these conditions are not possible, save the last 64KB of decoded
- * data into a safe buffer, and indicate where it is saved
- * using LZ4_setStreamDecode()
- *
- * Return: number of bytes decompressed into destination buffer
- *	(necessarily <= maxDecompressedSize)
- *	or a negative result in case of error
- */
-int LZ4_decompress_fast_continue(LZ4_streamDecode_t *LZ4_streamDecode,
-	const char *source, char *dest, int originalSize);
-
 /**
  * LZ4_decompress_safe_usingDict() - Same as LZ4_setStreamDecode()
  *	followed by LZ4_decompress_safe_continue()
@@ -623,28 +564,6 @@ int LZ4_decompress_safe_usingDict(const char *source, char *dest,
 	int compressedSize, int maxDecompressedSize, const char *dictStart,
 	int dictSize);
 
-/**
- * LZ4_decompress_fast_usingDict() - Same as LZ4_setStreamDecode()
- *	followed by LZ4_decompress_fast_continue()
- * @source: source address of the compressed data
- * @dest: output buffer address of the uncompressed data
- *	which must be already allocated with 'originalSize' bytes
- * @originalSize: is the original and therefore uncompressed size
- * @dictStart: pointer to the start of the dictionary in memory
- * @dictSize: size of dictionary
- *
- * This decoding function works the same as
- * a combination of LZ4_setStreamDecode() followed by
- * LZ4_decompress_fast_continue()
- * It is stand-alone, and doesn't need an LZ4_streamDecode_t structure.
- *
- * Return: number of bytes decompressed into destination buffer
- *	(necessarily <= maxDecompressedSize)
- *	or a negative result in case of error
- */
-int LZ4_decompress_fast_usingDict(const char *source, char *dest,
-	int originalSize, const char *dictStart, int dictSize);
-
 #define LZ4_DECOMPRESS_INPLACE_MARGIN(compressedSize)          (((compressedSize) >> 8) + 32)
 
 #ifndef LZ4_DISTANCE_MAX	/* history window size; can be user-defined at compile time */
diff --git a/lib/lz4/lz4_decompress.c b/lib/lz4/lz4_decompress.c
index 3a2cd9acada4a09ef70a344090d48167c46088a1..f89b0e547f2156990c16770f69f89e0d8b46156d 100644
--- a/lib/lz4/lz4_decompress.c
+++ b/lib/lz4/lz4_decompress.c
@@ -474,14 +474,6 @@ int LZ4_decompress_safe_partial(const char *src, char *dst,
 				      noDict, (BYTE *)dst, NULL, 0);
 }
 
-int LZ4_decompress_fast(const char *source, char *dest, int originalSize)
-{
-	return LZ4_decompress_generic(source, dest, 0, originalSize,
-				      endOnOutputSize, decode_full_block,
-				      withPrefix64k,
-				      (BYTE *)dest - 64 * KB, NULL, 0);
-}
-
 /* ===== Instantiate a few more decoding cases, used more than once. ===== */
 
 static int LZ4_decompress_safe_withPrefix64k(const char *source, char *dest,
@@ -517,17 +509,6 @@ static int LZ4_decompress_safe_forceExtDict(const char *source, char *dest,
 				      (const BYTE *)dictStart, dictSize);
 }
 
-static int LZ4_decompress_fast_extDict(const char *source, char *dest,
-				       int originalSize,
-				       const void *dictStart, size_t dictSize)
-{
-	return LZ4_decompress_generic(source, dest,
-				      0, originalSize,
-				      endOnOutputSize, decode_full_block,
-				      usingExtDict, (BYTE *)dest,
-				      (const BYTE *)dictStart, dictSize);
-}
-
 /*
  * The "double dictionary" mode, for use with e.g. ring buffers: the first part
  * of the dictionary is passed as prefix, and the second via dictStart + dictSize.
@@ -546,18 +527,6 @@ int LZ4_decompress_safe_doubleDict(const char *source, char *dest,
 				      (const BYTE *)dictStart, dictSize);
 }
 
-static FORCE_INLINE
-int LZ4_decompress_fast_doubleDict(const char *source, char *dest,
-				   int originalSize, size_t prefixSize,
-				   const void *dictStart, size_t dictSize)
-{
-	return LZ4_decompress_generic(source, dest,
-				      0, originalSize,
-				      endOnOutputSize, decode_full_block,
-				      usingExtDict, (BYTE *)dest - prefixSize,
-				      (const BYTE *)dictStart, dictSize);
-}
-
 /* ===== streaming decompression functions ===== */
 
 int LZ4_setStreamDecode(LZ4_streamDecode_t *LZ4_streamDecode,
@@ -636,45 +605,6 @@ int LZ4_decompress_safe_continue(LZ4_streamDecode_t *LZ4_streamDecode,
 	return result;
 }
 
-int LZ4_decompress_fast_continue(LZ4_streamDecode_t *LZ4_streamDecode,
-	const char *source, char *dest, int originalSize)
-{
-	LZ4_streamDecode_t_internal *lz4sd = &LZ4_streamDecode->internal_donotuse;
-	int result;
-
-	if (lz4sd->prefixSize == 0) {
-		assert(lz4sd->extDictSize == 0);
-		result = LZ4_decompress_fast(source, dest, originalSize);
-		if (result <= 0)
-			return result;
-		lz4sd->prefixSize = originalSize;
-		lz4sd->prefixEnd = (BYTE *)dest + originalSize;
-	} else if (lz4sd->prefixEnd == (BYTE *)dest) {
-		if (lz4sd->prefixSize >= 64 * KB - 1 ||
-		    lz4sd->extDictSize == 0)
-			result = LZ4_decompress_fast(source, dest,
-						     originalSize);
-		else
-			result = LZ4_decompress_fast_doubleDict(source, dest,
-				originalSize, lz4sd->prefixSize,
-				lz4sd->externalDict, lz4sd->extDictSize);
-		if (result <= 0)
-			return result;
-		lz4sd->prefixSize += originalSize;
-		lz4sd->prefixEnd  += originalSize;
-	} else {
-		lz4sd->extDictSize = lz4sd->prefixSize;
-		lz4sd->externalDict = lz4sd->prefixEnd - lz4sd->extDictSize;
-		result = LZ4_decompress_fast_extDict(source, dest,
-			originalSize, lz4sd->externalDict, lz4sd->extDictSize);
-		if (result <= 0)
-			return result;
-		lz4sd->prefixSize = originalSize;
-		lz4sd->prefixEnd = (BYTE *)dest + originalSize;
-	}
-	return result;
-}
-
 int LZ4_decompress_safe_usingDict(const char *source, char *dest,
 				  int compressedSize, int maxOutputSize,
 				  const char *dictStart, int dictSize)
@@ -693,26 +623,12 @@ int LZ4_decompress_safe_usingDict(const char *source, char *dest,
 		compressedSize, maxOutputSize, dictStart, dictSize);
 }
 
-int LZ4_decompress_fast_usingDict(const char *source, char *dest,
-				  int originalSize,
-				  const char *dictStart, int dictSize)
-{
-	if (dictSize == 0 || dictStart + dictSize == dest)
-		return LZ4_decompress_fast(source, dest, originalSize);
-
-	return LZ4_decompress_fast_extDict(source, dest, originalSize,
-		dictStart, dictSize);
-}
-
 #ifndef STATIC
 EXPORT_SYMBOL(LZ4_decompress_safe);
 EXPORT_SYMBOL(LZ4_decompress_safe_partial);
-EXPORT_SYMBOL(LZ4_decompress_fast);
 EXPORT_SYMBOL(LZ4_setStreamDecode);
 EXPORT_SYMBOL(LZ4_decompress_safe_continue);
-EXPORT_SYMBOL(LZ4_decompress_fast_continue);
 EXPORT_SYMBOL(LZ4_decompress_safe_usingDict);
-EXPORT_SYMBOL(LZ4_decompress_fast_usingDict);
 
 MODULE_LICENSE("Dual BSD/GPL");
 MODULE_DESCRIPTION("LZ4 decompressor");

-- 
2.34.1



_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

Re: [PATCH v2 02/15] lib/lz4: remove LZ4_decompress_fast() and its variants

[sashiko-bot]

> LZ4_decompress_fast(), LZ4_decompress_fast_continue() and
> LZ4_decompress_fast_usingDict() have no users left. Upstream has
> deprecated all three since v1.9: they take only the decompressed size,
> so a corrupt or malicious stream has nothing to stop the decoder from
> reading past the end of its input.
> 
> Remove them, their exports and their declarations in <linux/lz4.h>,
> along with the two helpers only they used. Doing it before lib/lz4 moves
> to upstream LZ4 keeps them out of the glue that move adds.
> 
> Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>

Sashiko has reviewed this patch and found no issues.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261106-lz4-vendor-upstream-v2-0-ee04c3f0d7dc@samsung.com?part=2

[PATCH v2 03/15] zram: lz4hc: reset the dictionary stream once, not per page

[Michal Wilczynski]

With a dictionary, lz4hc_compress() calls LZ4_resetStreamHC() and then
LZ4_loadDictHC() for every page. LZ4_loadDictHC() re-initialises the
stream itself and keeps the level, so one reset at creation is enough.

The reset is cheap today, but upstream LZ4, which lib/lz4 moves to later
in this series, clears the whole 256K state in both calls. Reset once in
lz4hc_create() instead.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 drivers/block/zram/backend_lz4hc.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/block/zram/backend_lz4hc.c b/drivers/block/zram/backend_lz4hc.c
index d8aa01bb258fec760b597a88296dfaa4e0049693..06286c805cf7c56bd760f460cc854dc4014f293b 100644
--- a/drivers/block/zram/backend_lz4hc.c
+++ b/drivers/block/zram/backend_lz4hc.c
@@ -72,6 +72,7 @@ static int lz4hc_create(struct zcomp_params *params, struct zcomp_ctx *ctx)
 		zctx->cstrm = kzalloc_obj(*zctx->cstrm);
 		if (!zctx->cstrm)
 			goto error;
+		LZ4_resetStreamHC(zctx->cstrm, params->level);
 	}
 
 	return 0;
@@ -92,8 +93,7 @@ static int lz4hc_compress(struct zcomp_params *params, struct zcomp_ctx *ctx,
 				      req->dst_len, params->level,
 				      zctx->mem);
 	} else {
-		/* Cstrm needs to be reset */
-		LZ4_resetStreamHC(zctx->cstrm, params->level);
+		/* Reloading the dict resets cstrm and keeps its level */
 		ret = LZ4_loadDictHC(zctx->cstrm, params->dict,
 				     params->dict_sz);
 		if (ret != params->dict_sz)

-- 
2.34.1

[f2fs-dev] [PATCH v2 03/15] zram: lz4hc: reset the dictionary stream once, not per page

[Michal Wilczynski]

With a dictionary, lz4hc_compress() calls LZ4_resetStreamHC() and then
LZ4_loadDictHC() for every page. LZ4_loadDictHC() re-initialises the
stream itself and keeps the level, so one reset at creation is enough.

The reset is cheap today, but upstream LZ4, which lib/lz4 moves to later
in this series, clears the whole 256K state in both calls. Reset once in
lz4hc_create() instead.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 drivers/block/zram/backend_lz4hc.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/block/zram/backend_lz4hc.c b/drivers/block/zram/backend_lz4hc.c
index d8aa01bb258fec760b597a88296dfaa4e0049693..06286c805cf7c56bd760f460cc854dc4014f293b 100644
--- a/drivers/block/zram/backend_lz4hc.c
+++ b/drivers/block/zram/backend_lz4hc.c
@@ -72,6 +72,7 @@ static int lz4hc_create(struct zcomp_params *params, struct zcomp_ctx *ctx)
 		zctx->cstrm = kzalloc_obj(*zctx->cstrm);
 		if (!zctx->cstrm)
 			goto error;
+		LZ4_resetStreamHC(zctx->cstrm, params->level);
 	}
 
 	return 0;
@@ -92,8 +93,7 @@ static int lz4hc_compress(struct zcomp_params *params, struct zcomp_ctx *ctx,
 				      req->dst_len, params->level,
 				      zctx->mem);
 	} else {
-		/* Cstrm needs to be reset */
-		LZ4_resetStreamHC(zctx->cstrm, params->level);
+		/* Reloading the dict resets cstrm and keeps its level */
 		ret = LZ4_loadDictHC(zctx->cstrm, params->dict,
 				     params->dict_sz);
 		if (ret != params->dict_sz)

-- 
2.34.1



_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

Re: [PATCH v2 03/15] zram: lz4hc: reset the dictionary stream once, not per page

[sashiko-bot]

Thank you for your contribution! Sashiko AI review found no new issues
in this patch, but noticed 1 related pre-existing issue it does not
introduce:
- [Medium] zram: high-order memory allocation using kzalloc in
  lz4hc_create() [1]
--

commit 2f51683dc7418bdcd0730283cb7003d8754fcd5f
Author: Michal Wilczynski <m.wilczynski@samsung.com>

zram: lz4hc: reset the dictionary stream once, not per page

This commit moves the LZ4_resetStreamHC() call for dictionary-backed
streams from lz4hc_compress() to lz4hc_create() so that the stream state is
only reset once at creation time rather than on every compressed page.

> diff --git a/drivers/block/zram/backend_lz4hc.c b/drivers/block/zram/backend_lz4hc.c
> index d8aa01bb258fe..06286c805cf7c 100644
> --- a/drivers/block/zram/backend_lz4hc.c
> +++ b/drivers/block/zram/backend_lz4hc.c
> @@ -72,6 +72,7 @@ static int lz4hc_create(struct zcomp_params *params, struct zcomp_ctx *ctx)
>  		zctx->cstrm = kzalloc_obj(*zctx->cstrm);
>  		if (!zctx->cstrm)
>  			goto error;
> +		LZ4_resetStreamHC(zctx->cstrm, params->level);
>  	}

[Severity: Medium]
This problem wasn't introduced by this patch, but should zctx->cstrm be
allocated with vmalloc() or kvzalloc() instead of kzalloc_obj() [1]?

LZ4_streamHC_t is 262,192 bytes (LZ4_STREAMHCSIZE), so kzalloc_obj()
requests an order-7 physically contiguous allocation.  Unlike the
non-dictionary path right above, which uses vmalloc(LZ4HC_MEM_COMPRESS),
this allocation can easily fail under normal memory fragmentation during
zram initialization or CPU hotplug.

[1] https://sashiko.dev/#/bug/linux-78ebc045-5e81-4f0b-985e-99483d454b25

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261106-lz4-vendor-upstream-v2-0-ee04c3f0d7dc@samsung.com?part=3

Re: [PATCH v2 03/15] zram: lz4hc: reset the dictionary stream once, not per page

[Sergey Senozhatsky]

On (26/11/06 14:49), sashiko-bot@kernel.org wrote:
> Thank you for your contribution! Sashiko AI review found no new issues
> in this patch, but noticed 1 related pre-existing issue it does not
> introduce:
> - [Medium] zram: high-order memory allocation using kzalloc in
>   lz4hc_create() [1]
> --
> 
> commit 2f51683dc7418bdcd0730283cb7003d8754fcd5f
> Author: Michal Wilczynski <m.wilczynski@samsung.com>
> 
> zram: lz4hc: reset the dictionary stream once, not per page
> 
> This commit moves the LZ4_resetStreamHC() call for dictionary-backed
> streams from lz4hc_compress() to lz4hc_create() so that the stream state is
> only reset once at creation time rather than on every compressed page.
> 
> > diff --git a/drivers/block/zram/backend_lz4hc.c b/drivers/block/zram/backend_lz4hc.c
> > index d8aa01bb258fe..06286c805cf7c 100644
> > --- a/drivers/block/zram/backend_lz4hc.c
> > +++ b/drivers/block/zram/backend_lz4hc.c
> > @@ -72,6 +72,7 @@ static int lz4hc_create(struct zcomp_params *params, struct zcomp_ctx *ctx)
> >  		zctx->cstrm = kzalloc_obj(*zctx->cstrm);
> >  		if (!zctx->cstrm)
> >  			goto error;
> > +		LZ4_resetStreamHC(zctx->cstrm, params->level);
> >  	}
> 
> [Severity: Medium]
> This problem wasn't introduced by this patch, but should zctx->cstrm be
> allocated with vmalloc() or kvzalloc() instead of kzalloc_obj() [1]?

That's how I have it in my local tree - vmalloc(LZ4HC_MEM_COMPRESS).
Michal, I suppose you don't base the series on top of linux-next or
Andrew's tree?

We reset the stream when we setup params, so I don't think we need
to do anything in lz4hc_create().

Re: [PATCH v2 03/15] zram: lz4hc: reset the dictionary stream once, not per page

[Michal Wilczynski]

On 11/10/26 06:31, Sergey Senozhatsky wrote:
> On (26/11/06 14:49), sashiko-bot@kernel.org wrote:
>> Thank you for your contribution! Sashiko AI review found no new issues
>> in this patch, but noticed 1 related pre-existing issue it does not
>> introduce:
>> - [Medium] zram: high-order memory allocation using kzalloc in
>>   lz4hc_create() [1]
>> --
>>
>> commit 2f51683dc7418bdcd0730283cb7003d8754fcd5f
>> Author: Michal Wilczynski <m.wilczynski@samsung.com>
>>
>> zram: lz4hc: reset the dictionary stream once, not per page
>>
>> This commit moves the LZ4_resetStreamHC() call for dictionary-backed
>> streams from lz4hc_compress() to lz4hc_create() so that the stream state is
>> only reset once at creation time rather than on every compressed page.
>>
>>> diff --git a/drivers/block/zram/backend_lz4hc.c b/drivers/block/zram/backend_lz4hc.c
>>> index d8aa01bb258fe..06286c805cf7c 100644
>>> --- a/drivers/block/zram/backend_lz4hc.c
>>> +++ b/drivers/block/zram/backend_lz4hc.c
>>> @@ -72,6 +72,7 @@ static int lz4hc_create(struct zcomp_params *params, struct zcomp_ctx *ctx)
>>>  		zctx->cstrm = kzalloc_obj(*zctx->cstrm);
>>>  		if (!zctx->cstrm)
>>>  			goto error;
>>> +		LZ4_resetStreamHC(zctx->cstrm, params->level);
>>>  	}
>>
>> [Severity: Medium]
>> This problem wasn't introduced by this patch, but should zctx->cstrm be
>> allocated with vmalloc() or kvzalloc() instead of kzalloc_obj() [1]?
> 
> That's how I have it in my local tree - vmalloc(LZ4HC_MEM_COMPRESS).
> Michal, I suppose you don't base the series on top of linux-next or
> Andrew's tree?

No it's based on v7.3-rc4. I'll rebase v3 on linux-next.

> 
> We reset the stream when we setup params, so I don't think we need
> to do anything in lz4hc_create().
> 

Agree your change covers it. I'll drop this patch.

Thank you for having a look.


Best regards,
-- 
Michal Wilczynski <m.wilczynski@samsung.com>

[PATCH v2 04/15] lib/lz4: import upstream LZ4 sources verbatim

[Michal Wilczynski]

Add the four upstream files under lib/lz4/upstream/ without modification.
Nothing builds them yet; a later patch adds the build environment.

Copied without modification from the v1.10.0 release tag of
https://github.com/lz4/lz4, tagged 2024-07-21.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 lib/lz4/upstream/lz4.c   | 2829 ++++++++++++++++++++++++++++++++++++++++++++++
 lib/lz4/upstream/lz4.h   |  884 +++++++++++++++
 lib/lz4/upstream/lz4hc.c | 2192 +++++++++++++++++++++++++++++++++++
 lib/lz4/upstream/lz4hc.h |  414 +++++++
 4 files changed, 6319 insertions(+)

diff --git a/lib/lz4/upstream/lz4.c b/lib/lz4/upstream/lz4.c
new file mode 100644
index 0000000000000000000000000000000000000000..a2f7abee19fb9a5c768f2a6c266acf5b571f0855
--- /dev/null
+++ b/lib/lz4/upstream/lz4.c
@@ -0,0 +1,2829 @@
+/*
+   LZ4 - Fast LZ compression algorithm
+   Copyright (C) 2011-2023, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+    - LZ4 homepage : http://www.lz4.org
+    - LZ4 source repository : https://github.com/lz4/lz4
+*/
+
+/*-************************************
+*  Tuning parameters
+**************************************/
+/*
+ * LZ4_HEAPMODE :
+ * Select how stateless compression functions like `LZ4_compress_default()`
+ * allocate memory for their hash table,
+ * in memory stack (0:default, fastest), or in memory heap (1:requires malloc()).
+ */
+#ifndef LZ4_HEAPMODE
+#  define LZ4_HEAPMODE 0
+#endif
+
+/*
+ * LZ4_ACCELERATION_DEFAULT :
+ * Select "acceleration" for LZ4_compress_fast() when parameter value <= 0
+ */
+#define LZ4_ACCELERATION_DEFAULT 1
+/*
+ * LZ4_ACCELERATION_MAX :
+ * Any "acceleration" value higher than this threshold
+ * get treated as LZ4_ACCELERATION_MAX instead (fix #876)
+ */
+#define LZ4_ACCELERATION_MAX 65537
+
+
+/*-************************************
+*  CPU Feature Detection
+**************************************/
+/* LZ4_FORCE_MEMORY_ACCESS
+ * By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable.
+ * Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal.
+ * The below switch allow to select different access method for improved performance.
+ * Method 0 (default) : use `memcpy()`. Safe and portable.
+ * Method 1 : `__packed` statement. It depends on compiler extension (ie, not portable).
+ *            This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`.
+ * Method 2 : direct access. This method is portable but violate C standard.
+ *            It can generate buggy code on targets which assembly generation depends on alignment.
+ *            But in some circumstances, it's the only known way to get the most performance (ie GCC + ARMv6)
+ * See https://fastcompression.blogspot.fr/2015/08/accessing-unaligned-memory.html for details.
+ * Prefer these methods in priority order (0 > 1 > 2)
+ */
+#ifndef LZ4_FORCE_MEMORY_ACCESS   /* can be defined externally */
+#  if defined(__GNUC__) && \
+  ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) \
+  || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
+#    define LZ4_FORCE_MEMORY_ACCESS 2
+#  elif (defined(__INTEL_COMPILER) && !defined(_WIN32)) || defined(__GNUC__) || defined(_MSC_VER)
+#    define LZ4_FORCE_MEMORY_ACCESS 1
+#  endif
+#endif
+
+/*
+ * LZ4_FORCE_SW_BITCOUNT
+ * Define this parameter if your target system or compiler does not support hardware bit count
+ */
+#if defined(_MSC_VER) && defined(_WIN32_WCE)   /* Visual Studio for WinCE doesn't support Hardware bit count */
+#  undef  LZ4_FORCE_SW_BITCOUNT  /* avoid double def */
+#  define LZ4_FORCE_SW_BITCOUNT
+#endif
+
+
+
+/*-************************************
+*  Dependency
+**************************************/
+/*
+ * LZ4_SRC_INCLUDED:
+ * Amalgamation flag, whether lz4.c is included
+ */
+#ifndef LZ4_SRC_INCLUDED
+#  define LZ4_SRC_INCLUDED 1
+#endif
+
+#ifndef LZ4_DISABLE_DEPRECATE_WARNINGS
+#  define LZ4_DISABLE_DEPRECATE_WARNINGS /* due to LZ4_decompress_safe_withPrefix64k */
+#endif
+
+#ifndef LZ4_STATIC_LINKING_ONLY
+#  define LZ4_STATIC_LINKING_ONLY
+#endif
+#include "lz4.h"
+/* see also "memory routines" below */
+
+
+/*-************************************
+*  Compiler Options
+**************************************/
+#if defined(_MSC_VER) && (_MSC_VER >= 1400)  /* Visual Studio 2005+ */
+#  include <intrin.h>               /* only present in VS2005+ */
+#  pragma warning(disable : 4127)   /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 6237)   /* disable: C6237: conditional expression is always 0 */
+#  pragma warning(disable : 6239)   /* disable: C6239: (<non-zero constant> && <expression>) always evaluates to the result of <expression> */
+#  pragma warning(disable : 6240)   /* disable: C6240: (<expression> && <non-zero constant>) always evaluates to the result of <expression> */
+#  pragma warning(disable : 6326)   /* disable: C6326: Potential comparison of a constant with another constant */
+#endif  /* _MSC_VER */
+
+#ifndef LZ4_FORCE_INLINE
+#  if defined (_MSC_VER) && !defined (__clang__)    /* MSVC */
+#    define LZ4_FORCE_INLINE static __forceinline
+#  else
+#    if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#      if defined (__GNUC__) || defined (__clang__)
+#        define LZ4_FORCE_INLINE static inline __attribute__((always_inline))
+#      else
+#        define LZ4_FORCE_INLINE static inline
+#      endif
+#    else
+#      define LZ4_FORCE_INLINE static
+#    endif /* __STDC_VERSION__ */
+#  endif  /* _MSC_VER */
+#endif /* LZ4_FORCE_INLINE */
+
+/* LZ4_FORCE_O2 and LZ4_FORCE_INLINE
+ * gcc on ppc64le generates an unrolled SIMDized loop for LZ4_wildCopy8,
+ * together with a simple 8-byte copy loop as a fall-back path.
+ * However, this optimization hurts the decompression speed by >30%,
+ * because the execution does not go to the optimized loop
+ * for typical compressible data, and all of the preamble checks
+ * before going to the fall-back path become useless overhead.
+ * This optimization happens only with the -O3 flag, and -O2 generates
+ * a simple 8-byte copy loop.
+ * With gcc on ppc64le, all of the LZ4_decompress_* and LZ4_wildCopy8
+ * functions are annotated with __attribute__((optimize("O2"))),
+ * and also LZ4_wildCopy8 is forcibly inlined, so that the O2 attribute
+ * of LZ4_wildCopy8 does not affect the compression speed.
+ */
+#if defined(__PPC64__) && defined(__LITTLE_ENDIAN__) && defined(__GNUC__) && !defined(__clang__)
+#  define LZ4_FORCE_O2  __attribute__((optimize("O2")))
+#  undef LZ4_FORCE_INLINE
+#  define LZ4_FORCE_INLINE  static __inline __attribute__((optimize("O2"),always_inline))
+#else
+#  define LZ4_FORCE_O2
+#endif
+
+#if (defined(__GNUC__) && (__GNUC__ >= 3)) || (defined(__INTEL_COMPILER) && (__INTEL_COMPILER >= 800)) || defined(__clang__)
+#  define expect(expr,value)    (__builtin_expect ((expr),(value)) )
+#else
+#  define expect(expr,value)    (expr)
+#endif
+
+#ifndef likely
+#define likely(expr)     expect((expr) != 0, 1)
+#endif
+#ifndef unlikely
+#define unlikely(expr)   expect((expr) != 0, 0)
+#endif
+
+/* Should the alignment test prove unreliable, for some reason,
+ * it can be disabled by setting LZ4_ALIGN_TEST to 0 */
+#ifndef LZ4_ALIGN_TEST  /* can be externally provided */
+# define LZ4_ALIGN_TEST 1
+#endif
+
+
+/*-************************************
+*  Memory routines
+**************************************/
+
+/*! LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION :
+ *  Disable relatively high-level LZ4/HC functions that use dynamic memory
+ *  allocation functions (malloc(), calloc(), free()).
+ *
+ *  Note that this is a compile-time switch. And since it disables
+ *  public/stable LZ4 v1 API functions, we don't recommend using this
+ *  symbol to generate a library for distribution.
+ *
+ *  The following public functions are removed when this symbol is defined.
+ *  - lz4   : LZ4_createStream, LZ4_freeStream,
+ *            LZ4_createStreamDecode, LZ4_freeStreamDecode, LZ4_create (deprecated)
+ *  - lz4hc : LZ4_createStreamHC, LZ4_freeStreamHC,
+ *            LZ4_createHC (deprecated), LZ4_freeHC  (deprecated)
+ *  - lz4frame, lz4file : All LZ4F_* functions
+ */
+#if defined(LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION)
+#  define ALLOC(s)          lz4_error_memory_allocation_is_disabled
+#  define ALLOC_AND_ZERO(s) lz4_error_memory_allocation_is_disabled
+#  define FREEMEM(p)        lz4_error_memory_allocation_is_disabled
+#elif defined(LZ4_USER_MEMORY_FUNCTIONS)
+/* memory management functions can be customized by user project.
+ * Below functions must exist somewhere in the Project
+ * and be available at link time */
+void* LZ4_malloc(size_t s);
+void* LZ4_calloc(size_t n, size_t s);
+void  LZ4_free(void* p);
+# define ALLOC(s)          LZ4_malloc(s)
+# define ALLOC_AND_ZERO(s) LZ4_calloc(1,s)
+# define FREEMEM(p)        LZ4_free(p)
+#else
+# include <stdlib.h>   /* malloc, calloc, free */
+# define ALLOC(s)          malloc(s)
+# define ALLOC_AND_ZERO(s) calloc(1,s)
+# define FREEMEM(p)        free(p)
+#endif
+
+#if ! LZ4_FREESTANDING
+#  include <string.h>   /* memset, memcpy */
+#endif
+#if !defined(LZ4_memset)
+#  define LZ4_memset(p,v,s) memset((p),(v),(s))
+#endif
+#define MEM_INIT(p,v,s)   LZ4_memset((p),(v),(s))
+
+
+/*-************************************
+*  Common Constants
+**************************************/
+#define MINMATCH 4
+
+#define WILDCOPYLENGTH 8
+#define LASTLITERALS   5   /* see ../doc/lz4_Block_format.md#parsing-restrictions */
+#define MFLIMIT       12   /* see ../doc/lz4_Block_format.md#parsing-restrictions */
+#define MATCH_SAFEGUARD_DISTANCE  ((2*WILDCOPYLENGTH) - MINMATCH)   /* ensure it's possible to write 2 x wildcopyLength without overflowing output buffer */
+#define FASTLOOP_SAFE_DISTANCE 64
+static const int LZ4_minLength = (MFLIMIT+1);
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define LZ4_DISTANCE_ABSOLUTE_MAX 65535
+#if (LZ4_DISTANCE_MAX > LZ4_DISTANCE_ABSOLUTE_MAX)   /* max supported by LZ4 format */
+#  error "LZ4_DISTANCE_MAX is too big : must be <= 65535"
+#endif
+
+#define ML_BITS  4
+#define ML_MASK  ((1U<<ML_BITS)-1)
+#define RUN_BITS (8-ML_BITS)
+#define RUN_MASK ((1U<<RUN_BITS)-1)
+
+
+/*-************************************
+*  Error detection
+**************************************/
+#if defined(LZ4_DEBUG) && (LZ4_DEBUG>=1)
+#  include <assert.h>
+#else
+#  ifndef assert
+#    define assert(condition) ((void)0)
+#  endif
+#endif
+
+#define LZ4_STATIC_ASSERT(c)   { enum { LZ4_static_assert = 1/(int)(!!(c)) }; }   /* use after variable declarations */
+
+#if defined(LZ4_DEBUG) && (LZ4_DEBUG>=2)
+#  include <stdio.h>
+   static int g_debuglog_enable = 1;
+#  define DEBUGLOG(l, ...) {                          \
+        if ((g_debuglog_enable) && (l<=LZ4_DEBUG)) {  \
+            fprintf(stderr, __FILE__  " %i: ", __LINE__); \
+            fprintf(stderr, __VA_ARGS__);             \
+            fprintf(stderr, " \n");                   \
+    }   }
+#else
+#  define DEBUGLOG(l, ...) {}    /* disabled */
+#endif
+
+static int LZ4_isAligned(const void* ptr, size_t alignment)
+{
+    return ((size_t)ptr & (alignment -1)) == 0;
+}
+
+
+/*-************************************
+*  Types
+**************************************/
+#include <limits.h>
+#if defined(__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+# include <stdint.h>
+  typedef  uint8_t BYTE;
+  typedef uint16_t U16;
+  typedef uint32_t U32;
+  typedef  int32_t S32;
+  typedef uint64_t U64;
+  typedef uintptr_t uptrval;
+#else
+# if UINT_MAX != 4294967295UL
+#   error "LZ4 code (when not C++ or C99) assumes that sizeof(int) == 4"
+# endif
+  typedef unsigned char       BYTE;
+  typedef unsigned short      U16;
+  typedef unsigned int        U32;
+  typedef   signed int        S32;
+  typedef unsigned long long  U64;
+  typedef size_t              uptrval;   /* generally true, except OpenVMS-64 */
+#endif
+
+#if defined(__x86_64__)
+  typedef U64    reg_t;   /* 64-bits in x32 mode */
+#else
+  typedef size_t reg_t;   /* 32-bits in x32 mode */
+#endif
+
+typedef enum {
+    notLimited = 0,
+    limitedOutput = 1,
+    fillOutput = 2
+} limitedOutput_directive;
+
+
+/*-************************************
+*  Reading and writing into memory
+**************************************/
+
+/**
+ * LZ4 relies on memcpy with a constant size being inlined. In freestanding
+ * environments, the compiler can't assume the implementation of memcpy() is
+ * standard compliant, so it can't apply its specialized memcpy() inlining
+ * logic. When possible, use __builtin_memcpy() to tell the compiler to analyze
+ * memcpy() as if it were standard compliant, so it can inline it in freestanding
+ * environments. This is needed when decompressing the Linux Kernel, for example.
+ */
+#if !defined(LZ4_memcpy)
+#  if defined(__GNUC__) && (__GNUC__ >= 4)
+#    define LZ4_memcpy(dst, src, size) __builtin_memcpy(dst, src, size)
+#  else
+#    define LZ4_memcpy(dst, src, size) memcpy(dst, src, size)
+#  endif
+#endif
+
+#if !defined(LZ4_memmove)
+#  if defined(__GNUC__) && (__GNUC__ >= 4)
+#    define LZ4_memmove __builtin_memmove
+#  else
+#    define LZ4_memmove memmove
+#  endif
+#endif
+
+static unsigned LZ4_isLittleEndian(void)
+{
+    const union { U32 u; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental */
+    return one.c[0];
+}
+
+#if defined(__GNUC__) || defined(__INTEL_COMPILER)
+#define LZ4_PACK( __Declaration__ ) __Declaration__ __attribute__((__packed__))
+#elif defined(_MSC_VER)
+#define LZ4_PACK( __Declaration__ ) __pragma( pack(push, 1) ) __Declaration__ __pragma( pack(pop))
+#endif
+
+#if defined(LZ4_FORCE_MEMORY_ACCESS) && (LZ4_FORCE_MEMORY_ACCESS==2)
+/* lie to the compiler about data alignment; use with caution */
+
+static U16 LZ4_read16(const void* memPtr) { return *(const U16*) memPtr; }
+static U32 LZ4_read32(const void* memPtr) { return *(const U32*) memPtr; }
+static reg_t LZ4_read_ARCH(const void* memPtr) { return *(const reg_t*) memPtr; }
+
+static void LZ4_write16(void* memPtr, U16 value) { *(U16*)memPtr = value; }
+static void LZ4_write32(void* memPtr, U32 value) { *(U32*)memPtr = value; }
+
+#elif defined(LZ4_FORCE_MEMORY_ACCESS) && (LZ4_FORCE_MEMORY_ACCESS==1)
+
+/* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
+/* currently only defined for gcc and icc */
+LZ4_PACK(typedef struct { U16 u16; }) LZ4_unalign16;
+LZ4_PACK(typedef struct { U32 u32; }) LZ4_unalign32;
+LZ4_PACK(typedef struct { reg_t uArch; }) LZ4_unalignST;
+
+static U16 LZ4_read16(const void* ptr) { return ((const LZ4_unalign16*)ptr)->u16; }
+static U32 LZ4_read32(const void* ptr) { return ((const LZ4_unalign32*)ptr)->u32; }
+static reg_t LZ4_read_ARCH(const void* ptr) { return ((const LZ4_unalignST*)ptr)->uArch; }
+
+static void LZ4_write16(void* memPtr, U16 value) { ((LZ4_unalign16*)memPtr)->u16 = value; }
+static void LZ4_write32(void* memPtr, U32 value) { ((LZ4_unalign32*)memPtr)->u32 = value; }
+
+#else  /* safe and portable access using memcpy() */
+
+static U16 LZ4_read16(const void* memPtr)
+{
+    U16 val; LZ4_memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+static U32 LZ4_read32(const void* memPtr)
+{
+    U32 val; LZ4_memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+static reg_t LZ4_read_ARCH(const void* memPtr)
+{
+    reg_t val; LZ4_memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+static void LZ4_write16(void* memPtr, U16 value)
+{
+    LZ4_memcpy(memPtr, &value, sizeof(value));
+}
+
+static void LZ4_write32(void* memPtr, U32 value)
+{
+    LZ4_memcpy(memPtr, &value, sizeof(value));
+}
+
+#endif /* LZ4_FORCE_MEMORY_ACCESS */
+
+
+static U16 LZ4_readLE16(const void* memPtr)
+{
+    if (LZ4_isLittleEndian()) {
+        return LZ4_read16(memPtr);
+    } else {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)((U16)p[0] | (p[1]<<8));
+    }
+}
+
+#ifdef LZ4_STATIC_LINKING_ONLY_ENDIANNESS_INDEPENDENT_OUTPUT
+static U32 LZ4_readLE32(const void* memPtr)
+{
+    if (LZ4_isLittleEndian()) {
+        return LZ4_read32(memPtr);
+    } else {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U32)p[0] | (p[1]<<8) | (p[2]<<16) | (p[3]<<24);
+    }
+}
+#endif
+
+static void LZ4_writeLE16(void* memPtr, U16 value)
+{
+    if (LZ4_isLittleEndian()) {
+        LZ4_write16(memPtr, value);
+    } else {
+        BYTE* p = (BYTE*)memPtr;
+        p[0] = (BYTE) value;
+        p[1] = (BYTE)(value>>8);
+    }
+}
+
+/* customized variant of memcpy, which can overwrite up to 8 bytes beyond dstEnd */
+LZ4_FORCE_INLINE
+void LZ4_wildCopy8(void* dstPtr, const void* srcPtr, void* dstEnd)
+{
+    BYTE* d = (BYTE*)dstPtr;
+    const BYTE* s = (const BYTE*)srcPtr;
+    BYTE* const e = (BYTE*)dstEnd;
+
+    do { LZ4_memcpy(d,s,8); d+=8; s+=8; } while (d<e);
+}
+
+static const unsigned inc32table[8] = {0, 1, 2,  1,  0,  4, 4, 4};
+static const int      dec64table[8] = {0, 0, 0, -1, -4,  1, 2, 3};
+
+
+#ifndef LZ4_FAST_DEC_LOOP
+#  if defined __i386__ || defined _M_IX86 || defined __x86_64__ || defined _M_X64
+#    define LZ4_FAST_DEC_LOOP 1
+#  elif defined(__aarch64__) && defined(__APPLE__)
+#    define LZ4_FAST_DEC_LOOP 1
+#  elif defined(__aarch64__) && !defined(__clang__)
+     /* On non-Apple aarch64, we disable this optimization for clang because
+      * on certain mobile chipsets, performance is reduced with clang. For
+      * more information refer to https://github.com/lz4/lz4/pull/707 */
+#    define LZ4_FAST_DEC_LOOP 1
+#  else
+#    define LZ4_FAST_DEC_LOOP 0
+#  endif
+#endif
+
+#if LZ4_FAST_DEC_LOOP
+
+LZ4_FORCE_INLINE void
+LZ4_memcpy_using_offset_base(BYTE* dstPtr, const BYTE* srcPtr, BYTE* dstEnd, const size_t offset)
+{
+    assert(srcPtr + offset == dstPtr);
+    if (offset < 8) {
+        LZ4_write32(dstPtr, 0);   /* silence an msan warning when offset==0 */
+        dstPtr[0] = srcPtr[0];
+        dstPtr[1] = srcPtr[1];
+        dstPtr[2] = srcPtr[2];
+        dstPtr[3] = srcPtr[3];
+        srcPtr += inc32table[offset];
+        LZ4_memcpy(dstPtr+4, srcPtr, 4);
+        srcPtr -= dec64table[offset];
+        dstPtr += 8;
+    } else {
+        LZ4_memcpy(dstPtr, srcPtr, 8);
+        dstPtr += 8;
+        srcPtr += 8;
+    }
+
+    LZ4_wildCopy8(dstPtr, srcPtr, dstEnd);
+}
+
+/* customized variant of memcpy, which can overwrite up to 32 bytes beyond dstEnd
+ * this version copies two times 16 bytes (instead of one time 32 bytes)
+ * because it must be compatible with offsets >= 16. */
+LZ4_FORCE_INLINE void
+LZ4_wildCopy32(void* dstPtr, const void* srcPtr, void* dstEnd)
+{
+    BYTE* d = (BYTE*)dstPtr;
+    const BYTE* s = (const BYTE*)srcPtr;
+    BYTE* const e = (BYTE*)dstEnd;
+
+    do { LZ4_memcpy(d,s,16); LZ4_memcpy(d+16,s+16,16); d+=32; s+=32; } while (d<e);
+}
+
+/* LZ4_memcpy_using_offset()  presumes :
+ * - dstEnd >= dstPtr + MINMATCH
+ * - there is at least 12 bytes available to write after dstEnd */
+LZ4_FORCE_INLINE void
+LZ4_memcpy_using_offset(BYTE* dstPtr, const BYTE* srcPtr, BYTE* dstEnd, const size_t offset)
+{
+    BYTE v[8];
+
+    assert(dstEnd >= dstPtr + MINMATCH);
+
+    switch(offset) {
+    case 1:
+        MEM_INIT(v, *srcPtr, 8);
+        break;
+    case 2:
+        LZ4_memcpy(v, srcPtr, 2);
+        LZ4_memcpy(&v[2], srcPtr, 2);
+#if defined(_MSC_VER) && (_MSC_VER <= 1937) /* MSVC 2022 ver 17.7 or earlier */
+#  pragma warning(push)
+#  pragma warning(disable : 6385) /* warning C6385: Reading invalid data from 'v'. */
+#endif
+        LZ4_memcpy(&v[4], v, 4);
+#if defined(_MSC_VER) && (_MSC_VER <= 1937) /* MSVC 2022 ver 17.7 or earlier */
+#  pragma warning(pop)
+#endif
+        break;
+    case 4:
+        LZ4_memcpy(v, srcPtr, 4);
+        LZ4_memcpy(&v[4], srcPtr, 4);
+        break;
+    default:
+        LZ4_memcpy_using_offset_base(dstPtr, srcPtr, dstEnd, offset);
+        return;
+    }
+
+    LZ4_memcpy(dstPtr, v, 8);
+    dstPtr += 8;
+    while (dstPtr < dstEnd) {
+        LZ4_memcpy(dstPtr, v, 8);
+        dstPtr += 8;
+    }
+}
+#endif
+
+
+/*-************************************
+*  Common functions
+**************************************/
+static unsigned LZ4_NbCommonBytes (reg_t val)
+{
+    assert(val != 0);
+    if (LZ4_isLittleEndian()) {
+        if (sizeof(val) == 8) {
+#       if defined(_MSC_VER) && (_MSC_VER >= 1800) && (defined(_M_AMD64) && !defined(_M_ARM64EC)) && !defined(LZ4_FORCE_SW_BITCOUNT)
+/*-*************************************************************************************************
+* ARM64EC is a Microsoft-designed ARM64 ABI compatible with AMD64 applications on ARM64 Windows 11.
+* The ARM64EC ABI does not support AVX/AVX2/AVX512 instructions, nor their relevant intrinsics
+* including _tzcnt_u64. Therefore, we need to neuter the _tzcnt_u64 code path for ARM64EC.
+****************************************************************************************************/
+#         if defined(__clang__) && (__clang_major__ < 10)
+            /* Avoid undefined clang-cl intrinsics issue.
+             * See https://github.com/lz4/lz4/pull/1017 for details. */
+            return (unsigned)__builtin_ia32_tzcnt_u64(val) >> 3;
+#         else
+            /* x64 CPUS without BMI support interpret `TZCNT` as `REP BSF` */
+            return (unsigned)_tzcnt_u64(val) >> 3;
+#         endif
+#       elif defined(_MSC_VER) && defined(_WIN64) && !defined(LZ4_FORCE_SW_BITCOUNT)
+            unsigned long r = 0;
+            _BitScanForward64(&r, (U64)val);
+            return (unsigned)r >> 3;
+#       elif (defined(__clang__) || (defined(__GNUC__) && ((__GNUC__ > 3) || \
+                            ((__GNUC__ == 3) && (__GNUC_MINOR__ >= 4))))) && \
+                                        !defined(LZ4_FORCE_SW_BITCOUNT)
+            return (unsigned)__builtin_ctzll((U64)val) >> 3;
+#       else
+            const U64 m = 0x0101010101010101ULL;
+            val ^= val - 1;
+            return (unsigned)(((U64)((val & (m - 1)) * m)) >> 56);
+#       endif
+        } else /* 32 bits */ {
+#       if defined(_MSC_VER) && (_MSC_VER >= 1400) && !defined(LZ4_FORCE_SW_BITCOUNT)
+            unsigned long r;
+            _BitScanForward(&r, (U32)val);
+            return (unsigned)r >> 3;
+#       elif (defined(__clang__) || (defined(__GNUC__) && ((__GNUC__ > 3) || \
+                            ((__GNUC__ == 3) && (__GNUC_MINOR__ >= 4))))) && \
+                        !defined(__TINYC__) && !defined(LZ4_FORCE_SW_BITCOUNT)
+            return (unsigned)__builtin_ctz((U32)val) >> 3;
+#       else
+            const U32 m = 0x01010101;
+            return (unsigned)((((val - 1) ^ val) & (m - 1)) * m) >> 24;
+#       endif
+        }
+    } else   /* Big Endian CPU */ {
+        if (sizeof(val)==8) {
+#       if (defined(__clang__) || (defined(__GNUC__) && ((__GNUC__ > 3) || \
+                            ((__GNUC__ == 3) && (__GNUC_MINOR__ >= 4))))) && \
+                        !defined(__TINYC__) && !defined(LZ4_FORCE_SW_BITCOUNT)
+            return (unsigned)__builtin_clzll((U64)val) >> 3;
+#       else
+#if 1
+            /* this method is probably faster,
+             * but adds a 128 bytes lookup table */
+            static const unsigned char ctz7_tab[128] = {
+                7, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+                4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+                5, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+                4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+                6, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+                4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+                5, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+                4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+            };
+            U64 const mask = 0x0101010101010101ULL;
+            U64 const t = (((val >> 8) - mask) | val) & mask;
+            return ctz7_tab[(t * 0x0080402010080402ULL) >> 57];
+#else
+            /* this method doesn't consume memory space like the previous one,
+             * but it contains several branches,
+             * that may end up slowing execution */
+            static const U32 by32 = sizeof(val)*4;  /* 32 on 64 bits (goal), 16 on 32 bits.
+            Just to avoid some static analyzer complaining about shift by 32 on 32-bits target.
+            Note that this code path is never triggered in 32-bits mode. */
+            unsigned r;
+            if (!(val>>by32)) { r=4; } else { r=0; val>>=by32; }
+            if (!(val>>16)) { r+=2; val>>=8; } else { val>>=24; }
+            r += (!val);
+            return r;
+#endif
+#       endif
+        } else /* 32 bits */ {
+#       if (defined(__clang__) || (defined(__GNUC__) && ((__GNUC__ > 3) || \
+                            ((__GNUC__ == 3) && (__GNUC_MINOR__ >= 4))))) && \
+                                        !defined(LZ4_FORCE_SW_BITCOUNT)
+            return (unsigned)__builtin_clz((U32)val) >> 3;
+#       else
+            val >>= 8;
+            val = ((((val + 0x00FFFF00) | 0x00FFFFFF) + val) |
+              (val + 0x00FF0000)) >> 24;
+            return (unsigned)val ^ 3;
+#       endif
+        }
+    }
+}
+
+
+#define STEPSIZE sizeof(reg_t)
+LZ4_FORCE_INLINE
+unsigned LZ4_count(const BYTE* pIn, const BYTE* pMatch, const BYTE* pInLimit)
+{
+    const BYTE* const pStart = pIn;
+
+    if (likely(pIn < pInLimit-(STEPSIZE-1))) {
+        reg_t const diff = LZ4_read_ARCH(pMatch) ^ LZ4_read_ARCH(pIn);
+        if (!diff) {
+            pIn+=STEPSIZE; pMatch+=STEPSIZE;
+        } else {
+            return LZ4_NbCommonBytes(diff);
+    }   }
+
+    while (likely(pIn < pInLimit-(STEPSIZE-1))) {
+        reg_t const diff = LZ4_read_ARCH(pMatch) ^ LZ4_read_ARCH(pIn);
+        if (!diff) { pIn+=STEPSIZE; pMatch+=STEPSIZE; continue; }
+        pIn += LZ4_NbCommonBytes(diff);
+        return (unsigned)(pIn - pStart);
+    }
+
+    if ((STEPSIZE==8) && (pIn<(pInLimit-3)) && (LZ4_read32(pMatch) == LZ4_read32(pIn))) { pIn+=4; pMatch+=4; }
+    if ((pIn<(pInLimit-1)) && (LZ4_read16(pMatch) == LZ4_read16(pIn))) { pIn+=2; pMatch+=2; }
+    if ((pIn<pInLimit) && (*pMatch == *pIn)) pIn++;
+    return (unsigned)(pIn - pStart);
+}
+
+
+#ifndef LZ4_COMMONDEFS_ONLY
+/*-************************************
+*  Local Constants
+**************************************/
+static const int LZ4_64Klimit = ((64 KB) + (MFLIMIT-1));
+static const U32 LZ4_skipTrigger = 6;  /* Increase this value ==> compression run slower on incompressible data */
+
+
+/*-************************************
+*  Local Structures and types
+**************************************/
+typedef enum { clearedTable = 0, byPtr, byU32, byU16 } tableType_t;
+
+/**
+ * This enum distinguishes several different modes of accessing previous
+ * content in the stream.
+ *
+ * - noDict        : There is no preceding content.
+ * - withPrefix64k : Table entries up to ctx->dictSize before the current blob
+ *                   blob being compressed are valid and refer to the preceding
+ *                   content (of length ctx->dictSize), which is available
+ *                   contiguously preceding in memory the content currently
+ *                   being compressed.
+ * - usingExtDict  : Like withPrefix64k, but the preceding content is somewhere
+ *                   else in memory, starting at ctx->dictionary with length
+ *                   ctx->dictSize.
+ * - usingDictCtx  : Everything concerning the preceding content is
+ *                   in a separate context, pointed to by ctx->dictCtx.
+ *                   ctx->dictionary, ctx->dictSize, and table entries
+ *                   in the current context that refer to positions
+ *                   preceding the beginning of the current compression are
+ *                   ignored. Instead, ctx->dictCtx->dictionary and ctx->dictCtx
+ *                   ->dictSize describe the location and size of the preceding
+ *                   content, and matches are found by looking in the ctx
+ *                   ->dictCtx->hashTable.
+ */
+typedef enum { noDict = 0, withPrefix64k, usingExtDict, usingDictCtx } dict_directive;
+typedef enum { noDictIssue = 0, dictSmall } dictIssue_directive;
+
+
+/*-************************************
+*  Local Utils
+**************************************/
+int LZ4_versionNumber (void) { return LZ4_VERSION_NUMBER; }
+const char* LZ4_versionString(void) { return LZ4_VERSION_STRING; }
+int LZ4_compressBound(int isize)  { return LZ4_COMPRESSBOUND(isize); }
+int LZ4_sizeofState(void) { return sizeof(LZ4_stream_t); }
+
+
+/*-****************************************
+*  Internal Definitions, used only in Tests
+*******************************************/
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+int LZ4_compress_forceExtDict (LZ4_stream_t* LZ4_dict, const char* source, char* dest, int srcSize);
+
+int LZ4_decompress_safe_forceExtDict(const char* source, char* dest,
+                                     int compressedSize, int maxOutputSize,
+                                     const void* dictStart, size_t dictSize);
+int LZ4_decompress_safe_partial_forceExtDict(const char* source, char* dest,
+                                     int compressedSize, int targetOutputSize, int dstCapacity,
+                                     const void* dictStart, size_t dictSize);
+#if defined (__cplusplus)
+}
+#endif
+
+/*-******************************
+*  Compression functions
+********************************/
+LZ4_FORCE_INLINE U32 LZ4_hash4(U32 sequence, tableType_t const tableType)
+{
+    if (tableType == byU16)
+        return ((sequence * 2654435761U) >> ((MINMATCH*8)-(LZ4_HASHLOG+1)));
+    else
+        return ((sequence * 2654435761U) >> ((MINMATCH*8)-LZ4_HASHLOG));
+}
+
+LZ4_FORCE_INLINE U32 LZ4_hash5(U64 sequence, tableType_t const tableType)
+{
+    const U32 hashLog = (tableType == byU16) ? LZ4_HASHLOG+1 : LZ4_HASHLOG;
+    if (LZ4_isLittleEndian()) {
+        const U64 prime5bytes = 889523592379ULL;
+        return (U32)(((sequence << 24) * prime5bytes) >> (64 - hashLog));
+    } else {
+        const U64 prime8bytes = 11400714785074694791ULL;
+        return (U32)(((sequence >> 24) * prime8bytes) >> (64 - hashLog));
+    }
+}
+
+LZ4_FORCE_INLINE U32 LZ4_hashPosition(const void* const p, tableType_t const tableType)
+{
+    if ((sizeof(reg_t)==8) && (tableType != byU16)) return LZ4_hash5(LZ4_read_ARCH(p), tableType);
+
+#ifdef LZ4_STATIC_LINKING_ONLY_ENDIANNESS_INDEPENDENT_OUTPUT
+    return LZ4_hash4(LZ4_readLE32(p), tableType);
+#else
+    return LZ4_hash4(LZ4_read32(p), tableType);
+#endif
+}
+
+LZ4_FORCE_INLINE void LZ4_clearHash(U32 h, void* tableBase, tableType_t const tableType)
+{
+    switch (tableType)
+    {
+    default: /* fallthrough */
+    case clearedTable: { /* illegal! */ assert(0); return; }
+    case byPtr: { const BYTE** hashTable = (const BYTE**)tableBase; hashTable[h] = NULL; return; }
+    case byU32: { U32* hashTable = (U32*) tableBase; hashTable[h] = 0; return; }
+    case byU16: { U16* hashTable = (U16*) tableBase; hashTable[h] = 0; return; }
+    }
+}
+
+LZ4_FORCE_INLINE void LZ4_putIndexOnHash(U32 idx, U32 h, void* tableBase, tableType_t const tableType)
+{
+    switch (tableType)
+    {
+    default: /* fallthrough */
+    case clearedTable: /* fallthrough */
+    case byPtr: { /* illegal! */ assert(0); return; }
+    case byU32: { U32* hashTable = (U32*) tableBase; hashTable[h] = idx; return; }
+    case byU16: { U16* hashTable = (U16*) tableBase; assert(idx < 65536); hashTable[h] = (U16)idx; return; }
+    }
+}
+
+/* LZ4_putPosition*() : only used in byPtr mode */
+LZ4_FORCE_INLINE void LZ4_putPositionOnHash(const BYTE* p, U32 h,
+                                  void* tableBase, tableType_t const tableType)
+{
+    const BYTE** const hashTable = (const BYTE**)tableBase;
+    assert(tableType == byPtr); (void)tableType;
+    hashTable[h] = p;
+}
+
+LZ4_FORCE_INLINE void LZ4_putPosition(const BYTE* p, void* tableBase, tableType_t tableType)
+{
+    U32 const h = LZ4_hashPosition(p, tableType);
+    LZ4_putPositionOnHash(p, h, tableBase, tableType);
+}
+
+/* LZ4_getIndexOnHash() :
+ * Index of match position registered in hash table.
+ * hash position must be calculated by using base+index, or dictBase+index.
+ * Assumption 1 : only valid if tableType == byU32 or byU16.
+ * Assumption 2 : h is presumed valid (within limits of hash table)
+ */
+LZ4_FORCE_INLINE U32 LZ4_getIndexOnHash(U32 h, const void* tableBase, tableType_t tableType)
+{
+    LZ4_STATIC_ASSERT(LZ4_MEMORY_USAGE > 2);
+    if (tableType == byU32) {
+        const U32* const hashTable = (const U32*) tableBase;
+        assert(h < (1U << (LZ4_MEMORY_USAGE-2)));
+        return hashTable[h];
+    }
+    if (tableType == byU16) {
+        const U16* const hashTable = (const U16*) tableBase;
+        assert(h < (1U << (LZ4_MEMORY_USAGE-1)));
+        return hashTable[h];
+    }
+    assert(0); return 0;  /* forbidden case */
+}
+
+static const BYTE* LZ4_getPositionOnHash(U32 h, const void* tableBase, tableType_t tableType)
+{
+    assert(tableType == byPtr); (void)tableType;
+    { const BYTE* const* hashTable = (const BYTE* const*) tableBase; return hashTable[h]; }
+}
+
+LZ4_FORCE_INLINE const BYTE*
+LZ4_getPosition(const BYTE* p,
+                const void* tableBase, tableType_t tableType)
+{
+    U32 const h = LZ4_hashPosition(p, tableType);
+    return LZ4_getPositionOnHash(h, tableBase, tableType);
+}
+
+LZ4_FORCE_INLINE void
+LZ4_prepareTable(LZ4_stream_t_internal* const cctx,
+           const int inputSize,
+           const tableType_t tableType) {
+    /* If the table hasn't been used, it's guaranteed to be zeroed out, and is
+     * therefore safe to use no matter what mode we're in. Otherwise, we figure
+     * out if it's safe to leave as is or whether it needs to be reset.
+     */
+    if ((tableType_t)cctx->tableType != clearedTable) {
+        assert(inputSize >= 0);
+        if ((tableType_t)cctx->tableType != tableType
+          || ((tableType == byU16) && cctx->currentOffset + (unsigned)inputSize >= 0xFFFFU)
+          || ((tableType == byU32) && cctx->currentOffset > 1 GB)
+          || tableType == byPtr
+          || inputSize >= 4 KB)
+        {
+            DEBUGLOG(4, "LZ4_prepareTable: Resetting table in %p", cctx);
+            MEM_INIT(cctx->hashTable, 0, LZ4_HASHTABLESIZE);
+            cctx->currentOffset = 0;
+            cctx->tableType = (U32)clearedTable;
+        } else {
+            DEBUGLOG(4, "LZ4_prepareTable: Re-use hash table (no reset)");
+        }
+    }
+
+    /* Adding a gap, so all previous entries are > LZ4_DISTANCE_MAX back,
+     * is faster than compressing without a gap.
+     * However, compressing with currentOffset == 0 is faster still,
+     * so we preserve that case.
+     */
+    if (cctx->currentOffset != 0 && tableType == byU32) {
+        DEBUGLOG(5, "LZ4_prepareTable: adding 64KB to currentOffset");
+        cctx->currentOffset += 64 KB;
+    }
+
+    /* Finally, clear history */
+    cctx->dictCtx = NULL;
+    cctx->dictionary = NULL;
+    cctx->dictSize = 0;
+}
+
+/** LZ4_compress_generic_validated() :
+ *  inlined, to ensure branches are decided at compilation time.
+ *  The following conditions are presumed already validated:
+ *  - source != NULL
+ *  - inputSize > 0
+ */
+LZ4_FORCE_INLINE int LZ4_compress_generic_validated(
+                 LZ4_stream_t_internal* const cctx,
+                 const char* const source,
+                 char* const dest,
+                 const int inputSize,
+                 int*  inputConsumed, /* only written when outputDirective == fillOutput */
+                 const int maxOutputSize,
+                 const limitedOutput_directive outputDirective,
+                 const tableType_t tableType,
+                 const dict_directive dictDirective,
+                 const dictIssue_directive dictIssue,
+                 const int acceleration)
+{
+    int result;
+    const BYTE* ip = (const BYTE*)source;
+
+    U32 const startIndex = cctx->currentOffset;
+    const BYTE* base = (const BYTE*)source - startIndex;
+    const BYTE* lowLimit;
+
+    const LZ4_stream_t_internal* dictCtx = (const LZ4_stream_t_internal*) cctx->dictCtx;
+    const BYTE* const dictionary =
+        dictDirective == usingDictCtx ? dictCtx->dictionary : cctx->dictionary;
+    const U32 dictSize =
+        dictDirective == usingDictCtx ? dictCtx->dictSize : cctx->dictSize;
+    const U32 dictDelta =
+        (dictDirective == usingDictCtx) ? startIndex - dictCtx->currentOffset : 0;   /* make indexes in dictCtx comparable with indexes in current context */
+
+    int const maybe_extMem = (dictDirective == usingExtDict) || (dictDirective == usingDictCtx);
+    U32 const prefixIdxLimit = startIndex - dictSize;   /* used when dictDirective == dictSmall */
+    const BYTE* const dictEnd = dictionary ? dictionary + dictSize : dictionary;
+    const BYTE* anchor = (const BYTE*) source;
+    const BYTE* const iend = ip + inputSize;
+    const BYTE* const mflimitPlusOne = iend - MFLIMIT + 1;
+    const BYTE* const matchlimit = iend - LASTLITERALS;
+
+    /* the dictCtx currentOffset is indexed on the start of the dictionary,
+     * while a dictionary in the current context precedes the currentOffset */
+    const BYTE* dictBase = (dictionary == NULL) ? NULL :
+                           (dictDirective == usingDictCtx) ?
+                            dictionary + dictSize - dictCtx->currentOffset :
+                            dictionary + dictSize - startIndex;
+
+    BYTE* op = (BYTE*) dest;
+    BYTE* const olimit = op + maxOutputSize;
+
+    U32 offset = 0;
+    U32 forwardH;
+
+    DEBUGLOG(5, "LZ4_compress_generic_validated: srcSize=%i, tableType=%u", inputSize, tableType);
+    assert(ip != NULL);
+    if (tableType == byU16) assert(inputSize<LZ4_64Klimit);  /* Size too large (not within 64K limit) */
+    if (tableType == byPtr) assert(dictDirective==noDict);   /* only supported use case with byPtr */
+    /* If init conditions are not met, we don't have to mark stream
+     * as having dirty context, since no action was taken yet */
+    if (outputDirective == fillOutput && maxOutputSize < 1) { return 0; } /* Impossible to store anything */
+    assert(acceleration >= 1);
+
+    lowLimit = (const BYTE*)source - (dictDirective == withPrefix64k ? dictSize : 0);
+
+    /* Update context state */
+    if (dictDirective == usingDictCtx) {
+        /* Subsequent linked blocks can't use the dictionary. */
+        /* Instead, they use the block we just compressed. */
+        cctx->dictCtx = NULL;
+        cctx->dictSize = (U32)inputSize;
+    } else {
+        cctx->dictSize += (U32)inputSize;
+    }
+    cctx->currentOffset += (U32)inputSize;
+    cctx->tableType = (U32)tableType;
+
+    if (inputSize<LZ4_minLength) goto _last_literals;        /* Input too small, no compression (all literals) */
+
+    /* First Byte */
+    {   U32 const h = LZ4_hashPosition(ip, tableType);
+        if (tableType == byPtr) {
+            LZ4_putPositionOnHash(ip, h, cctx->hashTable, byPtr);
+        } else {
+            LZ4_putIndexOnHash(startIndex, h, cctx->hashTable, tableType);
+    }   }
+    ip++; forwardH = LZ4_hashPosition(ip, tableType);
+
+    /* Main Loop */
+    for ( ; ; ) {
+        const BYTE* match;
+        BYTE* token;
+        const BYTE* filledIp;
+
+        /* Find a match */
+        if (tableType == byPtr) {
+            const BYTE* forwardIp = ip;
+            int step = 1;
+            int searchMatchNb = acceleration << LZ4_skipTrigger;
+            do {
+                U32 const h = forwardH;
+                ip = forwardIp;
+                forwardIp += step;
+                step = (searchMatchNb++ >> LZ4_skipTrigger);
+
+                if (unlikely(forwardIp > mflimitPlusOne)) goto _last_literals;
+                assert(ip < mflimitPlusOne);
+
+                match = LZ4_getPositionOnHash(h, cctx->hashTable, tableType);
+                forwardH = LZ4_hashPosition(forwardIp, tableType);
+                LZ4_putPositionOnHash(ip, h, cctx->hashTable, tableType);
+
+            } while ( (match+LZ4_DISTANCE_MAX < ip)
+                   || (LZ4_read32(match) != LZ4_read32(ip)) );
+
+        } else {   /* byU32, byU16 */
+
+            const BYTE* forwardIp = ip;
+            int step = 1;
+            int searchMatchNb = acceleration << LZ4_skipTrigger;
+            do {
+                U32 const h = forwardH;
+                U32 const current = (U32)(forwardIp - base);
+                U32 matchIndex = LZ4_getIndexOnHash(h, cctx->hashTable, tableType);
+                assert(matchIndex <= current);
+                assert(forwardIp - base < (ptrdiff_t)(2 GB - 1));
+                ip = forwardIp;
+                forwardIp += step;
+                step = (searchMatchNb++ >> LZ4_skipTrigger);
+
+                if (unlikely(forwardIp > mflimitPlusOne)) goto _last_literals;
+                assert(ip < mflimitPlusOne);
+
+                if (dictDirective == usingDictCtx) {
+                    if (matchIndex < startIndex) {
+                        /* there was no match, try the dictionary */
+                        assert(tableType == byU32);
+                        matchIndex = LZ4_getIndexOnHash(h, dictCtx->hashTable, byU32);
+                        match = dictBase + matchIndex;
+                        matchIndex += dictDelta;   /* make dictCtx index comparable with current context */
+                        lowLimit = dictionary;
+                    } else {
+                        match = base + matchIndex;
+                        lowLimit = (const BYTE*)source;
+                    }
+                } else if (dictDirective == usingExtDict) {
+                    if (matchIndex < startIndex) {
+                        DEBUGLOG(7, "extDict candidate: matchIndex=%5u  <  startIndex=%5u", matchIndex, startIndex);
+                        assert(startIndex - matchIndex >= MINMATCH);
+                        assert(dictBase);
+                        match = dictBase + matchIndex;
+                        lowLimit = dictionary;
+                    } else {
+                        match = base + matchIndex;
+                        lowLimit = (const BYTE*)source;
+                    }
+                } else {   /* single continuous memory segment */
+                    match = base + matchIndex;
+                }
+                forwardH = LZ4_hashPosition(forwardIp, tableType);
+                LZ4_putIndexOnHash(current, h, cctx->hashTable, tableType);
+
+                DEBUGLOG(7, "candidate at pos=%u  (offset=%u \n", matchIndex, current - matchIndex);
+                if ((dictIssue == dictSmall) && (matchIndex < prefixIdxLimit)) { continue; }    /* match outside of valid area */
+                assert(matchIndex < current);
+                if ( ((tableType != byU16) || (LZ4_DISTANCE_MAX < LZ4_DISTANCE_ABSOLUTE_MAX))
+                  && (matchIndex+LZ4_DISTANCE_MAX < current)) {
+                    continue;
+                } /* too far */
+                assert((current - matchIndex) <= LZ4_DISTANCE_MAX);  /* match now expected within distance */
+
+                if (LZ4_read32(match) == LZ4_read32(ip)) {
+                    if (maybe_extMem) offset = current - matchIndex;
+                    break;   /* match found */
+                }
+
+            } while(1);
+        }
+
+        /* Catch up */
+        filledIp = ip;
+        assert(ip > anchor); /* this is always true as ip has been advanced before entering the main loop */
+        if ((match > lowLimit) && unlikely(ip[-1] == match[-1])) {
+            do { ip--; match--; } while (((ip > anchor) & (match > lowLimit)) && (unlikely(ip[-1] == match[-1])));
+        }
+
+        /* Encode Literals */
+        {   unsigned const litLength = (unsigned)(ip - anchor);
+            token = op++;
+            if ((outputDirective == limitedOutput) &&  /* Check output buffer overflow */
+                (unlikely(op + litLength + (2 + 1 + LASTLITERALS) + (litLength/255) > olimit)) ) {
+                return 0;   /* cannot compress within `dst` budget. Stored indexes in hash table are nonetheless fine */
+            }
+            if ((outputDirective == fillOutput) &&
+                (unlikely(op + (litLength+240)/255 /* litlen */ + litLength /* literals */ + 2 /* offset */ + 1 /* token */ + MFLIMIT - MINMATCH /* min last literals so last match is <= end - MFLIMIT */ > olimit))) {
+                op--;
+                goto _last_literals;
+            }
+            if (litLength >= RUN_MASK) {
+                unsigned len = litLength - RUN_MASK;
+                *token = (RUN_MASK<<ML_BITS);
+                for(; len >= 255 ; len-=255) *op++ = 255;
+                *op++ = (BYTE)len;
+            }
+            else *token = (BYTE)(litLength<<ML_BITS);
+
+            /* Copy Literals */
+            LZ4_wildCopy8(op, anchor, op+litLength);
+            op+=litLength;
+            DEBUGLOG(6, "seq.start:%i, literals=%u, match.start:%i",
+                        (int)(anchor-(const BYTE*)source), litLength, (int)(ip-(const BYTE*)source));
+        }
+
+_next_match:
+        /* at this stage, the following variables must be correctly set :
+         * - ip : at start of LZ operation
+         * - match : at start of previous pattern occurrence; can be within current prefix, or within extDict
+         * - offset : if maybe_ext_memSegment==1 (constant)
+         * - lowLimit : must be == dictionary to mean "match is within extDict"; must be == source otherwise
+         * - token and *token : position to write 4-bits for match length; higher 4-bits for literal length supposed already written
+         */
+
+        if ((outputDirective == fillOutput) &&
+            (op + 2 /* offset */ + 1 /* token */ + MFLIMIT - MINMATCH /* min last literals so last match is <= end - MFLIMIT */ > olimit)) {
+            /* the match was too close to the end, rewind and go to last literals */
+            op = token;
+            goto _last_literals;
+        }
+
+        /* Encode Offset */
+        if (maybe_extMem) {   /* static test */
+            DEBUGLOG(6, "             with offset=%u  (ext if > %i)", offset, (int)(ip - (const BYTE*)source));
+            assert(offset <= LZ4_DISTANCE_MAX && offset > 0);
+            LZ4_writeLE16(op, (U16)offset); op+=2;
+        } else  {
+            DEBUGLOG(6, "             with offset=%u  (same segment)", (U32)(ip - match));
+            assert(ip-match <= LZ4_DISTANCE_MAX);
+            LZ4_writeLE16(op, (U16)(ip - match)); op+=2;
+        }
+
+        /* Encode MatchLength */
+        {   unsigned matchCode;
+
+            if ( (dictDirective==usingExtDict || dictDirective==usingDictCtx)
+              && (lowLimit==dictionary) /* match within extDict */ ) {
+                const BYTE* limit = ip + (dictEnd-match);
+                assert(dictEnd > match);
+                if (limit > matchlimit) limit = matchlimit;
+                matchCode = LZ4_count(ip+MINMATCH, match+MINMATCH, limit);
+                ip += (size_t)matchCode + MINMATCH;
+                if (ip==limit) {
+                    unsigned const more = LZ4_count(limit, (const BYTE*)source, matchlimit);
+                    matchCode += more;
+                    ip += more;
+                }
+                DEBUGLOG(6, "             with matchLength=%u starting in extDict", matchCode+MINMATCH);
+            } else {
+                matchCode = LZ4_count(ip+MINMATCH, match+MINMATCH, matchlimit);
+                ip += (size_t)matchCode + MINMATCH;
+                DEBUGLOG(6, "             with matchLength=%u", matchCode+MINMATCH);
+            }
+
+            if ((outputDirective) &&    /* Check output buffer overflow */
+                (unlikely(op + (1 + LASTLITERALS) + (matchCode+240)/255 > olimit)) ) {
+                if (outputDirective == fillOutput) {
+                    /* Match description too long : reduce it */
+                    U32 newMatchCode = 15 /* in token */ - 1 /* to avoid needing a zero byte */ + ((U32)(olimit - op) - 1 - LASTLITERALS) * 255;
+                    ip -= matchCode - newMatchCode;
+                    assert(newMatchCode < matchCode);
+                    matchCode = newMatchCode;
+                    if (unlikely(ip <= filledIp)) {
+                        /* We have already filled up to filledIp so if ip ends up less than filledIp
+                         * we have positions in the hash table beyond the current position. This is
+                         * a problem if we reuse the hash table. So we have to remove these positions
+                         * from the hash table.
+                         */
+                        const BYTE* ptr;
+                        DEBUGLOG(5, "Clearing %u positions", (U32)(filledIp - ip));
+                        for (ptr = ip; ptr <= filledIp; ++ptr) {
+                            U32 const h = LZ4_hashPosition(ptr, tableType);
+                            LZ4_clearHash(h, cctx->hashTable, tableType);
+                        }
+                    }
+                } else {
+                    assert(outputDirective == limitedOutput);
+                    return 0;   /* cannot compress within `dst` budget. Stored indexes in hash table are nonetheless fine */
+                }
+            }
+            if (matchCode >= ML_MASK) {
+                *token += ML_MASK;
+                matchCode -= ML_MASK;
+                LZ4_write32(op, 0xFFFFFFFF);
+                while (matchCode >= 4*255) {
+                    op+=4;
+                    LZ4_write32(op, 0xFFFFFFFF);
+                    matchCode -= 4*255;
+                }
+                op += matchCode / 255;
+                *op++ = (BYTE)(matchCode % 255);
+            } else
+                *token += (BYTE)(matchCode);
+        }
+        /* Ensure we have enough space for the last literals. */
+        assert(!(outputDirective == fillOutput && op + 1 + LASTLITERALS > olimit));
+
+        anchor = ip;
+
+        /* Test end of chunk */
+        if (ip >= mflimitPlusOne) break;
+
+        /* Fill table */
+        {   U32 const h = LZ4_hashPosition(ip-2, tableType);
+            if (tableType == byPtr) {
+                LZ4_putPositionOnHash(ip-2, h, cctx->hashTable, byPtr);
+            } else {
+                U32 const idx = (U32)((ip-2) - base);
+                LZ4_putIndexOnHash(idx, h, cctx->hashTable, tableType);
+        }   }
+
+        /* Test next position */
+        if (tableType == byPtr) {
+
+            match = LZ4_getPosition(ip, cctx->hashTable, tableType);
+            LZ4_putPosition(ip, cctx->hashTable, tableType);
+            if ( (match+LZ4_DISTANCE_MAX >= ip)
+              && (LZ4_read32(match) == LZ4_read32(ip)) )
+            { token=op++; *token=0; goto _next_match; }
+
+        } else {   /* byU32, byU16 */
+
+            U32 const h = LZ4_hashPosition(ip, tableType);
+            U32 const current = (U32)(ip-base);
+            U32 matchIndex = LZ4_getIndexOnHash(h, cctx->hashTable, tableType);
+            assert(matchIndex < current);
+            if (dictDirective == usingDictCtx) {
+                if (matchIndex < startIndex) {
+                    /* there was no match, try the dictionary */
+                    assert(tableType == byU32);
+                    matchIndex = LZ4_getIndexOnHash(h, dictCtx->hashTable, byU32);
+                    match = dictBase + matchIndex;
+                    lowLimit = dictionary;   /* required for match length counter */
+                    matchIndex += dictDelta;
+                } else {
+                    match = base + matchIndex;
+                    lowLimit = (const BYTE*)source;  /* required for match length counter */
+                }
+            } else if (dictDirective==usingExtDict) {
+                if (matchIndex < startIndex) {
+                    assert(dictBase);
+                    match = dictBase + matchIndex;
+                    lowLimit = dictionary;   /* required for match length counter */
+                } else {
+                    match = base + matchIndex;
+                    lowLimit = (const BYTE*)source;   /* required for match length counter */
+                }
+            } else {   /* single memory segment */
+                match = base + matchIndex;
+            }
+            LZ4_putIndexOnHash(current, h, cctx->hashTable, tableType);
+            assert(matchIndex < current);
+            if ( ((dictIssue==dictSmall) ? (matchIndex >= prefixIdxLimit) : 1)
+              && (((tableType==byU16) && (LZ4_DISTANCE_MAX == LZ4_DISTANCE_ABSOLUTE_MAX)) ? 1 : (matchIndex+LZ4_DISTANCE_MAX >= current))
+              && (LZ4_read32(match) == LZ4_read32(ip)) ) {
+                token=op++;
+                *token=0;
+                if (maybe_extMem) offset = current - matchIndex;
+                DEBUGLOG(6, "seq.start:%i, literals=%u, match.start:%i",
+                            (int)(anchor-(const BYTE*)source), 0, (int)(ip-(const BYTE*)source));
+                goto _next_match;
+            }
+        }
+
+        /* Prepare next loop */
+        forwardH = LZ4_hashPosition(++ip, tableType);
+
+    }
+
+_last_literals:
+    /* Encode Last Literals */
+    {   size_t lastRun = (size_t)(iend - anchor);
+        if ( (outputDirective) &&  /* Check output buffer overflow */
+            (op + lastRun + 1 + ((lastRun+255-RUN_MASK)/255) > olimit)) {
+            if (outputDirective == fillOutput) {
+                /* adapt lastRun to fill 'dst' */
+                assert(olimit >= op);
+                lastRun  = (size_t)(olimit-op) - 1/*token*/;
+                lastRun -= (lastRun + 256 - RUN_MASK) / 256;  /*additional length tokens*/
+            } else {
+                assert(outputDirective == limitedOutput);
+                return 0;   /* cannot compress within `dst` budget. Stored indexes in hash table are nonetheless fine */
+            }
+        }
+        DEBUGLOG(6, "Final literal run : %i literals", (int)lastRun);
+        if (lastRun >= RUN_MASK) {
+            size_t accumulator = lastRun - RUN_MASK;
+            *op++ = RUN_MASK << ML_BITS;
+            for(; accumulator >= 255 ; accumulator-=255) *op++ = 255;
+            *op++ = (BYTE) accumulator;
+        } else {
+            *op++ = (BYTE)(lastRun<<ML_BITS);
+        }
+        LZ4_memcpy(op, anchor, lastRun);
+        ip = anchor + lastRun;
+        op += lastRun;
+    }
+
+    if (outputDirective == fillOutput) {
+        *inputConsumed = (int) (((const char*)ip)-source);
+    }
+    result = (int)(((char*)op) - dest);
+    assert(result > 0);
+    DEBUGLOG(5, "LZ4_compress_generic: compressed %i bytes into %i bytes", inputSize, result);
+    return result;
+}
+
+/** LZ4_compress_generic() :
+ *  inlined, to ensure branches are decided at compilation time;
+ *  takes care of src == (NULL, 0)
+ *  and forward the rest to LZ4_compress_generic_validated */
+LZ4_FORCE_INLINE int LZ4_compress_generic(
+                 LZ4_stream_t_internal* const cctx,
+                 const char* const src,
+                 char* const dst,
+                 const int srcSize,
+                 int *inputConsumed, /* only written when outputDirective == fillOutput */
+                 const int dstCapacity,
+                 const limitedOutput_directive outputDirective,
+                 const tableType_t tableType,
+                 const dict_directive dictDirective,
+                 const dictIssue_directive dictIssue,
+                 const int acceleration)
+{
+    DEBUGLOG(5, "LZ4_compress_generic: srcSize=%i, dstCapacity=%i",
+                srcSize, dstCapacity);
+
+    if ((U32)srcSize > (U32)LZ4_MAX_INPUT_SIZE) { return 0; }  /* Unsupported srcSize, too large (or negative) */
+    if (srcSize == 0) {   /* src == NULL supported if srcSize == 0 */
+        if (outputDirective != notLimited && dstCapacity <= 0) return 0;  /* no output, can't write anything */
+        DEBUGLOG(5, "Generating an empty block");
+        assert(outputDirective == notLimited || dstCapacity >= 1);
+        assert(dst != NULL);
+        dst[0] = 0;
+        if (outputDirective == fillOutput) {
+            assert (inputConsumed != NULL);
+            *inputConsumed = 0;
+        }
+        return 1;
+    }
+    assert(src != NULL);
+
+    return LZ4_compress_generic_validated(cctx, src, dst, srcSize,
+                inputConsumed, /* only written into if outputDirective == fillOutput */
+                dstCapacity, outputDirective,
+                tableType, dictDirective, dictIssue, acceleration);
+}
+
+
+int LZ4_compress_fast_extState(void* state, const char* source, char* dest, int inputSize, int maxOutputSize, int acceleration)
+{
+    LZ4_stream_t_internal* const ctx = & LZ4_initStream(state, sizeof(LZ4_stream_t)) -> internal_donotuse;
+    assert(ctx != NULL);
+    if (acceleration < 1) acceleration = LZ4_ACCELERATION_DEFAULT;
+    if (acceleration > LZ4_ACCELERATION_MAX) acceleration = LZ4_ACCELERATION_MAX;
+    if (maxOutputSize >= LZ4_compressBound(inputSize)) {
+        if (inputSize < LZ4_64Klimit) {
+            return LZ4_compress_generic(ctx, source, dest, inputSize, NULL, 0, notLimited, byU16, noDict, noDictIssue, acceleration);
+        } else {
+            const tableType_t tableType = ((sizeof(void*)==4) && ((uptrval)source > LZ4_DISTANCE_MAX)) ? byPtr : byU32;
+            return LZ4_compress_generic(ctx, source, dest, inputSize, NULL, 0, notLimited, tableType, noDict, noDictIssue, acceleration);
+        }
+    } else {
+        if (inputSize < LZ4_64Klimit) {
+            return LZ4_compress_generic(ctx, source, dest, inputSize, NULL, maxOutputSize, limitedOutput, byU16, noDict, noDictIssue, acceleration);
+        } else {
+            const tableType_t tableType = ((sizeof(void*)==4) && ((uptrval)source > LZ4_DISTANCE_MAX)) ? byPtr : byU32;
+            return LZ4_compress_generic(ctx, source, dest, inputSize, NULL, maxOutputSize, limitedOutput, tableType, noDict, noDictIssue, acceleration);
+        }
+    }
+}
+
+/**
+ * LZ4_compress_fast_extState_fastReset() :
+ * A variant of LZ4_compress_fast_extState().
+ *
+ * Using this variant avoids an expensive initialization step. It is only safe
+ * to call if the state buffer is known to be correctly initialized already
+ * (see comment in lz4.h on LZ4_resetStream_fast() for a definition of
+ * "correctly initialized").
+ */
+int LZ4_compress_fast_extState_fastReset(void* state, const char* src, char* dst, int srcSize, int dstCapacity, int acceleration)
+{
+    LZ4_stream_t_internal* const ctx = &((LZ4_stream_t*)state)->internal_donotuse;
+    if (acceleration < 1) acceleration = LZ4_ACCELERATION_DEFAULT;
+    if (acceleration > LZ4_ACCELERATION_MAX) acceleration = LZ4_ACCELERATION_MAX;
+    assert(ctx != NULL);
+
+    if (dstCapacity >= LZ4_compressBound(srcSize)) {
+        if (srcSize < LZ4_64Klimit) {
+            const tableType_t tableType = byU16;
+            LZ4_prepareTable(ctx, srcSize, tableType);
+            if (ctx->currentOffset) {
+                return LZ4_compress_generic(ctx, src, dst, srcSize, NULL, 0, notLimited, tableType, noDict, dictSmall, acceleration);
+            } else {
+                return LZ4_compress_generic(ctx, src, dst, srcSize, NULL, 0, notLimited, tableType, noDict, noDictIssue, acceleration);
+            }
+        } else {
+            const tableType_t tableType = ((sizeof(void*)==4) && ((uptrval)src > LZ4_DISTANCE_MAX)) ? byPtr : byU32;
+            LZ4_prepareTable(ctx, srcSize, tableType);
+            return LZ4_compress_generic(ctx, src, dst, srcSize, NULL, 0, notLimited, tableType, noDict, noDictIssue, acceleration);
+        }
+    } else {
+        if (srcSize < LZ4_64Klimit) {
+            const tableType_t tableType = byU16;
+            LZ4_prepareTable(ctx, srcSize, tableType);
+            if (ctx->currentOffset) {
+                return LZ4_compress_generic(ctx, src, dst, srcSize, NULL, dstCapacity, limitedOutput, tableType, noDict, dictSmall, acceleration);
+            } else {
+                return LZ4_compress_generic(ctx, src, dst, srcSize, NULL, dstCapacity, limitedOutput, tableType, noDict, noDictIssue, acceleration);
+            }
+        } else {
+            const tableType_t tableType = ((sizeof(void*)==4) && ((uptrval)src > LZ4_DISTANCE_MAX)) ? byPtr : byU32;
+            LZ4_prepareTable(ctx, srcSize, tableType);
+            return LZ4_compress_generic(ctx, src, dst, srcSize, NULL, dstCapacity, limitedOutput, tableType, noDict, noDictIssue, acceleration);
+        }
+    }
+}
+
+
+int LZ4_compress_fast(const char* src, char* dest, int srcSize, int dstCapacity, int acceleration)
+{
+    int result;
+#if (LZ4_HEAPMODE)
+    LZ4_stream_t* const ctxPtr = (LZ4_stream_t*)ALLOC(sizeof(LZ4_stream_t));   /* malloc-calloc always properly aligned */
+    if (ctxPtr == NULL) return 0;
+#else
+    LZ4_stream_t ctx;
+    LZ4_stream_t* const ctxPtr = &ctx;
+#endif
+    result = LZ4_compress_fast_extState(ctxPtr, src, dest, srcSize, dstCapacity, acceleration);
+
+#if (LZ4_HEAPMODE)
+    FREEMEM(ctxPtr);
+#endif
+    return result;
+}
+
+
+int LZ4_compress_default(const char* src, char* dst, int srcSize, int dstCapacity)
+{
+    return LZ4_compress_fast(src, dst, srcSize, dstCapacity, 1);
+}
+
+
+/* Note!: This function leaves the stream in an unclean/broken state!
+ * It is not safe to subsequently use the same state with a _fastReset() or
+ * _continue() call without resetting it. */
+static int LZ4_compress_destSize_extState_internal(LZ4_stream_t* state, const char* src, char* dst, int* srcSizePtr, int targetDstSize, int acceleration)
+{
+    void* const s = LZ4_initStream(state, sizeof (*state));
+    assert(s != NULL); (void)s;
+
+    if (targetDstSize >= LZ4_compressBound(*srcSizePtr)) {  /* compression success is guaranteed */
+        return LZ4_compress_fast_extState(state, src, dst, *srcSizePtr, targetDstSize, acceleration);
+    } else {
+        if (*srcSizePtr < LZ4_64Klimit) {
+            return LZ4_compress_generic(&state->internal_donotuse, src, dst, *srcSizePtr, srcSizePtr, targetDstSize, fillOutput, byU16, noDict, noDictIssue, acceleration);
+        } else {
+            tableType_t const addrMode = ((sizeof(void*)==4) && ((uptrval)src > LZ4_DISTANCE_MAX)) ? byPtr : byU32;
+            return LZ4_compress_generic(&state->internal_donotuse, src, dst, *srcSizePtr, srcSizePtr, targetDstSize, fillOutput, addrMode, noDict, noDictIssue, acceleration);
+    }   }
+}
+
+int LZ4_compress_destSize_extState(void* state, const char* src, char* dst, int* srcSizePtr, int targetDstSize, int acceleration)
+{
+    int const r = LZ4_compress_destSize_extState_internal((LZ4_stream_t*)state, src, dst, srcSizePtr, targetDstSize, acceleration);
+    /* clean the state on exit */
+    LZ4_initStream(state, sizeof (LZ4_stream_t));
+    return r;
+}
+
+
+int LZ4_compress_destSize(const char* src, char* dst, int* srcSizePtr, int targetDstSize)
+{
+#if (LZ4_HEAPMODE)
+    LZ4_stream_t* const ctx = (LZ4_stream_t*)ALLOC(sizeof(LZ4_stream_t));   /* malloc-calloc always properly aligned */
+    if (ctx == NULL) return 0;
+#else
+    LZ4_stream_t ctxBody;
+    LZ4_stream_t* const ctx = &ctxBody;
+#endif
+
+    int result = LZ4_compress_destSize_extState_internal(ctx, src, dst, srcSizePtr, targetDstSize, 1);
+
+#if (LZ4_HEAPMODE)
+    FREEMEM(ctx);
+#endif
+    return result;
+}
+
+
+
+/*-******************************
+*  Streaming functions
+********************************/
+
+#if !defined(LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION)
+LZ4_stream_t* LZ4_createStream(void)
+{
+    LZ4_stream_t* const lz4s = (LZ4_stream_t*)ALLOC(sizeof(LZ4_stream_t));
+    LZ4_STATIC_ASSERT(sizeof(LZ4_stream_t) >= sizeof(LZ4_stream_t_internal));
+    DEBUGLOG(4, "LZ4_createStream %p", lz4s);
+    if (lz4s == NULL) return NULL;
+    LZ4_initStream(lz4s, sizeof(*lz4s));
+    return lz4s;
+}
+#endif
+
+static size_t LZ4_stream_t_alignment(void)
+{
+#if LZ4_ALIGN_TEST
+    typedef struct { char c; LZ4_stream_t t; } t_a;
+    return sizeof(t_a) - sizeof(LZ4_stream_t);
+#else
+    return 1;  /* effectively disabled */
+#endif
+}
+
+LZ4_stream_t* LZ4_initStream (void* buffer, size_t size)
+{
+    DEBUGLOG(5, "LZ4_initStream");
+    if (buffer == NULL) { return NULL; }
+    if (size < sizeof(LZ4_stream_t)) { return NULL; }
+    if (!LZ4_isAligned(buffer, LZ4_stream_t_alignment())) return NULL;
+    MEM_INIT(buffer, 0, sizeof(LZ4_stream_t_internal));
+    return (LZ4_stream_t*)buffer;
+}
+
+/* resetStream is now deprecated,
+ * prefer initStream() which is more general */
+void LZ4_resetStream (LZ4_stream_t* LZ4_stream)
+{
+    DEBUGLOG(5, "LZ4_resetStream (ctx:%p)", LZ4_stream);
+    MEM_INIT(LZ4_stream, 0, sizeof(LZ4_stream_t_internal));
+}
+
+void LZ4_resetStream_fast(LZ4_stream_t* ctx) {
+    LZ4_prepareTable(&(ctx->internal_donotuse), 0, byU32);
+}
+
+#if !defined(LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION)
+int LZ4_freeStream (LZ4_stream_t* LZ4_stream)
+{
+    if (!LZ4_stream) return 0;   /* support free on NULL */
+    DEBUGLOG(5, "LZ4_freeStream %p", LZ4_stream);
+    FREEMEM(LZ4_stream);
+    return (0);
+}
+#endif
+
+
+typedef enum { _ld_fast, _ld_slow } LoadDict_mode_e;
+#define HASH_UNIT sizeof(reg_t)
+int LZ4_loadDict_internal(LZ4_stream_t* LZ4_dict,
+                    const char* dictionary, int dictSize,
+                    LoadDict_mode_e _ld)
+{
+    LZ4_stream_t_internal* const dict = &LZ4_dict->internal_donotuse;
+    const tableType_t tableType = byU32;
+    const BYTE* p = (const BYTE*)dictionary;
+    const BYTE* const dictEnd = p + dictSize;
+    U32 idx32;
+
+    DEBUGLOG(4, "LZ4_loadDict (%i bytes from %p into %p)", dictSize, dictionary, LZ4_dict);
+
+    /* It's necessary to reset the context,
+     * and not just continue it with prepareTable()
+     * to avoid any risk of generating overflowing matchIndex
+     * when compressing using this dictionary */
+    LZ4_resetStream(LZ4_dict);
+
+    /* We always increment the offset by 64 KB, since, if the dict is longer,
+     * we truncate it to the last 64k, and if it's shorter, we still want to
+     * advance by a whole window length so we can provide the guarantee that
+     * there are only valid offsets in the window, which allows an optimization
+     * in LZ4_compress_fast_continue() where it uses noDictIssue even when the
+     * dictionary isn't a full 64k. */
+    dict->currentOffset += 64 KB;
+
+    if (dictSize < (int)HASH_UNIT) {
+        return 0;
+    }
+
+    if ((dictEnd - p) > 64 KB) p = dictEnd - 64 KB;
+    dict->dictionary = p;
+    dict->dictSize = (U32)(dictEnd - p);
+    dict->tableType = (U32)tableType;
+    idx32 = dict->currentOffset - dict->dictSize;
+
+    while (p <= dictEnd-HASH_UNIT) {
+        U32 const h = LZ4_hashPosition(p, tableType);
+        /* Note: overwriting => favors positions end of dictionary */
+        LZ4_putIndexOnHash(idx32, h, dict->hashTable, tableType);
+        p+=3; idx32+=3;
+    }
+
+    if (_ld == _ld_slow) {
+        /* Fill hash table with additional references, to improve compression capability */
+        p = dict->dictionary;
+        idx32 = dict->currentOffset - dict->dictSize;
+        while (p <= dictEnd-HASH_UNIT) {
+            U32 const h = LZ4_hashPosition(p, tableType);
+            U32 const limit = dict->currentOffset - 64 KB;
+            if (LZ4_getIndexOnHash(h, dict->hashTable, tableType) <= limit) {
+                /* Note: not overwriting => favors positions beginning of dictionary */
+                LZ4_putIndexOnHash(idx32, h, dict->hashTable, tableType);
+            }
+            p++; idx32++;
+        }
+    }
+
+    return (int)dict->dictSize;
+}
+
+int LZ4_loadDict(LZ4_stream_t* LZ4_dict, const char* dictionary, int dictSize)
+{
+    return LZ4_loadDict_internal(LZ4_dict, dictionary, dictSize, _ld_fast);
+}
+
+int LZ4_loadDictSlow(LZ4_stream_t* LZ4_dict, const char* dictionary, int dictSize)
+{
+    return LZ4_loadDict_internal(LZ4_dict, dictionary, dictSize, _ld_slow);
+}
+
+void LZ4_attach_dictionary(LZ4_stream_t* workingStream, const LZ4_stream_t* dictionaryStream)
+{
+    const LZ4_stream_t_internal* dictCtx = (dictionaryStream == NULL) ? NULL :
+        &(dictionaryStream->internal_donotuse);
+
+    DEBUGLOG(4, "LZ4_attach_dictionary (%p, %p, size %u)",
+             workingStream, dictionaryStream,
+             dictCtx != NULL ? dictCtx->dictSize : 0);
+
+    if (dictCtx != NULL) {
+        /* If the current offset is zero, we will never look in the
+         * external dictionary context, since there is no value a table
+         * entry can take that indicate a miss. In that case, we need
+         * to bump the offset to something non-zero.
+         */
+        if (workingStream->internal_donotuse.currentOffset == 0) {
+            workingStream->internal_donotuse.currentOffset = 64 KB;
+        }
+
+        /* Don't actually attach an empty dictionary.
+         */
+        if (dictCtx->dictSize == 0) {
+            dictCtx = NULL;
+        }
+    }
+    workingStream->internal_donotuse.dictCtx = dictCtx;
+}
+
+
+static void LZ4_renormDictT(LZ4_stream_t_internal* LZ4_dict, int nextSize)
+{
+    assert(nextSize >= 0);
+    if (LZ4_dict->currentOffset + (unsigned)nextSize > 0x80000000) {   /* potential ptrdiff_t overflow (32-bits mode) */
+        /* rescale hash table */
+        U32 const delta = LZ4_dict->currentOffset - 64 KB;
+        const BYTE* dictEnd = LZ4_dict->dictionary + LZ4_dict->dictSize;
+        int i;
+        DEBUGLOG(4, "LZ4_renormDictT");
+        for (i=0; i<LZ4_HASH_SIZE_U32; i++) {
+            if (LZ4_dict->hashTable[i] < delta) LZ4_dict->hashTable[i]=0;
+            else LZ4_dict->hashTable[i] -= delta;
+        }
+        LZ4_dict->currentOffset = 64 KB;
+        if (LZ4_dict->dictSize > 64 KB) LZ4_dict->dictSize = 64 KB;
+        LZ4_dict->dictionary = dictEnd - LZ4_dict->dictSize;
+    }
+}
+
+
+int LZ4_compress_fast_continue (LZ4_stream_t* LZ4_stream,
+                                const char* source, char* dest,
+                                int inputSize, int maxOutputSize,
+                                int acceleration)
+{
+    const tableType_t tableType = byU32;
+    LZ4_stream_t_internal* const streamPtr = &LZ4_stream->internal_donotuse;
+    const char* dictEnd = streamPtr->dictSize ? (const char*)streamPtr->dictionary + streamPtr->dictSize : NULL;
+
+    DEBUGLOG(5, "LZ4_compress_fast_continue (inputSize=%i, dictSize=%u)", inputSize, streamPtr->dictSize);
+
+    LZ4_renormDictT(streamPtr, inputSize);   /* fix index overflow */
+    if (acceleration < 1) acceleration = LZ4_ACCELERATION_DEFAULT;
+    if (acceleration > LZ4_ACCELERATION_MAX) acceleration = LZ4_ACCELERATION_MAX;
+
+    /* invalidate tiny dictionaries */
+    if ( (streamPtr->dictSize < 4)     /* tiny dictionary : not enough for a hash */
+      && (dictEnd != source)           /* prefix mode */
+      && (inputSize > 0)               /* tolerance : don't lose history, in case next invocation would use prefix mode */
+      && (streamPtr->dictCtx == NULL)  /* usingDictCtx */
+      ) {
+        DEBUGLOG(5, "LZ4_compress_fast_continue: dictSize(%u) at addr:%p is too small", streamPtr->dictSize, streamPtr->dictionary);
+        /* remove dictionary existence from history, to employ faster prefix mode */
+        streamPtr->dictSize = 0;
+        streamPtr->dictionary = (const BYTE*)source;
+        dictEnd = source;
+    }
+
+    /* Check overlapping input/dictionary space */
+    {   const char* const sourceEnd = source + inputSize;
+        if ((sourceEnd > (const char*)streamPtr->dictionary) && (sourceEnd < dictEnd)) {
+            streamPtr->dictSize = (U32)(dictEnd - sourceEnd);
+            if (streamPtr->dictSize > 64 KB) streamPtr->dictSize = 64 KB;
+            if (streamPtr->dictSize < 4) streamPtr->dictSize = 0;
+            streamPtr->dictionary = (const BYTE*)dictEnd - streamPtr->dictSize;
+        }
+    }
+
+    /* prefix mode : source data follows dictionary */
+    if (dictEnd == source) {
+        if ((streamPtr->dictSize < 64 KB) && (streamPtr->dictSize < streamPtr->currentOffset))
+            return LZ4_compress_generic(streamPtr, source, dest, inputSize, NULL, maxOutputSize, limitedOutput, tableType, withPrefix64k, dictSmall, acceleration);
+        else
+            return LZ4_compress_generic(streamPtr, source, dest, inputSize, NULL, maxOutputSize, limitedOutput, tableType, withPrefix64k, noDictIssue, acceleration);
+    }
+
+    /* external dictionary mode */
+    {   int result;
+        if (streamPtr->dictCtx) {
+            /* We depend here on the fact that dictCtx'es (produced by
+             * LZ4_loadDict) guarantee that their tables contain no references
+             * to offsets between dictCtx->currentOffset - 64 KB and
+             * dictCtx->currentOffset - dictCtx->dictSize. This makes it safe
+             * to use noDictIssue even when the dict isn't a full 64 KB.
+             */
+            if (inputSize > 4 KB) {
+                /* For compressing large blobs, it is faster to pay the setup
+                 * cost to copy the dictionary's tables into the active context,
+                 * so that the compression loop is only looking into one table.
+                 */
+                LZ4_memcpy(streamPtr, streamPtr->dictCtx, sizeof(*streamPtr));
+                result = LZ4_compress_generic(streamPtr, source, dest, inputSize, NULL, maxOutputSize, limitedOutput, tableType, usingExtDict, noDictIssue, acceleration);
+            } else {
+                result = LZ4_compress_generic(streamPtr, source, dest, inputSize, NULL, maxOutputSize, limitedOutput, tableType, usingDictCtx, noDictIssue, acceleration);
+            }
+        } else {  /* small data <= 4 KB */
+            if ((streamPtr->dictSize < 64 KB) && (streamPtr->dictSize < streamPtr->currentOffset)) {
+                result = LZ4_compress_generic(streamPtr, source, dest, inputSize, NULL, maxOutputSize, limitedOutput, tableType, usingExtDict, dictSmall, acceleration);
+            } else {
+                result = LZ4_compress_generic(streamPtr, source, dest, inputSize, NULL, maxOutputSize, limitedOutput, tableType, usingExtDict, noDictIssue, acceleration);
+            }
+        }
+        streamPtr->dictionary = (const BYTE*)source;
+        streamPtr->dictSize = (U32)inputSize;
+        return result;
+    }
+}
+
+
+/* Hidden debug function, to force-test external dictionary mode */
+int LZ4_compress_forceExtDict (LZ4_stream_t* LZ4_dict, const char* source, char* dest, int srcSize)
+{
+    LZ4_stream_t_internal* const streamPtr = &LZ4_dict->internal_donotuse;
+    int result;
+
+    LZ4_renormDictT(streamPtr, srcSize);
+
+    if ((streamPtr->dictSize < 64 KB) && (streamPtr->dictSize < streamPtr->currentOffset)) {
+        result = LZ4_compress_generic(streamPtr, source, dest, srcSize, NULL, 0, notLimited, byU32, usingExtDict, dictSmall, 1);
+    } else {
+        result = LZ4_compress_generic(streamPtr, source, dest, srcSize, NULL, 0, notLimited, byU32, usingExtDict, noDictIssue, 1);
+    }
+
+    streamPtr->dictionary = (const BYTE*)source;
+    streamPtr->dictSize = (U32)srcSize;
+
+    return result;
+}
+
+
+/*! LZ4_saveDict() :
+ *  If previously compressed data block is not guaranteed to remain available at its memory location,
+ *  save it into a safer place (char* safeBuffer).
+ *  Note : no need to call LZ4_loadDict() afterwards, dictionary is immediately usable,
+ *         one can therefore call LZ4_compress_fast_continue() right after.
+ * @return : saved dictionary size in bytes (necessarily <= dictSize), or 0 if error.
+ */
+int LZ4_saveDict (LZ4_stream_t* LZ4_dict, char* safeBuffer, int dictSize)
+{
+    LZ4_stream_t_internal* const dict = &LZ4_dict->internal_donotuse;
+
+    DEBUGLOG(5, "LZ4_saveDict : dictSize=%i, safeBuffer=%p", dictSize, safeBuffer);
+
+    if ((U32)dictSize > 64 KB) { dictSize = 64 KB; } /* useless to define a dictionary > 64 KB */
+    if ((U32)dictSize > dict->dictSize) { dictSize = (int)dict->dictSize; }
+
+    if (safeBuffer == NULL) assert(dictSize == 0);
+    if (dictSize > 0) {
+        const BYTE* const previousDictEnd = dict->dictionary + dict->dictSize;
+        assert(dict->dictionary);
+        LZ4_memmove(safeBuffer, previousDictEnd - dictSize, (size_t)dictSize);
+    }
+
+    dict->dictionary = (const BYTE*)safeBuffer;
+    dict->dictSize = (U32)dictSize;
+
+    return dictSize;
+}
+
+
+
+/*-*******************************
+ *  Decompression functions
+ ********************************/
+
+typedef enum { decode_full_block = 0, partial_decode = 1 } earlyEnd_directive;
+
+#undef MIN
+#define MIN(a,b)    ( (a) < (b) ? (a) : (b) )
+
+
+/* variant for decompress_unsafe()
+ * does not know end of input
+ * presumes input is well formed
+ * note : will consume at least one byte */
+static size_t read_long_length_no_check(const BYTE** pp)
+{
+    size_t b, l = 0;
+    do { b = **pp; (*pp)++; l += b; } while (b==255);
+    DEBUGLOG(6, "read_long_length_no_check: +length=%zu using %zu input bytes", l, l/255 + 1)
+    return l;
+}
+
+/* core decoder variant for LZ4_decompress_fast*()
+ * for legacy support only : these entry points are deprecated.
+ * - Presumes input is correctly formed (no defense vs malformed inputs)
+ * - Does not know input size (presume input buffer is "large enough")
+ * - Decompress a full block (only)
+ * @return : nb of bytes read from input.
+ * Note : this variant is not optimized for speed, just for maintenance.
+ *        the goal is to remove support of decompress_fast*() variants by v2.0
+**/
+LZ4_FORCE_INLINE int
+LZ4_decompress_unsafe_generic(
+                 const BYTE* const istart,
+                 BYTE* const ostart,
+                 int decompressedSize,
+
+                 size_t prefixSize,
+                 const BYTE* const dictStart,  /* only if dict==usingExtDict */
+                 const size_t dictSize         /* note: =0 if dictStart==NULL */
+                 )
+{
+    const BYTE* ip = istart;
+    BYTE* op = (BYTE*)ostart;
+    BYTE* const oend = ostart + decompressedSize;
+    const BYTE* const prefixStart = ostart - prefixSize;
+
+    DEBUGLOG(5, "LZ4_decompress_unsafe_generic");
+    if (dictStart == NULL) assert(dictSize == 0);
+
+    while (1) {
+        /* start new sequence */
+        unsigned token = *ip++;
+
+        /* literals */
+        {   size_t ll = token >> ML_BITS;
+            if (ll==15) {
+                /* long literal length */
+                ll += read_long_length_no_check(&ip);
+            }
+            if ((size_t)(oend-op) < ll) return -1; /* output buffer overflow */
+            LZ4_memmove(op, ip, ll); /* support in-place decompression */
+            op += ll;
+            ip += ll;
+            if ((size_t)(oend-op) < MFLIMIT) {
+                if (op==oend) break;  /* end of block */
+                DEBUGLOG(5, "invalid: literals end at distance %zi from end of block", oend-op);
+                /* incorrect end of block :
+                 * last match must start at least MFLIMIT==12 bytes before end of output block */
+                return -1;
+        }   }
+
+        /* match */
+        {   size_t ml = token & 15;
+            size_t const offset = LZ4_readLE16(ip);
+            ip+=2;
+
+            if (ml==15) {
+                /* long literal length */
+                ml += read_long_length_no_check(&ip);
+            }
+            ml += MINMATCH;
+
+            if ((size_t)(oend-op) < ml) return -1; /* output buffer overflow */
+
+            {   const BYTE* match = op - offset;
+
+                /* out of range */
+                if (offset > (size_t)(op - prefixStart) + dictSize) {
+                    DEBUGLOG(6, "offset out of range");
+                    return -1;
+                }
+
+                /* check special case : extDict */
+                if (offset > (size_t)(op - prefixStart)) {
+                    /* extDict scenario */
+                    const BYTE* const dictEnd = dictStart + dictSize;
+                    const BYTE* extMatch = dictEnd - (offset - (size_t)(op-prefixStart));
+                    size_t const extml = (size_t)(dictEnd - extMatch);
+                    if (extml > ml) {
+                        /* match entirely within extDict */
+                        LZ4_memmove(op, extMatch, ml);
+                        op += ml;
+                        ml = 0;
+                    } else {
+                        /* match split between extDict & prefix */
+                        LZ4_memmove(op, extMatch, extml);
+                        op += extml;
+                        ml -= extml;
+                    }
+                    match = prefixStart;
+                }
+
+                /* match copy - slow variant, supporting overlap copy */
+                {   size_t u;
+                    for (u=0; u<ml; u++) {
+                        op[u] = match[u];
+            }   }   }
+            op += ml;
+            if ((size_t)(oend-op) < LASTLITERALS) {
+                DEBUGLOG(5, "invalid: match ends at distance %zi from end of block", oend-op);
+                /* incorrect end of block :
+                 * last match must stop at least LASTLITERALS==5 bytes before end of output block */
+                return -1;
+            }
+        } /* match */
+    } /* main loop */
+    return (int)(ip - istart);
+}
+
+
+/* Read the variable-length literal or match length.
+ *
+ * @ip : input pointer
+ * @ilimit : position after which if length is not decoded, the input is necessarily corrupted.
+ * @initial_check - check ip >= ipmax before start of loop.  Returns initial_error if so.
+ * @error (output) - error code.  Must be set to 0 before call.
+**/
+typedef size_t Rvl_t;
+static const Rvl_t rvl_error = (Rvl_t)(-1);
+LZ4_FORCE_INLINE Rvl_t
+read_variable_length(const BYTE** ip, const BYTE* ilimit,
+                     int initial_check)
+{
+    Rvl_t s, length = 0;
+    assert(ip != NULL);
+    assert(*ip !=  NULL);
+    assert(ilimit != NULL);
+    if (initial_check && unlikely((*ip) >= ilimit)) {    /* read limit reached */
+        return rvl_error;
+    }
+    s = **ip;
+    (*ip)++;
+    length += s;
+    if (unlikely((*ip) > ilimit)) {    /* read limit reached */
+        return rvl_error;
+    }
+    /* accumulator overflow detection (32-bit mode only) */
+    if ((sizeof(length) < 8) && unlikely(length > ((Rvl_t)(-1)/2)) ) {
+        return rvl_error;
+    }
+    if (likely(s != 255)) return length;
+    do {
+        s = **ip;
+        (*ip)++;
+        length += s;
+        if (unlikely((*ip) > ilimit)) {    /* read limit reached */
+            return rvl_error;
+        }
+        /* accumulator overflow detection (32-bit mode only) */
+        if ((sizeof(length) < 8) && unlikely(length > ((Rvl_t)(-1)/2)) ) {
+            return rvl_error;
+        }
+    } while (s == 255);
+
+    return length;
+}
+
+/*! LZ4_decompress_generic() :
+ *  This generic decompression function covers all use cases.
+ *  It shall be instantiated several times, using different sets of directives.
+ *  Note that it is important for performance that this function really get inlined,
+ *  in order to remove useless branches during compilation optimization.
+ */
+LZ4_FORCE_INLINE int
+LZ4_decompress_generic(
+                 const char* const src,
+                 char* const dst,
+                 int srcSize,
+                 int outputSize,         /* If endOnInput==endOnInputSize, this value is `dstCapacity` */
+
+                 earlyEnd_directive partialDecoding,  /* full, partial */
+                 dict_directive dict,                 /* noDict, withPrefix64k, usingExtDict */
+                 const BYTE* const lowPrefix,  /* always <= dst, == dst when no prefix */
+                 const BYTE* const dictStart,  /* only if dict==usingExtDict */
+                 const size_t dictSize         /* note : = 0 if noDict */
+                 )
+{
+    if ((src == NULL) || (outputSize < 0)) { return -1; }
+
+    {   const BYTE* ip = (const BYTE*) src;
+        const BYTE* const iend = ip + srcSize;
+
+        BYTE* op = (BYTE*) dst;
+        BYTE* const oend = op + outputSize;
+        BYTE* cpy;
+
+        const BYTE* const dictEnd = (dictStart == NULL) ? NULL : dictStart + dictSize;
+
+        const int checkOffset = (dictSize < (int)(64 KB));
+
+
+        /* Set up the "end" pointers for the shortcut. */
+        const BYTE* const shortiend = iend - 14 /*maxLL*/ - 2 /*offset*/;
+        const BYTE* const shortoend = oend - 14 /*maxLL*/ - 18 /*maxML*/;
+
+        const BYTE* match;
+        size_t offset;
+        unsigned token;
+        size_t length;
+
+
+        DEBUGLOG(5, "LZ4_decompress_generic (srcSize:%i, dstSize:%i)", srcSize, outputSize);
+
+        /* Special cases */
+        assert(lowPrefix <= op);
+        if (unlikely(outputSize==0)) {
+            /* Empty output buffer */
+            if (partialDecoding) return 0;
+            return ((srcSize==1) && (*ip==0)) ? 0 : -1;
+        }
+        if (unlikely(srcSize==0)) { return -1; }
+
+    /* LZ4_FAST_DEC_LOOP:
+     * designed for modern OoO performance cpus,
+     * where copying reliably 32-bytes is preferable to an unpredictable branch.
+     * note : fast loop may show a regression for some client arm chips. */
+#if LZ4_FAST_DEC_LOOP
+        if ((oend - op) < FASTLOOP_SAFE_DISTANCE) {
+            DEBUGLOG(6, "move to safe decode loop");
+            goto safe_decode;
+        }
+
+        /* Fast loop : decode sequences as long as output < oend-FASTLOOP_SAFE_DISTANCE */
+        DEBUGLOG(6, "using fast decode loop");
+        while (1) {
+            /* Main fastloop assertion: We can always wildcopy FASTLOOP_SAFE_DISTANCE */
+            assert(oend - op >= FASTLOOP_SAFE_DISTANCE);
+            assert(ip < iend);
+            token = *ip++;
+            length = token >> ML_BITS;  /* literal length */
+            DEBUGLOG(7, "blockPos%6u: litLength token = %u", (unsigned)(op-(BYTE*)dst), (unsigned)length);
+
+            /* decode literal length */
+            if (length == RUN_MASK) {
+                size_t const addl = read_variable_length(&ip, iend-RUN_MASK, 1);
+                if (addl == rvl_error) {
+                    DEBUGLOG(6, "error reading long literal length");
+                    goto _output_error;
+                }
+                length += addl;
+                if (unlikely((uptrval)(op)+length<(uptrval)(op))) { goto _output_error; } /* overflow detection */
+                if (unlikely((uptrval)(ip)+length<(uptrval)(ip))) { goto _output_error; } /* overflow detection */
+
+                /* copy literals */
+                LZ4_STATIC_ASSERT(MFLIMIT >= WILDCOPYLENGTH);
+                if ((op+length>oend-32) || (ip+length>iend-32)) { goto safe_literal_copy; }
+                LZ4_wildCopy32(op, ip, op+length);
+                ip += length; op += length;
+            } else if (ip <= iend-(16 + 1/*max lit + offset + nextToken*/)) {
+                /* We don't need to check oend, since we check it once for each loop below */
+                DEBUGLOG(7, "copy %u bytes in a 16-bytes stripe", (unsigned)length);
+                /* Literals can only be <= 14, but hope compilers optimize better when copy by a register size */
+                LZ4_memcpy(op, ip, 16);
+                ip += length; op += length;
+            } else {
+                goto safe_literal_copy;
+            }
+
+            /* get offset */
+            offset = LZ4_readLE16(ip); ip+=2;
+            DEBUGLOG(6, "blockPos%6u: offset = %u", (unsigned)(op-(BYTE*)dst), (unsigned)offset);
+            match = op - offset;
+            assert(match <= op);  /* overflow check */
+
+            /* get matchlength */
+            length = token & ML_MASK;
+            DEBUGLOG(7, "  match length token = %u (len==%u)", (unsigned)length, (unsigned)length+MINMATCH);
+
+            if (length == ML_MASK) {
+                size_t const addl = read_variable_length(&ip, iend - LASTLITERALS + 1, 0);
+                if (addl == rvl_error) {
+                    DEBUGLOG(5, "error reading long match length");
+                    goto _output_error;
+                }
+                length += addl;
+                length += MINMATCH;
+                DEBUGLOG(7, "  long match length == %u", (unsigned)length);
+                if (unlikely((uptrval)(op)+length<(uptrval)op)) { goto _output_error; } /* overflow detection */
+                if (op + length >= oend - FASTLOOP_SAFE_DISTANCE) {
+                    goto safe_match_copy;
+                }
+            } else {
+                length += MINMATCH;
+                if (op + length >= oend - FASTLOOP_SAFE_DISTANCE) {
+                    DEBUGLOG(7, "moving to safe_match_copy (ml==%u)", (unsigned)length);
+                    goto safe_match_copy;
+                }
+
+                /* Fastpath check: skip LZ4_wildCopy32 when true */
+                if ((dict == withPrefix64k) || (match >= lowPrefix)) {
+                    if (offset >= 8) {
+                        assert(match >= lowPrefix);
+                        assert(match <= op);
+                        assert(op + 18 <= oend);
+
+                        LZ4_memcpy(op, match, 8);
+                        LZ4_memcpy(op+8, match+8, 8);
+                        LZ4_memcpy(op+16, match+16, 2);
+                        op += length;
+                        continue;
+            }   }   }
+
+            if ( checkOffset && (unlikely(match + dictSize < lowPrefix)) ) {
+                DEBUGLOG(5, "Error : pos=%zi, offset=%zi => outside buffers", op-lowPrefix, op-match);
+                goto _output_error;
+            }
+            /* match starting within external dictionary */
+            if ((dict==usingExtDict) && (match < lowPrefix)) {
+                assert(dictEnd != NULL);
+                if (unlikely(op+length > oend-LASTLITERALS)) {
+                    if (partialDecoding) {
+                        DEBUGLOG(7, "partialDecoding: dictionary match, close to dstEnd");
+                        length = MIN(length, (size_t)(oend-op));
+                    } else {
+                        DEBUGLOG(6, "end-of-block condition violated")
+                        goto _output_error;
+                }   }
+
+                if (length <= (size_t)(lowPrefix-match)) {
+                    /* match fits entirely within external dictionary : just copy */
+                    LZ4_memmove(op, dictEnd - (lowPrefix-match), length);
+                    op += length;
+                } else {
+                    /* match stretches into both external dictionary and current block */
+                    size_t const copySize = (size_t)(lowPrefix - match);
+                    size_t const restSize = length - copySize;
+                    LZ4_memcpy(op, dictEnd - copySize, copySize);
+                    op += copySize;
+                    if (restSize > (size_t)(op - lowPrefix)) {  /* overlap copy */
+                        BYTE* const endOfMatch = op + restSize;
+                        const BYTE* copyFrom = lowPrefix;
+                        while (op < endOfMatch) { *op++ = *copyFrom++; }
+                    } else {
+                        LZ4_memcpy(op, lowPrefix, restSize);
+                        op += restSize;
+                }   }
+                continue;
+            }
+
+            /* copy match within block */
+            cpy = op + length;
+
+            assert((op <= oend) && (oend-op >= 32));
+            if (unlikely(offset<16)) {
+                LZ4_memcpy_using_offset(op, match, cpy, offset);
+            } else {
+                LZ4_wildCopy32(op, match, cpy);
+            }
+
+            op = cpy;   /* wildcopy correction */
+        }
+    safe_decode:
+#endif
+
+        /* Main Loop : decode remaining sequences where output < FASTLOOP_SAFE_DISTANCE */
+        DEBUGLOG(6, "using safe decode loop");
+        while (1) {
+            assert(ip < iend);
+            token = *ip++;
+            length = token >> ML_BITS;  /* literal length */
+            DEBUGLOG(7, "blockPos%6u: litLength token = %u", (unsigned)(op-(BYTE*)dst), (unsigned)length);
+
+            /* A two-stage shortcut for the most common case:
+             * 1) If the literal length is 0..14, and there is enough space,
+             * enter the shortcut and copy 16 bytes on behalf of the literals
+             * (in the fast mode, only 8 bytes can be safely copied this way).
+             * 2) Further if the match length is 4..18, copy 18 bytes in a similar
+             * manner; but we ensure that there's enough space in the output for
+             * those 18 bytes earlier, upon entering the shortcut (in other words,
+             * there is a combined check for both stages).
+             */
+            if ( (length != RUN_MASK)
+                /* strictly "less than" on input, to re-enter the loop with at least one byte */
+              && likely((ip < shortiend) & (op <= shortoend)) ) {
+                /* Copy the literals */
+                LZ4_memcpy(op, ip, 16);
+                op += length; ip += length;
+
+                /* The second stage: prepare for match copying, decode full info.
+                 * If it doesn't work out, the info won't be wasted. */
+                length = token & ML_MASK; /* match length */
+                DEBUGLOG(7, "blockPos%6u: matchLength token = %u (len=%u)", (unsigned)(op-(BYTE*)dst), (unsigned)length, (unsigned)length + 4);
+                offset = LZ4_readLE16(ip); ip += 2;
+                match = op - offset;
+                assert(match <= op); /* check overflow */
+
+                /* Do not deal with overlapping matches. */
+                if ( (length != ML_MASK)
+                  && (offset >= 8)
+                  && (dict==withPrefix64k || match >= lowPrefix) ) {
+                    /* Copy the match. */
+                    LZ4_memcpy(op + 0, match + 0, 8);
+                    LZ4_memcpy(op + 8, match + 8, 8);
+                    LZ4_memcpy(op +16, match +16, 2);
+                    op += length + MINMATCH;
+                    /* Both stages worked, load the next token. */
+                    continue;
+                }
+
+                /* The second stage didn't work out, but the info is ready.
+                 * Propel it right to the point of match copying. */
+                goto _copy_match;
+            }
+
+            /* decode literal length */
+            if (length == RUN_MASK) {
+                size_t const addl = read_variable_length(&ip, iend-RUN_MASK, 1);
+                if (addl == rvl_error) { goto _output_error; }
+                length += addl;
+                if (unlikely((uptrval)(op)+length<(uptrval)(op))) { goto _output_error; } /* overflow detection */
+                if (unlikely((uptrval)(ip)+length<(uptrval)(ip))) { goto _output_error; } /* overflow detection */
+            }
+
+#if LZ4_FAST_DEC_LOOP
+        safe_literal_copy:
+#endif
+            /* copy literals */
+            cpy = op+length;
+
+            LZ4_STATIC_ASSERT(MFLIMIT >= WILDCOPYLENGTH);
+            if ((cpy>oend-MFLIMIT) || (ip+length>iend-(2+1+LASTLITERALS))) {
+                /* We've either hit the input parsing restriction or the output parsing restriction.
+                 * In the normal scenario, decoding a full block, it must be the last sequence,
+                 * otherwise it's an error (invalid input or dimensions).
+                 * In partialDecoding scenario, it's necessary to ensure there is no buffer overflow.
+                 */
+                if (partialDecoding) {
+                    /* Since we are partial decoding we may be in this block because of the output parsing
+                     * restriction, which is not valid since the output buffer is allowed to be undersized.
+                     */
+                    DEBUGLOG(7, "partialDecoding: copying literals, close to input or output end")
+                    DEBUGLOG(7, "partialDecoding: literal length = %u", (unsigned)length);
+                    DEBUGLOG(7, "partialDecoding: remaining space in dstBuffer : %i", (int)(oend - op));
+                    DEBUGLOG(7, "partialDecoding: remaining space in srcBuffer : %i", (int)(iend - ip));
+                    /* Finishing in the middle of a literals segment,
+                     * due to lack of input.
+                     */
+                    if (ip+length > iend) {
+                        length = (size_t)(iend-ip);
+                        cpy = op + length;
+                    }
+                    /* Finishing in the middle of a literals segment,
+                     * due to lack of output space.
+                     */
+                    if (cpy > oend) {
+                        cpy = oend;
+                        assert(op<=oend);
+                        length = (size_t)(oend-op);
+                    }
+                } else {
+                     /* We must be on the last sequence (or invalid) because of the parsing limitations
+                      * so check that we exactly consume the input and don't overrun the output buffer.
+                      */
+                    if ((ip+length != iend) || (cpy > oend)) {
+                        DEBUGLOG(5, "should have been last run of literals")
+                        DEBUGLOG(5, "ip(%p) + length(%i) = %p != iend (%p)", ip, (int)length, ip+length, iend);
+                        DEBUGLOG(5, "or cpy(%p) > (oend-MFLIMIT)(%p)", cpy, oend-MFLIMIT);
+                        DEBUGLOG(5, "after writing %u bytes / %i bytes available", (unsigned)(op-(BYTE*)dst), outputSize);
+                        goto _output_error;
+                    }
+                }
+                LZ4_memmove(op, ip, length);  /* supports overlapping memory regions, for in-place decompression scenarios */
+                ip += length;
+                op += length;
+                /* Necessarily EOF when !partialDecoding.
+                 * When partialDecoding, it is EOF if we've either
+                 * filled the output buffer or
+                 * can't proceed with reading an offset for following match.
+                 */
+                if (!partialDecoding || (cpy == oend) || (ip >= (iend-2))) {
+                    break;
+                }
+            } else {
+                LZ4_wildCopy8(op, ip, cpy);   /* can overwrite up to 8 bytes beyond cpy */
+                ip += length; op = cpy;
+            }
+
+            /* get offset */
+            offset = LZ4_readLE16(ip); ip+=2;
+            match = op - offset;
+
+            /* get matchlength */
+            length = token & ML_MASK;
+            DEBUGLOG(7, "blockPos%6u: matchLength token = %u", (unsigned)(op-(BYTE*)dst), (unsigned)length);
+
+    _copy_match:
+            if (length == ML_MASK) {
+                size_t const addl = read_variable_length(&ip, iend - LASTLITERALS + 1, 0);
+                if (addl == rvl_error) { goto _output_error; }
+                length += addl;
+                if (unlikely((uptrval)(op)+length<(uptrval)op)) goto _output_error;   /* overflow detection */
+            }
+            length += MINMATCH;
+
+#if LZ4_FAST_DEC_LOOP
+        safe_match_copy:
+#endif
+            if ((checkOffset) && (unlikely(match + dictSize < lowPrefix))) goto _output_error;   /* Error : offset outside buffers */
+            /* match starting within external dictionary */
+            if ((dict==usingExtDict) && (match < lowPrefix)) {
+                assert(dictEnd != NULL);
+                if (unlikely(op+length > oend-LASTLITERALS)) {
+                    if (partialDecoding) length = MIN(length, (size_t)(oend-op));
+                    else goto _output_error;   /* doesn't respect parsing restriction */
+                }
+
+                if (length <= (size_t)(lowPrefix-match)) {
+                    /* match fits entirely within external dictionary : just copy */
+                    LZ4_memmove(op, dictEnd - (lowPrefix-match), length);
+                    op += length;
+                } else {
+                    /* match stretches into both external dictionary and current block */
+                    size_t const copySize = (size_t)(lowPrefix - match);
+                    size_t const restSize = length - copySize;
+                    LZ4_memcpy(op, dictEnd - copySize, copySize);
+                    op += copySize;
+                    if (restSize > (size_t)(op - lowPrefix)) {  /* overlap copy */
+                        BYTE* const endOfMatch = op + restSize;
+                        const BYTE* copyFrom = lowPrefix;
+                        while (op < endOfMatch) *op++ = *copyFrom++;
+                    } else {
+                        LZ4_memcpy(op, lowPrefix, restSize);
+                        op += restSize;
+                }   }
+                continue;
+            }
+            assert(match >= lowPrefix);
+
+            /* copy match within block */
+            cpy = op + length;
+
+            /* partialDecoding : may end anywhere within the block */
+            assert(op<=oend);
+            if (partialDecoding && (cpy > oend-MATCH_SAFEGUARD_DISTANCE)) {
+                size_t const mlen = MIN(length, (size_t)(oend-op));
+                const BYTE* const matchEnd = match + mlen;
+                BYTE* const copyEnd = op + mlen;
+                if (matchEnd > op) {   /* overlap copy */
+                    while (op < copyEnd) { *op++ = *match++; }
+                } else {
+                    LZ4_memcpy(op, match, mlen);
+                }
+                op = copyEnd;
+                if (op == oend) { break; }
+                continue;
+            }
+
+            if (unlikely(offset<8)) {
+                LZ4_write32(op, 0);   /* silence msan warning when offset==0 */
+                op[0] = match[0];
+                op[1] = match[1];
+                op[2] = match[2];
+                op[3] = match[3];
+                match += inc32table[offset];
+                LZ4_memcpy(op+4, match, 4);
+                match -= dec64table[offset];
+            } else {
+                LZ4_memcpy(op, match, 8);
+                match += 8;
+            }
+            op += 8;
+
+            if (unlikely(cpy > oend-MATCH_SAFEGUARD_DISTANCE)) {
+                BYTE* const oCopyLimit = oend - (WILDCOPYLENGTH-1);
+                if (cpy > oend-LASTLITERALS) { goto _output_error; } /* Error : last LASTLITERALS bytes must be literals (uncompressed) */
+                if (op < oCopyLimit) {
+                    LZ4_wildCopy8(op, match, oCopyLimit);
+                    match += oCopyLimit - op;
+                    op = oCopyLimit;
+                }
+                while (op < cpy) { *op++ = *match++; }
+            } else {
+                LZ4_memcpy(op, match, 8);
+                if (length > 16) { LZ4_wildCopy8(op+8, match+8, cpy); }
+            }
+            op = cpy;   /* wildcopy correction */
+        }
+
+        /* end of decoding */
+        DEBUGLOG(5, "decoded %i bytes", (int) (((char*)op)-dst));
+        return (int) (((char*)op)-dst);     /* Nb of output bytes decoded */
+
+        /* Overflow error detected */
+    _output_error:
+        return (int) (-(((const char*)ip)-src))-1;
+    }
+}
+
+
+/*===== Instantiate the API decoding functions. =====*/
+
+LZ4_FORCE_O2
+int LZ4_decompress_safe(const char* source, char* dest, int compressedSize, int maxDecompressedSize)
+{
+    return LZ4_decompress_generic(source, dest, compressedSize, maxDecompressedSize,
+                                  decode_full_block, noDict,
+                                  (BYTE*)dest, NULL, 0);
+}
+
+LZ4_FORCE_O2
+int LZ4_decompress_safe_partial(const char* src, char* dst, int compressedSize, int targetOutputSize, int dstCapacity)
+{
+    dstCapacity = MIN(targetOutputSize, dstCapacity);
+    return LZ4_decompress_generic(src, dst, compressedSize, dstCapacity,
+                                  partial_decode,
+                                  noDict, (BYTE*)dst, NULL, 0);
+}
+
+LZ4_FORCE_O2
+int LZ4_decompress_fast(const char* source, char* dest, int originalSize)
+{
+    DEBUGLOG(5, "LZ4_decompress_fast");
+    return LZ4_decompress_unsafe_generic(
+                (const BYTE*)source, (BYTE*)dest, originalSize,
+                0, NULL, 0);
+}
+
+/*===== Instantiate a few more decoding cases, used more than once. =====*/
+
+LZ4_FORCE_O2 /* Exported, an obsolete API function. */
+int LZ4_decompress_safe_withPrefix64k(const char* source, char* dest, int compressedSize, int maxOutputSize)
+{
+    return LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize,
+                                  decode_full_block, withPrefix64k,
+                                  (BYTE*)dest - 64 KB, NULL, 0);
+}
+
+LZ4_FORCE_O2
+static int LZ4_decompress_safe_partial_withPrefix64k(const char* source, char* dest, int compressedSize, int targetOutputSize, int dstCapacity)
+{
+    dstCapacity = MIN(targetOutputSize, dstCapacity);
+    return LZ4_decompress_generic(source, dest, compressedSize, dstCapacity,
+                                  partial_decode, withPrefix64k,
+                                  (BYTE*)dest - 64 KB, NULL, 0);
+}
+
+/* Another obsolete API function, paired with the previous one. */
+int LZ4_decompress_fast_withPrefix64k(const char* source, char* dest, int originalSize)
+{
+    return LZ4_decompress_unsafe_generic(
+                (const BYTE*)source, (BYTE*)dest, originalSize,
+                64 KB, NULL, 0);
+}
+
+LZ4_FORCE_O2
+static int LZ4_decompress_safe_withSmallPrefix(const char* source, char* dest, int compressedSize, int maxOutputSize,
+                                               size_t prefixSize)
+{
+    return LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize,
+                                  decode_full_block, noDict,
+                                  (BYTE*)dest-prefixSize, NULL, 0);
+}
+
+LZ4_FORCE_O2
+static int LZ4_decompress_safe_partial_withSmallPrefix(const char* source, char* dest, int compressedSize, int targetOutputSize, int dstCapacity,
+                                               size_t prefixSize)
+{
+    dstCapacity = MIN(targetOutputSize, dstCapacity);
+    return LZ4_decompress_generic(source, dest, compressedSize, dstCapacity,
+                                  partial_decode, noDict,
+                                  (BYTE*)dest-prefixSize, NULL, 0);
+}
+
+LZ4_FORCE_O2
+int LZ4_decompress_safe_forceExtDict(const char* source, char* dest,
+                                     int compressedSize, int maxOutputSize,
+                                     const void* dictStart, size_t dictSize)
+{
+    DEBUGLOG(5, "LZ4_decompress_safe_forceExtDict");
+    return LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize,
+                                  decode_full_block, usingExtDict,
+                                  (BYTE*)dest, (const BYTE*)dictStart, dictSize);
+}
+
+LZ4_FORCE_O2
+int LZ4_decompress_safe_partial_forceExtDict(const char* source, char* dest,
+                                     int compressedSize, int targetOutputSize, int dstCapacity,
+                                     const void* dictStart, size_t dictSize)
+{
+    dstCapacity = MIN(targetOutputSize, dstCapacity);
+    return LZ4_decompress_generic(source, dest, compressedSize, dstCapacity,
+                                  partial_decode, usingExtDict,
+                                  (BYTE*)dest, (const BYTE*)dictStart, dictSize);
+}
+
+LZ4_FORCE_O2
+static int LZ4_decompress_fast_extDict(const char* source, char* dest, int originalSize,
+                                       const void* dictStart, size_t dictSize)
+{
+    return LZ4_decompress_unsafe_generic(
+                (const BYTE*)source, (BYTE*)dest, originalSize,
+                0, (const BYTE*)dictStart, dictSize);
+}
+
+/* The "double dictionary" mode, for use with e.g. ring buffers: the first part
+ * of the dictionary is passed as prefix, and the second via dictStart + dictSize.
+ * These routines are used only once, in LZ4_decompress_*_continue().
+ */
+LZ4_FORCE_INLINE
+int LZ4_decompress_safe_doubleDict(const char* source, char* dest, int compressedSize, int maxOutputSize,
+                                   size_t prefixSize, const void* dictStart, size_t dictSize)
+{
+    return LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize,
+                                  decode_full_block, usingExtDict,
+                                  (BYTE*)dest-prefixSize, (const BYTE*)dictStart, dictSize);
+}
+
+/*===== streaming decompression functions =====*/
+
+#if !defined(LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION)
+LZ4_streamDecode_t* LZ4_createStreamDecode(void)
+{
+    LZ4_STATIC_ASSERT(sizeof(LZ4_streamDecode_t) >= sizeof(LZ4_streamDecode_t_internal));
+    return (LZ4_streamDecode_t*) ALLOC_AND_ZERO(sizeof(LZ4_streamDecode_t));
+}
+
+int LZ4_freeStreamDecode (LZ4_streamDecode_t* LZ4_stream)
+{
+    if (LZ4_stream == NULL) { return 0; }  /* support free on NULL */
+    FREEMEM(LZ4_stream);
+    return 0;
+}
+#endif
+
+/*! LZ4_setStreamDecode() :
+ *  Use this function to instruct where to find the dictionary.
+ *  This function is not necessary if previous data is still available where it was decoded.
+ *  Loading a size of 0 is allowed (same effect as no dictionary).
+ * @return : 1 if OK, 0 if error
+ */
+int LZ4_setStreamDecode (LZ4_streamDecode_t* LZ4_streamDecode, const char* dictionary, int dictSize)
+{
+    LZ4_streamDecode_t_internal* lz4sd = &LZ4_streamDecode->internal_donotuse;
+    lz4sd->prefixSize = (size_t)dictSize;
+    if (dictSize) {
+        assert(dictionary != NULL);
+        lz4sd->prefixEnd = (const BYTE*) dictionary + dictSize;
+    } else {
+        lz4sd->prefixEnd = (const BYTE*) dictionary;
+    }
+    lz4sd->externalDict = NULL;
+    lz4sd->extDictSize  = 0;
+    return 1;
+}
+
+/*! LZ4_decoderRingBufferSize() :
+ *  when setting a ring buffer for streaming decompression (optional scenario),
+ *  provides the minimum size of this ring buffer
+ *  to be compatible with any source respecting maxBlockSize condition.
+ *  Note : in a ring buffer scenario,
+ *  blocks are presumed decompressed next to each other.
+ *  When not enough space remains for next block (remainingSize < maxBlockSize),
+ *  decoding resumes from beginning of ring buffer.
+ * @return : minimum ring buffer size,
+ *           or 0 if there is an error (invalid maxBlockSize).
+ */
+int LZ4_decoderRingBufferSize(int maxBlockSize)
+{
+    if (maxBlockSize < 0) return 0;
+    if (maxBlockSize > LZ4_MAX_INPUT_SIZE) return 0;
+    if (maxBlockSize < 16) maxBlockSize = 16;
+    return LZ4_DECODER_RING_BUFFER_SIZE(maxBlockSize);
+}
+
+/*
+*_continue() :
+    These decoding functions allow decompression of multiple blocks in "streaming" mode.
+    Previously decoded blocks must still be available at the memory position where they were decoded.
+    If it's not possible, save the relevant part of decoded data into a safe buffer,
+    and indicate where it stands using LZ4_setStreamDecode()
+*/
+LZ4_FORCE_O2
+int LZ4_decompress_safe_continue (LZ4_streamDecode_t* LZ4_streamDecode, const char* source, char* dest, int compressedSize, int maxOutputSize)
+{
+    LZ4_streamDecode_t_internal* lz4sd = &LZ4_streamDecode->internal_donotuse;
+    int result;
+
+    if (lz4sd->prefixSize == 0) {
+        /* The first call, no dictionary yet. */
+        assert(lz4sd->extDictSize == 0);
+        result = LZ4_decompress_safe(source, dest, compressedSize, maxOutputSize);
+        if (result <= 0) return result;
+        lz4sd->prefixSize = (size_t)result;
+        lz4sd->prefixEnd = (BYTE*)dest + result;
+    } else if (lz4sd->prefixEnd == (BYTE*)dest) {
+        /* They're rolling the current segment. */
+        if (lz4sd->prefixSize >= 64 KB - 1)
+            result = LZ4_decompress_safe_withPrefix64k(source, dest, compressedSize, maxOutputSize);
+        else if (lz4sd->extDictSize == 0)
+            result = LZ4_decompress_safe_withSmallPrefix(source, dest, compressedSize, maxOutputSize,
+                                                         lz4sd->prefixSize);
+        else
+            result = LZ4_decompress_safe_doubleDict(source, dest, compressedSize, maxOutputSize,
+                                                    lz4sd->prefixSize, lz4sd->externalDict, lz4sd->extDictSize);
+        if (result <= 0) return result;
+        lz4sd->prefixSize += (size_t)result;
+        lz4sd->prefixEnd  += result;
+    } else {
+        /* The buffer wraps around, or they're switching to another buffer. */
+        lz4sd->extDictSize = lz4sd->prefixSize;
+        lz4sd->externalDict = lz4sd->prefixEnd - lz4sd->extDictSize;
+        result = LZ4_decompress_safe_forceExtDict(source, dest, compressedSize, maxOutputSize,
+                                                  lz4sd->externalDict, lz4sd->extDictSize);
+        if (result <= 0) return result;
+        lz4sd->prefixSize = (size_t)result;
+        lz4sd->prefixEnd  = (BYTE*)dest + result;
+    }
+
+    return result;
+}
+
+LZ4_FORCE_O2 int
+LZ4_decompress_fast_continue (LZ4_streamDecode_t* LZ4_streamDecode,
+                        const char* source, char* dest, int originalSize)
+{
+    LZ4_streamDecode_t_internal* const lz4sd =
+        (assert(LZ4_streamDecode!=NULL), &LZ4_streamDecode->internal_donotuse);
+    int result;
+
+    DEBUGLOG(5, "LZ4_decompress_fast_continue (toDecodeSize=%i)", originalSize);
+    assert(originalSize >= 0);
+
+    if (lz4sd->prefixSize == 0) {
+        DEBUGLOG(5, "first invocation : no prefix nor extDict");
+        assert(lz4sd->extDictSize == 0);
+        result = LZ4_decompress_fast(source, dest, originalSize);
+        if (result <= 0) return result;
+        lz4sd->prefixSize = (size_t)originalSize;
+        lz4sd->prefixEnd = (BYTE*)dest + originalSize;
+    } else if (lz4sd->prefixEnd == (BYTE*)dest) {
+        DEBUGLOG(5, "continue using existing prefix");
+        result = LZ4_decompress_unsafe_generic(
+                        (const BYTE*)source, (BYTE*)dest, originalSize,
+                        lz4sd->prefixSize,
+                        lz4sd->externalDict, lz4sd->extDictSize);
+        if (result <= 0) return result;
+        lz4sd->prefixSize += (size_t)originalSize;
+        lz4sd->prefixEnd  += originalSize;
+    } else {
+        DEBUGLOG(5, "prefix becomes extDict");
+        lz4sd->extDictSize = lz4sd->prefixSize;
+        lz4sd->externalDict = lz4sd->prefixEnd - lz4sd->extDictSize;
+        result = LZ4_decompress_fast_extDict(source, dest, originalSize,
+                                             lz4sd->externalDict, lz4sd->extDictSize);
+        if (result <= 0) return result;
+        lz4sd->prefixSize = (size_t)originalSize;
+        lz4sd->prefixEnd  = (BYTE*)dest + originalSize;
+    }
+
+    return result;
+}
+
+
+/*
+Advanced decoding functions :
+*_usingDict() :
+    These decoding functions work the same as "_continue" ones,
+    the dictionary must be explicitly provided within parameters
+*/
+
+int LZ4_decompress_safe_usingDict(const char* source, char* dest, int compressedSize, int maxOutputSize, const char* dictStart, int dictSize)
+{
+    if (dictSize==0)
+        return LZ4_decompress_safe(source, dest, compressedSize, maxOutputSize);
+    if (dictStart+dictSize == dest) {
+        if (dictSize >= 64 KB - 1) {
+            return LZ4_decompress_safe_withPrefix64k(source, dest, compressedSize, maxOutputSize);
+        }
+        assert(dictSize >= 0);
+        return LZ4_decompress_safe_withSmallPrefix(source, dest, compressedSize, maxOutputSize, (size_t)dictSize);
+    }
+    assert(dictSize >= 0);
+    return LZ4_decompress_safe_forceExtDict(source, dest, compressedSize, maxOutputSize, dictStart, (size_t)dictSize);
+}
+
+int LZ4_decompress_safe_partial_usingDict(const char* source, char* dest, int compressedSize, int targetOutputSize, int dstCapacity, const char* dictStart, int dictSize)
+{
+    if (dictSize==0)
+        return LZ4_decompress_safe_partial(source, dest, compressedSize, targetOutputSize, dstCapacity);
+    if (dictStart+dictSize == dest) {
+        if (dictSize >= 64 KB - 1) {
+            return LZ4_decompress_safe_partial_withPrefix64k(source, dest, compressedSize, targetOutputSize, dstCapacity);
+        }
+        assert(dictSize >= 0);
+        return LZ4_decompress_safe_partial_withSmallPrefix(source, dest, compressedSize, targetOutputSize, dstCapacity, (size_t)dictSize);
+    }
+    assert(dictSize >= 0);
+    return LZ4_decompress_safe_partial_forceExtDict(source, dest, compressedSize, targetOutputSize, dstCapacity, dictStart, (size_t)dictSize);
+}
+
+int LZ4_decompress_fast_usingDict(const char* source, char* dest, int originalSize, const char* dictStart, int dictSize)
+{
+    if (dictSize==0 || dictStart+dictSize == dest)
+        return LZ4_decompress_unsafe_generic(
+                        (const BYTE*)source, (BYTE*)dest, originalSize,
+                        (size_t)dictSize, NULL, 0);
+    assert(dictSize >= 0);
+    return LZ4_decompress_fast_extDict(source, dest, originalSize, dictStart, (size_t)dictSize);
+}
+
+
+/*=*************************************************
+*  Obsolete Functions
+***************************************************/
+/* obsolete compression functions */
+int LZ4_compress_limitedOutput(const char* source, char* dest, int inputSize, int maxOutputSize)
+{
+    return LZ4_compress_default(source, dest, inputSize, maxOutputSize);
+}
+int LZ4_compress(const char* src, char* dest, int srcSize)
+{
+    return LZ4_compress_default(src, dest, srcSize, LZ4_compressBound(srcSize));
+}
+int LZ4_compress_limitedOutput_withState (void* state, const char* src, char* dst, int srcSize, int dstSize)
+{
+    return LZ4_compress_fast_extState(state, src, dst, srcSize, dstSize, 1);
+}
+int LZ4_compress_withState (void* state, const char* src, char* dst, int srcSize)
+{
+    return LZ4_compress_fast_extState(state, src, dst, srcSize, LZ4_compressBound(srcSize), 1);
+}
+int LZ4_compress_limitedOutput_continue (LZ4_stream_t* LZ4_stream, const char* src, char* dst, int srcSize, int dstCapacity)
+{
+    return LZ4_compress_fast_continue(LZ4_stream, src, dst, srcSize, dstCapacity, 1);
+}
+int LZ4_compress_continue (LZ4_stream_t* LZ4_stream, const char* source, char* dest, int inputSize)
+{
+    return LZ4_compress_fast_continue(LZ4_stream, source, dest, inputSize, LZ4_compressBound(inputSize), 1);
+}
+
+/*
+These decompression functions are deprecated and should no longer be used.
+They are only provided here for compatibility with older user programs.
+- LZ4_uncompress is totally equivalent to LZ4_decompress_fast
+- LZ4_uncompress_unknownOutputSize is totally equivalent to LZ4_decompress_safe
+*/
+int LZ4_uncompress (const char* source, char* dest, int outputSize)
+{
+    return LZ4_decompress_fast(source, dest, outputSize);
+}
+int LZ4_uncompress_unknownOutputSize (const char* source, char* dest, int isize, int maxOutputSize)
+{
+    return LZ4_decompress_safe(source, dest, isize, maxOutputSize);
+}
+
+/* Obsolete Streaming functions */
+
+int LZ4_sizeofStreamState(void) { return sizeof(LZ4_stream_t); }
+
+int LZ4_resetStreamState(void* state, char* inputBuffer)
+{
+    (void)inputBuffer;
+    LZ4_resetStream((LZ4_stream_t*)state);
+    return 0;
+}
+
+#if !defined(LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION)
+void* LZ4_create (char* inputBuffer)
+{
+    (void)inputBuffer;
+    return LZ4_createStream();
+}
+#endif
+
+char* LZ4_slideInputBuffer (void* state)
+{
+    /* avoid const char * -> char * conversion warning */
+    return (char *)(uptrval)((LZ4_stream_t*)state)->internal_donotuse.dictionary;
+}
+
+#endif   /* LZ4_COMMONDEFS_ONLY */
diff --git a/lib/lz4/upstream/lz4.h b/lib/lz4/upstream/lz4.h
new file mode 100644
index 0000000000000000000000000000000000000000..80e3e5ca04d22db9ec4980dfc43672a13984aa15
--- /dev/null
+++ b/lib/lz4/upstream/lz4.h
@@ -0,0 +1,884 @@
+/*
+ *  LZ4 - Fast LZ compression algorithm
+ *  Header File
+ *  Copyright (C) 2011-2023, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+    - LZ4 homepage : http://www.lz4.org
+    - LZ4 source repository : https://github.com/lz4/lz4
+*/
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#ifndef LZ4_H_2983827168210
+#define LZ4_H_2983827168210
+
+/* --- Dependency --- */
+#include <stddef.h>   /* size_t */
+
+
+/**
+  Introduction
+
+  LZ4 is lossless compression algorithm, providing compression speed >500 MB/s per core,
+  scalable with multi-cores CPU. It features an extremely fast decoder, with speed in
+  multiple GB/s per core, typically reaching RAM speed limits on multi-core systems.
+
+  The LZ4 compression library provides in-memory compression and decompression functions.
+  It gives full buffer control to user.
+  Compression can be done in:
+    - a single step (described as Simple Functions)
+    - a single step, reusing a context (described in Advanced Functions)
+    - unbounded multiple steps (described as Streaming compression)
+
+  lz4.h generates and decodes LZ4-compressed blocks (doc/lz4_Block_format.md).
+  Decompressing such a compressed block requires additional metadata.
+  Exact metadata depends on exact decompression function.
+  For the typical case of LZ4_decompress_safe(),
+  metadata includes block's compressed size, and maximum bound of decompressed size.
+  Each application is free to encode and pass such metadata in whichever way it wants.
+
+  lz4.h only handle blocks, it can not generate Frames.
+
+  Blocks are different from Frames (doc/lz4_Frame_format.md).
+  Frames bundle both blocks and metadata in a specified manner.
+  Embedding metadata is required for compressed data to be self-contained and portable.
+  Frame format is delivered through a companion API, declared in lz4frame.h.
+  The `lz4` CLI can only manage frames.
+*/
+
+/*^***************************************************************
+*  Export parameters
+*****************************************************************/
+/*
+*  LZ4_DLL_EXPORT :
+*  Enable exporting of functions when building a Windows DLL
+*  LZ4LIB_VISIBILITY :
+*  Control library symbols visibility.
+*/
+#ifndef LZ4LIB_VISIBILITY
+#  if defined(__GNUC__) && (__GNUC__ >= 4)
+#    define LZ4LIB_VISIBILITY __attribute__ ((visibility ("default")))
+#  else
+#    define LZ4LIB_VISIBILITY
+#  endif
+#endif
+#if defined(LZ4_DLL_EXPORT) && (LZ4_DLL_EXPORT==1)
+#  define LZ4LIB_API __declspec(dllexport) LZ4LIB_VISIBILITY
+#elif defined(LZ4_DLL_IMPORT) && (LZ4_DLL_IMPORT==1)
+#  define LZ4LIB_API __declspec(dllimport) LZ4LIB_VISIBILITY /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
+#else
+#  define LZ4LIB_API LZ4LIB_VISIBILITY
+#endif
+
+/*! LZ4_FREESTANDING :
+ *  When this macro is set to 1, it enables "freestanding mode" that is
+ *  suitable for typical freestanding environment which doesn't support
+ *  standard C library.
+ *
+ *  - LZ4_FREESTANDING is a compile-time switch.
+ *  - It requires the following macros to be defined:
+ *    LZ4_memcpy, LZ4_memmove, LZ4_memset.
+ *  - It only enables LZ4/HC functions which don't use heap.
+ *    All LZ4F_* functions are not supported.
+ *  - See tests/freestanding.c to check its basic setup.
+ */
+#if defined(LZ4_FREESTANDING) && (LZ4_FREESTANDING == 1)
+#  define LZ4_HEAPMODE 0
+#  define LZ4HC_HEAPMODE 0
+#  define LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION 1
+#  if !defined(LZ4_memcpy)
+#    error "LZ4_FREESTANDING requires macro 'LZ4_memcpy'."
+#  endif
+#  if !defined(LZ4_memset)
+#    error "LZ4_FREESTANDING requires macro 'LZ4_memset'."
+#  endif
+#  if !defined(LZ4_memmove)
+#    error "LZ4_FREESTANDING requires macro 'LZ4_memmove'."
+#  endif
+#elif ! defined(LZ4_FREESTANDING)
+#  define LZ4_FREESTANDING 0
+#endif
+
+
+/*------   Version   ------*/
+#define LZ4_VERSION_MAJOR    1    /* for breaking interface changes  */
+#define LZ4_VERSION_MINOR   10    /* for new (non-breaking) interface capabilities */
+#define LZ4_VERSION_RELEASE  0    /* for tweaks, bug-fixes, or development */
+
+#define LZ4_VERSION_NUMBER (LZ4_VERSION_MAJOR *100*100 + LZ4_VERSION_MINOR *100 + LZ4_VERSION_RELEASE)
+
+#define LZ4_LIB_VERSION LZ4_VERSION_MAJOR.LZ4_VERSION_MINOR.LZ4_VERSION_RELEASE
+#define LZ4_QUOTE(str) #str
+#define LZ4_EXPAND_AND_QUOTE(str) LZ4_QUOTE(str)
+#define LZ4_VERSION_STRING LZ4_EXPAND_AND_QUOTE(LZ4_LIB_VERSION)  /* requires v1.7.3+ */
+
+LZ4LIB_API int LZ4_versionNumber (void);  /**< library version number; useful to check dll version; requires v1.3.0+ */
+LZ4LIB_API const char* LZ4_versionString (void);   /**< library version string; useful to check dll version; requires v1.7.5+ */
+
+
+/*-************************************
+*  Tuning memory usage
+**************************************/
+/*!
+ * LZ4_MEMORY_USAGE :
+ * Can be selected at compile time, by setting LZ4_MEMORY_USAGE.
+ * Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB)
+ * Increasing memory usage improves compression ratio, generally at the cost of speed.
+ * Reduced memory usage may improve speed at the cost of ratio, thanks to better cache locality.
+ * Default value is 14, for 16KB, which nicely fits into most L1 caches.
+ */
+#ifndef LZ4_MEMORY_USAGE
+# define LZ4_MEMORY_USAGE LZ4_MEMORY_USAGE_DEFAULT
+#endif
+
+/* These are absolute limits, they should not be changed by users */
+#define LZ4_MEMORY_USAGE_MIN 10
+#define LZ4_MEMORY_USAGE_DEFAULT 14
+#define LZ4_MEMORY_USAGE_MAX 20
+
+#if (LZ4_MEMORY_USAGE < LZ4_MEMORY_USAGE_MIN)
+#  error "LZ4_MEMORY_USAGE is too small !"
+#endif
+
+#if (LZ4_MEMORY_USAGE > LZ4_MEMORY_USAGE_MAX)
+#  error "LZ4_MEMORY_USAGE is too large !"
+#endif
+
+/*-************************************
+*  Simple Functions
+**************************************/
+/*! LZ4_compress_default() :
+ *  Compresses 'srcSize' bytes from buffer 'src'
+ *  into already allocated 'dst' buffer of size 'dstCapacity'.
+ *  Compression is guaranteed to succeed if 'dstCapacity' >= LZ4_compressBound(srcSize).
+ *  It also runs faster, so it's a recommended setting.
+ *  If the function cannot compress 'src' into a more limited 'dst' budget,
+ *  compression stops *immediately*, and the function result is zero.
+ *  In which case, 'dst' content is undefined (invalid).
+ *      srcSize : max supported value is LZ4_MAX_INPUT_SIZE.
+ *      dstCapacity : size of buffer 'dst' (which must be already allocated)
+ *     @return  : the number of bytes written into buffer 'dst' (necessarily <= dstCapacity)
+ *                or 0 if compression fails
+ * Note : This function is protected against buffer overflow scenarios (never writes outside 'dst' buffer, nor read outside 'source' buffer).
+ */
+LZ4LIB_API int LZ4_compress_default(const char* src, char* dst, int srcSize, int dstCapacity);
+
+/*! LZ4_decompress_safe() :
+ * @compressedSize : is the exact complete size of the compressed block.
+ * @dstCapacity : is the size of destination buffer (which must be already allocated),
+ *                presumed an upper bound of decompressed size.
+ * @return : the number of bytes decompressed into destination buffer (necessarily <= dstCapacity)
+ *           If destination buffer is not large enough, decoding will stop and output an error code (negative value).
+ *           If the source stream is detected malformed, the function will stop decoding and return a negative result.
+ * Note 1 : This function is protected against malicious data packets :
+ *          it will never writes outside 'dst' buffer, nor read outside 'source' buffer,
+ *          even if the compressed block is maliciously modified to order the decoder to do these actions.
+ *          In such case, the decoder stops immediately, and considers the compressed block malformed.
+ * Note 2 : compressedSize and dstCapacity must be provided to the function, the compressed block does not contain them.
+ *          The implementation is free to send / store / derive this information in whichever way is most beneficial.
+ *          If there is a need for a different format which bundles together both compressed data and its metadata, consider looking at lz4frame.h instead.
+ */
+LZ4LIB_API int LZ4_decompress_safe (const char* src, char* dst, int compressedSize, int dstCapacity);
+
+
+/*-************************************
+*  Advanced Functions
+**************************************/
+#define LZ4_MAX_INPUT_SIZE        0x7E000000   /* 2 113 929 216 bytes */
+#define LZ4_COMPRESSBOUND(isize)  ((unsigned)(isize) > (unsigned)LZ4_MAX_INPUT_SIZE ? 0 : (isize) + ((isize)/255) + 16)
+
+/*! LZ4_compressBound() :
+    Provides the maximum size that LZ4 compression may output in a "worst case" scenario (input data not compressible)
+    This function is primarily useful for memory allocation purposes (destination buffer size).
+    Macro LZ4_COMPRESSBOUND() is also provided for compilation-time evaluation (stack memory allocation for example).
+    Note that LZ4_compress_default() compresses faster when dstCapacity is >= LZ4_compressBound(srcSize)
+        inputSize  : max supported value is LZ4_MAX_INPUT_SIZE
+        return : maximum output size in a "worst case" scenario
+              or 0, if input size is incorrect (too large or negative)
+*/
+LZ4LIB_API int LZ4_compressBound(int inputSize);
+
+/*! LZ4_compress_fast() :
+    Same as LZ4_compress_default(), but allows selection of "acceleration" factor.
+    The larger the acceleration value, the faster the algorithm, but also the lesser the compression.
+    It's a trade-off. It can be fine tuned, with each successive value providing roughly +~3% to speed.
+    An acceleration value of "1" is the same as regular LZ4_compress_default()
+    Values <= 0 will be replaced by LZ4_ACCELERATION_DEFAULT (currently == 1, see lz4.c).
+    Values > LZ4_ACCELERATION_MAX will be replaced by LZ4_ACCELERATION_MAX (currently == 65537, see lz4.c).
+*/
+LZ4LIB_API int LZ4_compress_fast (const char* src, char* dst, int srcSize, int dstCapacity, int acceleration);
+
+
+/*! LZ4_compress_fast_extState() :
+ *  Same as LZ4_compress_fast(), using an externally allocated memory space for its state.
+ *  Use LZ4_sizeofState() to know how much memory must be allocated,
+ *  and allocate it on 8-bytes boundaries (using `malloc()` typically).
+ *  Then, provide this buffer as `void* state` to compression function.
+ */
+LZ4LIB_API int LZ4_sizeofState(void);
+LZ4LIB_API int LZ4_compress_fast_extState (void* state, const char* src, char* dst, int srcSize, int dstCapacity, int acceleration);
+
+/*! LZ4_compress_destSize() :
+ *  Reverse the logic : compresses as much data as possible from 'src' buffer
+ *  into already allocated buffer 'dst', of size >= 'dstCapacity'.
+ *  This function either compresses the entire 'src' content into 'dst' if it's large enough,
+ *  or fill 'dst' buffer completely with as much data as possible from 'src'.
+ *  note: acceleration parameter is fixed to "default".
+ *
+ * *srcSizePtr : in+out parameter. Initially contains size of input.
+ *               Will be modified to indicate how many bytes where read from 'src' to fill 'dst'.
+ *               New value is necessarily <= input value.
+ * @return : Nb bytes written into 'dst' (necessarily <= dstCapacity)
+ *           or 0 if compression fails.
+ *
+ * Note : from v1.8.2 to v1.9.1, this function had a bug (fixed in v1.9.2+):
+ *        the produced compressed content could, in specific circumstances,
+ *        require to be decompressed into a destination buffer larger
+ *        by at least 1 byte than the content to decompress.
+ *        If an application uses `LZ4_compress_destSize()`,
+ *        it's highly recommended to update liblz4 to v1.9.2 or better.
+ *        If this can't be done or ensured,
+ *        the receiving decompression function should provide
+ *        a dstCapacity which is > decompressedSize, by at least 1 byte.
+ *        See https://github.com/lz4/lz4/issues/859 for details
+ */
+LZ4LIB_API int LZ4_compress_destSize(const char* src, char* dst, int* srcSizePtr, int targetDstSize);
+
+/*! LZ4_decompress_safe_partial() :
+ *  Decompress an LZ4 compressed block, of size 'srcSize' at position 'src',
+ *  into destination buffer 'dst' of size 'dstCapacity'.
+ *  Up to 'targetOutputSize' bytes will be decoded.
+ *  The function stops decoding on reaching this objective.
+ *  This can be useful to boost performance
+ *  whenever only the beginning of a block is required.
+ *
+ * @return : the number of bytes decoded in `dst` (necessarily <= targetOutputSize)
+ *           If source stream is detected malformed, function returns a negative result.
+ *
+ *  Note 1 : @return can be < targetOutputSize, if compressed block contains less data.
+ *
+ *  Note 2 : targetOutputSize must be <= dstCapacity
+ *
+ *  Note 3 : this function effectively stops decoding on reaching targetOutputSize,
+ *           so dstCapacity is kind of redundant.
+ *           This is because in older versions of this function,
+ *           decoding operation would still write complete sequences.
+ *           Therefore, there was no guarantee that it would stop writing at exactly targetOutputSize,
+ *           it could write more bytes, though only up to dstCapacity.
+ *           Some "margin" used to be required for this operation to work properly.
+ *           Thankfully, this is no longer necessary.
+ *           The function nonetheless keeps the same signature, in an effort to preserve API compatibility.
+ *
+ *  Note 4 : If srcSize is the exact size of the block,
+ *           then targetOutputSize can be any value,
+ *           including larger than the block's decompressed size.
+ *           The function will, at most, generate block's decompressed size.
+ *
+ *  Note 5 : If srcSize is _larger_ than block's compressed size,
+ *           then targetOutputSize **MUST** be <= block's decompressed size.
+ *           Otherwise, *silent corruption will occur*.
+ */
+LZ4LIB_API int LZ4_decompress_safe_partial (const char* src, char* dst, int srcSize, int targetOutputSize, int dstCapacity);
+
+
+/*-*********************************************
+*  Streaming Compression Functions
+***********************************************/
+typedef union LZ4_stream_u LZ4_stream_t;  /* incomplete type (defined later) */
+
+/*!
+ Note about RC_INVOKED
+
+ - RC_INVOKED is predefined symbol of rc.exe (the resource compiler which is part of MSVC/Visual Studio).
+   https://docs.microsoft.com/en-us/windows/win32/menurc/predefined-macros
+
+ - Since rc.exe is a legacy compiler, it truncates long symbol (> 30 chars)
+   and reports warning "RC4011: identifier truncated".
+
+ - To eliminate the warning, we surround long preprocessor symbol with
+   "#if !defined(RC_INVOKED) ... #endif" block that means
+   "skip this block when rc.exe is trying to read it".
+*/
+#if !defined(RC_INVOKED) /* https://docs.microsoft.com/en-us/windows/win32/menurc/predefined-macros */
+#if !defined(LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION)
+LZ4LIB_API LZ4_stream_t* LZ4_createStream(void);
+LZ4LIB_API int           LZ4_freeStream (LZ4_stream_t* streamPtr);
+#endif /* !defined(LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION) */
+#endif
+
+/*! LZ4_resetStream_fast() : v1.9.0+
+ *  Use this to prepare an LZ4_stream_t for a new chain of dependent blocks
+ *  (e.g., LZ4_compress_fast_continue()).
+ *
+ *  An LZ4_stream_t must be initialized once before usage.
+ *  This is automatically done when created by LZ4_createStream().
+ *  However, should the LZ4_stream_t be simply declared on stack (for example),
+ *  it's necessary to initialize it first, using LZ4_initStream().
+ *
+ *  After init, start any new stream with LZ4_resetStream_fast().
+ *  A same LZ4_stream_t can be re-used multiple times consecutively
+ *  and compress multiple streams,
+ *  provided that it starts each new stream with LZ4_resetStream_fast().
+ *
+ *  LZ4_resetStream_fast() is much faster than LZ4_initStream(),
+ *  but is not compatible with memory regions containing garbage data.
+ *
+ *  Note: it's only useful to call LZ4_resetStream_fast()
+ *        in the context of streaming compression.
+ *        The *extState* functions perform their own resets.
+ *        Invoking LZ4_resetStream_fast() before is redundant, and even counterproductive.
+ */
+LZ4LIB_API void LZ4_resetStream_fast (LZ4_stream_t* streamPtr);
+
+/*! LZ4_loadDict() :
+ *  Use this function to reference a static dictionary into LZ4_stream_t.
+ *  The dictionary must remain available during compression.
+ *  LZ4_loadDict() triggers a reset, so any previous data will be forgotten.
+ *  The same dictionary will have to be loaded on decompression side for successful decoding.
+ *  Dictionary are useful for better compression of small data (KB range).
+ *  While LZ4 itself accepts any input as dictionary, dictionary efficiency is also a topic.
+ *  When in doubt, employ the Zstandard's Dictionary Builder.
+ *  Loading a size of 0 is allowed, and is the same as reset.
+ * @return : loaded dictionary size, in bytes (note: only the last 64 KB are loaded)
+ */
+LZ4LIB_API int LZ4_loadDict (LZ4_stream_t* streamPtr, const char* dictionary, int dictSize);
+
+/*! LZ4_loadDictSlow() : v1.10.0+
+ *  Same as LZ4_loadDict(),
+ *  but uses a bit more cpu to reference the dictionary content more thoroughly.
+ *  This is expected to slightly improve compression ratio.
+ *  The extra-cpu cost is likely worth it if the dictionary is re-used across multiple sessions.
+ * @return : loaded dictionary size, in bytes (note: only the last 64 KB are loaded)
+ */
+LZ4LIB_API int LZ4_loadDictSlow(LZ4_stream_t* streamPtr, const char* dictionary, int dictSize);
+
+/*! LZ4_attach_dictionary() : stable since v1.10.0
+ *
+ *  This allows efficient re-use of a static dictionary multiple times.
+ *
+ *  Rather than re-loading the dictionary buffer into a working context before
+ *  each compression, or copying a pre-loaded dictionary's LZ4_stream_t into a
+ *  working LZ4_stream_t, this function introduces a no-copy setup mechanism,
+ *  in which the working stream references @dictionaryStream in-place.
+ *
+ *  Several assumptions are made about the state of @dictionaryStream.
+ *  Currently, only states which have been prepared by LZ4_loadDict() or
+ *  LZ4_loadDictSlow() should be expected to work.
+ *
+ *  Alternatively, the provided @dictionaryStream may be NULL,
+ *  in which case any existing dictionary stream is unset.
+ *
+ *  If a dictionary is provided, it replaces any pre-existing stream history.
+ *  The dictionary contents are the only history that can be referenced and
+ *  logically immediately precede the data compressed in the first subsequent
+ *  compression call.
+ *
+ *  The dictionary will only remain attached to the working stream through the
+ *  first compression call, at the end of which it is cleared.
+ * @dictionaryStream stream (and source buffer) must remain in-place / accessible / unchanged
+ *  through the completion of the compression session.
+ *
+ *  Note: there is no equivalent LZ4_attach_*() method on the decompression side
+ *  because there is no initialization cost, hence no need to share the cost across multiple sessions.
+ *  To decompress LZ4 blocks using dictionary, attached or not,
+ *  just employ the regular LZ4_setStreamDecode() for streaming,
+ *  or the stateless LZ4_decompress_safe_usingDict() for one-shot decompression.
+ */
+LZ4LIB_API void
+LZ4_attach_dictionary(LZ4_stream_t* workingStream,
+                const LZ4_stream_t* dictionaryStream);
+
+/*! LZ4_compress_fast_continue() :
+ *  Compress 'src' content using data from previously compressed blocks, for better compression ratio.
+ * 'dst' buffer must be already allocated.
+ *  If dstCapacity >= LZ4_compressBound(srcSize), compression is guaranteed to succeed, and runs faster.
+ *
+ * @return : size of compressed block
+ *           or 0 if there is an error (typically, cannot fit into 'dst').
+ *
+ *  Note 1 : Each invocation to LZ4_compress_fast_continue() generates a new block.
+ *           Each block has precise boundaries.
+ *           Each block must be decompressed separately, calling LZ4_decompress_*() with relevant metadata.
+ *           It's not possible to append blocks together and expect a single invocation of LZ4_decompress_*() to decompress them together.
+ *
+ *  Note 2 : The previous 64KB of source data is __assumed__ to remain present, unmodified, at same address in memory !
+ *
+ *  Note 3 : When input is structured as a double-buffer, each buffer can have any size, including < 64 KB.
+ *           Make sure that buffers are separated, by at least one byte.
+ *           This construction ensures that each block only depends on previous block.
+ *
+ *  Note 4 : If input buffer is a ring-buffer, it can have any size, including < 64 KB.
+ *
+ *  Note 5 : After an error, the stream status is undefined (invalid), it can only be reset or freed.
+ */
+LZ4LIB_API int LZ4_compress_fast_continue (LZ4_stream_t* streamPtr, const char* src, char* dst, int srcSize, int dstCapacity, int acceleration);
+
+/*! LZ4_saveDict() :
+ *  If last 64KB data cannot be guaranteed to remain available at its current memory location,
+ *  save it into a safer place (char* safeBuffer).
+ *  This is schematically equivalent to a memcpy() followed by LZ4_loadDict(),
+ *  but is much faster, because LZ4_saveDict() doesn't need to rebuild tables.
+ * @return : saved dictionary size in bytes (necessarily <= maxDictSize), or 0 if error.
+ */
+LZ4LIB_API int LZ4_saveDict (LZ4_stream_t* streamPtr, char* safeBuffer, int maxDictSize);
+
+
+/*-**********************************************
+*  Streaming Decompression Functions
+*  Bufferless synchronous API
+************************************************/
+typedef union LZ4_streamDecode_u LZ4_streamDecode_t;   /* tracking context */
+
+/*! LZ4_createStreamDecode() and LZ4_freeStreamDecode() :
+ *  creation / destruction of streaming decompression tracking context.
+ *  A tracking context can be re-used multiple times.
+ */
+#if !defined(RC_INVOKED) /* https://docs.microsoft.com/en-us/windows/win32/menurc/predefined-macros */
+#if !defined(LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION)
+LZ4LIB_API LZ4_streamDecode_t* LZ4_createStreamDecode(void);
+LZ4LIB_API int                 LZ4_freeStreamDecode (LZ4_streamDecode_t* LZ4_stream);
+#endif /* !defined(LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION) */
+#endif
+
+/*! LZ4_setStreamDecode() :
+ *  An LZ4_streamDecode_t context can be allocated once and re-used multiple times.
+ *  Use this function to start decompression of a new stream of blocks.
+ *  A dictionary can optionally be set. Use NULL or size 0 for a reset order.
+ *  Dictionary is presumed stable : it must remain accessible and unmodified during next decompression.
+ * @return : 1 if OK, 0 if error
+ */
+LZ4LIB_API int LZ4_setStreamDecode (LZ4_streamDecode_t* LZ4_streamDecode, const char* dictionary, int dictSize);
+
+/*! LZ4_decoderRingBufferSize() : v1.8.2+
+ *  Note : in a ring buffer scenario (optional),
+ *  blocks are presumed decompressed next to each other
+ *  up to the moment there is not enough remaining space for next block (remainingSize < maxBlockSize),
+ *  at which stage it resumes from beginning of ring buffer.
+ *  When setting such a ring buffer for streaming decompression,
+ *  provides the minimum size of this ring buffer
+ *  to be compatible with any source respecting maxBlockSize condition.
+ * @return : minimum ring buffer size,
+ *           or 0 if there is an error (invalid maxBlockSize).
+ */
+LZ4LIB_API int LZ4_decoderRingBufferSize(int maxBlockSize);
+#define LZ4_DECODER_RING_BUFFER_SIZE(maxBlockSize) (65536 + 14 + (maxBlockSize))  /* for static allocation; maxBlockSize presumed valid */
+
+/*! LZ4_decompress_safe_continue() :
+ *  This decoding function allows decompression of consecutive blocks in "streaming" mode.
+ *  The difference with the usual independent blocks is that
+ *  new blocks are allowed to find references into former blocks.
+ *  A block is an unsplittable entity, and must be presented entirely to the decompression function.
+ *  LZ4_decompress_safe_continue() only accepts one block at a time.
+ *  It's modeled after `LZ4_decompress_safe()` and behaves similarly.
+ *
+ * @LZ4_streamDecode : decompression state, tracking the position in memory of past data
+ * @compressedSize : exact complete size of one compressed block.
+ * @dstCapacity : size of destination buffer (which must be already allocated),
+ *                must be an upper bound of decompressed size.
+ * @return : number of bytes decompressed into destination buffer (necessarily <= dstCapacity)
+ *           If destination buffer is not large enough, decoding will stop and output an error code (negative value).
+ *           If the source stream is detected malformed, the function will stop decoding and return a negative result.
+ *
+ *  The last 64KB of previously decoded data *must* remain available and unmodified
+ *  at the memory position where they were previously decoded.
+ *  If less than 64KB of data has been decoded, all the data must be present.
+ *
+ *  Special : if decompression side sets a ring buffer, it must respect one of the following conditions :
+ *  - Decompression buffer size is _at least_ LZ4_decoderRingBufferSize(maxBlockSize).
+ *    maxBlockSize is the maximum size of any single block. It can have any value > 16 bytes.
+ *    In which case, encoding and decoding buffers do not need to be synchronized.
+ *    Actually, data can be produced by any source compliant with LZ4 format specification, and respecting maxBlockSize.
+ *  - Synchronized mode :
+ *    Decompression buffer size is _exactly_ the same as compression buffer size,
+ *    and follows exactly same update rule (block boundaries at same positions),
+ *    and decoding function is provided with exact decompressed size of each block (exception for last block of the stream),
+ *    _then_ decoding & encoding ring buffer can have any size, including small ones ( < 64 KB).
+ *  - Decompression buffer is larger than encoding buffer, by a minimum of maxBlockSize more bytes.
+ *    In which case, encoding and decoding buffers do not need to be synchronized,
+ *    and encoding ring buffer can have any size, including small ones ( < 64 KB).
+ *
+ *  Whenever these conditions are not possible,
+ *  save the last 64KB of decoded data into a safe buffer where it can't be modified during decompression,
+ *  then indicate where this data is saved using LZ4_setStreamDecode(), before decompressing next block.
+*/
+LZ4LIB_API int
+LZ4_decompress_safe_continue (LZ4_streamDecode_t* LZ4_streamDecode,
+                        const char* src, char* dst,
+                        int srcSize, int dstCapacity);
+
+
+/*! LZ4_decompress_safe_usingDict() :
+ *  Works the same as
+ *  a combination of LZ4_setStreamDecode() followed by LZ4_decompress_safe_continue()
+ *  However, it's stateless: it doesn't need any LZ4_streamDecode_t state.
+ *  Dictionary is presumed stable : it must remain accessible and unmodified during decompression.
+ *  Performance tip : Decompression speed can be substantially increased
+ *                    when dst == dictStart + dictSize.
+ */
+LZ4LIB_API int
+LZ4_decompress_safe_usingDict(const char* src, char* dst,
+                              int srcSize, int dstCapacity,
+                              const char* dictStart, int dictSize);
+
+/*! LZ4_decompress_safe_partial_usingDict() :
+ *  Behaves the same as LZ4_decompress_safe_partial()
+ *  with the added ability to specify a memory segment for past data.
+ *  Performance tip : Decompression speed can be substantially increased
+ *                    when dst == dictStart + dictSize.
+ */
+LZ4LIB_API int
+LZ4_decompress_safe_partial_usingDict(const char* src, char* dst,
+                                      int compressedSize,
+                                      int targetOutputSize, int maxOutputSize,
+                                      const char* dictStart, int dictSize);
+
+#endif /* LZ4_H_2983827168210 */
+
+
+/*^*************************************
+ * !!!!!!   STATIC LINKING ONLY   !!!!!!
+ ***************************************/
+
+/*-****************************************************************************
+ * Experimental section
+ *
+ * Symbols declared in this section must be considered unstable. Their
+ * signatures or semantics may change, or they may be removed altogether in the
+ * future. They are therefore only safe to depend on when the caller is
+ * statically linked against the library.
+ *
+ * To protect against unsafe usage, not only are the declarations guarded,
+ * the definitions are hidden by default
+ * when building LZ4 as a shared/dynamic library.
+ *
+ * In order to access these declarations,
+ * define LZ4_STATIC_LINKING_ONLY in your application
+ * before including LZ4's headers.
+ *
+ * In order to make their implementations accessible dynamically, you must
+ * define LZ4_PUBLISH_STATIC_FUNCTIONS when building the LZ4 library.
+ ******************************************************************************/
+
+#ifdef LZ4_STATIC_LINKING_ONLY
+
+#ifndef LZ4_STATIC_3504398509
+#define LZ4_STATIC_3504398509
+
+#ifdef LZ4_PUBLISH_STATIC_FUNCTIONS
+# define LZ4LIB_STATIC_API LZ4LIB_API
+#else
+# define LZ4LIB_STATIC_API
+#endif
+
+
+/*! LZ4_compress_fast_extState_fastReset() :
+ *  A variant of LZ4_compress_fast_extState().
+ *
+ *  Using this variant avoids an expensive initialization step.
+ *  It is only safe to call if the state buffer is known to be correctly initialized already
+ *  (see above comment on LZ4_resetStream_fast() for a definition of "correctly initialized").
+ *  From a high level, the difference is that
+ *  this function initializes the provided state with a call to something like LZ4_resetStream_fast()
+ *  while LZ4_compress_fast_extState() starts with a call to LZ4_resetStream().
+ */
+LZ4LIB_STATIC_API int LZ4_compress_fast_extState_fastReset (void* state, const char* src, char* dst, int srcSize, int dstCapacity, int acceleration);
+
+/*! LZ4_compress_destSize_extState() : introduced in v1.10.0
+ *  Same as LZ4_compress_destSize(), but using an externally allocated state.
+ *  Also: exposes @acceleration
+ */
+int LZ4_compress_destSize_extState(void* state, const char* src, char* dst, int* srcSizePtr, int targetDstSize, int acceleration);
+
+/*! In-place compression and decompression
+ *
+ * It's possible to have input and output sharing the same buffer,
+ * for highly constrained memory environments.
+ * In both cases, it requires input to lay at the end of the buffer,
+ * and decompression to start at beginning of the buffer.
+ * Buffer size must feature some margin, hence be larger than final size.
+ *
+ * |<------------------------buffer--------------------------------->|
+ *                             |<-----------compressed data--------->|
+ * |<-----------decompressed size------------------>|
+ *                                                  |<----margin---->|
+ *
+ * This technique is more useful for decompression,
+ * since decompressed size is typically larger,
+ * and margin is short.
+ *
+ * In-place decompression will work inside any buffer
+ * which size is >= LZ4_DECOMPRESS_INPLACE_BUFFER_SIZE(decompressedSize).
+ * This presumes that decompressedSize > compressedSize.
+ * Otherwise, it means compression actually expanded data,
+ * and it would be more efficient to store such data with a flag indicating it's not compressed.
+ * This can happen when data is not compressible (already compressed, or encrypted).
+ *
+ * For in-place compression, margin is larger, as it must be able to cope with both
+ * history preservation, requiring input data to remain unmodified up to LZ4_DISTANCE_MAX,
+ * and data expansion, which can happen when input is not compressible.
+ * As a consequence, buffer size requirements are much higher,
+ * and memory savings offered by in-place compression are more limited.
+ *
+ * There are ways to limit this cost for compression :
+ * - Reduce history size, by modifying LZ4_DISTANCE_MAX.
+ *   Note that it is a compile-time constant, so all compressions will apply this limit.
+ *   Lower values will reduce compression ratio, except when input_size < LZ4_DISTANCE_MAX,
+ *   so it's a reasonable trick when inputs are known to be small.
+ * - Require the compressor to deliver a "maximum compressed size".
+ *   This is the `dstCapacity` parameter in `LZ4_compress*()`.
+ *   When this size is < LZ4_COMPRESSBOUND(inputSize), then compression can fail,
+ *   in which case, the return code will be 0 (zero).
+ *   The caller must be ready for these cases to happen,
+ *   and typically design a backup scheme to send data uncompressed.
+ * The combination of both techniques can significantly reduce
+ * the amount of margin required for in-place compression.
+ *
+ * In-place compression can work in any buffer
+ * which size is >= (maxCompressedSize)
+ * with maxCompressedSize == LZ4_COMPRESSBOUND(srcSize) for guaranteed compression success.
+ * LZ4_COMPRESS_INPLACE_BUFFER_SIZE() depends on both maxCompressedSize and LZ4_DISTANCE_MAX,
+ * so it's possible to reduce memory requirements by playing with them.
+ */
+
+#define LZ4_DECOMPRESS_INPLACE_MARGIN(compressedSize)          (((compressedSize) >> 8) + 32)
+#define LZ4_DECOMPRESS_INPLACE_BUFFER_SIZE(decompressedSize)   ((decompressedSize) + LZ4_DECOMPRESS_INPLACE_MARGIN(decompressedSize))  /**< note: presumes that compressedSize < decompressedSize. note2: margin is overestimated a bit, since it could use compressedSize instead */
+
+#ifndef LZ4_DISTANCE_MAX   /* history window size; can be user-defined at compile time */
+#  define LZ4_DISTANCE_MAX 65535   /* set to maximum value by default */
+#endif
+
+#define LZ4_COMPRESS_INPLACE_MARGIN                           (LZ4_DISTANCE_MAX + 32)   /* LZ4_DISTANCE_MAX can be safely replaced by srcSize when it's smaller */
+#define LZ4_COMPRESS_INPLACE_BUFFER_SIZE(maxCompressedSize)   ((maxCompressedSize) + LZ4_COMPRESS_INPLACE_MARGIN)  /**< maxCompressedSize is generally LZ4_COMPRESSBOUND(inputSize), but can be set to any lower value, with the risk that compression can fail (return code 0(zero)) */
+
+#endif   /* LZ4_STATIC_3504398509 */
+#endif   /* LZ4_STATIC_LINKING_ONLY */
+
+
+
+#ifndef LZ4_H_98237428734687
+#define LZ4_H_98237428734687
+
+/*-************************************************************
+ *  Private Definitions
+ **************************************************************
+ * Do not use these definitions directly.
+ * They are only exposed to allow static allocation of `LZ4_stream_t` and `LZ4_streamDecode_t`.
+ * Accessing members will expose user code to API and/or ABI break in future versions of the library.
+ **************************************************************/
+#define LZ4_HASHLOG   (LZ4_MEMORY_USAGE-2)
+#define LZ4_HASHTABLESIZE (1 << LZ4_MEMORY_USAGE)
+#define LZ4_HASH_SIZE_U32 (1 << LZ4_HASHLOG)       /* required as macro for static allocation */
+
+#if defined(__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+# include <stdint.h>
+  typedef  int8_t  LZ4_i8;
+  typedef uint8_t  LZ4_byte;
+  typedef uint16_t LZ4_u16;
+  typedef uint32_t LZ4_u32;
+#else
+  typedef   signed char  LZ4_i8;
+  typedef unsigned char  LZ4_byte;
+  typedef unsigned short LZ4_u16;
+  typedef unsigned int   LZ4_u32;
+#endif
+
+/*! LZ4_stream_t :
+ *  Never ever use below internal definitions directly !
+ *  These definitions are not API/ABI safe, and may change in future versions.
+ *  If you need static allocation, declare or allocate an LZ4_stream_t object.
+**/
+
+typedef struct LZ4_stream_t_internal LZ4_stream_t_internal;
+struct LZ4_stream_t_internal {
+    LZ4_u32 hashTable[LZ4_HASH_SIZE_U32];
+    const LZ4_byte* dictionary;
+    const LZ4_stream_t_internal* dictCtx;
+    LZ4_u32 currentOffset;
+    LZ4_u32 tableType;
+    LZ4_u32 dictSize;
+    /* Implicit padding to ensure structure is aligned */
+};
+
+#define LZ4_STREAM_MINSIZE  ((1UL << (LZ4_MEMORY_USAGE)) + 32)  /* static size, for inter-version compatibility */
+union LZ4_stream_u {
+    char minStateSize[LZ4_STREAM_MINSIZE];
+    LZ4_stream_t_internal internal_donotuse;
+}; /* previously typedef'd to LZ4_stream_t */
+
+
+/*! LZ4_initStream() : v1.9.0+
+ *  An LZ4_stream_t structure must be initialized at least once.
+ *  This is automatically done when invoking LZ4_createStream(),
+ *  but it's not when the structure is simply declared on stack (for example).
+ *
+ *  Use LZ4_initStream() to properly initialize a newly declared LZ4_stream_t.
+ *  It can also initialize any arbitrary buffer of sufficient size,
+ *  and will @return a pointer of proper type upon initialization.
+ *
+ *  Note : initialization fails if size and alignment conditions are not respected.
+ *         In which case, the function will @return NULL.
+ *  Note2: An LZ4_stream_t structure guarantees correct alignment and size.
+ *  Note3: Before v1.9.0, use LZ4_resetStream() instead
+**/
+LZ4LIB_API LZ4_stream_t* LZ4_initStream (void* stateBuffer, size_t size);
+
+
+/*! LZ4_streamDecode_t :
+ *  Never ever use below internal definitions directly !
+ *  These definitions are not API/ABI safe, and may change in future versions.
+ *  If you need static allocation, declare or allocate an LZ4_streamDecode_t object.
+**/
+typedef struct {
+    const LZ4_byte* externalDict;
+    const LZ4_byte* prefixEnd;
+    size_t extDictSize;
+    size_t prefixSize;
+} LZ4_streamDecode_t_internal;
+
+#define LZ4_STREAMDECODE_MINSIZE 32
+union LZ4_streamDecode_u {
+    char minStateSize[LZ4_STREAMDECODE_MINSIZE];
+    LZ4_streamDecode_t_internal internal_donotuse;
+} ;   /* previously typedef'd to LZ4_streamDecode_t */
+
+
+
+/*-************************************
+*  Obsolete Functions
+**************************************/
+
+/*! Deprecation warnings
+ *
+ *  Deprecated functions make the compiler generate a warning when invoked.
+ *  This is meant to invite users to update their source code.
+ *  Should deprecation warnings be a problem, it is generally possible to disable them,
+ *  typically with -Wno-deprecated-declarations for gcc
+ *  or _CRT_SECURE_NO_WARNINGS in Visual.
+ *
+ *  Another method is to define LZ4_DISABLE_DEPRECATE_WARNINGS
+ *  before including the header file.
+ */
+#ifdef LZ4_DISABLE_DEPRECATE_WARNINGS
+#  define LZ4_DEPRECATED(message)   /* disable deprecation warnings */
+#else
+#  if defined (__cplusplus) && (__cplusplus >= 201402) /* C++14 or greater */
+#    define LZ4_DEPRECATED(message) [[deprecated(message)]]
+#  elif defined(_MSC_VER)
+#    define LZ4_DEPRECATED(message) __declspec(deprecated(message))
+#  elif defined(__clang__) || (defined(__GNUC__) && (__GNUC__ * 10 + __GNUC_MINOR__ >= 45))
+#    define LZ4_DEPRECATED(message) __attribute__((deprecated(message)))
+#  elif defined(__GNUC__) && (__GNUC__ * 10 + __GNUC_MINOR__ >= 31)
+#    define LZ4_DEPRECATED(message) __attribute__((deprecated))
+#  else
+#    pragma message("WARNING: LZ4_DEPRECATED needs custom implementation for this compiler")
+#    define LZ4_DEPRECATED(message)   /* disabled */
+#  endif
+#endif /* LZ4_DISABLE_DEPRECATE_WARNINGS */
+
+/*! Obsolete compression functions (since v1.7.3) */
+LZ4_DEPRECATED("use LZ4_compress_default() instead")       LZ4LIB_API int LZ4_compress               (const char* src, char* dest, int srcSize);
+LZ4_DEPRECATED("use LZ4_compress_default() instead")       LZ4LIB_API int LZ4_compress_limitedOutput (const char* src, char* dest, int srcSize, int maxOutputSize);
+LZ4_DEPRECATED("use LZ4_compress_fast_extState() instead") LZ4LIB_API int LZ4_compress_withState               (void* state, const char* source, char* dest, int inputSize);
+LZ4_DEPRECATED("use LZ4_compress_fast_extState() instead") LZ4LIB_API int LZ4_compress_limitedOutput_withState (void* state, const char* source, char* dest, int inputSize, int maxOutputSize);
+LZ4_DEPRECATED("use LZ4_compress_fast_continue() instead") LZ4LIB_API int LZ4_compress_continue                (LZ4_stream_t* LZ4_streamPtr, const char* source, char* dest, int inputSize);
+LZ4_DEPRECATED("use LZ4_compress_fast_continue() instead") LZ4LIB_API int LZ4_compress_limitedOutput_continue  (LZ4_stream_t* LZ4_streamPtr, const char* source, char* dest, int inputSize, int maxOutputSize);
+
+/*! Obsolete decompression functions (since v1.8.0) */
+LZ4_DEPRECATED("use LZ4_decompress_fast() instead") LZ4LIB_API int LZ4_uncompress (const char* source, char* dest, int outputSize);
+LZ4_DEPRECATED("use LZ4_decompress_safe() instead") LZ4LIB_API int LZ4_uncompress_unknownOutputSize (const char* source, char* dest, int isize, int maxOutputSize);
+
+/* Obsolete streaming functions (since v1.7.0)
+ * degraded functionality; do not use!
+ *
+ * In order to perform streaming compression, these functions depended on data
+ * that is no longer tracked in the state. They have been preserved as well as
+ * possible: using them will still produce a correct output. However, they don't
+ * actually retain any history between compression calls. The compression ratio
+ * achieved will therefore be no better than compressing each chunk
+ * independently.
+ */
+LZ4_DEPRECATED("Use LZ4_createStream() instead") LZ4LIB_API void* LZ4_create (char* inputBuffer);
+LZ4_DEPRECATED("Use LZ4_createStream() instead") LZ4LIB_API int   LZ4_sizeofStreamState(void);
+LZ4_DEPRECATED("Use LZ4_resetStream() instead")  LZ4LIB_API int   LZ4_resetStreamState(void* state, char* inputBuffer);
+LZ4_DEPRECATED("Use LZ4_saveDict() instead")     LZ4LIB_API char* LZ4_slideInputBuffer (void* state);
+
+/*! Obsolete streaming decoding functions (since v1.7.0) */
+LZ4_DEPRECATED("use LZ4_decompress_safe_usingDict() instead") LZ4LIB_API int LZ4_decompress_safe_withPrefix64k (const char* src, char* dst, int compressedSize, int maxDstSize);
+LZ4_DEPRECATED("use LZ4_decompress_fast_usingDict() instead") LZ4LIB_API int LZ4_decompress_fast_withPrefix64k (const char* src, char* dst, int originalSize);
+
+/*! Obsolete LZ4_decompress_fast variants (since v1.9.0) :
+ *  These functions used to be faster than LZ4_decompress_safe(),
+ *  but this is no longer the case. They are now slower.
+ *  This is because LZ4_decompress_fast() doesn't know the input size,
+ *  and therefore must progress more cautiously into the input buffer to not read beyond the end of block.
+ *  On top of that `LZ4_decompress_fast()` is not protected vs malformed or malicious inputs, making it a security liability.
+ *  As a consequence, LZ4_decompress_fast() is strongly discouraged, and deprecated.
+ *
+ *  The last remaining LZ4_decompress_fast() specificity is that
+ *  it can decompress a block without knowing its compressed size.
+ *  Such functionality can be achieved in a more secure manner
+ *  by employing LZ4_decompress_safe_partial().
+ *
+ *  Parameters:
+ *  originalSize : is the uncompressed size to regenerate.
+ *                 `dst` must be already allocated, its size must be >= 'originalSize' bytes.
+ * @return : number of bytes read from source buffer (== compressed size).
+ *           The function expects to finish at block's end exactly.
+ *           If the source stream is detected malformed, the function stops decoding and returns a negative result.
+ *  note : LZ4_decompress_fast*() requires originalSize. Thanks to this information, it never writes past the output buffer.
+ *         However, since it doesn't know its 'src' size, it may read an unknown amount of input, past input buffer bounds.
+ *         Also, since match offsets are not validated, match reads from 'src' may underflow too.
+ *         These issues never happen if input (compressed) data is correct.
+ *         But they may happen if input data is invalid (error or intentional tampering).
+ *         As a consequence, use these functions in trusted environments with trusted data **only**.
+ */
+LZ4_DEPRECATED("This function is deprecated and unsafe. Consider using LZ4_decompress_safe_partial() instead")
+LZ4LIB_API int LZ4_decompress_fast (const char* src, char* dst, int originalSize);
+LZ4_DEPRECATED("This function is deprecated and unsafe. Consider migrating towards LZ4_decompress_safe_continue() instead. "
+               "Note that the contract will change (requires block's compressed size, instead of decompressed size)")
+LZ4LIB_API int LZ4_decompress_fast_continue (LZ4_streamDecode_t* LZ4_streamDecode, const char* src, char* dst, int originalSize);
+LZ4_DEPRECATED("This function is deprecated and unsafe. Consider using LZ4_decompress_safe_partial_usingDict() instead")
+LZ4LIB_API int LZ4_decompress_fast_usingDict (const char* src, char* dst, int originalSize, const char* dictStart, int dictSize);
+
+/*! LZ4_resetStream() :
+ *  An LZ4_stream_t structure must be initialized at least once.
+ *  This is done with LZ4_initStream(), or LZ4_resetStream().
+ *  Consider switching to LZ4_initStream(),
+ *  invoking LZ4_resetStream() will trigger deprecation warnings in the future.
+ */
+LZ4LIB_API void LZ4_resetStream (LZ4_stream_t* streamPtr);
+
+
+#endif /* LZ4_H_98237428734687 */
+
+
+#if defined (__cplusplus)
+}
+#endif
diff --git a/lib/lz4/upstream/lz4hc.c b/lib/lz4/upstream/lz4hc.c
new file mode 100644
index 0000000000000000000000000000000000000000..4d8c36a6978fcae09e4c4a936572b4b271513466
--- /dev/null
+++ b/lib/lz4/upstream/lz4hc.c
@@ -0,0 +1,2192 @@
+/*
+    LZ4 HC - High Compression Mode of LZ4
+    Copyright (C) 2011-2020, Yann Collet.
+
+    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+       - LZ4 source repository : https://github.com/lz4/lz4
+       - LZ4 public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+/* note : lz4hc is not an independent module, it requires lz4.h/lz4.c for proper compilation */
+
+
+/* *************************************
+*  Tuning Parameter
+***************************************/
+
+/*! HEAPMODE :
+ *  Select how stateless HC compression functions like `LZ4_compress_HC()`
+ *  allocate memory for their workspace:
+ *  in stack (0:fastest), or in heap (1:default, requires malloc()).
+ *  Since workspace is rather large, heap mode is recommended.
+**/
+#ifndef LZ4HC_HEAPMODE
+#  define LZ4HC_HEAPMODE 1
+#endif
+
+
+/*===    Dependency    ===*/
+#define LZ4_HC_STATIC_LINKING_ONLY
+#include "lz4hc.h"
+#include <limits.h>
+
+
+/*===   Shared lz4.c code   ===*/
+#ifndef LZ4_SRC_INCLUDED
+# if defined(__GNUC__)
+#  pragma GCC diagnostic ignored "-Wunused-function"
+# endif
+# if defined (__clang__)
+#  pragma clang diagnostic ignored "-Wunused-function"
+# endif
+# define LZ4_COMMONDEFS_ONLY
+# include "lz4.c"   /* LZ4_count, constants, mem */
+#endif
+
+
+/*===   Enums   ===*/
+typedef enum { noDictCtx, usingDictCtxHc } dictCtx_directive;
+
+
+/*===   Constants   ===*/
+#define OPTIMAL_ML (int)((ML_MASK-1)+MINMATCH)
+#define LZ4_OPT_NUM   (1<<12)
+
+
+/*===   Macros   ===*/
+#define MIN(a,b)   ( (a) < (b) ? (a) : (b) )
+#define MAX(a,b)   ( (a) > (b) ? (a) : (b) )
+
+
+/*===   Levels definition   ===*/
+typedef enum { lz4mid, lz4hc, lz4opt } lz4hc_strat_e;
+typedef struct {
+    lz4hc_strat_e strat;
+    int nbSearches;
+    U32 targetLength;
+} cParams_t;
+static const cParams_t k_clTable[LZ4HC_CLEVEL_MAX+1] = {
+    { lz4mid,    2, 16 },  /* 0, unused */
+    { lz4mid,    2, 16 },  /* 1, unused */
+    { lz4mid,    2, 16 },  /* 2 */
+    { lz4hc,     4, 16 },  /* 3 */
+    { lz4hc,     8, 16 },  /* 4 */
+    { lz4hc,    16, 16 },  /* 5 */
+    { lz4hc,    32, 16 },  /* 6 */
+    { lz4hc,    64, 16 },  /* 7 */
+    { lz4hc,   128, 16 },  /* 8 */
+    { lz4hc,   256, 16 },  /* 9 */
+    { lz4opt,   96, 64 },  /*10==LZ4HC_CLEVEL_OPT_MIN*/
+    { lz4opt,  512,128 },  /*11 */
+    { lz4opt,16384,LZ4_OPT_NUM },  /* 12==LZ4HC_CLEVEL_MAX */
+};
+
+static cParams_t LZ4HC_getCLevelParams(int cLevel)
+{
+    /* note : clevel convention is a bit different from lz4frame,
+     * possibly something worth revisiting for consistency */
+    if (cLevel < 1)
+        cLevel = LZ4HC_CLEVEL_DEFAULT;
+    cLevel = MIN(LZ4HC_CLEVEL_MAX, cLevel);
+    return k_clTable[cLevel];
+}
+
+
+/*===   Hashing   ===*/
+#define LZ4HC_HASHSIZE 4
+#define HASH_FUNCTION(i)      (((i) * 2654435761U) >> ((MINMATCH*8)-LZ4HC_HASH_LOG))
+static U32 LZ4HC_hashPtr(const void* ptr) { return HASH_FUNCTION(LZ4_read32(ptr)); }
+
+#if defined(LZ4_FORCE_MEMORY_ACCESS) && (LZ4_FORCE_MEMORY_ACCESS==2)
+/* lie to the compiler about data alignment; use with caution */
+static U64 LZ4_read64(const void* memPtr) { return *(const U64*) memPtr; }
+
+#elif defined(LZ4_FORCE_MEMORY_ACCESS) && (LZ4_FORCE_MEMORY_ACCESS==1)
+/* __pack instructions are safer, but compiler specific */
+LZ4_PACK(typedef struct { U64 u64; }) LZ4_unalign64;
+static U64 LZ4_read64(const void* ptr) { return ((const LZ4_unalign64*)ptr)->u64; }
+
+#else  /* safe and portable access using memcpy() */
+static U64 LZ4_read64(const void* memPtr)
+{
+    U64 val; LZ4_memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+#endif /* LZ4_FORCE_MEMORY_ACCESS */
+
+#define LZ4MID_HASHSIZE 8
+#define LZ4MID_HASHLOG (LZ4HC_HASH_LOG-1)
+#define LZ4MID_HASHTABLESIZE (1 << LZ4MID_HASHLOG)
+
+static U32 LZ4MID_hash4(U32 v) { return (v * 2654435761U) >> (32-LZ4MID_HASHLOG); }
+static U32 LZ4MID_hash4Ptr(const void* ptr) { return LZ4MID_hash4(LZ4_read32(ptr)); }
+/* note: hash7 hashes the lower 56-bits.
+ * It presumes input was read using little endian.*/
+static U32 LZ4MID_hash7(U64 v) { return (U32)(((v  << (64-56)) * 58295818150454627ULL) >> (64-LZ4MID_HASHLOG)) ; }
+static U64 LZ4_readLE64(const void* memPtr);
+static U32 LZ4MID_hash8Ptr(const void* ptr) { return LZ4MID_hash7(LZ4_readLE64(ptr)); }
+
+static U64 LZ4_readLE64(const void* memPtr)
+{
+    if (LZ4_isLittleEndian()) {
+        return LZ4_read64(memPtr);
+    } else {
+        const BYTE* p = (const BYTE*)memPtr;
+        /* note: relies on the compiler to simplify this expression */
+        return (U64)p[0] | ((U64)p[1]<<8) | ((U64)p[2]<<16) | ((U64)p[3]<<24)
+            | ((U64)p[4]<<32) | ((U64)p[5]<<40) | ((U64)p[6]<<48) | ((U64)p[7]<<56);
+    }
+}
+
+
+/*===   Count match length   ===*/
+LZ4_FORCE_INLINE
+unsigned LZ4HC_NbCommonBytes32(U32 val)
+{
+    assert(val != 0);
+    if (LZ4_isLittleEndian()) {
+#     if defined(_MSC_VER) && (_MSC_VER >= 1400) && !defined(LZ4_FORCE_SW_BITCOUNT)
+        unsigned long r;
+        _BitScanReverse(&r, val);
+        return (unsigned)((31 - r) >> 3);
+#     elif (defined(__clang__) || (defined(__GNUC__) && ((__GNUC__ > 3) || \
+                            ((__GNUC__ == 3) && (__GNUC_MINOR__ >= 4))))) && \
+                                        !defined(LZ4_FORCE_SW_BITCOUNT)
+        return (unsigned)__builtin_clz(val) >> 3;
+#     else
+        val >>= 8;
+        val = ((((val + 0x00FFFF00) | 0x00FFFFFF) + val) |
+              (val + 0x00FF0000)) >> 24;
+        return (unsigned)val ^ 3;
+#     endif
+    } else {
+#     if defined(_MSC_VER) && (_MSC_VER >= 1400) && !defined(LZ4_FORCE_SW_BITCOUNT)
+        unsigned long r;
+        _BitScanForward(&r, val);
+        return (unsigned)(r >> 3);
+#     elif (defined(__clang__) || (defined(__GNUC__) && ((__GNUC__ > 3) || \
+                            ((__GNUC__ == 3) && (__GNUC_MINOR__ >= 4))))) && \
+                                        !defined(LZ4_FORCE_SW_BITCOUNT)
+        return (unsigned)__builtin_ctz(val) >> 3;
+#     else
+        const U32 m = 0x01010101;
+        return (unsigned)((((val - 1) ^ val) & (m - 1)) * m) >> 24;
+#     endif
+    }
+}
+
+/** LZ4HC_countBack() :
+ * @return : negative value, nb of common bytes before ip/match */
+LZ4_FORCE_INLINE
+int LZ4HC_countBack(const BYTE* const ip, const BYTE* const match,
+                    const BYTE* const iMin, const BYTE* const mMin)
+{
+    int back = 0;
+    int const min = (int)MAX(iMin - ip, mMin - match);
+    assert(min <= 0);
+    assert(ip >= iMin); assert((size_t)(ip-iMin) < (1U<<31));
+    assert(match >= mMin); assert((size_t)(match - mMin) < (1U<<31));
+
+    while ((back - min) > 3) {
+        U32 const v = LZ4_read32(ip + back - 4) ^ LZ4_read32(match + back - 4);
+        if (v) {
+            return (back - (int)LZ4HC_NbCommonBytes32(v));
+        } else back -= 4; /* 4-byte step */
+    }
+    /* check remainder if any */
+    while ( (back > min)
+         && (ip[back-1] == match[back-1]) )
+            back--;
+    return back;
+}
+
+/*===   Chain table updates   ===*/
+#define DELTANEXTU16(table, pos) table[(U16)(pos)]   /* faster */
+/* Make fields passed to, and updated by LZ4HC_encodeSequence explicit */
+#define UPDATABLE(ip, op, anchor) &ip, &op, &anchor
+
+
+/**************************************
+*  Init
+**************************************/
+static void LZ4HC_clearTables (LZ4HC_CCtx_internal* hc4)
+{
+    MEM_INIT(hc4->hashTable, 0, sizeof(hc4->hashTable));
+    MEM_INIT(hc4->chainTable, 0xFF, sizeof(hc4->chainTable));
+}
+
+static void LZ4HC_init_internal (LZ4HC_CCtx_internal* hc4, const BYTE* start)
+{
+    size_t const bufferSize = (size_t)(hc4->end - hc4->prefixStart);
+    size_t newStartingOffset = bufferSize + hc4->dictLimit;
+    DEBUGLOG(5, "LZ4HC_init_internal");
+    assert(newStartingOffset >= bufferSize);  /* check overflow */
+    if (newStartingOffset > 1 GB) {
+        LZ4HC_clearTables(hc4);
+        newStartingOffset = 0;
+    }
+    newStartingOffset += 64 KB;
+    hc4->nextToUpdate = (U32)newStartingOffset;
+    hc4->prefixStart = start;
+    hc4->end = start;
+    hc4->dictStart = start;
+    hc4->dictLimit = (U32)newStartingOffset;
+    hc4->lowLimit = (U32)newStartingOffset;
+}
+
+
+/**************************************
+*  Encode
+**************************************/
+/* LZ4HC_encodeSequence() :
+ * @return : 0 if ok,
+ *           1 if buffer issue detected */
+LZ4_FORCE_INLINE int LZ4HC_encodeSequence (
+    const BYTE** _ip,
+    BYTE** _op,
+    const BYTE** _anchor,
+    int matchLength,
+    int offset,
+    limitedOutput_directive limit,
+    BYTE* oend)
+{
+#define ip      (*_ip)
+#define op      (*_op)
+#define anchor  (*_anchor)
+
+    size_t length;
+    BYTE* const token = op++;
+
+#if defined(LZ4_DEBUG) && (LZ4_DEBUG >= 6)
+    static const BYTE* start = NULL;
+    static U32 totalCost = 0;
+    U32 const pos = (start==NULL) ? 0 : (U32)(anchor - start);
+    U32 const ll = (U32)(ip - anchor);
+    U32 const llAdd = (ll>=15) ? ((ll-15) / 255) + 1 : 0;
+    U32 const mlAdd = (matchLength>=19) ? ((matchLength-19) / 255) + 1 : 0;
+    U32 const cost = 1 + llAdd + ll + 2 + mlAdd;
+    if (start==NULL) start = anchor;  /* only works for single segment */
+    /* g_debuglog_enable = (pos >= 2228) & (pos <= 2262); */
+    DEBUGLOG(6, "pos:%7u -- literals:%4u, match:%4i, offset:%5i, cost:%4u + %5u",
+                pos,
+                (U32)(ip - anchor), matchLength, offset,
+                cost, totalCost);
+    totalCost += cost;
+#endif
+
+    /* Encode Literal length */
+    length = (size_t)(ip - anchor);
+    LZ4_STATIC_ASSERT(notLimited == 0);
+    /* Check output limit */
+    if (limit && ((op + (length / 255) + length + (2 + 1 + LASTLITERALS)) > oend)) {
+        DEBUGLOG(6, "Not enough room to write %i literals (%i bytes remaining)",
+                (int)length, (int)(oend - op));
+        return 1;
+    }
+    if (length >= RUN_MASK) {
+        size_t len = length - RUN_MASK;
+        *token = (RUN_MASK << ML_BITS);
+        for(; len >= 255 ; len -= 255) *op++ = 255;
+        *op++ = (BYTE)len;
+    } else {
+        *token = (BYTE)(length << ML_BITS);
+    }
+
+    /* Copy Literals */
+    LZ4_wildCopy8(op, anchor, op + length);
+    op += length;
+
+    /* Encode Offset */
+    assert(offset <= LZ4_DISTANCE_MAX );
+    assert(offset > 0);
+    LZ4_writeLE16(op, (U16)(offset)); op += 2;
+
+    /* Encode MatchLength */
+    assert(matchLength >= MINMATCH);
+    length = (size_t)matchLength - MINMATCH;
+    if (limit && (op + (length / 255) + (1 + LASTLITERALS) > oend)) {
+        DEBUGLOG(6, "Not enough room to write match length");
+        return 1;   /* Check output limit */
+    }
+    if (length >= ML_MASK) {
+        *token += ML_MASK;
+        length -= ML_MASK;
+        for(; length >= 510 ; length -= 510) { *op++ = 255; *op++ = 255; }
+        if (length >= 255) { length -= 255; *op++ = 255; }
+        *op++ = (BYTE)length;
+    } else {
+        *token += (BYTE)(length);
+    }
+
+    /* Prepare next loop */
+    ip += matchLength;
+    anchor = ip;
+
+    return 0;
+
+#undef ip
+#undef op
+#undef anchor
+}
+
+
+typedef struct {
+    int off;
+    int len;
+    int back;  /* negative value */
+} LZ4HC_match_t;
+
+LZ4HC_match_t LZ4HC_searchExtDict(const BYTE* ip, U32 ipIndex,
+        const BYTE* const iLowLimit, const BYTE* const iHighLimit,
+        const LZ4HC_CCtx_internal* dictCtx, U32 gDictEndIndex,
+        int currentBestML, int nbAttempts)
+{
+    size_t const lDictEndIndex = (size_t)(dictCtx->end - dictCtx->prefixStart) + dictCtx->dictLimit;
+    U32 lDictMatchIndex = dictCtx->hashTable[LZ4HC_hashPtr(ip)];
+    U32 matchIndex = lDictMatchIndex + gDictEndIndex - (U32)lDictEndIndex;
+    int offset = 0, sBack = 0;
+    assert(lDictEndIndex <= 1 GB);
+    if (lDictMatchIndex>0)
+        DEBUGLOG(7, "lDictEndIndex = %zu, lDictMatchIndex = %u", lDictEndIndex, lDictMatchIndex);
+    while (ipIndex - matchIndex <= LZ4_DISTANCE_MAX && nbAttempts--) {
+        const BYTE* const matchPtr = dictCtx->prefixStart - dictCtx->dictLimit + lDictMatchIndex;
+
+        if (LZ4_read32(matchPtr) == LZ4_read32(ip)) {
+            int mlt;
+            int back = 0;
+            const BYTE* vLimit = ip + (lDictEndIndex - lDictMatchIndex);
+            if (vLimit > iHighLimit) vLimit = iHighLimit;
+            mlt = (int)LZ4_count(ip+MINMATCH, matchPtr+MINMATCH, vLimit) + MINMATCH;
+            back = (ip > iLowLimit) ? LZ4HC_countBack(ip, matchPtr, iLowLimit, dictCtx->prefixStart) : 0;
+            mlt -= back;
+            if (mlt > currentBestML) {
+                currentBestML = mlt;
+                offset = (int)(ipIndex - matchIndex);
+                sBack = back;
+                DEBUGLOG(7, "found match of length %i within extDictCtx", currentBestML);
+        }   }
+
+        {   U32 const nextOffset = DELTANEXTU16(dictCtx->chainTable, lDictMatchIndex);
+            lDictMatchIndex -= nextOffset;
+            matchIndex -= nextOffset;
+    }   }
+
+    {   LZ4HC_match_t md;
+        md.len = currentBestML;
+        md.off = offset;
+        md.back = sBack;
+        return md;
+    }
+}
+
+typedef LZ4HC_match_t (*LZ4MID_searchIntoDict_f)(const BYTE* ip, U32 ipIndex,
+        const BYTE* const iHighLimit,
+        const LZ4HC_CCtx_internal* dictCtx, U32 gDictEndIndex);
+
+static LZ4HC_match_t LZ4MID_searchHCDict(const BYTE* ip, U32 ipIndex,
+        const BYTE* const iHighLimit,
+        const LZ4HC_CCtx_internal* dictCtx, U32 gDictEndIndex)
+{
+    return LZ4HC_searchExtDict(ip,ipIndex,
+                            ip, iHighLimit,
+                            dictCtx, gDictEndIndex,
+                            MINMATCH-1, 2);
+}
+
+static LZ4HC_match_t LZ4MID_searchExtDict(const BYTE* ip, U32 ipIndex,
+        const BYTE* const iHighLimit,
+        const LZ4HC_CCtx_internal* dictCtx, U32 gDictEndIndex)
+{
+    size_t const lDictEndIndex = (size_t)(dictCtx->end - dictCtx->prefixStart) + dictCtx->dictLimit;
+    const U32* const hash4Table = dictCtx->hashTable;
+    const U32* const hash8Table = hash4Table + LZ4MID_HASHTABLESIZE;
+    DEBUGLOG(7, "LZ4MID_searchExtDict (ipIdx=%u)", ipIndex);
+
+    /* search long match first */
+    {   U32 l8DictMatchIndex = hash8Table[LZ4MID_hash8Ptr(ip)];
+        U32 m8Index = l8DictMatchIndex + gDictEndIndex - (U32)lDictEndIndex;
+        assert(lDictEndIndex <= 1 GB);
+        if (ipIndex - m8Index <= LZ4_DISTANCE_MAX) {
+            const BYTE* const matchPtr = dictCtx->prefixStart - dictCtx->dictLimit + l8DictMatchIndex;
+            const size_t safeLen = MIN(lDictEndIndex - l8DictMatchIndex, (size_t)(iHighLimit - ip));
+            int mlt = (int)LZ4_count(ip, matchPtr, ip + safeLen);
+            if (mlt >= MINMATCH) {
+                LZ4HC_match_t md;
+                DEBUGLOG(7, "Found long ExtDict match of len=%u", mlt);
+                md.len = mlt;
+                md.off = (int)(ipIndex - m8Index);
+                md.back = 0;
+                return md;
+            }
+        }
+    }
+
+    /* search for short match second */
+    {   U32 l4DictMatchIndex = hash4Table[LZ4MID_hash4Ptr(ip)];
+        U32 m4Index = l4DictMatchIndex + gDictEndIndex - (U32)lDictEndIndex;
+        if (ipIndex - m4Index <= LZ4_DISTANCE_MAX) {
+            const BYTE* const matchPtr = dictCtx->prefixStart - dictCtx->dictLimit + l4DictMatchIndex;
+            const size_t safeLen = MIN(lDictEndIndex - l4DictMatchIndex, (size_t)(iHighLimit - ip));
+            int mlt = (int)LZ4_count(ip, matchPtr, ip + safeLen);
+            if (mlt >= MINMATCH) {
+                LZ4HC_match_t md;
+                DEBUGLOG(7, "Found short ExtDict match of len=%u", mlt);
+                md.len = mlt;
+                md.off = (int)(ipIndex - m4Index);
+                md.back = 0;
+                return md;
+            }
+        }
+    }
+
+    /* nothing found */
+    {   LZ4HC_match_t const md = {0, 0, 0 };
+        return md;
+    }
+}
+
+/**************************************
+*  Mid Compression (level 2)
+**************************************/
+
+LZ4_FORCE_INLINE void
+LZ4MID_addPosition(U32* hTable, U32 hValue, U32 index)
+{
+    hTable[hValue] = index;
+}
+
+#define ADDPOS8(_p, _idx) LZ4MID_addPosition(hash8Table, LZ4MID_hash8Ptr(_p), _idx)
+#define ADDPOS4(_p, _idx) LZ4MID_addPosition(hash4Table, LZ4MID_hash4Ptr(_p), _idx)
+
+/* Fill hash tables with references into dictionary.
+ * The resulting table is only exploitable by LZ4MID (level 2) */
+static void
+LZ4MID_fillHTable (LZ4HC_CCtx_internal* cctx, const void* dict, size_t size)
+{
+    U32* const hash4Table = cctx->hashTable;
+    U32* const hash8Table = hash4Table + LZ4MID_HASHTABLESIZE;
+    const BYTE* const prefixPtr = (const BYTE*)dict;
+    U32 const prefixIdx = cctx->dictLimit;
+    U32 const target = prefixIdx + (U32)size - LZ4MID_HASHSIZE;
+    U32 idx = cctx->nextToUpdate;
+    assert(dict == cctx->prefixStart);
+    DEBUGLOG(4, "LZ4MID_fillHTable (size:%zu)", size);
+    if (size <= LZ4MID_HASHSIZE)
+        return;
+
+    for (; idx < target; idx += 3) {
+        ADDPOS4(prefixPtr+idx-prefixIdx, idx);
+        ADDPOS8(prefixPtr+idx+1-prefixIdx, idx+1);
+    }
+
+    idx = (size > 32 KB + LZ4MID_HASHSIZE) ? target - 32 KB : cctx->nextToUpdate;
+    for (; idx < target; idx += 1) {
+        ADDPOS8(prefixPtr+idx-prefixIdx, idx);
+    }
+
+    cctx->nextToUpdate = target;
+}
+
+static LZ4MID_searchIntoDict_f select_searchDict_function(const LZ4HC_CCtx_internal* dictCtx)
+{
+    if (dictCtx == NULL) return NULL;
+    if (LZ4HC_getCLevelParams(dictCtx->compressionLevel).strat == lz4mid)
+        return LZ4MID_searchExtDict;
+    return LZ4MID_searchHCDict;
+}
+
+static int LZ4MID_compress (
+    LZ4HC_CCtx_internal* const ctx,
+    const char* const src,
+    char* const dst,
+    int* srcSizePtr,
+    int const maxOutputSize,
+    const limitedOutput_directive limit,
+    const dictCtx_directive dict
+    )
+{
+    U32* const hash4Table = ctx->hashTable;
+    U32* const hash8Table = hash4Table + LZ4MID_HASHTABLESIZE;
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* anchor = ip;
+    const BYTE* const iend = ip + *srcSizePtr;
+    const BYTE* const mflimit = iend - MFLIMIT;
+    const BYTE* const matchlimit = (iend - LASTLITERALS);
+    const BYTE* const ilimit = (iend - LZ4MID_HASHSIZE);
+    BYTE* op = (BYTE*)dst;
+    BYTE* oend = op + maxOutputSize;
+
+    const BYTE* const prefixPtr = ctx->prefixStart;
+    const U32 prefixIdx = ctx->dictLimit;
+    const U32 ilimitIdx = (U32)(ilimit - prefixPtr) + prefixIdx;
+    const BYTE* const dictStart = ctx->dictStart;
+    const U32 dictIdx = ctx->lowLimit;
+    const U32 gDictEndIndex = ctx->lowLimit;
+    const LZ4MID_searchIntoDict_f searchIntoDict = (dict == usingDictCtxHc) ? select_searchDict_function(ctx->dictCtx) : NULL;
+    unsigned matchLength;
+    unsigned matchDistance;
+
+    /* input sanitization */
+    DEBUGLOG(5, "LZ4MID_compress (%i bytes)", *srcSizePtr);
+    if (dict == usingDictCtxHc) DEBUGLOG(5, "usingDictCtxHc");
+    assert(*srcSizePtr >= 0);
+    if (*srcSizePtr) assert(src != NULL);
+    if (maxOutputSize) assert(dst != NULL);
+    if (*srcSizePtr < 0) return 0;  /* invalid */
+    if (maxOutputSize < 0) return 0; /* invalid */
+    if (*srcSizePtr > LZ4_MAX_INPUT_SIZE) {
+        /* forbidden: no input is allowed to be that large */
+        return 0;
+    }
+    if (limit == fillOutput) oend -= LASTLITERALS;  /* Hack for support LZ4 format restriction */
+    if (*srcSizePtr < LZ4_minLength)
+        goto _lz4mid_last_literals;  /* Input too small, no compression (all literals) */
+
+    /* main loop */
+    while (ip <= mflimit) {
+        const U32 ipIndex = (U32)(ip - prefixPtr) + prefixIdx;
+        /* search long match */
+        {   U32 const h8 = LZ4MID_hash8Ptr(ip);
+            U32 const pos8 = hash8Table[h8];
+            assert(h8 < LZ4MID_HASHTABLESIZE);
+            assert(pos8 < ipIndex);
+            LZ4MID_addPosition(hash8Table, h8, ipIndex);
+            if (ipIndex - pos8 <= LZ4_DISTANCE_MAX) {
+                /* match candidate found */
+                if (pos8 >= prefixIdx) {
+                    const BYTE* const matchPtr = prefixPtr + pos8 - prefixIdx;
+                    assert(matchPtr < ip);
+                    matchLength = LZ4_count(ip, matchPtr, matchlimit);
+                    if (matchLength >= MINMATCH) {
+                        DEBUGLOG(7, "found long match at pos %u (len=%u)", pos8, matchLength);
+                        matchDistance = ipIndex - pos8;
+                        goto _lz4mid_encode_sequence;
+                    }
+                } else {
+                    if (pos8 >= dictIdx) {
+                        /* extDict match candidate */
+                        const BYTE* const matchPtr = dictStart + (pos8 - dictIdx);
+                        const size_t safeLen = MIN(prefixIdx - pos8, (size_t)(matchlimit - ip));
+                        matchLength = LZ4_count(ip, matchPtr, ip + safeLen);
+                        if (matchLength >= MINMATCH) {
+                            DEBUGLOG(7, "found long match at ExtDict pos %u (len=%u)", pos8, matchLength);
+                            matchDistance = ipIndex - pos8;
+                            goto _lz4mid_encode_sequence;
+                        }
+                    }
+                }
+        }   }
+        /* search short match */
+        {   U32 const h4 = LZ4MID_hash4Ptr(ip);
+            U32 const pos4 = hash4Table[h4];
+            assert(h4 < LZ4MID_HASHTABLESIZE);
+            assert(pos4 < ipIndex);
+            LZ4MID_addPosition(hash4Table, h4, ipIndex);
+            if (ipIndex - pos4 <= LZ4_DISTANCE_MAX) {
+                /* match candidate found */
+                if (pos4 >= prefixIdx) {
+                /* only search within prefix */
+                    const BYTE* const matchPtr = prefixPtr + (pos4 - prefixIdx);
+                    assert(matchPtr < ip);
+                    assert(matchPtr >= prefixPtr);
+                    matchLength = LZ4_count(ip, matchPtr, matchlimit);
+                    if (matchLength >= MINMATCH) {
+                        /* short match found, let's just check ip+1 for longer */
+                        U32 const h8 = LZ4MID_hash8Ptr(ip+1);
+                        U32 const pos8 = hash8Table[h8];
+                        U32 const m2Distance = ipIndex + 1 - pos8;
+                        matchDistance = ipIndex - pos4;
+                        if ( m2Distance <= LZ4_DISTANCE_MAX
+                        && pos8 >= prefixIdx /* only search within prefix */
+                        && likely(ip < mflimit)
+                        ) {
+                            const BYTE* const m2Ptr = prefixPtr + (pos8 - prefixIdx);
+                            unsigned ml2 = LZ4_count(ip+1, m2Ptr, matchlimit);
+                            if (ml2 > matchLength) {
+                                LZ4MID_addPosition(hash8Table, h8, ipIndex+1);
+                                ip++;
+                                matchLength = ml2;
+                                matchDistance = m2Distance;
+                        }   }
+                        goto _lz4mid_encode_sequence;
+                    }
+                } else {
+                    if (pos4 >= dictIdx) {
+                        /* extDict match candidate */
+                        const BYTE* const matchPtr = dictStart + (pos4 - dictIdx);
+                        const size_t safeLen = MIN(prefixIdx - pos4, (size_t)(matchlimit - ip));
+                        matchLength = LZ4_count(ip, matchPtr, ip + safeLen);
+                        if (matchLength >= MINMATCH) {
+                            DEBUGLOG(7, "found match at ExtDict pos %u (len=%u)", pos4, matchLength);
+                            matchDistance = ipIndex - pos4;
+                            goto _lz4mid_encode_sequence;
+                        }
+                    }
+                }
+        }   }
+        /* no match found in prefix */
+        if ( (dict == usingDictCtxHc)
+          && (ipIndex - gDictEndIndex < LZ4_DISTANCE_MAX - 8) ) {
+            /* search a match into external dictionary */
+            LZ4HC_match_t dMatch = searchIntoDict(ip, ipIndex,
+                    matchlimit,
+                    ctx->dictCtx, gDictEndIndex);
+            if (dMatch.len >= MINMATCH) {
+                DEBUGLOG(7, "found Dictionary match (offset=%i)", dMatch.off);
+                assert(dMatch.back == 0);
+                matchLength = (unsigned)dMatch.len;
+                matchDistance = (unsigned)dMatch.off;
+                goto _lz4mid_encode_sequence;
+            }
+        }
+        /* no match found */
+        ip += 1 + ((ip-anchor) >> 9);  /* skip faster over incompressible data */
+        continue;
+
+_lz4mid_encode_sequence:
+        /* catch back */
+        while (((ip > anchor) & ((U32)(ip-prefixPtr) > matchDistance)) && (unlikely(ip[-1] == ip[-(int)matchDistance-1]))) {
+            ip--;  matchLength++;
+        };
+
+        /* fill table with beginning of match */
+        ADDPOS8(ip+1, ipIndex+1);
+        ADDPOS8(ip+2, ipIndex+2);
+        ADDPOS4(ip+1, ipIndex+1);
+
+        /* encode */
+        {   BYTE* const saved_op = op;
+            /* LZ4HC_encodeSequence always updates @op; on success, it updates @ip and @anchor */
+            if (LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor),
+                    (int)matchLength, (int)matchDistance,
+                    limit, oend) ) {
+                op = saved_op;  /* restore @op value before failed LZ4HC_encodeSequence */
+                goto _lz4mid_dest_overflow;
+            }
+        }
+
+        /* fill table with end of match */
+        {   U32 endMatchIdx = (U32)(ip-prefixPtr) + prefixIdx;
+            U32 pos_m2 = endMatchIdx - 2;
+            if (pos_m2 < ilimitIdx) {
+                if (likely(ip - prefixPtr > 5)) {
+                    ADDPOS8(ip-5, endMatchIdx - 5);
+                }
+                ADDPOS8(ip-3, endMatchIdx - 3);
+                ADDPOS8(ip-2, endMatchIdx - 2);
+                ADDPOS4(ip-2, endMatchIdx - 2);
+                ADDPOS4(ip-1, endMatchIdx - 1);
+            }
+        }
+    }
+
+_lz4mid_last_literals:
+    /* Encode Last Literals */
+    {   size_t lastRunSize = (size_t)(iend - anchor);  /* literals */
+        size_t llAdd = (lastRunSize + 255 - RUN_MASK) / 255;
+        size_t const totalSize = 1 + llAdd + lastRunSize;
+        if (limit == fillOutput) oend += LASTLITERALS;  /* restore correct value */
+        if (limit && (op + totalSize > oend)) {
+            if (limit == limitedOutput) return 0;  /* not enough space in @dst */
+            /* adapt lastRunSize to fill 'dest' */
+            lastRunSize  = (size_t)(oend - op) - 1 /*token*/;
+            llAdd = (lastRunSize + 256 - RUN_MASK) / 256;
+            lastRunSize -= llAdd;
+        }
+        DEBUGLOG(6, "Final literal run : %i literals", (int)lastRunSize);
+        ip = anchor + lastRunSize;  /* can be != iend if limit==fillOutput */
+
+        if (lastRunSize >= RUN_MASK) {
+            size_t accumulator = lastRunSize - RUN_MASK;
+            *op++ = (RUN_MASK << ML_BITS);
+            for(; accumulator >= 255 ; accumulator -= 255)
+                *op++ = 255;
+            *op++ = (BYTE) accumulator;
+        } else {
+            *op++ = (BYTE)(lastRunSize << ML_BITS);
+        }
+        assert(lastRunSize <= (size_t)(oend - op));
+        LZ4_memcpy(op, anchor, lastRunSize);
+        op += lastRunSize;
+    }
+
+    /* End */
+    DEBUGLOG(5, "compressed %i bytes into %i bytes", *srcSizePtr, (int)((char*)op - dst));
+    assert(ip >= (const BYTE*)src);
+    assert(ip <= iend);
+    *srcSizePtr = (int)(ip - (const BYTE*)src);
+    assert((char*)op >= dst);
+    assert(op <= oend);
+    assert((char*)op - dst < INT_MAX);
+    return (int)((char*)op - dst);
+
+_lz4mid_dest_overflow:
+    if (limit == fillOutput) {
+        /* Assumption : @ip, @anchor, @optr and @matchLength must be set correctly */
+        size_t const ll = (size_t)(ip - anchor);
+        size_t const ll_addbytes = (ll + 240) / 255;
+        size_t const ll_totalCost = 1 + ll_addbytes + ll;
+        BYTE* const maxLitPos = oend - 3; /* 2 for offset, 1 for token */
+        DEBUGLOG(6, "Last sequence is overflowing : %u literals, %u remaining space",
+                (unsigned)ll, (unsigned)(oend-op));
+        if (op + ll_totalCost <= maxLitPos) {
+            /* ll validated; now adjust match length */
+            size_t const bytesLeftForMl = (size_t)(maxLitPos - (op+ll_totalCost));
+            size_t const maxMlSize = MINMATCH + (ML_MASK-1) + (bytesLeftForMl * 255);
+            assert(maxMlSize < INT_MAX);
+            if ((size_t)matchLength > maxMlSize) matchLength= (unsigned)maxMlSize;
+            if ((oend + LASTLITERALS) - (op + ll_totalCost + 2) - 1 + matchLength >= MFLIMIT) {
+            DEBUGLOG(6, "Let's encode a last sequence (ll=%u, ml=%u)", (unsigned)ll, matchLength);
+                LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor),
+                        (int)matchLength, (int)matchDistance,
+                        notLimited, oend);
+        }   }
+        DEBUGLOG(6, "Let's finish with a run of literals (%u bytes left)", (unsigned)(oend-op));
+        goto _lz4mid_last_literals;
+    }
+    /* compression failed */
+    return 0;
+}
+
+
+/**************************************
+*  HC Compression - Search
+**************************************/
+
+/* Update chains up to ip (excluded) */
+LZ4_FORCE_INLINE void LZ4HC_Insert (LZ4HC_CCtx_internal* hc4, const BYTE* ip)
+{
+    U16* const chainTable = hc4->chainTable;
+    U32* const hashTable  = hc4->hashTable;
+    const BYTE* const prefixPtr = hc4->prefixStart;
+    U32 const prefixIdx = hc4->dictLimit;
+    U32 const target = (U32)(ip - prefixPtr) + prefixIdx;
+    U32 idx = hc4->nextToUpdate;
+    assert(ip >= prefixPtr);
+    assert(target >= prefixIdx);
+
+    while (idx < target) {
+        U32 const h = LZ4HC_hashPtr(prefixPtr+idx-prefixIdx);
+        size_t delta = idx - hashTable[h];
+        if (delta>LZ4_DISTANCE_MAX) delta = LZ4_DISTANCE_MAX;
+        DELTANEXTU16(chainTable, idx) = (U16)delta;
+        hashTable[h] = idx;
+        idx++;
+    }
+
+    hc4->nextToUpdate = target;
+}
+
+#if defined(_MSC_VER)
+#  define LZ4HC_rotl32(x,r) _rotl(x,r)
+#else
+#  define LZ4HC_rotl32(x,r) ((x << r) | (x >> (32 - r)))
+#endif
+
+
+static U32 LZ4HC_rotatePattern(size_t const rotate, U32 const pattern)
+{
+    size_t const bitsToRotate = (rotate & (sizeof(pattern) - 1)) << 3;
+    if (bitsToRotate == 0) return pattern;
+    return LZ4HC_rotl32(pattern, (int)bitsToRotate);
+}
+
+/* LZ4HC_countPattern() :
+ * pattern32 must be a sample of repetitive pattern of length 1, 2 or 4 (but not 3!) */
+static unsigned
+LZ4HC_countPattern(const BYTE* ip, const BYTE* const iEnd, U32 const pattern32)
+{
+    const BYTE* const iStart = ip;
+    reg_t const pattern = (sizeof(pattern)==8) ?
+        (reg_t)pattern32 + (((reg_t)pattern32) << (sizeof(pattern)*4)) : pattern32;
+
+    while (likely(ip < iEnd-(sizeof(pattern)-1))) {
+        reg_t const diff = LZ4_read_ARCH(ip) ^ pattern;
+        if (!diff) { ip+=sizeof(pattern); continue; }
+        ip += LZ4_NbCommonBytes(diff);
+        return (unsigned)(ip - iStart);
+    }
+
+    if (LZ4_isLittleEndian()) {
+        reg_t patternByte = pattern;
+        while ((ip<iEnd) && (*ip == (BYTE)patternByte)) {
+            ip++; patternByte >>= 8;
+        }
+    } else {  /* big endian */
+        U32 bitOffset = (sizeof(pattern)*8) - 8;
+        while (ip < iEnd) {
+            BYTE const byte = (BYTE)(pattern >> bitOffset);
+            if (*ip != byte) break;
+            ip ++; bitOffset -= 8;
+    }   }
+
+    return (unsigned)(ip - iStart);
+}
+
+/* LZ4HC_reverseCountPattern() :
+ * pattern must be a sample of repetitive pattern of length 1, 2 or 4 (but not 3!)
+ * read using natural platform endianness */
+static unsigned
+LZ4HC_reverseCountPattern(const BYTE* ip, const BYTE* const iLow, U32 pattern)
+{
+    const BYTE* const iStart = ip;
+
+    while (likely(ip >= iLow+4)) {
+        if (LZ4_read32(ip-4) != pattern) break;
+        ip -= 4;
+    }
+    {   const BYTE* bytePtr = (const BYTE*)(&pattern) + 3; /* works for any endianness */
+        while (likely(ip>iLow)) {
+            if (ip[-1] != *bytePtr) break;
+            ip--; bytePtr--;
+    }   }
+    return (unsigned)(iStart - ip);
+}
+
+/* LZ4HC_protectDictEnd() :
+ * Checks if the match is in the last 3 bytes of the dictionary, so reading the
+ * 4 byte MINMATCH would overflow.
+ * @returns true if the match index is okay.
+ */
+static int LZ4HC_protectDictEnd(U32 const dictLimit, U32 const matchIndex)
+{
+    return ((U32)((dictLimit - 1) - matchIndex) >= 3);
+}
+
+typedef enum { rep_untested, rep_not, rep_confirmed } repeat_state_e;
+typedef enum { favorCompressionRatio=0, favorDecompressionSpeed } HCfavor_e;
+
+
+LZ4_FORCE_INLINE LZ4HC_match_t
+LZ4HC_InsertAndGetWiderMatch (
+        LZ4HC_CCtx_internal* const hc4,
+        const BYTE* const ip,
+        const BYTE* const iLowLimit, const BYTE* const iHighLimit,
+        int longest,
+        const int maxNbAttempts,
+        const int patternAnalysis, const int chainSwap,
+        const dictCtx_directive dict,
+        const HCfavor_e favorDecSpeed)
+{
+    U16* const chainTable = hc4->chainTable;
+    U32* const hashTable = hc4->hashTable;
+    const LZ4HC_CCtx_internal* const dictCtx = hc4->dictCtx;
+    const BYTE* const prefixPtr = hc4->prefixStart;
+    const U32 prefixIdx = hc4->dictLimit;
+    const U32 ipIndex = (U32)(ip - prefixPtr) + prefixIdx;
+    const int withinStartDistance = (hc4->lowLimit + (LZ4_DISTANCE_MAX + 1) > ipIndex);
+    const U32 lowestMatchIndex = (withinStartDistance) ? hc4->lowLimit : ipIndex - LZ4_DISTANCE_MAX;
+    const BYTE* const dictStart = hc4->dictStart;
+    const U32 dictIdx = hc4->lowLimit;
+    const BYTE* const dictEnd = dictStart + prefixIdx - dictIdx;
+    int const lookBackLength = (int)(ip-iLowLimit);
+    int nbAttempts = maxNbAttempts;
+    U32 matchChainPos = 0;
+    U32 const pattern = LZ4_read32(ip);
+    U32 matchIndex;
+    repeat_state_e repeat = rep_untested;
+    size_t srcPatternLength = 0;
+    int offset = 0, sBack = 0;
+
+    DEBUGLOG(7, "LZ4HC_InsertAndGetWiderMatch");
+    /* First Match */
+    LZ4HC_Insert(hc4, ip);  /* insert all prior positions up to ip (excluded) */
+    matchIndex = hashTable[LZ4HC_hashPtr(ip)];
+    DEBUGLOG(7, "First candidate match for pos %u found at index %u / %u (lowestMatchIndex)",
+                ipIndex, matchIndex, lowestMatchIndex);
+
+    while ((matchIndex>=lowestMatchIndex) && (nbAttempts>0)) {
+        int matchLength=0;
+        nbAttempts--;
+        assert(matchIndex < ipIndex);
+        if (favorDecSpeed && (ipIndex - matchIndex < 8)) {
+            /* do nothing:
+             * favorDecSpeed intentionally skips matches with offset < 8 */
+        } else if (matchIndex >= prefixIdx) {   /* within current Prefix */
+            const BYTE* const matchPtr = prefixPtr + (matchIndex - prefixIdx);
+            assert(matchPtr < ip);
+            assert(longest >= 1);
+            if (LZ4_read16(iLowLimit + longest - 1) == LZ4_read16(matchPtr - lookBackLength + longest - 1)) {
+                if (LZ4_read32(matchPtr) == pattern) {
+                    int const back = lookBackLength ? LZ4HC_countBack(ip, matchPtr, iLowLimit, prefixPtr) : 0;
+                    matchLength = MINMATCH + (int)LZ4_count(ip+MINMATCH, matchPtr+MINMATCH, iHighLimit);
+                    matchLength -= back;
+                    if (matchLength > longest) {
+                        longest = matchLength;
+                        offset = (int)(ipIndex - matchIndex);
+                        sBack = back;
+                        DEBUGLOG(7, "Found match of len=%i within prefix, offset=%i, back=%i", longest, offset, -back);
+            }   }   }
+        } else {   /* lowestMatchIndex <= matchIndex < dictLimit : within Ext Dict */
+            const BYTE* const matchPtr = dictStart + (matchIndex - dictIdx);
+            assert(matchIndex >= dictIdx);
+            if ( likely(matchIndex <= prefixIdx - 4)
+              && (LZ4_read32(matchPtr) == pattern) ) {
+                int back = 0;
+                const BYTE* vLimit = ip + (prefixIdx - matchIndex);
+                if (vLimit > iHighLimit) vLimit = iHighLimit;
+                matchLength = (int)LZ4_count(ip+MINMATCH, matchPtr+MINMATCH, vLimit) + MINMATCH;
+                if ((ip+matchLength == vLimit) && (vLimit < iHighLimit))
+                    matchLength += LZ4_count(ip+matchLength, prefixPtr, iHighLimit);
+                back = lookBackLength ? LZ4HC_countBack(ip, matchPtr, iLowLimit, dictStart) : 0;
+                matchLength -= back;
+                if (matchLength > longest) {
+                    longest = matchLength;
+                    offset = (int)(ipIndex - matchIndex);
+                    sBack = back;
+                    DEBUGLOG(7, "Found match of len=%i within dict, offset=%i, back=%i", longest, offset, -back);
+        }   }   }
+
+        if (chainSwap && matchLength==longest) {   /* better match => select a better chain */
+            assert(lookBackLength==0);   /* search forward only */
+            if (matchIndex + (U32)longest <= ipIndex) {
+                int const kTrigger = 4;
+                U32 distanceToNextMatch = 1;
+                int const end = longest - MINMATCH + 1;
+                int step = 1;
+                int accel = 1 << kTrigger;
+                int pos;
+                for (pos = 0; pos < end; pos += step) {
+                    U32 const candidateDist = DELTANEXTU16(chainTable, matchIndex + (U32)pos);
+                    step = (accel++ >> kTrigger);
+                    if (candidateDist > distanceToNextMatch) {
+                        distanceToNextMatch = candidateDist;
+                        matchChainPos = (U32)pos;
+                        accel = 1 << kTrigger;
+                }   }
+                if (distanceToNextMatch > 1) {
+                    if (distanceToNextMatch > matchIndex) break;   /* avoid overflow */
+                    matchIndex -= distanceToNextMatch;
+                    continue;
+        }   }   }
+
+        {   U32 const distNextMatch = DELTANEXTU16(chainTable, matchIndex);
+            if (patternAnalysis && distNextMatch==1 && matchChainPos==0) {
+                U32 const matchCandidateIdx = matchIndex-1;
+                /* may be a repeated pattern */
+                if (repeat == rep_untested) {
+                    if ( ((pattern & 0xFFFF) == (pattern >> 16))
+                      &  ((pattern & 0xFF)   == (pattern >> 24)) ) {
+                        DEBUGLOG(7, "Repeat pattern detected, char %02X", pattern >> 24);
+                        repeat = rep_confirmed;
+                        srcPatternLength = LZ4HC_countPattern(ip+sizeof(pattern), iHighLimit, pattern) + sizeof(pattern);
+                    } else {
+                        repeat = rep_not;
+                }   }
+                if ( (repeat == rep_confirmed) && (matchCandidateIdx >= lowestMatchIndex)
+                  && LZ4HC_protectDictEnd(prefixIdx, matchCandidateIdx) ) {
+                    const int extDict = matchCandidateIdx < prefixIdx;
+                    const BYTE* const matchPtr = extDict ? dictStart + (matchCandidateIdx - dictIdx) : prefixPtr + (matchCandidateIdx - prefixIdx);
+                    if (LZ4_read32(matchPtr) == pattern) {  /* good candidate */
+                        const BYTE* const iLimit = extDict ? dictEnd : iHighLimit;
+                        size_t forwardPatternLength = LZ4HC_countPattern(matchPtr+sizeof(pattern), iLimit, pattern) + sizeof(pattern);
+                        if (extDict && matchPtr + forwardPatternLength == iLimit) {
+                            U32 const rotatedPattern = LZ4HC_rotatePattern(forwardPatternLength, pattern);
+                            forwardPatternLength += LZ4HC_countPattern(prefixPtr, iHighLimit, rotatedPattern);
+                        }
+                        {   const BYTE* const lowestMatchPtr = extDict ? dictStart : prefixPtr;
+                            size_t backLength = LZ4HC_reverseCountPattern(matchPtr, lowestMatchPtr, pattern);
+                            size_t currentSegmentLength;
+                            if (!extDict
+                              && matchPtr - backLength == prefixPtr
+                              && dictIdx < prefixIdx) {
+                                U32 const rotatedPattern = LZ4HC_rotatePattern((U32)(-(int)backLength), pattern);
+                                backLength += LZ4HC_reverseCountPattern(dictEnd, dictStart, rotatedPattern);
+                            }
+                            /* Limit backLength not go further than lowestMatchIndex */
+                            backLength = matchCandidateIdx - MAX(matchCandidateIdx - (U32)backLength, lowestMatchIndex);
+                            assert(matchCandidateIdx - backLength >= lowestMatchIndex);
+                            currentSegmentLength = backLength + forwardPatternLength;
+                            /* Adjust to end of pattern if the source pattern fits, otherwise the beginning of the pattern */
+                            if ( (currentSegmentLength >= srcPatternLength)   /* current pattern segment large enough to contain full srcPatternLength */
+                              && (forwardPatternLength <= srcPatternLength) ) { /* haven't reached this position yet */
+                                U32 const newMatchIndex = matchCandidateIdx + (U32)forwardPatternLength - (U32)srcPatternLength;  /* best position, full pattern, might be followed by more match */
+                                if (LZ4HC_protectDictEnd(prefixIdx, newMatchIndex))
+                                    matchIndex = newMatchIndex;
+                                else {
+                                    /* Can only happen if started in the prefix */
+                                    assert(newMatchIndex >= prefixIdx - 3 && newMatchIndex < prefixIdx && !extDict);
+                                    matchIndex = prefixIdx;
+                                }
+                            } else {
+                                U32 const newMatchIndex = matchCandidateIdx - (U32)backLength;   /* farthest position in current segment, will find a match of length currentSegmentLength + maybe some back */
+                                if (!LZ4HC_protectDictEnd(prefixIdx, newMatchIndex)) {
+                                    assert(newMatchIndex >= prefixIdx - 3 && newMatchIndex < prefixIdx && !extDict);
+                                    matchIndex = prefixIdx;
+                                } else {
+                                    matchIndex = newMatchIndex;
+                                    if (lookBackLength==0) {  /* no back possible */
+                                        size_t const maxML = MIN(currentSegmentLength, srcPatternLength);
+                                        if ((size_t)longest < maxML) {
+                                            assert(prefixPtr - prefixIdx + matchIndex != ip);
+                                            if ((size_t)(ip - prefixPtr) + prefixIdx - matchIndex > LZ4_DISTANCE_MAX) break;
+                                            assert(maxML < 2 GB);
+                                            longest = (int)maxML;
+                                            offset = (int)(ipIndex - matchIndex);
+                                            assert(sBack == 0);
+                                            DEBUGLOG(7, "Found repeat pattern match of len=%i, offset=%i", longest, offset);
+                                        }
+                                        {   U32 const distToNextPattern = DELTANEXTU16(chainTable, matchIndex);
+                                            if (distToNextPattern > matchIndex) break;  /* avoid overflow */
+                                            matchIndex -= distToNextPattern;
+                        }   }   }   }   }
+                        continue;
+                }   }
+        }   }   /* PA optimization */
+
+        /* follow current chain */
+        matchIndex -= DELTANEXTU16(chainTable, matchIndex + matchChainPos);
+
+    }  /* while ((matchIndex>=lowestMatchIndex) && (nbAttempts)) */
+
+    if ( dict == usingDictCtxHc
+      && nbAttempts > 0
+      && withinStartDistance) {
+        size_t const dictEndOffset = (size_t)(dictCtx->end - dictCtx->prefixStart) + dictCtx->dictLimit;
+        U32 dictMatchIndex = dictCtx->hashTable[LZ4HC_hashPtr(ip)];
+        assert(dictEndOffset <= 1 GB);
+        matchIndex = dictMatchIndex + lowestMatchIndex - (U32)dictEndOffset;
+        if (dictMatchIndex>0) DEBUGLOG(7, "dictEndOffset = %zu, dictMatchIndex = %u => relative matchIndex = %i", dictEndOffset, dictMatchIndex, (int)dictMatchIndex - (int)dictEndOffset);
+        while (ipIndex - matchIndex <= LZ4_DISTANCE_MAX && nbAttempts--) {
+            const BYTE* const matchPtr = dictCtx->prefixStart - dictCtx->dictLimit + dictMatchIndex;
+
+            if (LZ4_read32(matchPtr) == pattern) {
+                int mlt;
+                int back = 0;
+                const BYTE* vLimit = ip + (dictEndOffset - dictMatchIndex);
+                if (vLimit > iHighLimit) vLimit = iHighLimit;
+                mlt = (int)LZ4_count(ip+MINMATCH, matchPtr+MINMATCH, vLimit) + MINMATCH;
+                back = lookBackLength ? LZ4HC_countBack(ip, matchPtr, iLowLimit, dictCtx->prefixStart) : 0;
+                mlt -= back;
+                if (mlt > longest) {
+                    longest = mlt;
+                    offset = (int)(ipIndex - matchIndex);
+                    sBack = back;
+                    DEBUGLOG(7, "found match of length %i within extDictCtx", longest);
+            }   }
+
+            {   U32 const nextOffset = DELTANEXTU16(dictCtx->chainTable, dictMatchIndex);
+                dictMatchIndex -= nextOffset;
+                matchIndex -= nextOffset;
+    }   }   }
+
+    {   LZ4HC_match_t md;
+        assert(longest >= 0);
+        md.len = longest;
+        md.off = offset;
+        md.back = sBack;
+        return md;
+    }
+}
+
+LZ4_FORCE_INLINE LZ4HC_match_t
+LZ4HC_InsertAndFindBestMatch(LZ4HC_CCtx_internal* const hc4,   /* Index table will be updated */
+                       const BYTE* const ip, const BYTE* const iLimit,
+                       const int maxNbAttempts,
+                       const int patternAnalysis,
+                       const dictCtx_directive dict)
+{
+    DEBUGLOG(7, "LZ4HC_InsertAndFindBestMatch");
+    /* note : LZ4HC_InsertAndGetWiderMatch() is able to modify the starting position of a match (*startpos),
+     * but this won't be the case here, as we define iLowLimit==ip,
+     * so LZ4HC_InsertAndGetWiderMatch() won't be allowed to search past ip */
+    return LZ4HC_InsertAndGetWiderMatch(hc4, ip, ip, iLimit, MINMATCH-1, maxNbAttempts, patternAnalysis, 0 /*chainSwap*/, dict, favorCompressionRatio);
+}
+
+
+LZ4_FORCE_INLINE int LZ4HC_compress_hashChain (
+    LZ4HC_CCtx_internal* const ctx,
+    const char* const source,
+    char* const dest,
+    int* srcSizePtr,
+    int const maxOutputSize,
+    int maxNbAttempts,
+    const limitedOutput_directive limit,
+    const dictCtx_directive dict
+    )
+{
+    const int inputSize = *srcSizePtr;
+    const int patternAnalysis = (maxNbAttempts > 128);   /* levels 9+ */
+
+    const BYTE* ip = (const BYTE*) source;
+    const BYTE* anchor = ip;
+    const BYTE* const iend = ip + inputSize;
+    const BYTE* const mflimit = iend - MFLIMIT;
+    const BYTE* const matchlimit = (iend - LASTLITERALS);
+
+    BYTE* optr = (BYTE*) dest;
+    BYTE* op = (BYTE*) dest;
+    BYTE* oend = op + maxOutputSize;
+
+    const BYTE* start0;
+    const BYTE* start2 = NULL;
+    const BYTE* start3 = NULL;
+    LZ4HC_match_t m0, m1, m2, m3;
+    const LZ4HC_match_t nomatch = {0, 0, 0};
+
+    /* init */
+    DEBUGLOG(5, "LZ4HC_compress_hashChain (dict?=>%i)", dict);
+    *srcSizePtr = 0;
+    if (limit == fillOutput) oend -= LASTLITERALS;                  /* Hack for support LZ4 format restriction */
+    if (inputSize < LZ4_minLength) goto _last_literals;             /* Input too small, no compression (all literals) */
+
+    /* Main Loop */
+    while (ip <= mflimit) {
+        m1 = LZ4HC_InsertAndFindBestMatch(ctx, ip, matchlimit, maxNbAttempts, patternAnalysis, dict);
+        if (m1.len<MINMATCH) { ip++; continue; }
+
+        /* saved, in case we would skip too much */
+        start0 = ip; m0 = m1;
+
+_Search2:
+        DEBUGLOG(7, "_Search2 (currently found match of size %i)", m1.len);
+        if (ip+m1.len <= mflimit) {
+            start2 = ip + m1.len - 2;
+            m2 = LZ4HC_InsertAndGetWiderMatch(ctx,
+                            start2, ip + 0, matchlimit, m1.len,
+                            maxNbAttempts, patternAnalysis, 0, dict, favorCompressionRatio);
+            start2 += m2.back;
+        } else {
+            m2 = nomatch;  /* do not search further */
+        }
+
+        if (m2.len <= m1.len) { /* No better match => encode ML1 immediately */
+            optr = op;
+            if (LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor),
+                    m1.len, m1.off,
+                    limit, oend) )
+                goto _dest_overflow;
+            continue;
+        }
+
+        if (start0 < ip) {   /* first match was skipped at least once */
+            if (start2 < ip + m0.len) {  /* squeezing ML1 between ML0(original ML1) and ML2 */
+                ip = start0; m1 = m0;  /* restore initial Match1 */
+        }   }
+
+        /* Here, start0==ip */
+        if ((start2 - ip) < 3) {  /* First Match too small : removed */
+            ip = start2;
+            m1 = m2;
+            goto _Search2;
+        }
+
+_Search3:
+        if ((start2 - ip) < OPTIMAL_ML) {
+            int correction;
+            int new_ml = m1.len;
+            if (new_ml > OPTIMAL_ML) new_ml = OPTIMAL_ML;
+            if (ip+new_ml > start2 + m2.len - MINMATCH)
+                new_ml = (int)(start2 - ip) + m2.len - MINMATCH;
+            correction = new_ml - (int)(start2 - ip);
+            if (correction > 0) {
+                start2 += correction;
+                m2.len -= correction;
+            }
+        }
+
+        if (start2 + m2.len <= mflimit) {
+            start3 = start2 + m2.len - 3;
+            m3 = LZ4HC_InsertAndGetWiderMatch(ctx,
+                            start3, start2, matchlimit, m2.len,
+                            maxNbAttempts, patternAnalysis, 0, dict, favorCompressionRatio);
+            start3 += m3.back;
+        } else {
+            m3 = nomatch;  /* do not search further */
+        }
+
+        if (m3.len <= m2.len) {  /* No better match => encode ML1 and ML2 */
+            /* ip & ref are known; Now for ml */
+            if (start2 < ip+m1.len) m1.len = (int)(start2 - ip);
+            /* Now, encode 2 sequences */
+            optr = op;
+            if (LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor),
+                    m1.len, m1.off,
+                    limit, oend) )
+                goto _dest_overflow;
+            ip = start2;
+            optr = op;
+            if (LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor),
+                    m2.len, m2.off,
+                    limit, oend) ) {
+                m1 = m2;
+                goto _dest_overflow;
+            }
+            continue;
+        }
+
+        if (start3 < ip+m1.len+3) {  /* Not enough space for match 2 : remove it */
+            if (start3 >= (ip+m1.len)) {  /* can write Seq1 immediately ==> Seq2 is removed, so Seq3 becomes Seq1 */
+                if (start2 < ip+m1.len) {
+                    int correction = (int)(ip+m1.len - start2);
+                    start2 += correction;
+                    m2.len -= correction;
+                    if (m2.len < MINMATCH) {
+                        start2 = start3;
+                        m2 = m3;
+                    }
+                }
+
+                optr = op;
+                if (LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor),
+                        m1.len, m1.off,
+                        limit, oend) )
+                    goto _dest_overflow;
+                ip  = start3;
+                m1 = m3;
+
+                start0 = start2;
+                m0 = m2;
+                goto _Search2;
+            }
+
+            start2 = start3;
+            m2 = m3;
+            goto _Search3;
+        }
+
+        /*
+        * OK, now we have 3 ascending matches;
+        * let's write the first one ML1.
+        * ip & ref are known; Now decide ml.
+        */
+        if (start2 < ip+m1.len) {
+            if ((start2 - ip) < OPTIMAL_ML) {
+                int correction;
+                if (m1.len > OPTIMAL_ML) m1.len = OPTIMAL_ML;
+                if (ip + m1.len > start2 + m2.len - MINMATCH)
+                    m1.len = (int)(start2 - ip) + m2.len - MINMATCH;
+                correction = m1.len - (int)(start2 - ip);
+                if (correction > 0) {
+                    start2 += correction;
+                    m2.len -= correction;
+                }
+            } else {
+                m1.len = (int)(start2 - ip);
+            }
+        }
+        optr = op;
+        if ( LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor),
+                m1.len, m1.off,
+                limit, oend) )
+            goto _dest_overflow;
+
+        /* ML2 becomes ML1 */
+        ip = start2; m1 = m2;
+
+        /* ML3 becomes ML2 */
+        start2 = start3; m2 = m3;
+
+        /* let's find a new ML3 */
+        goto _Search3;
+    }
+
+_last_literals:
+    /* Encode Last Literals */
+    {   size_t lastRunSize = (size_t)(iend - anchor);  /* literals */
+        size_t llAdd = (lastRunSize + 255 - RUN_MASK) / 255;
+        size_t const totalSize = 1 + llAdd + lastRunSize;
+        if (limit == fillOutput) oend += LASTLITERALS;  /* restore correct value */
+        if (limit && (op + totalSize > oend)) {
+            if (limit == limitedOutput) return 0;
+            /* adapt lastRunSize to fill 'dest' */
+            lastRunSize  = (size_t)(oend - op) - 1 /*token*/;
+            llAdd = (lastRunSize + 256 - RUN_MASK) / 256;
+            lastRunSize -= llAdd;
+        }
+        DEBUGLOG(6, "Final literal run : %i literals", (int)lastRunSize);
+        ip = anchor + lastRunSize;  /* can be != iend if limit==fillOutput */
+
+        if (lastRunSize >= RUN_MASK) {
+            size_t accumulator = lastRunSize - RUN_MASK;
+            *op++ = (RUN_MASK << ML_BITS);
+            for(; accumulator >= 255 ; accumulator -= 255) *op++ = 255;
+            *op++ = (BYTE) accumulator;
+        } else {
+            *op++ = (BYTE)(lastRunSize << ML_BITS);
+        }
+        LZ4_memcpy(op, anchor, lastRunSize);
+        op += lastRunSize;
+    }
+
+    /* End */
+    *srcSizePtr = (int) (((const char*)ip) - source);
+    return (int) (((char*)op)-dest);
+
+_dest_overflow:
+    if (limit == fillOutput) {
+        /* Assumption : @ip, @anchor, @optr and @m1 must be set correctly */
+        size_t const ll = (size_t)(ip - anchor);
+        size_t const ll_addbytes = (ll + 240) / 255;
+        size_t const ll_totalCost = 1 + ll_addbytes + ll;
+        BYTE* const maxLitPos = oend - 3; /* 2 for offset, 1 for token */
+        DEBUGLOG(6, "Last sequence overflowing");
+        op = optr;  /* restore correct out pointer */
+        if (op + ll_totalCost <= maxLitPos) {
+            /* ll validated; now adjust match length */
+            size_t const bytesLeftForMl = (size_t)(maxLitPos - (op+ll_totalCost));
+            size_t const maxMlSize = MINMATCH + (ML_MASK-1) + (bytesLeftForMl * 255);
+            assert(maxMlSize < INT_MAX); assert(m1.len >= 0);
+            if ((size_t)m1.len > maxMlSize) m1.len = (int)maxMlSize;
+            if ((oend + LASTLITERALS) - (op + ll_totalCost + 2) - 1 + m1.len >= MFLIMIT) {
+                LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), m1.len, m1.off, notLimited, oend);
+        }   }
+        goto _last_literals;
+    }
+    /* compression failed */
+    return 0;
+}
+
+
+static int LZ4HC_compress_optimal( LZ4HC_CCtx_internal* ctx,
+    const char* const source, char* dst,
+    int* srcSizePtr, int dstCapacity,
+    int const nbSearches, size_t sufficient_len,
+    const limitedOutput_directive limit, int const fullUpdate,
+    const dictCtx_directive dict,
+    const HCfavor_e favorDecSpeed);
+
+LZ4_FORCE_INLINE int
+LZ4HC_compress_generic_internal (
+            LZ4HC_CCtx_internal* const ctx,
+            const char* const src,
+            char* const dst,
+            int* const srcSizePtr,
+            int const dstCapacity,
+            int cLevel,
+            const limitedOutput_directive limit,
+            const dictCtx_directive dict
+            )
+{
+    DEBUGLOG(5, "LZ4HC_compress_generic_internal(src=%p, srcSize=%d)",
+                src, *srcSizePtr);
+
+    if (limit == fillOutput && dstCapacity < 1) return 0;   /* Impossible to store anything */
+    if ((U32)*srcSizePtr > (U32)LZ4_MAX_INPUT_SIZE) return 0;  /* Unsupported input size (too large or negative) */
+
+    ctx->end += *srcSizePtr;
+    {   cParams_t const cParam = LZ4HC_getCLevelParams(cLevel);
+        HCfavor_e const favor = ctx->favorDecSpeed ? favorDecompressionSpeed : favorCompressionRatio;
+        int result;
+
+        if (cParam.strat == lz4mid) {
+            result = LZ4MID_compress(ctx,
+                                src, dst, srcSizePtr, dstCapacity,
+                                limit, dict);
+        } else if (cParam.strat == lz4hc) {
+            result = LZ4HC_compress_hashChain(ctx,
+                                src, dst, srcSizePtr, dstCapacity,
+                                cParam.nbSearches, limit, dict);
+        } else {
+            assert(cParam.strat == lz4opt);
+            result = LZ4HC_compress_optimal(ctx,
+                                src, dst, srcSizePtr, dstCapacity,
+                                cParam.nbSearches, cParam.targetLength, limit,
+                                cLevel >= LZ4HC_CLEVEL_MAX,   /* ultra mode */
+                                dict, favor);
+        }
+        if (result <= 0) ctx->dirty = 1;
+        return result;
+    }
+}
+
+static void LZ4HC_setExternalDict(LZ4HC_CCtx_internal* ctxPtr, const BYTE* newBlock);
+
+static int
+LZ4HC_compress_generic_noDictCtx (
+        LZ4HC_CCtx_internal* const ctx,
+        const char* const src,
+        char* const dst,
+        int* const srcSizePtr,
+        int const dstCapacity,
+        int cLevel,
+        limitedOutput_directive limit
+        )
+{
+    assert(ctx->dictCtx == NULL);
+    return LZ4HC_compress_generic_internal(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit, noDictCtx);
+}
+
+static int isStateCompatible(const LZ4HC_CCtx_internal* ctx1, const LZ4HC_CCtx_internal* ctx2)
+{
+    int const isMid1 = LZ4HC_getCLevelParams(ctx1->compressionLevel).strat == lz4mid;
+    int const isMid2 = LZ4HC_getCLevelParams(ctx2->compressionLevel).strat == lz4mid;
+    return !(isMid1 ^ isMid2);
+}
+
+static int
+LZ4HC_compress_generic_dictCtx (
+        LZ4HC_CCtx_internal* const ctx,
+        const char* const src,
+        char* const dst,
+        int* const srcSizePtr,
+        int const dstCapacity,
+        int cLevel,
+        limitedOutput_directive limit
+        )
+{
+    const size_t position = (size_t)(ctx->end - ctx->prefixStart) + (ctx->dictLimit - ctx->lowLimit);
+    assert(ctx->dictCtx != NULL);
+    if (position >= 64 KB) {
+        ctx->dictCtx = NULL;
+        return LZ4HC_compress_generic_noDictCtx(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit);
+    } else if (position == 0 && *srcSizePtr > 4 KB && isStateCompatible(ctx, ctx->dictCtx)) {
+        LZ4_memcpy(ctx, ctx->dictCtx, sizeof(LZ4HC_CCtx_internal));
+        LZ4HC_setExternalDict(ctx, (const BYTE *)src);
+        ctx->compressionLevel = (short)cLevel;
+        return LZ4HC_compress_generic_noDictCtx(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit);
+    } else {
+        return LZ4HC_compress_generic_internal(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit, usingDictCtxHc);
+    }
+}
+
+static int
+LZ4HC_compress_generic (
+        LZ4HC_CCtx_internal* const ctx,
+        const char* const src,
+        char* const dst,
+        int* const srcSizePtr,
+        int const dstCapacity,
+        int cLevel,
+        limitedOutput_directive limit
+        )
+{
+    if (ctx->dictCtx == NULL) {
+        return LZ4HC_compress_generic_noDictCtx(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit);
+    } else {
+        return LZ4HC_compress_generic_dictCtx(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit);
+    }
+}
+
+
+int LZ4_sizeofStateHC(void) { return (int)sizeof(LZ4_streamHC_t); }
+
+static size_t LZ4_streamHC_t_alignment(void)
+{
+#if LZ4_ALIGN_TEST
+    typedef struct { char c; LZ4_streamHC_t t; } t_a;
+    return sizeof(t_a) - sizeof(LZ4_streamHC_t);
+#else
+    return 1;  /* effectively disabled */
+#endif
+}
+
+/* state is presumed correctly initialized,
+ * in which case its size and alignment have already been validate */
+int LZ4_compress_HC_extStateHC_fastReset (void* state, const char* src, char* dst, int srcSize, int dstCapacity, int compressionLevel)
+{
+    LZ4HC_CCtx_internal* const ctx = &((LZ4_streamHC_t*)state)->internal_donotuse;
+    if (!LZ4_isAligned(state, LZ4_streamHC_t_alignment())) return 0;
+    LZ4_resetStreamHC_fast((LZ4_streamHC_t*)state, compressionLevel);
+    LZ4HC_init_internal (ctx, (const BYTE*)src);
+    if (dstCapacity < LZ4_compressBound(srcSize))
+        return LZ4HC_compress_generic (ctx, src, dst, &srcSize, dstCapacity, compressionLevel, limitedOutput);
+    else
+        return LZ4HC_compress_generic (ctx, src, dst, &srcSize, dstCapacity, compressionLevel, notLimited);
+}
+
+int LZ4_compress_HC_extStateHC (void* state, const char* src, char* dst, int srcSize, int dstCapacity, int compressionLevel)
+{
+    LZ4_streamHC_t* const ctx = LZ4_initStreamHC(state, sizeof(*ctx));
+    if (ctx==NULL) return 0;   /* init failure */
+    return LZ4_compress_HC_extStateHC_fastReset(state, src, dst, srcSize, dstCapacity, compressionLevel);
+}
+
+int LZ4_compress_HC(const char* src, char* dst, int srcSize, int dstCapacity, int compressionLevel)
+{
+    int cSize;
+#if defined(LZ4HC_HEAPMODE) && LZ4HC_HEAPMODE==1
+    LZ4_streamHC_t* const statePtr = (LZ4_streamHC_t*)ALLOC(sizeof(LZ4_streamHC_t));
+    if (statePtr==NULL) return 0;
+#else
+    LZ4_streamHC_t state;
+    LZ4_streamHC_t* const statePtr = &state;
+#endif
+    DEBUGLOG(5, "LZ4_compress_HC")
+    cSize = LZ4_compress_HC_extStateHC(statePtr, src, dst, srcSize, dstCapacity, compressionLevel);
+#if defined(LZ4HC_HEAPMODE) && LZ4HC_HEAPMODE==1
+    FREEMEM(statePtr);
+#endif
+    return cSize;
+}
+
+/* state is presumed sized correctly (>= sizeof(LZ4_streamHC_t)) */
+int LZ4_compress_HC_destSize(void* state, const char* source, char* dest, int* sourceSizePtr, int targetDestSize, int cLevel)
+{
+    LZ4_streamHC_t* const ctx = LZ4_initStreamHC(state, sizeof(*ctx));
+    if (ctx==NULL) return 0;   /* init failure */
+    LZ4HC_init_internal(&ctx->internal_donotuse, (const BYTE*) source);
+    LZ4_setCompressionLevel(ctx, cLevel);
+    return LZ4HC_compress_generic(&ctx->internal_donotuse, source, dest, sourceSizePtr, targetDestSize, cLevel, fillOutput);
+}
+
+
+
+/**************************************
+*  Streaming Functions
+**************************************/
+/* allocation */
+#if !defined(LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION)
+LZ4_streamHC_t* LZ4_createStreamHC(void)
+{
+    LZ4_streamHC_t* const state =
+        (LZ4_streamHC_t*)ALLOC_AND_ZERO(sizeof(LZ4_streamHC_t));
+    if (state == NULL) return NULL;
+    LZ4_setCompressionLevel(state, LZ4HC_CLEVEL_DEFAULT);
+    return state;
+}
+
+int LZ4_freeStreamHC (LZ4_streamHC_t* LZ4_streamHCPtr)
+{
+    DEBUGLOG(4, "LZ4_freeStreamHC(%p)", LZ4_streamHCPtr);
+    if (!LZ4_streamHCPtr) return 0;  /* support free on NULL */
+    FREEMEM(LZ4_streamHCPtr);
+    return 0;
+}
+#endif
+
+
+LZ4_streamHC_t* LZ4_initStreamHC (void* buffer, size_t size)
+{
+    LZ4_streamHC_t* const LZ4_streamHCPtr = (LZ4_streamHC_t*)buffer;
+    DEBUGLOG(4, "LZ4_initStreamHC(%p, %u)", buffer, (unsigned)size);
+    /* check conditions */
+    if (buffer == NULL) return NULL;
+    if (size < sizeof(LZ4_streamHC_t)) return NULL;
+    if (!LZ4_isAligned(buffer, LZ4_streamHC_t_alignment())) return NULL;
+    /* init */
+    { LZ4HC_CCtx_internal* const hcstate = &(LZ4_streamHCPtr->internal_donotuse);
+      MEM_INIT(hcstate, 0, sizeof(*hcstate)); }
+    LZ4_setCompressionLevel(LZ4_streamHCPtr, LZ4HC_CLEVEL_DEFAULT);
+    return LZ4_streamHCPtr;
+}
+
+/* just a stub */
+void LZ4_resetStreamHC (LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel)
+{
+    LZ4_initStreamHC(LZ4_streamHCPtr, sizeof(*LZ4_streamHCPtr));
+    LZ4_setCompressionLevel(LZ4_streamHCPtr, compressionLevel);
+}
+
+void LZ4_resetStreamHC_fast (LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel)
+{
+    LZ4HC_CCtx_internal* const s = &LZ4_streamHCPtr->internal_donotuse;
+    DEBUGLOG(5, "LZ4_resetStreamHC_fast(%p, %d)", LZ4_streamHCPtr, compressionLevel);
+    if (s->dirty) {
+        LZ4_initStreamHC(LZ4_streamHCPtr, sizeof(*LZ4_streamHCPtr));
+    } else {
+        assert(s->end >= s->prefixStart);
+        s->dictLimit += (U32)(s->end - s->prefixStart);
+        s->prefixStart = NULL;
+        s->end = NULL;
+        s->dictCtx = NULL;
+    }
+    LZ4_setCompressionLevel(LZ4_streamHCPtr, compressionLevel);
+}
+
+void LZ4_setCompressionLevel(LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel)
+{
+    DEBUGLOG(5, "LZ4_setCompressionLevel(%p, %d)", LZ4_streamHCPtr, compressionLevel);
+    if (compressionLevel < 1) compressionLevel = LZ4HC_CLEVEL_DEFAULT;
+    if (compressionLevel > LZ4HC_CLEVEL_MAX) compressionLevel = LZ4HC_CLEVEL_MAX;
+    LZ4_streamHCPtr->internal_donotuse.compressionLevel = (short)compressionLevel;
+}
+
+void LZ4_favorDecompressionSpeed(LZ4_streamHC_t* LZ4_streamHCPtr, int favor)
+{
+    LZ4_streamHCPtr->internal_donotuse.favorDecSpeed = (favor!=0);
+}
+
+/* LZ4_loadDictHC() :
+ * LZ4_streamHCPtr is presumed properly initialized */
+int LZ4_loadDictHC (LZ4_streamHC_t* LZ4_streamHCPtr,
+              const char* dictionary, int dictSize)
+{
+    LZ4HC_CCtx_internal* const ctxPtr = &LZ4_streamHCPtr->internal_donotuse;
+    cParams_t cp;
+    DEBUGLOG(4, "LZ4_loadDictHC(ctx:%p, dict:%p, dictSize:%d, clevel=%d)", LZ4_streamHCPtr, dictionary, dictSize, ctxPtr->compressionLevel);
+    assert(dictSize >= 0);
+    assert(LZ4_streamHCPtr != NULL);
+    if (dictSize > 64 KB) {
+        dictionary += (size_t)dictSize - 64 KB;
+        dictSize = 64 KB;
+    }
+    /* need a full initialization, there are bad side-effects when using resetFast() */
+    {   int const cLevel = ctxPtr->compressionLevel;
+        LZ4_initStreamHC(LZ4_streamHCPtr, sizeof(*LZ4_streamHCPtr));
+        LZ4_setCompressionLevel(LZ4_streamHCPtr, cLevel);
+        cp = LZ4HC_getCLevelParams(cLevel);
+    }
+    LZ4HC_init_internal (ctxPtr, (const BYTE*)dictionary);
+    ctxPtr->end = (const BYTE*)dictionary + dictSize;
+    if (cp.strat == lz4mid) {
+        LZ4MID_fillHTable (ctxPtr, dictionary, (size_t)dictSize);
+    } else {
+        if (dictSize >= LZ4HC_HASHSIZE) LZ4HC_Insert (ctxPtr, ctxPtr->end-3);
+    }
+    return dictSize;
+}
+
+void LZ4_attach_HC_dictionary(LZ4_streamHC_t *working_stream, const LZ4_streamHC_t *dictionary_stream) {
+    working_stream->internal_donotuse.dictCtx = dictionary_stream != NULL ? &(dictionary_stream->internal_donotuse) : NULL;
+}
+
+/* compression */
+
+static void LZ4HC_setExternalDict(LZ4HC_CCtx_internal* ctxPtr, const BYTE* newBlock)
+{
+    DEBUGLOG(4, "LZ4HC_setExternalDict(%p, %p)", ctxPtr, newBlock);
+    if ( (ctxPtr->end >= ctxPtr->prefixStart + 4)
+      && (LZ4HC_getCLevelParams(ctxPtr->compressionLevel).strat != lz4mid) ) {
+        LZ4HC_Insert (ctxPtr, ctxPtr->end-3);  /* Referencing remaining dictionary content */
+    }
+
+    /* Only one memory segment for extDict, so any previous extDict is lost at this stage */
+    ctxPtr->lowLimit  = ctxPtr->dictLimit;
+    ctxPtr->dictStart  = ctxPtr->prefixStart;
+    ctxPtr->dictLimit += (U32)(ctxPtr->end - ctxPtr->prefixStart);
+    ctxPtr->prefixStart = newBlock;
+    ctxPtr->end  = newBlock;
+    ctxPtr->nextToUpdate = ctxPtr->dictLimit;   /* match referencing will resume from there */
+
+    /* cannot reference an extDict and a dictCtx at the same time */
+    ctxPtr->dictCtx = NULL;
+}
+
+static int
+LZ4_compressHC_continue_generic (LZ4_streamHC_t* LZ4_streamHCPtr,
+                                 const char* src, char* dst,
+                                 int* srcSizePtr, int dstCapacity,
+                                 limitedOutput_directive limit)
+{
+    LZ4HC_CCtx_internal* const ctxPtr = &LZ4_streamHCPtr->internal_donotuse;
+    DEBUGLOG(5, "LZ4_compressHC_continue_generic(ctx=%p, src=%p, srcSize=%d, limit=%d)",
+                LZ4_streamHCPtr, src, *srcSizePtr, limit);
+    assert(ctxPtr != NULL);
+    /* auto-init if forgotten */
+    if (ctxPtr->prefixStart == NULL)
+        LZ4HC_init_internal (ctxPtr, (const BYTE*) src);
+
+    /* Check overflow */
+    if ((size_t)(ctxPtr->end - ctxPtr->prefixStart) + ctxPtr->dictLimit > 2 GB) {
+        size_t dictSize = (size_t)(ctxPtr->end - ctxPtr->prefixStart);
+        if (dictSize > 64 KB) dictSize = 64 KB;
+        LZ4_loadDictHC(LZ4_streamHCPtr, (const char*)(ctxPtr->end) - dictSize, (int)dictSize);
+    }
+
+    /* Check if blocks follow each other */
+    if ((const BYTE*)src != ctxPtr->end)
+        LZ4HC_setExternalDict(ctxPtr, (const BYTE*)src);
+
+    /* Check overlapping input/dictionary space */
+    {   const BYTE* sourceEnd = (const BYTE*) src + *srcSizePtr;
+        const BYTE* const dictBegin = ctxPtr->dictStart;
+        const BYTE* const dictEnd   = ctxPtr->dictStart + (ctxPtr->dictLimit - ctxPtr->lowLimit);
+        if ((sourceEnd > dictBegin) && ((const BYTE*)src < dictEnd)) {
+            if (sourceEnd > dictEnd) sourceEnd = dictEnd;
+            ctxPtr->lowLimit += (U32)(sourceEnd - ctxPtr->dictStart);
+            ctxPtr->dictStart += (U32)(sourceEnd - ctxPtr->dictStart);
+            /* invalidate dictionary is it's too small */
+            if (ctxPtr->dictLimit - ctxPtr->lowLimit < LZ4HC_HASHSIZE) {
+                ctxPtr->lowLimit = ctxPtr->dictLimit;
+                ctxPtr->dictStart = ctxPtr->prefixStart;
+    }   }   }
+
+    return LZ4HC_compress_generic (ctxPtr, src, dst, srcSizePtr, dstCapacity, ctxPtr->compressionLevel, limit);
+}
+
+int LZ4_compress_HC_continue (LZ4_streamHC_t* LZ4_streamHCPtr, const char* src, char* dst, int srcSize, int dstCapacity)
+{
+    DEBUGLOG(5, "LZ4_compress_HC_continue");
+    if (dstCapacity < LZ4_compressBound(srcSize))
+        return LZ4_compressHC_continue_generic (LZ4_streamHCPtr, src, dst, &srcSize, dstCapacity, limitedOutput);
+    else
+        return LZ4_compressHC_continue_generic (LZ4_streamHCPtr, src, dst, &srcSize, dstCapacity, notLimited);
+}
+
+int LZ4_compress_HC_continue_destSize (LZ4_streamHC_t* LZ4_streamHCPtr, const char* src, char* dst, int* srcSizePtr, int targetDestSize)
+{
+    return LZ4_compressHC_continue_generic(LZ4_streamHCPtr, src, dst, srcSizePtr, targetDestSize, fillOutput);
+}
+
+
+/* LZ4_saveDictHC :
+ * save history content
+ * into a user-provided buffer
+ * which is then used to continue compression
+ */
+int LZ4_saveDictHC (LZ4_streamHC_t* LZ4_streamHCPtr, char* safeBuffer, int dictSize)
+{
+    LZ4HC_CCtx_internal* const streamPtr = &LZ4_streamHCPtr->internal_donotuse;
+    int const prefixSize = (int)(streamPtr->end - streamPtr->prefixStart);
+    DEBUGLOG(5, "LZ4_saveDictHC(%p, %p, %d)", LZ4_streamHCPtr, safeBuffer, dictSize);
+    assert(prefixSize >= 0);
+    if (dictSize > 64 KB) dictSize = 64 KB;
+    if (dictSize < 4) dictSize = 0;
+    if (dictSize > prefixSize) dictSize = prefixSize;
+    if (safeBuffer == NULL) assert(dictSize == 0);
+    if (dictSize > 0)
+        LZ4_memmove(safeBuffer, streamPtr->end - dictSize, (size_t)dictSize);
+    {   U32 const endIndex = (U32)(streamPtr->end - streamPtr->prefixStart) + streamPtr->dictLimit;
+        streamPtr->end = (safeBuffer == NULL) ? NULL : (const BYTE*)safeBuffer + dictSize;
+        streamPtr->prefixStart = (const BYTE*)safeBuffer;
+        streamPtr->dictLimit = endIndex - (U32)dictSize;
+        streamPtr->lowLimit = endIndex - (U32)dictSize;
+        streamPtr->dictStart = streamPtr->prefixStart;
+        if (streamPtr->nextToUpdate < streamPtr->dictLimit)
+            streamPtr->nextToUpdate = streamPtr->dictLimit;
+    }
+    return dictSize;
+}
+
+
+/* ================================================
+ *  LZ4 Optimal parser (levels [LZ4HC_CLEVEL_OPT_MIN - LZ4HC_CLEVEL_MAX])
+ * ===============================================*/
+typedef struct {
+    int price;
+    int off;
+    int mlen;
+    int litlen;
+} LZ4HC_optimal_t;
+
+/* price in bytes */
+LZ4_FORCE_INLINE int LZ4HC_literalsPrice(int const litlen)
+{
+    int price = litlen;
+    assert(litlen >= 0);
+    if (litlen >= (int)RUN_MASK)
+        price += 1 + ((litlen-(int)RUN_MASK) / 255);
+    return price;
+}
+
+/* requires mlen >= MINMATCH */
+LZ4_FORCE_INLINE int LZ4HC_sequencePrice(int litlen, int mlen)
+{
+    int price = 1 + 2 ; /* token + 16-bit offset */
+    assert(litlen >= 0);
+    assert(mlen >= MINMATCH);
+
+    price += LZ4HC_literalsPrice(litlen);
+
+    if (mlen >= (int)(ML_MASK+MINMATCH))
+        price += 1 + ((mlen-(int)(ML_MASK+MINMATCH)) / 255);
+
+    return price;
+}
+
+LZ4_FORCE_INLINE LZ4HC_match_t
+LZ4HC_FindLongerMatch(LZ4HC_CCtx_internal* const ctx,
+                      const BYTE* ip, const BYTE* const iHighLimit,
+                      int minLen, int nbSearches,
+                      const dictCtx_directive dict,
+                      const HCfavor_e favorDecSpeed)
+{
+    LZ4HC_match_t const match0 = { 0 , 0, 0 };
+    /* note : LZ4HC_InsertAndGetWiderMatch() is able to modify the starting position of a match (*startpos),
+     * but this won't be the case here, as we define iLowLimit==ip,
+    ** so LZ4HC_InsertAndGetWiderMatch() won't be allowed to search past ip */
+    LZ4HC_match_t md = LZ4HC_InsertAndGetWiderMatch(ctx, ip, ip, iHighLimit, minLen, nbSearches, 1 /*patternAnalysis*/, 1 /*chainSwap*/, dict, favorDecSpeed);
+    assert(md.back == 0);
+    if (md.len <= minLen) return match0;
+    if (favorDecSpeed) {
+        if ((md.len>18) & (md.len<=36)) md.len=18;   /* favor dec.speed (shortcut) */
+    }
+    return md;
+}
+
+
+static int LZ4HC_compress_optimal ( LZ4HC_CCtx_internal* ctx,
+                                    const char* const source,
+                                    char* dst,
+                                    int* srcSizePtr,
+                                    int dstCapacity,
+                                    int const nbSearches,
+                                    size_t sufficient_len,
+                                    const limitedOutput_directive limit,
+                                    int const fullUpdate,
+                                    const dictCtx_directive dict,
+                                    const HCfavor_e favorDecSpeed)
+{
+    int retval = 0;
+#define TRAILING_LITERALS 3
+#if defined(LZ4HC_HEAPMODE) && LZ4HC_HEAPMODE==1
+    LZ4HC_optimal_t* const opt = (LZ4HC_optimal_t*)ALLOC(sizeof(LZ4HC_optimal_t) * (LZ4_OPT_NUM + TRAILING_LITERALS));
+#else
+    LZ4HC_optimal_t opt[LZ4_OPT_NUM + TRAILING_LITERALS];   /* ~64 KB, which is a bit large for stack... */
+#endif
+
+    const BYTE* ip = (const BYTE*) source;
+    const BYTE* anchor = ip;
+    const BYTE* const iend = ip + *srcSizePtr;
+    const BYTE* const mflimit = iend - MFLIMIT;
+    const BYTE* const matchlimit = iend - LASTLITERALS;
+    BYTE* op = (BYTE*) dst;
+    BYTE* opSaved = (BYTE*) dst;
+    BYTE* oend = op + dstCapacity;
+    int ovml = MINMATCH;  /* overflow - last sequence */
+    int ovoff = 0;
+
+    /* init */
+#if defined(LZ4HC_HEAPMODE) && LZ4HC_HEAPMODE==1
+    if (opt == NULL) goto _return_label;
+#endif
+    DEBUGLOG(5, "LZ4HC_compress_optimal(dst=%p, dstCapa=%u)", dst, (unsigned)dstCapacity);
+    *srcSizePtr = 0;
+    if (limit == fillOutput) oend -= LASTLITERALS;   /* Hack for support LZ4 format restriction */
+    if (sufficient_len >= LZ4_OPT_NUM) sufficient_len = LZ4_OPT_NUM-1;
+
+    /* Main Loop */
+    while (ip <= mflimit) {
+         int const llen = (int)(ip - anchor);
+         int best_mlen, best_off;
+         int cur, last_match_pos = 0;
+
+         LZ4HC_match_t const firstMatch = LZ4HC_FindLongerMatch(ctx, ip, matchlimit, MINMATCH-1, nbSearches, dict, favorDecSpeed);
+         if (firstMatch.len==0) { ip++; continue; }
+
+         if ((size_t)firstMatch.len > sufficient_len) {
+             /* good enough solution : immediate encoding */
+             int const firstML = firstMatch.len;
+             opSaved = op;
+             if ( LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), firstML, firstMatch.off, limit, oend) ) {  /* updates ip, op and anchor */
+                 ovml = firstML;
+                 ovoff = firstMatch.off;
+                 goto _dest_overflow;
+             }
+             continue;
+         }
+
+         /* set prices for first positions (literals) */
+         {   int rPos;
+             for (rPos = 0 ; rPos < MINMATCH ; rPos++) {
+                 int const cost = LZ4HC_literalsPrice(llen + rPos);
+                 opt[rPos].mlen = 1;
+                 opt[rPos].off = 0;
+                 opt[rPos].litlen = llen + rPos;
+                 opt[rPos].price = cost;
+                 DEBUGLOG(7, "rPos:%3i => price:%3i (litlen=%i) -- initial setup",
+                             rPos, cost, opt[rPos].litlen);
+         }   }
+         /* set prices using initial match */
+         {   int const matchML = firstMatch.len;   /* necessarily < sufficient_len < LZ4_OPT_NUM */
+             int const offset = firstMatch.off;
+             int mlen;
+             assert(matchML < LZ4_OPT_NUM);
+             for (mlen = MINMATCH ; mlen <= matchML ; mlen++) {
+                 int const cost = LZ4HC_sequencePrice(llen, mlen);
+                 opt[mlen].mlen = mlen;
+                 opt[mlen].off = offset;
+                 opt[mlen].litlen = llen;
+                 opt[mlen].price = cost;
+                 DEBUGLOG(7, "rPos:%3i => price:%3i (matchlen=%i) -- initial setup",
+                             mlen, cost, mlen);
+         }   }
+         last_match_pos = firstMatch.len;
+         {   int addLit;
+             for (addLit = 1; addLit <= TRAILING_LITERALS; addLit ++) {
+                 opt[last_match_pos+addLit].mlen = 1; /* literal */
+                 opt[last_match_pos+addLit].off = 0;
+                 opt[last_match_pos+addLit].litlen = addLit;
+                 opt[last_match_pos+addLit].price = opt[last_match_pos].price + LZ4HC_literalsPrice(addLit);
+                 DEBUGLOG(7, "rPos:%3i => price:%3i (litlen=%i) -- initial setup",
+                             last_match_pos+addLit, opt[last_match_pos+addLit].price, addLit);
+         }   }
+
+         /* check further positions */
+         for (cur = 1; cur < last_match_pos; cur++) {
+             const BYTE* const curPtr = ip + cur;
+             LZ4HC_match_t newMatch;
+
+             if (curPtr > mflimit) break;
+             DEBUGLOG(7, "rPos:%u[%u] vs [%u]%u",
+                     cur, opt[cur].price, opt[cur+1].price, cur+1);
+             if (fullUpdate) {
+                 /* not useful to search here if next position has same (or lower) cost */
+                 if ( (opt[cur+1].price <= opt[cur].price)
+                   /* in some cases, next position has same cost, but cost rises sharply after, so a small match would still be beneficial */
+                   && (opt[cur+MINMATCH].price < opt[cur].price + 3/*min seq price*/) )
+                     continue;
+             } else {
+                 /* not useful to search here if next position has same (or lower) cost */
+                 if (opt[cur+1].price <= opt[cur].price) continue;
+             }
+
+             DEBUGLOG(7, "search at rPos:%u", cur);
+             if (fullUpdate)
+                 newMatch = LZ4HC_FindLongerMatch(ctx, curPtr, matchlimit, MINMATCH-1, nbSearches, dict, favorDecSpeed);
+             else
+                 /* only test matches of minimum length; slightly faster, but misses a few bytes */
+                 newMatch = LZ4HC_FindLongerMatch(ctx, curPtr, matchlimit, last_match_pos - cur, nbSearches, dict, favorDecSpeed);
+             if (!newMatch.len) continue;
+
+             if ( ((size_t)newMatch.len > sufficient_len)
+               || (newMatch.len + cur >= LZ4_OPT_NUM) ) {
+                 /* immediate encoding */
+                 best_mlen = newMatch.len;
+                 best_off = newMatch.off;
+                 last_match_pos = cur + 1;
+                 goto encode;
+             }
+
+             /* before match : set price with literals at beginning */
+             {   int const baseLitlen = opt[cur].litlen;
+                 int litlen;
+                 for (litlen = 1; litlen < MINMATCH; litlen++) {
+                     int const price = opt[cur].price - LZ4HC_literalsPrice(baseLitlen) + LZ4HC_literalsPrice(baseLitlen+litlen);
+                     int const pos = cur + litlen;
+                     if (price < opt[pos].price) {
+                         opt[pos].mlen = 1; /* literal */
+                         opt[pos].off = 0;
+                         opt[pos].litlen = baseLitlen+litlen;
+                         opt[pos].price = price;
+                         DEBUGLOG(7, "rPos:%3i => price:%3i (litlen=%i)",
+                                     pos, price, opt[pos].litlen);
+             }   }   }
+
+             /* set prices using match at position = cur */
+             {   int const matchML = newMatch.len;
+                 int ml = MINMATCH;
+
+                 assert(cur + newMatch.len < LZ4_OPT_NUM);
+                 for ( ; ml <= matchML ; ml++) {
+                     int const pos = cur + ml;
+                     int const offset = newMatch.off;
+                     int price;
+                     int ll;
+                     DEBUGLOG(7, "testing price rPos %i (last_match_pos=%i)",
+                                 pos, last_match_pos);
+                     if (opt[cur].mlen == 1) {
+                         ll = opt[cur].litlen;
+                         price = ((cur > ll) ? opt[cur - ll].price : 0)
+                               + LZ4HC_sequencePrice(ll, ml);
+                     } else {
+                         ll = 0;
+                         price = opt[cur].price + LZ4HC_sequencePrice(0, ml);
+                     }
+
+                    assert((U32)favorDecSpeed <= 1);
+                     if (pos > last_match_pos+TRAILING_LITERALS
+                      || price <= opt[pos].price - (int)favorDecSpeed) {
+                         DEBUGLOG(7, "rPos:%3i => price:%3i (matchlen=%i)",
+                                     pos, price, ml);
+                         assert(pos < LZ4_OPT_NUM);
+                         if ( (ml == matchML)  /* last pos of last match */
+                           && (last_match_pos < pos) )
+                             last_match_pos = pos;
+                         opt[pos].mlen = ml;
+                         opt[pos].off = offset;
+                         opt[pos].litlen = ll;
+                         opt[pos].price = price;
+             }   }   }
+             /* complete following positions with literals */
+             {   int addLit;
+                 for (addLit = 1; addLit <= TRAILING_LITERALS; addLit ++) {
+                     opt[last_match_pos+addLit].mlen = 1; /* literal */
+                     opt[last_match_pos+addLit].off = 0;
+                     opt[last_match_pos+addLit].litlen = addLit;
+                     opt[last_match_pos+addLit].price = opt[last_match_pos].price + LZ4HC_literalsPrice(addLit);
+                     DEBUGLOG(7, "rPos:%3i => price:%3i (litlen=%i)", last_match_pos+addLit, opt[last_match_pos+addLit].price, addLit);
+             }   }
+         }  /* for (cur = 1; cur <= last_match_pos; cur++) */
+
+         assert(last_match_pos < LZ4_OPT_NUM + TRAILING_LITERALS);
+         best_mlen = opt[last_match_pos].mlen;
+         best_off = opt[last_match_pos].off;
+         cur = last_match_pos - best_mlen;
+
+encode: /* cur, last_match_pos, best_mlen, best_off must be set */
+         assert(cur < LZ4_OPT_NUM);
+         assert(last_match_pos >= 1);  /* == 1 when only one candidate */
+         DEBUGLOG(6, "reverse traversal, looking for shortest path (last_match_pos=%i)", last_match_pos);
+         {   int candidate_pos = cur;
+             int selected_matchLength = best_mlen;
+             int selected_offset = best_off;
+             while (1) {  /* from end to beginning */
+                 int const next_matchLength = opt[candidate_pos].mlen;  /* can be 1, means literal */
+                 int const next_offset = opt[candidate_pos].off;
+                 DEBUGLOG(7, "pos %i: sequence length %i", candidate_pos, selected_matchLength);
+                 opt[candidate_pos].mlen = selected_matchLength;
+                 opt[candidate_pos].off = selected_offset;
+                 selected_matchLength = next_matchLength;
+                 selected_offset = next_offset;
+                 if (next_matchLength > candidate_pos) break; /* last match elected, first match to encode */
+                 assert(next_matchLength > 0);  /* can be 1, means literal */
+                 candidate_pos -= next_matchLength;
+         }   }
+
+         /* encode all recorded sequences in order */
+         {   int rPos = 0;  /* relative position (to ip) */
+             while (rPos < last_match_pos) {
+                 int const ml = opt[rPos].mlen;
+                 int const offset = opt[rPos].off;
+                 if (ml == 1) { ip++; rPos++; continue; }  /* literal; note: can end up with several literals, in which case, skip them */
+                 rPos += ml;
+                 assert(ml >= MINMATCH);
+                 assert((offset >= 1) && (offset <= LZ4_DISTANCE_MAX));
+                 opSaved = op;
+                 if ( LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), ml, offset, limit, oend) ) {  /* updates ip, op and anchor */
+                     ovml = ml;
+                     ovoff = offset;
+                     goto _dest_overflow;
+         }   }   }
+     }  /* while (ip <= mflimit) */
+
+_last_literals:
+     /* Encode Last Literals */
+     {   size_t lastRunSize = (size_t)(iend - anchor);  /* literals */
+         size_t llAdd = (lastRunSize + 255 - RUN_MASK) / 255;
+         size_t const totalSize = 1 + llAdd + lastRunSize;
+         if (limit == fillOutput) oend += LASTLITERALS;  /* restore correct value */
+         if (limit && (op + totalSize > oend)) {
+             if (limit == limitedOutput) { /* Check output limit */
+                retval = 0;
+                goto _return_label;
+             }
+             /* adapt lastRunSize to fill 'dst' */
+             lastRunSize  = (size_t)(oend - op) - 1 /*token*/;
+             llAdd = (lastRunSize + 256 - RUN_MASK) / 256;
+             lastRunSize -= llAdd;
+         }
+         DEBUGLOG(6, "Final literal run : %i literals", (int)lastRunSize);
+         ip = anchor + lastRunSize; /* can be != iend if limit==fillOutput */
+
+         if (lastRunSize >= RUN_MASK) {
+             size_t accumulator = lastRunSize - RUN_MASK;
+             *op++ = (RUN_MASK << ML_BITS);
+             for(; accumulator >= 255 ; accumulator -= 255) *op++ = 255;
+             *op++ = (BYTE) accumulator;
+         } else {
+             *op++ = (BYTE)(lastRunSize << ML_BITS);
+         }
+         LZ4_memcpy(op, anchor, lastRunSize);
+         op += lastRunSize;
+     }
+
+     /* End */
+     *srcSizePtr = (int) (((const char*)ip) - source);
+     retval = (int) ((char*)op-dst);
+     goto _return_label;
+
+_dest_overflow:
+if (limit == fillOutput) {
+     /* Assumption : ip, anchor, ovml and ovref must be set correctly */
+     size_t const ll = (size_t)(ip - anchor);
+     size_t const ll_addbytes = (ll + 240) / 255;
+     size_t const ll_totalCost = 1 + ll_addbytes + ll;
+     BYTE* const maxLitPos = oend - 3; /* 2 for offset, 1 for token */
+     DEBUGLOG(6, "Last sequence overflowing (only %i bytes remaining)", (int)(oend-1-opSaved));
+     op = opSaved;  /* restore correct out pointer */
+     if (op + ll_totalCost <= maxLitPos) {
+         /* ll validated; now adjust match length */
+         size_t const bytesLeftForMl = (size_t)(maxLitPos - (op+ll_totalCost));
+         size_t const maxMlSize = MINMATCH + (ML_MASK-1) + (bytesLeftForMl * 255);
+         assert(maxMlSize < INT_MAX); assert(ovml >= 0);
+         if ((size_t)ovml > maxMlSize) ovml = (int)maxMlSize;
+         if ((oend + LASTLITERALS) - (op + ll_totalCost + 2) - 1 + ovml >= MFLIMIT) {
+             DEBUGLOG(6, "Space to end : %i + ml (%i)", (int)((oend + LASTLITERALS) - (op + ll_totalCost + 2) - 1), ovml);
+             DEBUGLOG(6, "Before : ip = %p, anchor = %p", ip, anchor);
+             LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), ovml, ovoff, notLimited, oend);
+             DEBUGLOG(6, "After : ip = %p, anchor = %p", ip, anchor);
+     }   }
+     goto _last_literals;
+}
+_return_label:
+#if defined(LZ4HC_HEAPMODE) && LZ4HC_HEAPMODE==1
+     if (opt) FREEMEM(opt);
+#endif
+     return retval;
+}
+
+
+/***************************************************
+*  Deprecated Functions
+***************************************************/
+
+/* These functions currently generate deprecation warnings */
+
+/* Wrappers for deprecated compression functions */
+int LZ4_compressHC(const char* src, char* dst, int srcSize) { return LZ4_compress_HC (src, dst, srcSize, LZ4_compressBound(srcSize), 0); }
+int LZ4_compressHC_limitedOutput(const char* src, char* dst, int srcSize, int maxDstSize) { return LZ4_compress_HC(src, dst, srcSize, maxDstSize, 0); }
+int LZ4_compressHC2(const char* src, char* dst, int srcSize, int cLevel) { return LZ4_compress_HC (src, dst, srcSize, LZ4_compressBound(srcSize), cLevel); }
+int LZ4_compressHC2_limitedOutput(const char* src, char* dst, int srcSize, int maxDstSize, int cLevel) { return LZ4_compress_HC(src, dst, srcSize, maxDstSize, cLevel); }
+int LZ4_compressHC_withStateHC (void* state, const char* src, char* dst, int srcSize) { return LZ4_compress_HC_extStateHC (state, src, dst, srcSize, LZ4_compressBound(srcSize), 0); }
+int LZ4_compressHC_limitedOutput_withStateHC (void* state, const char* src, char* dst, int srcSize, int maxDstSize) { return LZ4_compress_HC_extStateHC (state, src, dst, srcSize, maxDstSize, 0); }
+int LZ4_compressHC2_withStateHC (void* state, const char* src, char* dst, int srcSize, int cLevel) { return LZ4_compress_HC_extStateHC(state, src, dst, srcSize, LZ4_compressBound(srcSize), cLevel); }
+int LZ4_compressHC2_limitedOutput_withStateHC (void* state, const char* src, char* dst, int srcSize, int maxDstSize, int cLevel) { return LZ4_compress_HC_extStateHC(state, src, dst, srcSize, maxDstSize, cLevel); }
+int LZ4_compressHC_continue (LZ4_streamHC_t* ctx, const char* src, char* dst, int srcSize) { return LZ4_compress_HC_continue (ctx, src, dst, srcSize, LZ4_compressBound(srcSize)); }
+int LZ4_compressHC_limitedOutput_continue (LZ4_streamHC_t* ctx, const char* src, char* dst, int srcSize, int maxDstSize) { return LZ4_compress_HC_continue (ctx, src, dst, srcSize, maxDstSize); }
+
+
+/* Deprecated streaming functions */
+int LZ4_sizeofStreamStateHC(void) { return sizeof(LZ4_streamHC_t); }
+
+/* state is presumed correctly sized, aka >= sizeof(LZ4_streamHC_t)
+ * @return : 0 on success, !=0 if error */
+int LZ4_resetStreamStateHC(void* state, char* inputBuffer)
+{
+    LZ4_streamHC_t* const hc4 = LZ4_initStreamHC(state, sizeof(*hc4));
+    if (hc4 == NULL) return 1;   /* init failed */
+    LZ4HC_init_internal (&hc4->internal_donotuse, (const BYTE*)inputBuffer);
+    return 0;
+}
+
+#if !defined(LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION)
+void* LZ4_createHC (const char* inputBuffer)
+{
+    LZ4_streamHC_t* const hc4 = LZ4_createStreamHC();
+    if (hc4 == NULL) return NULL;   /* not enough memory */
+    LZ4HC_init_internal (&hc4->internal_donotuse, (const BYTE*)inputBuffer);
+    return hc4;
+}
+
+int LZ4_freeHC (void* LZ4HC_Data)
+{
+    if (!LZ4HC_Data) return 0;  /* support free on NULL */
+    FREEMEM(LZ4HC_Data);
+    return 0;
+}
+#endif
+
+int LZ4_compressHC2_continue (void* LZ4HC_Data, const char* src, char* dst, int srcSize, int cLevel)
+{
+    return LZ4HC_compress_generic (&((LZ4_streamHC_t*)LZ4HC_Data)->internal_donotuse, src, dst, &srcSize, 0, cLevel, notLimited);
+}
+
+int LZ4_compressHC2_limitedOutput_continue (void* LZ4HC_Data, const char* src, char* dst, int srcSize, int dstCapacity, int cLevel)
+{
+    return LZ4HC_compress_generic (&((LZ4_streamHC_t*)LZ4HC_Data)->internal_donotuse, src, dst, &srcSize, dstCapacity, cLevel, limitedOutput);
+}
+
+char* LZ4_slideInputBufferHC(void* LZ4HC_Data)
+{
+    LZ4HC_CCtx_internal* const s = &((LZ4_streamHC_t*)LZ4HC_Data)->internal_donotuse;
+    const BYTE* const bufferStart = s->prefixStart - s->dictLimit + s->lowLimit;
+    LZ4_resetStreamHC_fast((LZ4_streamHC_t*)LZ4HC_Data, s->compressionLevel);
+    /* ugly conversion trick, required to evade (const char*) -> (char*) cast-qual warning :( */
+    return (char*)(uptrval)bufferStart;
+}
diff --git a/lib/lz4/upstream/lz4hc.h b/lib/lz4/upstream/lz4hc.h
new file mode 100644
index 0000000000000000000000000000000000000000..992bc8cdd7703e7ff49b01137a00bf6f004683a6
--- /dev/null
+++ b/lib/lz4/upstream/lz4hc.h
@@ -0,0 +1,414 @@
+/*
+   LZ4 HC - High Compression Mode of LZ4
+   Header File
+   Copyright (C) 2011-2020, Yann Collet.
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - LZ4 source repository : https://github.com/lz4/lz4
+   - LZ4 public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+#ifndef LZ4_HC_H_19834876238432
+#define LZ4_HC_H_19834876238432
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* --- Dependency --- */
+/* note : lz4hc requires lz4.h/lz4.c for compilation */
+#include "lz4.h"   /* stddef, LZ4LIB_API, LZ4_DEPRECATED */
+
+
+/* --- Useful constants --- */
+#define LZ4HC_CLEVEL_MIN         2
+#define LZ4HC_CLEVEL_DEFAULT     9
+#define LZ4HC_CLEVEL_OPT_MIN    10
+#define LZ4HC_CLEVEL_MAX        12
+
+
+/*-************************************
+ *  Block Compression
+ **************************************/
+/*! LZ4_compress_HC() :
+ *  Compress data from `src` into `dst`, using the powerful but slower "HC" algorithm.
+ * `dst` must be already allocated.
+ *  Compression is guaranteed to succeed if `dstCapacity >= LZ4_compressBound(srcSize)` (see "lz4.h")
+ *  Max supported `srcSize` value is LZ4_MAX_INPUT_SIZE (see "lz4.h")
+ * `compressionLevel` : any value between 1 and LZ4HC_CLEVEL_MAX will work.
+ *                      Values > LZ4HC_CLEVEL_MAX behave the same as LZ4HC_CLEVEL_MAX.
+ * @return : the number of bytes written into 'dst'
+ *           or 0 if compression fails.
+ */
+LZ4LIB_API int LZ4_compress_HC (const char* src, char* dst, int srcSize, int dstCapacity, int compressionLevel);
+
+
+/* Note :
+ *   Decompression functions are provided within "lz4.h" (BSD license)
+ */
+
+
+/*! LZ4_compress_HC_extStateHC() :
+ *  Same as LZ4_compress_HC(), but using an externally allocated memory segment for `state`.
+ * `state` size is provided by LZ4_sizeofStateHC().
+ *  Memory segment must be aligned on 8-bytes boundaries (which a normal malloc() should do properly).
+ */
+LZ4LIB_API int LZ4_sizeofStateHC(void);
+LZ4LIB_API int LZ4_compress_HC_extStateHC(void* stateHC, const char* src, char* dst, int srcSize, int maxDstSize, int compressionLevel);
+
+
+/*! LZ4_compress_HC_destSize() : v1.9.0+
+ *  Will compress as much data as possible from `src`
+ *  to fit into `targetDstSize` budget.
+ *  Result is provided in 2 parts :
+ * @return : the number of bytes written into 'dst' (necessarily <= targetDstSize)
+ *           or 0 if compression fails.
+ * `srcSizePtr` : on success, *srcSizePtr is updated to indicate how much bytes were read from `src`
+ */
+LZ4LIB_API int LZ4_compress_HC_destSize(void* stateHC,
+                                  const char* src, char* dst,
+                                        int* srcSizePtr, int targetDstSize,
+                                        int compressionLevel);
+
+
+/*-************************************
+ *  Streaming Compression
+ *  Bufferless synchronous API
+ **************************************/
+ typedef union LZ4_streamHC_u LZ4_streamHC_t;   /* incomplete type (defined later) */
+
+/*! LZ4_createStreamHC() and LZ4_freeStreamHC() :
+ *  These functions create and release memory for LZ4 HC streaming state.
+ *  Newly created states are automatically initialized.
+ *  A same state can be used multiple times consecutively,
+ *  starting with LZ4_resetStreamHC_fast() to start a new stream of blocks.
+ */
+LZ4LIB_API LZ4_streamHC_t* LZ4_createStreamHC(void);
+LZ4LIB_API int             LZ4_freeStreamHC (LZ4_streamHC_t* streamHCPtr);
+
+/*
+  These functions compress data in successive blocks of any size,
+  using previous blocks as dictionary, to improve compression ratio.
+  One key assumption is that previous blocks (up to 64 KB) remain read-accessible while compressing next blocks.
+  There is an exception for ring buffers, which can be smaller than 64 KB.
+  Ring-buffer scenario is automatically detected and handled within LZ4_compress_HC_continue().
+
+  Before starting compression, state must be allocated and properly initialized.
+  LZ4_createStreamHC() does both, though compression level is set to LZ4HC_CLEVEL_DEFAULT.
+
+  Selecting the compression level can be done with LZ4_resetStreamHC_fast() (starts a new stream)
+  or LZ4_setCompressionLevel() (anytime, between blocks in the same stream) (experimental).
+  LZ4_resetStreamHC_fast() only works on states which have been properly initialized at least once,
+  which is automatically the case when state is created using LZ4_createStreamHC().
+
+  After reset, a first "fictional block" can be designated as initial dictionary,
+  using LZ4_loadDictHC() (Optional).
+  Note: In order for LZ4_loadDictHC() to create the correct data structure,
+  it is essential to set the compression level _before_ loading the dictionary.
+
+  Invoke LZ4_compress_HC_continue() to compress each successive block.
+  The number of blocks is unlimited.
+  Previous input blocks, including initial dictionary when present,
+  must remain accessible and unmodified during compression.
+
+  It's allowed to update compression level anytime between blocks,
+  using LZ4_setCompressionLevel() (experimental).
+
+ @dst buffer should be sized to handle worst case scenarios
+  (see LZ4_compressBound(), it ensures compression success).
+  In case of failure, the API does not guarantee recovery,
+  so the state _must_ be reset.
+  To ensure compression success
+  whenever @dst buffer size cannot be made >= LZ4_compressBound(),
+  consider using LZ4_compress_HC_continue_destSize().
+
+  Whenever previous input blocks can't be preserved unmodified in-place during compression of next blocks,
+  it's possible to copy the last blocks into a more stable memory space, using LZ4_saveDictHC().
+  Return value of LZ4_saveDictHC() is the size of dictionary effectively saved into 'safeBuffer' (<= 64 KB)
+
+  After completing a streaming compression,
+  it's possible to start a new stream of blocks, using the same LZ4_streamHC_t state,
+  just by resetting it, using LZ4_resetStreamHC_fast().
+*/
+
+LZ4LIB_API void LZ4_resetStreamHC_fast(LZ4_streamHC_t* streamHCPtr, int compressionLevel);   /* v1.9.0+ */
+LZ4LIB_API int  LZ4_loadDictHC (LZ4_streamHC_t* streamHCPtr, const char* dictionary, int dictSize);
+
+LZ4LIB_API int LZ4_compress_HC_continue (LZ4_streamHC_t* streamHCPtr,
+                                   const char* src, char* dst,
+                                         int srcSize, int maxDstSize);
+
+/*! LZ4_compress_HC_continue_destSize() : v1.9.0+
+ *  Similar to LZ4_compress_HC_continue(),
+ *  but will read as much data as possible from `src`
+ *  to fit into `targetDstSize` budget.
+ *  Result is provided into 2 parts :
+ * @return : the number of bytes written into 'dst' (necessarily <= targetDstSize)
+ *           or 0 if compression fails.
+ * `srcSizePtr` : on success, *srcSizePtr will be updated to indicate how much bytes were read from `src`.
+ *           Note that this function may not consume the entire input.
+ */
+LZ4LIB_API int LZ4_compress_HC_continue_destSize(LZ4_streamHC_t* LZ4_streamHCPtr,
+                                           const char* src, char* dst,
+                                                 int* srcSizePtr, int targetDstSize);
+
+LZ4LIB_API int LZ4_saveDictHC (LZ4_streamHC_t* streamHCPtr, char* safeBuffer, int maxDictSize);
+
+
+/*! LZ4_attach_HC_dictionary() : stable since v1.10.0
+ *  This API allows for the efficient re-use of a static dictionary many times.
+ *
+ *  Rather than re-loading the dictionary buffer into a working context before
+ *  each compression, or copying a pre-loaded dictionary's LZ4_streamHC_t into a
+ *  working LZ4_streamHC_t, this function introduces a no-copy setup mechanism,
+ *  in which the working stream references the dictionary stream in-place.
+ *
+ *  Several assumptions are made about the state of the dictionary stream.
+ *  Currently, only streams which have been prepared by LZ4_loadDictHC() should
+ *  be expected to work.
+ *
+ *  Alternatively, the provided dictionary stream pointer may be NULL, in which
+ *  case any existing dictionary stream is unset.
+ *
+ *  A dictionary should only be attached to a stream without any history (i.e.,
+ *  a stream that has just been reset).
+ *
+ *  The dictionary will remain attached to the working stream only for the
+ *  current stream session. Calls to LZ4_resetStreamHC(_fast) will remove the
+ *  dictionary context association from the working stream. The dictionary
+ *  stream (and source buffer) must remain in-place / accessible / unchanged
+ *  through the lifetime of the stream session.
+ */
+LZ4LIB_API void
+LZ4_attach_HC_dictionary(LZ4_streamHC_t* working_stream,
+                   const LZ4_streamHC_t* dictionary_stream);
+
+
+/*^**********************************************
+ * !!!!!!   STATIC LINKING ONLY   !!!!!!
+ ***********************************************/
+
+/*-******************************************************************
+ * PRIVATE DEFINITIONS :
+ * Do not use these definitions directly.
+ * They are merely exposed to allow static allocation of `LZ4_streamHC_t`.
+ * Declare an `LZ4_streamHC_t` directly, rather than any type below.
+ * Even then, only do so in the context of static linking, as definitions may change between versions.
+ ********************************************************************/
+
+#define LZ4HC_DICTIONARY_LOGSIZE 16
+#define LZ4HC_MAXD (1<<LZ4HC_DICTIONARY_LOGSIZE)
+#define LZ4HC_MAXD_MASK (LZ4HC_MAXD - 1)
+
+#define LZ4HC_HASH_LOG 15
+#define LZ4HC_HASHTABLESIZE (1 << LZ4HC_HASH_LOG)
+#define LZ4HC_HASH_MASK (LZ4HC_HASHTABLESIZE - 1)
+
+
+/* Never ever use these definitions directly !
+ * Declare or allocate an LZ4_streamHC_t instead.
+**/
+typedef struct LZ4HC_CCtx_internal LZ4HC_CCtx_internal;
+struct LZ4HC_CCtx_internal
+{
+    LZ4_u32 hashTable[LZ4HC_HASHTABLESIZE];
+    LZ4_u16 chainTable[LZ4HC_MAXD];
+    const LZ4_byte* end;     /* next block here to continue on current prefix */
+    const LZ4_byte* prefixStart;  /* Indexes relative to this position */
+    const LZ4_byte* dictStart; /* alternate reference for extDict */
+    LZ4_u32 dictLimit;       /* below that point, need extDict */
+    LZ4_u32 lowLimit;        /* below that point, no more history */
+    LZ4_u32 nextToUpdate;    /* index from which to continue dictionary update */
+    short   compressionLevel;
+    LZ4_i8  favorDecSpeed;   /* favor decompression speed if this flag set,
+                                otherwise, favor compression ratio */
+    LZ4_i8  dirty;           /* stream has to be fully reset if this flag is set */
+    const LZ4HC_CCtx_internal* dictCtx;
+};
+
+#define LZ4_STREAMHC_MINSIZE  262200  /* static size, for inter-version compatibility */
+union LZ4_streamHC_u {
+    char minStateSize[LZ4_STREAMHC_MINSIZE];
+    LZ4HC_CCtx_internal internal_donotuse;
+}; /* previously typedef'd to LZ4_streamHC_t */
+
+/* LZ4_streamHC_t :
+ * This structure allows static allocation of LZ4 HC streaming state.
+ * This can be used to allocate statically on stack, or as part of a larger structure.
+ *
+ * Such state **must** be initialized using LZ4_initStreamHC() before first use.
+ *
+ * Note that invoking LZ4_initStreamHC() is not required when
+ * the state was created using LZ4_createStreamHC() (which is recommended).
+ * Using the normal builder, a newly created state is automatically initialized.
+ *
+ * Static allocation shall only be used in combination with static linking.
+ */
+
+/* LZ4_initStreamHC() : v1.9.0+
+ * Required before first use of a statically allocated LZ4_streamHC_t.
+ * Before v1.9.0 : use LZ4_resetStreamHC() instead
+ */
+LZ4LIB_API LZ4_streamHC_t* LZ4_initStreamHC(void* buffer, size_t size);
+
+
+/*-************************************
+*  Deprecated Functions
+**************************************/
+/* see lz4.h LZ4_DISABLE_DEPRECATE_WARNINGS to turn off deprecation warnings */
+
+/* deprecated compression functions */
+LZ4_DEPRECATED("use LZ4_compress_HC() instead") LZ4LIB_API int LZ4_compressHC               (const char* source, char* dest, int inputSize);
+LZ4_DEPRECATED("use LZ4_compress_HC() instead") LZ4LIB_API int LZ4_compressHC_limitedOutput (const char* source, char* dest, int inputSize, int maxOutputSize);
+LZ4_DEPRECATED("use LZ4_compress_HC() instead") LZ4LIB_API int LZ4_compressHC2              (const char* source, char* dest, int inputSize, int compressionLevel);
+LZ4_DEPRECATED("use LZ4_compress_HC() instead") LZ4LIB_API int LZ4_compressHC2_limitedOutput(const char* source, char* dest, int inputSize, int maxOutputSize, int compressionLevel);
+LZ4_DEPRECATED("use LZ4_compress_HC_extStateHC() instead") LZ4LIB_API int LZ4_compressHC_withStateHC               (void* state, const char* source, char* dest, int inputSize);
+LZ4_DEPRECATED("use LZ4_compress_HC_extStateHC() instead") LZ4LIB_API int LZ4_compressHC_limitedOutput_withStateHC (void* state, const char* source, char* dest, int inputSize, int maxOutputSize);
+LZ4_DEPRECATED("use LZ4_compress_HC_extStateHC() instead") LZ4LIB_API int LZ4_compressHC2_withStateHC              (void* state, const char* source, char* dest, int inputSize, int compressionLevel);
+LZ4_DEPRECATED("use LZ4_compress_HC_extStateHC() instead") LZ4LIB_API int LZ4_compressHC2_limitedOutput_withStateHC(void* state, const char* source, char* dest, int inputSize, int maxOutputSize, int compressionLevel);
+LZ4_DEPRECATED("use LZ4_compress_HC_continue() instead") LZ4LIB_API int LZ4_compressHC_continue               (LZ4_streamHC_t* LZ4_streamHCPtr, const char* source, char* dest, int inputSize);
+LZ4_DEPRECATED("use LZ4_compress_HC_continue() instead") LZ4LIB_API int LZ4_compressHC_limitedOutput_continue (LZ4_streamHC_t* LZ4_streamHCPtr, const char* source, char* dest, int inputSize, int maxOutputSize);
+
+/* Obsolete streaming functions; degraded functionality; do not use!
+ *
+ * In order to perform streaming compression, these functions depended on data
+ * that is no longer tracked in the state. They have been preserved as well as
+ * possible: using them will still produce a correct output. However, use of
+ * LZ4_slideInputBufferHC() will truncate the history of the stream, rather
+ * than preserve a window-sized chunk of history.
+ */
+#if !defined(LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION)
+LZ4_DEPRECATED("use LZ4_createStreamHC() instead") LZ4LIB_API void* LZ4_createHC (const char* inputBuffer);
+LZ4_DEPRECATED("use LZ4_freeStreamHC() instead") LZ4LIB_API   int   LZ4_freeHC (void* LZ4HC_Data);
+#endif
+LZ4_DEPRECATED("use LZ4_saveDictHC() instead") LZ4LIB_API     char* LZ4_slideInputBufferHC (void* LZ4HC_Data);
+LZ4_DEPRECATED("use LZ4_compress_HC_continue() instead") LZ4LIB_API int LZ4_compressHC2_continue               (void* LZ4HC_Data, const char* source, char* dest, int inputSize, int compressionLevel);
+LZ4_DEPRECATED("use LZ4_compress_HC_continue() instead") LZ4LIB_API int LZ4_compressHC2_limitedOutput_continue (void* LZ4HC_Data, const char* source, char* dest, int inputSize, int maxOutputSize, int compressionLevel);
+LZ4_DEPRECATED("use LZ4_createStreamHC() instead") LZ4LIB_API int   LZ4_sizeofStreamStateHC(void);
+LZ4_DEPRECATED("use LZ4_initStreamHC() instead") LZ4LIB_API  int   LZ4_resetStreamStateHC(void* state, char* inputBuffer);
+
+
+/* LZ4_resetStreamHC() is now replaced by LZ4_initStreamHC().
+ * The intention is to emphasize the difference with LZ4_resetStreamHC_fast(),
+ * which is now the recommended function to start a new stream of blocks,
+ * but cannot be used to initialize a memory segment containing arbitrary garbage data.
+ *
+ * It is recommended to switch to LZ4_initStreamHC().
+ * LZ4_resetStreamHC() will generate deprecation warnings in a future version.
+ */
+LZ4LIB_API void LZ4_resetStreamHC (LZ4_streamHC_t* streamHCPtr, int compressionLevel);
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* LZ4_HC_H_19834876238432 */
+
+
+/*-**************************************************
+ * !!!!!     STATIC LINKING ONLY     !!!!!
+ * Following definitions are considered experimental.
+ * They should not be linked from DLL,
+ * as there is no guarantee of API stability yet.
+ * Prototypes will be promoted to "stable" status
+ * after successful usage in real-life scenarios.
+ ***************************************************/
+#ifdef LZ4_HC_STATIC_LINKING_ONLY   /* protection macro */
+#ifndef LZ4_HC_SLO_098092834
+#define LZ4_HC_SLO_098092834
+
+#define LZ4_STATIC_LINKING_ONLY   /* LZ4LIB_STATIC_API */
+#include "lz4.h"
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/*! LZ4_setCompressionLevel() : v1.8.0+ (experimental)
+ *  It's possible to change compression level
+ *  between successive invocations of LZ4_compress_HC_continue*()
+ *  for dynamic adaptation.
+ */
+LZ4LIB_STATIC_API void LZ4_setCompressionLevel(
+    LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel);
+
+/*! LZ4_favorDecompressionSpeed() : v1.8.2+ (experimental)
+ *  Opt. Parser will favor decompression speed over compression ratio.
+ *  Only applicable to levels >= LZ4HC_CLEVEL_OPT_MIN.
+ */
+LZ4LIB_STATIC_API void LZ4_favorDecompressionSpeed(
+    LZ4_streamHC_t* LZ4_streamHCPtr, int favor);
+
+/*! LZ4_resetStreamHC_fast() : v1.9.0+
+ *  When an LZ4_streamHC_t is known to be in a internally coherent state,
+ *  it can often be prepared for a new compression with almost no work, only
+ *  sometimes falling back to the full, expensive reset that is always required
+ *  when the stream is in an indeterminate state (i.e., the reset performed by
+ *  LZ4_resetStreamHC()).
+ *
+ *  LZ4_streamHCs are guaranteed to be in a valid state when:
+ *  - returned from LZ4_createStreamHC()
+ *  - reset by LZ4_resetStreamHC()
+ *  - memset(stream, 0, sizeof(LZ4_streamHC_t))
+ *  - the stream was in a valid state and was reset by LZ4_resetStreamHC_fast()
+ *  - the stream was in a valid state and was then used in any compression call
+ *    that returned success
+ *  - the stream was in an indeterminate state and was used in a compression
+ *    call that fully reset the state (LZ4_compress_HC_extStateHC()) and that
+ *    returned success
+ *
+ *  Note:
+ *  A stream that was last used in a compression call that returned an error
+ *  may be passed to this function. However, it will be fully reset, which will
+ *  clear any existing history and settings from the context.
+ */
+LZ4LIB_STATIC_API void LZ4_resetStreamHC_fast(
+    LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel);
+
+/*! LZ4_compress_HC_extStateHC_fastReset() :
+ *  A variant of LZ4_compress_HC_extStateHC().
+ *
+ *  Using this variant avoids an expensive initialization step. It is only safe
+ *  to call if the state buffer is known to be correctly initialized already
+ *  (see above comment on LZ4_resetStreamHC_fast() for a definition of
+ *  "correctly initialized"). From a high level, the difference is that this
+ *  function initializes the provided state with a call to
+ *  LZ4_resetStreamHC_fast() while LZ4_compress_HC_extStateHC() starts with a
+ *  call to LZ4_resetStreamHC().
+ */
+LZ4LIB_STATIC_API int LZ4_compress_HC_extStateHC_fastReset (
+    void* state,
+    const char* src, char* dst,
+    int srcSize, int dstCapacity,
+    int compressionLevel);
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif   /* LZ4_HC_SLO_098092834 */
+#endif   /* LZ4_HC_STATIC_LINKING_ONLY */

-- 
2.34.1

[f2fs-dev] [PATCH v2 04/15] lib/lz4: import upstream LZ4 sources verbatim

[Michal Wilczynski]

Add the four upstream files under lib/lz4/upstream/ without modification.
Nothing builds them yet; a later patch adds the build environment.

Copied without modification from the v1.10.0 release tag of
https://github.com/lz4/lz4, tagged 2024-07-21.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 lib/lz4/upstream/lz4.c   | 2829 ++++++++++++++++++++++++++++++++++++++++++++++
 lib/lz4/upstream/lz4.h   |  884 +++++++++++++++
 lib/lz4/upstream/lz4hc.c | 2192 +++++++++++++++++++++++++++++++++++
 lib/lz4/upstream/lz4hc.h |  414 +++++++
 4 files changed, 6319 insertions(+)

diff --git a/lib/lz4/upstream/lz4.c b/lib/lz4/upstream/lz4.c
new file mode 100644
index 0000000000000000000000000000000000000000..a2f7abee19fb9a5c768f2a6c266acf5b571f0855
--- /dev/null
+++ b/lib/lz4/upstream/lz4.c
@@ -0,0 +1,2829 @@
+/*
+   LZ4 - Fast LZ compression algorithm
+   Copyright (C) 2011-2023, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+    - LZ4 homepage : http://www.lz4.org
+    - LZ4 source repository : https://github.com/lz4/lz4
+*/
+
+/*-************************************
+*  Tuning parameters
+**************************************/
+/*
+ * LZ4_HEAPMODE :
+ * Select how stateless compression functions like `LZ4_compress_default()`
+ * allocate memory for their hash table,
+ * in memory stack (0:default, fastest), or in memory heap (1:requires malloc()).
+ */
+#ifndef LZ4_HEAPMODE
+#  define LZ4_HEAPMODE 0
+#endif
+
+/*
+ * LZ4_ACCELERATION_DEFAULT :
+ * Select "acceleration" for LZ4_compress_fast() when parameter value <= 0
+ */
+#define LZ4_ACCELERATION_DEFAULT 1
+/*
+ * LZ4_ACCELERATION_MAX :
+ * Any "acceleration" value higher than this threshold
+ * get treated as LZ4_ACCELERATION_MAX instead (fix #876)
+ */
+#define LZ4_ACCELERATION_MAX 65537
+
+
+/*-************************************
+*  CPU Feature Detection
+**************************************/
+/* LZ4_FORCE_MEMORY_ACCESS
+ * By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable.
+ * Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal.
+ * The below switch allow to select different access method for improved performance.
+ * Method 0 (default) : use `memcpy()`. Safe and portable.
+ * Method 1 : `__packed` statement. It depends on compiler extension (ie, not portable).
+ *            This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`.
+ * Method 2 : direct access. This method is portable but violate C standard.
+ *            It can generate buggy code on targets which assembly generation depends on alignment.
+ *            But in some circumstances, it's the only known way to get the most performance (ie GCC + ARMv6)
+ * See https://fastcompression.blogspot.fr/2015/08/accessing-unaligned-memory.html for details.
+ * Prefer these methods in priority order (0 > 1 > 2)
+ */
+#ifndef LZ4_FORCE_MEMORY_ACCESS   /* can be defined externally */
+#  if defined(__GNUC__) && \
+  ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) \
+  || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
+#    define LZ4_FORCE_MEMORY_ACCESS 2
+#  elif (defined(__INTEL_COMPILER) && !defined(_WIN32)) || defined(__GNUC__) || defined(_MSC_VER)
+#    define LZ4_FORCE_MEMORY_ACCESS 1
+#  endif
+#endif
+
+/*
+ * LZ4_FORCE_SW_BITCOUNT
+ * Define this parameter if your target system or compiler does not support hardware bit count
+ */
+#if defined(_MSC_VER) && defined(_WIN32_WCE)   /* Visual Studio for WinCE doesn't support Hardware bit count */
+#  undef  LZ4_FORCE_SW_BITCOUNT  /* avoid double def */
+#  define LZ4_FORCE_SW_BITCOUNT
+#endif
+
+
+
+/*-************************************
+*  Dependency
+**************************************/
+/*
+ * LZ4_SRC_INCLUDED:
+ * Amalgamation flag, whether lz4.c is included
+ */
+#ifndef LZ4_SRC_INCLUDED
+#  define LZ4_SRC_INCLUDED 1
+#endif
+
+#ifndef LZ4_DISABLE_DEPRECATE_WARNINGS
+#  define LZ4_DISABLE_DEPRECATE_WARNINGS /* due to LZ4_decompress_safe_withPrefix64k */
+#endif
+
+#ifndef LZ4_STATIC_LINKING_ONLY
+#  define LZ4_STATIC_LINKING_ONLY
+#endif
+#include "lz4.h"
+/* see also "memory routines" below */
+
+
+/*-************************************
+*  Compiler Options
+**************************************/
+#if defined(_MSC_VER) && (_MSC_VER >= 1400)  /* Visual Studio 2005+ */
+#  include <intrin.h>               /* only present in VS2005+ */
+#  pragma warning(disable : 4127)   /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 6237)   /* disable: C6237: conditional expression is always 0 */
+#  pragma warning(disable : 6239)   /* disable: C6239: (<non-zero constant> && <expression>) always evaluates to the result of <expression> */
+#  pragma warning(disable : 6240)   /* disable: C6240: (<expression> && <non-zero constant>) always evaluates to the result of <expression> */
+#  pragma warning(disable : 6326)   /* disable: C6326: Potential comparison of a constant with another constant */
+#endif  /* _MSC_VER */
+
+#ifndef LZ4_FORCE_INLINE
+#  if defined (_MSC_VER) && !defined (__clang__)    /* MSVC */
+#    define LZ4_FORCE_INLINE static __forceinline
+#  else
+#    if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#      if defined (__GNUC__) || defined (__clang__)
+#        define LZ4_FORCE_INLINE static inline __attribute__((always_inline))
+#      else
+#        define LZ4_FORCE_INLINE static inline
+#      endif
+#    else
+#      define LZ4_FORCE_INLINE static
+#    endif /* __STDC_VERSION__ */
+#  endif  /* _MSC_VER */
+#endif /* LZ4_FORCE_INLINE */
+
+/* LZ4_FORCE_O2 and LZ4_FORCE_INLINE
+ * gcc on ppc64le generates an unrolled SIMDized loop for LZ4_wildCopy8,
+ * together with a simple 8-byte copy loop as a fall-back path.
+ * However, this optimization hurts the decompression speed by >30%,
+ * because the execution does not go to the optimized loop
+ * for typical compressible data, and all of the preamble checks
+ * before going to the fall-back path become useless overhead.
+ * This optimization happens only with the -O3 flag, and -O2 generates
+ * a simple 8-byte copy loop.
+ * With gcc on ppc64le, all of the LZ4_decompress_* and LZ4_wildCopy8
+ * functions are annotated with __attribute__((optimize("O2"))),
+ * and also LZ4_wildCopy8 is forcibly inlined, so that the O2 attribute
+ * of LZ4_wildCopy8 does not affect the compression speed.
+ */
+#if defined(__PPC64__) && defined(__LITTLE_ENDIAN__) && defined(__GNUC__) && !defined(__clang__)
+#  define LZ4_FORCE_O2  __attribute__((optimize("O2")))
+#  undef LZ4_FORCE_INLINE
+#  define LZ4_FORCE_INLINE  static __inline __attribute__((optimize("O2"),always_inline))
+#else
+#  define LZ4_FORCE_O2
+#endif
+
+#if (defined(__GNUC__) && (__GNUC__ >= 3)) || (defined(__INTEL_COMPILER) && (__INTEL_COMPILER >= 800)) || defined(__clang__)
+#  define expect(expr,value)    (__builtin_expect ((expr),(value)) )
+#else
+#  define expect(expr,value)    (expr)
+#endif
+
+#ifndef likely
+#define likely(expr)     expect((expr) != 0, 1)
+#endif
+#ifndef unlikely
+#define unlikely(expr)   expect((expr) != 0, 0)
+#endif
+
+/* Should the alignment test prove unreliable, for some reason,
+ * it can be disabled by setting LZ4_ALIGN_TEST to 0 */
+#ifndef LZ4_ALIGN_TEST  /* can be externally provided */
+# define LZ4_ALIGN_TEST 1
+#endif
+
+
+/*-************************************
+*  Memory routines
+**************************************/
+
+/*! LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION :
+ *  Disable relatively high-level LZ4/HC functions that use dynamic memory
+ *  allocation functions (malloc(), calloc(), free()).
+ *
+ *  Note that this is a compile-time switch. And since it disables
+ *  public/stable LZ4 v1 API functions, we don't recommend using this
+ *  symbol to generate a library for distribution.
+ *
+ *  The following public functions are removed when this symbol is defined.
+ *  - lz4   : LZ4_createStream, LZ4_freeStream,
+ *            LZ4_createStreamDecode, LZ4_freeStreamDecode, LZ4_create (deprecated)
+ *  - lz4hc : LZ4_createStreamHC, LZ4_freeStreamHC,
+ *            LZ4_createHC (deprecated), LZ4_freeHC  (deprecated)
+ *  - lz4frame, lz4file : All LZ4F_* functions
+ */
+#if defined(LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION)
+#  define ALLOC(s)          lz4_error_memory_allocation_is_disabled
+#  define ALLOC_AND_ZERO(s) lz4_error_memory_allocation_is_disabled
+#  define FREEMEM(p)        lz4_error_memory_allocation_is_disabled
+#elif defined(LZ4_USER_MEMORY_FUNCTIONS)
+/* memory management functions can be customized by user project.
+ * Below functions must exist somewhere in the Project
+ * and be available at link time */
+void* LZ4_malloc(size_t s);
+void* LZ4_calloc(size_t n, size_t s);
+void  LZ4_free(void* p);
+# define ALLOC(s)          LZ4_malloc(s)
+# define ALLOC_AND_ZERO(s) LZ4_calloc(1,s)
+# define FREEMEM(p)        LZ4_free(p)
+#else
+# include <stdlib.h>   /* malloc, calloc, free */
+# define ALLOC(s)          malloc(s)
+# define ALLOC_AND_ZERO(s) calloc(1,s)
+# define FREEMEM(p)        free(p)
+#endif
+
+#if ! LZ4_FREESTANDING
+#  include <string.h>   /* memset, memcpy */
+#endif
+#if !defined(LZ4_memset)
+#  define LZ4_memset(p,v,s) memset((p),(v),(s))
+#endif
+#define MEM_INIT(p,v,s)   LZ4_memset((p),(v),(s))
+
+
+/*-************************************
+*  Common Constants
+**************************************/
+#define MINMATCH 4
+
+#define WILDCOPYLENGTH 8
+#define LASTLITERALS   5   /* see ../doc/lz4_Block_format.md#parsing-restrictions */
+#define MFLIMIT       12   /* see ../doc/lz4_Block_format.md#parsing-restrictions */
+#define MATCH_SAFEGUARD_DISTANCE  ((2*WILDCOPYLENGTH) - MINMATCH)   /* ensure it's possible to write 2 x wildcopyLength without overflowing output buffer */
+#define FASTLOOP_SAFE_DISTANCE 64
+static const int LZ4_minLength = (MFLIMIT+1);
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define LZ4_DISTANCE_ABSOLUTE_MAX 65535
+#if (LZ4_DISTANCE_MAX > LZ4_DISTANCE_ABSOLUTE_MAX)   /* max supported by LZ4 format */
+#  error "LZ4_DISTANCE_MAX is too big : must be <= 65535"
+#endif
+
+#define ML_BITS  4
+#define ML_MASK  ((1U<<ML_BITS)-1)
+#define RUN_BITS (8-ML_BITS)
+#define RUN_MASK ((1U<<RUN_BITS)-1)
+
+
+/*-************************************
+*  Error detection
+**************************************/
+#if defined(LZ4_DEBUG) && (LZ4_DEBUG>=1)
+#  include <assert.h>
+#else
+#  ifndef assert
+#    define assert(condition) ((void)0)
+#  endif
+#endif
+
+#define LZ4_STATIC_ASSERT(c)   { enum { LZ4_static_assert = 1/(int)(!!(c)) }; }   /* use after variable declarations */
+
+#if defined(LZ4_DEBUG) && (LZ4_DEBUG>=2)
+#  include <stdio.h>
+   static int g_debuglog_enable = 1;
+#  define DEBUGLOG(l, ...) {                          \
+        if ((g_debuglog_enable) && (l<=LZ4_DEBUG)) {  \
+            fprintf(stderr, __FILE__  " %i: ", __LINE__); \
+            fprintf(stderr, __VA_ARGS__);             \
+            fprintf(stderr, " \n");                   \
+    }   }
+#else
+#  define DEBUGLOG(l, ...) {}    /* disabled */
+#endif
+
+static int LZ4_isAligned(const void* ptr, size_t alignment)
+{
+    return ((size_t)ptr & (alignment -1)) == 0;
+}
+
+
+/*-************************************
+*  Types
+**************************************/
+#include <limits.h>
+#if defined(__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+# include <stdint.h>
+  typedef  uint8_t BYTE;
+  typedef uint16_t U16;
+  typedef uint32_t U32;
+  typedef  int32_t S32;
+  typedef uint64_t U64;
+  typedef uintptr_t uptrval;
+#else
+# if UINT_MAX != 4294967295UL
+#   error "LZ4 code (when not C++ or C99) assumes that sizeof(int) == 4"
+# endif
+  typedef unsigned char       BYTE;
+  typedef unsigned short      U16;
+  typedef unsigned int        U32;
+  typedef   signed int        S32;
+  typedef unsigned long long  U64;
+  typedef size_t              uptrval;   /* generally true, except OpenVMS-64 */
+#endif
+
+#if defined(__x86_64__)
+  typedef U64    reg_t;   /* 64-bits in x32 mode */
+#else
+  typedef size_t reg_t;   /* 32-bits in x32 mode */
+#endif
+
+typedef enum {
+    notLimited = 0,
+    limitedOutput = 1,
+    fillOutput = 2
+} limitedOutput_directive;
+
+
+/*-************************************
+*  Reading and writing into memory
+**************************************/
+
+/**
+ * LZ4 relies on memcpy with a constant size being inlined. In freestanding
+ * environments, the compiler can't assume the implementation of memcpy() is
+ * standard compliant, so it can't apply its specialized memcpy() inlining
+ * logic. When possible, use __builtin_memcpy() to tell the compiler to analyze
+ * memcpy() as if it were standard compliant, so it can inline it in freestanding
+ * environments. This is needed when decompressing the Linux Kernel, for example.
+ */
+#if !defined(LZ4_memcpy)
+#  if defined(__GNUC__) && (__GNUC__ >= 4)
+#    define LZ4_memcpy(dst, src, size) __builtin_memcpy(dst, src, size)
+#  else
+#    define LZ4_memcpy(dst, src, size) memcpy(dst, src, size)
+#  endif
+#endif
+
+#if !defined(LZ4_memmove)
+#  if defined(__GNUC__) && (__GNUC__ >= 4)
+#    define LZ4_memmove __builtin_memmove
+#  else
+#    define LZ4_memmove memmove
+#  endif
+#endif
+
+static unsigned LZ4_isLittleEndian(void)
+{
+    const union { U32 u; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental */
+    return one.c[0];
+}
+
+#if defined(__GNUC__) || defined(__INTEL_COMPILER)
+#define LZ4_PACK( __Declaration__ ) __Declaration__ __attribute__((__packed__))
+#elif defined(_MSC_VER)
+#define LZ4_PACK( __Declaration__ ) __pragma( pack(push, 1) ) __Declaration__ __pragma( pack(pop))
+#endif
+
+#if defined(LZ4_FORCE_MEMORY_ACCESS) && (LZ4_FORCE_MEMORY_ACCESS==2)
+/* lie to the compiler about data alignment; use with caution */
+
+static U16 LZ4_read16(const void* memPtr) { return *(const U16*) memPtr; }
+static U32 LZ4_read32(const void* memPtr) { return *(const U32*) memPtr; }
+static reg_t LZ4_read_ARCH(const void* memPtr) { return *(const reg_t*) memPtr; }
+
+static void LZ4_write16(void* memPtr, U16 value) { *(U16*)memPtr = value; }
+static void LZ4_write32(void* memPtr, U32 value) { *(U32*)memPtr = value; }
+
+#elif defined(LZ4_FORCE_MEMORY_ACCESS) && (LZ4_FORCE_MEMORY_ACCESS==1)
+
+/* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
+/* currently only defined for gcc and icc */
+LZ4_PACK(typedef struct { U16 u16; }) LZ4_unalign16;
+LZ4_PACK(typedef struct { U32 u32; }) LZ4_unalign32;
+LZ4_PACK(typedef struct { reg_t uArch; }) LZ4_unalignST;
+
+static U16 LZ4_read16(const void* ptr) { return ((const LZ4_unalign16*)ptr)->u16; }
+static U32 LZ4_read32(const void* ptr) { return ((const LZ4_unalign32*)ptr)->u32; }
+static reg_t LZ4_read_ARCH(const void* ptr) { return ((const LZ4_unalignST*)ptr)->uArch; }
+
+static void LZ4_write16(void* memPtr, U16 value) { ((LZ4_unalign16*)memPtr)->u16 = value; }
+static void LZ4_write32(void* memPtr, U32 value) { ((LZ4_unalign32*)memPtr)->u32 = value; }
+
+#else  /* safe and portable access using memcpy() */
+
+static U16 LZ4_read16(const void* memPtr)
+{
+    U16 val; LZ4_memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+static U32 LZ4_read32(const void* memPtr)
+{
+    U32 val; LZ4_memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+static reg_t LZ4_read_ARCH(const void* memPtr)
+{
+    reg_t val; LZ4_memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+static void LZ4_write16(void* memPtr, U16 value)
+{
+    LZ4_memcpy(memPtr, &value, sizeof(value));
+}
+
+static void LZ4_write32(void* memPtr, U32 value)
+{
+    LZ4_memcpy(memPtr, &value, sizeof(value));
+}
+
+#endif /* LZ4_FORCE_MEMORY_ACCESS */
+
+
+static U16 LZ4_readLE16(const void* memPtr)
+{
+    if (LZ4_isLittleEndian()) {
+        return LZ4_read16(memPtr);
+    } else {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)((U16)p[0] | (p[1]<<8));
+    }
+}
+
+#ifdef LZ4_STATIC_LINKING_ONLY_ENDIANNESS_INDEPENDENT_OUTPUT
+static U32 LZ4_readLE32(const void* memPtr)
+{
+    if (LZ4_isLittleEndian()) {
+        return LZ4_read32(memPtr);
+    } else {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U32)p[0] | (p[1]<<8) | (p[2]<<16) | (p[3]<<24);
+    }
+}
+#endif
+
+static void LZ4_writeLE16(void* memPtr, U16 value)
+{
+    if (LZ4_isLittleEndian()) {
+        LZ4_write16(memPtr, value);
+    } else {
+        BYTE* p = (BYTE*)memPtr;
+        p[0] = (BYTE) value;
+        p[1] = (BYTE)(value>>8);
+    }
+}
+
+/* customized variant of memcpy, which can overwrite up to 8 bytes beyond dstEnd */
+LZ4_FORCE_INLINE
+void LZ4_wildCopy8(void* dstPtr, const void* srcPtr, void* dstEnd)
+{
+    BYTE* d = (BYTE*)dstPtr;
+    const BYTE* s = (const BYTE*)srcPtr;
+    BYTE* const e = (BYTE*)dstEnd;
+
+    do { LZ4_memcpy(d,s,8); d+=8; s+=8; } while (d<e);
+}
+
+static const unsigned inc32table[8] = {0, 1, 2,  1,  0,  4, 4, 4};
+static const int      dec64table[8] = {0, 0, 0, -1, -4,  1, 2, 3};
+
+
+#ifndef LZ4_FAST_DEC_LOOP
+#  if defined __i386__ || defined _M_IX86 || defined __x86_64__ || defined _M_X64
+#    define LZ4_FAST_DEC_LOOP 1
+#  elif defined(__aarch64__) && defined(__APPLE__)
+#    define LZ4_FAST_DEC_LOOP 1
+#  elif defined(__aarch64__) && !defined(__clang__)
+     /* On non-Apple aarch64, we disable this optimization for clang because
+      * on certain mobile chipsets, performance is reduced with clang. For
+      * more information refer to https://github.com/lz4/lz4/pull/707 */
+#    define LZ4_FAST_DEC_LOOP 1
+#  else
+#    define LZ4_FAST_DEC_LOOP 0
+#  endif
+#endif
+
+#if LZ4_FAST_DEC_LOOP
+
+LZ4_FORCE_INLINE void
+LZ4_memcpy_using_offset_base(BYTE* dstPtr, const BYTE* srcPtr, BYTE* dstEnd, const size_t offset)
+{
+    assert(srcPtr + offset == dstPtr);
+    if (offset < 8) {
+        LZ4_write32(dstPtr, 0);   /* silence an msan warning when offset==0 */
+        dstPtr[0] = srcPtr[0];
+        dstPtr[1] = srcPtr[1];
+        dstPtr[2] = srcPtr[2];
+        dstPtr[3] = srcPtr[3];
+        srcPtr += inc32table[offset];
+        LZ4_memcpy(dstPtr+4, srcPtr, 4);
+        srcPtr -= dec64table[offset];
+        dstPtr += 8;
+    } else {
+        LZ4_memcpy(dstPtr, srcPtr, 8);
+        dstPtr += 8;
+        srcPtr += 8;
+    }
+
+    LZ4_wildCopy8(dstPtr, srcPtr, dstEnd);
+}
+
+/* customized variant of memcpy, which can overwrite up to 32 bytes beyond dstEnd
+ * this version copies two times 16 bytes (instead of one time 32 bytes)
+ * because it must be compatible with offsets >= 16. */
+LZ4_FORCE_INLINE void
+LZ4_wildCopy32(void* dstPtr, const void* srcPtr, void* dstEnd)
+{
+    BYTE* d = (BYTE*)dstPtr;
+    const BYTE* s = (const BYTE*)srcPtr;
+    BYTE* const e = (BYTE*)dstEnd;
+
+    do { LZ4_memcpy(d,s,16); LZ4_memcpy(d+16,s+16,16); d+=32; s+=32; } while (d<e);
+}
+
+/* LZ4_memcpy_using_offset()  presumes :
+ * - dstEnd >= dstPtr + MINMATCH
+ * - there is at least 12 bytes available to write after dstEnd */
+LZ4_FORCE_INLINE void
+LZ4_memcpy_using_offset(BYTE* dstPtr, const BYTE* srcPtr, BYTE* dstEnd, const size_t offset)
+{
+    BYTE v[8];
+
+    assert(dstEnd >= dstPtr + MINMATCH);
+
+    switch(offset) {
+    case 1:
+        MEM_INIT(v, *srcPtr, 8);
+        break;
+    case 2:
+        LZ4_memcpy(v, srcPtr, 2);
+        LZ4_memcpy(&v[2], srcPtr, 2);
+#if defined(_MSC_VER) && (_MSC_VER <= 1937) /* MSVC 2022 ver 17.7 or earlier */
+#  pragma warning(push)
+#  pragma warning(disable : 6385) /* warning C6385: Reading invalid data from 'v'. */
+#endif
+        LZ4_memcpy(&v[4], v, 4);
+#if defined(_MSC_VER) && (_MSC_VER <= 1937) /* MSVC 2022 ver 17.7 or earlier */
+#  pragma warning(pop)
+#endif
+        break;
+    case 4:
+        LZ4_memcpy(v, srcPtr, 4);
+        LZ4_memcpy(&v[4], srcPtr, 4);
+        break;
+    default:
+        LZ4_memcpy_using_offset_base(dstPtr, srcPtr, dstEnd, offset);
+        return;
+    }
+
+    LZ4_memcpy(dstPtr, v, 8);
+    dstPtr += 8;
+    while (dstPtr < dstEnd) {
+        LZ4_memcpy(dstPtr, v, 8);
+        dstPtr += 8;
+    }
+}
+#endif
+
+
+/*-************************************
+*  Common functions
+**************************************/
+static unsigned LZ4_NbCommonBytes (reg_t val)
+{
+    assert(val != 0);
+    if (LZ4_isLittleEndian()) {
+        if (sizeof(val) == 8) {
+#       if defined(_MSC_VER) && (_MSC_VER >= 1800) && (defined(_M_AMD64) && !defined(_M_ARM64EC)) && !defined(LZ4_FORCE_SW_BITCOUNT)
+/*-*************************************************************************************************
+* ARM64EC is a Microsoft-designed ARM64 ABI compatible with AMD64 applications on ARM64 Windows 11.
+* The ARM64EC ABI does not support AVX/AVX2/AVX512 instructions, nor their relevant intrinsics
+* including _tzcnt_u64. Therefore, we need to neuter the _tzcnt_u64 code path for ARM64EC.
+****************************************************************************************************/
+#         if defined(__clang__) && (__clang_major__ < 10)
+            /* Avoid undefined clang-cl intrinsics issue.
+             * See https://github.com/lz4/lz4/pull/1017 for details. */
+            return (unsigned)__builtin_ia32_tzcnt_u64(val) >> 3;
+#         else
+            /* x64 CPUS without BMI support interpret `TZCNT` as `REP BSF` */
+            return (unsigned)_tzcnt_u64(val) >> 3;
+#         endif
+#       elif defined(_MSC_VER) && defined(_WIN64) && !defined(LZ4_FORCE_SW_BITCOUNT)
+            unsigned long r = 0;
+            _BitScanForward64(&r, (U64)val);
+            return (unsigned)r >> 3;
+#       elif (defined(__clang__) || (defined(__GNUC__) && ((__GNUC__ > 3) || \
+                            ((__GNUC__ == 3) && (__GNUC_MINOR__ >= 4))))) && \
+                                        !defined(LZ4_FORCE_SW_BITCOUNT)
+            return (unsigned)__builtin_ctzll((U64)val) >> 3;
+#       else
+            const U64 m = 0x0101010101010101ULL;
+            val ^= val - 1;
+            return (unsigned)(((U64)((val & (m - 1)) * m)) >> 56);
+#       endif
+        } else /* 32 bits */ {
+#       if defined(_MSC_VER) && (_MSC_VER >= 1400) && !defined(LZ4_FORCE_SW_BITCOUNT)
+            unsigned long r;
+            _BitScanForward(&r, (U32)val);
+            return (unsigned)r >> 3;
+#       elif (defined(__clang__) || (defined(__GNUC__) && ((__GNUC__ > 3) || \
+                            ((__GNUC__ == 3) && (__GNUC_MINOR__ >= 4))))) && \
+                        !defined(__TINYC__) && !defined(LZ4_FORCE_SW_BITCOUNT)
+            return (unsigned)__builtin_ctz((U32)val) >> 3;
+#       else
+            const U32 m = 0x01010101;
+            return (unsigned)((((val - 1) ^ val) & (m - 1)) * m) >> 24;
+#       endif
+        }
+    } else   /* Big Endian CPU */ {
+        if (sizeof(val)==8) {
+#       if (defined(__clang__) || (defined(__GNUC__) && ((__GNUC__ > 3) || \
+                            ((__GNUC__ == 3) && (__GNUC_MINOR__ >= 4))))) && \
+                        !defined(__TINYC__) && !defined(LZ4_FORCE_SW_BITCOUNT)
+            return (unsigned)__builtin_clzll((U64)val) >> 3;
+#       else
+#if 1
+            /* this method is probably faster,
+             * but adds a 128 bytes lookup table */
+            static const unsigned char ctz7_tab[128] = {
+                7, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+                4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+                5, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+                4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+                6, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+                4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+                5, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+                4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+            };
+            U64 const mask = 0x0101010101010101ULL;
+            U64 const t = (((val >> 8) - mask) | val) & mask;
+            return ctz7_tab[(t * 0x0080402010080402ULL) >> 57];
+#else
+            /* this method doesn't consume memory space like the previous one,
+             * but it contains several branches,
+             * that may end up slowing execution */
+            static const U32 by32 = sizeof(val)*4;  /* 32 on 64 bits (goal), 16 on 32 bits.
+            Just to avoid some static analyzer complaining about shift by 32 on 32-bits target.
+            Note that this code path is never triggered in 32-bits mode. */
+            unsigned r;
+            if (!(val>>by32)) { r=4; } else { r=0; val>>=by32; }
+            if (!(val>>16)) { r+=2; val>>=8; } else { val>>=24; }
+            r += (!val);
+            return r;
+#endif
+#       endif
+        } else /* 32 bits */ {
+#       if (defined(__clang__) || (defined(__GNUC__) && ((__GNUC__ > 3) || \
+                            ((__GNUC__ == 3) && (__GNUC_MINOR__ >= 4))))) && \
+                                        !defined(LZ4_FORCE_SW_BITCOUNT)
+            return (unsigned)__builtin_clz((U32)val) >> 3;
+#       else
+            val >>= 8;
+            val = ((((val + 0x00FFFF00) | 0x00FFFFFF) + val) |
+              (val + 0x00FF0000)) >> 24;
+            return (unsigned)val ^ 3;
+#       endif
+        }
+    }
+}
+
+
+#define STEPSIZE sizeof(reg_t)
+LZ4_FORCE_INLINE
+unsigned LZ4_count(const BYTE* pIn, const BYTE* pMatch, const BYTE* pInLimit)
+{
+    const BYTE* const pStart = pIn;
+
+    if (likely(pIn < pInLimit-(STEPSIZE-1))) {
+        reg_t const diff = LZ4_read_ARCH(pMatch) ^ LZ4_read_ARCH(pIn);
+        if (!diff) {
+            pIn+=STEPSIZE; pMatch+=STEPSIZE;
+        } else {
+            return LZ4_NbCommonBytes(diff);
+    }   }
+
+    while (likely(pIn < pInLimit-(STEPSIZE-1))) {
+        reg_t const diff = LZ4_read_ARCH(pMatch) ^ LZ4_read_ARCH(pIn);
+        if (!diff) { pIn+=STEPSIZE; pMatch+=STEPSIZE; continue; }
+        pIn += LZ4_NbCommonBytes(diff);
+        return (unsigned)(pIn - pStart);
+    }
+
+    if ((STEPSIZE==8) && (pIn<(pInLimit-3)) && (LZ4_read32(pMatch) == LZ4_read32(pIn))) { pIn+=4; pMatch+=4; }
+    if ((pIn<(pInLimit-1)) && (LZ4_read16(pMatch) == LZ4_read16(pIn))) { pIn+=2; pMatch+=2; }
+    if ((pIn<pInLimit) && (*pMatch == *pIn)) pIn++;
+    return (unsigned)(pIn - pStart);
+}
+
+
+#ifndef LZ4_COMMONDEFS_ONLY
+/*-************************************
+*  Local Constants
+**************************************/
+static const int LZ4_64Klimit = ((64 KB) + (MFLIMIT-1));
+static const U32 LZ4_skipTrigger = 6;  /* Increase this value ==> compression run slower on incompressible data */
+
+
+/*-************************************
+*  Local Structures and types
+**************************************/
+typedef enum { clearedTable = 0, byPtr, byU32, byU16 } tableType_t;
+
+/**
+ * This enum distinguishes several different modes of accessing previous
+ * content in the stream.
+ *
+ * - noDict        : There is no preceding content.
+ * - withPrefix64k : Table entries up to ctx->dictSize before the current blob
+ *                   blob being compressed are valid and refer to the preceding
+ *                   content (of length ctx->dictSize), which is available
+ *                   contiguously preceding in memory the content currently
+ *                   being compressed.
+ * - usingExtDict  : Like withPrefix64k, but the preceding content is somewhere
+ *                   else in memory, starting at ctx->dictionary with length
+ *                   ctx->dictSize.
+ * - usingDictCtx  : Everything concerning the preceding content is
+ *                   in a separate context, pointed to by ctx->dictCtx.
+ *                   ctx->dictionary, ctx->dictSize, and table entries
+ *                   in the current context that refer to positions
+ *                   preceding the beginning of the current compression are
+ *                   ignored. Instead, ctx->dictCtx->dictionary and ctx->dictCtx
+ *                   ->dictSize describe the location and size of the preceding
+ *                   content, and matches are found by looking in the ctx
+ *                   ->dictCtx->hashTable.
+ */
+typedef enum { noDict = 0, withPrefix64k, usingExtDict, usingDictCtx } dict_directive;
+typedef enum { noDictIssue = 0, dictSmall } dictIssue_directive;
+
+
+/*-************************************
+*  Local Utils
+**************************************/
+int LZ4_versionNumber (void) { return LZ4_VERSION_NUMBER; }
+const char* LZ4_versionString(void) { return LZ4_VERSION_STRING; }
+int LZ4_compressBound(int isize)  { return LZ4_COMPRESSBOUND(isize); }
+int LZ4_sizeofState(void) { return sizeof(LZ4_stream_t); }
+
+
+/*-****************************************
+*  Internal Definitions, used only in Tests
+*******************************************/
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+int LZ4_compress_forceExtDict (LZ4_stream_t* LZ4_dict, const char* source, char* dest, int srcSize);
+
+int LZ4_decompress_safe_forceExtDict(const char* source, char* dest,
+                                     int compressedSize, int maxOutputSize,
+                                     const void* dictStart, size_t dictSize);
+int LZ4_decompress_safe_partial_forceExtDict(const char* source, char* dest,
+                                     int compressedSize, int targetOutputSize, int dstCapacity,
+                                     const void* dictStart, size_t dictSize);
+#if defined (__cplusplus)
+}
+#endif
+
+/*-******************************
+*  Compression functions
+********************************/
+LZ4_FORCE_INLINE U32 LZ4_hash4(U32 sequence, tableType_t const tableType)
+{
+    if (tableType == byU16)
+        return ((sequence * 2654435761U) >> ((MINMATCH*8)-(LZ4_HASHLOG+1)));
+    else
+        return ((sequence * 2654435761U) >> ((MINMATCH*8)-LZ4_HASHLOG));
+}
+
+LZ4_FORCE_INLINE U32 LZ4_hash5(U64 sequence, tableType_t const tableType)
+{
+    const U32 hashLog = (tableType == byU16) ? LZ4_HASHLOG+1 : LZ4_HASHLOG;
+    if (LZ4_isLittleEndian()) {
+        const U64 prime5bytes = 889523592379ULL;
+        return (U32)(((sequence << 24) * prime5bytes) >> (64 - hashLog));
+    } else {
+        const U64 prime8bytes = 11400714785074694791ULL;
+        return (U32)(((sequence >> 24) * prime8bytes) >> (64 - hashLog));
+    }
+}
+
+LZ4_FORCE_INLINE U32 LZ4_hashPosition(const void* const p, tableType_t const tableType)
+{
+    if ((sizeof(reg_t)==8) && (tableType != byU16)) return LZ4_hash5(LZ4_read_ARCH(p), tableType);
+
+#ifdef LZ4_STATIC_LINKING_ONLY_ENDIANNESS_INDEPENDENT_OUTPUT
+    return LZ4_hash4(LZ4_readLE32(p), tableType);
+#else
+    return LZ4_hash4(LZ4_read32(p), tableType);
+#endif
+}
+
+LZ4_FORCE_INLINE void LZ4_clearHash(U32 h, void* tableBase, tableType_t const tableType)
+{
+    switch (tableType)
+    {
+    default: /* fallthrough */
+    case clearedTable: { /* illegal! */ assert(0); return; }
+    case byPtr: { const BYTE** hashTable = (const BYTE**)tableBase; hashTable[h] = NULL; return; }
+    case byU32: { U32* hashTable = (U32*) tableBase; hashTable[h] = 0; return; }
+    case byU16: { U16* hashTable = (U16*) tableBase; hashTable[h] = 0; return; }
+    }
+}
+
+LZ4_FORCE_INLINE void LZ4_putIndexOnHash(U32 idx, U32 h, void* tableBase, tableType_t const tableType)
+{
+    switch (tableType)
+    {
+    default: /* fallthrough */
+    case clearedTable: /* fallthrough */
+    case byPtr: { /* illegal! */ assert(0); return; }
+    case byU32: { U32* hashTable = (U32*) tableBase; hashTable[h] = idx; return; }
+    case byU16: { U16* hashTable = (U16*) tableBase; assert(idx < 65536); hashTable[h] = (U16)idx; return; }
+    }
+}
+
+/* LZ4_putPosition*() : only used in byPtr mode */
+LZ4_FORCE_INLINE void LZ4_putPositionOnHash(const BYTE* p, U32 h,
+                                  void* tableBase, tableType_t const tableType)
+{
+    const BYTE** const hashTable = (const BYTE**)tableBase;
+    assert(tableType == byPtr); (void)tableType;
+    hashTable[h] = p;
+}
+
+LZ4_FORCE_INLINE void LZ4_putPosition(const BYTE* p, void* tableBase, tableType_t tableType)
+{
+    U32 const h = LZ4_hashPosition(p, tableType);
+    LZ4_putPositionOnHash(p, h, tableBase, tableType);
+}
+
+/* LZ4_getIndexOnHash() :
+ * Index of match position registered in hash table.
+ * hash position must be calculated by using base+index, or dictBase+index.
+ * Assumption 1 : only valid if tableType == byU32 or byU16.
+ * Assumption 2 : h is presumed valid (within limits of hash table)
+ */
+LZ4_FORCE_INLINE U32 LZ4_getIndexOnHash(U32 h, const void* tableBase, tableType_t tableType)
+{
+    LZ4_STATIC_ASSERT(LZ4_MEMORY_USAGE > 2);
+    if (tableType == byU32) {
+        const U32* const hashTable = (const U32*) tableBase;
+        assert(h < (1U << (LZ4_MEMORY_USAGE-2)));
+        return hashTable[h];
+    }
+    if (tableType == byU16) {
+        const U16* const hashTable = (const U16*) tableBase;
+        assert(h < (1U << (LZ4_MEMORY_USAGE-1)));
+        return hashTable[h];
+    }
+    assert(0); return 0;  /* forbidden case */
+}
+
+static const BYTE* LZ4_getPositionOnHash(U32 h, const void* tableBase, tableType_t tableType)
+{
+    assert(tableType == byPtr); (void)tableType;
+    { const BYTE* const* hashTable = (const BYTE* const*) tableBase; return hashTable[h]; }
+}
+
+LZ4_FORCE_INLINE const BYTE*
+LZ4_getPosition(const BYTE* p,
+                const void* tableBase, tableType_t tableType)
+{
+    U32 const h = LZ4_hashPosition(p, tableType);
+    return LZ4_getPositionOnHash(h, tableBase, tableType);
+}
+
+LZ4_FORCE_INLINE void
+LZ4_prepareTable(LZ4_stream_t_internal* const cctx,
+           const int inputSize,
+           const tableType_t tableType) {
+    /* If the table hasn't been used, it's guaranteed to be zeroed out, and is
+     * therefore safe to use no matter what mode we're in. Otherwise, we figure
+     * out if it's safe to leave as is or whether it needs to be reset.
+     */
+    if ((tableType_t)cctx->tableType != clearedTable) {
+        assert(inputSize >= 0);
+        if ((tableType_t)cctx->tableType != tableType
+          || ((tableType == byU16) && cctx->currentOffset + (unsigned)inputSize >= 0xFFFFU)
+          || ((tableType == byU32) && cctx->currentOffset > 1 GB)
+          || tableType == byPtr
+          || inputSize >= 4 KB)
+        {
+            DEBUGLOG(4, "LZ4_prepareTable: Resetting table in %p", cctx);
+            MEM_INIT(cctx->hashTable, 0, LZ4_HASHTABLESIZE);
+            cctx->currentOffset = 0;
+            cctx->tableType = (U32)clearedTable;
+        } else {
+            DEBUGLOG(4, "LZ4_prepareTable: Re-use hash table (no reset)");
+        }
+    }
+
+    /* Adding a gap, so all previous entries are > LZ4_DISTANCE_MAX back,
+     * is faster than compressing without a gap.
+     * However, compressing with currentOffset == 0 is faster still,
+     * so we preserve that case.
+     */
+    if (cctx->currentOffset != 0 && tableType == byU32) {
+        DEBUGLOG(5, "LZ4_prepareTable: adding 64KB to currentOffset");
+        cctx->currentOffset += 64 KB;
+    }
+
+    /* Finally, clear history */
+    cctx->dictCtx = NULL;
+    cctx->dictionary = NULL;
+    cctx->dictSize = 0;
+}
+
+/** LZ4_compress_generic_validated() :
+ *  inlined, to ensure branches are decided at compilation time.
+ *  The following conditions are presumed already validated:
+ *  - source != NULL
+ *  - inputSize > 0
+ */
+LZ4_FORCE_INLINE int LZ4_compress_generic_validated(
+                 LZ4_stream_t_internal* const cctx,
+                 const char* const source,
+                 char* const dest,
+                 const int inputSize,
+                 int*  inputConsumed, /* only written when outputDirective == fillOutput */
+                 const int maxOutputSize,
+                 const limitedOutput_directive outputDirective,
+                 const tableType_t tableType,
+                 const dict_directive dictDirective,
+                 const dictIssue_directive dictIssue,
+                 const int acceleration)
+{
+    int result;
+    const BYTE* ip = (const BYTE*)source;
+
+    U32 const startIndex = cctx->currentOffset;
+    const BYTE* base = (const BYTE*)source - startIndex;
+    const BYTE* lowLimit;
+
+    const LZ4_stream_t_internal* dictCtx = (const LZ4_stream_t_internal*) cctx->dictCtx;
+    const BYTE* const dictionary =
+        dictDirective == usingDictCtx ? dictCtx->dictionary : cctx->dictionary;
+    const U32 dictSize =
+        dictDirective == usingDictCtx ? dictCtx->dictSize : cctx->dictSize;
+    const U32 dictDelta =
+        (dictDirective == usingDictCtx) ? startIndex - dictCtx->currentOffset : 0;   /* make indexes in dictCtx comparable with indexes in current context */
+
+    int const maybe_extMem = (dictDirective == usingExtDict) || (dictDirective == usingDictCtx);
+    U32 const prefixIdxLimit = startIndex - dictSize;   /* used when dictDirective == dictSmall */
+    const BYTE* const dictEnd = dictionary ? dictionary + dictSize : dictionary;
+    const BYTE* anchor = (const BYTE*) source;
+    const BYTE* const iend = ip + inputSize;
+    const BYTE* const mflimitPlusOne = iend - MFLIMIT + 1;
+    const BYTE* const matchlimit = iend - LASTLITERALS;
+
+    /* the dictCtx currentOffset is indexed on the start of the dictionary,
+     * while a dictionary in the current context precedes the currentOffset */
+    const BYTE* dictBase = (dictionary == NULL) ? NULL :
+                           (dictDirective == usingDictCtx) ?
+                            dictionary + dictSize - dictCtx->currentOffset :
+                            dictionary + dictSize - startIndex;
+
+    BYTE* op = (BYTE*) dest;
+    BYTE* const olimit = op + maxOutputSize;
+
+    U32 offset = 0;
+    U32 forwardH;
+
+    DEBUGLOG(5, "LZ4_compress_generic_validated: srcSize=%i, tableType=%u", inputSize, tableType);
+    assert(ip != NULL);
+    if (tableType == byU16) assert(inputSize<LZ4_64Klimit);  /* Size too large (not within 64K limit) */
+    if (tableType == byPtr) assert(dictDirective==noDict);   /* only supported use case with byPtr */
+    /* If init conditions are not met, we don't have to mark stream
+     * as having dirty context, since no action was taken yet */
+    if (outputDirective == fillOutput && maxOutputSize < 1) { return 0; } /* Impossible to store anything */
+    assert(acceleration >= 1);
+
+    lowLimit = (const BYTE*)source - (dictDirective == withPrefix64k ? dictSize : 0);
+
+    /* Update context state */
+    if (dictDirective == usingDictCtx) {
+        /* Subsequent linked blocks can't use the dictionary. */
+        /* Instead, they use the block we just compressed. */
+        cctx->dictCtx = NULL;
+        cctx->dictSize = (U32)inputSize;
+    } else {
+        cctx->dictSize += (U32)inputSize;
+    }
+    cctx->currentOffset += (U32)inputSize;
+    cctx->tableType = (U32)tableType;
+
+    if (inputSize<LZ4_minLength) goto _last_literals;        /* Input too small, no compression (all literals) */
+
+    /* First Byte */
+    {   U32 const h = LZ4_hashPosition(ip, tableType);
+        if (tableType == byPtr) {
+            LZ4_putPositionOnHash(ip, h, cctx->hashTable, byPtr);
+        } else {
+            LZ4_putIndexOnHash(startIndex, h, cctx->hashTable, tableType);
+    }   }
+    ip++; forwardH = LZ4_hashPosition(ip, tableType);
+
+    /* Main Loop */
+    for ( ; ; ) {
+        const BYTE* match;
+        BYTE* token;
+        const BYTE* filledIp;
+
+        /* Find a match */
+        if (tableType == byPtr) {
+            const BYTE* forwardIp = ip;
+            int step = 1;
+            int searchMatchNb = acceleration << LZ4_skipTrigger;
+            do {
+                U32 const h = forwardH;
+                ip = forwardIp;
+                forwardIp += step;
+                step = (searchMatchNb++ >> LZ4_skipTrigger);
+
+                if (unlikely(forwardIp > mflimitPlusOne)) goto _last_literals;
+                assert(ip < mflimitPlusOne);
+
+                match = LZ4_getPositionOnHash(h, cctx->hashTable, tableType);
+                forwardH = LZ4_hashPosition(forwardIp, tableType);
+                LZ4_putPositionOnHash(ip, h, cctx->hashTable, tableType);
+
+            } while ( (match+LZ4_DISTANCE_MAX < ip)
+                   || (LZ4_read32(match) != LZ4_read32(ip)) );
+
+        } else {   /* byU32, byU16 */
+
+            const BYTE* forwardIp = ip;
+            int step = 1;
+            int searchMatchNb = acceleration << LZ4_skipTrigger;
+            do {
+                U32 const h = forwardH;
+                U32 const current = (U32)(forwardIp - base);
+                U32 matchIndex = LZ4_getIndexOnHash(h, cctx->hashTable, tableType);
+                assert(matchIndex <= current);
+                assert(forwardIp - base < (ptrdiff_t)(2 GB - 1));
+                ip = forwardIp;
+                forwardIp += step;
+                step = (searchMatchNb++ >> LZ4_skipTrigger);
+
+                if (unlikely(forwardIp > mflimitPlusOne)) goto _last_literals;
+                assert(ip < mflimitPlusOne);
+
+                if (dictDirective == usingDictCtx) {
+                    if (matchIndex < startIndex) {
+                        /* there was no match, try the dictionary */
+                        assert(tableType == byU32);
+                        matchIndex = LZ4_getIndexOnHash(h, dictCtx->hashTable, byU32);
+                        match = dictBase + matchIndex;
+                        matchIndex += dictDelta;   /* make dictCtx index comparable with current context */
+                        lowLimit = dictionary;
+                    } else {
+                        match = base + matchIndex;
+                        lowLimit = (const BYTE*)source;
+                    }
+                } else if (dictDirective == usingExtDict) {
+                    if (matchIndex < startIndex) {
+                        DEBUGLOG(7, "extDict candidate: matchIndex=%5u  <  startIndex=%5u", matchIndex, startIndex);
+                        assert(startIndex - matchIndex >= MINMATCH);
+                        assert(dictBase);
+                        match = dictBase + matchIndex;
+                        lowLimit = dictionary;
+                    } else {
+                        match = base + matchIndex;
+                        lowLimit = (const BYTE*)source;
+                    }
+                } else {   /* single continuous memory segment */
+                    match = base + matchIndex;
+                }
+                forwardH = LZ4_hashPosition(forwardIp, tableType);
+                LZ4_putIndexOnHash(current, h, cctx->hashTable, tableType);
+
+                DEBUGLOG(7, "candidate at pos=%u  (offset=%u \n", matchIndex, current - matchIndex);
+                if ((dictIssue == dictSmall) && (matchIndex < prefixIdxLimit)) { continue; }    /* match outside of valid area */
+                assert(matchIndex < current);
+                if ( ((tableType != byU16) || (LZ4_DISTANCE_MAX < LZ4_DISTANCE_ABSOLUTE_MAX))
+                  && (matchIndex+LZ4_DISTANCE_MAX < current)) {
+                    continue;
+                } /* too far */
+                assert((current - matchIndex) <= LZ4_DISTANCE_MAX);  /* match now expected within distance */
+
+                if (LZ4_read32(match) == LZ4_read32(ip)) {
+                    if (maybe_extMem) offset = current - matchIndex;
+                    break;   /* match found */
+                }
+
+            } while(1);
+        }
+
+        /* Catch up */
+        filledIp = ip;
+        assert(ip > anchor); /* this is always true as ip has been advanced before entering the main loop */
+        if ((match > lowLimit) && unlikely(ip[-1] == match[-1])) {
+            do { ip--; match--; } while (((ip > anchor) & (match > lowLimit)) && (unlikely(ip[-1] == match[-1])));
+        }
+
+        /* Encode Literals */
+        {   unsigned const litLength = (unsigned)(ip - anchor);
+            token = op++;
+            if ((outputDirective == limitedOutput) &&  /* Check output buffer overflow */
+                (unlikely(op + litLength + (2 + 1 + LASTLITERALS) + (litLength/255) > olimit)) ) {
+                return 0;   /* cannot compress within `dst` budget. Stored indexes in hash table are nonetheless fine */
+            }
+            if ((outputDirective == fillOutput) &&
+                (unlikely(op + (litLength+240)/255 /* litlen */ + litLength /* literals */ + 2 /* offset */ + 1 /* token */ + MFLIMIT - MINMATCH /* min last literals so last match is <= end - MFLIMIT */ > olimit))) {
+                op--;
+                goto _last_literals;
+            }
+            if (litLength >= RUN_MASK) {
+                unsigned len = litLength - RUN_MASK;
+                *token = (RUN_MASK<<ML_BITS);
+                for(; len >= 255 ; len-=255) *op++ = 255;
+                *op++ = (BYTE)len;
+            }
+            else *token = (BYTE)(litLength<<ML_BITS);
+
+            /* Copy Literals */
+            LZ4_wildCopy8(op, anchor, op+litLength);
+            op+=litLength;
+            DEBUGLOG(6, "seq.start:%i, literals=%u, match.start:%i",
+                        (int)(anchor-(const BYTE*)source), litLength, (int)(ip-(const BYTE*)source));
+        }
+
+_next_match:
+        /* at this stage, the following variables must be correctly set :
+         * - ip : at start of LZ operation
+         * - match : at start of previous pattern occurrence; can be within current prefix, or within extDict
+         * - offset : if maybe_ext_memSegment==1 (constant)
+         * - lowLimit : must be == dictionary to mean "match is within extDict"; must be == source otherwise
+         * - token and *token : position to write 4-bits for match length; higher 4-bits for literal length supposed already written
+         */
+
+        if ((outputDirective == fillOutput) &&
+            (op + 2 /* offset */ + 1 /* token */ + MFLIMIT - MINMATCH /* min last literals so last match is <= end - MFLIMIT */ > olimit)) {
+            /* the match was too close to the end, rewind and go to last literals */
+            op = token;
+            goto _last_literals;
+        }
+
+        /* Encode Offset */
+        if (maybe_extMem) {   /* static test */
+            DEBUGLOG(6, "             with offset=%u  (ext if > %i)", offset, (int)(ip - (const BYTE*)source));
+            assert(offset <= LZ4_DISTANCE_MAX && offset > 0);
+            LZ4_writeLE16(op, (U16)offset); op+=2;
+        } else  {
+            DEBUGLOG(6, "             with offset=%u  (same segment)", (U32)(ip - match));
+            assert(ip-match <= LZ4_DISTANCE_MAX);
+            LZ4_writeLE16(op, (U16)(ip - match)); op+=2;
+        }
+
+        /* Encode MatchLength */
+        {   unsigned matchCode;
+
+            if ( (dictDirective==usingExtDict || dictDirective==usingDictCtx)
+              && (lowLimit==dictionary) /* match within extDict */ ) {
+                const BYTE* limit = ip + (dictEnd-match);
+                assert(dictEnd > match);
+                if (limit > matchlimit) limit = matchlimit;
+                matchCode = LZ4_count(ip+MINMATCH, match+MINMATCH, limit);
+                ip += (size_t)matchCode + MINMATCH;
+                if (ip==limit) {
+                    unsigned const more = LZ4_count(limit, (const BYTE*)source, matchlimit);
+                    matchCode += more;
+                    ip += more;
+                }
+                DEBUGLOG(6, "             with matchLength=%u starting in extDict", matchCode+MINMATCH);
+            } else {
+                matchCode = LZ4_count(ip+MINMATCH, match+MINMATCH, matchlimit);
+                ip += (size_t)matchCode + MINMATCH;
+                DEBUGLOG(6, "             with matchLength=%u", matchCode+MINMATCH);
+            }
+
+            if ((outputDirective) &&    /* Check output buffer overflow */
+                (unlikely(op + (1 + LASTLITERALS) + (matchCode+240)/255 > olimit)) ) {
+                if (outputDirective == fillOutput) {
+                    /* Match description too long : reduce it */
+                    U32 newMatchCode = 15 /* in token */ - 1 /* to avoid needing a zero byte */ + ((U32)(olimit - op) - 1 - LASTLITERALS) * 255;
+                    ip -= matchCode - newMatchCode;
+                    assert(newMatchCode < matchCode);
+                    matchCode = newMatchCode;
+                    if (unlikely(ip <= filledIp)) {
+                        /* We have already filled up to filledIp so if ip ends up less than filledIp
+                         * we have positions in the hash table beyond the current position. This is
+                         * a problem if we reuse the hash table. So we have to remove these positions
+                         * from the hash table.
+                         */
+                        const BYTE* ptr;
+                        DEBUGLOG(5, "Clearing %u positions", (U32)(filledIp - ip));
+                        for (ptr = ip; ptr <= filledIp; ++ptr) {
+                            U32 const h = LZ4_hashPosition(ptr, tableType);
+                            LZ4_clearHash(h, cctx->hashTable, tableType);
+                        }
+                    }
+                } else {
+                    assert(outputDirective == limitedOutput);
+                    return 0;   /* cannot compress within `dst` budget. Stored indexes in hash table are nonetheless fine */
+                }
+            }
+            if (matchCode >= ML_MASK) {
+                *token += ML_MASK;
+                matchCode -= ML_MASK;
+                LZ4_write32(op, 0xFFFFFFFF);
+                while (matchCode >= 4*255) {
+                    op+=4;
+                    LZ4_write32(op, 0xFFFFFFFF);
+                    matchCode -= 4*255;
+                }
+                op += matchCode / 255;
+                *op++ = (BYTE)(matchCode % 255);
+            } else
+                *token += (BYTE)(matchCode);
+        }
+        /* Ensure we have enough space for the last literals. */
+        assert(!(outputDirective == fillOutput && op + 1 + LASTLITERALS > olimit));
+
+        anchor = ip;
+
+        /* Test end of chunk */
+        if (ip >= mflimitPlusOne) break;
+
+        /* Fill table */
+        {   U32 const h = LZ4_hashPosition(ip-2, tableType);
+            if (tableType == byPtr) {
+                LZ4_putPositionOnHash(ip-2, h, cctx->hashTable, byPtr);
+            } else {
+                U32 const idx = (U32)((ip-2) - base);
+                LZ4_putIndexOnHash(idx, h, cctx->hashTable, tableType);
+        }   }
+
+        /* Test next position */
+        if (tableType == byPtr) {
+
+            match = LZ4_getPosition(ip, cctx->hashTable, tableType);
+            LZ4_putPosition(ip, cctx->hashTable, tableType);
+            if ( (match+LZ4_DISTANCE_MAX >= ip)
+              && (LZ4_read32(match) == LZ4_read32(ip)) )
+            { token=op++; *token=0; goto _next_match; }
+
+        } else {   /* byU32, byU16 */
+
+            U32 const h = LZ4_hashPosition(ip, tableType);
+            U32 const current = (U32)(ip-base);
+            U32 matchIndex = LZ4_getIndexOnHash(h, cctx->hashTable, tableType);
+            assert(matchIndex < current);
+            if (dictDirective == usingDictCtx) {
+                if (matchIndex < startIndex) {
+                    /* there was no match, try the dictionary */
+                    assert(tableType == byU32);
+                    matchIndex = LZ4_getIndexOnHash(h, dictCtx->hashTable, byU32);
+                    match = dictBase + matchIndex;
+                    lowLimit = dictionary;   /* required for match length counter */
+                    matchIndex += dictDelta;
+                } else {
+                    match = base + matchIndex;
+                    lowLimit = (const BYTE*)source;  /* required for match length counter */
+                }
+            } else if (dictDirective==usingExtDict) {
+                if (matchIndex < startIndex) {
+                    assert(dictBase);
+                    match = dictBase + matchIndex;
+                    lowLimit = dictionary;   /* required for match length counter */
+                } else {
+                    match = base + matchIndex;
+                    lowLimit = (const BYTE*)source;   /* required for match length counter */
+                }
+            } else {   /* single memory segment */
+                match = base + matchIndex;
+            }
+            LZ4_putIndexOnHash(current, h, cctx->hashTable, tableType);
+            assert(matchIndex < current);
+            if ( ((dictIssue==dictSmall) ? (matchIndex >= prefixIdxLimit) : 1)
+              && (((tableType==byU16) && (LZ4_DISTANCE_MAX == LZ4_DISTANCE_ABSOLUTE_MAX)) ? 1 : (matchIndex+LZ4_DISTANCE_MAX >= current))
+              && (LZ4_read32(match) == LZ4_read32(ip)) ) {
+                token=op++;
+                *token=0;
+                if (maybe_extMem) offset = current - matchIndex;
+                DEBUGLOG(6, "seq.start:%i, literals=%u, match.start:%i",
+                            (int)(anchor-(const BYTE*)source), 0, (int)(ip-(const BYTE*)source));
+                goto _next_match;
+            }
+        }
+
+        /* Prepare next loop */
+        forwardH = LZ4_hashPosition(++ip, tableType);
+
+    }
+
+_last_literals:
+    /* Encode Last Literals */
+    {   size_t lastRun = (size_t)(iend - anchor);
+        if ( (outputDirective) &&  /* Check output buffer overflow */
+            (op + lastRun + 1 + ((lastRun+255-RUN_MASK)/255) > olimit)) {
+            if (outputDirective == fillOutput) {
+                /* adapt lastRun to fill 'dst' */
+                assert(olimit >= op);
+                lastRun  = (size_t)(olimit-op) - 1/*token*/;
+                lastRun -= (lastRun + 256 - RUN_MASK) / 256;  /*additional length tokens*/
+            } else {
+                assert(outputDirective == limitedOutput);
+                return 0;   /* cannot compress within `dst` budget. Stored indexes in hash table are nonetheless fine */
+            }
+        }
+        DEBUGLOG(6, "Final literal run : %i literals", (int)lastRun);
+        if (lastRun >= RUN_MASK) {
+            size_t accumulator = lastRun - RUN_MASK;
+            *op++ = RUN_MASK << ML_BITS;
+            for(; accumulator >= 255 ; accumulator-=255) *op++ = 255;
+            *op++ = (BYTE) accumulator;
+        } else {
+            *op++ = (BYTE)(lastRun<<ML_BITS);
+        }
+        LZ4_memcpy(op, anchor, lastRun);
+        ip = anchor + lastRun;
+        op += lastRun;
+    }
+
+    if (outputDirective == fillOutput) {
+        *inputConsumed = (int) (((const char*)ip)-source);
+    }
+    result = (int)(((char*)op) - dest);
+    assert(result > 0);
+    DEBUGLOG(5, "LZ4_compress_generic: compressed %i bytes into %i bytes", inputSize, result);
+    return result;
+}
+
+/** LZ4_compress_generic() :
+ *  inlined, to ensure branches are decided at compilation time;
+ *  takes care of src == (NULL, 0)
+ *  and forward the rest to LZ4_compress_generic_validated */
+LZ4_FORCE_INLINE int LZ4_compress_generic(
+                 LZ4_stream_t_internal* const cctx,
+                 const char* const src,
+                 char* const dst,
+                 const int srcSize,
+                 int *inputConsumed, /* only written when outputDirective == fillOutput */
+                 const int dstCapacity,
+                 const limitedOutput_directive outputDirective,
+                 const tableType_t tableType,
+                 const dict_directive dictDirective,
+                 const dictIssue_directive dictIssue,
+                 const int acceleration)
+{
+    DEBUGLOG(5, "LZ4_compress_generic: srcSize=%i, dstCapacity=%i",
+                srcSize, dstCapacity);
+
+    if ((U32)srcSize > (U32)LZ4_MAX_INPUT_SIZE) { return 0; }  /* Unsupported srcSize, too large (or negative) */
+    if (srcSize == 0) {   /* src == NULL supported if srcSize == 0 */
+        if (outputDirective != notLimited && dstCapacity <= 0) return 0;  /* no output, can't write anything */
+        DEBUGLOG(5, "Generating an empty block");
+        assert(outputDirective == notLimited || dstCapacity >= 1);
+        assert(dst != NULL);
+        dst[0] = 0;
+        if (outputDirective == fillOutput) {
+            assert (inputConsumed != NULL);
+            *inputConsumed = 0;
+        }
+        return 1;
+    }
+    assert(src != NULL);
+
+    return LZ4_compress_generic_validated(cctx, src, dst, srcSize,
+                inputConsumed, /* only written into if outputDirective == fillOutput */
+                dstCapacity, outputDirective,
+                tableType, dictDirective, dictIssue, acceleration);
+}
+
+
+int LZ4_compress_fast_extState(void* state, const char* source, char* dest, int inputSize, int maxOutputSize, int acceleration)
+{
+    LZ4_stream_t_internal* const ctx = & LZ4_initStream(state, sizeof(LZ4_stream_t)) -> internal_donotuse;
+    assert(ctx != NULL);
+    if (acceleration < 1) acceleration = LZ4_ACCELERATION_DEFAULT;
+    if (acceleration > LZ4_ACCELERATION_MAX) acceleration = LZ4_ACCELERATION_MAX;
+    if (maxOutputSize >= LZ4_compressBound(inputSize)) {
+        if (inputSize < LZ4_64Klimit) {
+            return LZ4_compress_generic(ctx, source, dest, inputSize, NULL, 0, notLimited, byU16, noDict, noDictIssue, acceleration);
+        } else {
+            const tableType_t tableType = ((sizeof(void*)==4) && ((uptrval)source > LZ4_DISTANCE_MAX)) ? byPtr : byU32;
+            return LZ4_compress_generic(ctx, source, dest, inputSize, NULL, 0, notLimited, tableType, noDict, noDictIssue, acceleration);
+        }
+    } else {
+        if (inputSize < LZ4_64Klimit) {
+            return LZ4_compress_generic(ctx, source, dest, inputSize, NULL, maxOutputSize, limitedOutput, byU16, noDict, noDictIssue, acceleration);
+        } else {
+            const tableType_t tableType = ((sizeof(void*)==4) && ((uptrval)source > LZ4_DISTANCE_MAX)) ? byPtr : byU32;
+            return LZ4_compress_generic(ctx, source, dest, inputSize, NULL, maxOutputSize, limitedOutput, tableType, noDict, noDictIssue, acceleration);
+        }
+    }
+}
+
+/**
+ * LZ4_compress_fast_extState_fastReset() :
+ * A variant of LZ4_compress_fast_extState().
+ *
+ * Using this variant avoids an expensive initialization step. It is only safe
+ * to call if the state buffer is known to be correctly initialized already
+ * (see comment in lz4.h on LZ4_resetStream_fast() for a definition of
+ * "correctly initialized").
+ */
+int LZ4_compress_fast_extState_fastReset(void* state, const char* src, char* dst, int srcSize, int dstCapacity, int acceleration)
+{
+    LZ4_stream_t_internal* const ctx = &((LZ4_stream_t*)state)->internal_donotuse;
+    if (acceleration < 1) acceleration = LZ4_ACCELERATION_DEFAULT;
+    if (acceleration > LZ4_ACCELERATION_MAX) acceleration = LZ4_ACCELERATION_MAX;
+    assert(ctx != NULL);
+
+    if (dstCapacity >= LZ4_compressBound(srcSize)) {
+        if (srcSize < LZ4_64Klimit) {
+            const tableType_t tableType = byU16;
+            LZ4_prepareTable(ctx, srcSize, tableType);
+            if (ctx->currentOffset) {
+                return LZ4_compress_generic(ctx, src, dst, srcSize, NULL, 0, notLimited, tableType, noDict, dictSmall, acceleration);
+            } else {
+                return LZ4_compress_generic(ctx, src, dst, srcSize, NULL, 0, notLimited, tableType, noDict, noDictIssue, acceleration);
+            }
+        } else {
+            const tableType_t tableType = ((sizeof(void*)==4) && ((uptrval)src > LZ4_DISTANCE_MAX)) ? byPtr : byU32;
+            LZ4_prepareTable(ctx, srcSize, tableType);
+            return LZ4_compress_generic(ctx, src, dst, srcSize, NULL, 0, notLimited, tableType, noDict, noDictIssue, acceleration);
+        }
+    } else {
+        if (srcSize < LZ4_64Klimit) {
+            const tableType_t tableType = byU16;
+            LZ4_prepareTable(ctx, srcSize, tableType);
+            if (ctx->currentOffset) {
+                return LZ4_compress_generic(ctx, src, dst, srcSize, NULL, dstCapacity, limitedOutput, tableType, noDict, dictSmall, acceleration);
+            } else {
+                return LZ4_compress_generic(ctx, src, dst, srcSize, NULL, dstCapacity, limitedOutput, tableType, noDict, noDictIssue, acceleration);
+            }
+        } else {
+            const tableType_t tableType = ((sizeof(void*)==4) && ((uptrval)src > LZ4_DISTANCE_MAX)) ? byPtr : byU32;
+            LZ4_prepareTable(ctx, srcSize, tableType);
+            return LZ4_compress_generic(ctx, src, dst, srcSize, NULL, dstCapacity, limitedOutput, tableType, noDict, noDictIssue, acceleration);
+        }
+    }
+}
+
+
+int LZ4_compress_fast(const char* src, char* dest, int srcSize, int dstCapacity, int acceleration)
+{
+    int result;
+#if (LZ4_HEAPMODE)
+    LZ4_stream_t* const ctxPtr = (LZ4_stream_t*)ALLOC(sizeof(LZ4_stream_t));   /* malloc-calloc always properly aligned */
+    if (ctxPtr == NULL) return 0;
+#else
+    LZ4_stream_t ctx;
+    LZ4_stream_t* const ctxPtr = &ctx;
+#endif
+    result = LZ4_compress_fast_extState(ctxPtr, src, dest, srcSize, dstCapacity, acceleration);
+
+#if (LZ4_HEAPMODE)
+    FREEMEM(ctxPtr);
+#endif
+    return result;
+}
+
+
+int LZ4_compress_default(const char* src, char* dst, int srcSize, int dstCapacity)
+{
+    return LZ4_compress_fast(src, dst, srcSize, dstCapacity, 1);
+}
+
+
+/* Note!: This function leaves the stream in an unclean/broken state!
+ * It is not safe to subsequently use the same state with a _fastReset() or
+ * _continue() call without resetting it. */
+static int LZ4_compress_destSize_extState_internal(LZ4_stream_t* state, const char* src, char* dst, int* srcSizePtr, int targetDstSize, int acceleration)
+{
+    void* const s = LZ4_initStream(state, sizeof (*state));
+    assert(s != NULL); (void)s;
+
+    if (targetDstSize >= LZ4_compressBound(*srcSizePtr)) {  /* compression success is guaranteed */
+        return LZ4_compress_fast_extState(state, src, dst, *srcSizePtr, targetDstSize, acceleration);
+    } else {
+        if (*srcSizePtr < LZ4_64Klimit) {
+            return LZ4_compress_generic(&state->internal_donotuse, src, dst, *srcSizePtr, srcSizePtr, targetDstSize, fillOutput, byU16, noDict, noDictIssue, acceleration);
+        } else {
+            tableType_t const addrMode = ((sizeof(void*)==4) && ((uptrval)src > LZ4_DISTANCE_MAX)) ? byPtr : byU32;
+            return LZ4_compress_generic(&state->internal_donotuse, src, dst, *srcSizePtr, srcSizePtr, targetDstSize, fillOutput, addrMode, noDict, noDictIssue, acceleration);
+    }   }
+}
+
+int LZ4_compress_destSize_extState(void* state, const char* src, char* dst, int* srcSizePtr, int targetDstSize, int acceleration)
+{
+    int const r = LZ4_compress_destSize_extState_internal((LZ4_stream_t*)state, src, dst, srcSizePtr, targetDstSize, acceleration);
+    /* clean the state on exit */
+    LZ4_initStream(state, sizeof (LZ4_stream_t));
+    return r;
+}
+
+
+int LZ4_compress_destSize(const char* src, char* dst, int* srcSizePtr, int targetDstSize)
+{
+#if (LZ4_HEAPMODE)
+    LZ4_stream_t* const ctx = (LZ4_stream_t*)ALLOC(sizeof(LZ4_stream_t));   /* malloc-calloc always properly aligned */
+    if (ctx == NULL) return 0;
+#else
+    LZ4_stream_t ctxBody;
+    LZ4_stream_t* const ctx = &ctxBody;
+#endif
+
+    int result = LZ4_compress_destSize_extState_internal(ctx, src, dst, srcSizePtr, targetDstSize, 1);
+
+#if (LZ4_HEAPMODE)
+    FREEMEM(ctx);
+#endif
+    return result;
+}
+
+
+
+/*-******************************
+*  Streaming functions
+********************************/
+
+#if !defined(LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION)
+LZ4_stream_t* LZ4_createStream(void)
+{
+    LZ4_stream_t* const lz4s = (LZ4_stream_t*)ALLOC(sizeof(LZ4_stream_t));
+    LZ4_STATIC_ASSERT(sizeof(LZ4_stream_t) >= sizeof(LZ4_stream_t_internal));
+    DEBUGLOG(4, "LZ4_createStream %p", lz4s);
+    if (lz4s == NULL) return NULL;
+    LZ4_initStream(lz4s, sizeof(*lz4s));
+    return lz4s;
+}
+#endif
+
+static size_t LZ4_stream_t_alignment(void)
+{
+#if LZ4_ALIGN_TEST
+    typedef struct { char c; LZ4_stream_t t; } t_a;
+    return sizeof(t_a) - sizeof(LZ4_stream_t);
+#else
+    return 1;  /* effectively disabled */
+#endif
+}
+
+LZ4_stream_t* LZ4_initStream (void* buffer, size_t size)
+{
+    DEBUGLOG(5, "LZ4_initStream");
+    if (buffer == NULL) { return NULL; }
+    if (size < sizeof(LZ4_stream_t)) { return NULL; }
+    if (!LZ4_isAligned(buffer, LZ4_stream_t_alignment())) return NULL;
+    MEM_INIT(buffer, 0, sizeof(LZ4_stream_t_internal));
+    return (LZ4_stream_t*)buffer;
+}
+
+/* resetStream is now deprecated,
+ * prefer initStream() which is more general */
+void LZ4_resetStream (LZ4_stream_t* LZ4_stream)
+{
+    DEBUGLOG(5, "LZ4_resetStream (ctx:%p)", LZ4_stream);
+    MEM_INIT(LZ4_stream, 0, sizeof(LZ4_stream_t_internal));
+}
+
+void LZ4_resetStream_fast(LZ4_stream_t* ctx) {
+    LZ4_prepareTable(&(ctx->internal_donotuse), 0, byU32);
+}
+
+#if !defined(LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION)
+int LZ4_freeStream (LZ4_stream_t* LZ4_stream)
+{
+    if (!LZ4_stream) return 0;   /* support free on NULL */
+    DEBUGLOG(5, "LZ4_freeStream %p", LZ4_stream);
+    FREEMEM(LZ4_stream);
+    return (0);
+}
+#endif
+
+
+typedef enum { _ld_fast, _ld_slow } LoadDict_mode_e;
+#define HASH_UNIT sizeof(reg_t)
+int LZ4_loadDict_internal(LZ4_stream_t* LZ4_dict,
+                    const char* dictionary, int dictSize,
+                    LoadDict_mode_e _ld)
+{
+    LZ4_stream_t_internal* const dict = &LZ4_dict->internal_donotuse;
+    const tableType_t tableType = byU32;
+    const BYTE* p = (const BYTE*)dictionary;
+    const BYTE* const dictEnd = p + dictSize;
+    U32 idx32;
+
+    DEBUGLOG(4, "LZ4_loadDict (%i bytes from %p into %p)", dictSize, dictionary, LZ4_dict);
+
+    /* It's necessary to reset the context,
+     * and not just continue it with prepareTable()
+     * to avoid any risk of generating overflowing matchIndex
+     * when compressing using this dictionary */
+    LZ4_resetStream(LZ4_dict);
+
+    /* We always increment the offset by 64 KB, since, if the dict is longer,
+     * we truncate it to the last 64k, and if it's shorter, we still want to
+     * advance by a whole window length so we can provide the guarantee that
+     * there are only valid offsets in the window, which allows an optimization
+     * in LZ4_compress_fast_continue() where it uses noDictIssue even when the
+     * dictionary isn't a full 64k. */
+    dict->currentOffset += 64 KB;
+
+    if (dictSize < (int)HASH_UNIT) {
+        return 0;
+    }
+
+    if ((dictEnd - p) > 64 KB) p = dictEnd - 64 KB;
+    dict->dictionary = p;
+    dict->dictSize = (U32)(dictEnd - p);
+    dict->tableType = (U32)tableType;
+    idx32 = dict->currentOffset - dict->dictSize;
+
+    while (p <= dictEnd-HASH_UNIT) {
+        U32 const h = LZ4_hashPosition(p, tableType);
+        /* Note: overwriting => favors positions end of dictionary */
+        LZ4_putIndexOnHash(idx32, h, dict->hashTable, tableType);
+        p+=3; idx32+=3;
+    }
+
+    if (_ld == _ld_slow) {
+        /* Fill hash table with additional references, to improve compression capability */
+        p = dict->dictionary;
+        idx32 = dict->currentOffset - dict->dictSize;
+        while (p <= dictEnd-HASH_UNIT) {
+            U32 const h = LZ4_hashPosition(p, tableType);
+            U32 const limit = dict->currentOffset - 64 KB;
+            if (LZ4_getIndexOnHash(h, dict->hashTable, tableType) <= limit) {
+                /* Note: not overwriting => favors positions beginning of dictionary */
+                LZ4_putIndexOnHash(idx32, h, dict->hashTable, tableType);
+            }
+            p++; idx32++;
+        }
+    }
+
+    return (int)dict->dictSize;
+}
+
+int LZ4_loadDict(LZ4_stream_t* LZ4_dict, const char* dictionary, int dictSize)
+{
+    return LZ4_loadDict_internal(LZ4_dict, dictionary, dictSize, _ld_fast);
+}
+
+int LZ4_loadDictSlow(LZ4_stream_t* LZ4_dict, const char* dictionary, int dictSize)
+{
+    return LZ4_loadDict_internal(LZ4_dict, dictionary, dictSize, _ld_slow);
+}
+
+void LZ4_attach_dictionary(LZ4_stream_t* workingStream, const LZ4_stream_t* dictionaryStream)
+{
+    const LZ4_stream_t_internal* dictCtx = (dictionaryStream == NULL) ? NULL :
+        &(dictionaryStream->internal_donotuse);
+
+    DEBUGLOG(4, "LZ4_attach_dictionary (%p, %p, size %u)",
+             workingStream, dictionaryStream,
+             dictCtx != NULL ? dictCtx->dictSize : 0);
+
+    if (dictCtx != NULL) {
+        /* If the current offset is zero, we will never look in the
+         * external dictionary context, since there is no value a table
+         * entry can take that indicate a miss. In that case, we need
+         * to bump the offset to something non-zero.
+         */
+        if (workingStream->internal_donotuse.currentOffset == 0) {
+            workingStream->internal_donotuse.currentOffset = 64 KB;
+        }
+
+        /* Don't actually attach an empty dictionary.
+         */
+        if (dictCtx->dictSize == 0) {
+            dictCtx = NULL;
+        }
+    }
+    workingStream->internal_donotuse.dictCtx = dictCtx;
+}
+
+
+static void LZ4_renormDictT(LZ4_stream_t_internal* LZ4_dict, int nextSize)
+{
+    assert(nextSize >= 0);
+    if (LZ4_dict->currentOffset + (unsigned)nextSize > 0x80000000) {   /* potential ptrdiff_t overflow (32-bits mode) */
+        /* rescale hash table */
+        U32 const delta = LZ4_dict->currentOffset - 64 KB;
+        const BYTE* dictEnd = LZ4_dict->dictionary + LZ4_dict->dictSize;
+        int i;
+        DEBUGLOG(4, "LZ4_renormDictT");
+        for (i=0; i<LZ4_HASH_SIZE_U32; i++) {
+            if (LZ4_dict->hashTable[i] < delta) LZ4_dict->hashTable[i]=0;
+            else LZ4_dict->hashTable[i] -= delta;
+        }
+        LZ4_dict->currentOffset = 64 KB;
+        if (LZ4_dict->dictSize > 64 KB) LZ4_dict->dictSize = 64 KB;
+        LZ4_dict->dictionary = dictEnd - LZ4_dict->dictSize;
+    }
+}
+
+
+int LZ4_compress_fast_continue (LZ4_stream_t* LZ4_stream,
+                                const char* source, char* dest,
+                                int inputSize, int maxOutputSize,
+                                int acceleration)
+{
+    const tableType_t tableType = byU32;
+    LZ4_stream_t_internal* const streamPtr = &LZ4_stream->internal_donotuse;
+    const char* dictEnd = streamPtr->dictSize ? (const char*)streamPtr->dictionary + streamPtr->dictSize : NULL;
+
+    DEBUGLOG(5, "LZ4_compress_fast_continue (inputSize=%i, dictSize=%u)", inputSize, streamPtr->dictSize);
+
+    LZ4_renormDictT(streamPtr, inputSize);   /* fix index overflow */
+    if (acceleration < 1) acceleration = LZ4_ACCELERATION_DEFAULT;
+    if (acceleration > LZ4_ACCELERATION_MAX) acceleration = LZ4_ACCELERATION_MAX;
+
+    /* invalidate tiny dictionaries */
+    if ( (streamPtr->dictSize < 4)     /* tiny dictionary : not enough for a hash */
+      && (dictEnd != source)           /* prefix mode */
+      && (inputSize > 0)               /* tolerance : don't lose history, in case next invocation would use prefix mode */
+      && (streamPtr->dictCtx == NULL)  /* usingDictCtx */
+      ) {
+        DEBUGLOG(5, "LZ4_compress_fast_continue: dictSize(%u) at addr:%p is too small", streamPtr->dictSize, streamPtr->dictionary);
+        /* remove dictionary existence from history, to employ faster prefix mode */
+        streamPtr->dictSize = 0;
+        streamPtr->dictionary = (const BYTE*)source;
+        dictEnd = source;
+    }
+
+    /* Check overlapping input/dictionary space */
+    {   const char* const sourceEnd = source + inputSize;
+        if ((sourceEnd > (const char*)streamPtr->dictionary) && (sourceEnd < dictEnd)) {
+            streamPtr->dictSize = (U32)(dictEnd - sourceEnd);
+            if (streamPtr->dictSize > 64 KB) streamPtr->dictSize = 64 KB;
+            if (streamPtr->dictSize < 4) streamPtr->dictSize = 0;
+            streamPtr->dictionary = (const BYTE*)dictEnd - streamPtr->dictSize;
+        }
+    }
+
+    /* prefix mode : source data follows dictionary */
+    if (dictEnd == source) {
+        if ((streamPtr->dictSize < 64 KB) && (streamPtr->dictSize < streamPtr->currentOffset))
+            return LZ4_compress_generic(streamPtr, source, dest, inputSize, NULL, maxOutputSize, limitedOutput, tableType, withPrefix64k, dictSmall, acceleration);
+        else
+            return LZ4_compress_generic(streamPtr, source, dest, inputSize, NULL, maxOutputSize, limitedOutput, tableType, withPrefix64k, noDictIssue, acceleration);
+    }
+
+    /* external dictionary mode */
+    {   int result;
+        if (streamPtr->dictCtx) {
+            /* We depend here on the fact that dictCtx'es (produced by
+             * LZ4_loadDict) guarantee that their tables contain no references
+             * to offsets between dictCtx->currentOffset - 64 KB and
+             * dictCtx->currentOffset - dictCtx->dictSize. This makes it safe
+             * to use noDictIssue even when the dict isn't a full 64 KB.
+             */
+            if (inputSize > 4 KB) {
+                /* For compressing large blobs, it is faster to pay the setup
+                 * cost to copy the dictionary's tables into the active context,
+                 * so that the compression loop is only looking into one table.
+                 */
+                LZ4_memcpy(streamPtr, streamPtr->dictCtx, sizeof(*streamPtr));
+                result = LZ4_compress_generic(streamPtr, source, dest, inputSize, NULL, maxOutputSize, limitedOutput, tableType, usingExtDict, noDictIssue, acceleration);
+            } else {
+                result = LZ4_compress_generic(streamPtr, source, dest, inputSize, NULL, maxOutputSize, limitedOutput, tableType, usingDictCtx, noDictIssue, acceleration);
+            }
+        } else {  /* small data <= 4 KB */
+            if ((streamPtr->dictSize < 64 KB) && (streamPtr->dictSize < streamPtr->currentOffset)) {
+                result = LZ4_compress_generic(streamPtr, source, dest, inputSize, NULL, maxOutputSize, limitedOutput, tableType, usingExtDict, dictSmall, acceleration);
+            } else {
+                result = LZ4_compress_generic(streamPtr, source, dest, inputSize, NULL, maxOutputSize, limitedOutput, tableType, usingExtDict, noDictIssue, acceleration);
+            }
+        }
+        streamPtr->dictionary = (const BYTE*)source;
+        streamPtr->dictSize = (U32)inputSize;
+        return result;
+    }
+}
+
+
+/* Hidden debug function, to force-test external dictionary mode */
+int LZ4_compress_forceExtDict (LZ4_stream_t* LZ4_dict, const char* source, char* dest, int srcSize)
+{
+    LZ4_stream_t_internal* const streamPtr = &LZ4_dict->internal_donotuse;
+    int result;
+
+    LZ4_renormDictT(streamPtr, srcSize);
+
+    if ((streamPtr->dictSize < 64 KB) && (streamPtr->dictSize < streamPtr->currentOffset)) {
+        result = LZ4_compress_generic(streamPtr, source, dest, srcSize, NULL, 0, notLimited, byU32, usingExtDict, dictSmall, 1);
+    } else {
+        result = LZ4_compress_generic(streamPtr, source, dest, srcSize, NULL, 0, notLimited, byU32, usingExtDict, noDictIssue, 1);
+    }
+
+    streamPtr->dictionary = (const BYTE*)source;
+    streamPtr->dictSize = (U32)srcSize;
+
+    return result;
+}
+
+
+/*! LZ4_saveDict() :
+ *  If previously compressed data block is not guaranteed to remain available at its memory location,
+ *  save it into a safer place (char* safeBuffer).
+ *  Note : no need to call LZ4_loadDict() afterwards, dictionary is immediately usable,
+ *         one can therefore call LZ4_compress_fast_continue() right after.
+ * @return : saved dictionary size in bytes (necessarily <= dictSize), or 0 if error.
+ */
+int LZ4_saveDict (LZ4_stream_t* LZ4_dict, char* safeBuffer, int dictSize)
+{
+    LZ4_stream_t_internal* const dict = &LZ4_dict->internal_donotuse;
+
+    DEBUGLOG(5, "LZ4_saveDict : dictSize=%i, safeBuffer=%p", dictSize, safeBuffer);
+
+    if ((U32)dictSize > 64 KB) { dictSize = 64 KB; } /* useless to define a dictionary > 64 KB */
+    if ((U32)dictSize > dict->dictSize) { dictSize = (int)dict->dictSize; }
+
+    if (safeBuffer == NULL) assert(dictSize == 0);
+    if (dictSize > 0) {
+        const BYTE* const previousDictEnd = dict->dictionary + dict->dictSize;
+        assert(dict->dictionary);
+        LZ4_memmove(safeBuffer, previousDictEnd - dictSize, (size_t)dictSize);
+    }
+
+    dict->dictionary = (const BYTE*)safeBuffer;
+    dict->dictSize = (U32)dictSize;
+
+    return dictSize;
+}
+
+
+
+/*-*******************************
+ *  Decompression functions
+ ********************************/
+
+typedef enum { decode_full_block = 0, partial_decode = 1 } earlyEnd_directive;
+
+#undef MIN
+#define MIN(a,b)    ( (a) < (b) ? (a) : (b) )
+
+
+/* variant for decompress_unsafe()
+ * does not know end of input
+ * presumes input is well formed
+ * note : will consume at least one byte */
+static size_t read_long_length_no_check(const BYTE** pp)
+{
+    size_t b, l = 0;
+    do { b = **pp; (*pp)++; l += b; } while (b==255);
+    DEBUGLOG(6, "read_long_length_no_check: +length=%zu using %zu input bytes", l, l/255 + 1)
+    return l;
+}
+
+/* core decoder variant for LZ4_decompress_fast*()
+ * for legacy support only : these entry points are deprecated.
+ * - Presumes input is correctly formed (no defense vs malformed inputs)
+ * - Does not know input size (presume input buffer is "large enough")
+ * - Decompress a full block (only)
+ * @return : nb of bytes read from input.
+ * Note : this variant is not optimized for speed, just for maintenance.
+ *        the goal is to remove support of decompress_fast*() variants by v2.0
+**/
+LZ4_FORCE_INLINE int
+LZ4_decompress_unsafe_generic(
+                 const BYTE* const istart,
+                 BYTE* const ostart,
+                 int decompressedSize,
+
+                 size_t prefixSize,
+                 const BYTE* const dictStart,  /* only if dict==usingExtDict */
+                 const size_t dictSize         /* note: =0 if dictStart==NULL */
+                 )
+{
+    const BYTE* ip = istart;
+    BYTE* op = (BYTE*)ostart;
+    BYTE* const oend = ostart + decompressedSize;
+    const BYTE* const prefixStart = ostart - prefixSize;
+
+    DEBUGLOG(5, "LZ4_decompress_unsafe_generic");
+    if (dictStart == NULL) assert(dictSize == 0);
+
+    while (1) {
+        /* start new sequence */
+        unsigned token = *ip++;
+
+        /* literals */
+        {   size_t ll = token >> ML_BITS;
+            if (ll==15) {
+                /* long literal length */
+                ll += read_long_length_no_check(&ip);
+            }
+            if ((size_t)(oend-op) < ll) return -1; /* output buffer overflow */
+            LZ4_memmove(op, ip, ll); /* support in-place decompression */
+            op += ll;
+            ip += ll;
+            if ((size_t)(oend-op) < MFLIMIT) {
+                if (op==oend) break;  /* end of block */
+                DEBUGLOG(5, "invalid: literals end at distance %zi from end of block", oend-op);
+                /* incorrect end of block :
+                 * last match must start at least MFLIMIT==12 bytes before end of output block */
+                return -1;
+        }   }
+
+        /* match */
+        {   size_t ml = token & 15;
+            size_t const offset = LZ4_readLE16(ip);
+            ip+=2;
+
+            if (ml==15) {
+                /* long literal length */
+                ml += read_long_length_no_check(&ip);
+            }
+            ml += MINMATCH;
+
+            if ((size_t)(oend-op) < ml) return -1; /* output buffer overflow */
+
+            {   const BYTE* match = op - offset;
+
+                /* out of range */
+                if (offset > (size_t)(op - prefixStart) + dictSize) {
+                    DEBUGLOG(6, "offset out of range");
+                    return -1;
+                }
+
+                /* check special case : extDict */
+                if (offset > (size_t)(op - prefixStart)) {
+                    /* extDict scenario */
+                    const BYTE* const dictEnd = dictStart + dictSize;
+                    const BYTE* extMatch = dictEnd - (offset - (size_t)(op-prefixStart));
+                    size_t const extml = (size_t)(dictEnd - extMatch);
+                    if (extml > ml) {
+                        /* match entirely within extDict */
+                        LZ4_memmove(op, extMatch, ml);
+                        op += ml;
+                        ml = 0;
+                    } else {
+                        /* match split between extDict & prefix */
+                        LZ4_memmove(op, extMatch, extml);
+                        op += extml;
+                        ml -= extml;
+                    }
+                    match = prefixStart;
+                }
+
+                /* match copy - slow variant, supporting overlap copy */
+                {   size_t u;
+                    for (u=0; u<ml; u++) {
+                        op[u] = match[u];
+            }   }   }
+            op += ml;
+            if ((size_t)(oend-op) < LASTLITERALS) {
+                DEBUGLOG(5, "invalid: match ends at distance %zi from end of block", oend-op);
+                /* incorrect end of block :
+                 * last match must stop at least LASTLITERALS==5 bytes before end of output block */
+                return -1;
+            }
+        } /* match */
+    } /* main loop */
+    return (int)(ip - istart);
+}
+
+
+/* Read the variable-length literal or match length.
+ *
+ * @ip : input pointer
+ * @ilimit : position after which if length is not decoded, the input is necessarily corrupted.
+ * @initial_check - check ip >= ipmax before start of loop.  Returns initial_error if so.
+ * @error (output) - error code.  Must be set to 0 before call.
+**/
+typedef size_t Rvl_t;
+static const Rvl_t rvl_error = (Rvl_t)(-1);
+LZ4_FORCE_INLINE Rvl_t
+read_variable_length(const BYTE** ip, const BYTE* ilimit,
+                     int initial_check)
+{
+    Rvl_t s, length = 0;
+    assert(ip != NULL);
+    assert(*ip !=  NULL);
+    assert(ilimit != NULL);
+    if (initial_check && unlikely((*ip) >= ilimit)) {    /* read limit reached */
+        return rvl_error;
+    }
+    s = **ip;
+    (*ip)++;
+    length += s;
+    if (unlikely((*ip) > ilimit)) {    /* read limit reached */
+        return rvl_error;
+    }
+    /* accumulator overflow detection (32-bit mode only) */
+    if ((sizeof(length) < 8) && unlikely(length > ((Rvl_t)(-1)/2)) ) {
+        return rvl_error;
+    }
+    if (likely(s != 255)) return length;
+    do {
+        s = **ip;
+        (*ip)++;
+        length += s;
+        if (unlikely((*ip) > ilimit)) {    /* read limit reached */
+            return rvl_error;
+        }
+        /* accumulator overflow detection (32-bit mode only) */
+        if ((sizeof(length) < 8) && unlikely(length > ((Rvl_t)(-1)/2)) ) {
+            return rvl_error;
+        }
+    } while (s == 255);
+
+    return length;
+}
+
+/*! LZ4_decompress_generic() :
+ *  This generic decompression function covers all use cases.
+ *  It shall be instantiated several times, using different sets of directives.
+ *  Note that it is important for performance that this function really get inlined,
+ *  in order to remove useless branches during compilation optimization.
+ */
+LZ4_FORCE_INLINE int
+LZ4_decompress_generic(
+                 const char* const src,
+                 char* const dst,
+                 int srcSize,
+                 int outputSize,         /* If endOnInput==endOnInputSize, this value is `dstCapacity` */
+
+                 earlyEnd_directive partialDecoding,  /* full, partial */
+                 dict_directive dict,                 /* noDict, withPrefix64k, usingExtDict */
+                 const BYTE* const lowPrefix,  /* always <= dst, == dst when no prefix */
+                 const BYTE* const dictStart,  /* only if dict==usingExtDict */
+                 const size_t dictSize         /* note : = 0 if noDict */
+                 )
+{
+    if ((src == NULL) || (outputSize < 0)) { return -1; }
+
+    {   const BYTE* ip = (const BYTE*) src;
+        const BYTE* const iend = ip + srcSize;
+
+        BYTE* op = (BYTE*) dst;
+        BYTE* const oend = op + outputSize;
+        BYTE* cpy;
+
+        const BYTE* const dictEnd = (dictStart == NULL) ? NULL : dictStart + dictSize;
+
+        const int checkOffset = (dictSize < (int)(64 KB));
+
+
+        /* Set up the "end" pointers for the shortcut. */
+        const BYTE* const shortiend = iend - 14 /*maxLL*/ - 2 /*offset*/;
+        const BYTE* const shortoend = oend - 14 /*maxLL*/ - 18 /*maxML*/;
+
+        const BYTE* match;
+        size_t offset;
+        unsigned token;
+        size_t length;
+
+
+        DEBUGLOG(5, "LZ4_decompress_generic (srcSize:%i, dstSize:%i)", srcSize, outputSize);
+
+        /* Special cases */
+        assert(lowPrefix <= op);
+        if (unlikely(outputSize==0)) {
+            /* Empty output buffer */
+            if (partialDecoding) return 0;
+            return ((srcSize==1) && (*ip==0)) ? 0 : -1;
+        }
+        if (unlikely(srcSize==0)) { return -1; }
+
+    /* LZ4_FAST_DEC_LOOP:
+     * designed for modern OoO performance cpus,
+     * where copying reliably 32-bytes is preferable to an unpredictable branch.
+     * note : fast loop may show a regression for some client arm chips. */
+#if LZ4_FAST_DEC_LOOP
+        if ((oend - op) < FASTLOOP_SAFE_DISTANCE) {
+            DEBUGLOG(6, "move to safe decode loop");
+            goto safe_decode;
+        }
+
+        /* Fast loop : decode sequences as long as output < oend-FASTLOOP_SAFE_DISTANCE */
+        DEBUGLOG(6, "using fast decode loop");
+        while (1) {
+            /* Main fastloop assertion: We can always wildcopy FASTLOOP_SAFE_DISTANCE */
+            assert(oend - op >= FASTLOOP_SAFE_DISTANCE);
+            assert(ip < iend);
+            token = *ip++;
+            length = token >> ML_BITS;  /* literal length */
+            DEBUGLOG(7, "blockPos%6u: litLength token = %u", (unsigned)(op-(BYTE*)dst), (unsigned)length);
+
+            /* decode literal length */
+            if (length == RUN_MASK) {
+                size_t const addl = read_variable_length(&ip, iend-RUN_MASK, 1);
+                if (addl == rvl_error) {
+                    DEBUGLOG(6, "error reading long literal length");
+                    goto _output_error;
+                }
+                length += addl;
+                if (unlikely((uptrval)(op)+length<(uptrval)(op))) { goto _output_error; } /* overflow detection */
+                if (unlikely((uptrval)(ip)+length<(uptrval)(ip))) { goto _output_error; } /* overflow detection */
+
+                /* copy literals */
+                LZ4_STATIC_ASSERT(MFLIMIT >= WILDCOPYLENGTH);
+                if ((op+length>oend-32) || (ip+length>iend-32)) { goto safe_literal_copy; }
+                LZ4_wildCopy32(op, ip, op+length);
+                ip += length; op += length;
+            } else if (ip <= iend-(16 + 1/*max lit + offset + nextToken*/)) {
+                /* We don't need to check oend, since we check it once for each loop below */
+                DEBUGLOG(7, "copy %u bytes in a 16-bytes stripe", (unsigned)length);
+                /* Literals can only be <= 14, but hope compilers optimize better when copy by a register size */
+                LZ4_memcpy(op, ip, 16);
+                ip += length; op += length;
+            } else {
+                goto safe_literal_copy;
+            }
+
+            /* get offset */
+            offset = LZ4_readLE16(ip); ip+=2;
+            DEBUGLOG(6, "blockPos%6u: offset = %u", (unsigned)(op-(BYTE*)dst), (unsigned)offset);
+            match = op - offset;
+            assert(match <= op);  /* overflow check */
+
+            /* get matchlength */
+            length = token & ML_MASK;
+            DEBUGLOG(7, "  match length token = %u (len==%u)", (unsigned)length, (unsigned)length+MINMATCH);
+
+            if (length == ML_MASK) {
+                size_t const addl = read_variable_length(&ip, iend - LASTLITERALS + 1, 0);
+                if (addl == rvl_error) {
+                    DEBUGLOG(5, "error reading long match length");
+                    goto _output_error;
+                }
+                length += addl;
+                length += MINMATCH;
+                DEBUGLOG(7, "  long match length == %u", (unsigned)length);
+                if (unlikely((uptrval)(op)+length<(uptrval)op)) { goto _output_error; } /* overflow detection */
+                if (op + length >= oend - FASTLOOP_SAFE_DISTANCE) {
+                    goto safe_match_copy;
+                }
+            } else {
+                length += MINMATCH;
+                if (op + length >= oend - FASTLOOP_SAFE_DISTANCE) {
+                    DEBUGLOG(7, "moving to safe_match_copy (ml==%u)", (unsigned)length);
+                    goto safe_match_copy;
+                }
+
+                /* Fastpath check: skip LZ4_wildCopy32 when true */
+                if ((dict == withPrefix64k) || (match >= lowPrefix)) {
+                    if (offset >= 8) {
+                        assert(match >= lowPrefix);
+                        assert(match <= op);
+                        assert(op + 18 <= oend);
+
+                        LZ4_memcpy(op, match, 8);
+                        LZ4_memcpy(op+8, match+8, 8);
+                        LZ4_memcpy(op+16, match+16, 2);
+                        op += length;
+                        continue;
+            }   }   }
+
+            if ( checkOffset && (unlikely(match + dictSize < lowPrefix)) ) {
+                DEBUGLOG(5, "Error : pos=%zi, offset=%zi => outside buffers", op-lowPrefix, op-match);
+                goto _output_error;
+            }
+            /* match starting within external dictionary */
+            if ((dict==usingExtDict) && (match < lowPrefix)) {
+                assert(dictEnd != NULL);
+                if (unlikely(op+length > oend-LASTLITERALS)) {
+                    if (partialDecoding) {
+                        DEBUGLOG(7, "partialDecoding: dictionary match, close to dstEnd");
+                        length = MIN(length, (size_t)(oend-op));
+                    } else {
+                        DEBUGLOG(6, "end-of-block condition violated")
+                        goto _output_error;
+                }   }
+
+                if (length <= (size_t)(lowPrefix-match)) {
+                    /* match fits entirely within external dictionary : just copy */
+                    LZ4_memmove(op, dictEnd - (lowPrefix-match), length);
+                    op += length;
+                } else {
+                    /* match stretches into both external dictionary and current block */
+                    size_t const copySize = (size_t)(lowPrefix - match);
+                    size_t const restSize = length - copySize;
+                    LZ4_memcpy(op, dictEnd - copySize, copySize);
+                    op += copySize;
+                    if (restSize > (size_t)(op - lowPrefix)) {  /* overlap copy */
+                        BYTE* const endOfMatch = op + restSize;
+                        const BYTE* copyFrom = lowPrefix;
+                        while (op < endOfMatch) { *op++ = *copyFrom++; }
+                    } else {
+                        LZ4_memcpy(op, lowPrefix, restSize);
+                        op += restSize;
+                }   }
+                continue;
+            }
+
+            /* copy match within block */
+            cpy = op + length;
+
+            assert((op <= oend) && (oend-op >= 32));
+            if (unlikely(offset<16)) {
+                LZ4_memcpy_using_offset(op, match, cpy, offset);
+            } else {
+                LZ4_wildCopy32(op, match, cpy);
+            }
+
+            op = cpy;   /* wildcopy correction */
+        }
+    safe_decode:
+#endif
+
+        /* Main Loop : decode remaining sequences where output < FASTLOOP_SAFE_DISTANCE */
+        DEBUGLOG(6, "using safe decode loop");
+        while (1) {
+            assert(ip < iend);
+            token = *ip++;
+            length = token >> ML_BITS;  /* literal length */
+            DEBUGLOG(7, "blockPos%6u: litLength token = %u", (unsigned)(op-(BYTE*)dst), (unsigned)length);
+
+            /* A two-stage shortcut for the most common case:
+             * 1) If the literal length is 0..14, and there is enough space,
+             * enter the shortcut and copy 16 bytes on behalf of the literals
+             * (in the fast mode, only 8 bytes can be safely copied this way).
+             * 2) Further if the match length is 4..18, copy 18 bytes in a similar
+             * manner; but we ensure that there's enough space in the output for
+             * those 18 bytes earlier, upon entering the shortcut (in other words,
+             * there is a combined check for both stages).
+             */
+            if ( (length != RUN_MASK)
+                /* strictly "less than" on input, to re-enter the loop with at least one byte */
+              && likely((ip < shortiend) & (op <= shortoend)) ) {
+                /* Copy the literals */
+                LZ4_memcpy(op, ip, 16);
+                op += length; ip += length;
+
+                /* The second stage: prepare for match copying, decode full info.
+                 * If it doesn't work out, the info won't be wasted. */
+                length = token & ML_MASK; /* match length */
+                DEBUGLOG(7, "blockPos%6u: matchLength token = %u (len=%u)", (unsigned)(op-(BYTE*)dst), (unsigned)length, (unsigned)length + 4);
+                offset = LZ4_readLE16(ip); ip += 2;
+                match = op - offset;
+                assert(match <= op); /* check overflow */
+
+                /* Do not deal with overlapping matches. */
+                if ( (length != ML_MASK)
+                  && (offset >= 8)
+                  && (dict==withPrefix64k || match >= lowPrefix) ) {
+                    /* Copy the match. */
+                    LZ4_memcpy(op + 0, match + 0, 8);
+                    LZ4_memcpy(op + 8, match + 8, 8);
+                    LZ4_memcpy(op +16, match +16, 2);
+                    op += length + MINMATCH;
+                    /* Both stages worked, load the next token. */
+                    continue;
+                }
+
+                /* The second stage didn't work out, but the info is ready.
+                 * Propel it right to the point of match copying. */
+                goto _copy_match;
+            }
+
+            /* decode literal length */
+            if (length == RUN_MASK) {
+                size_t const addl = read_variable_length(&ip, iend-RUN_MASK, 1);
+                if (addl == rvl_error) { goto _output_error; }
+                length += addl;
+                if (unlikely((uptrval)(op)+length<(uptrval)(op))) { goto _output_error; } /* overflow detection */
+                if (unlikely((uptrval)(ip)+length<(uptrval)(ip))) { goto _output_error; } /* overflow detection */
+            }
+
+#if LZ4_FAST_DEC_LOOP
+        safe_literal_copy:
+#endif
+            /* copy literals */
+            cpy = op+length;
+
+            LZ4_STATIC_ASSERT(MFLIMIT >= WILDCOPYLENGTH);
+            if ((cpy>oend-MFLIMIT) || (ip+length>iend-(2+1+LASTLITERALS))) {
+                /* We've either hit the input parsing restriction or the output parsing restriction.
+                 * In the normal scenario, decoding a full block, it must be the last sequence,
+                 * otherwise it's an error (invalid input or dimensions).
+                 * In partialDecoding scenario, it's necessary to ensure there is no buffer overflow.
+                 */
+                if (partialDecoding) {
+                    /* Since we are partial decoding we may be in this block because of the output parsing
+                     * restriction, which is not valid since the output buffer is allowed to be undersized.
+                     */
+                    DEBUGLOG(7, "partialDecoding: copying literals, close to input or output end")
+                    DEBUGLOG(7, "partialDecoding: literal length = %u", (unsigned)length);
+                    DEBUGLOG(7, "partialDecoding: remaining space in dstBuffer : %i", (int)(oend - op));
+                    DEBUGLOG(7, "partialDecoding: remaining space in srcBuffer : %i", (int)(iend - ip));
+                    /* Finishing in the middle of a literals segment,
+                     * due to lack of input.
+                     */
+                    if (ip+length > iend) {
+                        length = (size_t)(iend-ip);
+                        cpy = op + length;
+                    }
+                    /* Finishing in the middle of a literals segment,
+                     * due to lack of output space.
+                     */
+                    if (cpy > oend) {
+                        cpy = oend;
+                        assert(op<=oend);
+                        length = (size_t)(oend-op);
+                    }
+                } else {
+                     /* We must be on the last sequence (or invalid) because of the parsing limitations
+                      * so check that we exactly consume the input and don't overrun the output buffer.
+                      */
+                    if ((ip+length != iend) || (cpy > oend)) {
+                        DEBUGLOG(5, "should have been last run of literals")
+                        DEBUGLOG(5, "ip(%p) + length(%i) = %p != iend (%p)", ip, (int)length, ip+length, iend);
+                        DEBUGLOG(5, "or cpy(%p) > (oend-MFLIMIT)(%p)", cpy, oend-MFLIMIT);
+                        DEBUGLOG(5, "after writing %u bytes / %i bytes available", (unsigned)(op-(BYTE*)dst), outputSize);
+                        goto _output_error;
+                    }
+                }
+                LZ4_memmove(op, ip, length);  /* supports overlapping memory regions, for in-place decompression scenarios */
+                ip += length;
+                op += length;
+                /* Necessarily EOF when !partialDecoding.
+                 * When partialDecoding, it is EOF if we've either
+                 * filled the output buffer or
+                 * can't proceed with reading an offset for following match.
+                 */
+                if (!partialDecoding || (cpy == oend) || (ip >= (iend-2))) {
+                    break;
+                }
+            } else {
+                LZ4_wildCopy8(op, ip, cpy);   /* can overwrite up to 8 bytes beyond cpy */
+                ip += length; op = cpy;
+            }
+
+            /* get offset */
+            offset = LZ4_readLE16(ip); ip+=2;
+            match = op - offset;
+
+            /* get matchlength */
+            length = token & ML_MASK;
+            DEBUGLOG(7, "blockPos%6u: matchLength token = %u", (unsigned)(op-(BYTE*)dst), (unsigned)length);
+
+    _copy_match:
+            if (length == ML_MASK) {
+                size_t const addl = read_variable_length(&ip, iend - LASTLITERALS + 1, 0);
+                if (addl == rvl_error) { goto _output_error; }
+                length += addl;
+                if (unlikely((uptrval)(op)+length<(uptrval)op)) goto _output_error;   /* overflow detection */
+            }
+            length += MINMATCH;
+
+#if LZ4_FAST_DEC_LOOP
+        safe_match_copy:
+#endif
+            if ((checkOffset) && (unlikely(match + dictSize < lowPrefix))) goto _output_error;   /* Error : offset outside buffers */
+            /* match starting within external dictionary */
+            if ((dict==usingExtDict) && (match < lowPrefix)) {
+                assert(dictEnd != NULL);
+                if (unlikely(op+length > oend-LASTLITERALS)) {
+                    if (partialDecoding) length = MIN(length, (size_t)(oend-op));
+                    else goto _output_error;   /* doesn't respect parsing restriction */
+                }
+
+                if (length <= (size_t)(lowPrefix-match)) {
+                    /* match fits entirely within external dictionary : just copy */
+                    LZ4_memmove(op, dictEnd - (lowPrefix-match), length);
+                    op += length;
+                } else {
+                    /* match stretches into both external dictionary and current block */
+                    size_t const copySize = (size_t)(lowPrefix - match);
+                    size_t const restSize = length - copySize;
+                    LZ4_memcpy(op, dictEnd - copySize, copySize);
+                    op += copySize;
+                    if (restSize > (size_t)(op - lowPrefix)) {  /* overlap copy */
+                        BYTE* const endOfMatch = op + restSize;
+                        const BYTE* copyFrom = lowPrefix;
+                        while (op < endOfMatch) *op++ = *copyFrom++;
+                    } else {
+                        LZ4_memcpy(op, lowPrefix, restSize);
+                        op += restSize;
+                }   }
+                continue;
+            }
+            assert(match >= lowPrefix);
+
+            /* copy match within block */
+            cpy = op + length;
+
+            /* partialDecoding : may end anywhere within the block */
+            assert(op<=oend);
+            if (partialDecoding && (cpy > oend-MATCH_SAFEGUARD_DISTANCE)) {
+                size_t const mlen = MIN(length, (size_t)(oend-op));
+                const BYTE* const matchEnd = match + mlen;
+                BYTE* const copyEnd = op + mlen;
+                if (matchEnd > op) {   /* overlap copy */
+                    while (op < copyEnd) { *op++ = *match++; }
+                } else {
+                    LZ4_memcpy(op, match, mlen);
+                }
+                op = copyEnd;
+                if (op == oend) { break; }
+                continue;
+            }
+
+            if (unlikely(offset<8)) {
+                LZ4_write32(op, 0);   /* silence msan warning when offset==0 */
+                op[0] = match[0];
+                op[1] = match[1];
+                op[2] = match[2];
+                op[3] = match[3];
+                match += inc32table[offset];
+                LZ4_memcpy(op+4, match, 4);
+                match -= dec64table[offset];
+            } else {
+                LZ4_memcpy(op, match, 8);
+                match += 8;
+            }
+            op += 8;
+
+            if (unlikely(cpy > oend-MATCH_SAFEGUARD_DISTANCE)) {
+                BYTE* const oCopyLimit = oend - (WILDCOPYLENGTH-1);
+                if (cpy > oend-LASTLITERALS) { goto _output_error; } /* Error : last LASTLITERALS bytes must be literals (uncompressed) */
+                if (op < oCopyLimit) {
+                    LZ4_wildCopy8(op, match, oCopyLimit);
+                    match += oCopyLimit - op;
+                    op = oCopyLimit;
+                }
+                while (op < cpy) { *op++ = *match++; }
+            } else {
+                LZ4_memcpy(op, match, 8);
+                if (length > 16) { LZ4_wildCopy8(op+8, match+8, cpy); }
+            }
+            op = cpy;   /* wildcopy correction */
+        }
+
+        /* end of decoding */
+        DEBUGLOG(5, "decoded %i bytes", (int) (((char*)op)-dst));
+        return (int) (((char*)op)-dst);     /* Nb of output bytes decoded */
+
+        /* Overflow error detected */
+    _output_error:
+        return (int) (-(((const char*)ip)-src))-1;
+    }
+}
+
+
+/*===== Instantiate the API decoding functions. =====*/
+
+LZ4_FORCE_O2
+int LZ4_decompress_safe(const char* source, char* dest, int compressedSize, int maxDecompressedSize)
+{
+    return LZ4_decompress_generic(source, dest, compressedSize, maxDecompressedSize,
+                                  decode_full_block, noDict,
+                                  (BYTE*)dest, NULL, 0);
+}
+
+LZ4_FORCE_O2
+int LZ4_decompress_safe_partial(const char* src, char* dst, int compressedSize, int targetOutputSize, int dstCapacity)
+{
+    dstCapacity = MIN(targetOutputSize, dstCapacity);
+    return LZ4_decompress_generic(src, dst, compressedSize, dstCapacity,
+                                  partial_decode,
+                                  noDict, (BYTE*)dst, NULL, 0);
+}
+
+LZ4_FORCE_O2
+int LZ4_decompress_fast(const char* source, char* dest, int originalSize)
+{
+    DEBUGLOG(5, "LZ4_decompress_fast");
+    return LZ4_decompress_unsafe_generic(
+                (const BYTE*)source, (BYTE*)dest, originalSize,
+                0, NULL, 0);
+}
+
+/*===== Instantiate a few more decoding cases, used more than once. =====*/
+
+LZ4_FORCE_O2 /* Exported, an obsolete API function. */
+int LZ4_decompress_safe_withPrefix64k(const char* source, char* dest, int compressedSize, int maxOutputSize)
+{
+    return LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize,
+                                  decode_full_block, withPrefix64k,
+                                  (BYTE*)dest - 64 KB, NULL, 0);
+}
+
+LZ4_FORCE_O2
+static int LZ4_decompress_safe_partial_withPrefix64k(const char* source, char* dest, int compressedSize, int targetOutputSize, int dstCapacity)
+{
+    dstCapacity = MIN(targetOutputSize, dstCapacity);
+    return LZ4_decompress_generic(source, dest, compressedSize, dstCapacity,
+                                  partial_decode, withPrefix64k,
+                                  (BYTE*)dest - 64 KB, NULL, 0);
+}
+
+/* Another obsolete API function, paired with the previous one. */
+int LZ4_decompress_fast_withPrefix64k(const char* source, char* dest, int originalSize)
+{
+    return LZ4_decompress_unsafe_generic(
+                (const BYTE*)source, (BYTE*)dest, originalSize,
+                64 KB, NULL, 0);
+}
+
+LZ4_FORCE_O2
+static int LZ4_decompress_safe_withSmallPrefix(const char* source, char* dest, int compressedSize, int maxOutputSize,
+                                               size_t prefixSize)
+{
+    return LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize,
+                                  decode_full_block, noDict,
+                                  (BYTE*)dest-prefixSize, NULL, 0);
+}
+
+LZ4_FORCE_O2
+static int LZ4_decompress_safe_partial_withSmallPrefix(const char* source, char* dest, int compressedSize, int targetOutputSize, int dstCapacity,
+                                               size_t prefixSize)
+{
+    dstCapacity = MIN(targetOutputSize, dstCapacity);
+    return LZ4_decompress_generic(source, dest, compressedSize, dstCapacity,
+                                  partial_decode, noDict,
+                                  (BYTE*)dest-prefixSize, NULL, 0);
+}
+
+LZ4_FORCE_O2
+int LZ4_decompress_safe_forceExtDict(const char* source, char* dest,
+                                     int compressedSize, int maxOutputSize,
+                                     const void* dictStart, size_t dictSize)
+{
+    DEBUGLOG(5, "LZ4_decompress_safe_forceExtDict");
+    return LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize,
+                                  decode_full_block, usingExtDict,
+                                  (BYTE*)dest, (const BYTE*)dictStart, dictSize);
+}
+
+LZ4_FORCE_O2
+int LZ4_decompress_safe_partial_forceExtDict(const char* source, char* dest,
+                                     int compressedSize, int targetOutputSize, int dstCapacity,
+                                     const void* dictStart, size_t dictSize)
+{
+    dstCapacity = MIN(targetOutputSize, dstCapacity);
+    return LZ4_decompress_generic(source, dest, compressedSize, dstCapacity,
+                                  partial_decode, usingExtDict,
+                                  (BYTE*)dest, (const BYTE*)dictStart, dictSize);
+}
+
+LZ4_FORCE_O2
+static int LZ4_decompress_fast_extDict(const char* source, char* dest, int originalSize,
+                                       const void* dictStart, size_t dictSize)
+{
+    return LZ4_decompress_unsafe_generic(
+                (const BYTE*)source, (BYTE*)dest, originalSize,
+                0, (const BYTE*)dictStart, dictSize);
+}
+
+/* The "double dictionary" mode, for use with e.g. ring buffers: the first part
+ * of the dictionary is passed as prefix, and the second via dictStart + dictSize.
+ * These routines are used only once, in LZ4_decompress_*_continue().
+ */
+LZ4_FORCE_INLINE
+int LZ4_decompress_safe_doubleDict(const char* source, char* dest, int compressedSize, int maxOutputSize,
+                                   size_t prefixSize, const void* dictStart, size_t dictSize)
+{
+    return LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize,
+                                  decode_full_block, usingExtDict,
+                                  (BYTE*)dest-prefixSize, (const BYTE*)dictStart, dictSize);
+}
+
+/*===== streaming decompression functions =====*/
+
+#if !defined(LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION)
+LZ4_streamDecode_t* LZ4_createStreamDecode(void)
+{
+    LZ4_STATIC_ASSERT(sizeof(LZ4_streamDecode_t) >= sizeof(LZ4_streamDecode_t_internal));
+    return (LZ4_streamDecode_t*) ALLOC_AND_ZERO(sizeof(LZ4_streamDecode_t));
+}
+
+int LZ4_freeStreamDecode (LZ4_streamDecode_t* LZ4_stream)
+{
+    if (LZ4_stream == NULL) { return 0; }  /* support free on NULL */
+    FREEMEM(LZ4_stream);
+    return 0;
+}
+#endif
+
+/*! LZ4_setStreamDecode() :
+ *  Use this function to instruct where to find the dictionary.
+ *  This function is not necessary if previous data is still available where it was decoded.
+ *  Loading a size of 0 is allowed (same effect as no dictionary).
+ * @return : 1 if OK, 0 if error
+ */
+int LZ4_setStreamDecode (LZ4_streamDecode_t* LZ4_streamDecode, const char* dictionary, int dictSize)
+{
+    LZ4_streamDecode_t_internal* lz4sd = &LZ4_streamDecode->internal_donotuse;
+    lz4sd->prefixSize = (size_t)dictSize;
+    if (dictSize) {
+        assert(dictionary != NULL);
+        lz4sd->prefixEnd = (const BYTE*) dictionary + dictSize;
+    } else {
+        lz4sd->prefixEnd = (const BYTE*) dictionary;
+    }
+    lz4sd->externalDict = NULL;
+    lz4sd->extDictSize  = 0;
+    return 1;
+}
+
+/*! LZ4_decoderRingBufferSize() :
+ *  when setting a ring buffer for streaming decompression (optional scenario),
+ *  provides the minimum size of this ring buffer
+ *  to be compatible with any source respecting maxBlockSize condition.
+ *  Note : in a ring buffer scenario,
+ *  blocks are presumed decompressed next to each other.
+ *  When not enough space remains for next block (remainingSize < maxBlockSize),
+ *  decoding resumes from beginning of ring buffer.
+ * @return : minimum ring buffer size,
+ *           or 0 if there is an error (invalid maxBlockSize).
+ */
+int LZ4_decoderRingBufferSize(int maxBlockSize)
+{
+    if (maxBlockSize < 0) return 0;
+    if (maxBlockSize > LZ4_MAX_INPUT_SIZE) return 0;
+    if (maxBlockSize < 16) maxBlockSize = 16;
+    return LZ4_DECODER_RING_BUFFER_SIZE(maxBlockSize);
+}
+
+/*
+*_continue() :
+    These decoding functions allow decompression of multiple blocks in "streaming" mode.
+    Previously decoded blocks must still be available at the memory position where they were decoded.
+    If it's not possible, save the relevant part of decoded data into a safe buffer,
+    and indicate where it stands using LZ4_setStreamDecode()
+*/
+LZ4_FORCE_O2
+int LZ4_decompress_safe_continue (LZ4_streamDecode_t* LZ4_streamDecode, const char* source, char* dest, int compressedSize, int maxOutputSize)
+{
+    LZ4_streamDecode_t_internal* lz4sd = &LZ4_streamDecode->internal_donotuse;
+    int result;
+
+    if (lz4sd->prefixSize == 0) {
+        /* The first call, no dictionary yet. */
+        assert(lz4sd->extDictSize == 0);
+        result = LZ4_decompress_safe(source, dest, compressedSize, maxOutputSize);
+        if (result <= 0) return result;
+        lz4sd->prefixSize = (size_t)result;
+        lz4sd->prefixEnd = (BYTE*)dest + result;
+    } else if (lz4sd->prefixEnd == (BYTE*)dest) {
+        /* They're rolling the current segment. */
+        if (lz4sd->prefixSize >= 64 KB - 1)
+            result = LZ4_decompress_safe_withPrefix64k(source, dest, compressedSize, maxOutputSize);
+        else if (lz4sd->extDictSize == 0)
+            result = LZ4_decompress_safe_withSmallPrefix(source, dest, compressedSize, maxOutputSize,
+                                                         lz4sd->prefixSize);
+        else
+            result = LZ4_decompress_safe_doubleDict(source, dest, compressedSize, maxOutputSize,
+                                                    lz4sd->prefixSize, lz4sd->externalDict, lz4sd->extDictSize);
+        if (result <= 0) return result;
+        lz4sd->prefixSize += (size_t)result;
+        lz4sd->prefixEnd  += result;
+    } else {
+        /* The buffer wraps around, or they're switching to another buffer. */
+        lz4sd->extDictSize = lz4sd->prefixSize;
+        lz4sd->externalDict = lz4sd->prefixEnd - lz4sd->extDictSize;
+        result = LZ4_decompress_safe_forceExtDict(source, dest, compressedSize, maxOutputSize,
+                                                  lz4sd->externalDict, lz4sd->extDictSize);
+        if (result <= 0) return result;
+        lz4sd->prefixSize = (size_t)result;
+        lz4sd->prefixEnd  = (BYTE*)dest + result;
+    }
+
+    return result;
+}
+
+LZ4_FORCE_O2 int
+LZ4_decompress_fast_continue (LZ4_streamDecode_t* LZ4_streamDecode,
+                        const char* source, char* dest, int originalSize)
+{
+    LZ4_streamDecode_t_internal* const lz4sd =
+        (assert(LZ4_streamDecode!=NULL), &LZ4_streamDecode->internal_donotuse);
+    int result;
+
+    DEBUGLOG(5, "LZ4_decompress_fast_continue (toDecodeSize=%i)", originalSize);
+    assert(originalSize >= 0);
+
+    if (lz4sd->prefixSize == 0) {
+        DEBUGLOG(5, "first invocation : no prefix nor extDict");
+        assert(lz4sd->extDictSize == 0);
+        result = LZ4_decompress_fast(source, dest, originalSize);
+        if (result <= 0) return result;
+        lz4sd->prefixSize = (size_t)originalSize;
+        lz4sd->prefixEnd = (BYTE*)dest + originalSize;
+    } else if (lz4sd->prefixEnd == (BYTE*)dest) {
+        DEBUGLOG(5, "continue using existing prefix");
+        result = LZ4_decompress_unsafe_generic(
+                        (const BYTE*)source, (BYTE*)dest, originalSize,
+                        lz4sd->prefixSize,
+                        lz4sd->externalDict, lz4sd->extDictSize);
+        if (result <= 0) return result;
+        lz4sd->prefixSize += (size_t)originalSize;
+        lz4sd->prefixEnd  += originalSize;
+    } else {
+        DEBUGLOG(5, "prefix becomes extDict");
+        lz4sd->extDictSize = lz4sd->prefixSize;
+        lz4sd->externalDict = lz4sd->prefixEnd - lz4sd->extDictSize;
+        result = LZ4_decompress_fast_extDict(source, dest, originalSize,
+                                             lz4sd->externalDict, lz4sd->extDictSize);
+        if (result <= 0) return result;
+        lz4sd->prefixSize = (size_t)originalSize;
+        lz4sd->prefixEnd  = (BYTE*)dest + originalSize;
+    }
+
+    return result;
+}
+
+
+/*
+Advanced decoding functions :
+*_usingDict() :
+    These decoding functions work the same as "_continue" ones,
+    the dictionary must be explicitly provided within parameters
+*/
+
+int LZ4_decompress_safe_usingDict(const char* source, char* dest, int compressedSize, int maxOutputSize, const char* dictStart, int dictSize)
+{
+    if (dictSize==0)
+        return LZ4_decompress_safe(source, dest, compressedSize, maxOutputSize);
+    if (dictStart+dictSize == dest) {
+        if (dictSize >= 64 KB - 1) {
+            return LZ4_decompress_safe_withPrefix64k(source, dest, compressedSize, maxOutputSize);
+        }
+        assert(dictSize >= 0);
+        return LZ4_decompress_safe_withSmallPrefix(source, dest, compressedSize, maxOutputSize, (size_t)dictSize);
+    }
+    assert(dictSize >= 0);
+    return LZ4_decompress_safe_forceExtDict(source, dest, compressedSize, maxOutputSize, dictStart, (size_t)dictSize);
+}
+
+int LZ4_decompress_safe_partial_usingDict(const char* source, char* dest, int compressedSize, int targetOutputSize, int dstCapacity, const char* dictStart, int dictSize)
+{
+    if (dictSize==0)
+        return LZ4_decompress_safe_partial(source, dest, compressedSize, targetOutputSize, dstCapacity);
+    if (dictStart+dictSize == dest) {
+        if (dictSize >= 64 KB - 1) {
+            return LZ4_decompress_safe_partial_withPrefix64k(source, dest, compressedSize, targetOutputSize, dstCapacity);
+        }
+        assert(dictSize >= 0);
+        return LZ4_decompress_safe_partial_withSmallPrefix(source, dest, compressedSize, targetOutputSize, dstCapacity, (size_t)dictSize);
+    }
+    assert(dictSize >= 0);
+    return LZ4_decompress_safe_partial_forceExtDict(source, dest, compressedSize, targetOutputSize, dstCapacity, dictStart, (size_t)dictSize);
+}
+
+int LZ4_decompress_fast_usingDict(const char* source, char* dest, int originalSize, const char* dictStart, int dictSize)
+{
+    if (dictSize==0 || dictStart+dictSize == dest)
+        return LZ4_decompress_unsafe_generic(
+                        (const BYTE*)source, (BYTE*)dest, originalSize,
+                        (size_t)dictSize, NULL, 0);
+    assert(dictSize >= 0);
+    return LZ4_decompress_fast_extDict(source, dest, originalSize, dictStart, (size_t)dictSize);
+}
+
+
+/*=*************************************************
+*  Obsolete Functions
+***************************************************/
+/* obsolete compression functions */
+int LZ4_compress_limitedOutput(const char* source, char* dest, int inputSize, int maxOutputSize)
+{
+    return LZ4_compress_default(source, dest, inputSize, maxOutputSize);
+}
+int LZ4_compress(const char* src, char* dest, int srcSize)
+{
+    return LZ4_compress_default(src, dest, srcSize, LZ4_compressBound(srcSize));
+}
+int LZ4_compress_limitedOutput_withState (void* state, const char* src, char* dst, int srcSize, int dstSize)
+{
+    return LZ4_compress_fast_extState(state, src, dst, srcSize, dstSize, 1);
+}
+int LZ4_compress_withState (void* state, const char* src, char* dst, int srcSize)
+{
+    return LZ4_compress_fast_extState(state, src, dst, srcSize, LZ4_compressBound(srcSize), 1);
+}
+int LZ4_compress_limitedOutput_continue (LZ4_stream_t* LZ4_stream, const char* src, char* dst, int srcSize, int dstCapacity)
+{
+    return LZ4_compress_fast_continue(LZ4_stream, src, dst, srcSize, dstCapacity, 1);
+}
+int LZ4_compress_continue (LZ4_stream_t* LZ4_stream, const char* source, char* dest, int inputSize)
+{
+    return LZ4_compress_fast_continue(LZ4_stream, source, dest, inputSize, LZ4_compressBound(inputSize), 1);
+}
+
+/*
+These decompression functions are deprecated and should no longer be used.
+They are only provided here for compatibility with older user programs.
+- LZ4_uncompress is totally equivalent to LZ4_decompress_fast
+- LZ4_uncompress_unknownOutputSize is totally equivalent to LZ4_decompress_safe
+*/
+int LZ4_uncompress (const char* source, char* dest, int outputSize)
+{
+    return LZ4_decompress_fast(source, dest, outputSize);
+}
+int LZ4_uncompress_unknownOutputSize (const char* source, char* dest, int isize, int maxOutputSize)
+{
+    return LZ4_decompress_safe(source, dest, isize, maxOutputSize);
+}
+
+/* Obsolete Streaming functions */
+
+int LZ4_sizeofStreamState(void) { return sizeof(LZ4_stream_t); }
+
+int LZ4_resetStreamState(void* state, char* inputBuffer)
+{
+    (void)inputBuffer;
+    LZ4_resetStream((LZ4_stream_t*)state);
+    return 0;
+}
+
+#if !defined(LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION)
+void* LZ4_create (char* inputBuffer)
+{
+    (void)inputBuffer;
+    return LZ4_createStream();
+}
+#endif
+
+char* LZ4_slideInputBuffer (void* state)
+{
+    /* avoid const char * -> char * conversion warning */
+    return (char *)(uptrval)((LZ4_stream_t*)state)->internal_donotuse.dictionary;
+}
+
+#endif   /* LZ4_COMMONDEFS_ONLY */
diff --git a/lib/lz4/upstream/lz4.h b/lib/lz4/upstream/lz4.h
new file mode 100644
index 0000000000000000000000000000000000000000..80e3e5ca04d22db9ec4980dfc43672a13984aa15
--- /dev/null
+++ b/lib/lz4/upstream/lz4.h
@@ -0,0 +1,884 @@
+/*
+ *  LZ4 - Fast LZ compression algorithm
+ *  Header File
+ *  Copyright (C) 2011-2023, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+    - LZ4 homepage : http://www.lz4.org
+    - LZ4 source repository : https://github.com/lz4/lz4
+*/
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#ifndef LZ4_H_2983827168210
+#define LZ4_H_2983827168210
+
+/* --- Dependency --- */
+#include <stddef.h>   /* size_t */
+
+
+/**
+  Introduction
+
+  LZ4 is lossless compression algorithm, providing compression speed >500 MB/s per core,
+  scalable with multi-cores CPU. It features an extremely fast decoder, with speed in
+  multiple GB/s per core, typically reaching RAM speed limits on multi-core systems.
+
+  The LZ4 compression library provides in-memory compression and decompression functions.
+  It gives full buffer control to user.
+  Compression can be done in:
+    - a single step (described as Simple Functions)
+    - a single step, reusing a context (described in Advanced Functions)
+    - unbounded multiple steps (described as Streaming compression)
+
+  lz4.h generates and decodes LZ4-compressed blocks (doc/lz4_Block_format.md).
+  Decompressing such a compressed block requires additional metadata.
+  Exact metadata depends on exact decompression function.
+  For the typical case of LZ4_decompress_safe(),
+  metadata includes block's compressed size, and maximum bound of decompressed size.
+  Each application is free to encode and pass such metadata in whichever way it wants.
+
+  lz4.h only handle blocks, it can not generate Frames.
+
+  Blocks are different from Frames (doc/lz4_Frame_format.md).
+  Frames bundle both blocks and metadata in a specified manner.
+  Embedding metadata is required for compressed data to be self-contained and portable.
+  Frame format is delivered through a companion API, declared in lz4frame.h.
+  The `lz4` CLI can only manage frames.
+*/
+
+/*^***************************************************************
+*  Export parameters
+*****************************************************************/
+/*
+*  LZ4_DLL_EXPORT :
+*  Enable exporting of functions when building a Windows DLL
+*  LZ4LIB_VISIBILITY :
+*  Control library symbols visibility.
+*/
+#ifndef LZ4LIB_VISIBILITY
+#  if defined(__GNUC__) && (__GNUC__ >= 4)
+#    define LZ4LIB_VISIBILITY __attribute__ ((visibility ("default")))
+#  else
+#    define LZ4LIB_VISIBILITY
+#  endif
+#endif
+#if defined(LZ4_DLL_EXPORT) && (LZ4_DLL_EXPORT==1)
+#  define LZ4LIB_API __declspec(dllexport) LZ4LIB_VISIBILITY
+#elif defined(LZ4_DLL_IMPORT) && (LZ4_DLL_IMPORT==1)
+#  define LZ4LIB_API __declspec(dllimport) LZ4LIB_VISIBILITY /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
+#else
+#  define LZ4LIB_API LZ4LIB_VISIBILITY
+#endif
+
+/*! LZ4_FREESTANDING :
+ *  When this macro is set to 1, it enables "freestanding mode" that is
+ *  suitable for typical freestanding environment which doesn't support
+ *  standard C library.
+ *
+ *  - LZ4_FREESTANDING is a compile-time switch.
+ *  - It requires the following macros to be defined:
+ *    LZ4_memcpy, LZ4_memmove, LZ4_memset.
+ *  - It only enables LZ4/HC functions which don't use heap.
+ *    All LZ4F_* functions are not supported.
+ *  - See tests/freestanding.c to check its basic setup.
+ */
+#if defined(LZ4_FREESTANDING) && (LZ4_FREESTANDING == 1)
+#  define LZ4_HEAPMODE 0
+#  define LZ4HC_HEAPMODE 0
+#  define LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION 1
+#  if !defined(LZ4_memcpy)
+#    error "LZ4_FREESTANDING requires macro 'LZ4_memcpy'."
+#  endif
+#  if !defined(LZ4_memset)
+#    error "LZ4_FREESTANDING requires macro 'LZ4_memset'."
+#  endif
+#  if !defined(LZ4_memmove)
+#    error "LZ4_FREESTANDING requires macro 'LZ4_memmove'."
+#  endif
+#elif ! defined(LZ4_FREESTANDING)
+#  define LZ4_FREESTANDING 0
+#endif
+
+
+/*------   Version   ------*/
+#define LZ4_VERSION_MAJOR    1    /* for breaking interface changes  */
+#define LZ4_VERSION_MINOR   10    /* for new (non-breaking) interface capabilities */
+#define LZ4_VERSION_RELEASE  0    /* for tweaks, bug-fixes, or development */
+
+#define LZ4_VERSION_NUMBER (LZ4_VERSION_MAJOR *100*100 + LZ4_VERSION_MINOR *100 + LZ4_VERSION_RELEASE)
+
+#define LZ4_LIB_VERSION LZ4_VERSION_MAJOR.LZ4_VERSION_MINOR.LZ4_VERSION_RELEASE
+#define LZ4_QUOTE(str) #str
+#define LZ4_EXPAND_AND_QUOTE(str) LZ4_QUOTE(str)
+#define LZ4_VERSION_STRING LZ4_EXPAND_AND_QUOTE(LZ4_LIB_VERSION)  /* requires v1.7.3+ */
+
+LZ4LIB_API int LZ4_versionNumber (void);  /**< library version number; useful to check dll version; requires v1.3.0+ */
+LZ4LIB_API const char* LZ4_versionString (void);   /**< library version string; useful to check dll version; requires v1.7.5+ */
+
+
+/*-************************************
+*  Tuning memory usage
+**************************************/
+/*!
+ * LZ4_MEMORY_USAGE :
+ * Can be selected at compile time, by setting LZ4_MEMORY_USAGE.
+ * Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB)
+ * Increasing memory usage improves compression ratio, generally at the cost of speed.
+ * Reduced memory usage may improve speed at the cost of ratio, thanks to better cache locality.
+ * Default value is 14, for 16KB, which nicely fits into most L1 caches.
+ */
+#ifndef LZ4_MEMORY_USAGE
+# define LZ4_MEMORY_USAGE LZ4_MEMORY_USAGE_DEFAULT
+#endif
+
+/* These are absolute limits, they should not be changed by users */
+#define LZ4_MEMORY_USAGE_MIN 10
+#define LZ4_MEMORY_USAGE_DEFAULT 14
+#define LZ4_MEMORY_USAGE_MAX 20
+
+#if (LZ4_MEMORY_USAGE < LZ4_MEMORY_USAGE_MIN)
+#  error "LZ4_MEMORY_USAGE is too small !"
+#endif
+
+#if (LZ4_MEMORY_USAGE > LZ4_MEMORY_USAGE_MAX)
+#  error "LZ4_MEMORY_USAGE is too large !"
+#endif
+
+/*-************************************
+*  Simple Functions
+**************************************/
+/*! LZ4_compress_default() :
+ *  Compresses 'srcSize' bytes from buffer 'src'
+ *  into already allocated 'dst' buffer of size 'dstCapacity'.
+ *  Compression is guaranteed to succeed if 'dstCapacity' >= LZ4_compressBound(srcSize).
+ *  It also runs faster, so it's a recommended setting.
+ *  If the function cannot compress 'src' into a more limited 'dst' budget,
+ *  compression stops *immediately*, and the function result is zero.
+ *  In which case, 'dst' content is undefined (invalid).
+ *      srcSize : max supported value is LZ4_MAX_INPUT_SIZE.
+ *      dstCapacity : size of buffer 'dst' (which must be already allocated)
+ *     @return  : the number of bytes written into buffer 'dst' (necessarily <= dstCapacity)
+ *                or 0 if compression fails
+ * Note : This function is protected against buffer overflow scenarios (never writes outside 'dst' buffer, nor read outside 'source' buffer).
+ */
+LZ4LIB_API int LZ4_compress_default(const char* src, char* dst, int srcSize, int dstCapacity);
+
+/*! LZ4_decompress_safe() :
+ * @compressedSize : is the exact complete size of the compressed block.
+ * @dstCapacity : is the size of destination buffer (which must be already allocated),
+ *                presumed an upper bound of decompressed size.
+ * @return : the number of bytes decompressed into destination buffer (necessarily <= dstCapacity)
+ *           If destination buffer is not large enough, decoding will stop and output an error code (negative value).
+ *           If the source stream is detected malformed, the function will stop decoding and return a negative result.
+ * Note 1 : This function is protected against malicious data packets :
+ *          it will never writes outside 'dst' buffer, nor read outside 'source' buffer,
+ *          even if the compressed block is maliciously modified to order the decoder to do these actions.
+ *          In such case, the decoder stops immediately, and considers the compressed block malformed.
+ * Note 2 : compressedSize and dstCapacity must be provided to the function, the compressed block does not contain them.
+ *          The implementation is free to send / store / derive this information in whichever way is most beneficial.
+ *          If there is a need for a different format which bundles together both compressed data and its metadata, consider looking at lz4frame.h instead.
+ */
+LZ4LIB_API int LZ4_decompress_safe (const char* src, char* dst, int compressedSize, int dstCapacity);
+
+
+/*-************************************
+*  Advanced Functions
+**************************************/
+#define LZ4_MAX_INPUT_SIZE        0x7E000000   /* 2 113 929 216 bytes */
+#define LZ4_COMPRESSBOUND(isize)  ((unsigned)(isize) > (unsigned)LZ4_MAX_INPUT_SIZE ? 0 : (isize) + ((isize)/255) + 16)
+
+/*! LZ4_compressBound() :
+    Provides the maximum size that LZ4 compression may output in a "worst case" scenario (input data not compressible)
+    This function is primarily useful for memory allocation purposes (destination buffer size).
+    Macro LZ4_COMPRESSBOUND() is also provided for compilation-time evaluation (stack memory allocation for example).
+    Note that LZ4_compress_default() compresses faster when dstCapacity is >= LZ4_compressBound(srcSize)
+        inputSize  : max supported value is LZ4_MAX_INPUT_SIZE
+        return : maximum output size in a "worst case" scenario
+              or 0, if input size is incorrect (too large or negative)
+*/
+LZ4LIB_API int LZ4_compressBound(int inputSize);
+
+/*! LZ4_compress_fast() :
+    Same as LZ4_compress_default(), but allows selection of "acceleration" factor.
+    The larger the acceleration value, the faster the algorithm, but also the lesser the compression.
+    It's a trade-off. It can be fine tuned, with each successive value providing roughly +~3% to speed.
+    An acceleration value of "1" is the same as regular LZ4_compress_default()
+    Values <= 0 will be replaced by LZ4_ACCELERATION_DEFAULT (currently == 1, see lz4.c).
+    Values > LZ4_ACCELERATION_MAX will be replaced by LZ4_ACCELERATION_MAX (currently == 65537, see lz4.c).
+*/
+LZ4LIB_API int LZ4_compress_fast (const char* src, char* dst, int srcSize, int dstCapacity, int acceleration);
+
+
+/*! LZ4_compress_fast_extState() :
+ *  Same as LZ4_compress_fast(), using an externally allocated memory space for its state.
+ *  Use LZ4_sizeofState() to know how much memory must be allocated,
+ *  and allocate it on 8-bytes boundaries (using `malloc()` typically).
+ *  Then, provide this buffer as `void* state` to compression function.
+ */
+LZ4LIB_API int LZ4_sizeofState(void);
+LZ4LIB_API int LZ4_compress_fast_extState (void* state, const char* src, char* dst, int srcSize, int dstCapacity, int acceleration);
+
+/*! LZ4_compress_destSize() :
+ *  Reverse the logic : compresses as much data as possible from 'src' buffer
+ *  into already allocated buffer 'dst', of size >= 'dstCapacity'.
+ *  This function either compresses the entire 'src' content into 'dst' if it's large enough,
+ *  or fill 'dst' buffer completely with as much data as possible from 'src'.
+ *  note: acceleration parameter is fixed to "default".
+ *
+ * *srcSizePtr : in+out parameter. Initially contains size of input.
+ *               Will be modified to indicate how many bytes where read from 'src' to fill 'dst'.
+ *               New value is necessarily <= input value.
+ * @return : Nb bytes written into 'dst' (necessarily <= dstCapacity)
+ *           or 0 if compression fails.
+ *
+ * Note : from v1.8.2 to v1.9.1, this function had a bug (fixed in v1.9.2+):
+ *        the produced compressed content could, in specific circumstances,
+ *        require to be decompressed into a destination buffer larger
+ *        by at least 1 byte than the content to decompress.
+ *        If an application uses `LZ4_compress_destSize()`,
+ *        it's highly recommended to update liblz4 to v1.9.2 or better.
+ *        If this can't be done or ensured,
+ *        the receiving decompression function should provide
+ *        a dstCapacity which is > decompressedSize, by at least 1 byte.
+ *        See https://github.com/lz4/lz4/issues/859 for details
+ */
+LZ4LIB_API int LZ4_compress_destSize(const char* src, char* dst, int* srcSizePtr, int targetDstSize);
+
+/*! LZ4_decompress_safe_partial() :
+ *  Decompress an LZ4 compressed block, of size 'srcSize' at position 'src',
+ *  into destination buffer 'dst' of size 'dstCapacity'.
+ *  Up to 'targetOutputSize' bytes will be decoded.
+ *  The function stops decoding on reaching this objective.
+ *  This can be useful to boost performance
+ *  whenever only the beginning of a block is required.
+ *
+ * @return : the number of bytes decoded in `dst` (necessarily <= targetOutputSize)
+ *           If source stream is detected malformed, function returns a negative result.
+ *
+ *  Note 1 : @return can be < targetOutputSize, if compressed block contains less data.
+ *
+ *  Note 2 : targetOutputSize must be <= dstCapacity
+ *
+ *  Note 3 : this function effectively stops decoding on reaching targetOutputSize,
+ *           so dstCapacity is kind of redundant.
+ *           This is because in older versions of this function,
+ *           decoding operation would still write complete sequences.
+ *           Therefore, there was no guarantee that it would stop writing at exactly targetOutputSize,
+ *           it could write more bytes, though only up to dstCapacity.
+ *           Some "margin" used to be required for this operation to work properly.
+ *           Thankfully, this is no longer necessary.
+ *           The function nonetheless keeps the same signature, in an effort to preserve API compatibility.
+ *
+ *  Note 4 : If srcSize is the exact size of the block,
+ *           then targetOutputSize can be any value,
+ *           including larger than the block's decompressed size.
+ *           The function will, at most, generate block's decompressed size.
+ *
+ *  Note 5 : If srcSize is _larger_ than block's compressed size,
+ *           then targetOutputSize **MUST** be <= block's decompressed size.
+ *           Otherwise, *silent corruption will occur*.
+ */
+LZ4LIB_API int LZ4_decompress_safe_partial (const char* src, char* dst, int srcSize, int targetOutputSize, int dstCapacity);
+
+
+/*-*********************************************
+*  Streaming Compression Functions
+***********************************************/
+typedef union LZ4_stream_u LZ4_stream_t;  /* incomplete type (defined later) */
+
+/*!
+ Note about RC_INVOKED
+
+ - RC_INVOKED is predefined symbol of rc.exe (the resource compiler which is part of MSVC/Visual Studio).
+   https://docs.microsoft.com/en-us/windows/win32/menurc/predefined-macros
+
+ - Since rc.exe is a legacy compiler, it truncates long symbol (> 30 chars)
+   and reports warning "RC4011: identifier truncated".
+
+ - To eliminate the warning, we surround long preprocessor symbol with
+   "#if !defined(RC_INVOKED) ... #endif" block that means
+   "skip this block when rc.exe is trying to read it".
+*/
+#if !defined(RC_INVOKED) /* https://docs.microsoft.com/en-us/windows/win32/menurc/predefined-macros */
+#if !defined(LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION)
+LZ4LIB_API LZ4_stream_t* LZ4_createStream(void);
+LZ4LIB_API int           LZ4_freeStream (LZ4_stream_t* streamPtr);
+#endif /* !defined(LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION) */
+#endif
+
+/*! LZ4_resetStream_fast() : v1.9.0+
+ *  Use this to prepare an LZ4_stream_t for a new chain of dependent blocks
+ *  (e.g., LZ4_compress_fast_continue()).
+ *
+ *  An LZ4_stream_t must be initialized once before usage.
+ *  This is automatically done when created by LZ4_createStream().
+ *  However, should the LZ4_stream_t be simply declared on stack (for example),
+ *  it's necessary to initialize it first, using LZ4_initStream().
+ *
+ *  After init, start any new stream with LZ4_resetStream_fast().
+ *  A same LZ4_stream_t can be re-used multiple times consecutively
+ *  and compress multiple streams,
+ *  provided that it starts each new stream with LZ4_resetStream_fast().
+ *
+ *  LZ4_resetStream_fast() is much faster than LZ4_initStream(),
+ *  but is not compatible with memory regions containing garbage data.
+ *
+ *  Note: it's only useful to call LZ4_resetStream_fast()
+ *        in the context of streaming compression.
+ *        The *extState* functions perform their own resets.
+ *        Invoking LZ4_resetStream_fast() before is redundant, and even counterproductive.
+ */
+LZ4LIB_API void LZ4_resetStream_fast (LZ4_stream_t* streamPtr);
+
+/*! LZ4_loadDict() :
+ *  Use this function to reference a static dictionary into LZ4_stream_t.
+ *  The dictionary must remain available during compression.
+ *  LZ4_loadDict() triggers a reset, so any previous data will be forgotten.
+ *  The same dictionary will have to be loaded on decompression side for successful decoding.
+ *  Dictionary are useful for better compression of small data (KB range).
+ *  While LZ4 itself accepts any input as dictionary, dictionary efficiency is also a topic.
+ *  When in doubt, employ the Zstandard's Dictionary Builder.
+ *  Loading a size of 0 is allowed, and is the same as reset.
+ * @return : loaded dictionary size, in bytes (note: only the last 64 KB are loaded)
+ */
+LZ4LIB_API int LZ4_loadDict (LZ4_stream_t* streamPtr, const char* dictionary, int dictSize);
+
+/*! LZ4_loadDictSlow() : v1.10.0+
+ *  Same as LZ4_loadDict(),
+ *  but uses a bit more cpu to reference the dictionary content more thoroughly.
+ *  This is expected to slightly improve compression ratio.
+ *  The extra-cpu cost is likely worth it if the dictionary is re-used across multiple sessions.
+ * @return : loaded dictionary size, in bytes (note: only the last 64 KB are loaded)
+ */
+LZ4LIB_API int LZ4_loadDictSlow(LZ4_stream_t* streamPtr, const char* dictionary, int dictSize);
+
+/*! LZ4_attach_dictionary() : stable since v1.10.0
+ *
+ *  This allows efficient re-use of a static dictionary multiple times.
+ *
+ *  Rather than re-loading the dictionary buffer into a working context before
+ *  each compression, or copying a pre-loaded dictionary's LZ4_stream_t into a
+ *  working LZ4_stream_t, this function introduces a no-copy setup mechanism,
+ *  in which the working stream references @dictionaryStream in-place.
+ *
+ *  Several assumptions are made about the state of @dictionaryStream.
+ *  Currently, only states which have been prepared by LZ4_loadDict() or
+ *  LZ4_loadDictSlow() should be expected to work.
+ *
+ *  Alternatively, the provided @dictionaryStream may be NULL,
+ *  in which case any existing dictionary stream is unset.
+ *
+ *  If a dictionary is provided, it replaces any pre-existing stream history.
+ *  The dictionary contents are the only history that can be referenced and
+ *  logically immediately precede the data compressed in the first subsequent
+ *  compression call.
+ *
+ *  The dictionary will only remain attached to the working stream through the
+ *  first compression call, at the end of which it is cleared.
+ * @dictionaryStream stream (and source buffer) must remain in-place / accessible / unchanged
+ *  through the completion of the compression session.
+ *
+ *  Note: there is no equivalent LZ4_attach_*() method on the decompression side
+ *  because there is no initialization cost, hence no need to share the cost across multiple sessions.
+ *  To decompress LZ4 blocks using dictionary, attached or not,
+ *  just employ the regular LZ4_setStreamDecode() for streaming,
+ *  or the stateless LZ4_decompress_safe_usingDict() for one-shot decompression.
+ */
+LZ4LIB_API void
+LZ4_attach_dictionary(LZ4_stream_t* workingStream,
+                const LZ4_stream_t* dictionaryStream);
+
+/*! LZ4_compress_fast_continue() :
+ *  Compress 'src' content using data from previously compressed blocks, for better compression ratio.
+ * 'dst' buffer must be already allocated.
+ *  If dstCapacity >= LZ4_compressBound(srcSize), compression is guaranteed to succeed, and runs faster.
+ *
+ * @return : size of compressed block
+ *           or 0 if there is an error (typically, cannot fit into 'dst').
+ *
+ *  Note 1 : Each invocation to LZ4_compress_fast_continue() generates a new block.
+ *           Each block has precise boundaries.
+ *           Each block must be decompressed separately, calling LZ4_decompress_*() with relevant metadata.
+ *           It's not possible to append blocks together and expect a single invocation of LZ4_decompress_*() to decompress them together.
+ *
+ *  Note 2 : The previous 64KB of source data is __assumed__ to remain present, unmodified, at same address in memory !
+ *
+ *  Note 3 : When input is structured as a double-buffer, each buffer can have any size, including < 64 KB.
+ *           Make sure that buffers are separated, by at least one byte.
+ *           This construction ensures that each block only depends on previous block.
+ *
+ *  Note 4 : If input buffer is a ring-buffer, it can have any size, including < 64 KB.
+ *
+ *  Note 5 : After an error, the stream status is undefined (invalid), it can only be reset or freed.
+ */
+LZ4LIB_API int LZ4_compress_fast_continue (LZ4_stream_t* streamPtr, const char* src, char* dst, int srcSize, int dstCapacity, int acceleration);
+
+/*! LZ4_saveDict() :
+ *  If last 64KB data cannot be guaranteed to remain available at its current memory location,
+ *  save it into a safer place (char* safeBuffer).
+ *  This is schematically equivalent to a memcpy() followed by LZ4_loadDict(),
+ *  but is much faster, because LZ4_saveDict() doesn't need to rebuild tables.
+ * @return : saved dictionary size in bytes (necessarily <= maxDictSize), or 0 if error.
+ */
+LZ4LIB_API int LZ4_saveDict (LZ4_stream_t* streamPtr, char* safeBuffer, int maxDictSize);
+
+
+/*-**********************************************
+*  Streaming Decompression Functions
+*  Bufferless synchronous API
+************************************************/
+typedef union LZ4_streamDecode_u LZ4_streamDecode_t;   /* tracking context */
+
+/*! LZ4_createStreamDecode() and LZ4_freeStreamDecode() :
+ *  creation / destruction of streaming decompression tracking context.
+ *  A tracking context can be re-used multiple times.
+ */
+#if !defined(RC_INVOKED) /* https://docs.microsoft.com/en-us/windows/win32/menurc/predefined-macros */
+#if !defined(LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION)
+LZ4LIB_API LZ4_streamDecode_t* LZ4_createStreamDecode(void);
+LZ4LIB_API int                 LZ4_freeStreamDecode (LZ4_streamDecode_t* LZ4_stream);
+#endif /* !defined(LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION) */
+#endif
+
+/*! LZ4_setStreamDecode() :
+ *  An LZ4_streamDecode_t context can be allocated once and re-used multiple times.
+ *  Use this function to start decompression of a new stream of blocks.
+ *  A dictionary can optionally be set. Use NULL or size 0 for a reset order.
+ *  Dictionary is presumed stable : it must remain accessible and unmodified during next decompression.
+ * @return : 1 if OK, 0 if error
+ */
+LZ4LIB_API int LZ4_setStreamDecode (LZ4_streamDecode_t* LZ4_streamDecode, const char* dictionary, int dictSize);
+
+/*! LZ4_decoderRingBufferSize() : v1.8.2+
+ *  Note : in a ring buffer scenario (optional),
+ *  blocks are presumed decompressed next to each other
+ *  up to the moment there is not enough remaining space for next block (remainingSize < maxBlockSize),
+ *  at which stage it resumes from beginning of ring buffer.
+ *  When setting such a ring buffer for streaming decompression,
+ *  provides the minimum size of this ring buffer
+ *  to be compatible with any source respecting maxBlockSize condition.
+ * @return : minimum ring buffer size,
+ *           or 0 if there is an error (invalid maxBlockSize).
+ */
+LZ4LIB_API int LZ4_decoderRingBufferSize(int maxBlockSize);
+#define LZ4_DECODER_RING_BUFFER_SIZE(maxBlockSize) (65536 + 14 + (maxBlockSize))  /* for static allocation; maxBlockSize presumed valid */
+
+/*! LZ4_decompress_safe_continue() :
+ *  This decoding function allows decompression of consecutive blocks in "streaming" mode.
+ *  The difference with the usual independent blocks is that
+ *  new blocks are allowed to find references into former blocks.
+ *  A block is an unsplittable entity, and must be presented entirely to the decompression function.
+ *  LZ4_decompress_safe_continue() only accepts one block at a time.
+ *  It's modeled after `LZ4_decompress_safe()` and behaves similarly.
+ *
+ * @LZ4_streamDecode : decompression state, tracking the position in memory of past data
+ * @compressedSize : exact complete size of one compressed block.
+ * @dstCapacity : size of destination buffer (which must be already allocated),
+ *                must be an upper bound of decompressed size.
+ * @return : number of bytes decompressed into destination buffer (necessarily <= dstCapacity)
+ *           If destination buffer is not large enough, decoding will stop and output an error code (negative value).
+ *           If the source stream is detected malformed, the function will stop decoding and return a negative result.
+ *
+ *  The last 64KB of previously decoded data *must* remain available and unmodified
+ *  at the memory position where they were previously decoded.
+ *  If less than 64KB of data has been decoded, all the data must be present.
+ *
+ *  Special : if decompression side sets a ring buffer, it must respect one of the following conditions :
+ *  - Decompression buffer size is _at least_ LZ4_decoderRingBufferSize(maxBlockSize).
+ *    maxBlockSize is the maximum size of any single block. It can have any value > 16 bytes.
+ *    In which case, encoding and decoding buffers do not need to be synchronized.
+ *    Actually, data can be produced by any source compliant with LZ4 format specification, and respecting maxBlockSize.
+ *  - Synchronized mode :
+ *    Decompression buffer size is _exactly_ the same as compression buffer size,
+ *    and follows exactly same update rule (block boundaries at same positions),
+ *    and decoding function is provided with exact decompressed size of each block (exception for last block of the stream),
+ *    _then_ decoding & encoding ring buffer can have any size, including small ones ( < 64 KB).
+ *  - Decompression buffer is larger than encoding buffer, by a minimum of maxBlockSize more bytes.
+ *    In which case, encoding and decoding buffers do not need to be synchronized,
+ *    and encoding ring buffer can have any size, including small ones ( < 64 KB).
+ *
+ *  Whenever these conditions are not possible,
+ *  save the last 64KB of decoded data into a safe buffer where it can't be modified during decompression,
+ *  then indicate where this data is saved using LZ4_setStreamDecode(), before decompressing next block.
+*/
+LZ4LIB_API int
+LZ4_decompress_safe_continue (LZ4_streamDecode_t* LZ4_streamDecode,
+                        const char* src, char* dst,
+                        int srcSize, int dstCapacity);
+
+
+/*! LZ4_decompress_safe_usingDict() :
+ *  Works the same as
+ *  a combination of LZ4_setStreamDecode() followed by LZ4_decompress_safe_continue()
+ *  However, it's stateless: it doesn't need any LZ4_streamDecode_t state.
+ *  Dictionary is presumed stable : it must remain accessible and unmodified during decompression.
+ *  Performance tip : Decompression speed can be substantially increased
+ *                    when dst == dictStart + dictSize.
+ */
+LZ4LIB_API int
+LZ4_decompress_safe_usingDict(const char* src, char* dst,
+                              int srcSize, int dstCapacity,
+                              const char* dictStart, int dictSize);
+
+/*! LZ4_decompress_safe_partial_usingDict() :
+ *  Behaves the same as LZ4_decompress_safe_partial()
+ *  with the added ability to specify a memory segment for past data.
+ *  Performance tip : Decompression speed can be substantially increased
+ *                    when dst == dictStart + dictSize.
+ */
+LZ4LIB_API int
+LZ4_decompress_safe_partial_usingDict(const char* src, char* dst,
+                                      int compressedSize,
+                                      int targetOutputSize, int maxOutputSize,
+                                      const char* dictStart, int dictSize);
+
+#endif /* LZ4_H_2983827168210 */
+
+
+/*^*************************************
+ * !!!!!!   STATIC LINKING ONLY   !!!!!!
+ ***************************************/
+
+/*-****************************************************************************
+ * Experimental section
+ *
+ * Symbols declared in this section must be considered unstable. Their
+ * signatures or semantics may change, or they may be removed altogether in the
+ * future. They are therefore only safe to depend on when the caller is
+ * statically linked against the library.
+ *
+ * To protect against unsafe usage, not only are the declarations guarded,
+ * the definitions are hidden by default
+ * when building LZ4 as a shared/dynamic library.
+ *
+ * In order to access these declarations,
+ * define LZ4_STATIC_LINKING_ONLY in your application
+ * before including LZ4's headers.
+ *
+ * In order to make their implementations accessible dynamically, you must
+ * define LZ4_PUBLISH_STATIC_FUNCTIONS when building the LZ4 library.
+ ******************************************************************************/
+
+#ifdef LZ4_STATIC_LINKING_ONLY
+
+#ifndef LZ4_STATIC_3504398509
+#define LZ4_STATIC_3504398509
+
+#ifdef LZ4_PUBLISH_STATIC_FUNCTIONS
+# define LZ4LIB_STATIC_API LZ4LIB_API
+#else
+# define LZ4LIB_STATIC_API
+#endif
+
+
+/*! LZ4_compress_fast_extState_fastReset() :
+ *  A variant of LZ4_compress_fast_extState().
+ *
+ *  Using this variant avoids an expensive initialization step.
+ *  It is only safe to call if the state buffer is known to be correctly initialized already
+ *  (see above comment on LZ4_resetStream_fast() for a definition of "correctly initialized").
+ *  From a high level, the difference is that
+ *  this function initializes the provided state with a call to something like LZ4_resetStream_fast()
+ *  while LZ4_compress_fast_extState() starts with a call to LZ4_resetStream().
+ */
+LZ4LIB_STATIC_API int LZ4_compress_fast_extState_fastReset (void* state, const char* src, char* dst, int srcSize, int dstCapacity, int acceleration);
+
+/*! LZ4_compress_destSize_extState() : introduced in v1.10.0
+ *  Same as LZ4_compress_destSize(), but using an externally allocated state.
+ *  Also: exposes @acceleration
+ */
+int LZ4_compress_destSize_extState(void* state, const char* src, char* dst, int* srcSizePtr, int targetDstSize, int acceleration);
+
+/*! In-place compression and decompression
+ *
+ * It's possible to have input and output sharing the same buffer,
+ * for highly constrained memory environments.
+ * In both cases, it requires input to lay at the end of the buffer,
+ * and decompression to start at beginning of the buffer.
+ * Buffer size must feature some margin, hence be larger than final size.
+ *
+ * |<------------------------buffer--------------------------------->|
+ *                             |<-----------compressed data--------->|
+ * |<-----------decompressed size------------------>|
+ *                                                  |<----margin---->|
+ *
+ * This technique is more useful for decompression,
+ * since decompressed size is typically larger,
+ * and margin is short.
+ *
+ * In-place decompression will work inside any buffer
+ * which size is >= LZ4_DECOMPRESS_INPLACE_BUFFER_SIZE(decompressedSize).
+ * This presumes that decompressedSize > compressedSize.
+ * Otherwise, it means compression actually expanded data,
+ * and it would be more efficient to store such data with a flag indicating it's not compressed.
+ * This can happen when data is not compressible (already compressed, or encrypted).
+ *
+ * For in-place compression, margin is larger, as it must be able to cope with both
+ * history preservation, requiring input data to remain unmodified up to LZ4_DISTANCE_MAX,
+ * and data expansion, which can happen when input is not compressible.
+ * As a consequence, buffer size requirements are much higher,
+ * and memory savings offered by in-place compression are more limited.
+ *
+ * There are ways to limit this cost for compression :
+ * - Reduce history size, by modifying LZ4_DISTANCE_MAX.
+ *   Note that it is a compile-time constant, so all compressions will apply this limit.
+ *   Lower values will reduce compression ratio, except when input_size < LZ4_DISTANCE_MAX,
+ *   so it's a reasonable trick when inputs are known to be small.
+ * - Require the compressor to deliver a "maximum compressed size".
+ *   This is the `dstCapacity` parameter in `LZ4_compress*()`.
+ *   When this size is < LZ4_COMPRESSBOUND(inputSize), then compression can fail,
+ *   in which case, the return code will be 0 (zero).
+ *   The caller must be ready for these cases to happen,
+ *   and typically design a backup scheme to send data uncompressed.
+ * The combination of both techniques can significantly reduce
+ * the amount of margin required for in-place compression.
+ *
+ * In-place compression can work in any buffer
+ * which size is >= (maxCompressedSize)
+ * with maxCompressedSize == LZ4_COMPRESSBOUND(srcSize) for guaranteed compression success.
+ * LZ4_COMPRESS_INPLACE_BUFFER_SIZE() depends on both maxCompressedSize and LZ4_DISTANCE_MAX,
+ * so it's possible to reduce memory requirements by playing with them.
+ */
+
+#define LZ4_DECOMPRESS_INPLACE_MARGIN(compressedSize)          (((compressedSize) >> 8) + 32)
+#define LZ4_DECOMPRESS_INPLACE_BUFFER_SIZE(decompressedSize)   ((decompressedSize) + LZ4_DECOMPRESS_INPLACE_MARGIN(decompressedSize))  /**< note: presumes that compressedSize < decompressedSize. note2: margin is overestimated a bit, since it could use compressedSize instead */
+
+#ifndef LZ4_DISTANCE_MAX   /* history window size; can be user-defined at compile time */
+#  define LZ4_DISTANCE_MAX 65535   /* set to maximum value by default */
+#endif
+
+#define LZ4_COMPRESS_INPLACE_MARGIN                           (LZ4_DISTANCE_MAX + 32)   /* LZ4_DISTANCE_MAX can be safely replaced by srcSize when it's smaller */
+#define LZ4_COMPRESS_INPLACE_BUFFER_SIZE(maxCompressedSize)   ((maxCompressedSize) + LZ4_COMPRESS_INPLACE_MARGIN)  /**< maxCompressedSize is generally LZ4_COMPRESSBOUND(inputSize), but can be set to any lower value, with the risk that compression can fail (return code 0(zero)) */
+
+#endif   /* LZ4_STATIC_3504398509 */
+#endif   /* LZ4_STATIC_LINKING_ONLY */
+
+
+
+#ifndef LZ4_H_98237428734687
+#define LZ4_H_98237428734687
+
+/*-************************************************************
+ *  Private Definitions
+ **************************************************************
+ * Do not use these definitions directly.
+ * They are only exposed to allow static allocation of `LZ4_stream_t` and `LZ4_streamDecode_t`.
+ * Accessing members will expose user code to API and/or ABI break in future versions of the library.
+ **************************************************************/
+#define LZ4_HASHLOG   (LZ4_MEMORY_USAGE-2)
+#define LZ4_HASHTABLESIZE (1 << LZ4_MEMORY_USAGE)
+#define LZ4_HASH_SIZE_U32 (1 << LZ4_HASHLOG)       /* required as macro for static allocation */
+
+#if defined(__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+# include <stdint.h>
+  typedef  int8_t  LZ4_i8;
+  typedef uint8_t  LZ4_byte;
+  typedef uint16_t LZ4_u16;
+  typedef uint32_t LZ4_u32;
+#else
+  typedef   signed char  LZ4_i8;
+  typedef unsigned char  LZ4_byte;
+  typedef unsigned short LZ4_u16;
+  typedef unsigned int   LZ4_u32;
+#endif
+
+/*! LZ4_stream_t :
+ *  Never ever use below internal definitions directly !
+ *  These definitions are not API/ABI safe, and may change in future versions.
+ *  If you need static allocation, declare or allocate an LZ4_stream_t object.
+**/
+
+typedef struct LZ4_stream_t_internal LZ4_stream_t_internal;
+struct LZ4_stream_t_internal {
+    LZ4_u32 hashTable[LZ4_HASH_SIZE_U32];
+    const LZ4_byte* dictionary;
+    const LZ4_stream_t_internal* dictCtx;
+    LZ4_u32 currentOffset;
+    LZ4_u32 tableType;
+    LZ4_u32 dictSize;
+    /* Implicit padding to ensure structure is aligned */
+};
+
+#define LZ4_STREAM_MINSIZE  ((1UL << (LZ4_MEMORY_USAGE)) + 32)  /* static size, for inter-version compatibility */
+union LZ4_stream_u {
+    char minStateSize[LZ4_STREAM_MINSIZE];
+    LZ4_stream_t_internal internal_donotuse;
+}; /* previously typedef'd to LZ4_stream_t */
+
+
+/*! LZ4_initStream() : v1.9.0+
+ *  An LZ4_stream_t structure must be initialized at least once.
+ *  This is automatically done when invoking LZ4_createStream(),
+ *  but it's not when the structure is simply declared on stack (for example).
+ *
+ *  Use LZ4_initStream() to properly initialize a newly declared LZ4_stream_t.
+ *  It can also initialize any arbitrary buffer of sufficient size,
+ *  and will @return a pointer of proper type upon initialization.
+ *
+ *  Note : initialization fails if size and alignment conditions are not respected.
+ *         In which case, the function will @return NULL.
+ *  Note2: An LZ4_stream_t structure guarantees correct alignment and size.
+ *  Note3: Before v1.9.0, use LZ4_resetStream() instead
+**/
+LZ4LIB_API LZ4_stream_t* LZ4_initStream (void* stateBuffer, size_t size);
+
+
+/*! LZ4_streamDecode_t :
+ *  Never ever use below internal definitions directly !
+ *  These definitions are not API/ABI safe, and may change in future versions.
+ *  If you need static allocation, declare or allocate an LZ4_streamDecode_t object.
+**/
+typedef struct {
+    const LZ4_byte* externalDict;
+    const LZ4_byte* prefixEnd;
+    size_t extDictSize;
+    size_t prefixSize;
+} LZ4_streamDecode_t_internal;
+
+#define LZ4_STREAMDECODE_MINSIZE 32
+union LZ4_streamDecode_u {
+    char minStateSize[LZ4_STREAMDECODE_MINSIZE];
+    LZ4_streamDecode_t_internal internal_donotuse;
+} ;   /* previously typedef'd to LZ4_streamDecode_t */
+
+
+
+/*-************************************
+*  Obsolete Functions
+**************************************/
+
+/*! Deprecation warnings
+ *
+ *  Deprecated functions make the compiler generate a warning when invoked.
+ *  This is meant to invite users to update their source code.
+ *  Should deprecation warnings be a problem, it is generally possible to disable them,
+ *  typically with -Wno-deprecated-declarations for gcc
+ *  or _CRT_SECURE_NO_WARNINGS in Visual.
+ *
+ *  Another method is to define LZ4_DISABLE_DEPRECATE_WARNINGS
+ *  before including the header file.
+ */
+#ifdef LZ4_DISABLE_DEPRECATE_WARNINGS
+#  define LZ4_DEPRECATED(message)   /* disable deprecation warnings */
+#else
+#  if defined (__cplusplus) && (__cplusplus >= 201402) /* C++14 or greater */
+#    define LZ4_DEPRECATED(message) [[deprecated(message)]]
+#  elif defined(_MSC_VER)
+#    define LZ4_DEPRECATED(message) __declspec(deprecated(message))
+#  elif defined(__clang__) || (defined(__GNUC__) && (__GNUC__ * 10 + __GNUC_MINOR__ >= 45))
+#    define LZ4_DEPRECATED(message) __attribute__((deprecated(message)))
+#  elif defined(__GNUC__) && (__GNUC__ * 10 + __GNUC_MINOR__ >= 31)
+#    define LZ4_DEPRECATED(message) __attribute__((deprecated))
+#  else
+#    pragma message("WARNING: LZ4_DEPRECATED needs custom implementation for this compiler")
+#    define LZ4_DEPRECATED(message)   /* disabled */
+#  endif
+#endif /* LZ4_DISABLE_DEPRECATE_WARNINGS */
+
+/*! Obsolete compression functions (since v1.7.3) */
+LZ4_DEPRECATED("use LZ4_compress_default() instead")       LZ4LIB_API int LZ4_compress               (const char* src, char* dest, int srcSize);
+LZ4_DEPRECATED("use LZ4_compress_default() instead")       LZ4LIB_API int LZ4_compress_limitedOutput (const char* src, char* dest, int srcSize, int maxOutputSize);
+LZ4_DEPRECATED("use LZ4_compress_fast_extState() instead") LZ4LIB_API int LZ4_compress_withState               (void* state, const char* source, char* dest, int inputSize);
+LZ4_DEPRECATED("use LZ4_compress_fast_extState() instead") LZ4LIB_API int LZ4_compress_limitedOutput_withState (void* state, const char* source, char* dest, int inputSize, int maxOutputSize);
+LZ4_DEPRECATED("use LZ4_compress_fast_continue() instead") LZ4LIB_API int LZ4_compress_continue                (LZ4_stream_t* LZ4_streamPtr, const char* source, char* dest, int inputSize);
+LZ4_DEPRECATED("use LZ4_compress_fast_continue() instead") LZ4LIB_API int LZ4_compress_limitedOutput_continue  (LZ4_stream_t* LZ4_streamPtr, const char* source, char* dest, int inputSize, int maxOutputSize);
+
+/*! Obsolete decompression functions (since v1.8.0) */
+LZ4_DEPRECATED("use LZ4_decompress_fast() instead") LZ4LIB_API int LZ4_uncompress (const char* source, char* dest, int outputSize);
+LZ4_DEPRECATED("use LZ4_decompress_safe() instead") LZ4LIB_API int LZ4_uncompress_unknownOutputSize (const char* source, char* dest, int isize, int maxOutputSize);
+
+/* Obsolete streaming functions (since v1.7.0)
+ * degraded functionality; do not use!
+ *
+ * In order to perform streaming compression, these functions depended on data
+ * that is no longer tracked in the state. They have been preserved as well as
+ * possible: using them will still produce a correct output. However, they don't
+ * actually retain any history between compression calls. The compression ratio
+ * achieved will therefore be no better than compressing each chunk
+ * independently.
+ */
+LZ4_DEPRECATED("Use LZ4_createStream() instead") LZ4LIB_API void* LZ4_create (char* inputBuffer);
+LZ4_DEPRECATED("Use LZ4_createStream() instead") LZ4LIB_API int   LZ4_sizeofStreamState(void);
+LZ4_DEPRECATED("Use LZ4_resetStream() instead")  LZ4LIB_API int   LZ4_resetStreamState(void* state, char* inputBuffer);
+LZ4_DEPRECATED("Use LZ4_saveDict() instead")     LZ4LIB_API char* LZ4_slideInputBuffer (void* state);
+
+/*! Obsolete streaming decoding functions (since v1.7.0) */
+LZ4_DEPRECATED("use LZ4_decompress_safe_usingDict() instead") LZ4LIB_API int LZ4_decompress_safe_withPrefix64k (const char* src, char* dst, int compressedSize, int maxDstSize);
+LZ4_DEPRECATED("use LZ4_decompress_fast_usingDict() instead") LZ4LIB_API int LZ4_decompress_fast_withPrefix64k (const char* src, char* dst, int originalSize);
+
+/*! Obsolete LZ4_decompress_fast variants (since v1.9.0) :
+ *  These functions used to be faster than LZ4_decompress_safe(),
+ *  but this is no longer the case. They are now slower.
+ *  This is because LZ4_decompress_fast() doesn't know the input size,
+ *  and therefore must progress more cautiously into the input buffer to not read beyond the end of block.
+ *  On top of that `LZ4_decompress_fast()` is not protected vs malformed or malicious inputs, making it a security liability.
+ *  As a consequence, LZ4_decompress_fast() is strongly discouraged, and deprecated.
+ *
+ *  The last remaining LZ4_decompress_fast() specificity is that
+ *  it can decompress a block without knowing its compressed size.
+ *  Such functionality can be achieved in a more secure manner
+ *  by employing LZ4_decompress_safe_partial().
+ *
+ *  Parameters:
+ *  originalSize : is the uncompressed size to regenerate.
+ *                 `dst` must be already allocated, its size must be >= 'originalSize' bytes.
+ * @return : number of bytes read from source buffer (== compressed size).
+ *           The function expects to finish at block's end exactly.
+ *           If the source stream is detected malformed, the function stops decoding and returns a negative result.
+ *  note : LZ4_decompress_fast*() requires originalSize. Thanks to this information, it never writes past the output buffer.
+ *         However, since it doesn't know its 'src' size, it may read an unknown amount of input, past input buffer bounds.
+ *         Also, since match offsets are not validated, match reads from 'src' may underflow too.
+ *         These issues never happen if input (compressed) data is correct.
+ *         But they may happen if input data is invalid (error or intentional tampering).
+ *         As a consequence, use these functions in trusted environments with trusted data **only**.
+ */
+LZ4_DEPRECATED("This function is deprecated and unsafe. Consider using LZ4_decompress_safe_partial() instead")
+LZ4LIB_API int LZ4_decompress_fast (const char* src, char* dst, int originalSize);
+LZ4_DEPRECATED("This function is deprecated and unsafe. Consider migrating towards LZ4_decompress_safe_continue() instead. "
+               "Note that the contract will change (requires block's compressed size, instead of decompressed size)")
+LZ4LIB_API int LZ4_decompress_fast_continue (LZ4_streamDecode_t* LZ4_streamDecode, const char* src, char* dst, int originalSize);
+LZ4_DEPRECATED("This function is deprecated and unsafe. Consider using LZ4_decompress_safe_partial_usingDict() instead")
+LZ4LIB_API int LZ4_decompress_fast_usingDict (const char* src, char* dst, int originalSize, const char* dictStart, int dictSize);
+
+/*! LZ4_resetStream() :
+ *  An LZ4_stream_t structure must be initialized at least once.
+ *  This is done with LZ4_initStream(), or LZ4_resetStream().
+ *  Consider switching to LZ4_initStream(),
+ *  invoking LZ4_resetStream() will trigger deprecation warnings in the future.
+ */
+LZ4LIB_API void LZ4_resetStream (LZ4_stream_t* streamPtr);
+
+
+#endif /* LZ4_H_98237428734687 */
+
+
+#if defined (__cplusplus)
+}
+#endif
diff --git a/lib/lz4/upstream/lz4hc.c b/lib/lz4/upstream/lz4hc.c
new file mode 100644
index 0000000000000000000000000000000000000000..4d8c36a6978fcae09e4c4a936572b4b271513466
--- /dev/null
+++ b/lib/lz4/upstream/lz4hc.c
@@ -0,0 +1,2192 @@
+/*
+    LZ4 HC - High Compression Mode of LZ4
+    Copyright (C) 2011-2020, Yann Collet.
+
+    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+       - LZ4 source repository : https://github.com/lz4/lz4
+       - LZ4 public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+/* note : lz4hc is not an independent module, it requires lz4.h/lz4.c for proper compilation */
+
+
+/* *************************************
+*  Tuning Parameter
+***************************************/
+
+/*! HEAPMODE :
+ *  Select how stateless HC compression functions like `LZ4_compress_HC()`
+ *  allocate memory for their workspace:
+ *  in stack (0:fastest), or in heap (1:default, requires malloc()).
+ *  Since workspace is rather large, heap mode is recommended.
+**/
+#ifndef LZ4HC_HEAPMODE
+#  define LZ4HC_HEAPMODE 1
+#endif
+
+
+/*===    Dependency    ===*/
+#define LZ4_HC_STATIC_LINKING_ONLY
+#include "lz4hc.h"
+#include <limits.h>
+
+
+/*===   Shared lz4.c code   ===*/
+#ifndef LZ4_SRC_INCLUDED
+# if defined(__GNUC__)
+#  pragma GCC diagnostic ignored "-Wunused-function"
+# endif
+# if defined (__clang__)
+#  pragma clang diagnostic ignored "-Wunused-function"
+# endif
+# define LZ4_COMMONDEFS_ONLY
+# include "lz4.c"   /* LZ4_count, constants, mem */
+#endif
+
+
+/*===   Enums   ===*/
+typedef enum { noDictCtx, usingDictCtxHc } dictCtx_directive;
+
+
+/*===   Constants   ===*/
+#define OPTIMAL_ML (int)((ML_MASK-1)+MINMATCH)
+#define LZ4_OPT_NUM   (1<<12)
+
+
+/*===   Macros   ===*/
+#define MIN(a,b)   ( (a) < (b) ? (a) : (b) )
+#define MAX(a,b)   ( (a) > (b) ? (a) : (b) )
+
+
+/*===   Levels definition   ===*/
+typedef enum { lz4mid, lz4hc, lz4opt } lz4hc_strat_e;
+typedef struct {
+    lz4hc_strat_e strat;
+    int nbSearches;
+    U32 targetLength;
+} cParams_t;
+static const cParams_t k_clTable[LZ4HC_CLEVEL_MAX+1] = {
+    { lz4mid,    2, 16 },  /* 0, unused */
+    { lz4mid,    2, 16 },  /* 1, unused */
+    { lz4mid,    2, 16 },  /* 2 */
+    { lz4hc,     4, 16 },  /* 3 */
+    { lz4hc,     8, 16 },  /* 4 */
+    { lz4hc,    16, 16 },  /* 5 */
+    { lz4hc,    32, 16 },  /* 6 */
+    { lz4hc,    64, 16 },  /* 7 */
+    { lz4hc,   128, 16 },  /* 8 */
+    { lz4hc,   256, 16 },  /* 9 */
+    { lz4opt,   96, 64 },  /*10==LZ4HC_CLEVEL_OPT_MIN*/
+    { lz4opt,  512,128 },  /*11 */
+    { lz4opt,16384,LZ4_OPT_NUM },  /* 12==LZ4HC_CLEVEL_MAX */
+};
+
+static cParams_t LZ4HC_getCLevelParams(int cLevel)
+{
+    /* note : clevel convention is a bit different from lz4frame,
+     * possibly something worth revisiting for consistency */
+    if (cLevel < 1)
+        cLevel = LZ4HC_CLEVEL_DEFAULT;
+    cLevel = MIN(LZ4HC_CLEVEL_MAX, cLevel);
+    return k_clTable[cLevel];
+}
+
+
+/*===   Hashing   ===*/
+#define LZ4HC_HASHSIZE 4
+#define HASH_FUNCTION(i)      (((i) * 2654435761U) >> ((MINMATCH*8)-LZ4HC_HASH_LOG))
+static U32 LZ4HC_hashPtr(const void* ptr) { return HASH_FUNCTION(LZ4_read32(ptr)); }
+
+#if defined(LZ4_FORCE_MEMORY_ACCESS) && (LZ4_FORCE_MEMORY_ACCESS==2)
+/* lie to the compiler about data alignment; use with caution */
+static U64 LZ4_read64(const void* memPtr) { return *(const U64*) memPtr; }
+
+#elif defined(LZ4_FORCE_MEMORY_ACCESS) && (LZ4_FORCE_MEMORY_ACCESS==1)
+/* __pack instructions are safer, but compiler specific */
+LZ4_PACK(typedef struct { U64 u64; }) LZ4_unalign64;
+static U64 LZ4_read64(const void* ptr) { return ((const LZ4_unalign64*)ptr)->u64; }
+
+#else  /* safe and portable access using memcpy() */
+static U64 LZ4_read64(const void* memPtr)
+{
+    U64 val; LZ4_memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+#endif /* LZ4_FORCE_MEMORY_ACCESS */
+
+#define LZ4MID_HASHSIZE 8
+#define LZ4MID_HASHLOG (LZ4HC_HASH_LOG-1)
+#define LZ4MID_HASHTABLESIZE (1 << LZ4MID_HASHLOG)
+
+static U32 LZ4MID_hash4(U32 v) { return (v * 2654435761U) >> (32-LZ4MID_HASHLOG); }
+static U32 LZ4MID_hash4Ptr(const void* ptr) { return LZ4MID_hash4(LZ4_read32(ptr)); }
+/* note: hash7 hashes the lower 56-bits.
+ * It presumes input was read using little endian.*/
+static U32 LZ4MID_hash7(U64 v) { return (U32)(((v  << (64-56)) * 58295818150454627ULL) >> (64-LZ4MID_HASHLOG)) ; }
+static U64 LZ4_readLE64(const void* memPtr);
+static U32 LZ4MID_hash8Ptr(const void* ptr) { return LZ4MID_hash7(LZ4_readLE64(ptr)); }
+
+static U64 LZ4_readLE64(const void* memPtr)
+{
+    if (LZ4_isLittleEndian()) {
+        return LZ4_read64(memPtr);
+    } else {
+        const BYTE* p = (const BYTE*)memPtr;
+        /* note: relies on the compiler to simplify this expression */
+        return (U64)p[0] | ((U64)p[1]<<8) | ((U64)p[2]<<16) | ((U64)p[3]<<24)
+            | ((U64)p[4]<<32) | ((U64)p[5]<<40) | ((U64)p[6]<<48) | ((U64)p[7]<<56);
+    }
+}
+
+
+/*===   Count match length   ===*/
+LZ4_FORCE_INLINE
+unsigned LZ4HC_NbCommonBytes32(U32 val)
+{
+    assert(val != 0);
+    if (LZ4_isLittleEndian()) {
+#     if defined(_MSC_VER) && (_MSC_VER >= 1400) && !defined(LZ4_FORCE_SW_BITCOUNT)
+        unsigned long r;
+        _BitScanReverse(&r, val);
+        return (unsigned)((31 - r) >> 3);
+#     elif (defined(__clang__) || (defined(__GNUC__) && ((__GNUC__ > 3) || \
+                            ((__GNUC__ == 3) && (__GNUC_MINOR__ >= 4))))) && \
+                                        !defined(LZ4_FORCE_SW_BITCOUNT)
+        return (unsigned)__builtin_clz(val) >> 3;
+#     else
+        val >>= 8;
+        val = ((((val + 0x00FFFF00) | 0x00FFFFFF) + val) |
+              (val + 0x00FF0000)) >> 24;
+        return (unsigned)val ^ 3;
+#     endif
+    } else {
+#     if defined(_MSC_VER) && (_MSC_VER >= 1400) && !defined(LZ4_FORCE_SW_BITCOUNT)
+        unsigned long r;
+        _BitScanForward(&r, val);
+        return (unsigned)(r >> 3);
+#     elif (defined(__clang__) || (defined(__GNUC__) && ((__GNUC__ > 3) || \
+                            ((__GNUC__ == 3) && (__GNUC_MINOR__ >= 4))))) && \
+                                        !defined(LZ4_FORCE_SW_BITCOUNT)
+        return (unsigned)__builtin_ctz(val) >> 3;
+#     else
+        const U32 m = 0x01010101;
+        return (unsigned)((((val - 1) ^ val) & (m - 1)) * m) >> 24;
+#     endif
+    }
+}
+
+/** LZ4HC_countBack() :
+ * @return : negative value, nb of common bytes before ip/match */
+LZ4_FORCE_INLINE
+int LZ4HC_countBack(const BYTE* const ip, const BYTE* const match,
+                    const BYTE* const iMin, const BYTE* const mMin)
+{
+    int back = 0;
+    int const min = (int)MAX(iMin - ip, mMin - match);
+    assert(min <= 0);
+    assert(ip >= iMin); assert((size_t)(ip-iMin) < (1U<<31));
+    assert(match >= mMin); assert((size_t)(match - mMin) < (1U<<31));
+
+    while ((back - min) > 3) {
+        U32 const v = LZ4_read32(ip + back - 4) ^ LZ4_read32(match + back - 4);
+        if (v) {
+            return (back - (int)LZ4HC_NbCommonBytes32(v));
+        } else back -= 4; /* 4-byte step */
+    }
+    /* check remainder if any */
+    while ( (back > min)
+         && (ip[back-1] == match[back-1]) )
+            back--;
+    return back;
+}
+
+/*===   Chain table updates   ===*/
+#define DELTANEXTU16(table, pos) table[(U16)(pos)]   /* faster */
+/* Make fields passed to, and updated by LZ4HC_encodeSequence explicit */
+#define UPDATABLE(ip, op, anchor) &ip, &op, &anchor
+
+
+/**************************************
+*  Init
+**************************************/
+static void LZ4HC_clearTables (LZ4HC_CCtx_internal* hc4)
+{
+    MEM_INIT(hc4->hashTable, 0, sizeof(hc4->hashTable));
+    MEM_INIT(hc4->chainTable, 0xFF, sizeof(hc4->chainTable));
+}
+
+static void LZ4HC_init_internal (LZ4HC_CCtx_internal* hc4, const BYTE* start)
+{
+    size_t const bufferSize = (size_t)(hc4->end - hc4->prefixStart);
+    size_t newStartingOffset = bufferSize + hc4->dictLimit;
+    DEBUGLOG(5, "LZ4HC_init_internal");
+    assert(newStartingOffset >= bufferSize);  /* check overflow */
+    if (newStartingOffset > 1 GB) {
+        LZ4HC_clearTables(hc4);
+        newStartingOffset = 0;
+    }
+    newStartingOffset += 64 KB;
+    hc4->nextToUpdate = (U32)newStartingOffset;
+    hc4->prefixStart = start;
+    hc4->end = start;
+    hc4->dictStart = start;
+    hc4->dictLimit = (U32)newStartingOffset;
+    hc4->lowLimit = (U32)newStartingOffset;
+}
+
+
+/**************************************
+*  Encode
+**************************************/
+/* LZ4HC_encodeSequence() :
+ * @return : 0 if ok,
+ *           1 if buffer issue detected */
+LZ4_FORCE_INLINE int LZ4HC_encodeSequence (
+    const BYTE** _ip,
+    BYTE** _op,
+    const BYTE** _anchor,
+    int matchLength,
+    int offset,
+    limitedOutput_directive limit,
+    BYTE* oend)
+{
+#define ip      (*_ip)
+#define op      (*_op)
+#define anchor  (*_anchor)
+
+    size_t length;
+    BYTE* const token = op++;
+
+#if defined(LZ4_DEBUG) && (LZ4_DEBUG >= 6)
+    static const BYTE* start = NULL;
+    static U32 totalCost = 0;
+    U32 const pos = (start==NULL) ? 0 : (U32)(anchor - start);
+    U32 const ll = (U32)(ip - anchor);
+    U32 const llAdd = (ll>=15) ? ((ll-15) / 255) + 1 : 0;
+    U32 const mlAdd = (matchLength>=19) ? ((matchLength-19) / 255) + 1 : 0;
+    U32 const cost = 1 + llAdd + ll + 2 + mlAdd;
+    if (start==NULL) start = anchor;  /* only works for single segment */
+    /* g_debuglog_enable = (pos >= 2228) & (pos <= 2262); */
+    DEBUGLOG(6, "pos:%7u -- literals:%4u, match:%4i, offset:%5i, cost:%4u + %5u",
+                pos,
+                (U32)(ip - anchor), matchLength, offset,
+                cost, totalCost);
+    totalCost += cost;
+#endif
+
+    /* Encode Literal length */
+    length = (size_t)(ip - anchor);
+    LZ4_STATIC_ASSERT(notLimited == 0);
+    /* Check output limit */
+    if (limit && ((op + (length / 255) + length + (2 + 1 + LASTLITERALS)) > oend)) {
+        DEBUGLOG(6, "Not enough room to write %i literals (%i bytes remaining)",
+                (int)length, (int)(oend - op));
+        return 1;
+    }
+    if (length >= RUN_MASK) {
+        size_t len = length - RUN_MASK;
+        *token = (RUN_MASK << ML_BITS);
+        for(; len >= 255 ; len -= 255) *op++ = 255;
+        *op++ = (BYTE)len;
+    } else {
+        *token = (BYTE)(length << ML_BITS);
+    }
+
+    /* Copy Literals */
+    LZ4_wildCopy8(op, anchor, op + length);
+    op += length;
+
+    /* Encode Offset */
+    assert(offset <= LZ4_DISTANCE_MAX );
+    assert(offset > 0);
+    LZ4_writeLE16(op, (U16)(offset)); op += 2;
+
+    /* Encode MatchLength */
+    assert(matchLength >= MINMATCH);
+    length = (size_t)matchLength - MINMATCH;
+    if (limit && (op + (length / 255) + (1 + LASTLITERALS) > oend)) {
+        DEBUGLOG(6, "Not enough room to write match length");
+        return 1;   /* Check output limit */
+    }
+    if (length >= ML_MASK) {
+        *token += ML_MASK;
+        length -= ML_MASK;
+        for(; length >= 510 ; length -= 510) { *op++ = 255; *op++ = 255; }
+        if (length >= 255) { length -= 255; *op++ = 255; }
+        *op++ = (BYTE)length;
+    } else {
+        *token += (BYTE)(length);
+    }
+
+    /* Prepare next loop */
+    ip += matchLength;
+    anchor = ip;
+
+    return 0;
+
+#undef ip
+#undef op
+#undef anchor
+}
+
+
+typedef struct {
+    int off;
+    int len;
+    int back;  /* negative value */
+} LZ4HC_match_t;
+
+LZ4HC_match_t LZ4HC_searchExtDict(const BYTE* ip, U32 ipIndex,
+        const BYTE* const iLowLimit, const BYTE* const iHighLimit,
+        const LZ4HC_CCtx_internal* dictCtx, U32 gDictEndIndex,
+        int currentBestML, int nbAttempts)
+{
+    size_t const lDictEndIndex = (size_t)(dictCtx->end - dictCtx->prefixStart) + dictCtx->dictLimit;
+    U32 lDictMatchIndex = dictCtx->hashTable[LZ4HC_hashPtr(ip)];
+    U32 matchIndex = lDictMatchIndex + gDictEndIndex - (U32)lDictEndIndex;
+    int offset = 0, sBack = 0;
+    assert(lDictEndIndex <= 1 GB);
+    if (lDictMatchIndex>0)
+        DEBUGLOG(7, "lDictEndIndex = %zu, lDictMatchIndex = %u", lDictEndIndex, lDictMatchIndex);
+    while (ipIndex - matchIndex <= LZ4_DISTANCE_MAX && nbAttempts--) {
+        const BYTE* const matchPtr = dictCtx->prefixStart - dictCtx->dictLimit + lDictMatchIndex;
+
+        if (LZ4_read32(matchPtr) == LZ4_read32(ip)) {
+            int mlt;
+            int back = 0;
+            const BYTE* vLimit = ip + (lDictEndIndex - lDictMatchIndex);
+            if (vLimit > iHighLimit) vLimit = iHighLimit;
+            mlt = (int)LZ4_count(ip+MINMATCH, matchPtr+MINMATCH, vLimit) + MINMATCH;
+            back = (ip > iLowLimit) ? LZ4HC_countBack(ip, matchPtr, iLowLimit, dictCtx->prefixStart) : 0;
+            mlt -= back;
+            if (mlt > currentBestML) {
+                currentBestML = mlt;
+                offset = (int)(ipIndex - matchIndex);
+                sBack = back;
+                DEBUGLOG(7, "found match of length %i within extDictCtx", currentBestML);
+        }   }
+
+        {   U32 const nextOffset = DELTANEXTU16(dictCtx->chainTable, lDictMatchIndex);
+            lDictMatchIndex -= nextOffset;
+            matchIndex -= nextOffset;
+    }   }
+
+    {   LZ4HC_match_t md;
+        md.len = currentBestML;
+        md.off = offset;
+        md.back = sBack;
+        return md;
+    }
+}
+
+typedef LZ4HC_match_t (*LZ4MID_searchIntoDict_f)(const BYTE* ip, U32 ipIndex,
+        const BYTE* const iHighLimit,
+        const LZ4HC_CCtx_internal* dictCtx, U32 gDictEndIndex);
+
+static LZ4HC_match_t LZ4MID_searchHCDict(const BYTE* ip, U32 ipIndex,
+        const BYTE* const iHighLimit,
+        const LZ4HC_CCtx_internal* dictCtx, U32 gDictEndIndex)
+{
+    return LZ4HC_searchExtDict(ip,ipIndex,
+                            ip, iHighLimit,
+                            dictCtx, gDictEndIndex,
+                            MINMATCH-1, 2);
+}
+
+static LZ4HC_match_t LZ4MID_searchExtDict(const BYTE* ip, U32 ipIndex,
+        const BYTE* const iHighLimit,
+        const LZ4HC_CCtx_internal* dictCtx, U32 gDictEndIndex)
+{
+    size_t const lDictEndIndex = (size_t)(dictCtx->end - dictCtx->prefixStart) + dictCtx->dictLimit;
+    const U32* const hash4Table = dictCtx->hashTable;
+    const U32* const hash8Table = hash4Table + LZ4MID_HASHTABLESIZE;
+    DEBUGLOG(7, "LZ4MID_searchExtDict (ipIdx=%u)", ipIndex);
+
+    /* search long match first */
+    {   U32 l8DictMatchIndex = hash8Table[LZ4MID_hash8Ptr(ip)];
+        U32 m8Index = l8DictMatchIndex + gDictEndIndex - (U32)lDictEndIndex;
+        assert(lDictEndIndex <= 1 GB);
+        if (ipIndex - m8Index <= LZ4_DISTANCE_MAX) {
+            const BYTE* const matchPtr = dictCtx->prefixStart - dictCtx->dictLimit + l8DictMatchIndex;
+            const size_t safeLen = MIN(lDictEndIndex - l8DictMatchIndex, (size_t)(iHighLimit - ip));
+            int mlt = (int)LZ4_count(ip, matchPtr, ip + safeLen);
+            if (mlt >= MINMATCH) {
+                LZ4HC_match_t md;
+                DEBUGLOG(7, "Found long ExtDict match of len=%u", mlt);
+                md.len = mlt;
+                md.off = (int)(ipIndex - m8Index);
+                md.back = 0;
+                return md;
+            }
+        }
+    }
+
+    /* search for short match second */
+    {   U32 l4DictMatchIndex = hash4Table[LZ4MID_hash4Ptr(ip)];
+        U32 m4Index = l4DictMatchIndex + gDictEndIndex - (U32)lDictEndIndex;
+        if (ipIndex - m4Index <= LZ4_DISTANCE_MAX) {
+            const BYTE* const matchPtr = dictCtx->prefixStart - dictCtx->dictLimit + l4DictMatchIndex;
+            const size_t safeLen = MIN(lDictEndIndex - l4DictMatchIndex, (size_t)(iHighLimit - ip));
+            int mlt = (int)LZ4_count(ip, matchPtr, ip + safeLen);
+            if (mlt >= MINMATCH) {
+                LZ4HC_match_t md;
+                DEBUGLOG(7, "Found short ExtDict match of len=%u", mlt);
+                md.len = mlt;
+                md.off = (int)(ipIndex - m4Index);
+                md.back = 0;
+                return md;
+            }
+        }
+    }
+
+    /* nothing found */
+    {   LZ4HC_match_t const md = {0, 0, 0 };
+        return md;
+    }
+}
+
+/**************************************
+*  Mid Compression (level 2)
+**************************************/
+
+LZ4_FORCE_INLINE void
+LZ4MID_addPosition(U32* hTable, U32 hValue, U32 index)
+{
+    hTable[hValue] = index;
+}
+
+#define ADDPOS8(_p, _idx) LZ4MID_addPosition(hash8Table, LZ4MID_hash8Ptr(_p), _idx)
+#define ADDPOS4(_p, _idx) LZ4MID_addPosition(hash4Table, LZ4MID_hash4Ptr(_p), _idx)
+
+/* Fill hash tables with references into dictionary.
+ * The resulting table is only exploitable by LZ4MID (level 2) */
+static void
+LZ4MID_fillHTable (LZ4HC_CCtx_internal* cctx, const void* dict, size_t size)
+{
+    U32* const hash4Table = cctx->hashTable;
+    U32* const hash8Table = hash4Table + LZ4MID_HASHTABLESIZE;
+    const BYTE* const prefixPtr = (const BYTE*)dict;
+    U32 const prefixIdx = cctx->dictLimit;
+    U32 const target = prefixIdx + (U32)size - LZ4MID_HASHSIZE;
+    U32 idx = cctx->nextToUpdate;
+    assert(dict == cctx->prefixStart);
+    DEBUGLOG(4, "LZ4MID_fillHTable (size:%zu)", size);
+    if (size <= LZ4MID_HASHSIZE)
+        return;
+
+    for (; idx < target; idx += 3) {
+        ADDPOS4(prefixPtr+idx-prefixIdx, idx);
+        ADDPOS8(prefixPtr+idx+1-prefixIdx, idx+1);
+    }
+
+    idx = (size > 32 KB + LZ4MID_HASHSIZE) ? target - 32 KB : cctx->nextToUpdate;
+    for (; idx < target; idx += 1) {
+        ADDPOS8(prefixPtr+idx-prefixIdx, idx);
+    }
+
+    cctx->nextToUpdate = target;
+}
+
+static LZ4MID_searchIntoDict_f select_searchDict_function(const LZ4HC_CCtx_internal* dictCtx)
+{
+    if (dictCtx == NULL) return NULL;
+    if (LZ4HC_getCLevelParams(dictCtx->compressionLevel).strat == lz4mid)
+        return LZ4MID_searchExtDict;
+    return LZ4MID_searchHCDict;
+}
+
+static int LZ4MID_compress (
+    LZ4HC_CCtx_internal* const ctx,
+    const char* const src,
+    char* const dst,
+    int* srcSizePtr,
+    int const maxOutputSize,
+    const limitedOutput_directive limit,
+    const dictCtx_directive dict
+    )
+{
+    U32* const hash4Table = ctx->hashTable;
+    U32* const hash8Table = hash4Table + LZ4MID_HASHTABLESIZE;
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* anchor = ip;
+    const BYTE* const iend = ip + *srcSizePtr;
+    const BYTE* const mflimit = iend - MFLIMIT;
+    const BYTE* const matchlimit = (iend - LASTLITERALS);
+    const BYTE* const ilimit = (iend - LZ4MID_HASHSIZE);
+    BYTE* op = (BYTE*)dst;
+    BYTE* oend = op + maxOutputSize;
+
+    const BYTE* const prefixPtr = ctx->prefixStart;
+    const U32 prefixIdx = ctx->dictLimit;
+    const U32 ilimitIdx = (U32)(ilimit - prefixPtr) + prefixIdx;
+    const BYTE* const dictStart = ctx->dictStart;
+    const U32 dictIdx = ctx->lowLimit;
+    const U32 gDictEndIndex = ctx->lowLimit;
+    const LZ4MID_searchIntoDict_f searchIntoDict = (dict == usingDictCtxHc) ? select_searchDict_function(ctx->dictCtx) : NULL;
+    unsigned matchLength;
+    unsigned matchDistance;
+
+    /* input sanitization */
+    DEBUGLOG(5, "LZ4MID_compress (%i bytes)", *srcSizePtr);
+    if (dict == usingDictCtxHc) DEBUGLOG(5, "usingDictCtxHc");
+    assert(*srcSizePtr >= 0);
+    if (*srcSizePtr) assert(src != NULL);
+    if (maxOutputSize) assert(dst != NULL);
+    if (*srcSizePtr < 0) return 0;  /* invalid */
+    if (maxOutputSize < 0) return 0; /* invalid */
+    if (*srcSizePtr > LZ4_MAX_INPUT_SIZE) {
+        /* forbidden: no input is allowed to be that large */
+        return 0;
+    }
+    if (limit == fillOutput) oend -= LASTLITERALS;  /* Hack for support LZ4 format restriction */
+    if (*srcSizePtr < LZ4_minLength)
+        goto _lz4mid_last_literals;  /* Input too small, no compression (all literals) */
+
+    /* main loop */
+    while (ip <= mflimit) {
+        const U32 ipIndex = (U32)(ip - prefixPtr) + prefixIdx;
+        /* search long match */
+        {   U32 const h8 = LZ4MID_hash8Ptr(ip);
+            U32 const pos8 = hash8Table[h8];
+            assert(h8 < LZ4MID_HASHTABLESIZE);
+            assert(pos8 < ipIndex);
+            LZ4MID_addPosition(hash8Table, h8, ipIndex);
+            if (ipIndex - pos8 <= LZ4_DISTANCE_MAX) {
+                /* match candidate found */
+                if (pos8 >= prefixIdx) {
+                    const BYTE* const matchPtr = prefixPtr + pos8 - prefixIdx;
+                    assert(matchPtr < ip);
+                    matchLength = LZ4_count(ip, matchPtr, matchlimit);
+                    if (matchLength >= MINMATCH) {
+                        DEBUGLOG(7, "found long match at pos %u (len=%u)", pos8, matchLength);
+                        matchDistance = ipIndex - pos8;
+                        goto _lz4mid_encode_sequence;
+                    }
+                } else {
+                    if (pos8 >= dictIdx) {
+                        /* extDict match candidate */
+                        const BYTE* const matchPtr = dictStart + (pos8 - dictIdx);
+                        const size_t safeLen = MIN(prefixIdx - pos8, (size_t)(matchlimit - ip));
+                        matchLength = LZ4_count(ip, matchPtr, ip + safeLen);
+                        if (matchLength >= MINMATCH) {
+                            DEBUGLOG(7, "found long match at ExtDict pos %u (len=%u)", pos8, matchLength);
+                            matchDistance = ipIndex - pos8;
+                            goto _lz4mid_encode_sequence;
+                        }
+                    }
+                }
+        }   }
+        /* search short match */
+        {   U32 const h4 = LZ4MID_hash4Ptr(ip);
+            U32 const pos4 = hash4Table[h4];
+            assert(h4 < LZ4MID_HASHTABLESIZE);
+            assert(pos4 < ipIndex);
+            LZ4MID_addPosition(hash4Table, h4, ipIndex);
+            if (ipIndex - pos4 <= LZ4_DISTANCE_MAX) {
+                /* match candidate found */
+                if (pos4 >= prefixIdx) {
+                /* only search within prefix */
+                    const BYTE* const matchPtr = prefixPtr + (pos4 - prefixIdx);
+                    assert(matchPtr < ip);
+                    assert(matchPtr >= prefixPtr);
+                    matchLength = LZ4_count(ip, matchPtr, matchlimit);
+                    if (matchLength >= MINMATCH) {
+                        /* short match found, let's just check ip+1 for longer */
+                        U32 const h8 = LZ4MID_hash8Ptr(ip+1);
+                        U32 const pos8 = hash8Table[h8];
+                        U32 const m2Distance = ipIndex + 1 - pos8;
+                        matchDistance = ipIndex - pos4;
+                        if ( m2Distance <= LZ4_DISTANCE_MAX
+                        && pos8 >= prefixIdx /* only search within prefix */
+                        && likely(ip < mflimit)
+                        ) {
+                            const BYTE* const m2Ptr = prefixPtr + (pos8 - prefixIdx);
+                            unsigned ml2 = LZ4_count(ip+1, m2Ptr, matchlimit);
+                            if (ml2 > matchLength) {
+                                LZ4MID_addPosition(hash8Table, h8, ipIndex+1);
+                                ip++;
+                                matchLength = ml2;
+                                matchDistance = m2Distance;
+                        }   }
+                        goto _lz4mid_encode_sequence;
+                    }
+                } else {
+                    if (pos4 >= dictIdx) {
+                        /* extDict match candidate */
+                        const BYTE* const matchPtr = dictStart + (pos4 - dictIdx);
+                        const size_t safeLen = MIN(prefixIdx - pos4, (size_t)(matchlimit - ip));
+                        matchLength = LZ4_count(ip, matchPtr, ip + safeLen);
+                        if (matchLength >= MINMATCH) {
+                            DEBUGLOG(7, "found match at ExtDict pos %u (len=%u)", pos4, matchLength);
+                            matchDistance = ipIndex - pos4;
+                            goto _lz4mid_encode_sequence;
+                        }
+                    }
+                }
+        }   }
+        /* no match found in prefix */
+        if ( (dict == usingDictCtxHc)
+          && (ipIndex - gDictEndIndex < LZ4_DISTANCE_MAX - 8) ) {
+            /* search a match into external dictionary */
+            LZ4HC_match_t dMatch = searchIntoDict(ip, ipIndex,
+                    matchlimit,
+                    ctx->dictCtx, gDictEndIndex);
+            if (dMatch.len >= MINMATCH) {
+                DEBUGLOG(7, "found Dictionary match (offset=%i)", dMatch.off);
+                assert(dMatch.back == 0);
+                matchLength = (unsigned)dMatch.len;
+                matchDistance = (unsigned)dMatch.off;
+                goto _lz4mid_encode_sequence;
+            }
+        }
+        /* no match found */
+        ip += 1 + ((ip-anchor) >> 9);  /* skip faster over incompressible data */
+        continue;
+
+_lz4mid_encode_sequence:
+        /* catch back */
+        while (((ip > anchor) & ((U32)(ip-prefixPtr) > matchDistance)) && (unlikely(ip[-1] == ip[-(int)matchDistance-1]))) {
+            ip--;  matchLength++;
+        };
+
+        /* fill table with beginning of match */
+        ADDPOS8(ip+1, ipIndex+1);
+        ADDPOS8(ip+2, ipIndex+2);
+        ADDPOS4(ip+1, ipIndex+1);
+
+        /* encode */
+        {   BYTE* const saved_op = op;
+            /* LZ4HC_encodeSequence always updates @op; on success, it updates @ip and @anchor */
+            if (LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor),
+                    (int)matchLength, (int)matchDistance,
+                    limit, oend) ) {
+                op = saved_op;  /* restore @op value before failed LZ4HC_encodeSequence */
+                goto _lz4mid_dest_overflow;
+            }
+        }
+
+        /* fill table with end of match */
+        {   U32 endMatchIdx = (U32)(ip-prefixPtr) + prefixIdx;
+            U32 pos_m2 = endMatchIdx - 2;
+            if (pos_m2 < ilimitIdx) {
+                if (likely(ip - prefixPtr > 5)) {
+                    ADDPOS8(ip-5, endMatchIdx - 5);
+                }
+                ADDPOS8(ip-3, endMatchIdx - 3);
+                ADDPOS8(ip-2, endMatchIdx - 2);
+                ADDPOS4(ip-2, endMatchIdx - 2);
+                ADDPOS4(ip-1, endMatchIdx - 1);
+            }
+        }
+    }
+
+_lz4mid_last_literals:
+    /* Encode Last Literals */
+    {   size_t lastRunSize = (size_t)(iend - anchor);  /* literals */
+        size_t llAdd = (lastRunSize + 255 - RUN_MASK) / 255;
+        size_t const totalSize = 1 + llAdd + lastRunSize;
+        if (limit == fillOutput) oend += LASTLITERALS;  /* restore correct value */
+        if (limit && (op + totalSize > oend)) {
+            if (limit == limitedOutput) return 0;  /* not enough space in @dst */
+            /* adapt lastRunSize to fill 'dest' */
+            lastRunSize  = (size_t)(oend - op) - 1 /*token*/;
+            llAdd = (lastRunSize + 256 - RUN_MASK) / 256;
+            lastRunSize -= llAdd;
+        }
+        DEBUGLOG(6, "Final literal run : %i literals", (int)lastRunSize);
+        ip = anchor + lastRunSize;  /* can be != iend if limit==fillOutput */
+
+        if (lastRunSize >= RUN_MASK) {
+            size_t accumulator = lastRunSize - RUN_MASK;
+            *op++ = (RUN_MASK << ML_BITS);
+            for(; accumulator >= 255 ; accumulator -= 255)
+                *op++ = 255;
+            *op++ = (BYTE) accumulator;
+        } else {
+            *op++ = (BYTE)(lastRunSize << ML_BITS);
+        }
+        assert(lastRunSize <= (size_t)(oend - op));
+        LZ4_memcpy(op, anchor, lastRunSize);
+        op += lastRunSize;
+    }
+
+    /* End */
+    DEBUGLOG(5, "compressed %i bytes into %i bytes", *srcSizePtr, (int)((char*)op - dst));
+    assert(ip >= (const BYTE*)src);
+    assert(ip <= iend);
+    *srcSizePtr = (int)(ip - (const BYTE*)src);
+    assert((char*)op >= dst);
+    assert(op <= oend);
+    assert((char*)op - dst < INT_MAX);
+    return (int)((char*)op - dst);
+
+_lz4mid_dest_overflow:
+    if (limit == fillOutput) {
+        /* Assumption : @ip, @anchor, @optr and @matchLength must be set correctly */
+        size_t const ll = (size_t)(ip - anchor);
+        size_t const ll_addbytes = (ll + 240) / 255;
+        size_t const ll_totalCost = 1 + ll_addbytes + ll;
+        BYTE* const maxLitPos = oend - 3; /* 2 for offset, 1 for token */
+        DEBUGLOG(6, "Last sequence is overflowing : %u literals, %u remaining space",
+                (unsigned)ll, (unsigned)(oend-op));
+        if (op + ll_totalCost <= maxLitPos) {
+            /* ll validated; now adjust match length */
+            size_t const bytesLeftForMl = (size_t)(maxLitPos - (op+ll_totalCost));
+            size_t const maxMlSize = MINMATCH + (ML_MASK-1) + (bytesLeftForMl * 255);
+            assert(maxMlSize < INT_MAX);
+            if ((size_t)matchLength > maxMlSize) matchLength= (unsigned)maxMlSize;
+            if ((oend + LASTLITERALS) - (op + ll_totalCost + 2) - 1 + matchLength >= MFLIMIT) {
+            DEBUGLOG(6, "Let's encode a last sequence (ll=%u, ml=%u)", (unsigned)ll, matchLength);
+                LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor),
+                        (int)matchLength, (int)matchDistance,
+                        notLimited, oend);
+        }   }
+        DEBUGLOG(6, "Let's finish with a run of literals (%u bytes left)", (unsigned)(oend-op));
+        goto _lz4mid_last_literals;
+    }
+    /* compression failed */
+    return 0;
+}
+
+
+/**************************************
+*  HC Compression - Search
+**************************************/
+
+/* Update chains up to ip (excluded) */
+LZ4_FORCE_INLINE void LZ4HC_Insert (LZ4HC_CCtx_internal* hc4, const BYTE* ip)
+{
+    U16* const chainTable = hc4->chainTable;
+    U32* const hashTable  = hc4->hashTable;
+    const BYTE* const prefixPtr = hc4->prefixStart;
+    U32 const prefixIdx = hc4->dictLimit;
+    U32 const target = (U32)(ip - prefixPtr) + prefixIdx;
+    U32 idx = hc4->nextToUpdate;
+    assert(ip >= prefixPtr);
+    assert(target >= prefixIdx);
+
+    while (idx < target) {
+        U32 const h = LZ4HC_hashPtr(prefixPtr+idx-prefixIdx);
+        size_t delta = idx - hashTable[h];
+        if (delta>LZ4_DISTANCE_MAX) delta = LZ4_DISTANCE_MAX;
+        DELTANEXTU16(chainTable, idx) = (U16)delta;
+        hashTable[h] = idx;
+        idx++;
+    }
+
+    hc4->nextToUpdate = target;
+}
+
+#if defined(_MSC_VER)
+#  define LZ4HC_rotl32(x,r) _rotl(x,r)
+#else
+#  define LZ4HC_rotl32(x,r) ((x << r) | (x >> (32 - r)))
+#endif
+
+
+static U32 LZ4HC_rotatePattern(size_t const rotate, U32 const pattern)
+{
+    size_t const bitsToRotate = (rotate & (sizeof(pattern) - 1)) << 3;
+    if (bitsToRotate == 0) return pattern;
+    return LZ4HC_rotl32(pattern, (int)bitsToRotate);
+}
+
+/* LZ4HC_countPattern() :
+ * pattern32 must be a sample of repetitive pattern of length 1, 2 or 4 (but not 3!) */
+static unsigned
+LZ4HC_countPattern(const BYTE* ip, const BYTE* const iEnd, U32 const pattern32)
+{
+    const BYTE* const iStart = ip;
+    reg_t const pattern = (sizeof(pattern)==8) ?
+        (reg_t)pattern32 + (((reg_t)pattern32) << (sizeof(pattern)*4)) : pattern32;
+
+    while (likely(ip < iEnd-(sizeof(pattern)-1))) {
+        reg_t const diff = LZ4_read_ARCH(ip) ^ pattern;
+        if (!diff) { ip+=sizeof(pattern); continue; }
+        ip += LZ4_NbCommonBytes(diff);
+        return (unsigned)(ip - iStart);
+    }
+
+    if (LZ4_isLittleEndian()) {
+        reg_t patternByte = pattern;
+        while ((ip<iEnd) && (*ip == (BYTE)patternByte)) {
+            ip++; patternByte >>= 8;
+        }
+    } else {  /* big endian */
+        U32 bitOffset = (sizeof(pattern)*8) - 8;
+        while (ip < iEnd) {
+            BYTE const byte = (BYTE)(pattern >> bitOffset);
+            if (*ip != byte) break;
+            ip ++; bitOffset -= 8;
+    }   }
+
+    return (unsigned)(ip - iStart);
+}
+
+/* LZ4HC_reverseCountPattern() :
+ * pattern must be a sample of repetitive pattern of length 1, 2 or 4 (but not 3!)
+ * read using natural platform endianness */
+static unsigned
+LZ4HC_reverseCountPattern(const BYTE* ip, const BYTE* const iLow, U32 pattern)
+{
+    const BYTE* const iStart = ip;
+
+    while (likely(ip >= iLow+4)) {
+        if (LZ4_read32(ip-4) != pattern) break;
+        ip -= 4;
+    }
+    {   const BYTE* bytePtr = (const BYTE*)(&pattern) + 3; /* works for any endianness */
+        while (likely(ip>iLow)) {
+            if (ip[-1] != *bytePtr) break;
+            ip--; bytePtr--;
+    }   }
+    return (unsigned)(iStart - ip);
+}
+
+/* LZ4HC_protectDictEnd() :
+ * Checks if the match is in the last 3 bytes of the dictionary, so reading the
+ * 4 byte MINMATCH would overflow.
+ * @returns true if the match index is okay.
+ */
+static int LZ4HC_protectDictEnd(U32 const dictLimit, U32 const matchIndex)
+{
+    return ((U32)((dictLimit - 1) - matchIndex) >= 3);
+}
+
+typedef enum { rep_untested, rep_not, rep_confirmed } repeat_state_e;
+typedef enum { favorCompressionRatio=0, favorDecompressionSpeed } HCfavor_e;
+
+
+LZ4_FORCE_INLINE LZ4HC_match_t
+LZ4HC_InsertAndGetWiderMatch (
+        LZ4HC_CCtx_internal* const hc4,
+        const BYTE* const ip,
+        const BYTE* const iLowLimit, const BYTE* const iHighLimit,
+        int longest,
+        const int maxNbAttempts,
+        const int patternAnalysis, const int chainSwap,
+        const dictCtx_directive dict,
+        const HCfavor_e favorDecSpeed)
+{
+    U16* const chainTable = hc4->chainTable;
+    U32* const hashTable = hc4->hashTable;
+    const LZ4HC_CCtx_internal* const dictCtx = hc4->dictCtx;
+    const BYTE* const prefixPtr = hc4->prefixStart;
+    const U32 prefixIdx = hc4->dictLimit;
+    const U32 ipIndex = (U32)(ip - prefixPtr) + prefixIdx;
+    const int withinStartDistance = (hc4->lowLimit + (LZ4_DISTANCE_MAX + 1) > ipIndex);
+    const U32 lowestMatchIndex = (withinStartDistance) ? hc4->lowLimit : ipIndex - LZ4_DISTANCE_MAX;
+    const BYTE* const dictStart = hc4->dictStart;
+    const U32 dictIdx = hc4->lowLimit;
+    const BYTE* const dictEnd = dictStart + prefixIdx - dictIdx;
+    int const lookBackLength = (int)(ip-iLowLimit);
+    int nbAttempts = maxNbAttempts;
+    U32 matchChainPos = 0;
+    U32 const pattern = LZ4_read32(ip);
+    U32 matchIndex;
+    repeat_state_e repeat = rep_untested;
+    size_t srcPatternLength = 0;
+    int offset = 0, sBack = 0;
+
+    DEBUGLOG(7, "LZ4HC_InsertAndGetWiderMatch");
+    /* First Match */
+    LZ4HC_Insert(hc4, ip);  /* insert all prior positions up to ip (excluded) */
+    matchIndex = hashTable[LZ4HC_hashPtr(ip)];
+    DEBUGLOG(7, "First candidate match for pos %u found at index %u / %u (lowestMatchIndex)",
+                ipIndex, matchIndex, lowestMatchIndex);
+
+    while ((matchIndex>=lowestMatchIndex) && (nbAttempts>0)) {
+        int matchLength=0;
+        nbAttempts--;
+        assert(matchIndex < ipIndex);
+        if (favorDecSpeed && (ipIndex - matchIndex < 8)) {
+            /* do nothing:
+             * favorDecSpeed intentionally skips matches with offset < 8 */
+        } else if (matchIndex >= prefixIdx) {   /* within current Prefix */
+            const BYTE* const matchPtr = prefixPtr + (matchIndex - prefixIdx);
+            assert(matchPtr < ip);
+            assert(longest >= 1);
+            if (LZ4_read16(iLowLimit + longest - 1) == LZ4_read16(matchPtr - lookBackLength + longest - 1)) {
+                if (LZ4_read32(matchPtr) == pattern) {
+                    int const back = lookBackLength ? LZ4HC_countBack(ip, matchPtr, iLowLimit, prefixPtr) : 0;
+                    matchLength = MINMATCH + (int)LZ4_count(ip+MINMATCH, matchPtr+MINMATCH, iHighLimit);
+                    matchLength -= back;
+                    if (matchLength > longest) {
+                        longest = matchLength;
+                        offset = (int)(ipIndex - matchIndex);
+                        sBack = back;
+                        DEBUGLOG(7, "Found match of len=%i within prefix, offset=%i, back=%i", longest, offset, -back);
+            }   }   }
+        } else {   /* lowestMatchIndex <= matchIndex < dictLimit : within Ext Dict */
+            const BYTE* const matchPtr = dictStart + (matchIndex - dictIdx);
+            assert(matchIndex >= dictIdx);
+            if ( likely(matchIndex <= prefixIdx - 4)
+              && (LZ4_read32(matchPtr) == pattern) ) {
+                int back = 0;
+                const BYTE* vLimit = ip + (prefixIdx - matchIndex);
+                if (vLimit > iHighLimit) vLimit = iHighLimit;
+                matchLength = (int)LZ4_count(ip+MINMATCH, matchPtr+MINMATCH, vLimit) + MINMATCH;
+                if ((ip+matchLength == vLimit) && (vLimit < iHighLimit))
+                    matchLength += LZ4_count(ip+matchLength, prefixPtr, iHighLimit);
+                back = lookBackLength ? LZ4HC_countBack(ip, matchPtr, iLowLimit, dictStart) : 0;
+                matchLength -= back;
+                if (matchLength > longest) {
+                    longest = matchLength;
+                    offset = (int)(ipIndex - matchIndex);
+                    sBack = back;
+                    DEBUGLOG(7, "Found match of len=%i within dict, offset=%i, back=%i", longest, offset, -back);
+        }   }   }
+
+        if (chainSwap && matchLength==longest) {   /* better match => select a better chain */
+            assert(lookBackLength==0);   /* search forward only */
+            if (matchIndex + (U32)longest <= ipIndex) {
+                int const kTrigger = 4;
+                U32 distanceToNextMatch = 1;
+                int const end = longest - MINMATCH + 1;
+                int step = 1;
+                int accel = 1 << kTrigger;
+                int pos;
+                for (pos = 0; pos < end; pos += step) {
+                    U32 const candidateDist = DELTANEXTU16(chainTable, matchIndex + (U32)pos);
+                    step = (accel++ >> kTrigger);
+                    if (candidateDist > distanceToNextMatch) {
+                        distanceToNextMatch = candidateDist;
+                        matchChainPos = (U32)pos;
+                        accel = 1 << kTrigger;
+                }   }
+                if (distanceToNextMatch > 1) {
+                    if (distanceToNextMatch > matchIndex) break;   /* avoid overflow */
+                    matchIndex -= distanceToNextMatch;
+                    continue;
+        }   }   }
+
+        {   U32 const distNextMatch = DELTANEXTU16(chainTable, matchIndex);
+            if (patternAnalysis && distNextMatch==1 && matchChainPos==0) {
+                U32 const matchCandidateIdx = matchIndex-1;
+                /* may be a repeated pattern */
+                if (repeat == rep_untested) {
+                    if ( ((pattern & 0xFFFF) == (pattern >> 16))
+                      &  ((pattern & 0xFF)   == (pattern >> 24)) ) {
+                        DEBUGLOG(7, "Repeat pattern detected, char %02X", pattern >> 24);
+                        repeat = rep_confirmed;
+                        srcPatternLength = LZ4HC_countPattern(ip+sizeof(pattern), iHighLimit, pattern) + sizeof(pattern);
+                    } else {
+                        repeat = rep_not;
+                }   }
+                if ( (repeat == rep_confirmed) && (matchCandidateIdx >= lowestMatchIndex)
+                  && LZ4HC_protectDictEnd(prefixIdx, matchCandidateIdx) ) {
+                    const int extDict = matchCandidateIdx < prefixIdx;
+                    const BYTE* const matchPtr = extDict ? dictStart + (matchCandidateIdx - dictIdx) : prefixPtr + (matchCandidateIdx - prefixIdx);
+                    if (LZ4_read32(matchPtr) == pattern) {  /* good candidate */
+                        const BYTE* const iLimit = extDict ? dictEnd : iHighLimit;
+                        size_t forwardPatternLength = LZ4HC_countPattern(matchPtr+sizeof(pattern), iLimit, pattern) + sizeof(pattern);
+                        if (extDict && matchPtr + forwardPatternLength == iLimit) {
+                            U32 const rotatedPattern = LZ4HC_rotatePattern(forwardPatternLength, pattern);
+                            forwardPatternLength += LZ4HC_countPattern(prefixPtr, iHighLimit, rotatedPattern);
+                        }
+                        {   const BYTE* const lowestMatchPtr = extDict ? dictStart : prefixPtr;
+                            size_t backLength = LZ4HC_reverseCountPattern(matchPtr, lowestMatchPtr, pattern);
+                            size_t currentSegmentLength;
+                            if (!extDict
+                              && matchPtr - backLength == prefixPtr
+                              && dictIdx < prefixIdx) {
+                                U32 const rotatedPattern = LZ4HC_rotatePattern((U32)(-(int)backLength), pattern);
+                                backLength += LZ4HC_reverseCountPattern(dictEnd, dictStart, rotatedPattern);
+                            }
+                            /* Limit backLength not go further than lowestMatchIndex */
+                            backLength = matchCandidateIdx - MAX(matchCandidateIdx - (U32)backLength, lowestMatchIndex);
+                            assert(matchCandidateIdx - backLength >= lowestMatchIndex);
+                            currentSegmentLength = backLength + forwardPatternLength;
+                            /* Adjust to end of pattern if the source pattern fits, otherwise the beginning of the pattern */
+                            if ( (currentSegmentLength >= srcPatternLength)   /* current pattern segment large enough to contain full srcPatternLength */
+                              && (forwardPatternLength <= srcPatternLength) ) { /* haven't reached this position yet */
+                                U32 const newMatchIndex = matchCandidateIdx + (U32)forwardPatternLength - (U32)srcPatternLength;  /* best position, full pattern, might be followed by more match */
+                                if (LZ4HC_protectDictEnd(prefixIdx, newMatchIndex))
+                                    matchIndex = newMatchIndex;
+                                else {
+                                    /* Can only happen if started in the prefix */
+                                    assert(newMatchIndex >= prefixIdx - 3 && newMatchIndex < prefixIdx && !extDict);
+                                    matchIndex = prefixIdx;
+                                }
+                            } else {
+                                U32 const newMatchIndex = matchCandidateIdx - (U32)backLength;   /* farthest position in current segment, will find a match of length currentSegmentLength + maybe some back */
+                                if (!LZ4HC_protectDictEnd(prefixIdx, newMatchIndex)) {
+                                    assert(newMatchIndex >= prefixIdx - 3 && newMatchIndex < prefixIdx && !extDict);
+                                    matchIndex = prefixIdx;
+                                } else {
+                                    matchIndex = newMatchIndex;
+                                    if (lookBackLength==0) {  /* no back possible */
+                                        size_t const maxML = MIN(currentSegmentLength, srcPatternLength);
+                                        if ((size_t)longest < maxML) {
+                                            assert(prefixPtr - prefixIdx + matchIndex != ip);
+                                            if ((size_t)(ip - prefixPtr) + prefixIdx - matchIndex > LZ4_DISTANCE_MAX) break;
+                                            assert(maxML < 2 GB);
+                                            longest = (int)maxML;
+                                            offset = (int)(ipIndex - matchIndex);
+                                            assert(sBack == 0);
+                                            DEBUGLOG(7, "Found repeat pattern match of len=%i, offset=%i", longest, offset);
+                                        }
+                                        {   U32 const distToNextPattern = DELTANEXTU16(chainTable, matchIndex);
+                                            if (distToNextPattern > matchIndex) break;  /* avoid overflow */
+                                            matchIndex -= distToNextPattern;
+                        }   }   }   }   }
+                        continue;
+                }   }
+        }   }   /* PA optimization */
+
+        /* follow current chain */
+        matchIndex -= DELTANEXTU16(chainTable, matchIndex + matchChainPos);
+
+    }  /* while ((matchIndex>=lowestMatchIndex) && (nbAttempts)) */
+
+    if ( dict == usingDictCtxHc
+      && nbAttempts > 0
+      && withinStartDistance) {
+        size_t const dictEndOffset = (size_t)(dictCtx->end - dictCtx->prefixStart) + dictCtx->dictLimit;
+        U32 dictMatchIndex = dictCtx->hashTable[LZ4HC_hashPtr(ip)];
+        assert(dictEndOffset <= 1 GB);
+        matchIndex = dictMatchIndex + lowestMatchIndex - (U32)dictEndOffset;
+        if (dictMatchIndex>0) DEBUGLOG(7, "dictEndOffset = %zu, dictMatchIndex = %u => relative matchIndex = %i", dictEndOffset, dictMatchIndex, (int)dictMatchIndex - (int)dictEndOffset);
+        while (ipIndex - matchIndex <= LZ4_DISTANCE_MAX && nbAttempts--) {
+            const BYTE* const matchPtr = dictCtx->prefixStart - dictCtx->dictLimit + dictMatchIndex;
+
+            if (LZ4_read32(matchPtr) == pattern) {
+                int mlt;
+                int back = 0;
+                const BYTE* vLimit = ip + (dictEndOffset - dictMatchIndex);
+                if (vLimit > iHighLimit) vLimit = iHighLimit;
+                mlt = (int)LZ4_count(ip+MINMATCH, matchPtr+MINMATCH, vLimit) + MINMATCH;
+                back = lookBackLength ? LZ4HC_countBack(ip, matchPtr, iLowLimit, dictCtx->prefixStart) : 0;
+                mlt -= back;
+                if (mlt > longest) {
+                    longest = mlt;
+                    offset = (int)(ipIndex - matchIndex);
+                    sBack = back;
+                    DEBUGLOG(7, "found match of length %i within extDictCtx", longest);
+            }   }
+
+            {   U32 const nextOffset = DELTANEXTU16(dictCtx->chainTable, dictMatchIndex);
+                dictMatchIndex -= nextOffset;
+                matchIndex -= nextOffset;
+    }   }   }
+
+    {   LZ4HC_match_t md;
+        assert(longest >= 0);
+        md.len = longest;
+        md.off = offset;
+        md.back = sBack;
+        return md;
+    }
+}
+
+LZ4_FORCE_INLINE LZ4HC_match_t
+LZ4HC_InsertAndFindBestMatch(LZ4HC_CCtx_internal* const hc4,   /* Index table will be updated */
+                       const BYTE* const ip, const BYTE* const iLimit,
+                       const int maxNbAttempts,
+                       const int patternAnalysis,
+                       const dictCtx_directive dict)
+{
+    DEBUGLOG(7, "LZ4HC_InsertAndFindBestMatch");
+    /* note : LZ4HC_InsertAndGetWiderMatch() is able to modify the starting position of a match (*startpos),
+     * but this won't be the case here, as we define iLowLimit==ip,
+     * so LZ4HC_InsertAndGetWiderMatch() won't be allowed to search past ip */
+    return LZ4HC_InsertAndGetWiderMatch(hc4, ip, ip, iLimit, MINMATCH-1, maxNbAttempts, patternAnalysis, 0 /*chainSwap*/, dict, favorCompressionRatio);
+}
+
+
+LZ4_FORCE_INLINE int LZ4HC_compress_hashChain (
+    LZ4HC_CCtx_internal* const ctx,
+    const char* const source,
+    char* const dest,
+    int* srcSizePtr,
+    int const maxOutputSize,
+    int maxNbAttempts,
+    const limitedOutput_directive limit,
+    const dictCtx_directive dict
+    )
+{
+    const int inputSize = *srcSizePtr;
+    const int patternAnalysis = (maxNbAttempts > 128);   /* levels 9+ */
+
+    const BYTE* ip = (const BYTE*) source;
+    const BYTE* anchor = ip;
+    const BYTE* const iend = ip + inputSize;
+    const BYTE* const mflimit = iend - MFLIMIT;
+    const BYTE* const matchlimit = (iend - LASTLITERALS);
+
+    BYTE* optr = (BYTE*) dest;
+    BYTE* op = (BYTE*) dest;
+    BYTE* oend = op + maxOutputSize;
+
+    const BYTE* start0;
+    const BYTE* start2 = NULL;
+    const BYTE* start3 = NULL;
+    LZ4HC_match_t m0, m1, m2, m3;
+    const LZ4HC_match_t nomatch = {0, 0, 0};
+
+    /* init */
+    DEBUGLOG(5, "LZ4HC_compress_hashChain (dict?=>%i)", dict);
+    *srcSizePtr = 0;
+    if (limit == fillOutput) oend -= LASTLITERALS;                  /* Hack for support LZ4 format restriction */
+    if (inputSize < LZ4_minLength) goto _last_literals;             /* Input too small, no compression (all literals) */
+
+    /* Main Loop */
+    while (ip <= mflimit) {
+        m1 = LZ4HC_InsertAndFindBestMatch(ctx, ip, matchlimit, maxNbAttempts, patternAnalysis, dict);
+        if (m1.len<MINMATCH) { ip++; continue; }
+
+        /* saved, in case we would skip too much */
+        start0 = ip; m0 = m1;
+
+_Search2:
+        DEBUGLOG(7, "_Search2 (currently found match of size %i)", m1.len);
+        if (ip+m1.len <= mflimit) {
+            start2 = ip + m1.len - 2;
+            m2 = LZ4HC_InsertAndGetWiderMatch(ctx,
+                            start2, ip + 0, matchlimit, m1.len,
+                            maxNbAttempts, patternAnalysis, 0, dict, favorCompressionRatio);
+            start2 += m2.back;
+        } else {
+            m2 = nomatch;  /* do not search further */
+        }
+
+        if (m2.len <= m1.len) { /* No better match => encode ML1 immediately */
+            optr = op;
+            if (LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor),
+                    m1.len, m1.off,
+                    limit, oend) )
+                goto _dest_overflow;
+            continue;
+        }
+
+        if (start0 < ip) {   /* first match was skipped at least once */
+            if (start2 < ip + m0.len) {  /* squeezing ML1 between ML0(original ML1) and ML2 */
+                ip = start0; m1 = m0;  /* restore initial Match1 */
+        }   }
+
+        /* Here, start0==ip */
+        if ((start2 - ip) < 3) {  /* First Match too small : removed */
+            ip = start2;
+            m1 = m2;
+            goto _Search2;
+        }
+
+_Search3:
+        if ((start2 - ip) < OPTIMAL_ML) {
+            int correction;
+            int new_ml = m1.len;
+            if (new_ml > OPTIMAL_ML) new_ml = OPTIMAL_ML;
+            if (ip+new_ml > start2 + m2.len - MINMATCH)
+                new_ml = (int)(start2 - ip) + m2.len - MINMATCH;
+            correction = new_ml - (int)(start2 - ip);
+            if (correction > 0) {
+                start2 += correction;
+                m2.len -= correction;
+            }
+        }
+
+        if (start2 + m2.len <= mflimit) {
+            start3 = start2 + m2.len - 3;
+            m3 = LZ4HC_InsertAndGetWiderMatch(ctx,
+                            start3, start2, matchlimit, m2.len,
+                            maxNbAttempts, patternAnalysis, 0, dict, favorCompressionRatio);
+            start3 += m3.back;
+        } else {
+            m3 = nomatch;  /* do not search further */
+        }
+
+        if (m3.len <= m2.len) {  /* No better match => encode ML1 and ML2 */
+            /* ip & ref are known; Now for ml */
+            if (start2 < ip+m1.len) m1.len = (int)(start2 - ip);
+            /* Now, encode 2 sequences */
+            optr = op;
+            if (LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor),
+                    m1.len, m1.off,
+                    limit, oend) )
+                goto _dest_overflow;
+            ip = start2;
+            optr = op;
+            if (LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor),
+                    m2.len, m2.off,
+                    limit, oend) ) {
+                m1 = m2;
+                goto _dest_overflow;
+            }
+            continue;
+        }
+
+        if (start3 < ip+m1.len+3) {  /* Not enough space for match 2 : remove it */
+            if (start3 >= (ip+m1.len)) {  /* can write Seq1 immediately ==> Seq2 is removed, so Seq3 becomes Seq1 */
+                if (start2 < ip+m1.len) {
+                    int correction = (int)(ip+m1.len - start2);
+                    start2 += correction;
+                    m2.len -= correction;
+                    if (m2.len < MINMATCH) {
+                        start2 = start3;
+                        m2 = m3;
+                    }
+                }
+
+                optr = op;
+                if (LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor),
+                        m1.len, m1.off,
+                        limit, oend) )
+                    goto _dest_overflow;
+                ip  = start3;
+                m1 = m3;
+
+                start0 = start2;
+                m0 = m2;
+                goto _Search2;
+            }
+
+            start2 = start3;
+            m2 = m3;
+            goto _Search3;
+        }
+
+        /*
+        * OK, now we have 3 ascending matches;
+        * let's write the first one ML1.
+        * ip & ref are known; Now decide ml.
+        */
+        if (start2 < ip+m1.len) {
+            if ((start2 - ip) < OPTIMAL_ML) {
+                int correction;
+                if (m1.len > OPTIMAL_ML) m1.len = OPTIMAL_ML;
+                if (ip + m1.len > start2 + m2.len - MINMATCH)
+                    m1.len = (int)(start2 - ip) + m2.len - MINMATCH;
+                correction = m1.len - (int)(start2 - ip);
+                if (correction > 0) {
+                    start2 += correction;
+                    m2.len -= correction;
+                }
+            } else {
+                m1.len = (int)(start2 - ip);
+            }
+        }
+        optr = op;
+        if ( LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor),
+                m1.len, m1.off,
+                limit, oend) )
+            goto _dest_overflow;
+
+        /* ML2 becomes ML1 */
+        ip = start2; m1 = m2;
+
+        /* ML3 becomes ML2 */
+        start2 = start3; m2 = m3;
+
+        /* let's find a new ML3 */
+        goto _Search3;
+    }
+
+_last_literals:
+    /* Encode Last Literals */
+    {   size_t lastRunSize = (size_t)(iend - anchor);  /* literals */
+        size_t llAdd = (lastRunSize + 255 - RUN_MASK) / 255;
+        size_t const totalSize = 1 + llAdd + lastRunSize;
+        if (limit == fillOutput) oend += LASTLITERALS;  /* restore correct value */
+        if (limit && (op + totalSize > oend)) {
+            if (limit == limitedOutput) return 0;
+            /* adapt lastRunSize to fill 'dest' */
+            lastRunSize  = (size_t)(oend - op) - 1 /*token*/;
+            llAdd = (lastRunSize + 256 - RUN_MASK) / 256;
+            lastRunSize -= llAdd;
+        }
+        DEBUGLOG(6, "Final literal run : %i literals", (int)lastRunSize);
+        ip = anchor + lastRunSize;  /* can be != iend if limit==fillOutput */
+
+        if (lastRunSize >= RUN_MASK) {
+            size_t accumulator = lastRunSize - RUN_MASK;
+            *op++ = (RUN_MASK << ML_BITS);
+            for(; accumulator >= 255 ; accumulator -= 255) *op++ = 255;
+            *op++ = (BYTE) accumulator;
+        } else {
+            *op++ = (BYTE)(lastRunSize << ML_BITS);
+        }
+        LZ4_memcpy(op, anchor, lastRunSize);
+        op += lastRunSize;
+    }
+
+    /* End */
+    *srcSizePtr = (int) (((const char*)ip) - source);
+    return (int) (((char*)op)-dest);
+
+_dest_overflow:
+    if (limit == fillOutput) {
+        /* Assumption : @ip, @anchor, @optr and @m1 must be set correctly */
+        size_t const ll = (size_t)(ip - anchor);
+        size_t const ll_addbytes = (ll + 240) / 255;
+        size_t const ll_totalCost = 1 + ll_addbytes + ll;
+        BYTE* const maxLitPos = oend - 3; /* 2 for offset, 1 for token */
+        DEBUGLOG(6, "Last sequence overflowing");
+        op = optr;  /* restore correct out pointer */
+        if (op + ll_totalCost <= maxLitPos) {
+            /* ll validated; now adjust match length */
+            size_t const bytesLeftForMl = (size_t)(maxLitPos - (op+ll_totalCost));
+            size_t const maxMlSize = MINMATCH + (ML_MASK-1) + (bytesLeftForMl * 255);
+            assert(maxMlSize < INT_MAX); assert(m1.len >= 0);
+            if ((size_t)m1.len > maxMlSize) m1.len = (int)maxMlSize;
+            if ((oend + LASTLITERALS) - (op + ll_totalCost + 2) - 1 + m1.len >= MFLIMIT) {
+                LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), m1.len, m1.off, notLimited, oend);
+        }   }
+        goto _last_literals;
+    }
+    /* compression failed */
+    return 0;
+}
+
+
+static int LZ4HC_compress_optimal( LZ4HC_CCtx_internal* ctx,
+    const char* const source, char* dst,
+    int* srcSizePtr, int dstCapacity,
+    int const nbSearches, size_t sufficient_len,
+    const limitedOutput_directive limit, int const fullUpdate,
+    const dictCtx_directive dict,
+    const HCfavor_e favorDecSpeed);
+
+LZ4_FORCE_INLINE int
+LZ4HC_compress_generic_internal (
+            LZ4HC_CCtx_internal* const ctx,
+            const char* const src,
+            char* const dst,
+            int* const srcSizePtr,
+            int const dstCapacity,
+            int cLevel,
+            const limitedOutput_directive limit,
+            const dictCtx_directive dict
+            )
+{
+    DEBUGLOG(5, "LZ4HC_compress_generic_internal(src=%p, srcSize=%d)",
+                src, *srcSizePtr);
+
+    if (limit == fillOutput && dstCapacity < 1) return 0;   /* Impossible to store anything */
+    if ((U32)*srcSizePtr > (U32)LZ4_MAX_INPUT_SIZE) return 0;  /* Unsupported input size (too large or negative) */
+
+    ctx->end += *srcSizePtr;
+    {   cParams_t const cParam = LZ4HC_getCLevelParams(cLevel);
+        HCfavor_e const favor = ctx->favorDecSpeed ? favorDecompressionSpeed : favorCompressionRatio;
+        int result;
+
+        if (cParam.strat == lz4mid) {
+            result = LZ4MID_compress(ctx,
+                                src, dst, srcSizePtr, dstCapacity,
+                                limit, dict);
+        } else if (cParam.strat == lz4hc) {
+            result = LZ4HC_compress_hashChain(ctx,
+                                src, dst, srcSizePtr, dstCapacity,
+                                cParam.nbSearches, limit, dict);
+        } else {
+            assert(cParam.strat == lz4opt);
+            result = LZ4HC_compress_optimal(ctx,
+                                src, dst, srcSizePtr, dstCapacity,
+                                cParam.nbSearches, cParam.targetLength, limit,
+                                cLevel >= LZ4HC_CLEVEL_MAX,   /* ultra mode */
+                                dict, favor);
+        }
+        if (result <= 0) ctx->dirty = 1;
+        return result;
+    }
+}
+
+static void LZ4HC_setExternalDict(LZ4HC_CCtx_internal* ctxPtr, const BYTE* newBlock);
+
+static int
+LZ4HC_compress_generic_noDictCtx (
+        LZ4HC_CCtx_internal* const ctx,
+        const char* const src,
+        char* const dst,
+        int* const srcSizePtr,
+        int const dstCapacity,
+        int cLevel,
+        limitedOutput_directive limit
+        )
+{
+    assert(ctx->dictCtx == NULL);
+    return LZ4HC_compress_generic_internal(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit, noDictCtx);
+}
+
+static int isStateCompatible(const LZ4HC_CCtx_internal* ctx1, const LZ4HC_CCtx_internal* ctx2)
+{
+    int const isMid1 = LZ4HC_getCLevelParams(ctx1->compressionLevel).strat == lz4mid;
+    int const isMid2 = LZ4HC_getCLevelParams(ctx2->compressionLevel).strat == lz4mid;
+    return !(isMid1 ^ isMid2);
+}
+
+static int
+LZ4HC_compress_generic_dictCtx (
+        LZ4HC_CCtx_internal* const ctx,
+        const char* const src,
+        char* const dst,
+        int* const srcSizePtr,
+        int const dstCapacity,
+        int cLevel,
+        limitedOutput_directive limit
+        )
+{
+    const size_t position = (size_t)(ctx->end - ctx->prefixStart) + (ctx->dictLimit - ctx->lowLimit);
+    assert(ctx->dictCtx != NULL);
+    if (position >= 64 KB) {
+        ctx->dictCtx = NULL;
+        return LZ4HC_compress_generic_noDictCtx(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit);
+    } else if (position == 0 && *srcSizePtr > 4 KB && isStateCompatible(ctx, ctx->dictCtx)) {
+        LZ4_memcpy(ctx, ctx->dictCtx, sizeof(LZ4HC_CCtx_internal));
+        LZ4HC_setExternalDict(ctx, (const BYTE *)src);
+        ctx->compressionLevel = (short)cLevel;
+        return LZ4HC_compress_generic_noDictCtx(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit);
+    } else {
+        return LZ4HC_compress_generic_internal(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit, usingDictCtxHc);
+    }
+}
+
+static int
+LZ4HC_compress_generic (
+        LZ4HC_CCtx_internal* const ctx,
+        const char* const src,
+        char* const dst,
+        int* const srcSizePtr,
+        int const dstCapacity,
+        int cLevel,
+        limitedOutput_directive limit
+        )
+{
+    if (ctx->dictCtx == NULL) {
+        return LZ4HC_compress_generic_noDictCtx(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit);
+    } else {
+        return LZ4HC_compress_generic_dictCtx(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit);
+    }
+}
+
+
+int LZ4_sizeofStateHC(void) { return (int)sizeof(LZ4_streamHC_t); }
+
+static size_t LZ4_streamHC_t_alignment(void)
+{
+#if LZ4_ALIGN_TEST
+    typedef struct { char c; LZ4_streamHC_t t; } t_a;
+    return sizeof(t_a) - sizeof(LZ4_streamHC_t);
+#else
+    return 1;  /* effectively disabled */
+#endif
+}
+
+/* state is presumed correctly initialized,
+ * in which case its size and alignment have already been validate */
+int LZ4_compress_HC_extStateHC_fastReset (void* state, const char* src, char* dst, int srcSize, int dstCapacity, int compressionLevel)
+{
+    LZ4HC_CCtx_internal* const ctx = &((LZ4_streamHC_t*)state)->internal_donotuse;
+    if (!LZ4_isAligned(state, LZ4_streamHC_t_alignment())) return 0;
+    LZ4_resetStreamHC_fast((LZ4_streamHC_t*)state, compressionLevel);
+    LZ4HC_init_internal (ctx, (const BYTE*)src);
+    if (dstCapacity < LZ4_compressBound(srcSize))
+        return LZ4HC_compress_generic (ctx, src, dst, &srcSize, dstCapacity, compressionLevel, limitedOutput);
+    else
+        return LZ4HC_compress_generic (ctx, src, dst, &srcSize, dstCapacity, compressionLevel, notLimited);
+}
+
+int LZ4_compress_HC_extStateHC (void* state, const char* src, char* dst, int srcSize, int dstCapacity, int compressionLevel)
+{
+    LZ4_streamHC_t* const ctx = LZ4_initStreamHC(state, sizeof(*ctx));
+    if (ctx==NULL) return 0;   /* init failure */
+    return LZ4_compress_HC_extStateHC_fastReset(state, src, dst, srcSize, dstCapacity, compressionLevel);
+}
+
+int LZ4_compress_HC(const char* src, char* dst, int srcSize, int dstCapacity, int compressionLevel)
+{
+    int cSize;
+#if defined(LZ4HC_HEAPMODE) && LZ4HC_HEAPMODE==1
+    LZ4_streamHC_t* const statePtr = (LZ4_streamHC_t*)ALLOC(sizeof(LZ4_streamHC_t));
+    if (statePtr==NULL) return 0;
+#else
+    LZ4_streamHC_t state;
+    LZ4_streamHC_t* const statePtr = &state;
+#endif
+    DEBUGLOG(5, "LZ4_compress_HC")
+    cSize = LZ4_compress_HC_extStateHC(statePtr, src, dst, srcSize, dstCapacity, compressionLevel);
+#if defined(LZ4HC_HEAPMODE) && LZ4HC_HEAPMODE==1
+    FREEMEM(statePtr);
+#endif
+    return cSize;
+}
+
+/* state is presumed sized correctly (>= sizeof(LZ4_streamHC_t)) */
+int LZ4_compress_HC_destSize(void* state, const char* source, char* dest, int* sourceSizePtr, int targetDestSize, int cLevel)
+{
+    LZ4_streamHC_t* const ctx = LZ4_initStreamHC(state, sizeof(*ctx));
+    if (ctx==NULL) return 0;   /* init failure */
+    LZ4HC_init_internal(&ctx->internal_donotuse, (const BYTE*) source);
+    LZ4_setCompressionLevel(ctx, cLevel);
+    return LZ4HC_compress_generic(&ctx->internal_donotuse, source, dest, sourceSizePtr, targetDestSize, cLevel, fillOutput);
+}
+
+
+
+/**************************************
+*  Streaming Functions
+**************************************/
+/* allocation */
+#if !defined(LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION)
+LZ4_streamHC_t* LZ4_createStreamHC(void)
+{
+    LZ4_streamHC_t* const state =
+        (LZ4_streamHC_t*)ALLOC_AND_ZERO(sizeof(LZ4_streamHC_t));
+    if (state == NULL) return NULL;
+    LZ4_setCompressionLevel(state, LZ4HC_CLEVEL_DEFAULT);
+    return state;
+}
+
+int LZ4_freeStreamHC (LZ4_streamHC_t* LZ4_streamHCPtr)
+{
+    DEBUGLOG(4, "LZ4_freeStreamHC(%p)", LZ4_streamHCPtr);
+    if (!LZ4_streamHCPtr) return 0;  /* support free on NULL */
+    FREEMEM(LZ4_streamHCPtr);
+    return 0;
+}
+#endif
+
+
+LZ4_streamHC_t* LZ4_initStreamHC (void* buffer, size_t size)
+{
+    LZ4_streamHC_t* const LZ4_streamHCPtr = (LZ4_streamHC_t*)buffer;
+    DEBUGLOG(4, "LZ4_initStreamHC(%p, %u)", buffer, (unsigned)size);
+    /* check conditions */
+    if (buffer == NULL) return NULL;
+    if (size < sizeof(LZ4_streamHC_t)) return NULL;
+    if (!LZ4_isAligned(buffer, LZ4_streamHC_t_alignment())) return NULL;
+    /* init */
+    { LZ4HC_CCtx_internal* const hcstate = &(LZ4_streamHCPtr->internal_donotuse);
+      MEM_INIT(hcstate, 0, sizeof(*hcstate)); }
+    LZ4_setCompressionLevel(LZ4_streamHCPtr, LZ4HC_CLEVEL_DEFAULT);
+    return LZ4_streamHCPtr;
+}
+
+/* just a stub */
+void LZ4_resetStreamHC (LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel)
+{
+    LZ4_initStreamHC(LZ4_streamHCPtr, sizeof(*LZ4_streamHCPtr));
+    LZ4_setCompressionLevel(LZ4_streamHCPtr, compressionLevel);
+}
+
+void LZ4_resetStreamHC_fast (LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel)
+{
+    LZ4HC_CCtx_internal* const s = &LZ4_streamHCPtr->internal_donotuse;
+    DEBUGLOG(5, "LZ4_resetStreamHC_fast(%p, %d)", LZ4_streamHCPtr, compressionLevel);
+    if (s->dirty) {
+        LZ4_initStreamHC(LZ4_streamHCPtr, sizeof(*LZ4_streamHCPtr));
+    } else {
+        assert(s->end >= s->prefixStart);
+        s->dictLimit += (U32)(s->end - s->prefixStart);
+        s->prefixStart = NULL;
+        s->end = NULL;
+        s->dictCtx = NULL;
+    }
+    LZ4_setCompressionLevel(LZ4_streamHCPtr, compressionLevel);
+}
+
+void LZ4_setCompressionLevel(LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel)
+{
+    DEBUGLOG(5, "LZ4_setCompressionLevel(%p, %d)", LZ4_streamHCPtr, compressionLevel);
+    if (compressionLevel < 1) compressionLevel = LZ4HC_CLEVEL_DEFAULT;
+    if (compressionLevel > LZ4HC_CLEVEL_MAX) compressionLevel = LZ4HC_CLEVEL_MAX;
+    LZ4_streamHCPtr->internal_donotuse.compressionLevel = (short)compressionLevel;
+}
+
+void LZ4_favorDecompressionSpeed(LZ4_streamHC_t* LZ4_streamHCPtr, int favor)
+{
+    LZ4_streamHCPtr->internal_donotuse.favorDecSpeed = (favor!=0);
+}
+
+/* LZ4_loadDictHC() :
+ * LZ4_streamHCPtr is presumed properly initialized */
+int LZ4_loadDictHC (LZ4_streamHC_t* LZ4_streamHCPtr,
+              const char* dictionary, int dictSize)
+{
+    LZ4HC_CCtx_internal* const ctxPtr = &LZ4_streamHCPtr->internal_donotuse;
+    cParams_t cp;
+    DEBUGLOG(4, "LZ4_loadDictHC(ctx:%p, dict:%p, dictSize:%d, clevel=%d)", LZ4_streamHCPtr, dictionary, dictSize, ctxPtr->compressionLevel);
+    assert(dictSize >= 0);
+    assert(LZ4_streamHCPtr != NULL);
+    if (dictSize > 64 KB) {
+        dictionary += (size_t)dictSize - 64 KB;
+        dictSize = 64 KB;
+    }
+    /* need a full initialization, there are bad side-effects when using resetFast() */
+    {   int const cLevel = ctxPtr->compressionLevel;
+        LZ4_initStreamHC(LZ4_streamHCPtr, sizeof(*LZ4_streamHCPtr));
+        LZ4_setCompressionLevel(LZ4_streamHCPtr, cLevel);
+        cp = LZ4HC_getCLevelParams(cLevel);
+    }
+    LZ4HC_init_internal (ctxPtr, (const BYTE*)dictionary);
+    ctxPtr->end = (const BYTE*)dictionary + dictSize;
+    if (cp.strat == lz4mid) {
+        LZ4MID_fillHTable (ctxPtr, dictionary, (size_t)dictSize);
+    } else {
+        if (dictSize >= LZ4HC_HASHSIZE) LZ4HC_Insert (ctxPtr, ctxPtr->end-3);
+    }
+    return dictSize;
+}
+
+void LZ4_attach_HC_dictionary(LZ4_streamHC_t *working_stream, const LZ4_streamHC_t *dictionary_stream) {
+    working_stream->internal_donotuse.dictCtx = dictionary_stream != NULL ? &(dictionary_stream->internal_donotuse) : NULL;
+}
+
+/* compression */
+
+static void LZ4HC_setExternalDict(LZ4HC_CCtx_internal* ctxPtr, const BYTE* newBlock)
+{
+    DEBUGLOG(4, "LZ4HC_setExternalDict(%p, %p)", ctxPtr, newBlock);
+    if ( (ctxPtr->end >= ctxPtr->prefixStart + 4)
+      && (LZ4HC_getCLevelParams(ctxPtr->compressionLevel).strat != lz4mid) ) {
+        LZ4HC_Insert (ctxPtr, ctxPtr->end-3);  /* Referencing remaining dictionary content */
+    }
+
+    /* Only one memory segment for extDict, so any previous extDict is lost at this stage */
+    ctxPtr->lowLimit  = ctxPtr->dictLimit;
+    ctxPtr->dictStart  = ctxPtr->prefixStart;
+    ctxPtr->dictLimit += (U32)(ctxPtr->end - ctxPtr->prefixStart);
+    ctxPtr->prefixStart = newBlock;
+    ctxPtr->end  = newBlock;
+    ctxPtr->nextToUpdate = ctxPtr->dictLimit;   /* match referencing will resume from there */
+
+    /* cannot reference an extDict and a dictCtx at the same time */
+    ctxPtr->dictCtx = NULL;
+}
+
+static int
+LZ4_compressHC_continue_generic (LZ4_streamHC_t* LZ4_streamHCPtr,
+                                 const char* src, char* dst,
+                                 int* srcSizePtr, int dstCapacity,
+                                 limitedOutput_directive limit)
+{
+    LZ4HC_CCtx_internal* const ctxPtr = &LZ4_streamHCPtr->internal_donotuse;
+    DEBUGLOG(5, "LZ4_compressHC_continue_generic(ctx=%p, src=%p, srcSize=%d, limit=%d)",
+                LZ4_streamHCPtr, src, *srcSizePtr, limit);
+    assert(ctxPtr != NULL);
+    /* auto-init if forgotten */
+    if (ctxPtr->prefixStart == NULL)
+        LZ4HC_init_internal (ctxPtr, (const BYTE*) src);
+
+    /* Check overflow */
+    if ((size_t)(ctxPtr->end - ctxPtr->prefixStart) + ctxPtr->dictLimit > 2 GB) {
+        size_t dictSize = (size_t)(ctxPtr->end - ctxPtr->prefixStart);
+        if (dictSize > 64 KB) dictSize = 64 KB;
+        LZ4_loadDictHC(LZ4_streamHCPtr, (const char*)(ctxPtr->end) - dictSize, (int)dictSize);
+    }
+
+    /* Check if blocks follow each other */
+    if ((const BYTE*)src != ctxPtr->end)
+        LZ4HC_setExternalDict(ctxPtr, (const BYTE*)src);
+
+    /* Check overlapping input/dictionary space */
+    {   const BYTE* sourceEnd = (const BYTE*) src + *srcSizePtr;
+        const BYTE* const dictBegin = ctxPtr->dictStart;
+        const BYTE* const dictEnd   = ctxPtr->dictStart + (ctxPtr->dictLimit - ctxPtr->lowLimit);
+        if ((sourceEnd > dictBegin) && ((const BYTE*)src < dictEnd)) {
+            if (sourceEnd > dictEnd) sourceEnd = dictEnd;
+            ctxPtr->lowLimit += (U32)(sourceEnd - ctxPtr->dictStart);
+            ctxPtr->dictStart += (U32)(sourceEnd - ctxPtr->dictStart);
+            /* invalidate dictionary is it's too small */
+            if (ctxPtr->dictLimit - ctxPtr->lowLimit < LZ4HC_HASHSIZE) {
+                ctxPtr->lowLimit = ctxPtr->dictLimit;
+                ctxPtr->dictStart = ctxPtr->prefixStart;
+    }   }   }
+
+    return LZ4HC_compress_generic (ctxPtr, src, dst, srcSizePtr, dstCapacity, ctxPtr->compressionLevel, limit);
+}
+
+int LZ4_compress_HC_continue (LZ4_streamHC_t* LZ4_streamHCPtr, const char* src, char* dst, int srcSize, int dstCapacity)
+{
+    DEBUGLOG(5, "LZ4_compress_HC_continue");
+    if (dstCapacity < LZ4_compressBound(srcSize))
+        return LZ4_compressHC_continue_generic (LZ4_streamHCPtr, src, dst, &srcSize, dstCapacity, limitedOutput);
+    else
+        return LZ4_compressHC_continue_generic (LZ4_streamHCPtr, src, dst, &srcSize, dstCapacity, notLimited);
+}
+
+int LZ4_compress_HC_continue_destSize (LZ4_streamHC_t* LZ4_streamHCPtr, const char* src, char* dst, int* srcSizePtr, int targetDestSize)
+{
+    return LZ4_compressHC_continue_generic(LZ4_streamHCPtr, src, dst, srcSizePtr, targetDestSize, fillOutput);
+}
+
+
+/* LZ4_saveDictHC :
+ * save history content
+ * into a user-provided buffer
+ * which is then used to continue compression
+ */
+int LZ4_saveDictHC (LZ4_streamHC_t* LZ4_streamHCPtr, char* safeBuffer, int dictSize)
+{
+    LZ4HC_CCtx_internal* const streamPtr = &LZ4_streamHCPtr->internal_donotuse;
+    int const prefixSize = (int)(streamPtr->end - streamPtr->prefixStart);
+    DEBUGLOG(5, "LZ4_saveDictHC(%p, %p, %d)", LZ4_streamHCPtr, safeBuffer, dictSize);
+    assert(prefixSize >= 0);
+    if (dictSize > 64 KB) dictSize = 64 KB;
+    if (dictSize < 4) dictSize = 0;
+    if (dictSize > prefixSize) dictSize = prefixSize;
+    if (safeBuffer == NULL) assert(dictSize == 0);
+    if (dictSize > 0)
+        LZ4_memmove(safeBuffer, streamPtr->end - dictSize, (size_t)dictSize);
+    {   U32 const endIndex = (U32)(streamPtr->end - streamPtr->prefixStart) + streamPtr->dictLimit;
+        streamPtr->end = (safeBuffer == NULL) ? NULL : (const BYTE*)safeBuffer + dictSize;
+        streamPtr->prefixStart = (const BYTE*)safeBuffer;
+        streamPtr->dictLimit = endIndex - (U32)dictSize;
+        streamPtr->lowLimit = endIndex - (U32)dictSize;
+        streamPtr->dictStart = streamPtr->prefixStart;
+        if (streamPtr->nextToUpdate < streamPtr->dictLimit)
+            streamPtr->nextToUpdate = streamPtr->dictLimit;
+    }
+    return dictSize;
+}
+
+
+/* ================================================
+ *  LZ4 Optimal parser (levels [LZ4HC_CLEVEL_OPT_MIN - LZ4HC_CLEVEL_MAX])
+ * ===============================================*/
+typedef struct {
+    int price;
+    int off;
+    int mlen;
+    int litlen;
+} LZ4HC_optimal_t;
+
+/* price in bytes */
+LZ4_FORCE_INLINE int LZ4HC_literalsPrice(int const litlen)
+{
+    int price = litlen;
+    assert(litlen >= 0);
+    if (litlen >= (int)RUN_MASK)
+        price += 1 + ((litlen-(int)RUN_MASK) / 255);
+    return price;
+}
+
+/* requires mlen >= MINMATCH */
+LZ4_FORCE_INLINE int LZ4HC_sequencePrice(int litlen, int mlen)
+{
+    int price = 1 + 2 ; /* token + 16-bit offset */
+    assert(litlen >= 0);
+    assert(mlen >= MINMATCH);
+
+    price += LZ4HC_literalsPrice(litlen);
+
+    if (mlen >= (int)(ML_MASK+MINMATCH))
+        price += 1 + ((mlen-(int)(ML_MASK+MINMATCH)) / 255);
+
+    return price;
+}
+
+LZ4_FORCE_INLINE LZ4HC_match_t
+LZ4HC_FindLongerMatch(LZ4HC_CCtx_internal* const ctx,
+                      const BYTE* ip, const BYTE* const iHighLimit,
+                      int minLen, int nbSearches,
+                      const dictCtx_directive dict,
+                      const HCfavor_e favorDecSpeed)
+{
+    LZ4HC_match_t const match0 = { 0 , 0, 0 };
+    /* note : LZ4HC_InsertAndGetWiderMatch() is able to modify the starting position of a match (*startpos),
+     * but this won't be the case here, as we define iLowLimit==ip,
+    ** so LZ4HC_InsertAndGetWiderMatch() won't be allowed to search past ip */
+    LZ4HC_match_t md = LZ4HC_InsertAndGetWiderMatch(ctx, ip, ip, iHighLimit, minLen, nbSearches, 1 /*patternAnalysis*/, 1 /*chainSwap*/, dict, favorDecSpeed);
+    assert(md.back == 0);
+    if (md.len <= minLen) return match0;
+    if (favorDecSpeed) {
+        if ((md.len>18) & (md.len<=36)) md.len=18;   /* favor dec.speed (shortcut) */
+    }
+    return md;
+}
+
+
+static int LZ4HC_compress_optimal ( LZ4HC_CCtx_internal* ctx,
+                                    const char* const source,
+                                    char* dst,
+                                    int* srcSizePtr,
+                                    int dstCapacity,
+                                    int const nbSearches,
+                                    size_t sufficient_len,
+                                    const limitedOutput_directive limit,
+                                    int const fullUpdate,
+                                    const dictCtx_directive dict,
+                                    const HCfavor_e favorDecSpeed)
+{
+    int retval = 0;
+#define TRAILING_LITERALS 3
+#if defined(LZ4HC_HEAPMODE) && LZ4HC_HEAPMODE==1
+    LZ4HC_optimal_t* const opt = (LZ4HC_optimal_t*)ALLOC(sizeof(LZ4HC_optimal_t) * (LZ4_OPT_NUM + TRAILING_LITERALS));
+#else
+    LZ4HC_optimal_t opt[LZ4_OPT_NUM + TRAILING_LITERALS];   /* ~64 KB, which is a bit large for stack... */
+#endif
+
+    const BYTE* ip = (const BYTE*) source;
+    const BYTE* anchor = ip;
+    const BYTE* const iend = ip + *srcSizePtr;
+    const BYTE* const mflimit = iend - MFLIMIT;
+    const BYTE* const matchlimit = iend - LASTLITERALS;
+    BYTE* op = (BYTE*) dst;
+    BYTE* opSaved = (BYTE*) dst;
+    BYTE* oend = op + dstCapacity;
+    int ovml = MINMATCH;  /* overflow - last sequence */
+    int ovoff = 0;
+
+    /* init */
+#if defined(LZ4HC_HEAPMODE) && LZ4HC_HEAPMODE==1
+    if (opt == NULL) goto _return_label;
+#endif
+    DEBUGLOG(5, "LZ4HC_compress_optimal(dst=%p, dstCapa=%u)", dst, (unsigned)dstCapacity);
+    *srcSizePtr = 0;
+    if (limit == fillOutput) oend -= LASTLITERALS;   /* Hack for support LZ4 format restriction */
+    if (sufficient_len >= LZ4_OPT_NUM) sufficient_len = LZ4_OPT_NUM-1;
+
+    /* Main Loop */
+    while (ip <= mflimit) {
+         int const llen = (int)(ip - anchor);
+         int best_mlen, best_off;
+         int cur, last_match_pos = 0;
+
+         LZ4HC_match_t const firstMatch = LZ4HC_FindLongerMatch(ctx, ip, matchlimit, MINMATCH-1, nbSearches, dict, favorDecSpeed);
+         if (firstMatch.len==0) { ip++; continue; }
+
+         if ((size_t)firstMatch.len > sufficient_len) {
+             /* good enough solution : immediate encoding */
+             int const firstML = firstMatch.len;
+             opSaved = op;
+             if ( LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), firstML, firstMatch.off, limit, oend) ) {  /* updates ip, op and anchor */
+                 ovml = firstML;
+                 ovoff = firstMatch.off;
+                 goto _dest_overflow;
+             }
+             continue;
+         }
+
+         /* set prices for first positions (literals) */
+         {   int rPos;
+             for (rPos = 0 ; rPos < MINMATCH ; rPos++) {
+                 int const cost = LZ4HC_literalsPrice(llen + rPos);
+                 opt[rPos].mlen = 1;
+                 opt[rPos].off = 0;
+                 opt[rPos].litlen = llen + rPos;
+                 opt[rPos].price = cost;
+                 DEBUGLOG(7, "rPos:%3i => price:%3i (litlen=%i) -- initial setup",
+                             rPos, cost, opt[rPos].litlen);
+         }   }
+         /* set prices using initial match */
+         {   int const matchML = firstMatch.len;   /* necessarily < sufficient_len < LZ4_OPT_NUM */
+             int const offset = firstMatch.off;
+             int mlen;
+             assert(matchML < LZ4_OPT_NUM);
+             for (mlen = MINMATCH ; mlen <= matchML ; mlen++) {
+                 int const cost = LZ4HC_sequencePrice(llen, mlen);
+                 opt[mlen].mlen = mlen;
+                 opt[mlen].off = offset;
+                 opt[mlen].litlen = llen;
+                 opt[mlen].price = cost;
+                 DEBUGLOG(7, "rPos:%3i => price:%3i (matchlen=%i) -- initial setup",
+                             mlen, cost, mlen);
+         }   }
+         last_match_pos = firstMatch.len;
+         {   int addLit;
+             for (addLit = 1; addLit <= TRAILING_LITERALS; addLit ++) {
+                 opt[last_match_pos+addLit].mlen = 1; /* literal */
+                 opt[last_match_pos+addLit].off = 0;
+                 opt[last_match_pos+addLit].litlen = addLit;
+                 opt[last_match_pos+addLit].price = opt[last_match_pos].price + LZ4HC_literalsPrice(addLit);
+                 DEBUGLOG(7, "rPos:%3i => price:%3i (litlen=%i) -- initial setup",
+                             last_match_pos+addLit, opt[last_match_pos+addLit].price, addLit);
+         }   }
+
+         /* check further positions */
+         for (cur = 1; cur < last_match_pos; cur++) {
+             const BYTE* const curPtr = ip + cur;
+             LZ4HC_match_t newMatch;
+
+             if (curPtr > mflimit) break;
+             DEBUGLOG(7, "rPos:%u[%u] vs [%u]%u",
+                     cur, opt[cur].price, opt[cur+1].price, cur+1);
+             if (fullUpdate) {
+                 /* not useful to search here if next position has same (or lower) cost */
+                 if ( (opt[cur+1].price <= opt[cur].price)
+                   /* in some cases, next position has same cost, but cost rises sharply after, so a small match would still be beneficial */
+                   && (opt[cur+MINMATCH].price < opt[cur].price + 3/*min seq price*/) )
+                     continue;
+             } else {
+                 /* not useful to search here if next position has same (or lower) cost */
+                 if (opt[cur+1].price <= opt[cur].price) continue;
+             }
+
+             DEBUGLOG(7, "search at rPos:%u", cur);
+             if (fullUpdate)
+                 newMatch = LZ4HC_FindLongerMatch(ctx, curPtr, matchlimit, MINMATCH-1, nbSearches, dict, favorDecSpeed);
+             else
+                 /* only test matches of minimum length; slightly faster, but misses a few bytes */
+                 newMatch = LZ4HC_FindLongerMatch(ctx, curPtr, matchlimit, last_match_pos - cur, nbSearches, dict, favorDecSpeed);
+             if (!newMatch.len) continue;
+
+             if ( ((size_t)newMatch.len > sufficient_len)
+               || (newMatch.len + cur >= LZ4_OPT_NUM) ) {
+                 /* immediate encoding */
+                 best_mlen = newMatch.len;
+                 best_off = newMatch.off;
+                 last_match_pos = cur + 1;
+                 goto encode;
+             }
+
+             /* before match : set price with literals at beginning */
+             {   int const baseLitlen = opt[cur].litlen;
+                 int litlen;
+                 for (litlen = 1; litlen < MINMATCH; litlen++) {
+                     int const price = opt[cur].price - LZ4HC_literalsPrice(baseLitlen) + LZ4HC_literalsPrice(baseLitlen+litlen);
+                     int const pos = cur + litlen;
+                     if (price < opt[pos].price) {
+                         opt[pos].mlen = 1; /* literal */
+                         opt[pos].off = 0;
+                         opt[pos].litlen = baseLitlen+litlen;
+                         opt[pos].price = price;
+                         DEBUGLOG(7, "rPos:%3i => price:%3i (litlen=%i)",
+                                     pos, price, opt[pos].litlen);
+             }   }   }
+
+             /* set prices using match at position = cur */
+             {   int const matchML = newMatch.len;
+                 int ml = MINMATCH;
+
+                 assert(cur + newMatch.len < LZ4_OPT_NUM);
+                 for ( ; ml <= matchML ; ml++) {
+                     int const pos = cur + ml;
+                     int const offset = newMatch.off;
+                     int price;
+                     int ll;
+                     DEBUGLOG(7, "testing price rPos %i (last_match_pos=%i)",
+                                 pos, last_match_pos);
+                     if (opt[cur].mlen == 1) {
+                         ll = opt[cur].litlen;
+                         price = ((cur > ll) ? opt[cur - ll].price : 0)
+                               + LZ4HC_sequencePrice(ll, ml);
+                     } else {
+                         ll = 0;
+                         price = opt[cur].price + LZ4HC_sequencePrice(0, ml);
+                     }
+
+                    assert((U32)favorDecSpeed <= 1);
+                     if (pos > last_match_pos+TRAILING_LITERALS
+                      || price <= opt[pos].price - (int)favorDecSpeed) {
+                         DEBUGLOG(7, "rPos:%3i => price:%3i (matchlen=%i)",
+                                     pos, price, ml);
+                         assert(pos < LZ4_OPT_NUM);
+                         if ( (ml == matchML)  /* last pos of last match */
+                           && (last_match_pos < pos) )
+                             last_match_pos = pos;
+                         opt[pos].mlen = ml;
+                         opt[pos].off = offset;
+                         opt[pos].litlen = ll;
+                         opt[pos].price = price;
+             }   }   }
+             /* complete following positions with literals */
+             {   int addLit;
+                 for (addLit = 1; addLit <= TRAILING_LITERALS; addLit ++) {
+                     opt[last_match_pos+addLit].mlen = 1; /* literal */
+                     opt[last_match_pos+addLit].off = 0;
+                     opt[last_match_pos+addLit].litlen = addLit;
+                     opt[last_match_pos+addLit].price = opt[last_match_pos].price + LZ4HC_literalsPrice(addLit);
+                     DEBUGLOG(7, "rPos:%3i => price:%3i (litlen=%i)", last_match_pos+addLit, opt[last_match_pos+addLit].price, addLit);
+             }   }
+         }  /* for (cur = 1; cur <= last_match_pos; cur++) */
+
+         assert(last_match_pos < LZ4_OPT_NUM + TRAILING_LITERALS);
+         best_mlen = opt[last_match_pos].mlen;
+         best_off = opt[last_match_pos].off;
+         cur = last_match_pos - best_mlen;
+
+encode: /* cur, last_match_pos, best_mlen, best_off must be set */
+         assert(cur < LZ4_OPT_NUM);
+         assert(last_match_pos >= 1);  /* == 1 when only one candidate */
+         DEBUGLOG(6, "reverse traversal, looking for shortest path (last_match_pos=%i)", last_match_pos);
+         {   int candidate_pos = cur;
+             int selected_matchLength = best_mlen;
+             int selected_offset = best_off;
+             while (1) {  /* from end to beginning */
+                 int const next_matchLength = opt[candidate_pos].mlen;  /* can be 1, means literal */
+                 int const next_offset = opt[candidate_pos].off;
+                 DEBUGLOG(7, "pos %i: sequence length %i", candidate_pos, selected_matchLength);
+                 opt[candidate_pos].mlen = selected_matchLength;
+                 opt[candidate_pos].off = selected_offset;
+                 selected_matchLength = next_matchLength;
+                 selected_offset = next_offset;
+                 if (next_matchLength > candidate_pos) break; /* last match elected, first match to encode */
+                 assert(next_matchLength > 0);  /* can be 1, means literal */
+                 candidate_pos -= next_matchLength;
+         }   }
+
+         /* encode all recorded sequences in order */
+         {   int rPos = 0;  /* relative position (to ip) */
+             while (rPos < last_match_pos) {
+                 int const ml = opt[rPos].mlen;
+                 int const offset = opt[rPos].off;
+                 if (ml == 1) { ip++; rPos++; continue; }  /* literal; note: can end up with several literals, in which case, skip them */
+                 rPos += ml;
+                 assert(ml >= MINMATCH);
+                 assert((offset >= 1) && (offset <= LZ4_DISTANCE_MAX));
+                 opSaved = op;
+                 if ( LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), ml, offset, limit, oend) ) {  /* updates ip, op and anchor */
+                     ovml = ml;
+                     ovoff = offset;
+                     goto _dest_overflow;
+         }   }   }
+     }  /* while (ip <= mflimit) */
+
+_last_literals:
+     /* Encode Last Literals */
+     {   size_t lastRunSize = (size_t)(iend - anchor);  /* literals */
+         size_t llAdd = (lastRunSize + 255 - RUN_MASK) / 255;
+         size_t const totalSize = 1 + llAdd + lastRunSize;
+         if (limit == fillOutput) oend += LASTLITERALS;  /* restore correct value */
+         if (limit && (op + totalSize > oend)) {
+             if (limit == limitedOutput) { /* Check output limit */
+                retval = 0;
+                goto _return_label;
+             }
+             /* adapt lastRunSize to fill 'dst' */
+             lastRunSize  = (size_t)(oend - op) - 1 /*token*/;
+             llAdd = (lastRunSize + 256 - RUN_MASK) / 256;
+             lastRunSize -= llAdd;
+         }
+         DEBUGLOG(6, "Final literal run : %i literals", (int)lastRunSize);
+         ip = anchor + lastRunSize; /* can be != iend if limit==fillOutput */
+
+         if (lastRunSize >= RUN_MASK) {
+             size_t accumulator = lastRunSize - RUN_MASK;
+             *op++ = (RUN_MASK << ML_BITS);
+             for(; accumulator >= 255 ; accumulator -= 255) *op++ = 255;
+             *op++ = (BYTE) accumulator;
+         } else {
+             *op++ = (BYTE)(lastRunSize << ML_BITS);
+         }
+         LZ4_memcpy(op, anchor, lastRunSize);
+         op += lastRunSize;
+     }
+
+     /* End */
+     *srcSizePtr = (int) (((const char*)ip) - source);
+     retval = (int) ((char*)op-dst);
+     goto _return_label;
+
+_dest_overflow:
+if (limit == fillOutput) {
+     /* Assumption : ip, anchor, ovml and ovref must be set correctly */
+     size_t const ll = (size_t)(ip - anchor);
+     size_t const ll_addbytes = (ll + 240) / 255;
+     size_t const ll_totalCost = 1 + ll_addbytes + ll;
+     BYTE* const maxLitPos = oend - 3; /* 2 for offset, 1 for token */
+     DEBUGLOG(6, "Last sequence overflowing (only %i bytes remaining)", (int)(oend-1-opSaved));
+     op = opSaved;  /* restore correct out pointer */
+     if (op + ll_totalCost <= maxLitPos) {
+         /* ll validated; now adjust match length */
+         size_t const bytesLeftForMl = (size_t)(maxLitPos - (op+ll_totalCost));
+         size_t const maxMlSize = MINMATCH + (ML_MASK-1) + (bytesLeftForMl * 255);
+         assert(maxMlSize < INT_MAX); assert(ovml >= 0);
+         if ((size_t)ovml > maxMlSize) ovml = (int)maxMlSize;
+         if ((oend + LASTLITERALS) - (op + ll_totalCost + 2) - 1 + ovml >= MFLIMIT) {
+             DEBUGLOG(6, "Space to end : %i + ml (%i)", (int)((oend + LASTLITERALS) - (op + ll_totalCost + 2) - 1), ovml);
+             DEBUGLOG(6, "Before : ip = %p, anchor = %p", ip, anchor);
+             LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), ovml, ovoff, notLimited, oend);
+             DEBUGLOG(6, "After : ip = %p, anchor = %p", ip, anchor);
+     }   }
+     goto _last_literals;
+}
+_return_label:
+#if defined(LZ4HC_HEAPMODE) && LZ4HC_HEAPMODE==1
+     if (opt) FREEMEM(opt);
+#endif
+     return retval;
+}
+
+
+/***************************************************
+*  Deprecated Functions
+***************************************************/
+
+/* These functions currently generate deprecation warnings */
+
+/* Wrappers for deprecated compression functions */
+int LZ4_compressHC(const char* src, char* dst, int srcSize) { return LZ4_compress_HC (src, dst, srcSize, LZ4_compressBound(srcSize), 0); }
+int LZ4_compressHC_limitedOutput(const char* src, char* dst, int srcSize, int maxDstSize) { return LZ4_compress_HC(src, dst, srcSize, maxDstSize, 0); }
+int LZ4_compressHC2(const char* src, char* dst, int srcSize, int cLevel) { return LZ4_compress_HC (src, dst, srcSize, LZ4_compressBound(srcSize), cLevel); }
+int LZ4_compressHC2_limitedOutput(const char* src, char* dst, int srcSize, int maxDstSize, int cLevel) { return LZ4_compress_HC(src, dst, srcSize, maxDstSize, cLevel); }
+int LZ4_compressHC_withStateHC (void* state, const char* src, char* dst, int srcSize) { return LZ4_compress_HC_extStateHC (state, src, dst, srcSize, LZ4_compressBound(srcSize), 0); }
+int LZ4_compressHC_limitedOutput_withStateHC (void* state, const char* src, char* dst, int srcSize, int maxDstSize) { return LZ4_compress_HC_extStateHC (state, src, dst, srcSize, maxDstSize, 0); }
+int LZ4_compressHC2_withStateHC (void* state, const char* src, char* dst, int srcSize, int cLevel) { return LZ4_compress_HC_extStateHC(state, src, dst, srcSize, LZ4_compressBound(srcSize), cLevel); }
+int LZ4_compressHC2_limitedOutput_withStateHC (void* state, const char* src, char* dst, int srcSize, int maxDstSize, int cLevel) { return LZ4_compress_HC_extStateHC(state, src, dst, srcSize, maxDstSize, cLevel); }
+int LZ4_compressHC_continue (LZ4_streamHC_t* ctx, const char* src, char* dst, int srcSize) { return LZ4_compress_HC_continue (ctx, src, dst, srcSize, LZ4_compressBound(srcSize)); }
+int LZ4_compressHC_limitedOutput_continue (LZ4_streamHC_t* ctx, const char* src, char* dst, int srcSize, int maxDstSize) { return LZ4_compress_HC_continue (ctx, src, dst, srcSize, maxDstSize); }
+
+
+/* Deprecated streaming functions */
+int LZ4_sizeofStreamStateHC(void) { return sizeof(LZ4_streamHC_t); }
+
+/* state is presumed correctly sized, aka >= sizeof(LZ4_streamHC_t)
+ * @return : 0 on success, !=0 if error */
+int LZ4_resetStreamStateHC(void* state, char* inputBuffer)
+{
+    LZ4_streamHC_t* const hc4 = LZ4_initStreamHC(state, sizeof(*hc4));
+    if (hc4 == NULL) return 1;   /* init failed */
+    LZ4HC_init_internal (&hc4->internal_donotuse, (const BYTE*)inputBuffer);
+    return 0;
+}
+
+#if !defined(LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION)
+void* LZ4_createHC (const char* inputBuffer)
+{
+    LZ4_streamHC_t* const hc4 = LZ4_createStreamHC();
+    if (hc4 == NULL) return NULL;   /* not enough memory */
+    LZ4HC_init_internal (&hc4->internal_donotuse, (const BYTE*)inputBuffer);
+    return hc4;
+}
+
+int LZ4_freeHC (void* LZ4HC_Data)
+{
+    if (!LZ4HC_Data) return 0;  /* support free on NULL */
+    FREEMEM(LZ4HC_Data);
+    return 0;
+}
+#endif
+
+int LZ4_compressHC2_continue (void* LZ4HC_Data, const char* src, char* dst, int srcSize, int cLevel)
+{
+    return LZ4HC_compress_generic (&((LZ4_streamHC_t*)LZ4HC_Data)->internal_donotuse, src, dst, &srcSize, 0, cLevel, notLimited);
+}
+
+int LZ4_compressHC2_limitedOutput_continue (void* LZ4HC_Data, const char* src, char* dst, int srcSize, int dstCapacity, int cLevel)
+{
+    return LZ4HC_compress_generic (&((LZ4_streamHC_t*)LZ4HC_Data)->internal_donotuse, src, dst, &srcSize, dstCapacity, cLevel, limitedOutput);
+}
+
+char* LZ4_slideInputBufferHC(void* LZ4HC_Data)
+{
+    LZ4HC_CCtx_internal* const s = &((LZ4_streamHC_t*)LZ4HC_Data)->internal_donotuse;
+    const BYTE* const bufferStart = s->prefixStart - s->dictLimit + s->lowLimit;
+    LZ4_resetStreamHC_fast((LZ4_streamHC_t*)LZ4HC_Data, s->compressionLevel);
+    /* ugly conversion trick, required to evade (const char*) -> (char*) cast-qual warning :( */
+    return (char*)(uptrval)bufferStart;
+}
diff --git a/lib/lz4/upstream/lz4hc.h b/lib/lz4/upstream/lz4hc.h
new file mode 100644
index 0000000000000000000000000000000000000000..992bc8cdd7703e7ff49b01137a00bf6f004683a6
--- /dev/null
+++ b/lib/lz4/upstream/lz4hc.h
@@ -0,0 +1,414 @@
+/*
+   LZ4 HC - High Compression Mode of LZ4
+   Header File
+   Copyright (C) 2011-2020, Yann Collet.
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - LZ4 source repository : https://github.com/lz4/lz4
+   - LZ4 public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+#ifndef LZ4_HC_H_19834876238432
+#define LZ4_HC_H_19834876238432
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* --- Dependency --- */
+/* note : lz4hc requires lz4.h/lz4.c for compilation */
+#include "lz4.h"   /* stddef, LZ4LIB_API, LZ4_DEPRECATED */
+
+
+/* --- Useful constants --- */
+#define LZ4HC_CLEVEL_MIN         2
+#define LZ4HC_CLEVEL_DEFAULT     9
+#define LZ4HC_CLEVEL_OPT_MIN    10
+#define LZ4HC_CLEVEL_MAX        12
+
+
+/*-************************************
+ *  Block Compression
+ **************************************/
+/*! LZ4_compress_HC() :
+ *  Compress data from `src` into `dst`, using the powerful but slower "HC" algorithm.
+ * `dst` must be already allocated.
+ *  Compression is guaranteed to succeed if `dstCapacity >= LZ4_compressBound(srcSize)` (see "lz4.h")
+ *  Max supported `srcSize` value is LZ4_MAX_INPUT_SIZE (see "lz4.h")
+ * `compressionLevel` : any value between 1 and LZ4HC_CLEVEL_MAX will work.
+ *                      Values > LZ4HC_CLEVEL_MAX behave the same as LZ4HC_CLEVEL_MAX.
+ * @return : the number of bytes written into 'dst'
+ *           or 0 if compression fails.
+ */
+LZ4LIB_API int LZ4_compress_HC (const char* src, char* dst, int srcSize, int dstCapacity, int compressionLevel);
+
+
+/* Note :
+ *   Decompression functions are provided within "lz4.h" (BSD license)
+ */
+
+
+/*! LZ4_compress_HC_extStateHC() :
+ *  Same as LZ4_compress_HC(), but using an externally allocated memory segment for `state`.
+ * `state` size is provided by LZ4_sizeofStateHC().
+ *  Memory segment must be aligned on 8-bytes boundaries (which a normal malloc() should do properly).
+ */
+LZ4LIB_API int LZ4_sizeofStateHC(void);
+LZ4LIB_API int LZ4_compress_HC_extStateHC(void* stateHC, const char* src, char* dst, int srcSize, int maxDstSize, int compressionLevel);
+
+
+/*! LZ4_compress_HC_destSize() : v1.9.0+
+ *  Will compress as much data as possible from `src`
+ *  to fit into `targetDstSize` budget.
+ *  Result is provided in 2 parts :
+ * @return : the number of bytes written into 'dst' (necessarily <= targetDstSize)
+ *           or 0 if compression fails.
+ * `srcSizePtr` : on success, *srcSizePtr is updated to indicate how much bytes were read from `src`
+ */
+LZ4LIB_API int LZ4_compress_HC_destSize(void* stateHC,
+                                  const char* src, char* dst,
+                                        int* srcSizePtr, int targetDstSize,
+                                        int compressionLevel);
+
+
+/*-************************************
+ *  Streaming Compression
+ *  Bufferless synchronous API
+ **************************************/
+ typedef union LZ4_streamHC_u LZ4_streamHC_t;   /* incomplete type (defined later) */
+
+/*! LZ4_createStreamHC() and LZ4_freeStreamHC() :
+ *  These functions create and release memory for LZ4 HC streaming state.
+ *  Newly created states are automatically initialized.
+ *  A same state can be used multiple times consecutively,
+ *  starting with LZ4_resetStreamHC_fast() to start a new stream of blocks.
+ */
+LZ4LIB_API LZ4_streamHC_t* LZ4_createStreamHC(void);
+LZ4LIB_API int             LZ4_freeStreamHC (LZ4_streamHC_t* streamHCPtr);
+
+/*
+  These functions compress data in successive blocks of any size,
+  using previous blocks as dictionary, to improve compression ratio.
+  One key assumption is that previous blocks (up to 64 KB) remain read-accessible while compressing next blocks.
+  There is an exception for ring buffers, which can be smaller than 64 KB.
+  Ring-buffer scenario is automatically detected and handled within LZ4_compress_HC_continue().
+
+  Before starting compression, state must be allocated and properly initialized.
+  LZ4_createStreamHC() does both, though compression level is set to LZ4HC_CLEVEL_DEFAULT.
+
+  Selecting the compression level can be done with LZ4_resetStreamHC_fast() (starts a new stream)
+  or LZ4_setCompressionLevel() (anytime, between blocks in the same stream) (experimental).
+  LZ4_resetStreamHC_fast() only works on states which have been properly initialized at least once,
+  which is automatically the case when state is created using LZ4_createStreamHC().
+
+  After reset, a first "fictional block" can be designated as initial dictionary,
+  using LZ4_loadDictHC() (Optional).
+  Note: In order for LZ4_loadDictHC() to create the correct data structure,
+  it is essential to set the compression level _before_ loading the dictionary.
+
+  Invoke LZ4_compress_HC_continue() to compress each successive block.
+  The number of blocks is unlimited.
+  Previous input blocks, including initial dictionary when present,
+  must remain accessible and unmodified during compression.
+
+  It's allowed to update compression level anytime between blocks,
+  using LZ4_setCompressionLevel() (experimental).
+
+ @dst buffer should be sized to handle worst case scenarios
+  (see LZ4_compressBound(), it ensures compression success).
+  In case of failure, the API does not guarantee recovery,
+  so the state _must_ be reset.
+  To ensure compression success
+  whenever @dst buffer size cannot be made >= LZ4_compressBound(),
+  consider using LZ4_compress_HC_continue_destSize().
+
+  Whenever previous input blocks can't be preserved unmodified in-place during compression of next blocks,
+  it's possible to copy the last blocks into a more stable memory space, using LZ4_saveDictHC().
+  Return value of LZ4_saveDictHC() is the size of dictionary effectively saved into 'safeBuffer' (<= 64 KB)
+
+  After completing a streaming compression,
+  it's possible to start a new stream of blocks, using the same LZ4_streamHC_t state,
+  just by resetting it, using LZ4_resetStreamHC_fast().
+*/
+
+LZ4LIB_API void LZ4_resetStreamHC_fast(LZ4_streamHC_t* streamHCPtr, int compressionLevel);   /* v1.9.0+ */
+LZ4LIB_API int  LZ4_loadDictHC (LZ4_streamHC_t* streamHCPtr, const char* dictionary, int dictSize);
+
+LZ4LIB_API int LZ4_compress_HC_continue (LZ4_streamHC_t* streamHCPtr,
+                                   const char* src, char* dst,
+                                         int srcSize, int maxDstSize);
+
+/*! LZ4_compress_HC_continue_destSize() : v1.9.0+
+ *  Similar to LZ4_compress_HC_continue(),
+ *  but will read as much data as possible from `src`
+ *  to fit into `targetDstSize` budget.
+ *  Result is provided into 2 parts :
+ * @return : the number of bytes written into 'dst' (necessarily <= targetDstSize)
+ *           or 0 if compression fails.
+ * `srcSizePtr` : on success, *srcSizePtr will be updated to indicate how much bytes were read from `src`.
+ *           Note that this function may not consume the entire input.
+ */
+LZ4LIB_API int LZ4_compress_HC_continue_destSize(LZ4_streamHC_t* LZ4_streamHCPtr,
+                                           const char* src, char* dst,
+                                                 int* srcSizePtr, int targetDstSize);
+
+LZ4LIB_API int LZ4_saveDictHC (LZ4_streamHC_t* streamHCPtr, char* safeBuffer, int maxDictSize);
+
+
+/*! LZ4_attach_HC_dictionary() : stable since v1.10.0
+ *  This API allows for the efficient re-use of a static dictionary many times.
+ *
+ *  Rather than re-loading the dictionary buffer into a working context before
+ *  each compression, or copying a pre-loaded dictionary's LZ4_streamHC_t into a
+ *  working LZ4_streamHC_t, this function introduces a no-copy setup mechanism,
+ *  in which the working stream references the dictionary stream in-place.
+ *
+ *  Several assumptions are made about the state of the dictionary stream.
+ *  Currently, only streams which have been prepared by LZ4_loadDictHC() should
+ *  be expected to work.
+ *
+ *  Alternatively, the provided dictionary stream pointer may be NULL, in which
+ *  case any existing dictionary stream is unset.
+ *
+ *  A dictionary should only be attached to a stream without any history (i.e.,
+ *  a stream that has just been reset).
+ *
+ *  The dictionary will remain attached to the working stream only for the
+ *  current stream session. Calls to LZ4_resetStreamHC(_fast) will remove the
+ *  dictionary context association from the working stream. The dictionary
+ *  stream (and source buffer) must remain in-place / accessible / unchanged
+ *  through the lifetime of the stream session.
+ */
+LZ4LIB_API void
+LZ4_attach_HC_dictionary(LZ4_streamHC_t* working_stream,
+                   const LZ4_streamHC_t* dictionary_stream);
+
+
+/*^**********************************************
+ * !!!!!!   STATIC LINKING ONLY   !!!!!!
+ ***********************************************/
+
+/*-******************************************************************
+ * PRIVATE DEFINITIONS :
+ * Do not use these definitions directly.
+ * They are merely exposed to allow static allocation of `LZ4_streamHC_t`.
+ * Declare an `LZ4_streamHC_t` directly, rather than any type below.
+ * Even then, only do so in the context of static linking, as definitions may change between versions.
+ ********************************************************************/
+
+#define LZ4HC_DICTIONARY_LOGSIZE 16
+#define LZ4HC_MAXD (1<<LZ4HC_DICTIONARY_LOGSIZE)
+#define LZ4HC_MAXD_MASK (LZ4HC_MAXD - 1)
+
+#define LZ4HC_HASH_LOG 15
+#define LZ4HC_HASHTABLESIZE (1 << LZ4HC_HASH_LOG)
+#define LZ4HC_HASH_MASK (LZ4HC_HASHTABLESIZE - 1)
+
+
+/* Never ever use these definitions directly !
+ * Declare or allocate an LZ4_streamHC_t instead.
+**/
+typedef struct LZ4HC_CCtx_internal LZ4HC_CCtx_internal;
+struct LZ4HC_CCtx_internal
+{
+    LZ4_u32 hashTable[LZ4HC_HASHTABLESIZE];
+    LZ4_u16 chainTable[LZ4HC_MAXD];
+    const LZ4_byte* end;     /* next block here to continue on current prefix */
+    const LZ4_byte* prefixStart;  /* Indexes relative to this position */
+    const LZ4_byte* dictStart; /* alternate reference for extDict */
+    LZ4_u32 dictLimit;       /* below that point, need extDict */
+    LZ4_u32 lowLimit;        /* below that point, no more history */
+    LZ4_u32 nextToUpdate;    /* index from which to continue dictionary update */
+    short   compressionLevel;
+    LZ4_i8  favorDecSpeed;   /* favor decompression speed if this flag set,
+                                otherwise, favor compression ratio */
+    LZ4_i8  dirty;           /* stream has to be fully reset if this flag is set */
+    const LZ4HC_CCtx_internal* dictCtx;
+};
+
+#define LZ4_STREAMHC_MINSIZE  262200  /* static size, for inter-version compatibility */
+union LZ4_streamHC_u {
+    char minStateSize[LZ4_STREAMHC_MINSIZE];
+    LZ4HC_CCtx_internal internal_donotuse;
+}; /* previously typedef'd to LZ4_streamHC_t */
+
+/* LZ4_streamHC_t :
+ * This structure allows static allocation of LZ4 HC streaming state.
+ * This can be used to allocate statically on stack, or as part of a larger structure.
+ *
+ * Such state **must** be initialized using LZ4_initStreamHC() before first use.
+ *
+ * Note that invoking LZ4_initStreamHC() is not required when
+ * the state was created using LZ4_createStreamHC() (which is recommended).
+ * Using the normal builder, a newly created state is automatically initialized.
+ *
+ * Static allocation shall only be used in combination with static linking.
+ */
+
+/* LZ4_initStreamHC() : v1.9.0+
+ * Required before first use of a statically allocated LZ4_streamHC_t.
+ * Before v1.9.0 : use LZ4_resetStreamHC() instead
+ */
+LZ4LIB_API LZ4_streamHC_t* LZ4_initStreamHC(void* buffer, size_t size);
+
+
+/*-************************************
+*  Deprecated Functions
+**************************************/
+/* see lz4.h LZ4_DISABLE_DEPRECATE_WARNINGS to turn off deprecation warnings */
+
+/* deprecated compression functions */
+LZ4_DEPRECATED("use LZ4_compress_HC() instead") LZ4LIB_API int LZ4_compressHC               (const char* source, char* dest, int inputSize);
+LZ4_DEPRECATED("use LZ4_compress_HC() instead") LZ4LIB_API int LZ4_compressHC_limitedOutput (const char* source, char* dest, int inputSize, int maxOutputSize);
+LZ4_DEPRECATED("use LZ4_compress_HC() instead") LZ4LIB_API int LZ4_compressHC2              (const char* source, char* dest, int inputSize, int compressionLevel);
+LZ4_DEPRECATED("use LZ4_compress_HC() instead") LZ4LIB_API int LZ4_compressHC2_limitedOutput(const char* source, char* dest, int inputSize, int maxOutputSize, int compressionLevel);
+LZ4_DEPRECATED("use LZ4_compress_HC_extStateHC() instead") LZ4LIB_API int LZ4_compressHC_withStateHC               (void* state, const char* source, char* dest, int inputSize);
+LZ4_DEPRECATED("use LZ4_compress_HC_extStateHC() instead") LZ4LIB_API int LZ4_compressHC_limitedOutput_withStateHC (void* state, const char* source, char* dest, int inputSize, int maxOutputSize);
+LZ4_DEPRECATED("use LZ4_compress_HC_extStateHC() instead") LZ4LIB_API int LZ4_compressHC2_withStateHC              (void* state, const char* source, char* dest, int inputSize, int compressionLevel);
+LZ4_DEPRECATED("use LZ4_compress_HC_extStateHC() instead") LZ4LIB_API int LZ4_compressHC2_limitedOutput_withStateHC(void* state, const char* source, char* dest, int inputSize, int maxOutputSize, int compressionLevel);
+LZ4_DEPRECATED("use LZ4_compress_HC_continue() instead") LZ4LIB_API int LZ4_compressHC_continue               (LZ4_streamHC_t* LZ4_streamHCPtr, const char* source, char* dest, int inputSize);
+LZ4_DEPRECATED("use LZ4_compress_HC_continue() instead") LZ4LIB_API int LZ4_compressHC_limitedOutput_continue (LZ4_streamHC_t* LZ4_streamHCPtr, const char* source, char* dest, int inputSize, int maxOutputSize);
+
+/* Obsolete streaming functions; degraded functionality; do not use!
+ *
+ * In order to perform streaming compression, these functions depended on data
+ * that is no longer tracked in the state. They have been preserved as well as
+ * possible: using them will still produce a correct output. However, use of
+ * LZ4_slideInputBufferHC() will truncate the history of the stream, rather
+ * than preserve a window-sized chunk of history.
+ */
+#if !defined(LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION)
+LZ4_DEPRECATED("use LZ4_createStreamHC() instead") LZ4LIB_API void* LZ4_createHC (const char* inputBuffer);
+LZ4_DEPRECATED("use LZ4_freeStreamHC() instead") LZ4LIB_API   int   LZ4_freeHC (void* LZ4HC_Data);
+#endif
+LZ4_DEPRECATED("use LZ4_saveDictHC() instead") LZ4LIB_API     char* LZ4_slideInputBufferHC (void* LZ4HC_Data);
+LZ4_DEPRECATED("use LZ4_compress_HC_continue() instead") LZ4LIB_API int LZ4_compressHC2_continue               (void* LZ4HC_Data, const char* source, char* dest, int inputSize, int compressionLevel);
+LZ4_DEPRECATED("use LZ4_compress_HC_continue() instead") LZ4LIB_API int LZ4_compressHC2_limitedOutput_continue (void* LZ4HC_Data, const char* source, char* dest, int inputSize, int maxOutputSize, int compressionLevel);
+LZ4_DEPRECATED("use LZ4_createStreamHC() instead") LZ4LIB_API int   LZ4_sizeofStreamStateHC(void);
+LZ4_DEPRECATED("use LZ4_initStreamHC() instead") LZ4LIB_API  int   LZ4_resetStreamStateHC(void* state, char* inputBuffer);
+
+
+/* LZ4_resetStreamHC() is now replaced by LZ4_initStreamHC().
+ * The intention is to emphasize the difference with LZ4_resetStreamHC_fast(),
+ * which is now the recommended function to start a new stream of blocks,
+ * but cannot be used to initialize a memory segment containing arbitrary garbage data.
+ *
+ * It is recommended to switch to LZ4_initStreamHC().
+ * LZ4_resetStreamHC() will generate deprecation warnings in a future version.
+ */
+LZ4LIB_API void LZ4_resetStreamHC (LZ4_streamHC_t* streamHCPtr, int compressionLevel);
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* LZ4_HC_H_19834876238432 */
+
+
+/*-**************************************************
+ * !!!!!     STATIC LINKING ONLY     !!!!!
+ * Following definitions are considered experimental.
+ * They should not be linked from DLL,
+ * as there is no guarantee of API stability yet.
+ * Prototypes will be promoted to "stable" status
+ * after successful usage in real-life scenarios.
+ ***************************************************/
+#ifdef LZ4_HC_STATIC_LINKING_ONLY   /* protection macro */
+#ifndef LZ4_HC_SLO_098092834
+#define LZ4_HC_SLO_098092834
+
+#define LZ4_STATIC_LINKING_ONLY   /* LZ4LIB_STATIC_API */
+#include "lz4.h"
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/*! LZ4_setCompressionLevel() : v1.8.0+ (experimental)
+ *  It's possible to change compression level
+ *  between successive invocations of LZ4_compress_HC_continue*()
+ *  for dynamic adaptation.
+ */
+LZ4LIB_STATIC_API void LZ4_setCompressionLevel(
+    LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel);
+
+/*! LZ4_favorDecompressionSpeed() : v1.8.2+ (experimental)
+ *  Opt. Parser will favor decompression speed over compression ratio.
+ *  Only applicable to levels >= LZ4HC_CLEVEL_OPT_MIN.
+ */
+LZ4LIB_STATIC_API void LZ4_favorDecompressionSpeed(
+    LZ4_streamHC_t* LZ4_streamHCPtr, int favor);
+
+/*! LZ4_resetStreamHC_fast() : v1.9.0+
+ *  When an LZ4_streamHC_t is known to be in a internally coherent state,
+ *  it can often be prepared for a new compression with almost no work, only
+ *  sometimes falling back to the full, expensive reset that is always required
+ *  when the stream is in an indeterminate state (i.e., the reset performed by
+ *  LZ4_resetStreamHC()).
+ *
+ *  LZ4_streamHCs are guaranteed to be in a valid state when:
+ *  - returned from LZ4_createStreamHC()
+ *  - reset by LZ4_resetStreamHC()
+ *  - memset(stream, 0, sizeof(LZ4_streamHC_t))
+ *  - the stream was in a valid state and was reset by LZ4_resetStreamHC_fast()
+ *  - the stream was in a valid state and was then used in any compression call
+ *    that returned success
+ *  - the stream was in an indeterminate state and was used in a compression
+ *    call that fully reset the state (LZ4_compress_HC_extStateHC()) and that
+ *    returned success
+ *
+ *  Note:
+ *  A stream that was last used in a compression call that returned an error
+ *  may be passed to this function. However, it will be fully reset, which will
+ *  clear any existing history and settings from the context.
+ */
+LZ4LIB_STATIC_API void LZ4_resetStreamHC_fast(
+    LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel);
+
+/*! LZ4_compress_HC_extStateHC_fastReset() :
+ *  A variant of LZ4_compress_HC_extStateHC().
+ *
+ *  Using this variant avoids an expensive initialization step. It is only safe
+ *  to call if the state buffer is known to be correctly initialized already
+ *  (see above comment on LZ4_resetStreamHC_fast() for a definition of
+ *  "correctly initialized"). From a high level, the difference is that this
+ *  function initializes the provided state with a call to
+ *  LZ4_resetStreamHC_fast() while LZ4_compress_HC_extStateHC() starts with a
+ *  call to LZ4_resetStreamHC().
+ */
+LZ4LIB_STATIC_API int LZ4_compress_HC_extStateHC_fastReset (
+    void* state,
+    const char* src, char* dst,
+    int srcSize, int dstCapacity,
+    int compressionLevel);
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif   /* LZ4_HC_SLO_098092834 */
+#endif   /* LZ4_HC_STATIC_LINKING_ONLY */

-- 
2.34.1



_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

Re: [PATCH v2 04/15] lib/lz4: import upstream LZ4 sources verbatim

[sashiko-bot]

> Add the four upstream files under lib/lz4/upstream/ without modification.
> Nothing builds them yet; a later patch adds the build environment.
> 
> Copied without modification from the v1.10.0 release tag of
> https://github.com/lz4/lz4, tagged 2024-07-21.
> 
> Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>

Sashiko has reviewed this patch and found no issues.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261106-lz4-vendor-upstream-v2-0-ee04c3f0d7dc@samsung.com?part=4

[PATCH v2 05/15] lib/lz4: backport upstream's -Wmissing-prototypes fix

[Michal Wilczynski]

v1.10.0 defines LZ4_loadDict_internal() in lz4.c and LZ4HC_searchExtDict()
in lz4hc.c with external linkage and no prototype, which the kernel's
-Wmissing-prototypes rejects. The first would also prevent linking once
lz4_compress.c and lz4_decompress.c both include lz4.c, later in this
series, since each would define it.

Neither can be forward declared static in the lz4_deps.h a later patch
adds, as the other bare helpers are, because their signatures use types
declared inside the .c files.

Take the two hunks of upstream commit 5ef1f16929b5
("fix -Wmissing-prototypes warnings") that apply to the files we vendor.
This goes away at the next re-sync.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 lib/lz4/upstream/lz4.c   | 2 +-
 lib/lz4/upstream/lz4hc.c | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/lib/lz4/upstream/lz4.c b/lib/lz4/upstream/lz4.c
index a2f7abee19fb9a5c768f2a6c266acf5b571f0855..0d474be0236a698d297ab9a430f5082485eb3709 100644
--- a/lib/lz4/upstream/lz4.c
+++ b/lib/lz4/upstream/lz4.c
@@ -1584,7 +1584,7 @@ int LZ4_freeStream (LZ4_stream_t* LZ4_stream)
 
 typedef enum { _ld_fast, _ld_slow } LoadDict_mode_e;
 #define HASH_UNIT sizeof(reg_t)
-int LZ4_loadDict_internal(LZ4_stream_t* LZ4_dict,
+static int LZ4_loadDict_internal(LZ4_stream_t* LZ4_dict,
                     const char* dictionary, int dictSize,
                     LoadDict_mode_e _ld)
 {
diff --git a/lib/lz4/upstream/lz4hc.c b/lib/lz4/upstream/lz4hc.c
index 4d8c36a6978fcae09e4c4a936572b4b271513466..32b29f382268d961d65383b24c7225494d9044ad 100644
--- a/lib/lz4/upstream/lz4hc.c
+++ b/lib/lz4/upstream/lz4hc.c
@@ -360,7 +360,7 @@ typedef struct {
     int back;  /* negative value */
 } LZ4HC_match_t;
 
-LZ4HC_match_t LZ4HC_searchExtDict(const BYTE* ip, U32 ipIndex,
+static LZ4HC_match_t LZ4HC_searchExtDict(const BYTE* ip, U32 ipIndex,
         const BYTE* const iLowLimit, const BYTE* const iHighLimit,
         const LZ4HC_CCtx_internal* dictCtx, U32 gDictEndIndex,
         int currentBestML, int nbAttempts)

-- 
2.34.1

[f2fs-dev] [PATCH v2 05/15] lib/lz4: backport upstream's -Wmissing-prototypes fix

[Michal Wilczynski]

v1.10.0 defines LZ4_loadDict_internal() in lz4.c and LZ4HC_searchExtDict()
in lz4hc.c with external linkage and no prototype, which the kernel's
-Wmissing-prototypes rejects. The first would also prevent linking once
lz4_compress.c and lz4_decompress.c both include lz4.c, later in this
series, since each would define it.

Neither can be forward declared static in the lz4_deps.h a later patch
adds, as the other bare helpers are, because their signatures use types
declared inside the .c files.

Take the two hunks of upstream commit 5ef1f16929b5
("fix -Wmissing-prototypes warnings") that apply to the files we vendor.
This goes away at the next re-sync.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 lib/lz4/upstream/lz4.c   | 2 +-
 lib/lz4/upstream/lz4hc.c | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/lib/lz4/upstream/lz4.c b/lib/lz4/upstream/lz4.c
index a2f7abee19fb9a5c768f2a6c266acf5b571f0855..0d474be0236a698d297ab9a430f5082485eb3709 100644
--- a/lib/lz4/upstream/lz4.c
+++ b/lib/lz4/upstream/lz4.c
@@ -1584,7 +1584,7 @@ int LZ4_freeStream (LZ4_stream_t* LZ4_stream)
 
 typedef enum { _ld_fast, _ld_slow } LoadDict_mode_e;
 #define HASH_UNIT sizeof(reg_t)
-int LZ4_loadDict_internal(LZ4_stream_t* LZ4_dict,
+static int LZ4_loadDict_internal(LZ4_stream_t* LZ4_dict,
                     const char* dictionary, int dictSize,
                     LoadDict_mode_e _ld)
 {
diff --git a/lib/lz4/upstream/lz4hc.c b/lib/lz4/upstream/lz4hc.c
index 4d8c36a6978fcae09e4c4a936572b4b271513466..32b29f382268d961d65383b24c7225494d9044ad 100644
--- a/lib/lz4/upstream/lz4hc.c
+++ b/lib/lz4/upstream/lz4hc.c
@@ -360,7 +360,7 @@ typedef struct {
     int back;  /* negative value */
 } LZ4HC_match_t;
 
-LZ4HC_match_t LZ4HC_searchExtDict(const BYTE* ip, U32 ipIndex,
+static LZ4HC_match_t LZ4HC_searchExtDict(const BYTE* ip, U32 ipIndex,
         const BYTE* const iLowLimit, const BYTE* const iHighLimit,
         const LZ4HC_CCtx_internal* dictCtx, U32 gDictEndIndex,
         int currentBestML, int nbAttempts)

-- 
2.34.1



_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

Re: [PATCH v2 05/15] lib/lz4: backport upstream's -Wmissing-prototypes fix

[sashiko-bot]

> v1.10.0 defines LZ4_loadDict_internal() in lz4.c and LZ4HC_searchExtDict()
> in lz4hc.c with external linkage and no prototype, which the kernel's
> -Wmissing-prototypes rejects. The first would also prevent linking once
> lz4_compress.c and lz4_decompress.c both include lz4.c, later in this
> series, since each would define it.
> 
> Neither can be forward declared static in the lz4_deps.h a later patch
> adds, as the other bare helpers are, because their signatures use types
> declared inside the .c files.
> 
> Take the two hunks of upstream commit 5ef1f16929b5
> ("fix -Wmissing-prototypes warnings") that apply to the files we vendor.
> This goes away at the next re-sync.
> 
> Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>

Sashiko has reviewed this patch and found no issues.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261106-lz4-vendor-upstream-v2-0-ee04c3f0d7dc@samsung.com?part=5

[PATCH v2 06/15] lib/lz4: backport upstream's aarch64 fast decode loop enablement

[Michal Wilczynski]

v1.10.0 leaves LZ4_FAST_DEC_LOOP off for aarch64 built with clang, after
a slowdown reported on some Android phones in 2019 (upstream PR #707).
An arm64 kernel built with clang therefore decodes with the slower loop.

Upstream has since enabled it for all aarch64, in three unreleased
commits:

commit 0aa33884faf5 ("[LZ4] Limit disabling LZ4_FAST_DEC_LOOP to aarch64+clang+android")
commit 51f8af506b4a ("Fix missing endif")
commit ed86bc1ed802 ("AArch64: Enable LZ4_FAST_DEC_LOOP for Android")

Arm measured +3.5% to +15.6% on Cortex-A510 to X4 with clang. Take the
lz4.c hunks of all three. This goes away at the next re-sync.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 lib/lz4/upstream/lz4.c | 7 +------
 1 file changed, 1 insertion(+), 6 deletions(-)

diff --git a/lib/lz4/upstream/lz4.c b/lib/lz4/upstream/lz4.c
index 0d474be0236a698d297ab9a430f5082485eb3709..ee69d57aae112f4057c050bbdab4cc0f2d30f9d9 100644
--- a/lib/lz4/upstream/lz4.c
+++ b/lib/lz4/upstream/lz4.c
@@ -478,12 +478,7 @@ static const int      dec64table[8] = {0, 0, 0, -1, -4,  1, 2, 3};
 #ifndef LZ4_FAST_DEC_LOOP
 #  if defined __i386__ || defined _M_IX86 || defined __x86_64__ || defined _M_X64
 #    define LZ4_FAST_DEC_LOOP 1
-#  elif defined(__aarch64__) && defined(__APPLE__)
-#    define LZ4_FAST_DEC_LOOP 1
-#  elif defined(__aarch64__) && !defined(__clang__)
-     /* On non-Apple aarch64, we disable this optimization for clang because
-      * on certain mobile chipsets, performance is reduced with clang. For
-      * more information refer to https://github.com/lz4/lz4/pull/707 */
+#  elif defined(__aarch64__)
 #    define LZ4_FAST_DEC_LOOP 1
 #  else
 #    define LZ4_FAST_DEC_LOOP 0

-- 
2.34.1

[f2fs-dev] [PATCH v2 06/15] lib/lz4: backport upstream's aarch64 fast decode loop enablement

[Michal Wilczynski]

v1.10.0 leaves LZ4_FAST_DEC_LOOP off for aarch64 built with clang, after
a slowdown reported on some Android phones in 2019 (upstream PR #707).
An arm64 kernel built with clang therefore decodes with the slower loop.

Upstream has since enabled it for all aarch64, in three unreleased
commits:

commit 0aa33884faf5 ("[LZ4] Limit disabling LZ4_FAST_DEC_LOOP to aarch64+clang+android")
commit 51f8af506b4a ("Fix missing endif")
commit ed86bc1ed802 ("AArch64: Enable LZ4_FAST_DEC_LOOP for Android")

Arm measured +3.5% to +15.6% on Cortex-A510 to X4 with clang. Take the
lz4.c hunks of all three. This goes away at the next re-sync.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 lib/lz4/upstream/lz4.c | 7 +------
 1 file changed, 1 insertion(+), 6 deletions(-)

diff --git a/lib/lz4/upstream/lz4.c b/lib/lz4/upstream/lz4.c
index 0d474be0236a698d297ab9a430f5082485eb3709..ee69d57aae112f4057c050bbdab4cc0f2d30f9d9 100644
--- a/lib/lz4/upstream/lz4.c
+++ b/lib/lz4/upstream/lz4.c
@@ -478,12 +478,7 @@ static const int      dec64table[8] = {0, 0, 0, -1, -4,  1, 2, 3};
 #ifndef LZ4_FAST_DEC_LOOP
 #  if defined __i386__ || defined _M_IX86 || defined __x86_64__ || defined _M_X64
 #    define LZ4_FAST_DEC_LOOP 1
-#  elif defined(__aarch64__) && defined(__APPLE__)
-#    define LZ4_FAST_DEC_LOOP 1
-#  elif defined(__aarch64__) && !defined(__clang__)
-     /* On non-Apple aarch64, we disable this optimization for clang because
-      * on certain mobile chipsets, performance is reduced with clang. For
-      * more information refer to https://github.com/lz4/lz4/pull/707 */
+#  elif defined(__aarch64__)
 #    define LZ4_FAST_DEC_LOOP 1
 #  else
 #    define LZ4_FAST_DEC_LOOP 0

-- 
2.34.1



_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

Re: [PATCH v2 06/15] lib/lz4: backport upstream's aarch64 fast decode loop enablement

[sashiko-bot]

> v1.10.0 leaves LZ4_FAST_DEC_LOOP off for aarch64 built with clang, after
> a slowdown reported on some Android phones in 2019 (upstream PR #707).
> An arm64 kernel built with clang therefore decodes with the slower loop.
> 
> Upstream has since enabled it for all aarch64, in three unreleased
> commits:
> 
> commit 0aa33884faf5 ("[LZ4] Limit disabling LZ4_FAST_DEC_LOOP to aarch64+clang+android")
> commit 51f8af506b4a ("Fix missing endif")
> commit ed86bc1ed802 ("AArch64: Enable LZ4_FAST_DEC_LOOP for Android")
> 
> Arm measured +3.5% to +15.6% on Cortex-A510 to X4 with clang. Take the
> lz4.c hunks of all three. This goes away at the next re-sync.
> 
> Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>

Sashiko has reviewed this patch and found no issues.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261106-lz4-vendor-upstream-v2-0-ee04c3f0d7dc@samsung.com?part=6

[f2fs-dev] [PATCH v2 07/15] lib/lz4: backport upstream's match length limit fix

[Michal Wilczynski]

v1.10.0 lets a match length's extension bytes run to four bytes before
the end of the block, which leaves room for only a token and three
literals where the format requires five. The fork rejected such blocks,
v1.10.0 decodes them: 1f 41 01 00 ff 00 30 7a 7a 7a decodes to 278
bytes instead of failing.

Upstream fixed this in 587ed9557708 ("add comments explaining
read_variable_length() ilimit bounds, fix match length limit"), on top
of a read_variable_length() rework v1.10.0 does not have. Both versions
accept the same final ip for a given limit, so change the limit at the
two match length call sites to iend - (1 + LASTLITERALS), as upstream
does. This is one byte stricter than the fork; no compressor produces
the blocks it newly rejects.

This goes away at the next re-sync.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 lib/lz4/upstream/lz4.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/lib/lz4/upstream/lz4.c b/lib/lz4/upstream/lz4.c
index ee69d57aae112f4057c050bbdab4cc0f2d30f9d9..a45f3828141eec9baea59a5e97a3ff37d6474d6e 100644
--- a/lib/lz4/upstream/lz4.c
+++ b/lib/lz4/upstream/lz4.c
@@ -2120,7 +2120,7 @@ LZ4_decompress_generic(
             DEBUGLOG(7, "  match length token = %u (len==%u)", (unsigned)length, (unsigned)length+MINMATCH);
 
             if (length == ML_MASK) {
-                size_t const addl = read_variable_length(&ip, iend - LASTLITERALS + 1, 0);
+                size_t const addl = read_variable_length(&ip, iend - (1 + LASTLITERALS), 0);
                 if (addl == rvl_error) {
                     DEBUGLOG(5, "error reading long match length");
                     goto _output_error;
@@ -2338,7 +2338,7 @@ LZ4_decompress_generic(
 
     _copy_match:
             if (length == ML_MASK) {
-                size_t const addl = read_variable_length(&ip, iend - LASTLITERALS + 1, 0);
+                size_t const addl = read_variable_length(&ip, iend - (1 + LASTLITERALS), 0);
                 if (addl == rvl_error) { goto _output_error; }
                 length += addl;
                 if (unlikely((uptrval)(op)+length<(uptrval)op)) goto _output_error;   /* overflow detection */

-- 
2.34.1



_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

[PATCH v2 07/15] lib/lz4: backport upstream's match length limit fix

[Michal Wilczynski]

v1.10.0 lets a match length's extension bytes run to four bytes before
the end of the block, which leaves room for only a token and three
literals where the format requires five. The fork rejected such blocks,
v1.10.0 decodes them: 1f 41 01 00 ff 00 30 7a 7a 7a decodes to 278
bytes instead of failing.

Upstream fixed this in 587ed9557708 ("add comments explaining
read_variable_length() ilimit bounds, fix match length limit"), on top
of a read_variable_length() rework v1.10.0 does not have. Both versions
accept the same final ip for a given limit, so change the limit at the
two match length call sites to iend - (1 + LASTLITERALS), as upstream
does. This is one byte stricter than the fork; no compressor produces
the blocks it newly rejects.

This goes away at the next re-sync.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 lib/lz4/upstream/lz4.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/lib/lz4/upstream/lz4.c b/lib/lz4/upstream/lz4.c
index ee69d57aae112f4057c050bbdab4cc0f2d30f9d9..a45f3828141eec9baea59a5e97a3ff37d6474d6e 100644
--- a/lib/lz4/upstream/lz4.c
+++ b/lib/lz4/upstream/lz4.c
@@ -2120,7 +2120,7 @@ LZ4_decompress_generic(
             DEBUGLOG(7, "  match length token = %u (len==%u)", (unsigned)length, (unsigned)length+MINMATCH);
 
             if (length == ML_MASK) {
-                size_t const addl = read_variable_length(&ip, iend - LASTLITERALS + 1, 0);
+                size_t const addl = read_variable_length(&ip, iend - (1 + LASTLITERALS), 0);
                 if (addl == rvl_error) {
                     DEBUGLOG(5, "error reading long match length");
                     goto _output_error;
@@ -2338,7 +2338,7 @@ LZ4_decompress_generic(
 
     _copy_match:
             if (length == ML_MASK) {
-                size_t const addl = read_variable_length(&ip, iend - LASTLITERALS + 1, 0);
+                size_t const addl = read_variable_length(&ip, iend - (1 + LASTLITERALS), 0);
                 if (addl == rvl_error) { goto _output_error; }
                 length += addl;
                 if (unlikely((uptrval)(op)+length<(uptrval)op)) goto _output_error;   /* overflow detection */

-- 
2.34.1

Re: [PATCH v2 07/15] lib/lz4: backport upstream's match length limit fix

[sashiko-bot]

> v1.10.0 lets a match length's extension bytes run to four bytes before
> the end of the block, which leaves room for only a token and three
> literals where the format requires five. The fork rejected such blocks,
> v1.10.0 decodes them: 1f 41 01 00 ff 00 30 7a 7a 7a decodes to 278
> bytes instead of failing.
> 
> Upstream fixed this in 587ed9557708 ("add comments explaining
> read_variable_length() ilimit bounds, fix match length limit"), on top
> of a read_variable_length() rework v1.10.0 does not have. Both versions
> accept the same final ip for a given limit, so change the limit at the
> two match length call sites to iend - (1 + LASTLITERALS), as upstream
> does. This is one byte stricter than the fork; no compressor produces
> the blocks it newly rejects.
> 
> This goes away at the next re-sync.
> 
> Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>

Sashiko has reviewed this patch and found no issues.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261106-lz4-vendor-upstream-v2-0-ee04c3f0d7dc@samsung.com?part=7

[f2fs-dev] [PATCH v2 08/15] lib/lz4: add the build environment for the vendored sources

[Michal Wilczynski]

lz4_deps.h makes the vendored upstream sources build in the kernel. It
gives every upstream entry point internal linkage, so an object carries
only the codec it forwards, selects heap mode with failing allocators so
LZ4HC's ~64K optimal-parser buffer stays off the stack, and renames the
entry points the kernel re-exports so the kernel's own definitions can
use the real names. LZ4LIB_VISIBILITY carries __maybe_unused for the
functions an object does not use.

freestanding/ forwards the four ISO C headers upstream includes to their
<linux/> equivalents. lz4_kernel_api.h undoes the renaming; it carries
its own copy of the exported prototypes until a later patch can include
<linux/lz4.h> instead.

Two settings upstream leaves to the build:

 - Unaligned access goes through memcpy() where the architecture has
   efficient unaligned access, and through packed structs, as
   get_unaligned() does, elsewhere. Left alone, upstream uses plain
   misaligned loads on ARMv6.

 - On ppc64le, lz4.c marks functions __attribute__((optimize("O2"))) for
   GCC, which drops command line options such as -fno-strict-aliasing
   for them; see commit 080b6f407635 ("bpf: Don't rely on GCC
   __attribute__((optimize)) to disable GCSE"). lz4_deps.h defines
   optimize() to nothing for upstream's sources.

Nothing includes either header yet.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 lib/lz4/freestanding/limits.h |  14 +++++
 lib/lz4/freestanding/stddef.h |  15 ++++++
 lib/lz4/freestanding/stdint.h |  14 +++++
 lib/lz4/freestanding/string.h |  14 +++++
 lib/lz4/lz4_deps.h            | 118 ++++++++++++++++++++++++++++++++++++++++++
 lib/lz4/lz4_kernel_api.h      |  90 ++++++++++++++++++++++++++++++++
 6 files changed, 265 insertions(+)

diff --git a/lib/lz4/freestanding/limits.h b/lib/lz4/freestanding/limits.h
new file mode 100644
index 0000000000000000000000000000000000000000..0d1cafbc77eb5d2f564b51feaa911748446abf60
--- /dev/null
+++ b/lib/lz4/freestanding/limits.h
@@ -0,0 +1,14 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (c) 2026 Samsung Electronics Co., Ltd.
+ * Author: Michal Wilczynski <m.wilczynski@samsung.com>
+ *
+ * Freestanding <limits.h> for the vendored upstream LZ4 sources; lz4.c needs
+ * UINT_MAX and lz4hc.c needs INT_MAX.
+ */
+#ifndef __LZ4_FREESTANDING_LIMITS_H__
+#define __LZ4_FREESTANDING_LIMITS_H__
+
+#include <linux/limits.h>
+
+#endif
diff --git a/lib/lz4/freestanding/stddef.h b/lib/lz4/freestanding/stddef.h
new file mode 100644
index 0000000000000000000000000000000000000000..aaafd2413849210fa4ec4fa03ba0e9453a9fe026
--- /dev/null
+++ b/lib/lz4/freestanding/stddef.h
@@ -0,0 +1,15 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (c) 2026 Samsung Electronics Co., Ltd.
+ * Author: Michal Wilczynski <m.wilczynski@samsung.com>
+ *
+ * Freestanding <stddef.h> for the vendored upstream LZ4 sources; -nostdinc
+ * drops the compiler's copy, and lz4.h needs size_t and NULL.
+ */
+#ifndef __LZ4_FREESTANDING_STDDEF_H__
+#define __LZ4_FREESTANDING_STDDEF_H__
+
+#include <linux/stddef.h>	/* NULL, offsetof */
+#include <linux/types.h>	/* size_t, ptrdiff_t */
+
+#endif
diff --git a/lib/lz4/freestanding/stdint.h b/lib/lz4/freestanding/stdint.h
new file mode 100644
index 0000000000000000000000000000000000000000..923db05b19f23fb4051d16e5df2bf134cd5ac11e
--- /dev/null
+++ b/lib/lz4/freestanding/stdint.h
@@ -0,0 +1,14 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (c) 2026 Samsung Electronics Co., Ltd.
+ * Author: Michal Wilczynski <m.wilczynski@samsung.com>
+ *
+ * Freestanding <stdint.h> for the vendored upstream LZ4 sources; forwards to
+ * <linux/types.h>.
+ */
+#ifndef __LZ4_FREESTANDING_STDINT_H__
+#define __LZ4_FREESTANDING_STDINT_H__
+
+#include <linux/types.h>
+
+#endif
diff --git a/lib/lz4/freestanding/string.h b/lib/lz4/freestanding/string.h
new file mode 100644
index 0000000000000000000000000000000000000000..8f3ea3b63013608a780a52fe9c33bf5df557584e
--- /dev/null
+++ b/lib/lz4/freestanding/string.h
@@ -0,0 +1,14 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (c) 2026 Samsung Electronics Co., Ltd.
+ * Author: Michal Wilczynski <m.wilczynski@samsung.com>
+ *
+ * Freestanding <string.h> for the vendored upstream LZ4 sources; lz4.c
+ * includes it for memcpy()/memset(), which must resolve under -nostdinc.
+ */
+#ifndef __LZ4_FREESTANDING_STRING_H__
+#define __LZ4_FREESTANDING_STRING_H__
+
+#include <linux/string.h>
+
+#endif
diff --git a/lib/lz4/lz4_deps.h b/lib/lz4/lz4_deps.h
new file mode 100644
index 0000000000000000000000000000000000000000..23d69c75867f71224cb533e9341bd38439b9dee8
--- /dev/null
+++ b/lib/lz4/lz4_deps.h
@@ -0,0 +1,118 @@
+/* SPDX-License-Identifier: GPL-2.0-only OR BSD-2-Clause */
+/*
+ * Copyright (c) 2026 Samsung Electronics Co., Ltd.
+ * Author: Michal Wilczynski <m.wilczynski@samsung.com>
+ *
+ * Build environment for the vendored upstream LZ4 sources.
+ *
+ * upstream/ is LZ4 v1.10.0 plus three backports: 5ef1f16929b5, the aarch64
+ * fast loop (0aa33884faf5, 51f8af506b4a, ed86bc1ed802) and the match length
+ * limit of 587ed9557708, adapted.  Replace it with a later release rather
+ * than patch it further.  The three entry point files include this header
+ * first.
+ */
+
+#ifndef __LZ4_DEPS_H__
+#define __LZ4_DEPS_H__
+
+#include <linux/build_bug.h>	/* static_assert */
+#include <linux/compiler.h>	/* __maybe_unused */
+#include <linux/string.h>
+#include <linux/types.h>
+
+/* lz4.c uses "current" as a local; <asm/current.h> breaks the build. */
+#undef current
+
+/* Static so an object carries only the codec it forwards.  Each object
+ * forwards only some of them, so mark them __maybe_unused rather than turn
+ * off -Wunused-function, which the pre-boot builds would not see anyway.
+ */
+#define LZ4LIB_VISIBILITY static __maybe_unused
+
+/* Route upstream's "static linking only" entry points through
+ * LZ4LIB_VISIBILITY too.
+ */
+#define LZ4_PUBLISH_STATIC_FUNCTIONS
+
+/* Heap mode only: the stack mode puts LZ4HC's ~64K opt[] in a frame.
+ * Allocation goes through the failing stubs below; upstream handles
+ * NULL by returning 0.
+ */
+#define LZ4_USER_MEMORY_FUNCTIONS
+#define LZ4_HEAPMODE 1
+#define LZ4HC_HEAPMODE 1
+
+/* Failing stubs; nothing in the kernel reaches these. */
+static void *LZ4_malloc(size_t s) { return NULL; }
+static void *LZ4_calloc(size_t n, size_t s) { return NULL; }
+static void LZ4_free(void *p) { }
+
+/* Left alone, upstream reads unaligned words through packed structs, and
+ * on ARMv6 through plain misaligned dereferences.  Use memcpy(), upstream's
+ * documented default, where unaligned access is cheap, and packed structs,
+ * as get_unaligned() does, where it is not: there GCC builds a memcpy()'d
+ * word through a stack slot.
+ */
+#ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
+#define LZ4_FORCE_MEMORY_ACCESS 0
+#else
+#define LZ4_FORCE_MEMORY_ACCESS 1
+#endif
+
+/* On ppc64le GCC, lz4.c marks its decompression entry points and every
+ * LZ4_FORCE_INLINE helper optimize("O2"), which makes GCC forget command
+ * line options such as -fno-strict-aliasing for those functions (see commit
+ * 080b6f407635).  Expand the attribute to nothing; lz4_kernel_api.h undoes
+ * this.
+ */
+#define optimize(level)
+
+/* lz4.c sets this before including lz4.h, since its own code calls
+ * deprecated entry points.  lz4.h is included below, ahead of lz4.c.
+ */
+#define LZ4_DISABLE_DEPRECATE_WARNINGS 1
+
+/* Rename upstream's entry points out of the way; four of them take a
+ * trailing wrkmem argument in the kernel.
+ */
+#define LZ4_compress_fast		__lz4_compress_fast
+#define LZ4_compress_default		__lz4_compress_default
+#define LZ4_compress_destSize		__lz4_compress_destSize
+#define LZ4_resetStream			__lz4_resetStream
+#define LZ4_loadDict			__lz4_loadDict
+#define LZ4_saveDict			__lz4_saveDict
+#define LZ4_compress_fast_continue	__lz4_compress_fast_continue
+
+#define LZ4_decompress_safe		__lz4_decompress_safe
+#define LZ4_decompress_safe_partial	__lz4_decompress_safe_partial
+#define LZ4_setStreamDecode		__lz4_setStreamDecode
+#define LZ4_decompress_safe_continue	__lz4_decompress_safe_continue
+#define LZ4_decompress_safe_usingDict	__lz4_decompress_safe_usingDict
+
+#define LZ4_compress_HC			__lz4_compress_HC
+#define LZ4_resetStreamHC		__lz4_resetStreamHC
+#define LZ4_loadDictHC			__lz4_loadDictHC
+#define LZ4_compress_HC_continue	__lz4_compress_HC_continue
+#define LZ4_saveDictHC			__lz4_saveDictHC
+
+/* Upstream declares these bare, so both entry point objects would define
+ * them.  Declaring them static first gives them internal linkage (C11
+ * 6.2.2p4): this one in lz4.h, the three below in lz4.c.
+ */
+static __maybe_unused int LZ4_compress_destSize_extState(void *state, const char *src,
+		char *dst, int *srcSizePtr, int targetDstSize,
+		int acceleration);
+
+#include "upstream/lz4.h"
+
+static __maybe_unused int LZ4_compress_forceExtDict(LZ4_stream_t *LZ4_dict,
+		const char *source, char *dest, int srcSize);
+static __maybe_unused int LZ4_decompress_safe_forceExtDict(const char *source,
+		char *dest,
+		int compressedSize, int maxOutputSize,
+		const void *dictStart, size_t dictSize);
+static __maybe_unused int LZ4_decompress_safe_partial_forceExtDict(const char *source,
+		char *dest, int compressedSize, int targetOutputSize,
+		int dstCapacity, const void *dictStart, size_t dictSize);
+
+#endif /* __LZ4_DEPS_H__ */
diff --git a/lib/lz4/lz4_kernel_api.h b/lib/lz4/lz4_kernel_api.h
new file mode 100644
index 0000000000000000000000000000000000000000..de75e5d976eb9dd318da5318a26b91d7aed81042
--- /dev/null
+++ b/lib/lz4/lz4_kernel_api.h
@@ -0,0 +1,90 @@
+/* SPDX-License-Identifier: GPL-2.0-only OR BSD-2-Clause */
+/*
+ * Copyright (c) 2026 Samsung Electronics Co., Ltd.
+ * Author: Michal Wilczynski <m.wilczynski@samsung.com>
+ *
+ * lz4_kernel_api.h -- the LZ4 API this directory exports
+ *
+ * lz4_deps.h renames upstream's entry points to __lz4_* so that the kernel's
+ * own definitions can use the real names; this header undoes that renaming
+ * and the optimize() override, and declares what lib/lz4 exports.
+ *
+ * The same functions are declared to callers in <linux/lz4.h>, in terms of
+ * opaque stream types.  We cannot include that header here -- it and
+ * upstream's lz4.h describe the same library and collide -- so the
+ * declarations are repeated, expressed in upstream's own types.  Keep the two
+ * in step; <linux/lz4.h> is the one callers compile against.
+ *
+ * Include after upstream/lz4.c or upstream/lz4hc.c.
+ */
+
+#undef LZ4_compress_fast
+#undef LZ4_compress_default
+#undef LZ4_compress_destSize
+#undef LZ4_resetStream
+#undef LZ4_loadDict
+#undef LZ4_saveDict
+#undef LZ4_compress_fast_continue
+
+#undef LZ4_decompress_safe
+#undef LZ4_decompress_safe_partial
+#undef LZ4_setStreamDecode
+#undef LZ4_decompress_safe_continue
+#undef LZ4_decompress_safe_usingDict
+
+#undef LZ4_compress_HC
+#undef LZ4_resetStreamHC
+#undef LZ4_loadDictHC
+#undef LZ4_compress_HC_continue
+#undef LZ4_saveDictHC
+
+#undef optimize
+
+/*
+ * The objects that do not build lz4hc.c never see this type, and it is only
+ * ever used through a pointer here.  Upstream declares it the same way.
+ */
+typedef union LZ4_streamHC_u LZ4_streamHC_t;
+
+/* Compression.  wrkmem is LZ4_MEM_COMPRESS bytes, supplied by the caller. */
+int LZ4_compress_default(const char *source, char *dest, int inputSize,
+			 int maxOutputSize, void *wrkmem);
+int LZ4_compress_fast(const char *source, char *dest, int inputSize,
+		      int maxOutputSize, int acceleration, void *wrkmem);
+int LZ4_compress_destSize(const char *source, char *dest, int *sourceSizePtr,
+			  int targetDestSize, void *wrkmem);
+
+/* Streaming compression. */
+void LZ4_resetStream(LZ4_stream_t *LZ4_stream);
+int LZ4_loadDict(LZ4_stream_t *streamPtr, const char *dictionary,
+		 int dictSize);
+int LZ4_saveDict(LZ4_stream_t *streamPtr, char *safeBuffer, int dictSize);
+int LZ4_compress_fast_continue(LZ4_stream_t *streamPtr, const char *src,
+			       char *dst, int srcSize, int maxDstSize,
+			       int acceleration);
+
+/* Decompression. */
+int LZ4_decompress_safe(const char *source, char *dest, int compressedSize,
+			int maxDecompressedSize);
+int LZ4_decompress_safe_partial(const char *source, char *dest,
+				int compressedSize, int targetOutputSize,
+				int maxDecompressedSize);
+int LZ4_setStreamDecode(LZ4_streamDecode_t *LZ4_streamDecode,
+			const char *dictionary, int dictSize);
+int LZ4_decompress_safe_continue(LZ4_streamDecode_t *LZ4_streamDecode,
+				 const char *source, char *dest,
+				 int compressedSize, int maxDecompressedSize);
+int LZ4_decompress_safe_usingDict(const char *source, char *dest,
+				  int compressedSize, int maxDecompressedSize,
+				  const char *dictStart, int dictSize);
+
+/* HC compression.  wrkmem is LZ4HC_MEM_COMPRESS bytes. */
+int LZ4_compress_HC(const char *src, char *dst, int srcSize, int dstCapacity,
+		    int compressionLevel, void *wrkmem);
+void LZ4_resetStreamHC(LZ4_streamHC_t *streamHCPtr, int compressionLevel);
+int LZ4_loadDictHC(LZ4_streamHC_t *streamHCPtr, const char *dictionary,
+		   int dictSize);
+int LZ4_compress_HC_continue(LZ4_streamHC_t *streamHCPtr, const char *src,
+			     char *dst, int srcSize, int maxDstSize);
+int LZ4_saveDictHC(LZ4_streamHC_t *streamHCPtr, char *safeBuffer,
+		   int maxDictSize);

-- 
2.34.1



_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

[PATCH v2 08/15] lib/lz4: add the build environment for the vendored sources

[Michal Wilczynski]

lz4_deps.h makes the vendored upstream sources build in the kernel. It
gives every upstream entry point internal linkage, so an object carries
only the codec it forwards, selects heap mode with failing allocators so
LZ4HC's ~64K optimal-parser buffer stays off the stack, and renames the
entry points the kernel re-exports so the kernel's own definitions can
use the real names. LZ4LIB_VISIBILITY carries __maybe_unused for the
functions an object does not use.

freestanding/ forwards the four ISO C headers upstream includes to their
<linux/> equivalents. lz4_kernel_api.h undoes the renaming; it carries
its own copy of the exported prototypes until a later patch can include
<linux/lz4.h> instead.

Two settings upstream leaves to the build:

 - Unaligned access goes through memcpy() where the architecture has
   efficient unaligned access, and through packed structs, as
   get_unaligned() does, elsewhere. Left alone, upstream uses plain
   misaligned loads on ARMv6.

 - On ppc64le, lz4.c marks functions __attribute__((optimize("O2"))) for
   GCC, which drops command line options such as -fno-strict-aliasing
   for them; see commit 080b6f407635 ("bpf: Don't rely on GCC
   __attribute__((optimize)) to disable GCSE"). lz4_deps.h defines
   optimize() to nothing for upstream's sources.

Nothing includes either header yet.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 lib/lz4/freestanding/limits.h |  14 +++++
 lib/lz4/freestanding/stddef.h |  15 ++++++
 lib/lz4/freestanding/stdint.h |  14 +++++
 lib/lz4/freestanding/string.h |  14 +++++
 lib/lz4/lz4_deps.h            | 118 ++++++++++++++++++++++++++++++++++++++++++
 lib/lz4/lz4_kernel_api.h      |  90 ++++++++++++++++++++++++++++++++
 6 files changed, 265 insertions(+)

diff --git a/lib/lz4/freestanding/limits.h b/lib/lz4/freestanding/limits.h
new file mode 100644
index 0000000000000000000000000000000000000000..0d1cafbc77eb5d2f564b51feaa911748446abf60
--- /dev/null
+++ b/lib/lz4/freestanding/limits.h
@@ -0,0 +1,14 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (c) 2026 Samsung Electronics Co., Ltd.
+ * Author: Michal Wilczynski <m.wilczynski@samsung.com>
+ *
+ * Freestanding <limits.h> for the vendored upstream LZ4 sources; lz4.c needs
+ * UINT_MAX and lz4hc.c needs INT_MAX.
+ */
+#ifndef __LZ4_FREESTANDING_LIMITS_H__
+#define __LZ4_FREESTANDING_LIMITS_H__
+
+#include <linux/limits.h>
+
+#endif
diff --git a/lib/lz4/freestanding/stddef.h b/lib/lz4/freestanding/stddef.h
new file mode 100644
index 0000000000000000000000000000000000000000..aaafd2413849210fa4ec4fa03ba0e9453a9fe026
--- /dev/null
+++ b/lib/lz4/freestanding/stddef.h
@@ -0,0 +1,15 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (c) 2026 Samsung Electronics Co., Ltd.
+ * Author: Michal Wilczynski <m.wilczynski@samsung.com>
+ *
+ * Freestanding <stddef.h> for the vendored upstream LZ4 sources; -nostdinc
+ * drops the compiler's copy, and lz4.h needs size_t and NULL.
+ */
+#ifndef __LZ4_FREESTANDING_STDDEF_H__
+#define __LZ4_FREESTANDING_STDDEF_H__
+
+#include <linux/stddef.h>	/* NULL, offsetof */
+#include <linux/types.h>	/* size_t, ptrdiff_t */
+
+#endif
diff --git a/lib/lz4/freestanding/stdint.h b/lib/lz4/freestanding/stdint.h
new file mode 100644
index 0000000000000000000000000000000000000000..923db05b19f23fb4051d16e5df2bf134cd5ac11e
--- /dev/null
+++ b/lib/lz4/freestanding/stdint.h
@@ -0,0 +1,14 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (c) 2026 Samsung Electronics Co., Ltd.
+ * Author: Michal Wilczynski <m.wilczynski@samsung.com>
+ *
+ * Freestanding <stdint.h> for the vendored upstream LZ4 sources; forwards to
+ * <linux/types.h>.
+ */
+#ifndef __LZ4_FREESTANDING_STDINT_H__
+#define __LZ4_FREESTANDING_STDINT_H__
+
+#include <linux/types.h>
+
+#endif
diff --git a/lib/lz4/freestanding/string.h b/lib/lz4/freestanding/string.h
new file mode 100644
index 0000000000000000000000000000000000000000..8f3ea3b63013608a780a52fe9c33bf5df557584e
--- /dev/null
+++ b/lib/lz4/freestanding/string.h
@@ -0,0 +1,14 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (c) 2026 Samsung Electronics Co., Ltd.
+ * Author: Michal Wilczynski <m.wilczynski@samsung.com>
+ *
+ * Freestanding <string.h> for the vendored upstream LZ4 sources; lz4.c
+ * includes it for memcpy()/memset(), which must resolve under -nostdinc.
+ */
+#ifndef __LZ4_FREESTANDING_STRING_H__
+#define __LZ4_FREESTANDING_STRING_H__
+
+#include <linux/string.h>
+
+#endif
diff --git a/lib/lz4/lz4_deps.h b/lib/lz4/lz4_deps.h
new file mode 100644
index 0000000000000000000000000000000000000000..23d69c75867f71224cb533e9341bd38439b9dee8
--- /dev/null
+++ b/lib/lz4/lz4_deps.h
@@ -0,0 +1,118 @@
+/* SPDX-License-Identifier: GPL-2.0-only OR BSD-2-Clause */
+/*
+ * Copyright (c) 2026 Samsung Electronics Co., Ltd.
+ * Author: Michal Wilczynski <m.wilczynski@samsung.com>
+ *
+ * Build environment for the vendored upstream LZ4 sources.
+ *
+ * upstream/ is LZ4 v1.10.0 plus three backports: 5ef1f16929b5, the aarch64
+ * fast loop (0aa33884faf5, 51f8af506b4a, ed86bc1ed802) and the match length
+ * limit of 587ed9557708, adapted.  Replace it with a later release rather
+ * than patch it further.  The three entry point files include this header
+ * first.
+ */
+
+#ifndef __LZ4_DEPS_H__
+#define __LZ4_DEPS_H__
+
+#include <linux/build_bug.h>	/* static_assert */
+#include <linux/compiler.h>	/* __maybe_unused */
+#include <linux/string.h>
+#include <linux/types.h>
+
+/* lz4.c uses "current" as a local; <asm/current.h> breaks the build. */
+#undef current
+
+/* Static so an object carries only the codec it forwards.  Each object
+ * forwards only some of them, so mark them __maybe_unused rather than turn
+ * off -Wunused-function, which the pre-boot builds would not see anyway.
+ */
+#define LZ4LIB_VISIBILITY static __maybe_unused
+
+/* Route upstream's "static linking only" entry points through
+ * LZ4LIB_VISIBILITY too.
+ */
+#define LZ4_PUBLISH_STATIC_FUNCTIONS
+
+/* Heap mode only: the stack mode puts LZ4HC's ~64K opt[] in a frame.
+ * Allocation goes through the failing stubs below; upstream handles
+ * NULL by returning 0.
+ */
+#define LZ4_USER_MEMORY_FUNCTIONS
+#define LZ4_HEAPMODE 1
+#define LZ4HC_HEAPMODE 1
+
+/* Failing stubs; nothing in the kernel reaches these. */
+static void *LZ4_malloc(size_t s) { return NULL; }
+static void *LZ4_calloc(size_t n, size_t s) { return NULL; }
+static void LZ4_free(void *p) { }
+
+/* Left alone, upstream reads unaligned words through packed structs, and
+ * on ARMv6 through plain misaligned dereferences.  Use memcpy(), upstream's
+ * documented default, where unaligned access is cheap, and packed structs,
+ * as get_unaligned() does, where it is not: there GCC builds a memcpy()'d
+ * word through a stack slot.
+ */
+#ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
+#define LZ4_FORCE_MEMORY_ACCESS 0
+#else
+#define LZ4_FORCE_MEMORY_ACCESS 1
+#endif
+
+/* On ppc64le GCC, lz4.c marks its decompression entry points and every
+ * LZ4_FORCE_INLINE helper optimize("O2"), which makes GCC forget command
+ * line options such as -fno-strict-aliasing for those functions (see commit
+ * 080b6f407635).  Expand the attribute to nothing; lz4_kernel_api.h undoes
+ * this.
+ */
+#define optimize(level)
+
+/* lz4.c sets this before including lz4.h, since its own code calls
+ * deprecated entry points.  lz4.h is included below, ahead of lz4.c.
+ */
+#define LZ4_DISABLE_DEPRECATE_WARNINGS 1
+
+/* Rename upstream's entry points out of the way; four of them take a
+ * trailing wrkmem argument in the kernel.
+ */
+#define LZ4_compress_fast		__lz4_compress_fast
+#define LZ4_compress_default		__lz4_compress_default
+#define LZ4_compress_destSize		__lz4_compress_destSize
+#define LZ4_resetStream			__lz4_resetStream
+#define LZ4_loadDict			__lz4_loadDict
+#define LZ4_saveDict			__lz4_saveDict
+#define LZ4_compress_fast_continue	__lz4_compress_fast_continue
+
+#define LZ4_decompress_safe		__lz4_decompress_safe
+#define LZ4_decompress_safe_partial	__lz4_decompress_safe_partial
+#define LZ4_setStreamDecode		__lz4_setStreamDecode
+#define LZ4_decompress_safe_continue	__lz4_decompress_safe_continue
+#define LZ4_decompress_safe_usingDict	__lz4_decompress_safe_usingDict
+
+#define LZ4_compress_HC			__lz4_compress_HC
+#define LZ4_resetStreamHC		__lz4_resetStreamHC
+#define LZ4_loadDictHC			__lz4_loadDictHC
+#define LZ4_compress_HC_continue	__lz4_compress_HC_continue
+#define LZ4_saveDictHC			__lz4_saveDictHC
+
+/* Upstream declares these bare, so both entry point objects would define
+ * them.  Declaring them static first gives them internal linkage (C11
+ * 6.2.2p4): this one in lz4.h, the three below in lz4.c.
+ */
+static __maybe_unused int LZ4_compress_destSize_extState(void *state, const char *src,
+		char *dst, int *srcSizePtr, int targetDstSize,
+		int acceleration);
+
+#include "upstream/lz4.h"
+
+static __maybe_unused int LZ4_compress_forceExtDict(LZ4_stream_t *LZ4_dict,
+		const char *source, char *dest, int srcSize);
+static __maybe_unused int LZ4_decompress_safe_forceExtDict(const char *source,
+		char *dest,
+		int compressedSize, int maxOutputSize,
+		const void *dictStart, size_t dictSize);
+static __maybe_unused int LZ4_decompress_safe_partial_forceExtDict(const char *source,
+		char *dest, int compressedSize, int targetOutputSize,
+		int dstCapacity, const void *dictStart, size_t dictSize);
+
+#endif /* __LZ4_DEPS_H__ */
diff --git a/lib/lz4/lz4_kernel_api.h b/lib/lz4/lz4_kernel_api.h
new file mode 100644
index 0000000000000000000000000000000000000000..de75e5d976eb9dd318da5318a26b91d7aed81042
--- /dev/null
+++ b/lib/lz4/lz4_kernel_api.h
@@ -0,0 +1,90 @@
+/* SPDX-License-Identifier: GPL-2.0-only OR BSD-2-Clause */
+/*
+ * Copyright (c) 2026 Samsung Electronics Co., Ltd.
+ * Author: Michal Wilczynski <m.wilczynski@samsung.com>
+ *
+ * lz4_kernel_api.h -- the LZ4 API this directory exports
+ *
+ * lz4_deps.h renames upstream's entry points to __lz4_* so that the kernel's
+ * own definitions can use the real names; this header undoes that renaming
+ * and the optimize() override, and declares what lib/lz4 exports.
+ *
+ * The same functions are declared to callers in <linux/lz4.h>, in terms of
+ * opaque stream types.  We cannot include that header here -- it and
+ * upstream's lz4.h describe the same library and collide -- so the
+ * declarations are repeated, expressed in upstream's own types.  Keep the two
+ * in step; <linux/lz4.h> is the one callers compile against.
+ *
+ * Include after upstream/lz4.c or upstream/lz4hc.c.
+ */
+
+#undef LZ4_compress_fast
+#undef LZ4_compress_default
+#undef LZ4_compress_destSize
+#undef LZ4_resetStream
+#undef LZ4_loadDict
+#undef LZ4_saveDict
+#undef LZ4_compress_fast_continue
+
+#undef LZ4_decompress_safe
+#undef LZ4_decompress_safe_partial
+#undef LZ4_setStreamDecode
+#undef LZ4_decompress_safe_continue
+#undef LZ4_decompress_safe_usingDict
+
+#undef LZ4_compress_HC
+#undef LZ4_resetStreamHC
+#undef LZ4_loadDictHC
+#undef LZ4_compress_HC_continue
+#undef LZ4_saveDictHC
+
+#undef optimize
+
+/*
+ * The objects that do not build lz4hc.c never see this type, and it is only
+ * ever used through a pointer here.  Upstream declares it the same way.
+ */
+typedef union LZ4_streamHC_u LZ4_streamHC_t;
+
+/* Compression.  wrkmem is LZ4_MEM_COMPRESS bytes, supplied by the caller. */
+int LZ4_compress_default(const char *source, char *dest, int inputSize,
+			 int maxOutputSize, void *wrkmem);
+int LZ4_compress_fast(const char *source, char *dest, int inputSize,
+		      int maxOutputSize, int acceleration, void *wrkmem);
+int LZ4_compress_destSize(const char *source, char *dest, int *sourceSizePtr,
+			  int targetDestSize, void *wrkmem);
+
+/* Streaming compression. */
+void LZ4_resetStream(LZ4_stream_t *LZ4_stream);
+int LZ4_loadDict(LZ4_stream_t *streamPtr, const char *dictionary,
+		 int dictSize);
+int LZ4_saveDict(LZ4_stream_t *streamPtr, char *safeBuffer, int dictSize);
+int LZ4_compress_fast_continue(LZ4_stream_t *streamPtr, const char *src,
+			       char *dst, int srcSize, int maxDstSize,
+			       int acceleration);
+
+/* Decompression. */
+int LZ4_decompress_safe(const char *source, char *dest, int compressedSize,
+			int maxDecompressedSize);
+int LZ4_decompress_safe_partial(const char *source, char *dest,
+				int compressedSize, int targetOutputSize,
+				int maxDecompressedSize);
+int LZ4_setStreamDecode(LZ4_streamDecode_t *LZ4_streamDecode,
+			const char *dictionary, int dictSize);
+int LZ4_decompress_safe_continue(LZ4_streamDecode_t *LZ4_streamDecode,
+				 const char *source, char *dest,
+				 int compressedSize, int maxDecompressedSize);
+int LZ4_decompress_safe_usingDict(const char *source, char *dest,
+				  int compressedSize, int maxDecompressedSize,
+				  const char *dictStart, int dictSize);
+
+/* HC compression.  wrkmem is LZ4HC_MEM_COMPRESS bytes. */
+int LZ4_compress_HC(const char *src, char *dst, int srcSize, int dstCapacity,
+		    int compressionLevel, void *wrkmem);
+void LZ4_resetStreamHC(LZ4_streamHC_t *streamHCPtr, int compressionLevel);
+int LZ4_loadDictHC(LZ4_streamHC_t *streamHCPtr, const char *dictionary,
+		   int dictSize);
+int LZ4_compress_HC_continue(LZ4_streamHC_t *streamHCPtr, const char *src,
+			     char *dst, int srcSize, int maxDstSize);
+int LZ4_saveDictHC(LZ4_streamHC_t *streamHCPtr, char *safeBuffer,
+		   int maxDictSize);

-- 
2.34.1

Re: [PATCH v2 08/15] lib/lz4: add the build environment for the vendored sources

[sashiko-bot]

> lz4_deps.h makes the vendored upstream sources build in the kernel. It
> gives every upstream entry point internal linkage, so an object carries
> only the codec it forwards, selects heap mode with failing allocators so
> LZ4HC's ~64K optimal-parser buffer stays off the stack, and renames the
> entry points the kernel re-exports so the kernel's own definitions can
> use the real names. LZ4LIB_VISIBILITY carries __maybe_unused for the
> functions an object does not use.
> 
> freestanding/ forwards the four ISO C headers upstream includes to their
> <linux/> equivalents. lz4_kernel_api.h undoes the renaming; it carries
> its own copy of the exported prototypes until a later patch can include
> <linux/lz4.h> instead.
> 
> Two settings upstream leaves to the build:
> 
>  - Unaligned access goes through memcpy() where the architecture has
>    efficient unaligned access, and through packed structs, as
>    get_unaligned() does, elsewhere. Left alone, upstream uses plain
>    misaligned loads on ARMv6.
> 
>  - On ppc64le, lz4.c marks functions __attribute__((optimize("O2"))) for
>    GCC, which drops command line options such as -fno-strict-aliasing
>    for them; see commit 080b6f407635 ("bpf: Don't rely on GCC
>    __attribute__((optimize)) to disable GCSE"). lz4_deps.h defines
>    optimize() to nothing for upstream's sources.
> 
> Nothing includes either header yet.
> 
> Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>

Sashiko has reviewed this patch and found no issues.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261106-lz4-vendor-upstream-v2-0-ee04c3f0d7dc@samsung.com?part=8

[f2fs-dev] [PATCH v2 09/15] arch: boot: put the LZ4 freestanding headers on the decompressor path

[Michal Wilczynski]

lib/decompress_unlz4.c pulls lib/lz4/lz4_decompress.c into the pre-boot
decompressor on these five arches. A later patch makes that file include
the vendored LZ4 sources, which include ISO C headers that -nostdinc does
not provide, so put lib/lz4/freestanding on the include path first.

Only the one object that includes decompress_unlz4.c gets it, so the
rest of each boot directory keeps -nostdinc's view.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 arch/arm/boot/compressed/Makefile    | 3 +++
 arch/mips/boot/compressed/Makefile   | 4 ++++
 arch/parisc/boot/compressed/Makefile | 3 +++
 arch/s390/boot/Makefile              | 3 +++
 arch/x86/boot/compressed/Makefile    | 3 +++
 5 files changed, 16 insertions(+)

diff --git a/arch/arm/boot/compressed/Makefile b/arch/arm/boot/compressed/Makefile
index e3f550d6285786c8405b16768255af87b4dc8222..713225008aa71e614af8c9c6472fda180354569b 100644
--- a/arch/arm/boot/compressed/Makefile
+++ b/arch/arm/boot/compressed/Makefile
@@ -92,6 +92,9 @@ targets       := vmlinux vmlinux.lds piggy_data piggy.o \
 		 head.o $(OBJS)
 
 KBUILD_CFLAGS += -DDISABLE_BRANCH_PROFILING
+# decompress.c includes decompress_unlz4.c, which builds the vendored LZ4
+# sources; they include a few ISO C headers -nostdinc drops.
+CFLAGS_decompress.o += -I$(srctree)/lib/lz4/freestanding
 
 ccflags-y := -fpic $(call cc-option,-mno-single-pic-base,) -fno-builtin \
 	     -I$(srctree)/scripts/dtc/libfdt -fno-stack-protector \
diff --git a/arch/mips/boot/compressed/Makefile b/arch/mips/boot/compressed/Makefile
index e0b8ec9a9516281933b6cc5b175e82f4013c2f54..48f8a63c54a66826889634a54d3537d735f63512 100644
--- a/arch/mips/boot/compressed/Makefile
+++ b/arch/mips/boot/compressed/Makefile
@@ -30,6 +30,10 @@ endif
 KBUILD_CFLAGS := $(KBUILD_CFLAGS) -D__KERNEL__ -D__DISABLE_EXPORTS \
 	-DBOOT_HEAP_SIZE=$(BOOT_HEAP_SIZE) -D"VMLINUX_LOAD_ADDRESS_ULL=$(VMLINUX_LOAD_ADDRESS)ull"
 
+# decompress.c includes decompress_unlz4.c, which builds the vendored LZ4
+# sources; they include a few ISO C headers -nostdinc drops.
+CFLAGS_decompress.o += -I$(srctree)/lib/lz4/freestanding
+
 KBUILD_AFLAGS := $(KBUILD_AFLAGS) -D__ASSEMBLY__ \
 	-DBOOT_HEAP_SIZE=$(BOOT_HEAP_SIZE) \
 	-DKERNEL_ENTRY=$(VMLINUX_ENTRY_ADDRESS)
diff --git a/arch/parisc/boot/compressed/Makefile b/arch/parisc/boot/compressed/Makefile
index 14eefb5ed5d1d781a4ee0cb2ef61ab86c905cc56..9e82f9d1285f62b164073bdacc0d4a1828d4965c 100644
--- a/arch/parisc/boot/compressed/Makefile
+++ b/arch/parisc/boot/compressed/Makefile
@@ -12,6 +12,9 @@ targets += $(OBJECTS) sizes.h
 
 KBUILD_CFLAGS := -D__KERNEL__ -O2 -DBOOTLOADER
 KBUILD_CFLAGS += -DDISABLE_BRANCH_PROFILING
+# misc.c includes decompress_unlz4.c, which builds the vendored LZ4
+# sources; they include a few ISO C headers -nostdinc drops.
+CFLAGS_misc.o += -I$(srctree)/lib/lz4/freestanding
 KBUILD_CFLAGS += -fno-strict-aliasing
 KBUILD_CFLAGS += $(cflags-y) -fno-delete-null-pointer-checks -fno-builtin-printf
 KBUILD_CFLAGS += -fno-PIE -mno-space-regs -mdisable-fpregs -Os
diff --git a/arch/s390/boot/Makefile b/arch/s390/boot/Makefile
index 10b75e053a6f6bb9083548e8f82796f0aa9a9958..44355051a8c7f96b5c4dab5d110bc8aaf79c8f55 100644
--- a/arch/s390/boot/Makefile
+++ b/arch/s390/boot/Makefile
@@ -21,6 +21,9 @@ KBUILD_AFLAGS := $(filter-out $(CC_FLAGS_MARCH),$(KBUILD_AFLAGS_DECOMPRESSOR))
 KBUILD_CFLAGS := $(filter-out $(CC_FLAGS_MARCH),$(KBUILD_CFLAGS_DECOMPRESSOR))
 KBUILD_AFLAGS += $(CC_FLAGS_MARCH_MINIMUM) -D__DISABLE_EXPORTS
 KBUILD_CFLAGS += $(CC_FLAGS_MARCH_MINIMUM) -D__DISABLE_EXPORTS
+# decompressor.c includes decompress_unlz4.c, which builds the vendored LZ4
+# sources; they include a few ISO C headers -nostdinc drops.
+CFLAGS_decompressor.o += -I$(srctree)/lib/lz4/freestanding
 KBUILD_CFLAGS += $(call cc-option, -Wno-default-const-init-unsafe)
 
 CFLAGS_sclp_early_core.o += -I$(srctree)/drivers/s390/char
diff --git a/arch/x86/boot/compressed/Makefile b/arch/x86/boot/compressed/Makefile
index 06934f9691d6a9d6952532ae5f7d96f5858719dd..328eafcfc35bd14a16bc4ec5472f6c6dc72336a1 100644
--- a/arch/x86/boot/compressed/Makefile
+++ b/arch/x86/boot/compressed/Makefile
@@ -30,6 +30,9 @@ KBUILD_CFLAGS += -fno-strict-aliasing -fPIE
 KBUILD_CFLAGS += -fno-jump-tables
 KBUILD_CFLAGS += -Wundef
 KBUILD_CFLAGS += -DDISABLE_BRANCH_PROFILING
+# misc.c includes decompress_unlz4.c, which builds the vendored LZ4
+# sources; they include a few ISO C headers -nostdinc drops.
+CFLAGS_misc.o += -I$(srctree)/lib/lz4/freestanding
 cflags-$(CONFIG_X86_32) := -march=i386
 cflags-$(CONFIG_X86_64) := -mcmodel=small -mno-red-zone
 KBUILD_CFLAGS += $(cflags-y)

-- 
2.34.1



_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

[PATCH v2 09/15] arch: boot: put the LZ4 freestanding headers on the decompressor path

[Michal Wilczynski]

lib/decompress_unlz4.c pulls lib/lz4/lz4_decompress.c into the pre-boot
decompressor on these five arches. A later patch makes that file include
the vendored LZ4 sources, which include ISO C headers that -nostdinc does
not provide, so put lib/lz4/freestanding on the include path first.

Only the one object that includes decompress_unlz4.c gets it, so the
rest of each boot directory keeps -nostdinc's view.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 arch/arm/boot/compressed/Makefile    | 3 +++
 arch/mips/boot/compressed/Makefile   | 4 ++++
 arch/parisc/boot/compressed/Makefile | 3 +++
 arch/s390/boot/Makefile              | 3 +++
 arch/x86/boot/compressed/Makefile    | 3 +++
 5 files changed, 16 insertions(+)

diff --git a/arch/arm/boot/compressed/Makefile b/arch/arm/boot/compressed/Makefile
index e3f550d6285786c8405b16768255af87b4dc8222..713225008aa71e614af8c9c6472fda180354569b 100644
--- a/arch/arm/boot/compressed/Makefile
+++ b/arch/arm/boot/compressed/Makefile
@@ -92,6 +92,9 @@ targets       := vmlinux vmlinux.lds piggy_data piggy.o \
 		 head.o $(OBJS)
 
 KBUILD_CFLAGS += -DDISABLE_BRANCH_PROFILING
+# decompress.c includes decompress_unlz4.c, which builds the vendored LZ4
+# sources; they include a few ISO C headers -nostdinc drops.
+CFLAGS_decompress.o += -I$(srctree)/lib/lz4/freestanding
 
 ccflags-y := -fpic $(call cc-option,-mno-single-pic-base,) -fno-builtin \
 	     -I$(srctree)/scripts/dtc/libfdt -fno-stack-protector \
diff --git a/arch/mips/boot/compressed/Makefile b/arch/mips/boot/compressed/Makefile
index e0b8ec9a9516281933b6cc5b175e82f4013c2f54..48f8a63c54a66826889634a54d3537d735f63512 100644
--- a/arch/mips/boot/compressed/Makefile
+++ b/arch/mips/boot/compressed/Makefile
@@ -30,6 +30,10 @@ endif
 KBUILD_CFLAGS := $(KBUILD_CFLAGS) -D__KERNEL__ -D__DISABLE_EXPORTS \
 	-DBOOT_HEAP_SIZE=$(BOOT_HEAP_SIZE) -D"VMLINUX_LOAD_ADDRESS_ULL=$(VMLINUX_LOAD_ADDRESS)ull"
 
+# decompress.c includes decompress_unlz4.c, which builds the vendored LZ4
+# sources; they include a few ISO C headers -nostdinc drops.
+CFLAGS_decompress.o += -I$(srctree)/lib/lz4/freestanding
+
 KBUILD_AFLAGS := $(KBUILD_AFLAGS) -D__ASSEMBLY__ \
 	-DBOOT_HEAP_SIZE=$(BOOT_HEAP_SIZE) \
 	-DKERNEL_ENTRY=$(VMLINUX_ENTRY_ADDRESS)
diff --git a/arch/parisc/boot/compressed/Makefile b/arch/parisc/boot/compressed/Makefile
index 14eefb5ed5d1d781a4ee0cb2ef61ab86c905cc56..9e82f9d1285f62b164073bdacc0d4a1828d4965c 100644
--- a/arch/parisc/boot/compressed/Makefile
+++ b/arch/parisc/boot/compressed/Makefile
@@ -12,6 +12,9 @@ targets += $(OBJECTS) sizes.h
 
 KBUILD_CFLAGS := -D__KERNEL__ -O2 -DBOOTLOADER
 KBUILD_CFLAGS += -DDISABLE_BRANCH_PROFILING
+# misc.c includes decompress_unlz4.c, which builds the vendored LZ4
+# sources; they include a few ISO C headers -nostdinc drops.
+CFLAGS_misc.o += -I$(srctree)/lib/lz4/freestanding
 KBUILD_CFLAGS += -fno-strict-aliasing
 KBUILD_CFLAGS += $(cflags-y) -fno-delete-null-pointer-checks -fno-builtin-printf
 KBUILD_CFLAGS += -fno-PIE -mno-space-regs -mdisable-fpregs -Os
diff --git a/arch/s390/boot/Makefile b/arch/s390/boot/Makefile
index 10b75e053a6f6bb9083548e8f82796f0aa9a9958..44355051a8c7f96b5c4dab5d110bc8aaf79c8f55 100644
--- a/arch/s390/boot/Makefile
+++ b/arch/s390/boot/Makefile
@@ -21,6 +21,9 @@ KBUILD_AFLAGS := $(filter-out $(CC_FLAGS_MARCH),$(KBUILD_AFLAGS_DECOMPRESSOR))
 KBUILD_CFLAGS := $(filter-out $(CC_FLAGS_MARCH),$(KBUILD_CFLAGS_DECOMPRESSOR))
 KBUILD_AFLAGS += $(CC_FLAGS_MARCH_MINIMUM) -D__DISABLE_EXPORTS
 KBUILD_CFLAGS += $(CC_FLAGS_MARCH_MINIMUM) -D__DISABLE_EXPORTS
+# decompressor.c includes decompress_unlz4.c, which builds the vendored LZ4
+# sources; they include a few ISO C headers -nostdinc drops.
+CFLAGS_decompressor.o += -I$(srctree)/lib/lz4/freestanding
 KBUILD_CFLAGS += $(call cc-option, -Wno-default-const-init-unsafe)
 
 CFLAGS_sclp_early_core.o += -I$(srctree)/drivers/s390/char
diff --git a/arch/x86/boot/compressed/Makefile b/arch/x86/boot/compressed/Makefile
index 06934f9691d6a9d6952532ae5f7d96f5858719dd..328eafcfc35bd14a16bc4ec5472f6c6dc72336a1 100644
--- a/arch/x86/boot/compressed/Makefile
+++ b/arch/x86/boot/compressed/Makefile
@@ -30,6 +30,9 @@ KBUILD_CFLAGS += -fno-strict-aliasing -fPIE
 KBUILD_CFLAGS += -fno-jump-tables
 KBUILD_CFLAGS += -Wundef
 KBUILD_CFLAGS += -DDISABLE_BRANCH_PROFILING
+# misc.c includes decompress_unlz4.c, which builds the vendored LZ4
+# sources; they include a few ISO C headers -nostdinc drops.
+CFLAGS_misc.o += -I$(srctree)/lib/lz4/freestanding
 cflags-$(CONFIG_X86_32) := -march=i386
 cflags-$(CONFIG_X86_64) := -mcmodel=small -mno-red-zone
 KBUILD_CFLAGS += $(cflags-y)

-- 
2.34.1

Re: [PATCH v2 09/15] arch: boot: put the LZ4 freestanding headers on the decompressor path

[sashiko-bot]

> lib/decompress_unlz4.c pulls lib/lz4/lz4_decompress.c into the pre-boot
> decompressor on these five arches. A later patch makes that file include
> the vendored LZ4 sources, which include ISO C headers that -nostdinc does
> not provide, so put lib/lz4/freestanding on the include path first.
> 
> Only the one object that includes decompress_unlz4.c gets it, so the
> rest of each boot directory keeps -nostdinc's view.
> 
> Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>

Sashiko has reviewed this patch and found no issues.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261106-lz4-vendor-upstream-v2-0-ee04c3f0d7dc@samsung.com?part=9

[f2fs-dev] [PATCH v2 10/15] lib/lz4: switch the compressor to the vendored sources

[Michal Wilczynski]

Replace the forked compressor with thin entry points over the vendored
lz4.c. LZ4_compress_default(), LZ4_compress_fast() and
LZ4_compress_destSize() keep the trailing wrkmem argument and forward to
upstream's *_extState() functions; the rest are plain forwarders.

Upstream's LZ4_initStream() returns NULL for a misaligned workspace,
which LZ4_compress_fast_extState() dereferences and
LZ4_compress_destSize_extState() ignores. Check the alignment in the
entry points, warn once and fail the call. In-tree callers allocate with
kmalloc() or vmalloc() and are not affected.

Compressed output differs from the fork's on a small fraction of inputs
by a byte or two; both decode each other's output. Compressing 4K
blocks, upstream is 2-4% faster on kernel source and binaries and up to
9% slower on datagen.

LZ4_stream_t becomes an incomplete type, as upstream declares it, so
<linux/lz4.h> and upstream's lz4.h can share a translation unit. The
macros that laid out the forked structure go, and LZ4_MEM_COMPRESS keeps
its value, 16416. zram, the only caller using sizeof(LZ4_stream_t), now
allocates LZ4_MEM_COMPRESS bytes.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 drivers/block/zram/backend_lz4.c |   6 +-
 include/linux/lz4.h              |  47 +-
 lib/lz4/Makefile                 |  12 +-
 lib/lz4/lz4_compress.c           | 946 +++------------------------------------
 4 files changed, 80 insertions(+), 931 deletions(-)

diff --git a/drivers/block/zram/backend_lz4.c b/drivers/block/zram/backend_lz4.c
index 1e4ad31d39a63f45b3bddd3d8a467d8d7d095beb..50265e3ce256490994cb6dae32a45da376e408e0 100644
--- a/drivers/block/zram/backend_lz4.c
+++ b/drivers/block/zram/backend_lz4.c
@@ -42,7 +42,7 @@ static int lz4_setup_params(struct zcomp_params *params)
 	if (!params->dict || !params->dict_sz)
 		return 0;
 
-	dict_stream = kzalloc_obj(*dict_stream);
+	dict_stream = kzalloc(LZ4_MEM_COMPRESS, GFP_KERNEL);
 	if (!dict_stream)
 		return -ENOMEM;
 
@@ -88,7 +88,7 @@ static int lz4_create(struct zcomp_params *params, struct zcomp_ctx *ctx)
 		if (!zctx->dstrm)
 			goto error;
 
-		zctx->cstrm = kzalloc_obj(*zctx->cstrm);
+		zctx->cstrm = kzalloc(LZ4_MEM_COMPRESS, GFP_KERNEL);
 		if (!zctx->cstrm)
 			goto error;
 	}
@@ -112,7 +112,7 @@ static int lz4_compress(struct zcomp_params *params, struct zcomp_ctx *ctx,
 					zctx->mem);
 	} else {
 		/* Cstrm needs to be reset */
-		memcpy(zctx->cstrm, params->drv_data, sizeof(*zctx->cstrm));
+		memcpy(zctx->cstrm, params->drv_data, LZ4_MEM_COMPRESS);
 		ret = LZ4_compress_fast_continue(zctx->cstrm, req->src,
 						 req->dst, req->src_len,
 						 req->dst_len, params->level);
diff --git a/include/linux/lz4.h b/include/linux/lz4.h
index 0df69e2b1ad523c291e3a96ab5595df9239f7252..031a410b8e3878e9a24cc49af36cfcbba399d991 100644
--- a/include/linux/lz4.h
+++ b/include/linux/lz4.h
@@ -47,16 +47,6 @@
 /*-************************************************************************
  *	CONSTANTS
  **************************************************************************/
-/*
- * LZ4_MEMORY_USAGE :
- * Memory usage formula : N->2^N Bytes
- * (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
- * Increasing memory usage improves compression ratio
- * Reduced memory usage can improve speed, due to cache effect
- * Default value is 14, for 16KB, which nicely fits into Intel x86 L1 cache
- */
-#define LZ4_MEMORY_USAGE 14
-
 #define LZ4_MAX_INPUT_SIZE	0x7E000000 /* 2 113 929 216 bytes */
 #define LZ4_COMPRESSBOUND(isize)	(\
 	(unsigned int)(isize) > (unsigned int)LZ4_MAX_INPUT_SIZE \
@@ -64,9 +54,6 @@
 	: (isize) + ((isize)/255) + 16)
 
 #define LZ4_ACCELERATION_DEFAULT 1
-#define LZ4_HASHLOG	 (LZ4_MEMORY_USAGE-2)
-#define LZ4_HASHTABLESIZE (1 << LZ4_MEMORY_USAGE)
-#define LZ4_HASH_SIZE_U32 (1 << LZ4_HASHLOG)
 
 #define LZ4HC_MIN_CLEVEL			3
 #define LZ4HC_DEFAULT_CLEVEL			9
@@ -82,9 +69,6 @@
 /*-************************************************************************
  *	STREAMING CONSTANTS AND STRUCTURES
  **************************************************************************/
-#define LZ4_STREAMSIZE_U64 ((1 << (LZ4_MEMORY_USAGE - 3)) + 4)
-#define LZ4_STREAMSIZE	(LZ4_STREAMSIZE_U64 * sizeof(unsigned long long))
-
 #define LZ4_STREAMHCSIZE        262192
 #define LZ4_STREAMHCSIZE_SIZET (262192 / sizeof(size_t))
 
@@ -93,20 +77,10 @@
 	sizeof(unsigned long long))
 
 /*
- * LZ4_stream_t - information structure to track an LZ4 stream.
+ * LZ4_stream_t - an LZ4 stream.  Incomplete: lib/lz4 owns the layout.
+ * Allocate LZ4_MEM_COMPRESS bytes and cast, do not sizeof().
  */
-typedef struct {
-	uint32_t hashTable[LZ4_HASH_SIZE_U32];
-	uint32_t currentOffset;
-	uint32_t initCheck;
-	const uint8_t *dictionary;
-	uint8_t *bufferStart;
-	uint32_t dictSize;
-} LZ4_stream_t_internal;
-typedef union {
-	unsigned long long table[LZ4_STREAMSIZE_U64];
-	LZ4_stream_t_internal internal_donotuse;
-} LZ4_stream_t;
+typedef union LZ4_stream_u LZ4_stream_t;
 
 /*
  * LZ4_streamHC_t - information structure to track an LZ4HC stream.
@@ -153,7 +127,12 @@ typedef union {
 /*-************************************************************************
  *	SIZE OF STATE
  **************************************************************************/
-#define LZ4_MEM_COMPRESS	LZ4_STREAMSIZE
+/*
+ * Working memory for the compressors.  It must be aligned to at least 8
+ * bytes; anything from kmalloc() or vmalloc() already is.  The entry
+ * points warn about a misaligned buffer and fail the call.
+ */
+#define LZ4_MEM_COMPRESS	16416
 #define LZ4HC_MEM_COMPRESS	LZ4_STREAMHCSIZE
 
 /*-************************************************************************
@@ -180,7 +159,7 @@ static inline int LZ4_compressBound(size_t isize)
  * @maxOutputSize: full or partial size of buffer 'dest'
  *	which must be already allocated
  * @wrkmem: address of the working memory.
- *	This requires 'workmem' of LZ4_MEM_COMPRESS.
+ *	This requires 'workmem' of LZ4_MEM_COMPRESS, aligned to 8 bytes.
  *
  * Compresses 'sourceSize' bytes from buffer 'source'
  * into already allocated 'dest' buffer of size 'maxOutputSize'.
@@ -206,7 +185,7 @@ int LZ4_compress_default(const char *source, char *dest, int inputSize,
  *	which must be already allocated
  * @acceleration: acceleration factor
  * @wrkmem: address of the working memory.
- *	This requires 'workmem' of LZ4_MEM_COMPRESS.
+ *	This requires 'workmem' of LZ4_MEM_COMPRESS, aligned to 8 bytes.
  *
  * Same as LZ4_compress_default(), but allows to select an "acceleration"
  * factor. The larger the acceleration value, the faster the algorithm,
@@ -230,7 +209,7 @@ int LZ4_compress_fast(const char *source, char *dest, int inputSize,
  *	from 'source' to fill 'dest'. New value is necessarily <= old value.
  * @targetDestSize: Size of buffer 'dest' which must be already allocated
  * @wrkmem: address of the working memory.
- *	This requires 'workmem' of LZ4_MEM_COMPRESS.
+ *	This requires 'workmem' of LZ4_MEM_COMPRESS, aligned to 8 bytes.
  *
  * Reverse the logic, by compressing as much data as possible
  * from 'source' buffer into already allocated buffer 'dest'
@@ -314,7 +293,7 @@ int LZ4_decompress_safe_partial(const char *source, char *dest,
  *	value between 1 and LZ4HC_MAX_CLEVEL will work.
  *	Values >LZ4HC_MAX_CLEVEL behave the same as 16.
  * @wrkmem: address of the working memory.
- *	This requires 'wrkmem' of size LZ4HC_MEM_COMPRESS.
+ *	This requires 'wrkmem' of size LZ4HC_MEM_COMPRESS, aligned to 8 bytes.
  *
  * Compress data from 'src' into 'dst', using the more powerful
  * but slower "HC" algorithm. Compression is guaranteed to succeed if
diff --git a/lib/lz4/Makefile b/lib/lz4/Makefile
index 5b42242afaa20deb165d8a2f0b1c0c842dfcdc39..8b4eb07694e67dd99ec7cc85212fb68138d3981e 100644
--- a/lib/lz4/Makefile
+++ b/lib/lz4/Makefile
@@ -1,6 +1,12 @@
 # SPDX-License-Identifier: GPL-2.0-only
 ccflags-y += -O3
 
-obj-$(CONFIG_LZ4_COMPRESS) += lz4_compress.o
-obj-$(CONFIG_LZ4HC_COMPRESS) += lz4hc_compress.o
-obj-$(CONFIG_LZ4_DECOMPRESS) += lz4_decompress.o
+# upstream/ holds the upstream LZ4 sources lz4_deps.h describes; replace them
+# with a later release rather than patch them further.  The three entry point
+# files include them.  freestanding/ supplies the ISO C headers upstream
+# includes that -nostdinc does not provide.
+ccflags-y += -I $(src)/freestanding
+
+obj-$(CONFIG_LZ4_COMPRESS)	+= lz4_compress.o
+obj-$(CONFIG_LZ4HC_COMPRESS)	+= lz4hc_compress.o
+obj-$(CONFIG_LZ4_DECOMPRESS)	+= lz4_decompress.o
diff --git a/lib/lz4/lz4_compress.c b/lib/lz4/lz4_compress.c
index 2a397bb2c661d9c2e4a7a67d5f50f06bfd42f1f2..941e32a19ac972aef543c9a03b447a8a98bf5215 100644
--- a/lib/lz4/lz4_compress.c
+++ b/lib/lz4/lz4_compress.c
@@ -1,937 +1,101 @@
+// SPDX-License-Identifier: GPL-2.0-only OR BSD-2-Clause
 /*
- * LZ4 - Fast LZ compression algorithm
  * Copyright (C) 2011 - 2016, Yann Collet.
- * BSD 2 - Clause License (http://www.opensource.org/licenses/bsd - license.php)
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions are
- * met:
- *	* Redistributions of source code must retain the above copyright
- *	  notice, this list of conditions and the following disclaimer.
- *	* Redistributions in binary form must reproduce the above
- * copyright notice, this list of conditions and the following disclaimer
- * in the documentation and/or other materials provided with the
- * distribution.
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
- * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
- * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
- * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
- * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
- * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
- * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
- * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
- * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- * You can contact the author at :
- *	- LZ4 homepage : http://www.lz4.org
- *	- LZ4 source repository : https://github.com/lz4/lz4
+ * Copyright (C) 2016, Sven Schmidt <4sschmid@informatik.uni-hamburg.de>
+ * Copyright (c) 2026 Samsung Electronics Co., Ltd.
+ * Author: Michal Wilczynski <m.wilczynski@samsung.com>
  *
- *	Changed for kernel usage by:
- *	Sven Schmidt <4sschmid@informatik.uni-hamburg.de>
+ * LZ4 compressor -- kernel entry points
+ *
+ * The compressor is upstream's, in the verbatim lz4.c included below.  The
+ * kernel has no heap at these call sites, so LZ4_compress_default(),
+ * LZ4_compress_fast() and LZ4_compress_destSize() take a trailing wrkmem
+ * argument and forward to upstream's *_extState() entry points.
  */
 
-/*-************************************
- *	Dependencies
- **************************************/
-#include "lz4defs.h"
-#include <linux/module.h>
-#include <linux/kernel.h>
-#include <linux/unaligned.h>
-
-static const int LZ4_minLength = (MFLIMIT + 1);
-static const int LZ4_64Klimit = ((64 * KB) + (MFLIMIT - 1));
-
-/*-******************************
- *	Compression functions
- ********************************/
-static FORCE_INLINE U32 LZ4_hash4(
-	U32 sequence,
-	tableType_t const tableType)
-{
-	if (tableType == byU16)
-		return ((sequence * 2654435761U)
-			>> ((MINMATCH * 8) - (LZ4_HASHLOG + 1)));
-	else
-		return ((sequence * 2654435761U)
-			>> ((MINMATCH * 8) - LZ4_HASHLOG));
-}
-
-static FORCE_INLINE U32 LZ4_hash5(
-	U64 sequence,
-	tableType_t const tableType)
-{
-	const U32 hashLog = (tableType == byU16)
-		? LZ4_HASHLOG + 1
-		: LZ4_HASHLOG;
-
-#if LZ4_LITTLE_ENDIAN
-	static const U64 prime5bytes = 889523592379ULL;
-
-	return (U32)(((sequence << 24) * prime5bytes) >> (64 - hashLog));
-#else
-	static const U64 prime8bytes = 11400714785074694791ULL;
-
-	return (U32)(((sequence >> 24) * prime8bytes) >> (64 - hashLog));
-#endif
-}
-
-static FORCE_INLINE U32 LZ4_hashPosition(
-	const void *p,
-	tableType_t const tableType)
-{
-#if LZ4_ARCH64
-	if (tableType == byU32)
-		return LZ4_hash5(LZ4_read_ARCH(p), tableType);
-#endif
-
-	return LZ4_hash4(LZ4_read32(p), tableType);
-}
-
-static void LZ4_putPositionOnHash(
-	const BYTE *p,
-	U32 h,
-	void *tableBase,
-	tableType_t const tableType,
-	const BYTE *srcBase)
-{
-	switch (tableType) {
-	case byPtr:
-	{
-		const BYTE **hashTable = (const BYTE **)tableBase;
-
-		hashTable[h] = p;
-		return;
-	}
-	case byU32:
-	{
-		U32 *hashTable = (U32 *) tableBase;
-
-		hashTable[h] = (U32)(p - srcBase);
-		return;
-	}
-	case byU16:
-	{
-		U16 *hashTable = (U16 *) tableBase;
-
-		hashTable[h] = (U16)(p - srcBase);
-		return;
-	}
-	}
-}
-
-static FORCE_INLINE void LZ4_putPosition(
-	const BYTE *p,
-	void *tableBase,
-	tableType_t tableType,
-	const BYTE *srcBase)
-{
-	U32 const h = LZ4_hashPosition(p, tableType);
+#include "lz4_deps.h"
+#include "upstream/lz4.c"
 
-	LZ4_putPositionOnHash(p, h, tableBase, tableType, srcBase);
-}
-
-static const BYTE *LZ4_getPositionOnHash(
-	U32 h,
-	void *tableBase,
-	tableType_t tableType,
-	const BYTE *srcBase)
-{
-	if (tableType == byPtr) {
-		const BYTE **hashTable = (const BYTE **) tableBase;
+/* Upstream's short helper macros; MIN and MAX clash with <linux/minmax.h>. */
+#undef MIN
+#undef MAX
+#undef KB
+#undef MB
+#undef GB
 
-		return hashTable[h];
-	}
-
-	if (tableType == byU32) {
-		const U32 * const hashTable = (U32 *) tableBase;
-
-		return hashTable[h] + srcBase;
-	}
-
-	{
-		/* default, to ensure a return */
-		const U16 * const hashTable = (U16 *) tableBase;
-
-		return hashTable[h] + srcBase;
-	}
-}
-
-static FORCE_INLINE const BYTE *LZ4_getPosition(
-	const BYTE *p,
-	void *tableBase,
-	tableType_t tableType,
-	const BYTE *srcBase)
-{
-	U32 const h = LZ4_hashPosition(p, tableType);
-
-	return LZ4_getPositionOnHash(h, tableBase, tableType, srcBase);
-}
+#include "lz4_kernel_api.h"
+#include <linux/bug.h>
+#include <linux/export.h>
+#include <linux/module.h>
 
+/* Catch any divergence from upstream's layout at build time. */
+static_assert(sizeof(LZ4_stream_t) == LZ4_STREAM_MINSIZE);
+static_assert(LZ4_STREAM_MINSIZE == 16416); /* LZ4_MEM_COMPRESS */
 
-/*
- * LZ4_compress_generic() :
- * inlined, to ensure branches are decided at compilation time
+/* LZ4_initStream() returns NULL for a misaligned state, which
+ * LZ4_compress_fast_extState() dereferences and
+ * LZ4_compress_destSize_extState() ignores, behind assert()s the kernel
+ * compiles out.  Fail the call instead.
  */
-static FORCE_INLINE int LZ4_compress_generic(
-	LZ4_stream_t_internal * const dictPtr,
-	const char * const source,
-	char * const dest,
-	const int inputSize,
-	const int maxOutputSize,
-	const limitedOutput_directive outputLimited,
-	const tableType_t tableType,
-	const dict_directive dict,
-	const dictIssue_directive dictIssue,
-	const U32 acceleration)
+static bool lz4_wrkmem_misaligned(const void *wrkmem)
 {
-	const BYTE *ip = (const BYTE *) source;
-	const BYTE *base;
-	const BYTE *lowLimit;
-	const BYTE * const lowRefLimit = ip - dictPtr->dictSize;
-	const BYTE * const dictionary = dictPtr->dictionary;
-	const BYTE * const dictEnd = dictionary + dictPtr->dictSize;
-	const size_t dictDelta = dictEnd - (const BYTE *)source;
-	const BYTE *anchor = (const BYTE *) source;
-	const BYTE * const iend = ip + inputSize;
-	const BYTE * const mflimit = iend - MFLIMIT;
-	const BYTE * const matchlimit = iend - LASTLITERALS;
-
-	BYTE *op = (BYTE *) dest;
-	BYTE * const olimit = op + maxOutputSize;
-
-	U32 forwardH;
-	size_t refDelta = 0;
-
-	/* Init conditions */
-	if ((U32)inputSize > (U32)LZ4_MAX_INPUT_SIZE) {
-		/* Unsupported inputSize, too large (or negative) */
-		return 0;
-	}
-
-	switch (dict) {
-	case noDict:
-	default:
-		base = (const BYTE *)source;
-		lowLimit = (const BYTE *)source;
-		break;
-	case withPrefix64k:
-		base = (const BYTE *)source - dictPtr->currentOffset;
-		lowLimit = (const BYTE *)source - dictPtr->dictSize;
-		break;
-	case usingExtDict:
-		base = (const BYTE *)source - dictPtr->currentOffset;
-		lowLimit = (const BYTE *)source;
-		break;
-	}
-
-	if ((tableType == byU16)
-		&& (inputSize >= LZ4_64Klimit)) {
-		/* Size too large (not within 64K limit) */
-		return 0;
-	}
-
-	if (inputSize < LZ4_minLength) {
-		/* Input too small, no compression (all literals) */
-		goto _last_literals;
-	}
-
-	/* First Byte */
-	LZ4_putPosition(ip, dictPtr->hashTable, tableType, base);
-	ip++;
-	forwardH = LZ4_hashPosition(ip, tableType);
-
-	/* Main Loop */
-	for ( ; ; ) {
-		const BYTE *match;
-		BYTE *token;
-
-		/* Find a match */
-		{
-			const BYTE *forwardIp = ip;
-			unsigned int step = 1;
-			unsigned int searchMatchNb = acceleration << LZ4_SKIPTRIGGER;
-
-			do {
-				U32 const h = forwardH;
-
-				ip = forwardIp;
-				forwardIp += step;
-				step = (searchMatchNb++ >> LZ4_SKIPTRIGGER);
-
-				if (unlikely(forwardIp > mflimit))
-					goto _last_literals;
-
-				match = LZ4_getPositionOnHash(h,
-					dictPtr->hashTable,
-					tableType, base);
-
-				if (dict == usingExtDict) {
-					if (match < (const BYTE *)source) {
-						refDelta = dictDelta;
-						lowLimit = dictionary;
-					} else {
-						refDelta = 0;
-						lowLimit = (const BYTE *)source;
-				}	 }
-
-				forwardH = LZ4_hashPosition(forwardIp,
-					tableType);
-
-				LZ4_putPositionOnHash(ip, h, dictPtr->hashTable,
-					tableType, base);
-			} while (((dictIssue == dictSmall)
-					? (match < lowRefLimit)
-					: 0)
-				|| ((tableType == byU16)
-					? 0
-					: (match + MAX_DISTANCE < ip))
-				|| (LZ4_read32(match + refDelta)
-					!= LZ4_read32(ip)));
-		}
-
-		/* Catch up */
-		while (((ip > anchor) & (match + refDelta > lowLimit))
-				&& (unlikely(ip[-1] == match[refDelta - 1]))) {
-			ip--;
-			match--;
-		}
-
-		/* Encode Literals */
-		{
-			unsigned const int litLength = (unsigned int)(ip - anchor);
-
-			token = op++;
-
-			if ((outputLimited) &&
-				/* Check output buffer overflow */
-				(unlikely(op + litLength +
-					(2 + 1 + LASTLITERALS) +
-					(litLength / 255) > olimit)))
-				return 0;
-
-			if (litLength >= RUN_MASK) {
-				int len = (int)litLength - RUN_MASK;
-
-				*token = (RUN_MASK << ML_BITS);
-
-				for (; len >= 255; len -= 255)
-					*op++ = 255;
-				*op++ = (BYTE)len;
-			} else
-				*token = (BYTE)(litLength << ML_BITS);
-
-			/* Copy Literals */
-			LZ4_wildCopy(op, anchor, op + litLength);
-			op += litLength;
-		}
-
-_next_match:
-		/* Encode Offset */
-		LZ4_writeLE16(op, (U16)(ip - match));
-		op += 2;
-
-		/* Encode MatchLength */
-		{
-			unsigned int matchCode;
-
-			if ((dict == usingExtDict)
-				&& (lowLimit == dictionary)) {
-				const BYTE *limit;
-
-				match += refDelta;
-				limit = ip + (dictEnd - match);
-
-				if (limit > matchlimit)
-					limit = matchlimit;
-
-				matchCode = LZ4_count(ip + MINMATCH,
-					match + MINMATCH, limit);
-
-				ip += MINMATCH + matchCode;
-
-				if (ip == limit) {
-					unsigned const int more = LZ4_count(ip,
-						(const BYTE *)source,
-						matchlimit);
-
-					matchCode += more;
-					ip += more;
-				}
-			} else {
-				matchCode = LZ4_count(ip + MINMATCH,
-					match + MINMATCH, matchlimit);
-				ip += MINMATCH + matchCode;
-			}
-
-			if (outputLimited &&
-				/* Check output buffer overflow */
-				(unlikely(op +
-					(1 + LASTLITERALS) +
-					(matchCode >> 8) > olimit)))
-				return 0;
-
-			if (matchCode >= ML_MASK) {
-				*token += ML_MASK;
-				matchCode -= ML_MASK;
-				LZ4_write32(op, 0xFFFFFFFF);
-
-				while (matchCode >= 4 * 255) {
-					op += 4;
-					LZ4_write32(op, 0xFFFFFFFF);
-					matchCode -= 4 * 255;
-				}
-
-				op += matchCode / 255;
-				*op++ = (BYTE)(matchCode % 255);
-			} else
-				*token += (BYTE)(matchCode);
-		}
-
-		anchor = ip;
-
-		/* Test end of chunk */
-		if (ip > mflimit)
-			break;
-
-		/* Fill table */
-		LZ4_putPosition(ip - 2, dictPtr->hashTable, tableType, base);
-
-		/* Test next position */
-		match = LZ4_getPosition(ip, dictPtr->hashTable,
-			tableType, base);
-
-		if (dict == usingExtDict) {
-			if (match < (const BYTE *)source) {
-				refDelta = dictDelta;
-				lowLimit = dictionary;
-			} else {
-				refDelta = 0;
-				lowLimit = (const BYTE *)source;
-			}
-		}
-
-		LZ4_putPosition(ip, dictPtr->hashTable, tableType, base);
-
-		if (((dictIssue == dictSmall) ? (match >= lowRefLimit) : 1)
-			&& (match + MAX_DISTANCE >= ip)
-			&& (LZ4_read32(match + refDelta) == LZ4_read32(ip))) {
-			token = op++;
-			*token = 0;
-			goto _next_match;
-		}
-
-		/* Prepare next loop */
-		forwardH = LZ4_hashPosition(++ip, tableType);
-	}
-
-_last_literals:
-	/* Encode Last Literals */
-	{
-		size_t const lastRun = (size_t)(iend - anchor);
-
-		if ((outputLimited) &&
-			/* Check output buffer overflow */
-			((op - (BYTE *)dest) + lastRun + 1 +
-			((lastRun + 255 - RUN_MASK) / 255) > (U32)maxOutputSize))
-			return 0;
-
-		if (lastRun >= RUN_MASK) {
-			size_t accumulator = lastRun - RUN_MASK;
-			*op++ = RUN_MASK << ML_BITS;
-			for (; accumulator >= 255; accumulator -= 255)
-				*op++ = 255;
-			*op++ = (BYTE) accumulator;
-		} else {
-			*op++ = (BYTE)(lastRun << ML_BITS);
-		}
-
-		LZ4_memcpy(op, anchor, lastRun);
-
-		op += lastRun;
-	}
-
-	/* End */
-	return (int) (((char *)op) - dest);
-}
-
-static int LZ4_compress_fast_extState(
-	void *state,
-	const char *source,
-	char *dest,
-	int inputSize,
-	int maxOutputSize,
-	int acceleration)
-{
-	LZ4_stream_t_internal *ctx = &((LZ4_stream_t *)state)->internal_donotuse;
-#if LZ4_ARCH64
-	const tableType_t tableType = byU32;
-#else
-	const tableType_t tableType = byPtr;
-#endif
-
-	LZ4_resetStream((LZ4_stream_t *)state);
-
-	if (acceleration < 1)
-		acceleration = LZ4_ACCELERATION_DEFAULT;
-
-	if (maxOutputSize >= LZ4_COMPRESSBOUND(inputSize)) {
-		if (inputSize < LZ4_64Klimit)
-			return LZ4_compress_generic(ctx, source,
-				dest, inputSize, 0,
-				noLimit, byU16, noDict,
-				noDictIssue, acceleration);
-		else
-			return LZ4_compress_generic(ctx, source,
-				dest, inputSize, 0,
-				noLimit, tableType, noDict,
-				noDictIssue, acceleration);
-	} else {
-		if (inputSize < LZ4_64Klimit)
-			return LZ4_compress_generic(ctx, source,
-				dest, inputSize,
-				maxOutputSize, limitedOutput, byU16, noDict,
-				noDictIssue, acceleration);
-		else
-			return LZ4_compress_generic(ctx, source,
-				dest, inputSize,
-				maxOutputSize, limitedOutput, tableType, noDict,
-				noDictIssue, acceleration);
-	}
+	return WARN_ON_ONCE(!LZ4_isAligned(wrkmem, LZ4_stream_t_alignment()));
 }
 
 int LZ4_compress_fast(const char *source, char *dest, int inputSize,
-	int maxOutputSize, int acceleration, void *wrkmem)
+		      int maxOutputSize, int acceleration, void *wrkmem)
 {
+	if (lz4_wrkmem_misaligned(wrkmem))
+		return 0;
+
 	return LZ4_compress_fast_extState(wrkmem, source, dest, inputSize,
-		maxOutputSize, acceleration);
+					  maxOutputSize, acceleration);
 }
 EXPORT_SYMBOL(LZ4_compress_fast);
 
 int LZ4_compress_default(const char *source, char *dest, int inputSize,
-	int maxOutputSize, void *wrkmem)
+			 int maxOutputSize, void *wrkmem)
 {
-	return LZ4_compress_fast(source, dest, inputSize,
-		maxOutputSize, LZ4_ACCELERATION_DEFAULT, wrkmem);
+	return LZ4_compress_fast(source, dest, inputSize, maxOutputSize,
+				 LZ4_ACCELERATION_DEFAULT, wrkmem);
 }
 EXPORT_SYMBOL(LZ4_compress_default);
 
-/*-******************************
- *	*_destSize() variant
- ********************************/
-static int LZ4_compress_destSize_generic(
-	LZ4_stream_t_internal * const ctx,
-	const char * const src,
-	char * const dst,
-	int * const srcSizePtr,
-	const int targetDstSize,
-	const tableType_t tableType)
+int LZ4_compress_destSize(const char *source, char *dest, int *sourceSizePtr,
+			  int targetDestSize, void *wrkmem)
 {
-	const BYTE *ip = (const BYTE *) src;
-	const BYTE *base = (const BYTE *) src;
-	const BYTE *lowLimit = (const BYTE *) src;
-	const BYTE *anchor = ip;
-	const BYTE * const iend = ip + *srcSizePtr;
-	const BYTE * const mflimit = iend - MFLIMIT;
-	const BYTE * const matchlimit = iend - LASTLITERALS;
-
-	BYTE *op = (BYTE *) dst;
-	BYTE * const oend = op + targetDstSize;
-	BYTE * const oMaxLit = op + targetDstSize - 2 /* offset */
-		- 8 /* because 8 + MINMATCH == MFLIMIT */ - 1 /* token */;
-	BYTE * const oMaxMatch = op + targetDstSize
-		- (LASTLITERALS + 1 /* token */);
-	BYTE * const oMaxSeq = oMaxLit - 1 /* token */;
-
-	U32 forwardH;
-
-	/* Init conditions */
-	/* Impossible to store anything */
-	if (targetDstSize < 1)
-		return 0;
-	/* Unsupported input size, too large (or negative) */
-	if ((U32)*srcSizePtr > (U32)LZ4_MAX_INPUT_SIZE)
-		return 0;
-	/* Size too large (not within 64K limit) */
-	if ((tableType == byU16) && (*srcSizePtr >= LZ4_64Klimit))
+	if (lz4_wrkmem_misaligned(wrkmem))
 		return 0;
-	/* Input too small, no compression (all literals) */
-	if (*srcSizePtr < LZ4_minLength)
-		goto _last_literals;
-
-	/* First Byte */
-	*srcSizePtr = 0;
-	LZ4_putPosition(ip, ctx->hashTable, tableType, base);
-	ip++; forwardH = LZ4_hashPosition(ip, tableType);
-
-	/* Main Loop */
-	for ( ; ; ) {
-		const BYTE *match;
-		BYTE *token;
-
-		/* Find a match */
-		{
-			const BYTE *forwardIp = ip;
-			unsigned int step = 1;
-			unsigned int searchMatchNb = 1 << LZ4_SKIPTRIGGER;
-
-			do {
-				U32 h = forwardH;
-
-				ip = forwardIp;
-				forwardIp += step;
-				step = (searchMatchNb++ >> LZ4_SKIPTRIGGER);
-
-				if (unlikely(forwardIp > mflimit))
-					goto _last_literals;
-
-				match = LZ4_getPositionOnHash(h, ctx->hashTable,
-					tableType, base);
-				forwardH = LZ4_hashPosition(forwardIp,
-					tableType);
-				LZ4_putPositionOnHash(ip, h,
-					ctx->hashTable, tableType,
-					base);
-
-			} while (((tableType == byU16)
-				? 0
-				: (match + MAX_DISTANCE < ip))
-				|| (LZ4_read32(match) != LZ4_read32(ip)));
-		}
-
-		/* Catch up */
-		while ((ip > anchor)
-			&& (match > lowLimit)
-			&& (unlikely(ip[-1] == match[-1]))) {
-			ip--;
-			match--;
-		}
-
-		/* Encode Literal length */
-		{
-			unsigned int litLength = (unsigned int)(ip - anchor);
-
-			token = op++;
-			if (op + ((litLength + 240) / 255)
-				+ litLength > oMaxLit) {
-				/* Not enough space for a last match */
-				op--;
-				goto _last_literals;
-			}
-			if (litLength >= RUN_MASK) {
-				unsigned int len = litLength - RUN_MASK;
-				*token = (RUN_MASK<<ML_BITS);
-				for (; len >= 255; len -= 255)
-					*op++ = 255;
-				*op++ = (BYTE)len;
-			} else
-				*token = (BYTE)(litLength << ML_BITS);
-
-			/* Copy Literals */
-			LZ4_wildCopy(op, anchor, op + litLength);
-			op += litLength;
-		}
-
-_next_match:
-		/* Encode Offset */
-		LZ4_writeLE16(op, (U16)(ip - match)); op += 2;
-
-		/* Encode MatchLength */
-		{
-			size_t matchLength = LZ4_count(ip + MINMATCH,
-			match + MINMATCH, matchlimit);
-
-			if (op + ((matchLength + 240)/255) > oMaxMatch) {
-				/* Match description too long : reduce it */
-				matchLength = (15 - 1) + (oMaxMatch - op) * 255;
-			}
-			ip += MINMATCH + matchLength;
-
-			if (matchLength >= ML_MASK) {
-				*token += ML_MASK;
-				matchLength -= ML_MASK;
-				while (matchLength >= 255) {
-					matchLength -= 255;
-					*op++ = 255;
-				}
-				*op++ = (BYTE)matchLength;
-			} else
-				*token += (BYTE)(matchLength);
-		}
-
-		anchor = ip;
-
-		/* Test end of block */
-		if (ip > mflimit)
-			break;
-		if (op > oMaxSeq)
-			break;
-
-		/* Fill table */
-		LZ4_putPosition(ip - 2, ctx->hashTable, tableType, base);
-
-		/* Test next position */
-		match = LZ4_getPosition(ip, ctx->hashTable, tableType, base);
-		LZ4_putPosition(ip, ctx->hashTable, tableType, base);
 
-		if ((match + MAX_DISTANCE >= ip)
-			&& (LZ4_read32(match) == LZ4_read32(ip))) {
-			token = op++; *token = 0;
-			goto _next_match;
-		}
-
-		/* Prepare next loop */
-		forwardH = LZ4_hashPosition(++ip, tableType);
-	}
-
-_last_literals:
-	/* Encode Last Literals */
-	{
-		size_t lastRunSize = (size_t)(iend - anchor);
-
-		if (op + 1 /* token */
-			+ ((lastRunSize + 240) / 255) /* litLength */
-			+ lastRunSize /* literals */ > oend) {
-			/* adapt lastRunSize to fill 'dst' */
-			lastRunSize	= (oend - op) - 1;
-			lastRunSize -= (lastRunSize + 240) / 255;
-		}
-		ip = anchor + lastRunSize;
-
-		if (lastRunSize >= RUN_MASK) {
-			size_t accumulator = lastRunSize - RUN_MASK;
-
-			*op++ = RUN_MASK << ML_BITS;
-			for (; accumulator >= 255; accumulator -= 255)
-				*op++ = 255;
-			*op++ = (BYTE) accumulator;
-		} else {
-			*op++ = (BYTE)(lastRunSize<<ML_BITS);
-		}
-		LZ4_memcpy(op, anchor, lastRunSize);
-		op += lastRunSize;
-	}
-
-	/* End */
-	*srcSizePtr = (int) (((const char *)ip) - src);
-	return (int) (((char *)op) - dst);
-}
-
-static int LZ4_compress_destSize_extState(
-	LZ4_stream_t *state,
-	const char *src,
-	char *dst,
-	int *srcSizePtr,
-	int targetDstSize)
-{
-#if LZ4_ARCH64
-	const tableType_t tableType = byU32;
-#else
-	const tableType_t tableType = byPtr;
-#endif
-
-	LZ4_resetStream(state);
-
-	if (targetDstSize >= LZ4_COMPRESSBOUND(*srcSizePtr)) {
-		/* compression success is guaranteed */
-		return LZ4_compress_fast_extState(
-			state, src, dst, *srcSizePtr,
-			targetDstSize, 1);
-	} else {
-		if (*srcSizePtr < LZ4_64Klimit)
-			return LZ4_compress_destSize_generic(
-				&state->internal_donotuse,
-				src, dst, srcSizePtr,
-				targetDstSize, byU16);
-		else
-			return LZ4_compress_destSize_generic(
-				&state->internal_donotuse,
-				src, dst, srcSizePtr,
-				targetDstSize, tableType);
-	}
-}
-
-
-int LZ4_compress_destSize(
-	const char *src,
-	char *dst,
-	int *srcSizePtr,
-	int targetDstSize,
-	void *wrkmem)
-{
-	return LZ4_compress_destSize_extState(wrkmem, src, dst, srcSizePtr,
-		targetDstSize);
+	return LZ4_compress_destSize_extState(wrkmem, source, dest,
+					      sourceSizePtr, targetDestSize,
+					      LZ4_ACCELERATION_DEFAULT);
 }
 EXPORT_SYMBOL(LZ4_compress_destSize);
 
-/*-******************************
- *	Streaming functions
- ********************************/
 void LZ4_resetStream(LZ4_stream_t *LZ4_stream)
 {
-	memset(LZ4_stream, 0, sizeof(LZ4_stream_t));
+	__lz4_resetStream(LZ4_stream);
 }
 
-int LZ4_loadDict(LZ4_stream_t *LZ4_dict,
-	const char *dictionary, int dictSize)
+int LZ4_loadDict(LZ4_stream_t *streamPtr, const char *dictionary, int dictSize)
 {
-	LZ4_stream_t_internal *dict = &LZ4_dict->internal_donotuse;
-	const BYTE *p = (const BYTE *)dictionary;
-	const BYTE * const dictEnd = p + dictSize;
-	const BYTE *base;
-
-	if ((dict->initCheck)
-		|| (dict->currentOffset > 1 * GB)) {
-		/* Uninitialized structure, or reuse overflow */
-		LZ4_resetStream(LZ4_dict);
-	}
-
-	if (dictSize < (int)HASH_UNIT) {
-		dict->dictionary = NULL;
-		dict->dictSize = 0;
-		return 0;
-	}
-
-	if ((dictEnd - p) > 64 * KB)
-		p = dictEnd - 64 * KB;
-	dict->currentOffset += 64 * KB;
-	base = p - dict->currentOffset;
-	dict->dictionary = p;
-	dict->dictSize = (U32)(dictEnd - p);
-	dict->currentOffset += dict->dictSize;
-
-	while (p <= dictEnd - HASH_UNIT) {
-		LZ4_putPosition(p, dict->hashTable, byU32, base);
-		p += 3;
-	}
-
-	return dict->dictSize;
+	return __lz4_loadDict(streamPtr, dictionary, dictSize);
 }
 EXPORT_SYMBOL(LZ4_loadDict);
 
-static void LZ4_renormDictT(LZ4_stream_t_internal *LZ4_dict,
-	const BYTE *src)
+int LZ4_saveDict(LZ4_stream_t *streamPtr, char *safeBuffer, int dictSize)
 {
-	if ((LZ4_dict->currentOffset > 0x80000000) ||
-		((uptrval)LZ4_dict->currentOffset > (uptrval)src)) {
-		/* address space overflow */
-		/* rescale hash table */
-		U32 const delta = LZ4_dict->currentOffset - 64 * KB;
-		const BYTE *dictEnd = LZ4_dict->dictionary + LZ4_dict->dictSize;
-		int i;
-
-		for (i = 0; i < LZ4_HASH_SIZE_U32; i++) {
-			if (LZ4_dict->hashTable[i] < delta)
-				LZ4_dict->hashTable[i] = 0;
-			else
-				LZ4_dict->hashTable[i] -= delta;
-		}
-		LZ4_dict->currentOffset = 64 * KB;
-		if (LZ4_dict->dictSize > 64 * KB)
-			LZ4_dict->dictSize = 64 * KB;
-		LZ4_dict->dictionary = dictEnd - LZ4_dict->dictSize;
-	}
-}
-
-int LZ4_saveDict(LZ4_stream_t *LZ4_dict, char *safeBuffer, int dictSize)
-{
-	LZ4_stream_t_internal * const dict = &LZ4_dict->internal_donotuse;
-	const BYTE * const previousDictEnd = dict->dictionary + dict->dictSize;
-
-	if ((U32)dictSize > 64 * KB) {
-		/* useless to define a dictionary > 64 * KB */
-		dictSize = 64 * KB;
-	}
-	if ((U32)dictSize > dict->dictSize)
-		dictSize = dict->dictSize;
-
-	memmove(safeBuffer, previousDictEnd - dictSize, dictSize);
-
-	dict->dictionary = (const BYTE *)safeBuffer;
-	dict->dictSize = (U32)dictSize;
-
-	return dictSize;
+	return __lz4_saveDict(streamPtr, safeBuffer, dictSize);
 }
 EXPORT_SYMBOL(LZ4_saveDict);
 
-int LZ4_compress_fast_continue(LZ4_stream_t *LZ4_stream, const char *source,
-	char *dest, int inputSize, int maxOutputSize, int acceleration)
+int LZ4_compress_fast_continue(LZ4_stream_t *streamPtr, const char *src,
+			       char *dst, int srcSize, int maxDstSize,
+			       int acceleration)
 {
-	LZ4_stream_t_internal *streamPtr = &LZ4_stream->internal_donotuse;
-	const BYTE * const dictEnd = streamPtr->dictionary
-		+ streamPtr->dictSize;
-
-	const BYTE *smallest = (const BYTE *) source;
-
-	if (streamPtr->initCheck) {
-		/* Uninitialized structure detected */
-		return 0;
-	}
-
-	if ((streamPtr->dictSize > 0) && (smallest > dictEnd))
-		smallest = dictEnd;
-
-	LZ4_renormDictT(streamPtr, smallest);
-
-	if (acceleration < 1)
-		acceleration = LZ4_ACCELERATION_DEFAULT;
-
-	/* Check overlapping input/dictionary space */
-	{
-		const BYTE *sourceEnd = (const BYTE *) source + inputSize;
-
-		if ((sourceEnd > streamPtr->dictionary)
-			&& (sourceEnd < dictEnd)) {
-			streamPtr->dictSize = (U32)(dictEnd - sourceEnd);
-			if (streamPtr->dictSize > 64 * KB)
-				streamPtr->dictSize = 64 * KB;
-			if (streamPtr->dictSize < 4)
-				streamPtr->dictSize = 0;
-			streamPtr->dictionary = dictEnd - streamPtr->dictSize;
-		}
-	}
-
-	/* prefix mode : source data follows dictionary */
-	if (dictEnd == (const BYTE *)source) {
-		int result;
-
-		if ((streamPtr->dictSize < 64 * KB) &&
-			(streamPtr->dictSize < streamPtr->currentOffset)) {
-			result = LZ4_compress_generic(
-				streamPtr, source, dest, inputSize,
-				maxOutputSize, limitedOutput, byU32,
-				withPrefix64k, dictSmall, acceleration);
-		} else {
-			result = LZ4_compress_generic(
-				streamPtr, source, dest, inputSize,
-				maxOutputSize, limitedOutput, byU32,
-				withPrefix64k, noDictIssue, acceleration);
-		}
-		streamPtr->dictSize += (U32)inputSize;
-		streamPtr->currentOffset += (U32)inputSize;
-		return result;
-	}
-
-	/* external dictionary mode */
-	{
-		int result;
-
-		if ((streamPtr->dictSize < 64 * KB) &&
-			(streamPtr->dictSize < streamPtr->currentOffset)) {
-			result = LZ4_compress_generic(
-				streamPtr, source, dest, inputSize,
-				maxOutputSize, limitedOutput, byU32,
-				usingExtDict, dictSmall, acceleration);
-		} else {
-			result = LZ4_compress_generic(
-				streamPtr, source, dest, inputSize,
-				maxOutputSize, limitedOutput, byU32,
-				usingExtDict, noDictIssue, acceleration);
-		}
-		streamPtr->dictionary = (const BYTE *)source;
-		streamPtr->dictSize = (U32)inputSize;
-		streamPtr->currentOffset += (U32)inputSize;
-		return result;
-	}
+	return __lz4_compress_fast_continue(streamPtr, src, dst, srcSize,
+					    maxDstSize, acceleration);
 }
 EXPORT_SYMBOL(LZ4_compress_fast_continue);
 

-- 
2.34.1



_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

[PATCH v2 10/15] lib/lz4: switch the compressor to the vendored sources

[Michal Wilczynski]

Replace the forked compressor with thin entry points over the vendored
lz4.c. LZ4_compress_default(), LZ4_compress_fast() and
LZ4_compress_destSize() keep the trailing wrkmem argument and forward to
upstream's *_extState() functions; the rest are plain forwarders.

Upstream's LZ4_initStream() returns NULL for a misaligned workspace,
which LZ4_compress_fast_extState() dereferences and
LZ4_compress_destSize_extState() ignores. Check the alignment in the
entry points, warn once and fail the call. In-tree callers allocate with
kmalloc() or vmalloc() and are not affected.

Compressed output differs from the fork's on a small fraction of inputs
by a byte or two; both decode each other's output. Compressing 4K
blocks, upstream is 2-4% faster on kernel source and binaries and up to
9% slower on datagen.

LZ4_stream_t becomes an incomplete type, as upstream declares it, so
<linux/lz4.h> and upstream's lz4.h can share a translation unit. The
macros that laid out the forked structure go, and LZ4_MEM_COMPRESS keeps
its value, 16416. zram, the only caller using sizeof(LZ4_stream_t), now
allocates LZ4_MEM_COMPRESS bytes.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 drivers/block/zram/backend_lz4.c |   6 +-
 include/linux/lz4.h              |  47 +-
 lib/lz4/Makefile                 |  12 +-
 lib/lz4/lz4_compress.c           | 946 +++------------------------------------
 4 files changed, 80 insertions(+), 931 deletions(-)

diff --git a/drivers/block/zram/backend_lz4.c b/drivers/block/zram/backend_lz4.c
index 1e4ad31d39a63f45b3bddd3d8a467d8d7d095beb..50265e3ce256490994cb6dae32a45da376e408e0 100644
--- a/drivers/block/zram/backend_lz4.c
+++ b/drivers/block/zram/backend_lz4.c
@@ -42,7 +42,7 @@ static int lz4_setup_params(struct zcomp_params *params)
 	if (!params->dict || !params->dict_sz)
 		return 0;
 
-	dict_stream = kzalloc_obj(*dict_stream);
+	dict_stream = kzalloc(LZ4_MEM_COMPRESS, GFP_KERNEL);
 	if (!dict_stream)
 		return -ENOMEM;
 
@@ -88,7 +88,7 @@ static int lz4_create(struct zcomp_params *params, struct zcomp_ctx *ctx)
 		if (!zctx->dstrm)
 			goto error;
 
-		zctx->cstrm = kzalloc_obj(*zctx->cstrm);
+		zctx->cstrm = kzalloc(LZ4_MEM_COMPRESS, GFP_KERNEL);
 		if (!zctx->cstrm)
 			goto error;
 	}
@@ -112,7 +112,7 @@ static int lz4_compress(struct zcomp_params *params, struct zcomp_ctx *ctx,
 					zctx->mem);
 	} else {
 		/* Cstrm needs to be reset */
-		memcpy(zctx->cstrm, params->drv_data, sizeof(*zctx->cstrm));
+		memcpy(zctx->cstrm, params->drv_data, LZ4_MEM_COMPRESS);
 		ret = LZ4_compress_fast_continue(zctx->cstrm, req->src,
 						 req->dst, req->src_len,
 						 req->dst_len, params->level);
diff --git a/include/linux/lz4.h b/include/linux/lz4.h
index 0df69e2b1ad523c291e3a96ab5595df9239f7252..031a410b8e3878e9a24cc49af36cfcbba399d991 100644
--- a/include/linux/lz4.h
+++ b/include/linux/lz4.h
@@ -47,16 +47,6 @@
 /*-************************************************************************
  *	CONSTANTS
  **************************************************************************/
-/*
- * LZ4_MEMORY_USAGE :
- * Memory usage formula : N->2^N Bytes
- * (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
- * Increasing memory usage improves compression ratio
- * Reduced memory usage can improve speed, due to cache effect
- * Default value is 14, for 16KB, which nicely fits into Intel x86 L1 cache
- */
-#define LZ4_MEMORY_USAGE 14
-
 #define LZ4_MAX_INPUT_SIZE	0x7E000000 /* 2 113 929 216 bytes */
 #define LZ4_COMPRESSBOUND(isize)	(\
 	(unsigned int)(isize) > (unsigned int)LZ4_MAX_INPUT_SIZE \
@@ -64,9 +54,6 @@
 	: (isize) + ((isize)/255) + 16)
 
 #define LZ4_ACCELERATION_DEFAULT 1
-#define LZ4_HASHLOG	 (LZ4_MEMORY_USAGE-2)
-#define LZ4_HASHTABLESIZE (1 << LZ4_MEMORY_USAGE)
-#define LZ4_HASH_SIZE_U32 (1 << LZ4_HASHLOG)
 
 #define LZ4HC_MIN_CLEVEL			3
 #define LZ4HC_DEFAULT_CLEVEL			9
@@ -82,9 +69,6 @@
 /*-************************************************************************
  *	STREAMING CONSTANTS AND STRUCTURES
  **************************************************************************/
-#define LZ4_STREAMSIZE_U64 ((1 << (LZ4_MEMORY_USAGE - 3)) + 4)
-#define LZ4_STREAMSIZE	(LZ4_STREAMSIZE_U64 * sizeof(unsigned long long))
-
 #define LZ4_STREAMHCSIZE        262192
 #define LZ4_STREAMHCSIZE_SIZET (262192 / sizeof(size_t))
 
@@ -93,20 +77,10 @@
 	sizeof(unsigned long long))
 
 /*
- * LZ4_stream_t - information structure to track an LZ4 stream.
+ * LZ4_stream_t - an LZ4 stream.  Incomplete: lib/lz4 owns the layout.
+ * Allocate LZ4_MEM_COMPRESS bytes and cast, do not sizeof().
  */
-typedef struct {
-	uint32_t hashTable[LZ4_HASH_SIZE_U32];
-	uint32_t currentOffset;
-	uint32_t initCheck;
-	const uint8_t *dictionary;
-	uint8_t *bufferStart;
-	uint32_t dictSize;
-} LZ4_stream_t_internal;
-typedef union {
-	unsigned long long table[LZ4_STREAMSIZE_U64];
-	LZ4_stream_t_internal internal_donotuse;
-} LZ4_stream_t;
+typedef union LZ4_stream_u LZ4_stream_t;
 
 /*
  * LZ4_streamHC_t - information structure to track an LZ4HC stream.
@@ -153,7 +127,12 @@ typedef union {
 /*-************************************************************************
  *	SIZE OF STATE
  **************************************************************************/
-#define LZ4_MEM_COMPRESS	LZ4_STREAMSIZE
+/*
+ * Working memory for the compressors.  It must be aligned to at least 8
+ * bytes; anything from kmalloc() or vmalloc() already is.  The entry
+ * points warn about a misaligned buffer and fail the call.
+ */
+#define LZ4_MEM_COMPRESS	16416
 #define LZ4HC_MEM_COMPRESS	LZ4_STREAMHCSIZE
 
 /*-************************************************************************
@@ -180,7 +159,7 @@ static inline int LZ4_compressBound(size_t isize)
  * @maxOutputSize: full or partial size of buffer 'dest'
  *	which must be already allocated
  * @wrkmem: address of the working memory.
- *	This requires 'workmem' of LZ4_MEM_COMPRESS.
+ *	This requires 'workmem' of LZ4_MEM_COMPRESS, aligned to 8 bytes.
  *
  * Compresses 'sourceSize' bytes from buffer 'source'
  * into already allocated 'dest' buffer of size 'maxOutputSize'.
@@ -206,7 +185,7 @@ int LZ4_compress_default(const char *source, char *dest, int inputSize,
  *	which must be already allocated
  * @acceleration: acceleration factor
  * @wrkmem: address of the working memory.
- *	This requires 'workmem' of LZ4_MEM_COMPRESS.
+ *	This requires 'workmem' of LZ4_MEM_COMPRESS, aligned to 8 bytes.
  *
  * Same as LZ4_compress_default(), but allows to select an "acceleration"
  * factor. The larger the acceleration value, the faster the algorithm,
@@ -230,7 +209,7 @@ int LZ4_compress_fast(const char *source, char *dest, int inputSize,
  *	from 'source' to fill 'dest'. New value is necessarily <= old value.
  * @targetDestSize: Size of buffer 'dest' which must be already allocated
  * @wrkmem: address of the working memory.
- *	This requires 'workmem' of LZ4_MEM_COMPRESS.
+ *	This requires 'workmem' of LZ4_MEM_COMPRESS, aligned to 8 bytes.
  *
  * Reverse the logic, by compressing as much data as possible
  * from 'source' buffer into already allocated buffer 'dest'
@@ -314,7 +293,7 @@ int LZ4_decompress_safe_partial(const char *source, char *dest,
  *	value between 1 and LZ4HC_MAX_CLEVEL will work.
  *	Values >LZ4HC_MAX_CLEVEL behave the same as 16.
  * @wrkmem: address of the working memory.
- *	This requires 'wrkmem' of size LZ4HC_MEM_COMPRESS.
+ *	This requires 'wrkmem' of size LZ4HC_MEM_COMPRESS, aligned to 8 bytes.
  *
  * Compress data from 'src' into 'dst', using the more powerful
  * but slower "HC" algorithm. Compression is guaranteed to succeed if
diff --git a/lib/lz4/Makefile b/lib/lz4/Makefile
index 5b42242afaa20deb165d8a2f0b1c0c842dfcdc39..8b4eb07694e67dd99ec7cc85212fb68138d3981e 100644
--- a/lib/lz4/Makefile
+++ b/lib/lz4/Makefile
@@ -1,6 +1,12 @@
 # SPDX-License-Identifier: GPL-2.0-only
 ccflags-y += -O3
 
-obj-$(CONFIG_LZ4_COMPRESS) += lz4_compress.o
-obj-$(CONFIG_LZ4HC_COMPRESS) += lz4hc_compress.o
-obj-$(CONFIG_LZ4_DECOMPRESS) += lz4_decompress.o
+# upstream/ holds the upstream LZ4 sources lz4_deps.h describes; replace them
+# with a later release rather than patch them further.  The three entry point
+# files include them.  freestanding/ supplies the ISO C headers upstream
+# includes that -nostdinc does not provide.
+ccflags-y += -I $(src)/freestanding
+
+obj-$(CONFIG_LZ4_COMPRESS)	+= lz4_compress.o
+obj-$(CONFIG_LZ4HC_COMPRESS)	+= lz4hc_compress.o
+obj-$(CONFIG_LZ4_DECOMPRESS)	+= lz4_decompress.o
diff --git a/lib/lz4/lz4_compress.c b/lib/lz4/lz4_compress.c
index 2a397bb2c661d9c2e4a7a67d5f50f06bfd42f1f2..941e32a19ac972aef543c9a03b447a8a98bf5215 100644
--- a/lib/lz4/lz4_compress.c
+++ b/lib/lz4/lz4_compress.c
@@ -1,937 +1,101 @@
+// SPDX-License-Identifier: GPL-2.0-only OR BSD-2-Clause
 /*
- * LZ4 - Fast LZ compression algorithm
  * Copyright (C) 2011 - 2016, Yann Collet.
- * BSD 2 - Clause License (http://www.opensource.org/licenses/bsd - license.php)
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions are
- * met:
- *	* Redistributions of source code must retain the above copyright
- *	  notice, this list of conditions and the following disclaimer.
- *	* Redistributions in binary form must reproduce the above
- * copyright notice, this list of conditions and the following disclaimer
- * in the documentation and/or other materials provided with the
- * distribution.
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
- * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
- * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
- * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
- * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
- * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
- * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
- * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
- * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- * You can contact the author at :
- *	- LZ4 homepage : http://www.lz4.org
- *	- LZ4 source repository : https://github.com/lz4/lz4
+ * Copyright (C) 2016, Sven Schmidt <4sschmid@informatik.uni-hamburg.de>
+ * Copyright (c) 2026 Samsung Electronics Co., Ltd.
+ * Author: Michal Wilczynski <m.wilczynski@samsung.com>
  *
- *	Changed for kernel usage by:
- *	Sven Schmidt <4sschmid@informatik.uni-hamburg.de>
+ * LZ4 compressor -- kernel entry points
+ *
+ * The compressor is upstream's, in the verbatim lz4.c included below.  The
+ * kernel has no heap at these call sites, so LZ4_compress_default(),
+ * LZ4_compress_fast() and LZ4_compress_destSize() take a trailing wrkmem
+ * argument and forward to upstream's *_extState() entry points.
  */
 
-/*-************************************
- *	Dependencies
- **************************************/
-#include "lz4defs.h"
-#include <linux/module.h>
-#include <linux/kernel.h>
-#include <linux/unaligned.h>
-
-static const int LZ4_minLength = (MFLIMIT + 1);
-static const int LZ4_64Klimit = ((64 * KB) + (MFLIMIT - 1));
-
-/*-******************************
- *	Compression functions
- ********************************/
-static FORCE_INLINE U32 LZ4_hash4(
-	U32 sequence,
-	tableType_t const tableType)
-{
-	if (tableType == byU16)
-		return ((sequence * 2654435761U)
-			>> ((MINMATCH * 8) - (LZ4_HASHLOG + 1)));
-	else
-		return ((sequence * 2654435761U)
-			>> ((MINMATCH * 8) - LZ4_HASHLOG));
-}
-
-static FORCE_INLINE U32 LZ4_hash5(
-	U64 sequence,
-	tableType_t const tableType)
-{
-	const U32 hashLog = (tableType == byU16)
-		? LZ4_HASHLOG + 1
-		: LZ4_HASHLOG;
-
-#if LZ4_LITTLE_ENDIAN
-	static const U64 prime5bytes = 889523592379ULL;
-
-	return (U32)(((sequence << 24) * prime5bytes) >> (64 - hashLog));
-#else
-	static const U64 prime8bytes = 11400714785074694791ULL;
-
-	return (U32)(((sequence >> 24) * prime8bytes) >> (64 - hashLog));
-#endif
-}
-
-static FORCE_INLINE U32 LZ4_hashPosition(
-	const void *p,
-	tableType_t const tableType)
-{
-#if LZ4_ARCH64
-	if (tableType == byU32)
-		return LZ4_hash5(LZ4_read_ARCH(p), tableType);
-#endif
-
-	return LZ4_hash4(LZ4_read32(p), tableType);
-}
-
-static void LZ4_putPositionOnHash(
-	const BYTE *p,
-	U32 h,
-	void *tableBase,
-	tableType_t const tableType,
-	const BYTE *srcBase)
-{
-	switch (tableType) {
-	case byPtr:
-	{
-		const BYTE **hashTable = (const BYTE **)tableBase;
-
-		hashTable[h] = p;
-		return;
-	}
-	case byU32:
-	{
-		U32 *hashTable = (U32 *) tableBase;
-
-		hashTable[h] = (U32)(p - srcBase);
-		return;
-	}
-	case byU16:
-	{
-		U16 *hashTable = (U16 *) tableBase;
-
-		hashTable[h] = (U16)(p - srcBase);
-		return;
-	}
-	}
-}
-
-static FORCE_INLINE void LZ4_putPosition(
-	const BYTE *p,
-	void *tableBase,
-	tableType_t tableType,
-	const BYTE *srcBase)
-{
-	U32 const h = LZ4_hashPosition(p, tableType);
+#include "lz4_deps.h"
+#include "upstream/lz4.c"
 
-	LZ4_putPositionOnHash(p, h, tableBase, tableType, srcBase);
-}
-
-static const BYTE *LZ4_getPositionOnHash(
-	U32 h,
-	void *tableBase,
-	tableType_t tableType,
-	const BYTE *srcBase)
-{
-	if (tableType == byPtr) {
-		const BYTE **hashTable = (const BYTE **) tableBase;
+/* Upstream's short helper macros; MIN and MAX clash with <linux/minmax.h>. */
+#undef MIN
+#undef MAX
+#undef KB
+#undef MB
+#undef GB
 
-		return hashTable[h];
-	}
-
-	if (tableType == byU32) {
-		const U32 * const hashTable = (U32 *) tableBase;
-
-		return hashTable[h] + srcBase;
-	}
-
-	{
-		/* default, to ensure a return */
-		const U16 * const hashTable = (U16 *) tableBase;
-
-		return hashTable[h] + srcBase;
-	}
-}
-
-static FORCE_INLINE const BYTE *LZ4_getPosition(
-	const BYTE *p,
-	void *tableBase,
-	tableType_t tableType,
-	const BYTE *srcBase)
-{
-	U32 const h = LZ4_hashPosition(p, tableType);
-
-	return LZ4_getPositionOnHash(h, tableBase, tableType, srcBase);
-}
+#include "lz4_kernel_api.h"
+#include <linux/bug.h>
+#include <linux/export.h>
+#include <linux/module.h>
 
+/* Catch any divergence from upstream's layout at build time. */
+static_assert(sizeof(LZ4_stream_t) == LZ4_STREAM_MINSIZE);
+static_assert(LZ4_STREAM_MINSIZE == 16416); /* LZ4_MEM_COMPRESS */
 
-/*
- * LZ4_compress_generic() :
- * inlined, to ensure branches are decided at compilation time
+/* LZ4_initStream() returns NULL for a misaligned state, which
+ * LZ4_compress_fast_extState() dereferences and
+ * LZ4_compress_destSize_extState() ignores, behind assert()s the kernel
+ * compiles out.  Fail the call instead.
  */
-static FORCE_INLINE int LZ4_compress_generic(
-	LZ4_stream_t_internal * const dictPtr,
-	const char * const source,
-	char * const dest,
-	const int inputSize,
-	const int maxOutputSize,
-	const limitedOutput_directive outputLimited,
-	const tableType_t tableType,
-	const dict_directive dict,
-	const dictIssue_directive dictIssue,
-	const U32 acceleration)
+static bool lz4_wrkmem_misaligned(const void *wrkmem)
 {
-	const BYTE *ip = (const BYTE *) source;
-	const BYTE *base;
-	const BYTE *lowLimit;
-	const BYTE * const lowRefLimit = ip - dictPtr->dictSize;
-	const BYTE * const dictionary = dictPtr->dictionary;
-	const BYTE * const dictEnd = dictionary + dictPtr->dictSize;
-	const size_t dictDelta = dictEnd - (const BYTE *)source;
-	const BYTE *anchor = (const BYTE *) source;
-	const BYTE * const iend = ip + inputSize;
-	const BYTE * const mflimit = iend - MFLIMIT;
-	const BYTE * const matchlimit = iend - LASTLITERALS;
-
-	BYTE *op = (BYTE *) dest;
-	BYTE * const olimit = op + maxOutputSize;
-
-	U32 forwardH;
-	size_t refDelta = 0;
-
-	/* Init conditions */
-	if ((U32)inputSize > (U32)LZ4_MAX_INPUT_SIZE) {
-		/* Unsupported inputSize, too large (or negative) */
-		return 0;
-	}
-
-	switch (dict) {
-	case noDict:
-	default:
-		base = (const BYTE *)source;
-		lowLimit = (const BYTE *)source;
-		break;
-	case withPrefix64k:
-		base = (const BYTE *)source - dictPtr->currentOffset;
-		lowLimit = (const BYTE *)source - dictPtr->dictSize;
-		break;
-	case usingExtDict:
-		base = (const BYTE *)source - dictPtr->currentOffset;
-		lowLimit = (const BYTE *)source;
-		break;
-	}
-
-	if ((tableType == byU16)
-		&& (inputSize >= LZ4_64Klimit)) {
-		/* Size too large (not within 64K limit) */
-		return 0;
-	}
-
-	if (inputSize < LZ4_minLength) {
-		/* Input too small, no compression (all literals) */
-		goto _last_literals;
-	}
-
-	/* First Byte */
-	LZ4_putPosition(ip, dictPtr->hashTable, tableType, base);
-	ip++;
-	forwardH = LZ4_hashPosition(ip, tableType);
-
-	/* Main Loop */
-	for ( ; ; ) {
-		const BYTE *match;
-		BYTE *token;
-
-		/* Find a match */
-		{
-			const BYTE *forwardIp = ip;
-			unsigned int step = 1;
-			unsigned int searchMatchNb = acceleration << LZ4_SKIPTRIGGER;
-
-			do {
-				U32 const h = forwardH;
-
-				ip = forwardIp;
-				forwardIp += step;
-				step = (searchMatchNb++ >> LZ4_SKIPTRIGGER);
-
-				if (unlikely(forwardIp > mflimit))
-					goto _last_literals;
-
-				match = LZ4_getPositionOnHash(h,
-					dictPtr->hashTable,
-					tableType, base);
-
-				if (dict == usingExtDict) {
-					if (match < (const BYTE *)source) {
-						refDelta = dictDelta;
-						lowLimit = dictionary;
-					} else {
-						refDelta = 0;
-						lowLimit = (const BYTE *)source;
-				}	 }
-
-				forwardH = LZ4_hashPosition(forwardIp,
-					tableType);
-
-				LZ4_putPositionOnHash(ip, h, dictPtr->hashTable,
-					tableType, base);
-			} while (((dictIssue == dictSmall)
-					? (match < lowRefLimit)
-					: 0)
-				|| ((tableType == byU16)
-					? 0
-					: (match + MAX_DISTANCE < ip))
-				|| (LZ4_read32(match + refDelta)
-					!= LZ4_read32(ip)));
-		}
-
-		/* Catch up */
-		while (((ip > anchor) & (match + refDelta > lowLimit))
-				&& (unlikely(ip[-1] == match[refDelta - 1]))) {
-			ip--;
-			match--;
-		}
-
-		/* Encode Literals */
-		{
-			unsigned const int litLength = (unsigned int)(ip - anchor);
-
-			token = op++;
-
-			if ((outputLimited) &&
-				/* Check output buffer overflow */
-				(unlikely(op + litLength +
-					(2 + 1 + LASTLITERALS) +
-					(litLength / 255) > olimit)))
-				return 0;
-
-			if (litLength >= RUN_MASK) {
-				int len = (int)litLength - RUN_MASK;
-
-				*token = (RUN_MASK << ML_BITS);
-
-				for (; len >= 255; len -= 255)
-					*op++ = 255;
-				*op++ = (BYTE)len;
-			} else
-				*token = (BYTE)(litLength << ML_BITS);
-
-			/* Copy Literals */
-			LZ4_wildCopy(op, anchor, op + litLength);
-			op += litLength;
-		}
-
-_next_match:
-		/* Encode Offset */
-		LZ4_writeLE16(op, (U16)(ip - match));
-		op += 2;
-
-		/* Encode MatchLength */
-		{
-			unsigned int matchCode;
-
-			if ((dict == usingExtDict)
-				&& (lowLimit == dictionary)) {
-				const BYTE *limit;
-
-				match += refDelta;
-				limit = ip + (dictEnd - match);
-
-				if (limit > matchlimit)
-					limit = matchlimit;
-
-				matchCode = LZ4_count(ip + MINMATCH,
-					match + MINMATCH, limit);
-
-				ip += MINMATCH + matchCode;
-
-				if (ip == limit) {
-					unsigned const int more = LZ4_count(ip,
-						(const BYTE *)source,
-						matchlimit);
-
-					matchCode += more;
-					ip += more;
-				}
-			} else {
-				matchCode = LZ4_count(ip + MINMATCH,
-					match + MINMATCH, matchlimit);
-				ip += MINMATCH + matchCode;
-			}
-
-			if (outputLimited &&
-				/* Check output buffer overflow */
-				(unlikely(op +
-					(1 + LASTLITERALS) +
-					(matchCode >> 8) > olimit)))
-				return 0;
-
-			if (matchCode >= ML_MASK) {
-				*token += ML_MASK;
-				matchCode -= ML_MASK;
-				LZ4_write32(op, 0xFFFFFFFF);
-
-				while (matchCode >= 4 * 255) {
-					op += 4;
-					LZ4_write32(op, 0xFFFFFFFF);
-					matchCode -= 4 * 255;
-				}
-
-				op += matchCode / 255;
-				*op++ = (BYTE)(matchCode % 255);
-			} else
-				*token += (BYTE)(matchCode);
-		}
-
-		anchor = ip;
-
-		/* Test end of chunk */
-		if (ip > mflimit)
-			break;
-
-		/* Fill table */
-		LZ4_putPosition(ip - 2, dictPtr->hashTable, tableType, base);
-
-		/* Test next position */
-		match = LZ4_getPosition(ip, dictPtr->hashTable,
-			tableType, base);
-
-		if (dict == usingExtDict) {
-			if (match < (const BYTE *)source) {
-				refDelta = dictDelta;
-				lowLimit = dictionary;
-			} else {
-				refDelta = 0;
-				lowLimit = (const BYTE *)source;
-			}
-		}
-
-		LZ4_putPosition(ip, dictPtr->hashTable, tableType, base);
-
-		if (((dictIssue == dictSmall) ? (match >= lowRefLimit) : 1)
-			&& (match + MAX_DISTANCE >= ip)
-			&& (LZ4_read32(match + refDelta) == LZ4_read32(ip))) {
-			token = op++;
-			*token = 0;
-			goto _next_match;
-		}
-
-		/* Prepare next loop */
-		forwardH = LZ4_hashPosition(++ip, tableType);
-	}
-
-_last_literals:
-	/* Encode Last Literals */
-	{
-		size_t const lastRun = (size_t)(iend - anchor);
-
-		if ((outputLimited) &&
-			/* Check output buffer overflow */
-			((op - (BYTE *)dest) + lastRun + 1 +
-			((lastRun + 255 - RUN_MASK) / 255) > (U32)maxOutputSize))
-			return 0;
-
-		if (lastRun >= RUN_MASK) {
-			size_t accumulator = lastRun - RUN_MASK;
-			*op++ = RUN_MASK << ML_BITS;
-			for (; accumulator >= 255; accumulator -= 255)
-				*op++ = 255;
-			*op++ = (BYTE) accumulator;
-		} else {
-			*op++ = (BYTE)(lastRun << ML_BITS);
-		}
-
-		LZ4_memcpy(op, anchor, lastRun);
-
-		op += lastRun;
-	}
-
-	/* End */
-	return (int) (((char *)op) - dest);
-}
-
-static int LZ4_compress_fast_extState(
-	void *state,
-	const char *source,
-	char *dest,
-	int inputSize,
-	int maxOutputSize,
-	int acceleration)
-{
-	LZ4_stream_t_internal *ctx = &((LZ4_stream_t *)state)->internal_donotuse;
-#if LZ4_ARCH64
-	const tableType_t tableType = byU32;
-#else
-	const tableType_t tableType = byPtr;
-#endif
-
-	LZ4_resetStream((LZ4_stream_t *)state);
-
-	if (acceleration < 1)
-		acceleration = LZ4_ACCELERATION_DEFAULT;
-
-	if (maxOutputSize >= LZ4_COMPRESSBOUND(inputSize)) {
-		if (inputSize < LZ4_64Klimit)
-			return LZ4_compress_generic(ctx, source,
-				dest, inputSize, 0,
-				noLimit, byU16, noDict,
-				noDictIssue, acceleration);
-		else
-			return LZ4_compress_generic(ctx, source,
-				dest, inputSize, 0,
-				noLimit, tableType, noDict,
-				noDictIssue, acceleration);
-	} else {
-		if (inputSize < LZ4_64Klimit)
-			return LZ4_compress_generic(ctx, source,
-				dest, inputSize,
-				maxOutputSize, limitedOutput, byU16, noDict,
-				noDictIssue, acceleration);
-		else
-			return LZ4_compress_generic(ctx, source,
-				dest, inputSize,
-				maxOutputSize, limitedOutput, tableType, noDict,
-				noDictIssue, acceleration);
-	}
+	return WARN_ON_ONCE(!LZ4_isAligned(wrkmem, LZ4_stream_t_alignment()));
 }
 
 int LZ4_compress_fast(const char *source, char *dest, int inputSize,
-	int maxOutputSize, int acceleration, void *wrkmem)
+		      int maxOutputSize, int acceleration, void *wrkmem)
 {
+	if (lz4_wrkmem_misaligned(wrkmem))
+		return 0;
+
 	return LZ4_compress_fast_extState(wrkmem, source, dest, inputSize,
-		maxOutputSize, acceleration);
+					  maxOutputSize, acceleration);
 }
 EXPORT_SYMBOL(LZ4_compress_fast);
 
 int LZ4_compress_default(const char *source, char *dest, int inputSize,
-	int maxOutputSize, void *wrkmem)
+			 int maxOutputSize, void *wrkmem)
 {
-	return LZ4_compress_fast(source, dest, inputSize,
-		maxOutputSize, LZ4_ACCELERATION_DEFAULT, wrkmem);
+	return LZ4_compress_fast(source, dest, inputSize, maxOutputSize,
+				 LZ4_ACCELERATION_DEFAULT, wrkmem);
 }
 EXPORT_SYMBOL(LZ4_compress_default);
 
-/*-******************************
- *	*_destSize() variant
- ********************************/
-static int LZ4_compress_destSize_generic(
-	LZ4_stream_t_internal * const ctx,
-	const char * const src,
-	char * const dst,
-	int * const srcSizePtr,
-	const int targetDstSize,
-	const tableType_t tableType)
+int LZ4_compress_destSize(const char *source, char *dest, int *sourceSizePtr,
+			  int targetDestSize, void *wrkmem)
 {
-	const BYTE *ip = (const BYTE *) src;
-	const BYTE *base = (const BYTE *) src;
-	const BYTE *lowLimit = (const BYTE *) src;
-	const BYTE *anchor = ip;
-	const BYTE * const iend = ip + *srcSizePtr;
-	const BYTE * const mflimit = iend - MFLIMIT;
-	const BYTE * const matchlimit = iend - LASTLITERALS;
-
-	BYTE *op = (BYTE *) dst;
-	BYTE * const oend = op + targetDstSize;
-	BYTE * const oMaxLit = op + targetDstSize - 2 /* offset */
-		- 8 /* because 8 + MINMATCH == MFLIMIT */ - 1 /* token */;
-	BYTE * const oMaxMatch = op + targetDstSize
-		- (LASTLITERALS + 1 /* token */);
-	BYTE * const oMaxSeq = oMaxLit - 1 /* token */;
-
-	U32 forwardH;
-
-	/* Init conditions */
-	/* Impossible to store anything */
-	if (targetDstSize < 1)
-		return 0;
-	/* Unsupported input size, too large (or negative) */
-	if ((U32)*srcSizePtr > (U32)LZ4_MAX_INPUT_SIZE)
-		return 0;
-	/* Size too large (not within 64K limit) */
-	if ((tableType == byU16) && (*srcSizePtr >= LZ4_64Klimit))
+	if (lz4_wrkmem_misaligned(wrkmem))
 		return 0;
-	/* Input too small, no compression (all literals) */
-	if (*srcSizePtr < LZ4_minLength)
-		goto _last_literals;
-
-	/* First Byte */
-	*srcSizePtr = 0;
-	LZ4_putPosition(ip, ctx->hashTable, tableType, base);
-	ip++; forwardH = LZ4_hashPosition(ip, tableType);
-
-	/* Main Loop */
-	for ( ; ; ) {
-		const BYTE *match;
-		BYTE *token;
-
-		/* Find a match */
-		{
-			const BYTE *forwardIp = ip;
-			unsigned int step = 1;
-			unsigned int searchMatchNb = 1 << LZ4_SKIPTRIGGER;
-
-			do {
-				U32 h = forwardH;
-
-				ip = forwardIp;
-				forwardIp += step;
-				step = (searchMatchNb++ >> LZ4_SKIPTRIGGER);
-
-				if (unlikely(forwardIp > mflimit))
-					goto _last_literals;
-
-				match = LZ4_getPositionOnHash(h, ctx->hashTable,
-					tableType, base);
-				forwardH = LZ4_hashPosition(forwardIp,
-					tableType);
-				LZ4_putPositionOnHash(ip, h,
-					ctx->hashTable, tableType,
-					base);
-
-			} while (((tableType == byU16)
-				? 0
-				: (match + MAX_DISTANCE < ip))
-				|| (LZ4_read32(match) != LZ4_read32(ip)));
-		}
-
-		/* Catch up */
-		while ((ip > anchor)
-			&& (match > lowLimit)
-			&& (unlikely(ip[-1] == match[-1]))) {
-			ip--;
-			match--;
-		}
-
-		/* Encode Literal length */
-		{
-			unsigned int litLength = (unsigned int)(ip - anchor);
-
-			token = op++;
-			if (op + ((litLength + 240) / 255)
-				+ litLength > oMaxLit) {
-				/* Not enough space for a last match */
-				op--;
-				goto _last_literals;
-			}
-			if (litLength >= RUN_MASK) {
-				unsigned int len = litLength - RUN_MASK;
-				*token = (RUN_MASK<<ML_BITS);
-				for (; len >= 255; len -= 255)
-					*op++ = 255;
-				*op++ = (BYTE)len;
-			} else
-				*token = (BYTE)(litLength << ML_BITS);
-
-			/* Copy Literals */
-			LZ4_wildCopy(op, anchor, op + litLength);
-			op += litLength;
-		}
-
-_next_match:
-		/* Encode Offset */
-		LZ4_writeLE16(op, (U16)(ip - match)); op += 2;
-
-		/* Encode MatchLength */
-		{
-			size_t matchLength = LZ4_count(ip + MINMATCH,
-			match + MINMATCH, matchlimit);
-
-			if (op + ((matchLength + 240)/255) > oMaxMatch) {
-				/* Match description too long : reduce it */
-				matchLength = (15 - 1) + (oMaxMatch - op) * 255;
-			}
-			ip += MINMATCH + matchLength;
-
-			if (matchLength >= ML_MASK) {
-				*token += ML_MASK;
-				matchLength -= ML_MASK;
-				while (matchLength >= 255) {
-					matchLength -= 255;
-					*op++ = 255;
-				}
-				*op++ = (BYTE)matchLength;
-			} else
-				*token += (BYTE)(matchLength);
-		}
-
-		anchor = ip;
-
-		/* Test end of block */
-		if (ip > mflimit)
-			break;
-		if (op > oMaxSeq)
-			break;
-
-		/* Fill table */
-		LZ4_putPosition(ip - 2, ctx->hashTable, tableType, base);
-
-		/* Test next position */
-		match = LZ4_getPosition(ip, ctx->hashTable, tableType, base);
-		LZ4_putPosition(ip, ctx->hashTable, tableType, base);
 
-		if ((match + MAX_DISTANCE >= ip)
-			&& (LZ4_read32(match) == LZ4_read32(ip))) {
-			token = op++; *token = 0;
-			goto _next_match;
-		}
-
-		/* Prepare next loop */
-		forwardH = LZ4_hashPosition(++ip, tableType);
-	}
-
-_last_literals:
-	/* Encode Last Literals */
-	{
-		size_t lastRunSize = (size_t)(iend - anchor);
-
-		if (op + 1 /* token */
-			+ ((lastRunSize + 240) / 255) /* litLength */
-			+ lastRunSize /* literals */ > oend) {
-			/* adapt lastRunSize to fill 'dst' */
-			lastRunSize	= (oend - op) - 1;
-			lastRunSize -= (lastRunSize + 240) / 255;
-		}
-		ip = anchor + lastRunSize;
-
-		if (lastRunSize >= RUN_MASK) {
-			size_t accumulator = lastRunSize - RUN_MASK;
-
-			*op++ = RUN_MASK << ML_BITS;
-			for (; accumulator >= 255; accumulator -= 255)
-				*op++ = 255;
-			*op++ = (BYTE) accumulator;
-		} else {
-			*op++ = (BYTE)(lastRunSize<<ML_BITS);
-		}
-		LZ4_memcpy(op, anchor, lastRunSize);
-		op += lastRunSize;
-	}
-
-	/* End */
-	*srcSizePtr = (int) (((const char *)ip) - src);
-	return (int) (((char *)op) - dst);
-}
-
-static int LZ4_compress_destSize_extState(
-	LZ4_stream_t *state,
-	const char *src,
-	char *dst,
-	int *srcSizePtr,
-	int targetDstSize)
-{
-#if LZ4_ARCH64
-	const tableType_t tableType = byU32;
-#else
-	const tableType_t tableType = byPtr;
-#endif
-
-	LZ4_resetStream(state);
-
-	if (targetDstSize >= LZ4_COMPRESSBOUND(*srcSizePtr)) {
-		/* compression success is guaranteed */
-		return LZ4_compress_fast_extState(
-			state, src, dst, *srcSizePtr,
-			targetDstSize, 1);
-	} else {
-		if (*srcSizePtr < LZ4_64Klimit)
-			return LZ4_compress_destSize_generic(
-				&state->internal_donotuse,
-				src, dst, srcSizePtr,
-				targetDstSize, byU16);
-		else
-			return LZ4_compress_destSize_generic(
-				&state->internal_donotuse,
-				src, dst, srcSizePtr,
-				targetDstSize, tableType);
-	}
-}
-
-
-int LZ4_compress_destSize(
-	const char *src,
-	char *dst,
-	int *srcSizePtr,
-	int targetDstSize,
-	void *wrkmem)
-{
-	return LZ4_compress_destSize_extState(wrkmem, src, dst, srcSizePtr,
-		targetDstSize);
+	return LZ4_compress_destSize_extState(wrkmem, source, dest,
+					      sourceSizePtr, targetDestSize,
+					      LZ4_ACCELERATION_DEFAULT);
 }
 EXPORT_SYMBOL(LZ4_compress_destSize);
 
-/*-******************************
- *	Streaming functions
- ********************************/
 void LZ4_resetStream(LZ4_stream_t *LZ4_stream)
 {
-	memset(LZ4_stream, 0, sizeof(LZ4_stream_t));
+	__lz4_resetStream(LZ4_stream);
 }
 
-int LZ4_loadDict(LZ4_stream_t *LZ4_dict,
-	const char *dictionary, int dictSize)
+int LZ4_loadDict(LZ4_stream_t *streamPtr, const char *dictionary, int dictSize)
 {
-	LZ4_stream_t_internal *dict = &LZ4_dict->internal_donotuse;
-	const BYTE *p = (const BYTE *)dictionary;
-	const BYTE * const dictEnd = p + dictSize;
-	const BYTE *base;
-
-	if ((dict->initCheck)
-		|| (dict->currentOffset > 1 * GB)) {
-		/* Uninitialized structure, or reuse overflow */
-		LZ4_resetStream(LZ4_dict);
-	}
-
-	if (dictSize < (int)HASH_UNIT) {
-		dict->dictionary = NULL;
-		dict->dictSize = 0;
-		return 0;
-	}
-
-	if ((dictEnd - p) > 64 * KB)
-		p = dictEnd - 64 * KB;
-	dict->currentOffset += 64 * KB;
-	base = p - dict->currentOffset;
-	dict->dictionary = p;
-	dict->dictSize = (U32)(dictEnd - p);
-	dict->currentOffset += dict->dictSize;
-
-	while (p <= dictEnd - HASH_UNIT) {
-		LZ4_putPosition(p, dict->hashTable, byU32, base);
-		p += 3;
-	}
-
-	return dict->dictSize;
+	return __lz4_loadDict(streamPtr, dictionary, dictSize);
 }
 EXPORT_SYMBOL(LZ4_loadDict);
 
-static void LZ4_renormDictT(LZ4_stream_t_internal *LZ4_dict,
-	const BYTE *src)
+int LZ4_saveDict(LZ4_stream_t *streamPtr, char *safeBuffer, int dictSize)
 {
-	if ((LZ4_dict->currentOffset > 0x80000000) ||
-		((uptrval)LZ4_dict->currentOffset > (uptrval)src)) {
-		/* address space overflow */
-		/* rescale hash table */
-		U32 const delta = LZ4_dict->currentOffset - 64 * KB;
-		const BYTE *dictEnd = LZ4_dict->dictionary + LZ4_dict->dictSize;
-		int i;
-
-		for (i = 0; i < LZ4_HASH_SIZE_U32; i++) {
-			if (LZ4_dict->hashTable[i] < delta)
-				LZ4_dict->hashTable[i] = 0;
-			else
-				LZ4_dict->hashTable[i] -= delta;
-		}
-		LZ4_dict->currentOffset = 64 * KB;
-		if (LZ4_dict->dictSize > 64 * KB)
-			LZ4_dict->dictSize = 64 * KB;
-		LZ4_dict->dictionary = dictEnd - LZ4_dict->dictSize;
-	}
-}
-
-int LZ4_saveDict(LZ4_stream_t *LZ4_dict, char *safeBuffer, int dictSize)
-{
-	LZ4_stream_t_internal * const dict = &LZ4_dict->internal_donotuse;
-	const BYTE * const previousDictEnd = dict->dictionary + dict->dictSize;
-
-	if ((U32)dictSize > 64 * KB) {
-		/* useless to define a dictionary > 64 * KB */
-		dictSize = 64 * KB;
-	}
-	if ((U32)dictSize > dict->dictSize)
-		dictSize = dict->dictSize;
-
-	memmove(safeBuffer, previousDictEnd - dictSize, dictSize);
-
-	dict->dictionary = (const BYTE *)safeBuffer;
-	dict->dictSize = (U32)dictSize;
-
-	return dictSize;
+	return __lz4_saveDict(streamPtr, safeBuffer, dictSize);
 }
 EXPORT_SYMBOL(LZ4_saveDict);
 
-int LZ4_compress_fast_continue(LZ4_stream_t *LZ4_stream, const char *source,
-	char *dest, int inputSize, int maxOutputSize, int acceleration)
+int LZ4_compress_fast_continue(LZ4_stream_t *streamPtr, const char *src,
+			       char *dst, int srcSize, int maxDstSize,
+			       int acceleration)
 {
-	LZ4_stream_t_internal *streamPtr = &LZ4_stream->internal_donotuse;
-	const BYTE * const dictEnd = streamPtr->dictionary
-		+ streamPtr->dictSize;
-
-	const BYTE *smallest = (const BYTE *) source;
-
-	if (streamPtr->initCheck) {
-		/* Uninitialized structure detected */
-		return 0;
-	}
-
-	if ((streamPtr->dictSize > 0) && (smallest > dictEnd))
-		smallest = dictEnd;
-
-	LZ4_renormDictT(streamPtr, smallest);
-
-	if (acceleration < 1)
-		acceleration = LZ4_ACCELERATION_DEFAULT;
-
-	/* Check overlapping input/dictionary space */
-	{
-		const BYTE *sourceEnd = (const BYTE *) source + inputSize;
-
-		if ((sourceEnd > streamPtr->dictionary)
-			&& (sourceEnd < dictEnd)) {
-			streamPtr->dictSize = (U32)(dictEnd - sourceEnd);
-			if (streamPtr->dictSize > 64 * KB)
-				streamPtr->dictSize = 64 * KB;
-			if (streamPtr->dictSize < 4)
-				streamPtr->dictSize = 0;
-			streamPtr->dictionary = dictEnd - streamPtr->dictSize;
-		}
-	}
-
-	/* prefix mode : source data follows dictionary */
-	if (dictEnd == (const BYTE *)source) {
-		int result;
-
-		if ((streamPtr->dictSize < 64 * KB) &&
-			(streamPtr->dictSize < streamPtr->currentOffset)) {
-			result = LZ4_compress_generic(
-				streamPtr, source, dest, inputSize,
-				maxOutputSize, limitedOutput, byU32,
-				withPrefix64k, dictSmall, acceleration);
-		} else {
-			result = LZ4_compress_generic(
-				streamPtr, source, dest, inputSize,
-				maxOutputSize, limitedOutput, byU32,
-				withPrefix64k, noDictIssue, acceleration);
-		}
-		streamPtr->dictSize += (U32)inputSize;
-		streamPtr->currentOffset += (U32)inputSize;
-		return result;
-	}
-
-	/* external dictionary mode */
-	{
-		int result;
-
-		if ((streamPtr->dictSize < 64 * KB) &&
-			(streamPtr->dictSize < streamPtr->currentOffset)) {
-			result = LZ4_compress_generic(
-				streamPtr, source, dest, inputSize,
-				maxOutputSize, limitedOutput, byU32,
-				usingExtDict, dictSmall, acceleration);
-		} else {
-			result = LZ4_compress_generic(
-				streamPtr, source, dest, inputSize,
-				maxOutputSize, limitedOutput, byU32,
-				usingExtDict, noDictIssue, acceleration);
-		}
-		streamPtr->dictionary = (const BYTE *)source;
-		streamPtr->dictSize = (U32)inputSize;
-		streamPtr->currentOffset += (U32)inputSize;
-		return result;
-	}
+	return __lz4_compress_fast_continue(streamPtr, src, dst, srcSize,
+					    maxDstSize, acceleration);
 }
 EXPORT_SYMBOL(LZ4_compress_fast_continue);
 

-- 
2.34.1

Re: [PATCH v2 10/15] lib/lz4: switch the compressor to the vendored sources

[sashiko-bot]

> Replace the forked compressor with thin entry points over the vendored
> lz4.c. LZ4_compress_default(), LZ4_compress_fast() and
> LZ4_compress_destSize() keep the trailing wrkmem argument and forward to
> upstream's *_extState() functions; the rest are plain forwarders.
> 
> Upstream's LZ4_initStream() returns NULL for a misaligned workspace,
> which LZ4_compress_fast_extState() dereferences and
> LZ4_compress_destSize_extState() ignores. Check the alignment in the
> entry points, warn once and fail the call. In-tree callers allocate with
> kmalloc() or vmalloc() and are not affected.
> 
> Compressed output differs from the fork's on a small fraction of inputs
> by a byte or two; both decode each other's output. Compressing 4K
> blocks, upstream is 2-4% faster on kernel source and binaries and up to
> 9% slower on datagen.
> 
> LZ4_stream_t becomes an incomplete type, as upstream declares it, so
> <linux/lz4.h> and upstream's lz4.h can share a translation unit. The
> macros that laid out the forked structure go, and LZ4_MEM_COMPRESS keeps
> its value, 16416. zram, the only caller using sizeof(LZ4_stream_t), now
> allocates LZ4_MEM_COMPRESS bytes.
> 
> Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>

Sashiko has reviewed this patch and found no issues.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261106-lz4-vendor-upstream-v2-0-ee04c3f0d7dc@samsung.com?part=10

Re: [PATCH v2 10/15] lib/lz4: switch the compressor to the vendored sources

[kernel test robot]

Hi Michal,

kernel test robot noticed the following build errors:

[auto build test ERROR on 93f51579e7df248780214094418f205253383cc5]

url:    https://github.com/intel-lab-lkp/linux/commits/Michal-Wilczynski/lib-decompress_unlz4-use-LZ4_decompress_safe_partial-before-boot/20261106-154241
base:   93f51579e7df248780214094418f205253383cc5
patch link:    https://lore.kernel.org/r/20261106-lz4-vendor-upstream-v2-10-ee04c3f0d7dc%40samsung.com
patch subject: [PATCH v2 10/15] lib/lz4: switch the compressor to the vendored sources
config: arc-allmodconfig (https://download.01.org/0day-ci/archive/20261107/202611070943.EDg3udaH-lkp@intel.com/config)
compiler: arc-linux-gcc (GCC) 16.2.0
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20261107/202611070943.EDg3udaH-lkp@intel.com/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <lkp@intel.com>
| Closes: https://lore.kernel.org/oe-kbuild-all/202611070943.EDg3udaH-lkp@intel.com/

All errors (new ones prefixed by >>):

   In file included from include/linux/rcupdate.h:27,
                    from include/linux/rbtree.h:24,
                    from include/linux/mm_types.h:11,
                    from include/linux/mmzone.h:22,
                    from include/linux/gfp.h:7,
                    from include/linux/umh.h:4,
                    from include/linux/kmod.h:9,
                    from include/linux/module.h:18,
                    from lib/lz4/lz4_compress.c:29:
   include/linux/sched.h: In function 'is_percpu_thread':
>> include/linux/sched.h:1860:17: error: 'current' undeclared (first use in this function)
    1860 |         return (current->flags & PF_NO_SETAFFINITY) &&
         |                 ^~~~~~~
   include/linux/sched.h:1860:17: note: each undeclared identifier is reported only once for each function it appears in
   include/linux/sched.h: In function 'current_restore_flags':
   include/linux/sched.h:1918:9: error: 'current' undeclared (first use in this function)
    1918 |         current->flags &= ~flags;
         |         ^~~~~~~
   include/linux/sched.h: In function 'set_need_resched_current':
   include/linux/sched.h:2127:30: error: 'current' undeclared (first use in this function)
    2127 |         set_tsk_need_resched(current);
         |                              ^~~~~~~
   In file included from include/asm-generic/bug.h:7,
                    from arch/arc/include/asm/bug.h:30,
                    from include/linux/bug.h:5,
                    from lib/lz4/lz4_compress.c:27:
   include/linux/sched.h: In function '__set_task_blocked_on':
   include/linux/sched.h:2231:27: error: 'current' undeclared (first use in this function)
    2231 |         WARN_ON_ONCE(p != current);
         |                           ^~~~~~~
   include/linux/once_lite.h:28:41: note: in definition of macro 'DO_ONCE_LITE_IF'
      28 |                 bool __ret_do_once = !!(condition);                     \
         |                                         ^~~~~~~~~
   include/linux/sched.h:2231:9: note: in expansion of macro 'WARN_ON_ONCE'
    2231 |         WARN_ON_ONCE(p != current);
         |         ^~~~~~~~~~~~
   In file included from include/linux/nodemask.h:92,
                    from include/linux/numa.h:6,
                    from include/linux/cpumask.h:15,
                    from include/linux/smp.h:13,
                    from include/linux/interrupt_rc.h:17,
                    from include/linux/spinlock.h:60,
                    from include/linux/mmzone.h:8:
   include/linux/sched.h: In function 'alloc_tag_save':
   include/linux/sched.h:2361:14: error: 'current' undeclared (first use in this function)
    2361 |         swap(current->alloc_tag, tag);
         |              ^~~~~~~
   include/linux/minmax.h:308:21: note: in definition of macro 'swap'
     308 |         do { typeof(a) __tmp = (a); (a) = (b); (b) = __tmp; } while (0)
         |                     ^
>> include/linux/minmax.h:308:52: error: assignment to 'struct alloc_tag *' from 'int' makes pointer from integer without a cast [-Wint-conversion]
     308 |         do { typeof(a) __tmp = (a); (a) = (b); (b) = __tmp; } while (0)
         |                                                    ^
   include/linux/sched.h:2361:9: note: in expansion of macro 'swap'
    2361 |         swap(current->alloc_tag, tag);
         |         ^~~~
   include/linux/sched.h: In function 'alloc_tag_restore':
   include/linux/sched.h:2368:14: error: 'current' undeclared (first use in this function)
    2368 |         WARN(current->alloc_tag != tag, "current->alloc_tag was changed:\n");
         |              ^~~~~~~
   include/asm-generic/bug.h:164:32: note: in definition of macro 'WARN'
     164 |         int __ret_warn_on = !!(condition);                              \
         |                                ^~~~~~~~~
   In file included from <command-line>:
   include/linux/rcupdate.h: In function 'cond_resched_tasks_rcu_qs':
>> include/linux/rcupdate.h:226:22: error: 'current' undeclared (first use in this function)
     226 |         rcu_tasks_qs(current, false);
         |                      ^~~~~~~
   include/linux/compiler_types.h:682:23: note: in definition of macro '__compiletime_assert'
     682 |                 if (!(condition))                                       \
         |                       ^~~~~~~~~
   include/linux/compiler_types.h:702:9: note: in expansion of macro '_compiletime_assert'
     702 |         _compiletime_assert(condition, msg, __compiletime_assert_, __COUNTER__)
         |         ^~~~~~~~~~~~~~~~~~~
   include/asm-generic/rwonce.h:36:9: note: in expansion of macro 'compiletime_assert'
      36 |         compiletime_assert(__native_word(t) || sizeof(t) == sizeof(long long),  \
         |         ^~~~~~~~~~~~~~~~~~
   include/asm-generic/rwonce.h:36:28: note: in expansion of macro '__native_word'
      36 |         compiletime_assert(__native_word(t) || sizeof(t) == sizeof(long long),  \
         |                            ^~~~~~~~~~~~~
   include/asm-generic/rwonce.h:49:9: note: in expansion of macro 'compiletime_assert_rwonce_type'
      49 |         compiletime_assert_rwonce_type(x);                              \
         |         ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   include/linux/rcupdate.h:185:35: note: in expansion of macro 'READ_ONCE'
     185 |                 if (!(preempt) && READ_ONCE((t)->rcu_tasks_holdout))    \
         |                                   ^~~~~~~~~
   include/linux/rcupdate.h:197:34: note: in expansion of macro 'rcu_tasks_classic_qs'
     197 | #define rcu_tasks_qs(t, preempt) rcu_tasks_classic_qs((t), (preempt))
         |                                  ^~~~~~~~~~~~~~~~~~~~
   include/linux/rcupdate.h:226:9: note: in expansion of macro 'rcu_tasks_qs'
     226 |         rcu_tasks_qs(current, false);
         |         ^~~~~~~~~~~~
   In file included from include/linux/local_lock.h:5,
                    from include/linux/mmzone.h:24:
   include/linux/local_lock_internal.h: In function 'local_lock_acquire':
>> include/linux/local_lock_internal.h:48:20: error: 'current' undeclared (first use in this function)
      48 |         l->owner = current;
         |                    ^~~~~~~
   include/linux/local_lock_internal.h: In function 'local_trylock_acquire':
   include/linux/local_lock_internal.h:55:20: error: 'current' undeclared (first use in this function)
      55 |         l->owner = current;
         |                    ^~~~~~~
   In file included from include/linux/build_bug.h:5,
                    from lib/lz4/lz4_deps.h:18,
                    from lib/lz4/lz4_compress.c:16:
   include/linux/local_lock_internal.h: In function 'local_lock_release':
   include/linux/local_lock_internal.h:60:41: error: 'current' undeclared (first use in this function)
      60 |         DEBUG_LOCKS_WARN_ON(l->owner != current);
         |                                         ^~~~~~~
   include/linux/compiler.h:77:45: note: in definition of macro 'unlikely'
      77 | # define unlikely(x)    __builtin_expect(!!(x), 0)
         |                                             ^
   include/linux/local_lock_internal.h:60:9: note: in expansion of macro 'DEBUG_LOCKS_WARN_ON'
      60 |         DEBUG_LOCKS_WARN_ON(l->owner != current);
         |         ^~~~~~~~~~~~~~~~~~~
   include/linux/sched/mm.h: In function 'current_gfp_context':
>> include/linux/sched/mm.h:250:41: error: 'current' undeclared (first use in this function)
     250 |         unsigned int pflags = READ_ONCE(current->flags);
         |                                         ^~~~~~~
   include/linux/compiler_types.h:682:23: note: in definition of macro '__compiletime_assert'
     682 |                 if (!(condition))                                       \
         |                       ^~~~~~~~~
   include/linux/compiler_types.h:702:9: note: in expansion of macro '_compiletime_assert'
     702 |         _compiletime_assert(condition, msg, __compiletime_assert_, __COUNTER__)
         |         ^~~~~~~~~~~~~~~~~~~
   include/asm-generic/rwonce.h:36:9: note: in expansion of macro 'compiletime_assert'
      36 |         compiletime_assert(__native_word(t) || sizeof(t) == sizeof(long long),  \
         |         ^~~~~~~~~~~~~~~~~~
   include/asm-generic/rwonce.h:36:28: note: in expansion of macro '__native_word'
      36 |         compiletime_assert(__native_word(t) || sizeof(t) == sizeof(long long),  \
         |                            ^~~~~~~~~~~~~
   include/asm-generic/rwonce.h:49:9: note: in expansion of macro 'compiletime_assert_rwonce_type'
      49 |         compiletime_assert_rwonce_type(x);                              \
         |         ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   include/linux/sched/mm.h:250:31: note: in expansion of macro 'READ_ONCE'
     250 |         unsigned int pflags = READ_ONCE(current->flags);
         |                               ^~~~~~~~~
   include/linux/sched/mm.h: In function 'memalloc_retry_wait':
   include/linux/sched.h:170:17: error: 'current' undeclared (first use in this function)
     170 |                 current->task_state_change = _THIS_IP_;                 \
         |                 ^~~~~~~
   include/linux/sched.h:242:17: note: in expansion of macro 'debug_normal_state_change'
     242 |                 debug_normal_state_change((state_value));               \
         |                 ^~~~~~~~~~~~~~~~~~~~~~~~~
   include/linux/sched/mm.h:293:9: note: in expansion of macro '__set_current_state'
     293 |         __set_current_state(TASK_UNINTERRUPTIBLE);
         |         ^~~~~~~~~~~~~~~~~~~
   In file included from include/linux/xarray.h:21,
                    from include/linux/radix-tree.h:21,
                    from include/linux/idr.h:15,
                    from include/linux/kernfs.h:12,
                    from include/linux/sysfs.h:16,
                    from include/linux/kobject.h:20,
                    from include/linux/module.h:22:
   include/linux/sched/mm.h: In function 'might_alloc':
   include/linux/sched/mm.h:319:13: error: 'current' undeclared (first use in this function)
     319 |         if (current->flags & PF_MEMALLOC)
         |             ^~~~~~~
   include/linux/sched/mm.h: In function 'memalloc_flags_save':
   include/linux/sched/mm.h:334:30: error: 'current' undeclared (first use in this function)
     334 |         unsigned oldflags = ~current->flags & flags;
         |                              ^~~~~~~
   include/linux/sched/mm.h: In function 'memalloc_flags_restore':
   include/linux/sched/mm.h:341:9: error: 'current' undeclared (first use in this function)
     341 |         current->flags &= ~flags;
         |         ^~~~~~~
   include/linux/sched/mm.h: In function 'set_active_memcg':
   include/linux/sched/mm.h:499:23: error: 'current' undeclared (first use in this function)
     499 |                 old = current->active_memcg;
         |                       ^~~~~~~
   include/linux/sched/mm.h: In function 'membarrier_mm_sync_core_before_usermode':
   include/linux/sched/mm.h:543:13: error: 'current' undeclared (first use in this function)
     543 |         if (current->mm != mm)
         |             ^~~~~~~


vim +308 include/linux/minmax.h

f9bff0e31881d03 Matthew Wilcox (Oracle  2023-08-02  285) 
f9bff0e31881d03 Matthew Wilcox (Oracle  2023-08-02  286) /**
f9bff0e31881d03 Matthew Wilcox (Oracle  2023-08-02  287)  * in_range - Determine if a value lies within a range.
f9bff0e31881d03 Matthew Wilcox (Oracle  2023-08-02  288)  * @val: Value to test.
f9bff0e31881d03 Matthew Wilcox (Oracle  2023-08-02  289)  * @start: First value in range.
f9bff0e31881d03 Matthew Wilcox (Oracle  2023-08-02  290)  * @len: Number of values in range.
f9bff0e31881d03 Matthew Wilcox (Oracle  2023-08-02  291)  *
f9bff0e31881d03 Matthew Wilcox (Oracle  2023-08-02  292)  * This is more efficient than "if (start <= val && val < (start + len))".
f9bff0e31881d03 Matthew Wilcox (Oracle  2023-08-02  293)  * It also gives a different answer if @start + @len overflows the size of
f9bff0e31881d03 Matthew Wilcox (Oracle  2023-08-02  294)  * the type by a sufficient amount to encompass @val.  Decide for yourself
f9bff0e31881d03 Matthew Wilcox (Oracle  2023-08-02  295)  * which behaviour you want, or prove that start + len never overflow.
f9bff0e31881d03 Matthew Wilcox (Oracle  2023-08-02  296)  * Do not blindly replace one form with the other.
f9bff0e31881d03 Matthew Wilcox (Oracle  2023-08-02  297)  */
f9bff0e31881d03 Matthew Wilcox (Oracle  2023-08-02  298) #define in_range(val, start, len)					\
f9bff0e31881d03 Matthew Wilcox (Oracle  2023-08-02  299) 	((sizeof(start) | sizeof(len) | sizeof(val)) <= sizeof(u32) ?	\
f9bff0e31881d03 Matthew Wilcox (Oracle  2023-08-02  300) 		in_range32(val, start, len) : in_range64(val, start, len))
f9bff0e31881d03 Matthew Wilcox (Oracle  2023-08-02  301) 
b296a6d53339a79 Andy Shevchenko         2020-10-15  302  /**
b296a6d53339a79 Andy Shevchenko         2020-10-15  303   * swap - swap values of @a and @b
b296a6d53339a79 Andy Shevchenko         2020-10-15  304   * @a: first value
b296a6d53339a79 Andy Shevchenko         2020-10-15  305   * @b: second value
b296a6d53339a79 Andy Shevchenko         2020-10-15  306   */
b296a6d53339a79 Andy Shevchenko         2020-10-15  307  #define swap(a, b) \
b296a6d53339a79 Andy Shevchenko         2020-10-15 @308  	do { typeof(a) __tmp = (a); (a) = (b); (b) = __tmp; } while (0)
b296a6d53339a79 Andy Shevchenko         2020-10-15  309  

--
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki

Re: [PATCH v2 10/15] lib/lz4: switch the compressor to the vendored sources

[Sergey Senozhatsky]

On (26/11/06 15:42), Michal Wilczynski wrote:
[..]
> diff --git a/drivers/block/zram/backend_lz4.c b/drivers/block/zram/backend_lz4.c
> index 1e4ad31d39a63f45b3bddd3d8a467d8d7d095beb..50265e3ce256490994cb6dae32a45da376e408e0 100644
> --- a/drivers/block/zram/backend_lz4.c
> +++ b/drivers/block/zram/backend_lz4.c
> @@ -42,7 +42,7 @@ static int lz4_setup_params(struct zcomp_params *params)
>  	if (!params->dict || !params->dict_sz)
>  		return 0;
>  
> -	dict_stream = kzalloc_obj(*dict_stream);
> +	dict_stream = kzalloc(LZ4_MEM_COMPRESS, GFP_KERNEL);
>  	if (!dict_stream)
>  		return -ENOMEM;
>  
> @@ -88,7 +88,7 @@ static int lz4_create(struct zcomp_params *params, struct zcomp_ctx *ctx)
>  		if (!zctx->dstrm)
>  			goto error;
>  
> -		zctx->cstrm = kzalloc_obj(*zctx->cstrm);
> +		zctx->cstrm = kzalloc(LZ4_MEM_COMPRESS, GFP_KERNEL);
>  		if (!zctx->cstrm)
>  			goto error;
>  	}
> @@ -112,7 +112,7 @@ static int lz4_compress(struct zcomp_params *params, struct zcomp_ctx *ctx,
>  					zctx->mem);
>  	} else {
>  		/* Cstrm needs to be reset */
> -		memcpy(zctx->cstrm, params->drv_data, sizeof(*zctx->cstrm));
> +		memcpy(zctx->cstrm, params->drv_data, LZ4_MEM_COMPRESS);
>  		ret = LZ4_compress_fast_continue(zctx->cstrm, req->src,
>  						 req->dst, req->src_len,
>  						 req->dst_len, params->level);

The zram code here is outdated, it has been reworked quite a bit
(patches are in linux-next and Andrew's trees).

Re: [f2fs-dev] [PATCH v2 10/15] lib/lz4: switch the compressor to the vendored sources

[Sergey Senozhatsky]

On (26/11/06 15:42), Michal Wilczynski wrote:
[..]
> diff --git a/drivers/block/zram/backend_lz4.c b/drivers/block/zram/backend_lz4.c
> index 1e4ad31d39a63f45b3bddd3d8a467d8d7d095beb..50265e3ce256490994cb6dae32a45da376e408e0 100644
> --- a/drivers/block/zram/backend_lz4.c
> +++ b/drivers/block/zram/backend_lz4.c
> @@ -42,7 +42,7 @@ static int lz4_setup_params(struct zcomp_params *params)
>  	if (!params->dict || !params->dict_sz)
>  		return 0;
>  
> -	dict_stream = kzalloc_obj(*dict_stream);
> +	dict_stream = kzalloc(LZ4_MEM_COMPRESS, GFP_KERNEL);
>  	if (!dict_stream)
>  		return -ENOMEM;
>  
> @@ -88,7 +88,7 @@ static int lz4_create(struct zcomp_params *params, struct zcomp_ctx *ctx)
>  		if (!zctx->dstrm)
>  			goto error;
>  
> -		zctx->cstrm = kzalloc_obj(*zctx->cstrm);
> +		zctx->cstrm = kzalloc(LZ4_MEM_COMPRESS, GFP_KERNEL);
>  		if (!zctx->cstrm)
>  			goto error;
>  	}
> @@ -112,7 +112,7 @@ static int lz4_compress(struct zcomp_params *params, struct zcomp_ctx *ctx,
>  					zctx->mem);
>  	} else {
>  		/* Cstrm needs to be reset */
> -		memcpy(zctx->cstrm, params->drv_data, sizeof(*zctx->cstrm));
> +		memcpy(zctx->cstrm, params->drv_data, LZ4_MEM_COMPRESS);
>  		ret = LZ4_compress_fast_continue(zctx->cstrm, req->src,
>  						 req->dst, req->src_len,
>  						 req->dst_len, params->level);

The zram code here is outdated, it has been reworked quite a bit
(patches are in linux-next and Andrew's trees).


_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

Re: [PATCH v2 10/15] lib/lz4: switch the compressor to the vendored sources

[Michal Wilczynski]

On 11/10/26 06:36, Sergey Senozhatsky wrote:
> On (26/11/06 15:42), Michal Wilczynski wrote:
> [..]
>> diff --git a/drivers/block/zram/backend_lz4.c b/drivers/block/zram/backend_lz4.c
>> index 1e4ad31d39a63f45b3bddd3d8a467d8d7d095beb..50265e3ce256490994cb6dae32a45da376e408e0 100644
>> --- a/drivers/block/zram/backend_lz4.c
>> +++ b/drivers/block/zram/backend_lz4.c
>> @@ -42,7 +42,7 @@ static int lz4_setup_params(struct zcomp_params *params)
>>  	if (!params->dict || !params->dict_sz)
>>  		return 0;
>>  
>> -	dict_stream = kzalloc_obj(*dict_stream);
>> +	dict_stream = kzalloc(LZ4_MEM_COMPRESS, GFP_KERNEL);
>>  	if (!dict_stream)
>>  		return -ENOMEM;
>>  
>> @@ -88,7 +88,7 @@ static int lz4_create(struct zcomp_params *params, struct zcomp_ctx *ctx)
>>  		if (!zctx->dstrm)
>>  			goto error;
>>  
>> -		zctx->cstrm = kzalloc_obj(*zctx->cstrm);
>> +		zctx->cstrm = kzalloc(LZ4_MEM_COMPRESS, GFP_KERNEL);
>>  		if (!zctx->cstrm)
>>  			goto error;
>>  	}
>> @@ -112,7 +112,7 @@ static int lz4_compress(struct zcomp_params *params, struct zcomp_ctx *ctx,
>>  					zctx->mem);
>>  	} else {
>>  		/* Cstrm needs to be reset */
>> -		memcpy(zctx->cstrm, params->drv_data, sizeof(*zctx->cstrm));
>> +		memcpy(zctx->cstrm, params->drv_data, LZ4_MEM_COMPRESS);
>>  		ret = LZ4_compress_fast_continue(zctx->cstrm, req->src,
>>  						 req->dst, req->src_len,
>>  						 req->dst_len, params->level);
> 
> The zram code here is outdated, it has been reworked quite a bit
> (patches are in linux-next and Andrew's trees).
> 

Right, I'll rebase on linux-next.

Best regards,
-- 
Michal Wilczynski <m.wilczynski@samsung.com>

Re: [f2fs-dev] [PATCH v2 10/15] lib/lz4: switch the compressor to the vendored sources

[Michal Wilczynski]

On 11/10/26 06:36, Sergey Senozhatsky wrote:
> On (26/11/06 15:42), Michal Wilczynski wrote:
> [..]
>> diff --git a/drivers/block/zram/backend_lz4.c b/drivers/block/zram/backend_lz4.c
>> index 1e4ad31d39a63f45b3bddd3d8a467d8d7d095beb..50265e3ce256490994cb6dae32a45da376e408e0 100644
>> --- a/drivers/block/zram/backend_lz4.c
>> +++ b/drivers/block/zram/backend_lz4.c
>> @@ -42,7 +42,7 @@ static int lz4_setup_params(struct zcomp_params *params)
>>  	if (!params->dict || !params->dict_sz)
>>  		return 0;
>>  
>> -	dict_stream = kzalloc_obj(*dict_stream);
>> +	dict_stream = kzalloc(LZ4_MEM_COMPRESS, GFP_KERNEL);
>>  	if (!dict_stream)
>>  		return -ENOMEM;
>>  
>> @@ -88,7 +88,7 @@ static int lz4_create(struct zcomp_params *params, struct zcomp_ctx *ctx)
>>  		if (!zctx->dstrm)
>>  			goto error;
>>  
>> -		zctx->cstrm = kzalloc_obj(*zctx->cstrm);
>> +		zctx->cstrm = kzalloc(LZ4_MEM_COMPRESS, GFP_KERNEL);
>>  		if (!zctx->cstrm)
>>  			goto error;
>>  	}
>> @@ -112,7 +112,7 @@ static int lz4_compress(struct zcomp_params *params, struct zcomp_ctx *ctx,
>>  					zctx->mem);
>>  	} else {
>>  		/* Cstrm needs to be reset */
>> -		memcpy(zctx->cstrm, params->drv_data, sizeof(*zctx->cstrm));
>> +		memcpy(zctx->cstrm, params->drv_data, LZ4_MEM_COMPRESS);
>>  		ret = LZ4_compress_fast_continue(zctx->cstrm, req->src,
>>  						 req->dst, req->src_len,
>>  						 req->dst_len, params->level);
> 
> The zram code here is outdated, it has been reworked quite a bit
> (patches are in linux-next and Andrew's trees).
> 

Right, I'll rebase on linux-next.

Best regards,
-- 
Michal Wilczynski <m.wilczynski@samsung.com>


_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

[f2fs-dev] [PATCH v2 11/15] lib/lz4: switch the HC compressor to the vendored sources

[Michal Wilczynski]

Replace the forked LZ4HC with thin entry points over the vendored
lz4hc.c.

Upstream routes levels 10 and up to its optimal parser, whose ~64K
workspace the kernel cannot provide. Clamp those levels to 9 rather than
reject them, so existing f2fs and zram settings keep working, and add
LZ4HC_CLAMP_CLEVEL to <linux/lz4.h>. Document the clamp in f2fs.rst.

Levels 1 and 2 now use upstream's lz4mid: on kernel source they are
50-84% faster for 4-9% larger output. Levels 3 to 9 compress the same,
but at 4K are up to 15% slower on poorly compressible data, including at
level 9, zram's lz4hc default.

LZ4_compress_HC() and LZ4_resetStreamHC() check the state's alignment
like the compressor's entry points.

LZ4_streamHC_t becomes incomplete and the macros that laid out the forked
structure go. LZ4HC_MEM_COMPRESS grows from 262192 to upstream's 262200,
static_asserted against LZ4_STREAMHC_MINSIZE. lz4hc.c pulls in lz4.c's
decoder tables, unused here when the fast decode loop is off, so silence
-Wunused-const-variable for this object.

This retires commit b08918fb3f27 ("lz4: do not export static symbol"):
the glue exports only the five LZ4HC entry points.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 Documentation/filesystems/f2fs.rst |   2 +
 drivers/block/zram/backend_lz4hc.c |   2 +-
 include/linux/lz4.h                |  54 +--
 lib/lz4/Makefile                   |   4 +
 lib/lz4/lz4hc_compress.c           | 792 ++++---------------------------------
 5 files changed, 91 insertions(+), 763 deletions(-)

diff --git a/Documentation/filesystems/f2fs.rst b/Documentation/filesystems/f2fs.rst
index 771216f45207a797356d5d2a10328b74d1ab302a..4a2d4545c2752cbcea4939839ba3d2457bf92299 100644
--- a/Documentation/filesystems/f2fs.rst
+++ b/Documentation/filesystems/f2fs.rst
@@ -318,6 +318,8 @@ compress_algorithm=%s:%d Control compress algorithm and its compress level, now,
 				 zstd           1 - 22
 				 =========      ===========
 
+			 lz4 levels above 9 compress as level 9.
+
 compress_log_size=%u	 Support configuring compress cluster size. The size will
 			 be 4KB * (1 << %u). The default and minimum sizes are 16KB.
 compress_extension=%s	 Support adding specified extension, so that f2fs can enable
diff --git a/drivers/block/zram/backend_lz4hc.c b/drivers/block/zram/backend_lz4hc.c
index 06286c805cf7c56bd760f460cc854dc4014f293b..e7ee3740219852f079d1e0866da6b213a3f3a583 100644
--- a/drivers/block/zram/backend_lz4hc.c
+++ b/drivers/block/zram/backend_lz4hc.c
@@ -69,7 +69,7 @@ static int lz4hc_create(struct zcomp_params *params, struct zcomp_ctx *ctx)
 		if (!zctx->dstrm)
 			goto error;
 
-		zctx->cstrm = kzalloc_obj(*zctx->cstrm);
+		zctx->cstrm = kzalloc(LZ4HC_MEM_COMPRESS, GFP_KERNEL);
 		if (!zctx->cstrm)
 			goto error;
 		LZ4_resetStreamHC(zctx->cstrm, params->level);
diff --git a/include/linux/lz4.h b/include/linux/lz4.h
index 031a410b8e3878e9a24cc49af36cfcbba399d991..05060b9790be4621ddbe744172fa0ae37f6191f4 100644
--- a/include/linux/lz4.h
+++ b/include/linux/lz4.h
@@ -59,19 +59,16 @@
 #define LZ4HC_DEFAULT_CLEVEL			9
 #define LZ4HC_MAX_CLEVEL			16
 
-#define LZ4HC_DICTIONARY_LOGSIZE 16
-#define LZ4HC_MAXD (1<<LZ4HC_DICTIONARY_LOGSIZE)
-#define LZ4HC_MAXD_MASK (LZ4HC_MAXD - 1)
-#define LZ4HC_HASH_LOG (LZ4HC_DICTIONARY_LOGSIZE - 1)
-#define LZ4HC_HASHTABLESIZE (1 << LZ4HC_HASH_LOG)
-#define LZ4HC_HASH_MASK (LZ4HC_HASHTABLESIZE - 1)
+/* Levels from here up select the optimal parser, which needs a ~64K
+ * workspace lib/lz4 cannot give it, so lib/lz4 compresses them as
+ * LZ4HC_CLAMP_CLEVEL - 1.  f2fs and zram still accept levels up to
+ * LZ4HC_MAX_CLEVEL.
+ */
+#define LZ4HC_CLAMP_CLEVEL			10
 
 /*-************************************************************************
  *	STREAMING CONSTANTS AND STRUCTURES
  **************************************************************************/
-#define LZ4_STREAMHCSIZE        262192
-#define LZ4_STREAMHCSIZE_SIZET (262192 / sizeof(size_t))
-
 #define LZ4_STREAMDECODESIZE_U64	4
 #define LZ4_STREAMDECODESIZE		 (LZ4_STREAMDECODESIZE_U64 * \
 	sizeof(unsigned long long))
@@ -83,29 +80,10 @@
 typedef union LZ4_stream_u LZ4_stream_t;
 
 /*
- * LZ4_streamHC_t - information structure to track an LZ4HC stream.
+ * LZ4_streamHC_t - an LZ4HC stream.  Incomplete: lib/lz4 owns the layout.
+ * Allocate LZ4HC_MEM_COMPRESS bytes and cast, do not sizeof().
  */
-typedef struct {
-	unsigned int	 hashTable[LZ4HC_HASHTABLESIZE];
-	unsigned short	 chainTable[LZ4HC_MAXD];
-	/* next block to continue on current prefix */
-	const unsigned char *end;
-	/* All index relative to this position */
-	const unsigned char *base;
-	/* alternate base for extDict */
-	const unsigned char *dictBase;
-	/* below that point, need extDict */
-	unsigned int	 dictLimit;
-	/* below that point, no more dict */
-	unsigned int	 lowLimit;
-	/* index from which to continue dict update */
-	unsigned int	 nextToUpdate;
-	unsigned int	 compressionLevel;
-} LZ4HC_CCtx_internal;
-typedef union {
-	size_t table[LZ4_STREAMHCSIZE_SIZET];
-	LZ4HC_CCtx_internal internal_donotuse;
-} LZ4_streamHC_t;
+typedef union LZ4_streamHC_u LZ4_streamHC_t;
 
 /*
  * LZ4_streamDecode_t - information structure to track an
@@ -133,7 +111,7 @@ typedef union {
  * points warn about a misaligned buffer and fail the call.
  */
 #define LZ4_MEM_COMPRESS	16416
-#define LZ4HC_MEM_COMPRESS	LZ4_STREAMHCSIZE
+#define LZ4HC_MEM_COMPRESS	262200
 
 /*-************************************************************************
  *	Compression Functions
@@ -289,9 +267,9 @@ int LZ4_decompress_safe_partial(const char *source, char *dest,
  * @srcSize: size of the input data. Max supported value is LZ4_MAX_INPUT_SIZE
  * @dstCapacity: full or partial size of buffer 'dst',
  *	which must be already allocated
- * @compressionLevel: Recommended values are between 4 and 9, although any
- *	value between 1 and LZ4HC_MAX_CLEVEL will work.
- *	Values >LZ4HC_MAX_CLEVEL behave the same as 16.
+ * @compressionLevel: Recommended values are between 4 and 9.  Levels of
+ *	LZ4HC_CLAMP_CLEVEL and above are clamped to LZ4HC_CLAMP_CLEVEL - 1;
+ *	see that macro.
  * @wrkmem: address of the working memory.
  *	This requires 'wrkmem' of size LZ4HC_MEM_COMPRESS, aligned to 8 bytes.
  *
@@ -307,9 +285,9 @@ int LZ4_compress_HC(const char *src, char *dst, int srcSize, int dstCapacity,
 /**
  * LZ4_resetStreamHC() - Init an allocated 'LZ4_streamHC_t' structure
  * @streamHCPtr: pointer to the 'LZ4_streamHC_t' structure
- * @compressionLevel: Recommended values are between 4 and 9, although any
- *	value between 1 and LZ4HC_MAX_CLEVEL will work.
- *	Values >LZ4HC_MAX_CLEVEL behave the same as 16.
+ * @compressionLevel: Recommended values are between 4 and 9.  Levels of
+ *	LZ4HC_CLAMP_CLEVEL and above are clamped to LZ4HC_CLAMP_CLEVEL - 1;
+ *	see that macro.
  *
  * An LZ4_streamHC_t structure can be allocated once
  * and re-used multiple times.
diff --git a/lib/lz4/Makefile b/lib/lz4/Makefile
index 8b4eb07694e67dd99ec7cc85212fb68138d3981e..5cbbe10beedce1199dee68992d542a1e52f89f8e 100644
--- a/lib/lz4/Makefile
+++ b/lib/lz4/Makefile
@@ -10,3 +10,7 @@ ccflags-y += -I $(src)/freestanding
 obj-$(CONFIG_LZ4_COMPRESS)	+= lz4_compress.o
 obj-$(CONFIG_LZ4HC_COMPRESS)	+= lz4hc_compress.o
 obj-$(CONFIG_LZ4_DECOMPRESS)	+= lz4_decompress.o
+
+# lz4hc.c pulls in lz4.c's common definitions, which include two decoder
+# tables nothing here uses when LZ4_FAST_DEC_LOOP is off.
+CFLAGS_lz4hc_compress.o += $(call cc-disable-warning, unused-const-variable)
diff --git a/lib/lz4/lz4hc_compress.c b/lib/lz4/lz4hc_compress.c
index 91936dc3d14bca31752a361693380aad8eb3b98a..4d56d45028110805628be766b2758a8dffd56f0c 100644
--- a/lib/lz4/lz4hc_compress.c
+++ b/lib/lz4/lz4hc_compress.c
@@ -1,766 +1,110 @@
+// SPDX-License-Identifier: GPL-2.0-only OR BSD-2-Clause
 /*
- * LZ4 HC - High Compression Mode of LZ4
- * Copyright (C) 2011-2015, Yann Collet.
+ * Copyright (C) 2011 - 2016, Yann Collet.
+ * Copyright (C) 2016, Sven Schmidt <4sschmid@informatik.uni-hamburg.de>
+ * Copyright (c) 2026 Samsung Electronics Co., Ltd.
+ * Author: Michal Wilczynski <m.wilczynski@samsung.com>
  *
- * BSD 2 - Clause License (http://www.opensource.org/licenses/bsd - license.php)
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions are
- * met:
- *	* Redistributions of source code must retain the above copyright
- *	  notice, this list of conditions and the following disclaimer.
- *	* Redistributions in binary form must reproduce the above
- * copyright notice, this list of conditions and the following disclaimer
- * in the documentation and/or other materials provided with the
- * distribution.
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
- * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
- * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
- * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
- * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
- * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
- * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
- * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
- * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- * You can contact the author at :
- *	- LZ4 homepage : http://www.lz4.org
- *	- LZ4 source repository : https://github.com/lz4/lz4
+ * LZ4 HC compressor -- kernel entry points
  *
- *	Changed for kernel usage by:
- *	Sven Schmidt <4sschmid@informatik.uni-hamburg.de>
+ * The compressor is upstream's, in the verbatim lz4hc.c included below.
+ * LZ4_compress_HC() takes the kernel's trailing wrkmem argument and forwards
+ * to LZ4_compress_HC_extStateHC(); the level-taking entry points clamp the
+ * level below LZ4HC_CLEVEL_OPT_MIN (see lz4hc_clamp_level()).
  */
 
-/*-************************************
- *	Dependencies
- **************************************/
-#include "lz4defs.h"
-#include <linux/module.h>
-#include <linux/kernel.h>
-#include <linux/string.h> /* memset */
-
-/* *************************************
- *	Local Constants and types
- ***************************************/
-
-#define OPTIMAL_ML (int)((ML_MASK - 1) + MINMATCH)
-
-#define HASH_FUNCTION(i)	(((i) * 2654435761U) \
-	>> ((MINMATCH*8) - LZ4HC_HASH_LOG))
-#define DELTANEXTU16(p)	chainTable[(U16)(p)] /* faster */
-
-static U32 LZ4HC_hashPtr(const void *ptr)
-{
-	return HASH_FUNCTION(LZ4_read32(ptr));
-}
+#include "lz4_deps.h"
 
-/**************************************
- *	HC Compression
- **************************************/
-static void LZ4HC_init(LZ4HC_CCtx_internal *hc4, const BYTE *start)
-{
-	memset((void *)hc4->hashTable, 0, sizeof(hc4->hashTable));
-	memset(hc4->chainTable, 0xFF, sizeof(hc4->chainTable));
-	hc4->nextToUpdate = 64 * KB;
-	hc4->base = start - 64 * KB;
-	hc4->end = start;
-	hc4->dictBase = start - 64 * KB;
-	hc4->dictLimit = 64 * KB;
-	hc4->lowLimit = 64 * KB;
-}
-
-/* Update chains up to ip (excluded) */
-static FORCE_INLINE void LZ4HC_Insert(LZ4HC_CCtx_internal *hc4,
-	const BYTE *ip)
+/* lz4hc.c includes lz4.c with LZ4_COMMONDEFS_ONLY, which stops short of
+ * LZ4_compressBound().  Supply it for lz4hc.c's internal use.
+ */
+static int LZ4_compressBound(int isize)
 {
-	U16 * const chainTable = hc4->chainTable;
-	U32 * const hashTable	= hc4->hashTable;
-	const BYTE * const base = hc4->base;
-	U32 const target = (U32)(ip - base);
-	U32 idx = hc4->nextToUpdate;
-
-	while (idx < target) {
-		U32 const h = LZ4HC_hashPtr(base + idx);
-		size_t delta = idx - hashTable[h];
-
-		if (delta > MAX_DISTANCE)
-			delta = MAX_DISTANCE;
-
-		DELTANEXTU16(idx) = (U16)delta;
-
-		hashTable[h] = idx;
-		idx++;
-	}
-
-	hc4->nextToUpdate = target;
+	return LZ4_COMPRESSBOUND(isize);
 }
 
-static FORCE_INLINE int LZ4HC_InsertAndFindBestMatch(
-	LZ4HC_CCtx_internal *hc4, /* Index table will be updated */
-	const BYTE *ip,
-	const BYTE * const iLimit,
-	const BYTE **matchpos,
-	const int maxNbAttempts)
-{
-	U16 * const chainTable = hc4->chainTable;
-	U32 * const HashTable = hc4->hashTable;
-	const BYTE * const base = hc4->base;
-	const BYTE * const dictBase = hc4->dictBase;
-	const U32 dictLimit = hc4->dictLimit;
-	const U32 lowLimit = (hc4->lowLimit + 64 * KB > (U32)(ip - base))
-		? hc4->lowLimit
-		: (U32)(ip - base) - (64 * KB - 1);
-	U32 matchIndex;
-	int nbAttempts = maxNbAttempts;
-	size_t ml = 0;
-
-	/* HC4 match finder */
-	LZ4HC_Insert(hc4, ip);
-	matchIndex = HashTable[LZ4HC_hashPtr(ip)];
-
-	while ((matchIndex >= lowLimit)
-		&& (nbAttempts)) {
-		nbAttempts--;
-		if (matchIndex >= dictLimit) {
-			const BYTE * const match = base + matchIndex;
-
-			if (*(match + ml) == *(ip + ml)
-				&& (LZ4_read32(match) == LZ4_read32(ip))) {
-				size_t const mlt = LZ4_count(ip + MINMATCH,
-					match + MINMATCH, iLimit) + MINMATCH;
-
-				if (mlt > ml) {
-					ml = mlt;
-					*matchpos = match;
-				}
-			}
-		} else {
-			const BYTE * const match = dictBase + matchIndex;
+#include "upstream/lz4hc.c"
 
-			if (LZ4_read32(match) == LZ4_read32(ip)) {
-				size_t mlt;
-				const BYTE *vLimit = ip
-					+ (dictLimit - matchIndex);
+/* Upstream's short helper macros; MIN and MAX clash with <linux/minmax.h>. */
+#undef MIN
+#undef MAX
+#undef KB
+#undef MB
+#undef GB
 
-				if (vLimit > iLimit)
-					vLimit = iLimit;
-				mlt = LZ4_count(ip + MINMATCH,
-					match + MINMATCH, vLimit) + MINMATCH;
-				if ((ip + mlt == vLimit)
-					&& (vLimit < iLimit))
-					mlt += LZ4_count(ip + mlt,
-						base + dictLimit,
-						iLimit);
-				if (mlt > ml) {
-					/* virtual matchpos */
-					ml = mlt;
-					*matchpos = base + matchIndex;
-				}
-			}
-		}
-		matchIndex -= DELTANEXTU16(matchIndex);
-	}
+#include "lz4_kernel_api.h"
+#include <linux/bug.h>
+#include <linux/export.h>
+#include <linux/module.h>
 
-	return (int)ml;
-}
+/* Catch any divergence from upstream's layout at build time. */
+static_assert(sizeof(LZ4_streamHC_t) == LZ4_STREAMHC_MINSIZE);
+static_assert(LZ4_STREAMHC_MINSIZE == 262200); /* LZ4HC_MEM_COMPRESS */
 
-static FORCE_INLINE int LZ4HC_InsertAndGetWiderMatch(
-	LZ4HC_CCtx_internal *hc4,
-	const BYTE * const ip,
-	const BYTE * const iLowLimit,
-	const BYTE * const iHighLimit,
-	int longest,
-	const BYTE **matchpos,
-	const BYTE **startpos,
-	const int maxNbAttempts)
+/* Levels >= LZ4HC_CLEVEL_OPT_MIN (10) reach the optimal parser, which needs
+ * a ~64K opt[]: too big for a kernel stack, and in the heap mode lz4_deps.h
+ * selects its allocation always fails, and so would the call.  Clamp them
+ * to 9 rather than reject them, so existing f2fs and zram settings keep
+ * working.  Levels 3..9 keep their meaning; 1 and 2 are upstream's lz4mid,
+ * faster and weaker than the shallow hash chain the fork used for them.
+ */
+static int lz4hc_clamp_level(int compressionLevel)
 {
-	U16 * const chainTable = hc4->chainTable;
-	U32 * const HashTable = hc4->hashTable;
-	const BYTE * const base = hc4->base;
-	const U32 dictLimit = hc4->dictLimit;
-	const BYTE * const lowPrefixPtr = base + dictLimit;
-	const U32 lowLimit = (hc4->lowLimit + 64 * KB > (U32)(ip - base))
-		? hc4->lowLimit
-		: (U32)(ip - base) - (64 * KB - 1);
-	const BYTE * const dictBase = hc4->dictBase;
-	U32 matchIndex;
-	int nbAttempts = maxNbAttempts;
-	int delta = (int)(ip - iLowLimit);
-
-	/* First Match */
-	LZ4HC_Insert(hc4, ip);
-	matchIndex = HashTable[LZ4HC_hashPtr(ip)];
-
-	while ((matchIndex >= lowLimit)
-		&& (nbAttempts)) {
-		nbAttempts--;
-		if (matchIndex >= dictLimit) {
-			const BYTE *matchPtr = base + matchIndex;
-
-			if (*(iLowLimit + longest)
-				== *(matchPtr - delta + longest)) {
-				if (LZ4_read32(matchPtr) == LZ4_read32(ip)) {
-					int mlt = MINMATCH + LZ4_count(
-						ip + MINMATCH,
-						matchPtr + MINMATCH,
-						iHighLimit);
-					int back = 0;
-
-					while ((ip + back > iLowLimit)
-						&& (matchPtr + back > lowPrefixPtr)
-						&& (ip[back - 1] == matchPtr[back - 1]))
-						back--;
+	if (compressionLevel >= LZ4HC_CLEVEL_OPT_MIN)
+		return LZ4HC_CLEVEL_OPT_MIN - 1;
 
-					mlt -= back;
-
-					if (mlt > longest) {
-						longest = (int)mlt;
-						*matchpos = matchPtr + back;
-						*startpos = ip + back;
-					}
-				}
-			}
-		} else {
-			const BYTE * const matchPtr = dictBase + matchIndex;
-
-			if (LZ4_read32(matchPtr) == LZ4_read32(ip)) {
-				size_t mlt;
-				int back = 0;
-				const BYTE *vLimit = ip + (dictLimit - matchIndex);
-
-				if (vLimit > iHighLimit)
-					vLimit = iHighLimit;
-
-				mlt = LZ4_count(ip + MINMATCH,
-					matchPtr + MINMATCH, vLimit) + MINMATCH;
-
-				if ((ip + mlt == vLimit) && (vLimit < iHighLimit))
-					mlt += LZ4_count(ip + mlt, base + dictLimit,
-						iHighLimit);
-				while ((ip + back > iLowLimit)
-					&& (matchIndex + back > lowLimit)
-					&& (ip[back - 1] == matchPtr[back - 1]))
-					back--;
-
-				mlt -= back;
-
-				if ((int)mlt > longest) {
-					longest = (int)mlt;
-					*matchpos = base + matchIndex + back;
-					*startpos = ip + back;
-				}
-			}
-		}
-
-		matchIndex -= DELTANEXTU16(matchIndex);
-	}
-
-	return longest;
+	return compressionLevel;
 }
 
-static FORCE_INLINE int LZ4HC_encodeSequence(
-	const BYTE **ip,
-	BYTE **op,
-	const BYTE **anchor,
-	int matchLength,
-	const BYTE * const match,
-	limitedOutput_directive limitedOutputBuffer,
-	BYTE *oend)
+/* LZ4_initStreamHC() returns NULL for a misaligned state.
+ * LZ4_compress_HC_extStateHC() then fails quietly, and LZ4_resetStreamHC()
+ * ignores it and leaves the state uninitialised.  Warn, and skip the reset.
+ */
+static bool lz4hc_state_misaligned(const void *state)
 {
-	int length;
-	BYTE *token;
-
-	/* Encode Literal length */
-	length = (int)(*ip - *anchor);
-	token = (*op)++;
-
-	if ((limitedOutputBuffer)
-		&& ((*op + (length>>8)
-			+ length + (2 + 1 + LASTLITERALS)) > oend)) {
-		/* Check output limit */
-		return 1;
-	}
-	if (length >= (int)RUN_MASK) {
-		int len;
-
-		*token = (RUN_MASK<<ML_BITS);
-		len = length - RUN_MASK;
-		for (; len > 254 ; len -= 255)
-			*(*op)++ = 255;
-		*(*op)++ = (BYTE)len;
-	} else
-		*token = (BYTE)(length<<ML_BITS);
-
-	/* Copy Literals */
-	LZ4_wildCopy(*op, *anchor, (*op) + length);
-	*op += length;
-
-	/* Encode Offset */
-	LZ4_writeLE16(*op, (U16)(*ip - match));
-	*op += 2;
-
-	/* Encode MatchLength */
-	length = (int)(matchLength - MINMATCH);
-
-	if ((limitedOutputBuffer)
-		&& (*op + (length>>8)
-			+ (1 + LASTLITERALS) > oend)) {
-		/* Check output limit */
-		return 1;
-	}
-
-	if (length >= (int)ML_MASK) {
-		*token += ML_MASK;
-		length -= ML_MASK;
-
-		for (; length > 509 ; length -= 510) {
-			*(*op)++ = 255;
-			*(*op)++ = 255;
-		}
-
-		if (length > 254) {
-			length -= 255;
-			*(*op)++ = 255;
-		}
-
-		*(*op)++ = (BYTE)length;
-	} else
-		*token += (BYTE)(length);
-
-	/* Prepare next loop */
-	*ip += matchLength;
-	*anchor = *ip;
-
-	return 0;
+	return WARN_ON_ONCE(!LZ4_isAligned(state, LZ4_streamHC_t_alignment()));
 }
 
-static int LZ4HC_compress_generic(
-	LZ4HC_CCtx_internal *const ctx,
-	const char * const source,
-	char * const dest,
-	int const inputSize,
-	int const maxOutputSize,
-	int compressionLevel,
-	limitedOutput_directive limit
-	)
+int LZ4_compress_HC(const char *src, char *dst, int srcSize, int dstCapacity,
+		    int compressionLevel, void *wrkmem)
 {
-	const BYTE *ip = (const BYTE *) source;
-	const BYTE *anchor = ip;
-	const BYTE * const iend = ip + inputSize;
-	const BYTE * const mflimit = iend - MFLIMIT;
-	const BYTE * const matchlimit = (iend - LASTLITERALS);
-
-	BYTE *op = (BYTE *) dest;
-	BYTE * const oend = op + maxOutputSize;
-
-	unsigned int maxNbAttempts;
-	int ml, ml2, ml3, ml0;
-	const BYTE *ref = NULL;
-	const BYTE *start2 = NULL;
-	const BYTE *ref2 = NULL;
-	const BYTE *start3 = NULL;
-	const BYTE *ref3 = NULL;
-	const BYTE *start0;
-	const BYTE *ref0;
-
-	/* init */
-	if (compressionLevel > LZ4HC_MAX_CLEVEL)
-		compressionLevel = LZ4HC_MAX_CLEVEL;
-	if (compressionLevel < 1)
-		compressionLevel = LZ4HC_DEFAULT_CLEVEL;
-	maxNbAttempts = 1 << (compressionLevel - 1);
-	ctx->end += inputSize;
-
-	ip++;
-
-	/* Main Loop */
-	while (ip < mflimit) {
-		ml = LZ4HC_InsertAndFindBestMatch(ctx, ip,
-			matchlimit, (&ref), maxNbAttempts);
-		if (!ml) {
-			ip++;
-			continue;
-		}
-
-		/* saved, in case we would skip too much */
-		start0 = ip;
-		ref0 = ref;
-		ml0 = ml;
-
-_Search2:
-		if (ip + ml < mflimit)
-			ml2 = LZ4HC_InsertAndGetWiderMatch(ctx,
-				ip + ml - 2, ip + 0,
-				matchlimit, ml, &ref2,
-				&start2, maxNbAttempts);
-		else
-			ml2 = ml;
-
-		if (ml2 == ml) {
-			/* No better match */
-			if (LZ4HC_encodeSequence(&ip, &op,
-				&anchor, ml, ref, limit, oend))
-				return 0;
-			continue;
-		}
-
-		if (start0 < ip) {
-			if (start2 < ip + ml0) {
-				/* empirical */
-				ip = start0;
-				ref = ref0;
-				ml = ml0;
-			}
-		}
-
-		/* Here, start0 == ip */
-		if ((start2 - ip) < 3) {
-			/* First Match too small : removed */
-			ml = ml2;
-			ip = start2;
-			ref = ref2;
-			goto _Search2;
-		}
-
-_Search3:
-		/*
-		* Currently we have :
-		* ml2 > ml1, and
-		* ip1 + 3 <= ip2 (usually < ip1 + ml1)
-		*/
-		if ((start2 - ip) < OPTIMAL_ML) {
-			int correction;
-			int new_ml = ml;
-
-			if (new_ml > OPTIMAL_ML)
-				new_ml = OPTIMAL_ML;
-			if (ip + new_ml > start2 + ml2 - MINMATCH)
-				new_ml = (int)(start2 - ip) + ml2 - MINMATCH;
-
-			correction = new_ml - (int)(start2 - ip);
-
-			if (correction > 0) {
-				start2 += correction;
-				ref2 += correction;
-				ml2 -= correction;
-			}
-		}
-		/*
-		 * Now, we have start2 = ip + new_ml,
-		 * with new_ml = min(ml, OPTIMAL_ML = 18)
-		 */
-
-		if (start2 + ml2 < mflimit)
-			ml3 = LZ4HC_InsertAndGetWiderMatch(ctx,
-				start2 + ml2 - 3, start2,
-				matchlimit, ml2, &ref3, &start3,
-				maxNbAttempts);
-		else
-			ml3 = ml2;
-
-		if (ml3 == ml2) {
-			/* No better match : 2 sequences to encode */
-			/* ip & ref are known; Now for ml */
-			if (start2 < ip + ml)
-				ml = (int)(start2 - ip);
-			/* Now, encode 2 sequences */
-			if (LZ4HC_encodeSequence(&ip, &op, &anchor,
-				ml, ref, limit, oend))
-				return 0;
-			ip = start2;
-			if (LZ4HC_encodeSequence(&ip, &op, &anchor,
-				ml2, ref2, limit, oend))
-				return 0;
-			continue;
-		}
-
-		if (start3 < ip + ml + 3) {
-			/* Not enough space for match 2 : remove it */
-			if (start3 >= (ip + ml)) {
-				/* can write Seq1 immediately
-				 * ==> Seq2 is removed,
-				 * so Seq3 becomes Seq1
-				 */
-				if (start2 < ip + ml) {
-					int correction = (int)(ip + ml - start2);
-
-					start2 += correction;
-					ref2 += correction;
-					ml2 -= correction;
-					if (ml2 < MINMATCH) {
-						start2 = start3;
-						ref2 = ref3;
-						ml2 = ml3;
-					}
-				}
-
-				if (LZ4HC_encodeSequence(&ip, &op, &anchor,
-					ml, ref, limit, oend))
-					return 0;
-				ip = start3;
-				ref = ref3;
-				ml = ml3;
-
-				start0 = start2;
-				ref0 = ref2;
-				ml0 = ml2;
-				goto _Search2;
-			}
-
-			start2 = start3;
-			ref2 = ref3;
-			ml2 = ml3;
-			goto _Search3;
-		}
-
-		/*
-		* OK, now we have 3 ascending matches;
-		* let's write at least the first one
-		* ip & ref are known; Now for ml
-		*/
-		if (start2 < ip + ml) {
-			if ((start2 - ip) < (int)ML_MASK) {
-				int correction;
-
-				if (ml > OPTIMAL_ML)
-					ml = OPTIMAL_ML;
-				if (ip + ml > start2 + ml2 - MINMATCH)
-					ml = (int)(start2 - ip) + ml2 - MINMATCH;
-				correction = ml - (int)(start2 - ip);
-				if (correction > 0) {
-					start2 += correction;
-					ref2 += correction;
-					ml2 -= correction;
-				}
-			} else
-				ml = (int)(start2 - ip);
-		}
-		if (LZ4HC_encodeSequence(&ip, &op, &anchor, ml,
-			ref, limit, oend))
-			return 0;
-
-		ip = start2;
-		ref = ref2;
-		ml = ml2;
-
-		start2 = start3;
-		ref2 = ref3;
-		ml2 = ml3;
-
-		goto _Search3;
-	}
-
-	/* Encode Last Literals */
-	{
-		int lastRun = (int)(iend - anchor);
-
-		if ((limit)
-			&& (((char *)op - dest) + lastRun + 1
-				+ ((lastRun + 255 - RUN_MASK)/255)
-					> (U32)maxOutputSize)) {
-			/* Check output limit */
-			return 0;
-		}
-		if (lastRun >= (int)RUN_MASK) {
-			*op++ = (RUN_MASK<<ML_BITS);
-			lastRun -= RUN_MASK;
-			for (; lastRun > 254 ; lastRun -= 255)
-				*op++ = 255;
-			*op++ = (BYTE) lastRun;
-		} else
-			*op++ = (BYTE)(lastRun<<ML_BITS);
-		LZ4_memcpy(op, anchor, iend - anchor);
-		op += iend - anchor;
-	}
-
-	/* End */
-	return (int) (((char *)op) - dest);
-}
-
-static int LZ4_compress_HC_extStateHC(
-	void *state,
-	const char *src,
-	char *dst,
-	int srcSize,
-	int maxDstSize,
-	int compressionLevel)
-{
-	LZ4HC_CCtx_internal *ctx = &((LZ4_streamHC_t *)state)->internal_donotuse;
-
-	if (((size_t)(state)&(sizeof(void *) - 1)) != 0) {
-		/* Error : state is not aligned
-		 * for pointers (32 or 64 bits)
-		 */
+	if (lz4hc_state_misaligned(wrkmem))
 		return 0;
-	}
 
-	LZ4HC_init(ctx, (const BYTE *)src);
-
-	if (maxDstSize < LZ4_compressBound(srcSize))
-		return LZ4HC_compress_generic(ctx, src, dst,
-			srcSize, maxDstSize, compressionLevel, limitedOutput);
-	else
-		return LZ4HC_compress_generic(ctx, src, dst,
-			srcSize, maxDstSize, compressionLevel, noLimit);
-}
-
-int LZ4_compress_HC(const char *src, char *dst, int srcSize,
-	int maxDstSize, int compressionLevel, void *wrkmem)
-{
-	return LZ4_compress_HC_extStateHC(wrkmem, src, dst,
-		srcSize, maxDstSize, compressionLevel);
+	return LZ4_compress_HC_extStateHC(wrkmem, src, dst, srcSize,
+					  dstCapacity,
+					  lz4hc_clamp_level(compressionLevel));
 }
 EXPORT_SYMBOL(LZ4_compress_HC);
 
-/**************************************
- *	Streaming Functions
- **************************************/
-void LZ4_resetStreamHC(LZ4_streamHC_t *LZ4_streamHCPtr, int compressionLevel)
+void LZ4_resetStreamHC(LZ4_streamHC_t *streamHCPtr, int compressionLevel)
 {
-	LZ4_streamHCPtr->internal_donotuse.base = NULL;
-	LZ4_streamHCPtr->internal_donotuse.compressionLevel = (unsigned int)compressionLevel;
+	if (lz4hc_state_misaligned(streamHCPtr))
+		return;
+
+	__lz4_resetStreamHC(streamHCPtr, lz4hc_clamp_level(compressionLevel));
 }
 EXPORT_SYMBOL(LZ4_resetStreamHC);
 
-int LZ4_loadDictHC(LZ4_streamHC_t *LZ4_streamHCPtr,
-	const char *dictionary,
-	int dictSize)
+int LZ4_loadDictHC(LZ4_streamHC_t *streamHCPtr, const char *dictionary,
+		   int dictSize)
 {
-	LZ4HC_CCtx_internal *ctxPtr = &LZ4_streamHCPtr->internal_donotuse;
-
-	if (dictSize > 64 * KB) {
-		dictionary += dictSize - 64 * KB;
-		dictSize = 64 * KB;
-	}
-	LZ4HC_init(ctxPtr, (const BYTE *)dictionary);
-	if (dictSize >= 4)
-		LZ4HC_Insert(ctxPtr, (const BYTE *)dictionary + (dictSize - 3));
-	ctxPtr->end = (const BYTE *)dictionary + dictSize;
-	return dictSize;
+	return __lz4_loadDictHC(streamHCPtr, dictionary, dictSize);
 }
 EXPORT_SYMBOL(LZ4_loadDictHC);
 
-/* compression */
-
-static void LZ4HC_setExternalDict(
-	LZ4HC_CCtx_internal *ctxPtr,
-	const BYTE *newBlock)
+int LZ4_compress_HC_continue(LZ4_streamHC_t *streamHCPtr, const char *src,
+			     char *dst, int srcSize, int maxDstSize)
 {
-	if (ctxPtr->end >= ctxPtr->base + 4) {
-		/* Referencing remaining dictionary content */
-		LZ4HC_Insert(ctxPtr, ctxPtr->end - 3);
-	}
-
-	/*
-	 * Only one memory segment for extDict,
-	 * so any previous extDict is lost at this stage
-	 */
-	ctxPtr->lowLimit	= ctxPtr->dictLimit;
-	ctxPtr->dictLimit = (U32)(ctxPtr->end - ctxPtr->base);
-	ctxPtr->dictBase	= ctxPtr->base;
-	ctxPtr->base = newBlock - ctxPtr->dictLimit;
-	ctxPtr->end	= newBlock;
-	/* match referencing will resume from there */
-	ctxPtr->nextToUpdate = ctxPtr->dictLimit;
-}
-
-static int LZ4_compressHC_continue_generic(
-	LZ4_streamHC_t *LZ4_streamHCPtr,
-	const char *source,
-	char *dest,
-	int inputSize,
-	int maxOutputSize,
-	limitedOutput_directive limit)
-{
-	LZ4HC_CCtx_internal *ctxPtr = &LZ4_streamHCPtr->internal_donotuse;
-
-	/* auto - init if forgotten */
-	if (ctxPtr->base == NULL)
-		LZ4HC_init(ctxPtr, (const BYTE *) source);
-
-	/* Check overflow */
-	if ((size_t)(ctxPtr->end - ctxPtr->base) > 2 * GB) {
-		size_t dictSize = (size_t)(ctxPtr->end - ctxPtr->base)
-			- ctxPtr->dictLimit;
-		if (dictSize > 64 * KB)
-			dictSize = 64 * KB;
-		LZ4_loadDictHC(LZ4_streamHCPtr,
-			(const char *)(ctxPtr->end) - dictSize, (int)dictSize);
-	}
-
-	/* Check if blocks follow each other */
-	if ((const BYTE *)source != ctxPtr->end)
-		LZ4HC_setExternalDict(ctxPtr, (const BYTE *)source);
-
-	/* Check overlapping input/dictionary space */
-	{
-		const BYTE *sourceEnd = (const BYTE *) source + inputSize;
-		const BYTE * const dictBegin = ctxPtr->dictBase + ctxPtr->lowLimit;
-		const BYTE * const dictEnd = ctxPtr->dictBase + ctxPtr->dictLimit;
-
-		if ((sourceEnd > dictBegin)
-			&& ((const BYTE *)source < dictEnd)) {
-			if (sourceEnd > dictEnd)
-				sourceEnd = dictEnd;
-			ctxPtr->lowLimit = (U32)(sourceEnd - ctxPtr->dictBase);
-
-			if (ctxPtr->dictLimit - ctxPtr->lowLimit < 4)
-				ctxPtr->lowLimit = ctxPtr->dictLimit;
-		}
-	}
-
-	return LZ4HC_compress_generic(ctxPtr, source, dest,
-		inputSize, maxOutputSize, ctxPtr->compressionLevel, limit);
-}
-
-int LZ4_compress_HC_continue(
-	LZ4_streamHC_t *LZ4_streamHCPtr,
-	const char *source,
-	char *dest,
-	int inputSize,
-	int maxOutputSize)
-{
-	if (maxOutputSize < LZ4_compressBound(inputSize))
-		return LZ4_compressHC_continue_generic(LZ4_streamHCPtr,
-			source, dest, inputSize, maxOutputSize, limitedOutput);
-	else
-		return LZ4_compressHC_continue_generic(LZ4_streamHCPtr,
-			source, dest, inputSize, maxOutputSize, noLimit);
+	return __lz4_compress_HC_continue(streamHCPtr, src, dst, srcSize,
+					  maxDstSize);
 }
 EXPORT_SYMBOL(LZ4_compress_HC_continue);
 
-/* dictionary saving */
-
-int LZ4_saveDictHC(
-	LZ4_streamHC_t *LZ4_streamHCPtr,
-	char *safeBuffer,
-	int dictSize)
+int LZ4_saveDictHC(LZ4_streamHC_t *streamHCPtr, char *safeBuffer,
+		   int maxDictSize)
 {
-	LZ4HC_CCtx_internal *const streamPtr = &LZ4_streamHCPtr->internal_donotuse;
-	int const prefixSize = (int)(streamPtr->end
-		- (streamPtr->base + streamPtr->dictLimit));
-
-	if (dictSize > 64 * KB)
-		dictSize = 64 * KB;
-	if (dictSize < 4)
-		dictSize = 0;
-	if (dictSize > prefixSize)
-		dictSize = prefixSize;
-
-	memmove(safeBuffer, streamPtr->end - dictSize, dictSize);
-
-	{
-		U32 const endIndex = (U32)(streamPtr->end - streamPtr->base);
-
-		streamPtr->end = (const BYTE *)safeBuffer + dictSize;
-		streamPtr->base = streamPtr->end - endIndex;
-		streamPtr->dictLimit = endIndex - dictSize;
-		streamPtr->lowLimit = endIndex - dictSize;
-
-		if (streamPtr->nextToUpdate < streamPtr->dictLimit)
-			streamPtr->nextToUpdate = streamPtr->dictLimit;
-	}
-	return dictSize;
+	return __lz4_saveDictHC(streamHCPtr, safeBuffer, maxDictSize);
 }
 EXPORT_SYMBOL(LZ4_saveDictHC);
 

-- 
2.34.1



_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

[PATCH v2 11/15] lib/lz4: switch the HC compressor to the vendored sources

[Michal Wilczynski]

Replace the forked LZ4HC with thin entry points over the vendored
lz4hc.c.

Upstream routes levels 10 and up to its optimal parser, whose ~64K
workspace the kernel cannot provide. Clamp those levels to 9 rather than
reject them, so existing f2fs and zram settings keep working, and add
LZ4HC_CLAMP_CLEVEL to <linux/lz4.h>. Document the clamp in f2fs.rst.

Levels 1 and 2 now use upstream's lz4mid: on kernel source they are
50-84% faster for 4-9% larger output. Levels 3 to 9 compress the same,
but at 4K are up to 15% slower on poorly compressible data, including at
level 9, zram's lz4hc default.

LZ4_compress_HC() and LZ4_resetStreamHC() check the state's alignment
like the compressor's entry points.

LZ4_streamHC_t becomes incomplete and the macros that laid out the forked
structure go. LZ4HC_MEM_COMPRESS grows from 262192 to upstream's 262200,
static_asserted against LZ4_STREAMHC_MINSIZE. lz4hc.c pulls in lz4.c's
decoder tables, unused here when the fast decode loop is off, so silence
-Wunused-const-variable for this object.

This retires commit b08918fb3f27 ("lz4: do not export static symbol"):
the glue exports only the five LZ4HC entry points.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 Documentation/filesystems/f2fs.rst |   2 +
 drivers/block/zram/backend_lz4hc.c |   2 +-
 include/linux/lz4.h                |  54 +--
 lib/lz4/Makefile                   |   4 +
 lib/lz4/lz4hc_compress.c           | 792 ++++---------------------------------
 5 files changed, 91 insertions(+), 763 deletions(-)

diff --git a/Documentation/filesystems/f2fs.rst b/Documentation/filesystems/f2fs.rst
index 771216f45207a797356d5d2a10328b74d1ab302a..4a2d4545c2752cbcea4939839ba3d2457bf92299 100644
--- a/Documentation/filesystems/f2fs.rst
+++ b/Documentation/filesystems/f2fs.rst
@@ -318,6 +318,8 @@ compress_algorithm=%s:%d Control compress algorithm and its compress level, now,
 				 zstd           1 - 22
 				 =========      ===========
 
+			 lz4 levels above 9 compress as level 9.
+
 compress_log_size=%u	 Support configuring compress cluster size. The size will
 			 be 4KB * (1 << %u). The default and minimum sizes are 16KB.
 compress_extension=%s	 Support adding specified extension, so that f2fs can enable
diff --git a/drivers/block/zram/backend_lz4hc.c b/drivers/block/zram/backend_lz4hc.c
index 06286c805cf7c56bd760f460cc854dc4014f293b..e7ee3740219852f079d1e0866da6b213a3f3a583 100644
--- a/drivers/block/zram/backend_lz4hc.c
+++ b/drivers/block/zram/backend_lz4hc.c
@@ -69,7 +69,7 @@ static int lz4hc_create(struct zcomp_params *params, struct zcomp_ctx *ctx)
 		if (!zctx->dstrm)
 			goto error;
 
-		zctx->cstrm = kzalloc_obj(*zctx->cstrm);
+		zctx->cstrm = kzalloc(LZ4HC_MEM_COMPRESS, GFP_KERNEL);
 		if (!zctx->cstrm)
 			goto error;
 		LZ4_resetStreamHC(zctx->cstrm, params->level);
diff --git a/include/linux/lz4.h b/include/linux/lz4.h
index 031a410b8e3878e9a24cc49af36cfcbba399d991..05060b9790be4621ddbe744172fa0ae37f6191f4 100644
--- a/include/linux/lz4.h
+++ b/include/linux/lz4.h
@@ -59,19 +59,16 @@
 #define LZ4HC_DEFAULT_CLEVEL			9
 #define LZ4HC_MAX_CLEVEL			16
 
-#define LZ4HC_DICTIONARY_LOGSIZE 16
-#define LZ4HC_MAXD (1<<LZ4HC_DICTIONARY_LOGSIZE)
-#define LZ4HC_MAXD_MASK (LZ4HC_MAXD - 1)
-#define LZ4HC_HASH_LOG (LZ4HC_DICTIONARY_LOGSIZE - 1)
-#define LZ4HC_HASHTABLESIZE (1 << LZ4HC_HASH_LOG)
-#define LZ4HC_HASH_MASK (LZ4HC_HASHTABLESIZE - 1)
+/* Levels from here up select the optimal parser, which needs a ~64K
+ * workspace lib/lz4 cannot give it, so lib/lz4 compresses them as
+ * LZ4HC_CLAMP_CLEVEL - 1.  f2fs and zram still accept levels up to
+ * LZ4HC_MAX_CLEVEL.
+ */
+#define LZ4HC_CLAMP_CLEVEL			10
 
 /*-************************************************************************
  *	STREAMING CONSTANTS AND STRUCTURES
  **************************************************************************/
-#define LZ4_STREAMHCSIZE        262192
-#define LZ4_STREAMHCSIZE_SIZET (262192 / sizeof(size_t))
-
 #define LZ4_STREAMDECODESIZE_U64	4
 #define LZ4_STREAMDECODESIZE		 (LZ4_STREAMDECODESIZE_U64 * \
 	sizeof(unsigned long long))
@@ -83,29 +80,10 @@
 typedef union LZ4_stream_u LZ4_stream_t;
 
 /*
- * LZ4_streamHC_t - information structure to track an LZ4HC stream.
+ * LZ4_streamHC_t - an LZ4HC stream.  Incomplete: lib/lz4 owns the layout.
+ * Allocate LZ4HC_MEM_COMPRESS bytes and cast, do not sizeof().
  */
-typedef struct {
-	unsigned int	 hashTable[LZ4HC_HASHTABLESIZE];
-	unsigned short	 chainTable[LZ4HC_MAXD];
-	/* next block to continue on current prefix */
-	const unsigned char *end;
-	/* All index relative to this position */
-	const unsigned char *base;
-	/* alternate base for extDict */
-	const unsigned char *dictBase;
-	/* below that point, need extDict */
-	unsigned int	 dictLimit;
-	/* below that point, no more dict */
-	unsigned int	 lowLimit;
-	/* index from which to continue dict update */
-	unsigned int	 nextToUpdate;
-	unsigned int	 compressionLevel;
-} LZ4HC_CCtx_internal;
-typedef union {
-	size_t table[LZ4_STREAMHCSIZE_SIZET];
-	LZ4HC_CCtx_internal internal_donotuse;
-} LZ4_streamHC_t;
+typedef union LZ4_streamHC_u LZ4_streamHC_t;
 
 /*
  * LZ4_streamDecode_t - information structure to track an
@@ -133,7 +111,7 @@ typedef union {
  * points warn about a misaligned buffer and fail the call.
  */
 #define LZ4_MEM_COMPRESS	16416
-#define LZ4HC_MEM_COMPRESS	LZ4_STREAMHCSIZE
+#define LZ4HC_MEM_COMPRESS	262200
 
 /*-************************************************************************
  *	Compression Functions
@@ -289,9 +267,9 @@ int LZ4_decompress_safe_partial(const char *source, char *dest,
  * @srcSize: size of the input data. Max supported value is LZ4_MAX_INPUT_SIZE
  * @dstCapacity: full or partial size of buffer 'dst',
  *	which must be already allocated
- * @compressionLevel: Recommended values are between 4 and 9, although any
- *	value between 1 and LZ4HC_MAX_CLEVEL will work.
- *	Values >LZ4HC_MAX_CLEVEL behave the same as 16.
+ * @compressionLevel: Recommended values are between 4 and 9.  Levels of
+ *	LZ4HC_CLAMP_CLEVEL and above are clamped to LZ4HC_CLAMP_CLEVEL - 1;
+ *	see that macro.
  * @wrkmem: address of the working memory.
  *	This requires 'wrkmem' of size LZ4HC_MEM_COMPRESS, aligned to 8 bytes.
  *
@@ -307,9 +285,9 @@ int LZ4_compress_HC(const char *src, char *dst, int srcSize, int dstCapacity,
 /**
  * LZ4_resetStreamHC() - Init an allocated 'LZ4_streamHC_t' structure
  * @streamHCPtr: pointer to the 'LZ4_streamHC_t' structure
- * @compressionLevel: Recommended values are between 4 and 9, although any
- *	value between 1 and LZ4HC_MAX_CLEVEL will work.
- *	Values >LZ4HC_MAX_CLEVEL behave the same as 16.
+ * @compressionLevel: Recommended values are between 4 and 9.  Levels of
+ *	LZ4HC_CLAMP_CLEVEL and above are clamped to LZ4HC_CLAMP_CLEVEL - 1;
+ *	see that macro.
  *
  * An LZ4_streamHC_t structure can be allocated once
  * and re-used multiple times.
diff --git a/lib/lz4/Makefile b/lib/lz4/Makefile
index 8b4eb07694e67dd99ec7cc85212fb68138d3981e..5cbbe10beedce1199dee68992d542a1e52f89f8e 100644
--- a/lib/lz4/Makefile
+++ b/lib/lz4/Makefile
@@ -10,3 +10,7 @@ ccflags-y += -I $(src)/freestanding
 obj-$(CONFIG_LZ4_COMPRESS)	+= lz4_compress.o
 obj-$(CONFIG_LZ4HC_COMPRESS)	+= lz4hc_compress.o
 obj-$(CONFIG_LZ4_DECOMPRESS)	+= lz4_decompress.o
+
+# lz4hc.c pulls in lz4.c's common definitions, which include two decoder
+# tables nothing here uses when LZ4_FAST_DEC_LOOP is off.
+CFLAGS_lz4hc_compress.o += $(call cc-disable-warning, unused-const-variable)
diff --git a/lib/lz4/lz4hc_compress.c b/lib/lz4/lz4hc_compress.c
index 91936dc3d14bca31752a361693380aad8eb3b98a..4d56d45028110805628be766b2758a8dffd56f0c 100644
--- a/lib/lz4/lz4hc_compress.c
+++ b/lib/lz4/lz4hc_compress.c
@@ -1,766 +1,110 @@
+// SPDX-License-Identifier: GPL-2.0-only OR BSD-2-Clause
 /*
- * LZ4 HC - High Compression Mode of LZ4
- * Copyright (C) 2011-2015, Yann Collet.
+ * Copyright (C) 2011 - 2016, Yann Collet.
+ * Copyright (C) 2016, Sven Schmidt <4sschmid@informatik.uni-hamburg.de>
+ * Copyright (c) 2026 Samsung Electronics Co., Ltd.
+ * Author: Michal Wilczynski <m.wilczynski@samsung.com>
  *
- * BSD 2 - Clause License (http://www.opensource.org/licenses/bsd - license.php)
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions are
- * met:
- *	* Redistributions of source code must retain the above copyright
- *	  notice, this list of conditions and the following disclaimer.
- *	* Redistributions in binary form must reproduce the above
- * copyright notice, this list of conditions and the following disclaimer
- * in the documentation and/or other materials provided with the
- * distribution.
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
- * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
- * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
- * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
- * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
- * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
- * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
- * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
- * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- * You can contact the author at :
- *	- LZ4 homepage : http://www.lz4.org
- *	- LZ4 source repository : https://github.com/lz4/lz4
+ * LZ4 HC compressor -- kernel entry points
  *
- *	Changed for kernel usage by:
- *	Sven Schmidt <4sschmid@informatik.uni-hamburg.de>
+ * The compressor is upstream's, in the verbatim lz4hc.c included below.
+ * LZ4_compress_HC() takes the kernel's trailing wrkmem argument and forwards
+ * to LZ4_compress_HC_extStateHC(); the level-taking entry points clamp the
+ * level below LZ4HC_CLEVEL_OPT_MIN (see lz4hc_clamp_level()).
  */
 
-/*-************************************
- *	Dependencies
- **************************************/
-#include "lz4defs.h"
-#include <linux/module.h>
-#include <linux/kernel.h>
-#include <linux/string.h> /* memset */
-
-/* *************************************
- *	Local Constants and types
- ***************************************/
-
-#define OPTIMAL_ML (int)((ML_MASK - 1) + MINMATCH)
-
-#define HASH_FUNCTION(i)	(((i) * 2654435761U) \
-	>> ((MINMATCH*8) - LZ4HC_HASH_LOG))
-#define DELTANEXTU16(p)	chainTable[(U16)(p)] /* faster */
-
-static U32 LZ4HC_hashPtr(const void *ptr)
-{
-	return HASH_FUNCTION(LZ4_read32(ptr));
-}
+#include "lz4_deps.h"
 
-/**************************************
- *	HC Compression
- **************************************/
-static void LZ4HC_init(LZ4HC_CCtx_internal *hc4, const BYTE *start)
-{
-	memset((void *)hc4->hashTable, 0, sizeof(hc4->hashTable));
-	memset(hc4->chainTable, 0xFF, sizeof(hc4->chainTable));
-	hc4->nextToUpdate = 64 * KB;
-	hc4->base = start - 64 * KB;
-	hc4->end = start;
-	hc4->dictBase = start - 64 * KB;
-	hc4->dictLimit = 64 * KB;
-	hc4->lowLimit = 64 * KB;
-}
-
-/* Update chains up to ip (excluded) */
-static FORCE_INLINE void LZ4HC_Insert(LZ4HC_CCtx_internal *hc4,
-	const BYTE *ip)
+/* lz4hc.c includes lz4.c with LZ4_COMMONDEFS_ONLY, which stops short of
+ * LZ4_compressBound().  Supply it for lz4hc.c's internal use.
+ */
+static int LZ4_compressBound(int isize)
 {
-	U16 * const chainTable = hc4->chainTable;
-	U32 * const hashTable	= hc4->hashTable;
-	const BYTE * const base = hc4->base;
-	U32 const target = (U32)(ip - base);
-	U32 idx = hc4->nextToUpdate;
-
-	while (idx < target) {
-		U32 const h = LZ4HC_hashPtr(base + idx);
-		size_t delta = idx - hashTable[h];
-
-		if (delta > MAX_DISTANCE)
-			delta = MAX_DISTANCE;
-
-		DELTANEXTU16(idx) = (U16)delta;
-
-		hashTable[h] = idx;
-		idx++;
-	}
-
-	hc4->nextToUpdate = target;
+	return LZ4_COMPRESSBOUND(isize);
 }
 
-static FORCE_INLINE int LZ4HC_InsertAndFindBestMatch(
-	LZ4HC_CCtx_internal *hc4, /* Index table will be updated */
-	const BYTE *ip,
-	const BYTE * const iLimit,
-	const BYTE **matchpos,
-	const int maxNbAttempts)
-{
-	U16 * const chainTable = hc4->chainTable;
-	U32 * const HashTable = hc4->hashTable;
-	const BYTE * const base = hc4->base;
-	const BYTE * const dictBase = hc4->dictBase;
-	const U32 dictLimit = hc4->dictLimit;
-	const U32 lowLimit = (hc4->lowLimit + 64 * KB > (U32)(ip - base))
-		? hc4->lowLimit
-		: (U32)(ip - base) - (64 * KB - 1);
-	U32 matchIndex;
-	int nbAttempts = maxNbAttempts;
-	size_t ml = 0;
-
-	/* HC4 match finder */
-	LZ4HC_Insert(hc4, ip);
-	matchIndex = HashTable[LZ4HC_hashPtr(ip)];
-
-	while ((matchIndex >= lowLimit)
-		&& (nbAttempts)) {
-		nbAttempts--;
-		if (matchIndex >= dictLimit) {
-			const BYTE * const match = base + matchIndex;
-
-			if (*(match + ml) == *(ip + ml)
-				&& (LZ4_read32(match) == LZ4_read32(ip))) {
-				size_t const mlt = LZ4_count(ip + MINMATCH,
-					match + MINMATCH, iLimit) + MINMATCH;
-
-				if (mlt > ml) {
-					ml = mlt;
-					*matchpos = match;
-				}
-			}
-		} else {
-			const BYTE * const match = dictBase + matchIndex;
+#include "upstream/lz4hc.c"
 
-			if (LZ4_read32(match) == LZ4_read32(ip)) {
-				size_t mlt;
-				const BYTE *vLimit = ip
-					+ (dictLimit - matchIndex);
+/* Upstream's short helper macros; MIN and MAX clash with <linux/minmax.h>. */
+#undef MIN
+#undef MAX
+#undef KB
+#undef MB
+#undef GB
 
-				if (vLimit > iLimit)
-					vLimit = iLimit;
-				mlt = LZ4_count(ip + MINMATCH,
-					match + MINMATCH, vLimit) + MINMATCH;
-				if ((ip + mlt == vLimit)
-					&& (vLimit < iLimit))
-					mlt += LZ4_count(ip + mlt,
-						base + dictLimit,
-						iLimit);
-				if (mlt > ml) {
-					/* virtual matchpos */
-					ml = mlt;
-					*matchpos = base + matchIndex;
-				}
-			}
-		}
-		matchIndex -= DELTANEXTU16(matchIndex);
-	}
+#include "lz4_kernel_api.h"
+#include <linux/bug.h>
+#include <linux/export.h>
+#include <linux/module.h>
 
-	return (int)ml;
-}
+/* Catch any divergence from upstream's layout at build time. */
+static_assert(sizeof(LZ4_streamHC_t) == LZ4_STREAMHC_MINSIZE);
+static_assert(LZ4_STREAMHC_MINSIZE == 262200); /* LZ4HC_MEM_COMPRESS */
 
-static FORCE_INLINE int LZ4HC_InsertAndGetWiderMatch(
-	LZ4HC_CCtx_internal *hc4,
-	const BYTE * const ip,
-	const BYTE * const iLowLimit,
-	const BYTE * const iHighLimit,
-	int longest,
-	const BYTE **matchpos,
-	const BYTE **startpos,
-	const int maxNbAttempts)
+/* Levels >= LZ4HC_CLEVEL_OPT_MIN (10) reach the optimal parser, which needs
+ * a ~64K opt[]: too big for a kernel stack, and in the heap mode lz4_deps.h
+ * selects its allocation always fails, and so would the call.  Clamp them
+ * to 9 rather than reject them, so existing f2fs and zram settings keep
+ * working.  Levels 3..9 keep their meaning; 1 and 2 are upstream's lz4mid,
+ * faster and weaker than the shallow hash chain the fork used for them.
+ */
+static int lz4hc_clamp_level(int compressionLevel)
 {
-	U16 * const chainTable = hc4->chainTable;
-	U32 * const HashTable = hc4->hashTable;
-	const BYTE * const base = hc4->base;
-	const U32 dictLimit = hc4->dictLimit;
-	const BYTE * const lowPrefixPtr = base + dictLimit;
-	const U32 lowLimit = (hc4->lowLimit + 64 * KB > (U32)(ip - base))
-		? hc4->lowLimit
-		: (U32)(ip - base) - (64 * KB - 1);
-	const BYTE * const dictBase = hc4->dictBase;
-	U32 matchIndex;
-	int nbAttempts = maxNbAttempts;
-	int delta = (int)(ip - iLowLimit);
-
-	/* First Match */
-	LZ4HC_Insert(hc4, ip);
-	matchIndex = HashTable[LZ4HC_hashPtr(ip)];
-
-	while ((matchIndex >= lowLimit)
-		&& (nbAttempts)) {
-		nbAttempts--;
-		if (matchIndex >= dictLimit) {
-			const BYTE *matchPtr = base + matchIndex;
-
-			if (*(iLowLimit + longest)
-				== *(matchPtr - delta + longest)) {
-				if (LZ4_read32(matchPtr) == LZ4_read32(ip)) {
-					int mlt = MINMATCH + LZ4_count(
-						ip + MINMATCH,
-						matchPtr + MINMATCH,
-						iHighLimit);
-					int back = 0;
-
-					while ((ip + back > iLowLimit)
-						&& (matchPtr + back > lowPrefixPtr)
-						&& (ip[back - 1] == matchPtr[back - 1]))
-						back--;
+	if (compressionLevel >= LZ4HC_CLEVEL_OPT_MIN)
+		return LZ4HC_CLEVEL_OPT_MIN - 1;
 
-					mlt -= back;
-
-					if (mlt > longest) {
-						longest = (int)mlt;
-						*matchpos = matchPtr + back;
-						*startpos = ip + back;
-					}
-				}
-			}
-		} else {
-			const BYTE * const matchPtr = dictBase + matchIndex;
-
-			if (LZ4_read32(matchPtr) == LZ4_read32(ip)) {
-				size_t mlt;
-				int back = 0;
-				const BYTE *vLimit = ip + (dictLimit - matchIndex);
-
-				if (vLimit > iHighLimit)
-					vLimit = iHighLimit;
-
-				mlt = LZ4_count(ip + MINMATCH,
-					matchPtr + MINMATCH, vLimit) + MINMATCH;
-
-				if ((ip + mlt == vLimit) && (vLimit < iHighLimit))
-					mlt += LZ4_count(ip + mlt, base + dictLimit,
-						iHighLimit);
-				while ((ip + back > iLowLimit)
-					&& (matchIndex + back > lowLimit)
-					&& (ip[back - 1] == matchPtr[back - 1]))
-					back--;
-
-				mlt -= back;
-
-				if ((int)mlt > longest) {
-					longest = (int)mlt;
-					*matchpos = base + matchIndex + back;
-					*startpos = ip + back;
-				}
-			}
-		}
-
-		matchIndex -= DELTANEXTU16(matchIndex);
-	}
-
-	return longest;
+	return compressionLevel;
 }
 
-static FORCE_INLINE int LZ4HC_encodeSequence(
-	const BYTE **ip,
-	BYTE **op,
-	const BYTE **anchor,
-	int matchLength,
-	const BYTE * const match,
-	limitedOutput_directive limitedOutputBuffer,
-	BYTE *oend)
+/* LZ4_initStreamHC() returns NULL for a misaligned state.
+ * LZ4_compress_HC_extStateHC() then fails quietly, and LZ4_resetStreamHC()
+ * ignores it and leaves the state uninitialised.  Warn, and skip the reset.
+ */
+static bool lz4hc_state_misaligned(const void *state)
 {
-	int length;
-	BYTE *token;
-
-	/* Encode Literal length */
-	length = (int)(*ip - *anchor);
-	token = (*op)++;
-
-	if ((limitedOutputBuffer)
-		&& ((*op + (length>>8)
-			+ length + (2 + 1 + LASTLITERALS)) > oend)) {
-		/* Check output limit */
-		return 1;
-	}
-	if (length >= (int)RUN_MASK) {
-		int len;
-
-		*token = (RUN_MASK<<ML_BITS);
-		len = length - RUN_MASK;
-		for (; len > 254 ; len -= 255)
-			*(*op)++ = 255;
-		*(*op)++ = (BYTE)len;
-	} else
-		*token = (BYTE)(length<<ML_BITS);
-
-	/* Copy Literals */
-	LZ4_wildCopy(*op, *anchor, (*op) + length);
-	*op += length;
-
-	/* Encode Offset */
-	LZ4_writeLE16(*op, (U16)(*ip - match));
-	*op += 2;
-
-	/* Encode MatchLength */
-	length = (int)(matchLength - MINMATCH);
-
-	if ((limitedOutputBuffer)
-		&& (*op + (length>>8)
-			+ (1 + LASTLITERALS) > oend)) {
-		/* Check output limit */
-		return 1;
-	}
-
-	if (length >= (int)ML_MASK) {
-		*token += ML_MASK;
-		length -= ML_MASK;
-
-		for (; length > 509 ; length -= 510) {
-			*(*op)++ = 255;
-			*(*op)++ = 255;
-		}
-
-		if (length > 254) {
-			length -= 255;
-			*(*op)++ = 255;
-		}
-
-		*(*op)++ = (BYTE)length;
-	} else
-		*token += (BYTE)(length);
-
-	/* Prepare next loop */
-	*ip += matchLength;
-	*anchor = *ip;
-
-	return 0;
+	return WARN_ON_ONCE(!LZ4_isAligned(state, LZ4_streamHC_t_alignment()));
 }
 
-static int LZ4HC_compress_generic(
-	LZ4HC_CCtx_internal *const ctx,
-	const char * const source,
-	char * const dest,
-	int const inputSize,
-	int const maxOutputSize,
-	int compressionLevel,
-	limitedOutput_directive limit
-	)
+int LZ4_compress_HC(const char *src, char *dst, int srcSize, int dstCapacity,
+		    int compressionLevel, void *wrkmem)
 {
-	const BYTE *ip = (const BYTE *) source;
-	const BYTE *anchor = ip;
-	const BYTE * const iend = ip + inputSize;
-	const BYTE * const mflimit = iend - MFLIMIT;
-	const BYTE * const matchlimit = (iend - LASTLITERALS);
-
-	BYTE *op = (BYTE *) dest;
-	BYTE * const oend = op + maxOutputSize;
-
-	unsigned int maxNbAttempts;
-	int ml, ml2, ml3, ml0;
-	const BYTE *ref = NULL;
-	const BYTE *start2 = NULL;
-	const BYTE *ref2 = NULL;
-	const BYTE *start3 = NULL;
-	const BYTE *ref3 = NULL;
-	const BYTE *start0;
-	const BYTE *ref0;
-
-	/* init */
-	if (compressionLevel > LZ4HC_MAX_CLEVEL)
-		compressionLevel = LZ4HC_MAX_CLEVEL;
-	if (compressionLevel < 1)
-		compressionLevel = LZ4HC_DEFAULT_CLEVEL;
-	maxNbAttempts = 1 << (compressionLevel - 1);
-	ctx->end += inputSize;
-
-	ip++;
-
-	/* Main Loop */
-	while (ip < mflimit) {
-		ml = LZ4HC_InsertAndFindBestMatch(ctx, ip,
-			matchlimit, (&ref), maxNbAttempts);
-		if (!ml) {
-			ip++;
-			continue;
-		}
-
-		/* saved, in case we would skip too much */
-		start0 = ip;
-		ref0 = ref;
-		ml0 = ml;
-
-_Search2:
-		if (ip + ml < mflimit)
-			ml2 = LZ4HC_InsertAndGetWiderMatch(ctx,
-				ip + ml - 2, ip + 0,
-				matchlimit, ml, &ref2,
-				&start2, maxNbAttempts);
-		else
-			ml2 = ml;
-
-		if (ml2 == ml) {
-			/* No better match */
-			if (LZ4HC_encodeSequence(&ip, &op,
-				&anchor, ml, ref, limit, oend))
-				return 0;
-			continue;
-		}
-
-		if (start0 < ip) {
-			if (start2 < ip + ml0) {
-				/* empirical */
-				ip = start0;
-				ref = ref0;
-				ml = ml0;
-			}
-		}
-
-		/* Here, start0 == ip */
-		if ((start2 - ip) < 3) {
-			/* First Match too small : removed */
-			ml = ml2;
-			ip = start2;
-			ref = ref2;
-			goto _Search2;
-		}
-
-_Search3:
-		/*
-		* Currently we have :
-		* ml2 > ml1, and
-		* ip1 + 3 <= ip2 (usually < ip1 + ml1)
-		*/
-		if ((start2 - ip) < OPTIMAL_ML) {
-			int correction;
-			int new_ml = ml;
-
-			if (new_ml > OPTIMAL_ML)
-				new_ml = OPTIMAL_ML;
-			if (ip + new_ml > start2 + ml2 - MINMATCH)
-				new_ml = (int)(start2 - ip) + ml2 - MINMATCH;
-
-			correction = new_ml - (int)(start2 - ip);
-
-			if (correction > 0) {
-				start2 += correction;
-				ref2 += correction;
-				ml2 -= correction;
-			}
-		}
-		/*
-		 * Now, we have start2 = ip + new_ml,
-		 * with new_ml = min(ml, OPTIMAL_ML = 18)
-		 */
-
-		if (start2 + ml2 < mflimit)
-			ml3 = LZ4HC_InsertAndGetWiderMatch(ctx,
-				start2 + ml2 - 3, start2,
-				matchlimit, ml2, &ref3, &start3,
-				maxNbAttempts);
-		else
-			ml3 = ml2;
-
-		if (ml3 == ml2) {
-			/* No better match : 2 sequences to encode */
-			/* ip & ref are known; Now for ml */
-			if (start2 < ip + ml)
-				ml = (int)(start2 - ip);
-			/* Now, encode 2 sequences */
-			if (LZ4HC_encodeSequence(&ip, &op, &anchor,
-				ml, ref, limit, oend))
-				return 0;
-			ip = start2;
-			if (LZ4HC_encodeSequence(&ip, &op, &anchor,
-				ml2, ref2, limit, oend))
-				return 0;
-			continue;
-		}
-
-		if (start3 < ip + ml + 3) {
-			/* Not enough space for match 2 : remove it */
-			if (start3 >= (ip + ml)) {
-				/* can write Seq1 immediately
-				 * ==> Seq2 is removed,
-				 * so Seq3 becomes Seq1
-				 */
-				if (start2 < ip + ml) {
-					int correction = (int)(ip + ml - start2);
-
-					start2 += correction;
-					ref2 += correction;
-					ml2 -= correction;
-					if (ml2 < MINMATCH) {
-						start2 = start3;
-						ref2 = ref3;
-						ml2 = ml3;
-					}
-				}
-
-				if (LZ4HC_encodeSequence(&ip, &op, &anchor,
-					ml, ref, limit, oend))
-					return 0;
-				ip = start3;
-				ref = ref3;
-				ml = ml3;
-
-				start0 = start2;
-				ref0 = ref2;
-				ml0 = ml2;
-				goto _Search2;
-			}
-
-			start2 = start3;
-			ref2 = ref3;
-			ml2 = ml3;
-			goto _Search3;
-		}
-
-		/*
-		* OK, now we have 3 ascending matches;
-		* let's write at least the first one
-		* ip & ref are known; Now for ml
-		*/
-		if (start2 < ip + ml) {
-			if ((start2 - ip) < (int)ML_MASK) {
-				int correction;
-
-				if (ml > OPTIMAL_ML)
-					ml = OPTIMAL_ML;
-				if (ip + ml > start2 + ml2 - MINMATCH)
-					ml = (int)(start2 - ip) + ml2 - MINMATCH;
-				correction = ml - (int)(start2 - ip);
-				if (correction > 0) {
-					start2 += correction;
-					ref2 += correction;
-					ml2 -= correction;
-				}
-			} else
-				ml = (int)(start2 - ip);
-		}
-		if (LZ4HC_encodeSequence(&ip, &op, &anchor, ml,
-			ref, limit, oend))
-			return 0;
-
-		ip = start2;
-		ref = ref2;
-		ml = ml2;
-
-		start2 = start3;
-		ref2 = ref3;
-		ml2 = ml3;
-
-		goto _Search3;
-	}
-
-	/* Encode Last Literals */
-	{
-		int lastRun = (int)(iend - anchor);
-
-		if ((limit)
-			&& (((char *)op - dest) + lastRun + 1
-				+ ((lastRun + 255 - RUN_MASK)/255)
-					> (U32)maxOutputSize)) {
-			/* Check output limit */
-			return 0;
-		}
-		if (lastRun >= (int)RUN_MASK) {
-			*op++ = (RUN_MASK<<ML_BITS);
-			lastRun -= RUN_MASK;
-			for (; lastRun > 254 ; lastRun -= 255)
-				*op++ = 255;
-			*op++ = (BYTE) lastRun;
-		} else
-			*op++ = (BYTE)(lastRun<<ML_BITS);
-		LZ4_memcpy(op, anchor, iend - anchor);
-		op += iend - anchor;
-	}
-
-	/* End */
-	return (int) (((char *)op) - dest);
-}
-
-static int LZ4_compress_HC_extStateHC(
-	void *state,
-	const char *src,
-	char *dst,
-	int srcSize,
-	int maxDstSize,
-	int compressionLevel)
-{
-	LZ4HC_CCtx_internal *ctx = &((LZ4_streamHC_t *)state)->internal_donotuse;
-
-	if (((size_t)(state)&(sizeof(void *) - 1)) != 0) {
-		/* Error : state is not aligned
-		 * for pointers (32 or 64 bits)
-		 */
+	if (lz4hc_state_misaligned(wrkmem))
 		return 0;
-	}
 
-	LZ4HC_init(ctx, (const BYTE *)src);
-
-	if (maxDstSize < LZ4_compressBound(srcSize))
-		return LZ4HC_compress_generic(ctx, src, dst,
-			srcSize, maxDstSize, compressionLevel, limitedOutput);
-	else
-		return LZ4HC_compress_generic(ctx, src, dst,
-			srcSize, maxDstSize, compressionLevel, noLimit);
-}
-
-int LZ4_compress_HC(const char *src, char *dst, int srcSize,
-	int maxDstSize, int compressionLevel, void *wrkmem)
-{
-	return LZ4_compress_HC_extStateHC(wrkmem, src, dst,
-		srcSize, maxDstSize, compressionLevel);
+	return LZ4_compress_HC_extStateHC(wrkmem, src, dst, srcSize,
+					  dstCapacity,
+					  lz4hc_clamp_level(compressionLevel));
 }
 EXPORT_SYMBOL(LZ4_compress_HC);
 
-/**************************************
- *	Streaming Functions
- **************************************/
-void LZ4_resetStreamHC(LZ4_streamHC_t *LZ4_streamHCPtr, int compressionLevel)
+void LZ4_resetStreamHC(LZ4_streamHC_t *streamHCPtr, int compressionLevel)
 {
-	LZ4_streamHCPtr->internal_donotuse.base = NULL;
-	LZ4_streamHCPtr->internal_donotuse.compressionLevel = (unsigned int)compressionLevel;
+	if (lz4hc_state_misaligned(streamHCPtr))
+		return;
+
+	__lz4_resetStreamHC(streamHCPtr, lz4hc_clamp_level(compressionLevel));
 }
 EXPORT_SYMBOL(LZ4_resetStreamHC);
 
-int LZ4_loadDictHC(LZ4_streamHC_t *LZ4_streamHCPtr,
-	const char *dictionary,
-	int dictSize)
+int LZ4_loadDictHC(LZ4_streamHC_t *streamHCPtr, const char *dictionary,
+		   int dictSize)
 {
-	LZ4HC_CCtx_internal *ctxPtr = &LZ4_streamHCPtr->internal_donotuse;
-
-	if (dictSize > 64 * KB) {
-		dictionary += dictSize - 64 * KB;
-		dictSize = 64 * KB;
-	}
-	LZ4HC_init(ctxPtr, (const BYTE *)dictionary);
-	if (dictSize >= 4)
-		LZ4HC_Insert(ctxPtr, (const BYTE *)dictionary + (dictSize - 3));
-	ctxPtr->end = (const BYTE *)dictionary + dictSize;
-	return dictSize;
+	return __lz4_loadDictHC(streamHCPtr, dictionary, dictSize);
 }
 EXPORT_SYMBOL(LZ4_loadDictHC);
 
-/* compression */
-
-static void LZ4HC_setExternalDict(
-	LZ4HC_CCtx_internal *ctxPtr,
-	const BYTE *newBlock)
+int LZ4_compress_HC_continue(LZ4_streamHC_t *streamHCPtr, const char *src,
+			     char *dst, int srcSize, int maxDstSize)
 {
-	if (ctxPtr->end >= ctxPtr->base + 4) {
-		/* Referencing remaining dictionary content */
-		LZ4HC_Insert(ctxPtr, ctxPtr->end - 3);
-	}
-
-	/*
-	 * Only one memory segment for extDict,
-	 * so any previous extDict is lost at this stage
-	 */
-	ctxPtr->lowLimit	= ctxPtr->dictLimit;
-	ctxPtr->dictLimit = (U32)(ctxPtr->end - ctxPtr->base);
-	ctxPtr->dictBase	= ctxPtr->base;
-	ctxPtr->base = newBlock - ctxPtr->dictLimit;
-	ctxPtr->end	= newBlock;
-	/* match referencing will resume from there */
-	ctxPtr->nextToUpdate = ctxPtr->dictLimit;
-}
-
-static int LZ4_compressHC_continue_generic(
-	LZ4_streamHC_t *LZ4_streamHCPtr,
-	const char *source,
-	char *dest,
-	int inputSize,
-	int maxOutputSize,
-	limitedOutput_directive limit)
-{
-	LZ4HC_CCtx_internal *ctxPtr = &LZ4_streamHCPtr->internal_donotuse;
-
-	/* auto - init if forgotten */
-	if (ctxPtr->base == NULL)
-		LZ4HC_init(ctxPtr, (const BYTE *) source);
-
-	/* Check overflow */
-	if ((size_t)(ctxPtr->end - ctxPtr->base) > 2 * GB) {
-		size_t dictSize = (size_t)(ctxPtr->end - ctxPtr->base)
-			- ctxPtr->dictLimit;
-		if (dictSize > 64 * KB)
-			dictSize = 64 * KB;
-		LZ4_loadDictHC(LZ4_streamHCPtr,
-			(const char *)(ctxPtr->end) - dictSize, (int)dictSize);
-	}
-
-	/* Check if blocks follow each other */
-	if ((const BYTE *)source != ctxPtr->end)
-		LZ4HC_setExternalDict(ctxPtr, (const BYTE *)source);
-
-	/* Check overlapping input/dictionary space */
-	{
-		const BYTE *sourceEnd = (const BYTE *) source + inputSize;
-		const BYTE * const dictBegin = ctxPtr->dictBase + ctxPtr->lowLimit;
-		const BYTE * const dictEnd = ctxPtr->dictBase + ctxPtr->dictLimit;
-
-		if ((sourceEnd > dictBegin)
-			&& ((const BYTE *)source < dictEnd)) {
-			if (sourceEnd > dictEnd)
-				sourceEnd = dictEnd;
-			ctxPtr->lowLimit = (U32)(sourceEnd - ctxPtr->dictBase);
-
-			if (ctxPtr->dictLimit - ctxPtr->lowLimit < 4)
-				ctxPtr->lowLimit = ctxPtr->dictLimit;
-		}
-	}
-
-	return LZ4HC_compress_generic(ctxPtr, source, dest,
-		inputSize, maxOutputSize, ctxPtr->compressionLevel, limit);
-}
-
-int LZ4_compress_HC_continue(
-	LZ4_streamHC_t *LZ4_streamHCPtr,
-	const char *source,
-	char *dest,
-	int inputSize,
-	int maxOutputSize)
-{
-	if (maxOutputSize < LZ4_compressBound(inputSize))
-		return LZ4_compressHC_continue_generic(LZ4_streamHCPtr,
-			source, dest, inputSize, maxOutputSize, limitedOutput);
-	else
-		return LZ4_compressHC_continue_generic(LZ4_streamHCPtr,
-			source, dest, inputSize, maxOutputSize, noLimit);
+	return __lz4_compress_HC_continue(streamHCPtr, src, dst, srcSize,
+					  maxDstSize);
 }
 EXPORT_SYMBOL(LZ4_compress_HC_continue);
 
-/* dictionary saving */
-
-int LZ4_saveDictHC(
-	LZ4_streamHC_t *LZ4_streamHCPtr,
-	char *safeBuffer,
-	int dictSize)
+int LZ4_saveDictHC(LZ4_streamHC_t *streamHCPtr, char *safeBuffer,
+		   int maxDictSize)
 {
-	LZ4HC_CCtx_internal *const streamPtr = &LZ4_streamHCPtr->internal_donotuse;
-	int const prefixSize = (int)(streamPtr->end
-		- (streamPtr->base + streamPtr->dictLimit));
-
-	if (dictSize > 64 * KB)
-		dictSize = 64 * KB;
-	if (dictSize < 4)
-		dictSize = 0;
-	if (dictSize > prefixSize)
-		dictSize = prefixSize;
-
-	memmove(safeBuffer, streamPtr->end - dictSize, dictSize);
-
-	{
-		U32 const endIndex = (U32)(streamPtr->end - streamPtr->base);
-
-		streamPtr->end = (const BYTE *)safeBuffer + dictSize;
-		streamPtr->base = streamPtr->end - endIndex;
-		streamPtr->dictLimit = endIndex - dictSize;
-		streamPtr->lowLimit = endIndex - dictSize;
-
-		if (streamPtr->nextToUpdate < streamPtr->dictLimit)
-			streamPtr->nextToUpdate = streamPtr->dictLimit;
-	}
-	return dictSize;
+	return __lz4_saveDictHC(streamHCPtr, safeBuffer, maxDictSize);
 }
 EXPORT_SYMBOL(LZ4_saveDictHC);
 

-- 
2.34.1

Re: [PATCH v2 11/15] lib/lz4: switch the HC compressor to the vendored sources

[sashiko-bot]

> Replace the forked LZ4HC with thin entry points over the vendored
> lz4hc.c.
> 
> Upstream routes levels 10 and up to its optimal parser, whose ~64K
> workspace the kernel cannot provide. Clamp those levels to 9 rather than
> reject them, so existing f2fs and zram settings keep working, and add
> LZ4HC_CLAMP_CLEVEL to <linux/lz4.h>. Document the clamp in f2fs.rst.
> 
> Levels 1 and 2 now use upstream's lz4mid: on kernel source they are
> 50-84% faster for 4-9% larger output. Levels 3 to 9 compress the same,
> but at 4K are up to 15% slower on poorly compressible data, including at
> level 9, zram's lz4hc default.
> 
> LZ4_compress_HC() and LZ4_resetStreamHC() check the state's alignment
> like the compressor's entry points.
> 
> LZ4_streamHC_t becomes incomplete and the macros that laid out the forked
> structure go. LZ4HC_MEM_COMPRESS grows from 262192 to upstream's 262200,
> static_asserted against LZ4_STREAMHC_MINSIZE. lz4hc.c pulls in lz4.c's
> decoder tables, unused here when the fast decode loop is off, so silence
> -Wunused-const-variable for this object.
> 
> This retires commit b08918fb3f27 ("lz4: do not export static symbol"):
> the glue exports only the five LZ4HC entry points.
> 
> Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>

Sashiko has reviewed this patch and found no issues.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261106-lz4-vendor-upstream-v2-0-ee04c3f0d7dc@samsung.com?part=11

[f2fs-dev] [PATCH v2 12/15] lib/lz4: widen LZ4_DECOMPRESS_INPLACE_MARGIN for the fast decode loop

[Michal Wilczynski]

erofs decompresses a pcluster in place when the slack after the output
is at least LZ4_DECOMPRESS_INPLACE_MARGIN(compressed size), upstream's
(compressedSize >> 8) + 32.

That holds for the fork, which overshoots its output by at most 8
bytes. Upstream's LZ4_FAST_DEC_LOOP, which the next patch brings in for
x86 and arm64, copies in 32-byte steps and can overwrite input it has
not read yet: of 172410 valid streams laid out as erofs does at exactly
that margin, 3660 fail to decode, none with the fork. The worst needs
(compressedSize >> 8) + 33. Images mkfs.erofs built did not hit this,
but nothing in the format rules it out.

Use (compressedSize >> 8) + (compressedSize >> 15) + 36, which takes a
4K pcluster's margin from 48 to 52 bytes. Upstream's lz4.h defines the
macro too, so undefine it first.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 include/linux/lz4.h | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)

diff --git a/include/linux/lz4.h b/include/linux/lz4.h
index 05060b9790be4621ddbe744172fa0ae37f6191f4..e809149747c3dc97d129e42e0ce4385b5a219930 100644
--- a/include/linux/lz4.h
+++ b/include/linux/lz4.h
@@ -521,7 +521,18 @@ int LZ4_decompress_safe_usingDict(const char *source, char *dest,
 	int compressedSize, int maxDecompressedSize, const char *dictStart,
 	int dictSize);
 
-#define LZ4_DECOMPRESS_INPLACE_MARGIN(compressedSize)          (((compressedSize) >> 8) + 32)
+/*
+ * Slack needed after the decompressed data for LZ4_decompress_safe() to
+ * decode in place, with the compressed data at the end of the buffer.
+ * Upstream's (compressedSize >> 8) + 32 can be a few bytes short: its fast
+ * decode loop copies in 32-byte steps, which may overwrite input it has
+ * not read yet.  The >> 15 term is extra headroom for large pclusters.
+ * Upstream's lz4.h, which lib/lz4's objects also see, defines the macro
+ * too.
+ */
+#undef LZ4_DECOMPRESS_INPLACE_MARGIN
+#define LZ4_DECOMPRESS_INPLACE_MARGIN(compressedSize)	\
+	(((compressedSize) >> 8) + ((compressedSize) >> 15) + 36)
 
 #ifndef LZ4_DISTANCE_MAX	/* history window size; can be user-defined at compile time */
 #define LZ4_DISTANCE_MAX 65535	/* set to maximum value by default */

-- 
2.34.1



_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

[PATCH v2 12/15] lib/lz4: widen LZ4_DECOMPRESS_INPLACE_MARGIN for the fast decode loop

[Michal Wilczynski]

erofs decompresses a pcluster in place when the slack after the output
is at least LZ4_DECOMPRESS_INPLACE_MARGIN(compressed size), upstream's
(compressedSize >> 8) + 32.

That holds for the fork, which overshoots its output by at most 8
bytes. Upstream's LZ4_FAST_DEC_LOOP, which the next patch brings in for
x86 and arm64, copies in 32-byte steps and can overwrite input it has
not read yet: of 172410 valid streams laid out as erofs does at exactly
that margin, 3660 fail to decode, none with the fork. The worst needs
(compressedSize >> 8) + 33. Images mkfs.erofs built did not hit this,
but nothing in the format rules it out.

Use (compressedSize >> 8) + (compressedSize >> 15) + 36, which takes a
4K pcluster's margin from 48 to 52 bytes. Upstream's lz4.h defines the
macro too, so undefine it first.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 include/linux/lz4.h | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)

diff --git a/include/linux/lz4.h b/include/linux/lz4.h
index 05060b9790be4621ddbe744172fa0ae37f6191f4..e809149747c3dc97d129e42e0ce4385b5a219930 100644
--- a/include/linux/lz4.h
+++ b/include/linux/lz4.h
@@ -521,7 +521,18 @@ int LZ4_decompress_safe_usingDict(const char *source, char *dest,
 	int compressedSize, int maxDecompressedSize, const char *dictStart,
 	int dictSize);
 
-#define LZ4_DECOMPRESS_INPLACE_MARGIN(compressedSize)          (((compressedSize) >> 8) + 32)
+/*
+ * Slack needed after the decompressed data for LZ4_decompress_safe() to
+ * decode in place, with the compressed data at the end of the buffer.
+ * Upstream's (compressedSize >> 8) + 32 can be a few bytes short: its fast
+ * decode loop copies in 32-byte steps, which may overwrite input it has
+ * not read yet.  The >> 15 term is extra headroom for large pclusters.
+ * Upstream's lz4.h, which lib/lz4's objects also see, defines the macro
+ * too.
+ */
+#undef LZ4_DECOMPRESS_INPLACE_MARGIN
+#define LZ4_DECOMPRESS_INPLACE_MARGIN(compressedSize)	\
+	(((compressedSize) >> 8) + ((compressedSize) >> 15) + 36)
 
 #ifndef LZ4_DISTANCE_MAX	/* history window size; can be user-defined at compile time */
 #define LZ4_DISTANCE_MAX 65535	/* set to maximum value by default */

-- 
2.34.1

Re: [PATCH v2 12/15] lib/lz4: widen LZ4_DECOMPRESS_INPLACE_MARGIN for the fast decode loop

[sashiko-bot]

> erofs decompresses a pcluster in place when the slack after the output
> is at least LZ4_DECOMPRESS_INPLACE_MARGIN(compressed size), upstream's
> (compressedSize >> 8) + 32.
> 
> That holds for the fork, which overshoots its output by at most 8
> bytes. Upstream's LZ4_FAST_DEC_LOOP, which the next patch brings in for
> x86 and arm64, copies in 32-byte steps and can overwrite input it has
> not read yet: of 172410 valid streams laid out as erofs does at exactly
> that margin, 3660 fail to decode, none with the fork. The worst needs
> (compressedSize >> 8) + 33. Images mkfs.erofs built did not hit this,
> but nothing in the format rules it out.
> 
> Use (compressedSize >> 8) + (compressedSize >> 15) + 36, which takes a
> 4K pcluster's margin from 48 to 52 bytes. Upstream's lz4.h defines the
> macro too, so undefine it first.
> 
> Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>

Sashiko has reviewed this patch and found no issues.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261106-lz4-vendor-upstream-v2-0-ee04c3f0d7dc@samsung.com?part=12

Re: [PATCH v2 12/15] lib/lz4: widen LZ4_DECOMPRESS_INPLACE_MARGIN for the fast decode loop

[Gao Xiang]

Hi Michal,

On Fri, Nov 06, 2026 at 03:42:52PM +0100, Michal Wilczynski wrote:
> erofs decompresses a pcluster in place when the slack after the output
> is at least LZ4_DECOMPRESS_INPLACE_MARGIN(compressed size), upstream's
> (compressedSize >> 8) + 32.
> 
> That holds for the fork, which overshoots its output by at most 8
> bytes. Upstream's LZ4_FAST_DEC_LOOP, which the next patch brings in for
> x86 and arm64, copies in 32-byte steps and can overwrite input it has
> not read yet: of 172410 valid streams laid out as erofs does at exactly
> that margin, 3660 fail to decode, none with the fork. The worst needs
> (compressedSize >> 8) + 33. Images mkfs.erofs built did not hit this,
> but nothing in the format rules it out.
> 
> Use (compressedSize >> 8) + (compressedSize >> 15) + 36, which takes a
> 4K pcluster's margin from 48 to 52 bytes. Upstream's lz4.h defines the
> macro too, so undefine it first.
> 
> Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>

First, I really think kernel LZ4 should be updated for many years, so
thanks for this!

But my personal thought is that could you just fold this patch into
the next one ("lib/lz4: switch the decompressor to the vendored sources")?
Because this one should be part of the fast loop update.

Although the current patch order is also fine, but my potential concern
is that people don't know the details maybe pick the random ones for
backporting, which is what I'd like to avoid so I hope this patch could
be just part of it, otherwise the inplace decompression could be broken.

Thanks,
Gao Xiang

Re: [f2fs-dev] [PATCH v2 12/15] lib/lz4: widen LZ4_DECOMPRESS_INPLACE_MARGIN for the fast decode loop

[Gao Xiang via Linux-f2fs-devel]

Hi Michal,

On Fri, Nov 06, 2026 at 03:42:52PM +0100, Michal Wilczynski wrote:
> erofs decompresses a pcluster in place when the slack after the output
> is at least LZ4_DECOMPRESS_INPLACE_MARGIN(compressed size), upstream's
> (compressedSize >> 8) + 32.
> 
> That holds for the fork, which overshoots its output by at most 8
> bytes. Upstream's LZ4_FAST_DEC_LOOP, which the next patch brings in for
> x86 and arm64, copies in 32-byte steps and can overwrite input it has
> not read yet: of 172410 valid streams laid out as erofs does at exactly
> that margin, 3660 fail to decode, none with the fork. The worst needs
> (compressedSize >> 8) + 33. Images mkfs.erofs built did not hit this,
> but nothing in the format rules it out.
> 
> Use (compressedSize >> 8) + (compressedSize >> 15) + 36, which takes a
> 4K pcluster's margin from 48 to 52 bytes. Upstream's lz4.h defines the
> macro too, so undefine it first.
> 
> Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>

First, I really think kernel LZ4 should be updated for many years, so
thanks for this!

But my personal thought is that could you just fold this patch into
the next one ("lib/lz4: switch the decompressor to the vendored sources")?
Because this one should be part of the fast loop update.

Although the current patch order is also fine, but my potential concern
is that people don't know the details maybe pick the random ones for
backporting, which is what I'd like to avoid so I hope this patch could
be just part of it, otherwise the inplace decompression could be broken.

Thanks,
Gao Xiang


_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

Re: [PATCH v2 12/15] lib/lz4: widen LZ4_DECOMPRESS_INPLACE_MARGIN for the fast decode loop

[Michal Wilczynski]

On 11/11/26 23:02, Gao Xiang wrote:
> Hi Michal,
> 
> On Fri, Nov 06, 2026 at 03:42:52PM +0100, Michal Wilczynski wrote:
>> erofs decompresses a pcluster in place when the slack after the output
>> is at least LZ4_DECOMPRESS_INPLACE_MARGIN(compressed size), upstream's
>> (compressedSize >> 8) + 32.
>>
>> That holds for the fork, which overshoots its output by at most 8
>> bytes. Upstream's LZ4_FAST_DEC_LOOP, which the next patch brings in for
>> x86 and arm64, copies in 32-byte steps and can overwrite input it has
>> not read yet: of 172410 valid streams laid out as erofs does at exactly
>> that margin, 3660 fail to decode, none with the fork. The worst needs
>> (compressedSize >> 8) + 33. Images mkfs.erofs built did not hit this,
>> but nothing in the format rules it out.
>>
>> Use (compressedSize >> 8) + (compressedSize >> 15) + 36, which takes a
>> 4K pcluster's margin from 48 to 52 bytes. Upstream's lz4.h defines the
>> macro too, so undefine it first.
>>
>> Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
> 
> First, I really think kernel LZ4 should be updated for many years, so
> thanks for this!
> 
> But my personal thought is that could you just fold this patch into
> the next one ("lib/lz4: switch the decompressor to the vendored sources")?
> Because this one should be part of the fast loop update.
> 
> Although the current patch order is also fine, but my potential concern
> is that people don't know the details maybe pick the random ones for
> backporting, which is what I'd like to avoid so I hope this patch could
> be just part of it, otherwise the inplace decompression could be broken.
> 

Make sense I'll fold it into the decompressor switch in v3 so the
wider margin and the fast loop can't be picked up separately.

Thanks for the review and the kind words

Best regards,
-- 
Michal Wilczynski <m.wilczynski@samsung.com>

Re: [f2fs-dev] [PATCH v2 12/15] lib/lz4: widen LZ4_DECOMPRESS_INPLACE_MARGIN for the fast decode loop

[Michal Wilczynski]

On 11/11/26 23:02, Gao Xiang wrote:
> Hi Michal,
> 
> On Fri, Nov 06, 2026 at 03:42:52PM +0100, Michal Wilczynski wrote:
>> erofs decompresses a pcluster in place when the slack after the output
>> is at least LZ4_DECOMPRESS_INPLACE_MARGIN(compressed size), upstream's
>> (compressedSize >> 8) + 32.
>>
>> That holds for the fork, which overshoots its output by at most 8
>> bytes. Upstream's LZ4_FAST_DEC_LOOP, which the next patch brings in for
>> x86 and arm64, copies in 32-byte steps and can overwrite input it has
>> not read yet: of 172410 valid streams laid out as erofs does at exactly
>> that margin, 3660 fail to decode, none with the fork. The worst needs
>> (compressedSize >> 8) + 33. Images mkfs.erofs built did not hit this,
>> but nothing in the format rules it out.
>>
>> Use (compressedSize >> 8) + (compressedSize >> 15) + 36, which takes a
>> 4K pcluster's margin from 48 to 52 bytes. Upstream's lz4.h defines the
>> macro too, so undefine it first.
>>
>> Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
> 
> First, I really think kernel LZ4 should be updated for many years, so
> thanks for this!
> 
> But my personal thought is that could you just fold this patch into
> the next one ("lib/lz4: switch the decompressor to the vendored sources")?
> Because this one should be part of the fast loop update.
> 
> Although the current patch order is also fine, but my potential concern
> is that people don't know the details maybe pick the random ones for
> backporting, which is what I'd like to avoid so I hope this patch could
> be just part of it, otherwise the inplace decompression could be broken.
> 

Make sense I'll fold it into the decompressor switch in v3 so the
wider margin and the fast loop can't be picked up separately.

Thanks for the review and the kind words

Best regards,
-- 
Michal Wilczynski <m.wilczynski@samsung.com>


_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

[f2fs-dev] [PATCH v2 13/15] lib/lz4: switch the decompressor to the vendored sources

[Michal Wilczynski]

Replace the forked decompressor with thin entry points over the vendored
lz4.c; the API is signature-compatible, so every export is a plain
forwarder. lz4defs.h has no users left and goes.

lib/decompress_unlz4.c includes this file before boot, together with
<linux/lz4.h>, so LZ4_streamDecode_t becomes incomplete as upstream
declares it, and LZ4_MEM_DECOMPRESS (32) replaces the
LZ4_STREAMDECODESIZE macros. LZ4_COMPRESSBOUND is defined by both
headers and gets a guard; the LZ4_compressBound() wrapper goes, as
lib/decompress_unlz4.c now uses the macro.

LZ4_decompress_safe_partial() on a truncated block can now return the
bytes decoded so far rather than an error, depending on where the cut
falls; the kernel-doc says so.

This retires six local patches:

commit 8cb5d7482810 ("lib/lz4: make arrays static const, reduces object code size")
  upstream 75fb878a9019 ("decompress_generic: Add fastpath for small offsets"), v1.9.0
commit b1a3e75e466d ("lz4: fix kernel decompression speed")
  upstream fe2a1b3707d5 ("Call LZ4_memcpy() instead of memcpy()"), v1.9.3
commit 89b158635ad7 ("lib/lz4: explicitly support in-place decompression")
  upstream 33cb8518ac38 ("decompress: changed final memcpy() into memmove()"), v1.9.2
commit 7fde9d6e839d ("lz4_decompress: declare LZ4_decompress_safe_withPrefix64k static")
  LZ4LIB_VISIBILITY in lz4_deps.h, which makes every upstream entry point static
commit eafc0a02391b ("lz4: fix LZ4_decompress_safe_partial read out of bound")
  upstream c5d6f8a8be39 ("fix #783"), v1.9.3
commit 2d8867f3e083 ("lib: make LZ4_decompress_safe_forceExtDict() static")
  the static forward declaration in lz4_deps.h

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 drivers/block/zram/backend_lz4.c   |   2 +-
 drivers/block/zram/backend_lz4hc.c |   2 +-
 include/linux/lz4.h                |  60 ++--
 lib/decompress_unlz4.c             |   4 +-
 lib/lz4/lz4_decompress.c           | 640 +++----------------------------------
 lib/lz4/lz4defs.h                  | 247 --------------
 6 files changed, 71 insertions(+), 884 deletions(-)

diff --git a/drivers/block/zram/backend_lz4.c b/drivers/block/zram/backend_lz4.c
index 50265e3ce256490994cb6dae32a45da376e408e0..b9b8a5f678c155442fa3fd2c1984e45182e6fc38 100644
--- a/drivers/block/zram/backend_lz4.c
+++ b/drivers/block/zram/backend_lz4.c
@@ -84,7 +84,7 @@ static int lz4_create(struct zcomp_params *params, struct zcomp_ctx *ctx)
 		if (!zctx->mem)
 			goto error;
 	} else {
-		zctx->dstrm = kzalloc_obj(*zctx->dstrm);
+		zctx->dstrm = kzalloc(LZ4_MEM_DECOMPRESS, GFP_KERNEL);
 		if (!zctx->dstrm)
 			goto error;
 
diff --git a/drivers/block/zram/backend_lz4hc.c b/drivers/block/zram/backend_lz4hc.c
index e7ee3740219852f079d1e0866da6b213a3f3a583..1e09c93016264259c149e2913b2819571b56fab6 100644
--- a/drivers/block/zram/backend_lz4hc.c
+++ b/drivers/block/zram/backend_lz4hc.c
@@ -65,7 +65,7 @@ static int lz4hc_create(struct zcomp_params *params, struct zcomp_ctx *ctx)
 		if (!zctx->mem)
 			goto error;
 	} else {
-		zctx->dstrm = kzalloc_obj(*zctx->dstrm);
+		zctx->dstrm = kzalloc(LZ4_MEM_DECOMPRESS, GFP_KERNEL);
 		if (!zctx->dstrm)
 			goto error;
 
diff --git a/include/linux/lz4.h b/include/linux/lz4.h
index e809149747c3dc97d129e42e0ce4385b5a219930..fd9d7a162e04655a9d19639515466a6ba8e1e892 100644
--- a/include/linux/lz4.h
+++ b/include/linux/lz4.h
@@ -48,10 +48,16 @@
  *	CONSTANTS
  **************************************************************************/
 #define LZ4_MAX_INPUT_SIZE	0x7E000000 /* 2 113 929 216 bytes */
+
+/* Upstream's lz4.h defines the same macro, and lib/lz4's entry points and
+ * the pre-boot lib/decompress_unlz4.c see both headers.
+ */
+#ifndef LZ4_COMPRESSBOUND
 #define LZ4_COMPRESSBOUND(isize)	(\
 	(unsigned int)(isize) > (unsigned int)LZ4_MAX_INPUT_SIZE \
 	? 0 \
 	: (isize) + ((isize)/255) + 16)
+#endif
 
 #define LZ4_ACCELERATION_DEFAULT 1
 
@@ -67,12 +73,8 @@
 #define LZ4HC_CLAMP_CLEVEL			10
 
 /*-************************************************************************
- *	STREAMING CONSTANTS AND STRUCTURES
+ *	STREAMING STRUCTURES
  **************************************************************************/
-#define LZ4_STREAMDECODESIZE_U64	4
-#define LZ4_STREAMDECODESIZE		 (LZ4_STREAMDECODESIZE_U64 * \
-	sizeof(unsigned long long))
-
 /*
  * LZ4_stream_t - an LZ4 stream.  Incomplete: lib/lz4 owns the layout.
  * Allocate LZ4_MEM_COMPRESS bytes and cast, do not sizeof().
@@ -86,21 +88,11 @@ typedef union LZ4_stream_u LZ4_stream_t;
 typedef union LZ4_streamHC_u LZ4_streamHC_t;
 
 /*
- * LZ4_streamDecode_t - information structure to track an
- *	LZ4 stream during decompression.
- *
- * init this structure using LZ4_setStreamDecode (or memset()) before first use
+ * LZ4_streamDecode_t - an LZ4 stream during decompression.  Incomplete:
+ * lib/lz4 owns the layout.  Allocate LZ4_MEM_DECOMPRESS bytes and cast, do
+ * not sizeof().  Init with LZ4_setStreamDecode() (or zero it) before use.
  */
-typedef struct {
-	const uint8_t *externalDict;
-	size_t extDictSize;
-	const uint8_t *prefixEnd;
-	size_t prefixSize;
-} LZ4_streamDecode_t_internal;
-typedef union {
-	unsigned long long table[LZ4_STREAMDECODESIZE_U64];
-	LZ4_streamDecode_t_internal internal_donotuse;
-} LZ4_streamDecode_t;
+typedef union LZ4_streamDecode_u LZ4_streamDecode_t;
 
 /*-************************************************************************
  *	SIZE OF STATE
@@ -112,23 +104,12 @@ typedef union {
  */
 #define LZ4_MEM_COMPRESS	16416
 #define LZ4HC_MEM_COMPRESS	262200
+#define LZ4_MEM_DECOMPRESS	32
 
 /*-************************************************************************
  *	Compression Functions
  **************************************************************************/
 
-/**
- * LZ4_compressBound() - Max. output size in worst case szenarios
- * @isize: Size of the input data
- *
- * Return: Max. size LZ4 may output in a "worst case" szenario
- * (data not compressible)
- */
-static inline int LZ4_compressBound(size_t isize)
-{
-	return LZ4_COMPRESSBOUND(isize);
-}
-
 /**
  * LZ4_compress_default() - Compress data from source to dest
  * @source: source address of the original data
@@ -142,7 +123,7 @@ static inline int LZ4_compressBound(size_t isize)
  * Compresses 'sourceSize' bytes from buffer 'source'
  * into already allocated 'dest' buffer of size 'maxOutputSize'.
  * Compression is guaranteed to succeed if
- * 'maxOutputSize' >= LZ4_compressBound(inputSize).
+ * 'maxOutputSize' >= LZ4_COMPRESSBOUND(inputSize).
  * It also runs faster, so it's a recommended setting.
  * If the function cannot compress 'source' into a more limited 'dest' budget,
  * compression stops *immediately*, and the function result is zero.
@@ -234,7 +215,7 @@ int LZ4_decompress_safe(const char *source, char *dest, int compressedSize,
  * @source: source address of the compressed data
  * @dest: output buffer address of the decompressed data which must be
  *	already allocated
- * @compressedSize: is the precise full size of the compressed block.
+ * @compressedSize: size of the compressed data at 'source'
  * @targetOutputSize: the decompression operation will try
  *	to stop as soon as 'targetOutputSize' has been reached
  * @maxDecompressedSize: is the size of destination buffer
@@ -249,8 +230,11 @@ int LZ4_decompress_safe(const char *source, char *dest, int compressedSize,
  * It is therefore protected against malicious data packets.
  *
  * Return: the number of bytes decoded in the destination buffer
- *	(necessarily <= maxDecompressedSize)
- *	or a negative result in case of error
+ *	(necessarily <= targetOutputSize and maxDecompressedSize)
+ *	or a negative result in case of error.  For a block cut short the
+ *	result is either the bytes decoded so far, below targetOutputSize,
+ *	or an error, depending on where the cut falls; check it against the
+ *	size you expect.
  *
  */
 int LZ4_decompress_safe_partial(const char *source, char *dest,
@@ -275,7 +259,7 @@ int LZ4_decompress_safe_partial(const char *source, char *dest,
  *
  * Compress data from 'src' into 'dst', using the more powerful
  * but slower "HC" algorithm. Compression is guaranteed to succeed if
- * `dstCapacity >= LZ4_compressBound(srcSize)
+ * `dstCapacity >= LZ4_COMPRESSBOUND(srcSize)
  *
  * Return : the number of bytes written into 'dst' or 0 if compression fails.
  */
@@ -339,7 +323,7 @@ int	LZ4_loadDictHC(LZ4_streamHC_t *streamHCPtr, const char *dictionary,
  * (including initial dictionary when present) must remain accessible
  * and unmodified during compression.
  * 'dst' buffer should be sized to handle worst case scenarios, using
- *  LZ4_compressBound(), to ensure operation success.
+ *  LZ4_COMPRESSBOUND(), to ensure operation success.
  *  If, for any reason, previous data blocks can't be preserved unmodified
  *  in memory during next compression block,
  *  you must save it to a safer memory space, using LZ4_saveDictHC().
@@ -435,7 +419,7 @@ int LZ4_saveDict(LZ4_stream_t *streamPtr, char *safeBuffer, int dictSize);
  * as dictionary to improve compression ratio.
  * Important : Previous data blocks are assumed to still
  * be present and unmodified !
- * If maxDstSize >= LZ4_compressBound(srcSize),
+ * If maxDstSize >= LZ4_COMPRESSBOUND(srcSize),
  * compression is guaranteed to succeed, and runs faster.
  *
  * Return: Number of bytes written into buffer 'dst'  or 0 if compression fails
diff --git a/lib/decompress_unlz4.c b/lib/decompress_unlz4.c
index dc20c22e087900e21da49f9fcb267e6b511e1072..87eea1fbdea063e241dd98f994d7fee078b7a910 100644
--- a/lib/decompress_unlz4.c
+++ b/lib/decompress_unlz4.c
@@ -69,7 +69,7 @@ STATIC inline int INIT unlz4(u8 *input, long in_len,
 		error("NULL input pointer and missing fill function");
 		goto exit_1;
 	} else {
-		inp = large_malloc(LZ4_compressBound(uncomp_chunksize));
+		inp = large_malloc(LZ4_COMPRESSBOUND(uncomp_chunksize));
 		if (!inp) {
 			error("Could not allocate input buffer");
 			goto exit_1;
@@ -140,7 +140,7 @@ STATIC inline int INIT unlz4(u8 *input, long in_len,
 			inp += 4;
 			size -= 4;
 		} else {
-			if (chunksize > LZ4_compressBound(uncomp_chunksize)) {
+			if (chunksize > LZ4_COMPRESSBOUND(uncomp_chunksize)) {
 				error("chunk length is longer than allocated");
 				goto exit_2;
 			}
diff --git a/lib/lz4/lz4_decompress.c b/lib/lz4/lz4_decompress.c
index f89b0e547f2156990c16770f69f89e0d8b46156d..60b57862a2e387aa89fa26f6f8e2b5485b5d9995 100644
--- a/lib/lz4/lz4_decompress.c
+++ b/lib/lz4/lz4_decompress.c
@@ -1,626 +1,76 @@
+// SPDX-License-Identifier: GPL-2.0-only OR BSD-2-Clause
 /*
- * LZ4 - Fast LZ compression algorithm
  * Copyright (C) 2011 - 2016, Yann Collet.
- * BSD 2 - Clause License (http://www.opensource.org/licenses/bsd - license.php)
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions are
- * met:
- *	* Redistributions of source code must retain the above copyright
- *	  notice, this list of conditions and the following disclaimer.
- *	* Redistributions in binary form must reproduce the above
- * copyright notice, this list of conditions and the following disclaimer
- * in the documentation and/or other materials provided with the
- * distribution.
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
- * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
- * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
- * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
- * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
- * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
- * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
- * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
- * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- * You can contact the author at :
- *	- LZ4 homepage : http://www.lz4.org
- *	- LZ4 source repository : https://github.com/lz4/lz4
+ * Copyright (C) 2016, Sven Schmidt <4sschmid@informatik.uni-hamburg.de>
+ * Copyright (c) 2026 Samsung Electronics Co., Ltd.
+ * Author: Michal Wilczynski <m.wilczynski@samsung.com>
  *
- *	Changed for kernel usage by:
- *	Sven Schmidt <4sschmid@informatik.uni-hamburg.de>
+ * LZ4 decompressor -- kernel entry points
+ *
+ * The decompressor is upstream's, in the verbatim lz4.c included below.  The
+ * API is signature-compatible, so every export is a plain forwarder.  This
+ * file is also included by lib/decompress_unlz4.c for the pre-boot
+ * decompressor, which defines STATIC.
  */
 
-/*-************************************
- *	Dependencies
- **************************************/
-#include "lz4defs.h"
-#include <linux/init.h>
-#include <linux/module.h>
-#include <linux/kernel.h>
-#include <linux/unaligned.h>
+#include "lz4_deps.h"
+#include "upstream/lz4.c"
 
-/*-*****************************
- *	Decompression functions
- *******************************/
+/* Upstream's short helper macros; MIN and MAX clash with <linux/minmax.h>. */
+#undef MIN
+#undef MAX
+#undef KB
+#undef MB
+#undef GB
 
-#define DEBUGLOG(l, ...) {}	/* disabled */
+#include "lz4_kernel_api.h"
 
-#ifndef assert
-#define assert(condition) ((void)0)
+#ifndef STATIC
+#include <linux/export.h>
+#include <linux/module.h>
 #endif
 
-/*
- * LZ4_decompress_generic() :
- * This generic decompression function covers all use cases.
- * It shall be instantiated several times, using different sets of directives.
- * Note that it is important for performance that this function really get inlined,
- * in order to remove useless branches during compilation optimization.
- */
-static FORCE_INLINE int LZ4_decompress_generic(
-	 const char * const src,
-	 char * const dst,
-	 int srcSize,
-		/*
-		 * If endOnInput == endOnInputSize,
-		 * this value is `dstCapacity`
-		 */
-	 int outputSize,
-	 /* endOnOutputSize, endOnInputSize */
-	 endCondition_directive endOnInput,
-	 /* full, partial */
-	 earlyEnd_directive partialDecoding,
-	 /* noDict, withPrefix64k, usingExtDict */
-	 dict_directive dict,
-	 /* always <= dst, == dst when no prefix */
-	 const BYTE * const lowPrefix,
-	 /* only if dict == usingExtDict */
-	 const BYTE * const dictStart,
-	 /* note : = 0 if noDict */
-	 const size_t dictSize
-	 )
-{
-	const BYTE *ip = (const BYTE *) src;
-	const BYTE * const iend = ip + srcSize;
-
-	BYTE *op = (BYTE *) dst;
-	BYTE * const oend = op + outputSize;
-	BYTE *cpy;
-
-	const BYTE * const dictEnd = (const BYTE *)dictStart + dictSize;
-	static const unsigned int inc32table[8] = {0, 1, 2, 1, 0, 4, 4, 4};
-	static const int dec64table[8] = {0, 0, 0, -1, -4, 1, 2, 3};
-
-	const int safeDecode = (endOnInput == endOnInputSize);
-	const int checkOffset = ((safeDecode) && (dictSize < (int)(64 * KB)));
-
-	/* Set up the "end" pointers for the shortcut. */
-	const BYTE *const shortiend = iend -
-		(endOnInput ? 14 : 8) /*maxLL*/ - 2 /*offset*/;
-	const BYTE *const shortoend = oend -
-		(endOnInput ? 14 : 8) /*maxLL*/ - 18 /*maxML*/;
-
-	DEBUGLOG(5, "%s (srcSize:%i, dstSize:%i)", __func__,
-		 srcSize, outputSize);
-
-	/* Special cases */
-	assert(lowPrefix <= op);
-	assert(src != NULL);
-
-	/* Empty output buffer */
-	if ((endOnInput) && (unlikely(outputSize == 0)))
-		return ((srcSize == 1) && (*ip == 0)) ? 0 : -1;
-
-	if ((!endOnInput) && (unlikely(outputSize == 0)))
-		return (*ip == 0 ? 1 : -1);
-
-	if ((endOnInput) && unlikely(srcSize == 0))
-		return -1;
-
-	/* Main Loop : decode sequences */
-	while (1) {
-		size_t length;
-		const BYTE *match;
-		size_t offset;
-
-		/* get literal length */
-		unsigned int const token = *ip++;
-		length = token>>ML_BITS;
-
-		/* ip < iend before the increment */
-		assert(!endOnInput || ip <= iend);
-
-		/*
-		 * A two-stage shortcut for the most common case:
-		 * 1) If the literal length is 0..14, and there is enough
-		 * space, enter the shortcut and copy 16 bytes on behalf
-		 * of the literals (in the fast mode, only 8 bytes can be
-		 * safely copied this way).
-		 * 2) Further if the match length is 4..18, copy 18 bytes
-		 * in a similar manner; but we ensure that there's enough
-		 * space in the output for those 18 bytes earlier, upon
-		 * entering the shortcut (in other words, there is a
-		 * combined check for both stages).
-		 *
-		 * The & in the likely() below is intentionally not && so that
-		 * some compilers can produce better parallelized runtime code
-		 */
-		if ((endOnInput ? length != RUN_MASK : length <= 8)
-		   /*
-		    * strictly "less than" on input, to re-enter
-		    * the loop with at least one byte
-		    */
-		   && likely((endOnInput ? ip < shortiend : 1) &
-			     (op <= shortoend))) {
-			/* Copy the literals */
-			LZ4_memcpy(op, ip, endOnInput ? 16 : 8);
-			op += length; ip += length;
-
-			/*
-			 * The second stage:
-			 * prepare for match copying, decode full info.
-			 * If it doesn't work out, the info won't be wasted.
-			 */
-			length = token & ML_MASK; /* match length */
-			offset = LZ4_readLE16(ip);
-			ip += 2;
-			match = op - offset;
-			assert(match <= op); /* check overflow */
-
-			/* Do not deal with overlapping matches. */
-			if ((length != ML_MASK) &&
-			    (offset >= 8) &&
-			    (dict == withPrefix64k || match >= lowPrefix)) {
-				/* Copy the match. */
-				LZ4_memcpy(op + 0, match + 0, 8);
-				LZ4_memcpy(op + 8, match + 8, 8);
-				LZ4_memcpy(op + 16, match + 16, 2);
-				op += length + MINMATCH;
-				/* Both stages worked, load the next token. */
-				continue;
-			}
-
-			/*
-			 * The second stage didn't work out, but the info
-			 * is ready. Propel it right to the point of match
-			 * copying.
-			 */
-			goto _copy_match;
-		}
-
-		/* decode literal length */
-		if (length == RUN_MASK) {
-			unsigned int s;
-
-			if (unlikely(endOnInput ? ip >= iend - RUN_MASK : 0)) {
-				/* overflow detection */
-				goto _output_error;
-			}
-			do {
-				s = *ip++;
-				length += s;
-			} while (likely(endOnInput
-				? ip < iend - RUN_MASK
-				: 1) & (s == 255));
-
-			if ((safeDecode)
-			    && unlikely((uptrval)(op) +
-					length < (uptrval)(op))) {
-				/* overflow detection */
-				goto _output_error;
-			}
-			if ((safeDecode)
-			    && unlikely((uptrval)(ip) +
-					length < (uptrval)(ip))) {
-				/* overflow detection */
-				goto _output_error;
-			}
-		}
-
-		/* copy literals */
-		cpy = op + length;
-		LZ4_STATIC_ASSERT(MFLIMIT >= WILDCOPYLENGTH);
-
-		if (((endOnInput) && ((cpy > oend - MFLIMIT)
-			|| (ip + length > iend - (2 + 1 + LASTLITERALS))))
-			|| ((!endOnInput) && (cpy > oend - WILDCOPYLENGTH))) {
-			if (partialDecoding) {
-				if (cpy > oend) {
-					/*
-					 * Partial decoding :
-					 * stop in the middle of literal segment
-					 */
-					cpy = oend;
-					length = oend - op;
-				}
-				if ((endOnInput)
-					&& (ip + length > iend)) {
-					/*
-					 * Error :
-					 * read attempt beyond
-					 * end of input buffer
-					 */
-					goto _output_error;
-				}
-			} else {
-				if ((!endOnInput)
-					&& (cpy != oend)) {
-					/*
-					 * Error :
-					 * block decoding must
-					 * stop exactly there
-					 */
-					goto _output_error;
-				}
-				if ((endOnInput)
-					&& ((ip + length != iend)
-					|| (cpy > oend))) {
-					/*
-					 * Error :
-					 * input must be consumed
-					 */
-					goto _output_error;
-				}
-			}
-
-			/*
-			 * supports overlapping memory regions; only matters
-			 * for in-place decompression scenarios
-			 */
-			LZ4_memmove(op, ip, length);
-			ip += length;
-			op += length;
-
-			/* Necessarily EOF when !partialDecoding.
-			 * When partialDecoding, it is EOF if we've either
-			 * filled the output buffer or
-			 * can't proceed with reading an offset for following match.
-			 */
-			if (!partialDecoding || (cpy == oend) || (ip >= (iend - 2)))
-				break;
-		} else {
-			/* may overwrite up to WILDCOPYLENGTH beyond cpy */
-			LZ4_wildCopy(op, ip, cpy);
-			ip += length;
-			op = cpy;
-		}
-
-		/* get offset */
-		offset = LZ4_readLE16(ip);
-		ip += 2;
-		match = op - offset;
-
-		/* get matchlength */
-		length = token & ML_MASK;
-
-_copy_match:
-		if ((checkOffset) && (unlikely(match + dictSize < lowPrefix))) {
-			/* Error : offset outside buffers */
-			goto _output_error;
-		}
-
-		/* costs ~1%; silence an msan warning when offset == 0 */
-		/*
-		 * note : when partialDecoding, there is no guarantee that
-		 * at least 4 bytes remain available in output buffer
-		 */
-		if (!partialDecoding) {
-			assert(oend > op);
-			assert(oend - op >= 4);
-
-			LZ4_write32(op, (U32)offset);
-		}
-
-		if (length == ML_MASK) {
-			unsigned int s;
-
-			do {
-				s = *ip++;
-
-				if ((endOnInput) && (ip > iend - LASTLITERALS))
-					goto _output_error;
-
-				length += s;
-			} while (s == 255);
+static_assert(sizeof(LZ4_streamDecode_t) == LZ4_STREAMDECODE_MINSIZE);
+static_assert(LZ4_STREAMDECODE_MINSIZE == 32); /* LZ4_MEM_DECOMPRESS */
 
-			if ((safeDecode)
-				&& unlikely(
-					(uptrval)(op) + length < (uptrval)op)) {
-				/* overflow detection */
-				goto _output_error;
-			}
-		}
-
-		length += MINMATCH;
-
-		/* match starting within external dictionary */
-		if ((dict == usingExtDict) && (match < lowPrefix)) {
-			if (unlikely(op + length > oend - LASTLITERALS)) {
-				/* doesn't respect parsing restriction */
-				if (!partialDecoding)
-					goto _output_error;
-				length = min(length, (size_t)(oend - op));
-			}
-
-			if (length <= (size_t)(lowPrefix - match)) {
-				/*
-				 * match fits entirely within external
-				 * dictionary : just copy
-				 */
-				memmove(op, dictEnd - (lowPrefix - match),
-					length);
-				op += length;
-			} else {
-				/*
-				 * match stretches into both external
-				 * dictionary and current block
-				 */
-				size_t const copySize = (size_t)(lowPrefix - match);
-				size_t const restSize = length - copySize;
-
-				LZ4_memcpy(op, dictEnd - copySize, copySize);
-				op += copySize;
-				if (restSize > (size_t)(op - lowPrefix)) {
-					/* overlap copy */
-					BYTE * const endOfMatch = op + restSize;
-					const BYTE *copyFrom = lowPrefix;
-
-					while (op < endOfMatch)
-						*op++ = *copyFrom++;
-				} else {
-					LZ4_memcpy(op, lowPrefix, restSize);
-					op += restSize;
-				}
-			}
-			continue;
-		}
-
-		/* copy match within block */
-		cpy = op + length;
-
-		/*
-		 * partialDecoding :
-		 * may not respect endBlock parsing restrictions
-		 */
-		assert(op <= oend);
-		if (partialDecoding &&
-		    (cpy > oend - MATCH_SAFEGUARD_DISTANCE)) {
-			size_t const mlen = min(length, (size_t)(oend - op));
-			const BYTE * const matchEnd = match + mlen;
-			BYTE * const copyEnd = op + mlen;
-
-			if (matchEnd > op) {
-				/* overlap copy */
-				while (op < copyEnd)
-					*op++ = *match++;
-			} else {
-				LZ4_memcpy(op, match, mlen);
-			}
-			op = copyEnd;
-			if (op == oend)
-				break;
-			continue;
-		}
-
-		if (unlikely(offset < 8)) {
-			op[0] = match[0];
-			op[1] = match[1];
-			op[2] = match[2];
-			op[3] = match[3];
-			match += inc32table[offset];
-			LZ4_memcpy(op + 4, match, 4);
-			match -= dec64table[offset];
-		} else {
-			LZ4_copy8(op, match);
-			match += 8;
-		}
-
-		op += 8;
-
-		if (unlikely(cpy > oend - MATCH_SAFEGUARD_DISTANCE)) {
-			BYTE * const oCopyLimit = oend - (WILDCOPYLENGTH - 1);
-
-			if (cpy > oend - LASTLITERALS) {
-				/*
-				 * Error : last LASTLITERALS bytes
-				 * must be literals (uncompressed)
-				 */
-				goto _output_error;
-			}
-
-			if (op < oCopyLimit) {
-				LZ4_wildCopy(op, match, oCopyLimit);
-				match += oCopyLimit - op;
-				op = oCopyLimit;
-			}
-			while (op < cpy)
-				*op++ = *match++;
-		} else {
-			LZ4_copy8(op, match);
-			if (length > 16)
-				LZ4_wildCopy(op + 8, match + 8, cpy);
-		}
-		op = cpy; /* wildcopy correction */
-	}
-
-	/* end of decoding */
-	if (endOnInput) {
-		/* Nb of output bytes decoded */
-		return (int) (((char *)op) - dst);
-	} else {
-		/* Nb of input bytes read */
-		return (int) (((const char *)ip) - src);
-	}
-
-	/* Overflow error detected */
-_output_error:
-	return (int) (-(((const char *)ip) - src)) - 1;
-}
-
-int LZ4_decompress_safe(const char *source, char *dest,
-	int compressedSize, int maxDecompressedSize)
-{
-	return LZ4_decompress_generic(source, dest,
-				      compressedSize, maxDecompressedSize,
-				      endOnInputSize, decode_full_block,
-				      noDict, (BYTE *)dest, NULL, 0);
-}
-
-int LZ4_decompress_safe_partial(const char *src, char *dst,
-	int compressedSize, int targetOutputSize, int dstCapacity)
-{
-	dstCapacity = min(targetOutputSize, dstCapacity);
-	return LZ4_decompress_generic(src, dst, compressedSize, dstCapacity,
-				      endOnInputSize, partial_decode,
-				      noDict, (BYTE *)dst, NULL, 0);
-}
-
-/* ===== Instantiate a few more decoding cases, used more than once. ===== */
-
-static int LZ4_decompress_safe_withPrefix64k(const char *source, char *dest,
-				      int compressedSize, int maxOutputSize)
+int LZ4_decompress_safe(const char *source, char *dest, int compressedSize,
+			int maxDecompressedSize)
 {
-	return LZ4_decompress_generic(source, dest,
-				      compressedSize, maxOutputSize,
-				      endOnInputSize, decode_full_block,
-				      withPrefix64k,
-				      (BYTE *)dest - 64 * KB, NULL, 0);
+	return __lz4_decompress_safe(source, dest, compressedSize,
+				     maxDecompressedSize);
 }
 
-static int LZ4_decompress_safe_withSmallPrefix(const char *source, char *dest,
-					       int compressedSize,
-					       int maxOutputSize,
-					       size_t prefixSize)
+int LZ4_decompress_safe_partial(const char *source, char *dest,
+				int compressedSize, int targetOutputSize,
+				int maxDecompressedSize)
 {
-	return LZ4_decompress_generic(source, dest,
-				      compressedSize, maxOutputSize,
-				      endOnInputSize, decode_full_block,
-				      noDict,
-				      (BYTE *)dest - prefixSize, NULL, 0);
+	return __lz4_decompress_safe_partial(source, dest, compressedSize,
+					     targetOutputSize,
+					     maxDecompressedSize);
 }
 
-static int LZ4_decompress_safe_forceExtDict(const char *source, char *dest,
-					    int compressedSize, int maxOutputSize,
-					    const void *dictStart, size_t dictSize)
-{
-	return LZ4_decompress_generic(source, dest,
-				      compressedSize, maxOutputSize,
-				      endOnInputSize, decode_full_block,
-				      usingExtDict, (BYTE *)dest,
-				      (const BYTE *)dictStart, dictSize);
-}
-
-/*
- * The "double dictionary" mode, for use with e.g. ring buffers: the first part
- * of the dictionary is passed as prefix, and the second via dictStart + dictSize.
- * These routines are used only once, in LZ4_decompress_*_continue().
- */
-static FORCE_INLINE
-int LZ4_decompress_safe_doubleDict(const char *source, char *dest,
-				   int compressedSize, int maxOutputSize,
-				   size_t prefixSize,
-				   const void *dictStart, size_t dictSize)
-{
-	return LZ4_decompress_generic(source, dest,
-				      compressedSize, maxOutputSize,
-				      endOnInputSize, decode_full_block,
-				      usingExtDict, (BYTE *)dest - prefixSize,
-				      (const BYTE *)dictStart, dictSize);
-}
-
-/* ===== streaming decompression functions ===== */
-
 int LZ4_setStreamDecode(LZ4_streamDecode_t *LZ4_streamDecode,
-	const char *dictionary, int dictSize)
+			const char *dictionary, int dictSize)
 {
-	LZ4_streamDecode_t_internal *lz4sd =
-		&LZ4_streamDecode->internal_donotuse;
-
-	lz4sd->prefixSize = (size_t) dictSize;
-	lz4sd->prefixEnd = (const BYTE *) dictionary + dictSize;
-	lz4sd->externalDict = NULL;
-	lz4sd->extDictSize	= 0;
-	return 1;
+	return __lz4_setStreamDecode(LZ4_streamDecode, dictionary, dictSize);
 }
 
-/*
- * *_continue() :
- * These decoding functions allow decompression of multiple blocks
- * in "streaming" mode.
- * Previously decoded blocks must still be available at the memory
- * position where they were decoded.
- * If it's not possible, save the relevant part of
- * decoded data into a safe buffer,
- * and indicate where it stands using LZ4_setStreamDecode()
- */
 int LZ4_decompress_safe_continue(LZ4_streamDecode_t *LZ4_streamDecode,
-	const char *source, char *dest, int compressedSize, int maxOutputSize)
+				 const char *source, char *dest,
+				 int compressedSize, int maxDecompressedSize)
 {
-	LZ4_streamDecode_t_internal *lz4sd =
-		&LZ4_streamDecode->internal_donotuse;
-	int result;
-
-	if (lz4sd->prefixSize == 0) {
-		/* The first call, no dictionary yet. */
-		assert(lz4sd->extDictSize == 0);
-		result = LZ4_decompress_safe(source, dest,
-			compressedSize, maxOutputSize);
-		if (result <= 0)
-			return result;
-		lz4sd->prefixSize = result;
-		lz4sd->prefixEnd = (BYTE *)dest + result;
-	} else if (lz4sd->prefixEnd == (BYTE *)dest) {
-		/* They're rolling the current segment. */
-		if (lz4sd->prefixSize >= 64 * KB - 1)
-			result = LZ4_decompress_safe_withPrefix64k(source, dest,
-				compressedSize, maxOutputSize);
-		else if (lz4sd->extDictSize == 0)
-			result = LZ4_decompress_safe_withSmallPrefix(source,
-				dest, compressedSize, maxOutputSize,
-				lz4sd->prefixSize);
-		else
-			result = LZ4_decompress_safe_doubleDict(source, dest,
-				compressedSize, maxOutputSize,
-				lz4sd->prefixSize,
-				lz4sd->externalDict, lz4sd->extDictSize);
-		if (result <= 0)
-			return result;
-		lz4sd->prefixSize += result;
-		lz4sd->prefixEnd  += result;
-	} else {
-		/*
-		 * The buffer wraps around, or they're
-		 * switching to another buffer.
-		 */
-		lz4sd->extDictSize = lz4sd->prefixSize;
-		lz4sd->externalDict = lz4sd->prefixEnd - lz4sd->extDictSize;
-		result = LZ4_decompress_safe_forceExtDict(source, dest,
-			compressedSize, maxOutputSize,
-			lz4sd->externalDict, lz4sd->extDictSize);
-		if (result <= 0)
-			return result;
-		lz4sd->prefixSize = result;
-		lz4sd->prefixEnd  = (BYTE *)dest + result;
-	}
-
-	return result;
+	return __lz4_decompress_safe_continue(LZ4_streamDecode, source, dest,
+					      compressedSize,
+					      maxDecompressedSize);
 }
 
 int LZ4_decompress_safe_usingDict(const char *source, char *dest,
-				  int compressedSize, int maxOutputSize,
+				  int compressedSize, int maxDecompressedSize,
 				  const char *dictStart, int dictSize)
 {
-	if (dictSize == 0)
-		return LZ4_decompress_safe(source, dest,
-					   compressedSize, maxOutputSize);
-	if (dictStart+dictSize == dest) {
-		if (dictSize >= 64 * KB - 1)
-			return LZ4_decompress_safe_withPrefix64k(source, dest,
-				compressedSize, maxOutputSize);
-		return LZ4_decompress_safe_withSmallPrefix(source, dest,
-			compressedSize, maxOutputSize, dictSize);
-	}
-	return LZ4_decompress_safe_forceExtDict(source, dest,
-		compressedSize, maxOutputSize, dictStart, dictSize);
+	return __lz4_decompress_safe_usingDict(source, dest, compressedSize,
+					       maxDecompressedSize, dictStart,
+					       dictSize);
 }
 
 #ifndef STATIC
diff --git a/lib/lz4/lz4defs.h b/lib/lz4/lz4defs.h
deleted file mode 100644
index 17277ec16919f3a554afc7d91ad515c493d745ca..0000000000000000000000000000000000000000
--- a/lib/lz4/lz4defs.h
+++ /dev/null
@@ -1,247 +0,0 @@
-#ifndef __LZ4DEFS_H__
-#define __LZ4DEFS_H__
-
-/*
- * lz4defs.h -- common and architecture specific defines for the kernel usage
-
- * LZ4 - Fast LZ compression algorithm
- * Copyright (C) 2011-2016, Yann Collet.
- * BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions are
- * met:
- *	* Redistributions of source code must retain the above copyright
- *	  notice, this list of conditions and the following disclaimer.
- *	* Redistributions in binary form must reproduce the above
- * copyright notice, this list of conditions and the following disclaimer
- * in the documentation and/or other materials provided with the
- * distribution.
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
- * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
- * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
- * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
- * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
- * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
- * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
- * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
- * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- * You can contact the author at :
- *	- LZ4 homepage : http://www.lz4.org
- *	- LZ4 source repository : https://github.com/lz4/lz4
- *
- *	Changed for kernel usage by:
- *	Sven Schmidt <4sschmid@informatik.uni-hamburg.de>
- */
-
-#include <linux/unaligned.h>
-
-#include <linux/bitops.h>
-#include <linux/string.h>	 /* memset, memcpy */
-#include <linux/lz4.h>
-
-#define FORCE_INLINE __always_inline
-
-/*-************************************
- *	Basic Types
- **************************************/
-#include <linux/types.h>
-
-typedef	uint8_t BYTE;
-typedef uint16_t U16;
-typedef uint32_t U32;
-typedef	int32_t S32;
-typedef uint64_t U64;
-typedef uintptr_t uptrval;
-
-/*-************************************
- *	Architecture specifics
- **************************************/
-#if defined(CONFIG_64BIT)
-#define LZ4_ARCH64 1
-#else
-#define LZ4_ARCH64 0
-#endif
-
-#if defined(__LITTLE_ENDIAN)
-#define LZ4_LITTLE_ENDIAN 1
-#else
-#define LZ4_LITTLE_ENDIAN 0
-#endif
-
-/*-************************************
- *	Constants
- **************************************/
-#define MINMATCH 4
-
-#define WILDCOPYLENGTH 8
-#define LASTLITERALS 5
-#define MFLIMIT (WILDCOPYLENGTH + MINMATCH)
-/*
- * ensure it's possible to write 2 x wildcopyLength
- * without overflowing output buffer
- */
-#define MATCH_SAFEGUARD_DISTANCE  ((2 * WILDCOPYLENGTH) - MINMATCH)
-
-/* Increase this value ==> compression run slower on incompressible data */
-#define LZ4_SKIPTRIGGER 6
-
-#define HASH_UNIT sizeof(size_t)
-
-#define KB (1 << 10)
-#define MB (1 << 20)
-#define GB (1U << 30)
-
-#define MAX_DISTANCE LZ4_DISTANCE_MAX
-#define STEPSIZE sizeof(size_t)
-
-#define ML_BITS	4
-#define ML_MASK	((1U << ML_BITS) - 1)
-#define RUN_BITS (8 - ML_BITS)
-#define RUN_MASK ((1U << RUN_BITS) - 1)
-
-/*-************************************
- *	Reading and writing into memory
- **************************************/
-static FORCE_INLINE U16 LZ4_read16(const void *ptr)
-{
-	return get_unaligned((const U16 *)ptr);
-}
-
-static FORCE_INLINE U32 LZ4_read32(const void *ptr)
-{
-	return get_unaligned((const U32 *)ptr);
-}
-
-static FORCE_INLINE size_t LZ4_read_ARCH(const void *ptr)
-{
-	return get_unaligned((const size_t *)ptr);
-}
-
-static FORCE_INLINE void LZ4_write16(void *memPtr, U16 value)
-{
-	put_unaligned(value, (U16 *)memPtr);
-}
-
-static FORCE_INLINE void LZ4_write32(void *memPtr, U32 value)
-{
-	put_unaligned(value, (U32 *)memPtr);
-}
-
-static FORCE_INLINE U16 LZ4_readLE16(const void *memPtr)
-{
-	return get_unaligned_le16(memPtr);
-}
-
-static FORCE_INLINE void LZ4_writeLE16(void *memPtr, U16 value)
-{
-	return put_unaligned_le16(value, memPtr);
-}
-
-/*
- * LZ4 relies on memcpy with a constant size being inlined. In freestanding
- * environments, the compiler can't assume the implementation of memcpy() is
- * standard compliant, so apply its specialized memcpy() inlining logic. When
- * possible, use __builtin_memcpy() to tell the compiler to analyze memcpy()
- * as-if it were standard compliant, so it can inline it in freestanding
- * environments. This is needed when decompressing the Linux Kernel, for example.
- */
-#define LZ4_memcpy(dst, src, size) __builtin_memcpy(dst, src, size)
-#define LZ4_memmove(dst, src, size) __builtin_memmove(dst, src, size)
-
-static FORCE_INLINE void LZ4_copy8(void *dst, const void *src)
-{
-#if LZ4_ARCH64
-	U64 a = get_unaligned((const U64 *)src);
-
-	put_unaligned(a, (U64 *)dst);
-#else
-	U32 a = get_unaligned((const U32 *)src);
-	U32 b = get_unaligned((const U32 *)src + 1);
-
-	put_unaligned(a, (U32 *)dst);
-	put_unaligned(b, (U32 *)dst + 1);
-#endif
-}
-
-/*
- * customized variant of memcpy,
- * which can overwrite up to 7 bytes beyond dstEnd
- */
-static FORCE_INLINE void LZ4_wildCopy(void *dstPtr,
-	const void *srcPtr, void *dstEnd)
-{
-	BYTE *d = (BYTE *)dstPtr;
-	const BYTE *s = (const BYTE *)srcPtr;
-	BYTE *const e = (BYTE *)dstEnd;
-
-	do {
-		LZ4_copy8(d, s);
-		d += 8;
-		s += 8;
-	} while (d < e);
-}
-
-static FORCE_INLINE unsigned int LZ4_NbCommonBytes(register size_t val)
-{
-#if LZ4_LITTLE_ENDIAN
-	return __ffs(val) >> 3;
-#else
-	return (BITS_PER_LONG - 1 - __fls(val)) >> 3;
-#endif
-}
-
-static FORCE_INLINE unsigned int LZ4_count(
-	const BYTE *pIn,
-	const BYTE *pMatch,
-	const BYTE *pInLimit)
-{
-	const BYTE *const pStart = pIn;
-
-	while (likely(pIn < pInLimit - (STEPSIZE - 1))) {
-		size_t const diff = LZ4_read_ARCH(pMatch) ^ LZ4_read_ARCH(pIn);
-
-		if (!diff) {
-			pIn += STEPSIZE;
-			pMatch += STEPSIZE;
-			continue;
-		}
-
-		pIn += LZ4_NbCommonBytes(diff);
-
-		return (unsigned int)(pIn - pStart);
-	}
-
-#if LZ4_ARCH64
-	if ((pIn < (pInLimit - 3))
-		&& (LZ4_read32(pMatch) == LZ4_read32(pIn))) {
-		pIn += 4;
-		pMatch += 4;
-	}
-#endif
-
-	if ((pIn < (pInLimit - 1))
-		&& (LZ4_read16(pMatch) == LZ4_read16(pIn))) {
-		pIn += 2;
-		pMatch += 2;
-	}
-
-	if ((pIn < pInLimit) && (*pMatch == *pIn))
-		pIn++;
-
-	return (unsigned int)(pIn - pStart);
-}
-
-typedef enum { noLimit = 0, limitedOutput = 1 } limitedOutput_directive;
-typedef enum { byPtr, byU32, byU16 } tableType_t;
-
-typedef enum { noDict = 0, withPrefix64k, usingExtDict } dict_directive;
-typedef enum { noDictIssue = 0, dictSmall } dictIssue_directive;
-
-typedef enum { endOnOutputSize = 0, endOnInputSize = 1 } endCondition_directive;
-typedef enum { decode_full_block = 0, partial_decode = 1 } earlyEnd_directive;
-
-#define LZ4_STATIC_ASSERT(c)	BUILD_BUG_ON(!(c))
-
-#endif

-- 
2.34.1



_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

[PATCH v2 13/15] lib/lz4: switch the decompressor to the vendored sources

[Michal Wilczynski]

Replace the forked decompressor with thin entry points over the vendored
lz4.c; the API is signature-compatible, so every export is a plain
forwarder. lz4defs.h has no users left and goes.

lib/decompress_unlz4.c includes this file before boot, together with
<linux/lz4.h>, so LZ4_streamDecode_t becomes incomplete as upstream
declares it, and LZ4_MEM_DECOMPRESS (32) replaces the
LZ4_STREAMDECODESIZE macros. LZ4_COMPRESSBOUND is defined by both
headers and gets a guard; the LZ4_compressBound() wrapper goes, as
lib/decompress_unlz4.c now uses the macro.

LZ4_decompress_safe_partial() on a truncated block can now return the
bytes decoded so far rather than an error, depending on where the cut
falls; the kernel-doc says so.

This retires six local patches:

commit 8cb5d7482810 ("lib/lz4: make arrays static const, reduces object code size")
  upstream 75fb878a9019 ("decompress_generic: Add fastpath for small offsets"), v1.9.0
commit b1a3e75e466d ("lz4: fix kernel decompression speed")
  upstream fe2a1b3707d5 ("Call LZ4_memcpy() instead of memcpy()"), v1.9.3
commit 89b158635ad7 ("lib/lz4: explicitly support in-place decompression")
  upstream 33cb8518ac38 ("decompress: changed final memcpy() into memmove()"), v1.9.2
commit 7fde9d6e839d ("lz4_decompress: declare LZ4_decompress_safe_withPrefix64k static")
  LZ4LIB_VISIBILITY in lz4_deps.h, which makes every upstream entry point static
commit eafc0a02391b ("lz4: fix LZ4_decompress_safe_partial read out of bound")
  upstream c5d6f8a8be39 ("fix #783"), v1.9.3
commit 2d8867f3e083 ("lib: make LZ4_decompress_safe_forceExtDict() static")
  the static forward declaration in lz4_deps.h

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 drivers/block/zram/backend_lz4.c   |   2 +-
 drivers/block/zram/backend_lz4hc.c |   2 +-
 include/linux/lz4.h                |  60 ++--
 lib/decompress_unlz4.c             |   4 +-
 lib/lz4/lz4_decompress.c           | 640 +++----------------------------------
 lib/lz4/lz4defs.h                  | 247 --------------
 6 files changed, 71 insertions(+), 884 deletions(-)

diff --git a/drivers/block/zram/backend_lz4.c b/drivers/block/zram/backend_lz4.c
index 50265e3ce256490994cb6dae32a45da376e408e0..b9b8a5f678c155442fa3fd2c1984e45182e6fc38 100644
--- a/drivers/block/zram/backend_lz4.c
+++ b/drivers/block/zram/backend_lz4.c
@@ -84,7 +84,7 @@ static int lz4_create(struct zcomp_params *params, struct zcomp_ctx *ctx)
 		if (!zctx->mem)
 			goto error;
 	} else {
-		zctx->dstrm = kzalloc_obj(*zctx->dstrm);
+		zctx->dstrm = kzalloc(LZ4_MEM_DECOMPRESS, GFP_KERNEL);
 		if (!zctx->dstrm)
 			goto error;
 
diff --git a/drivers/block/zram/backend_lz4hc.c b/drivers/block/zram/backend_lz4hc.c
index e7ee3740219852f079d1e0866da6b213a3f3a583..1e09c93016264259c149e2913b2819571b56fab6 100644
--- a/drivers/block/zram/backend_lz4hc.c
+++ b/drivers/block/zram/backend_lz4hc.c
@@ -65,7 +65,7 @@ static int lz4hc_create(struct zcomp_params *params, struct zcomp_ctx *ctx)
 		if (!zctx->mem)
 			goto error;
 	} else {
-		zctx->dstrm = kzalloc_obj(*zctx->dstrm);
+		zctx->dstrm = kzalloc(LZ4_MEM_DECOMPRESS, GFP_KERNEL);
 		if (!zctx->dstrm)
 			goto error;
 
diff --git a/include/linux/lz4.h b/include/linux/lz4.h
index e809149747c3dc97d129e42e0ce4385b5a219930..fd9d7a162e04655a9d19639515466a6ba8e1e892 100644
--- a/include/linux/lz4.h
+++ b/include/linux/lz4.h
@@ -48,10 +48,16 @@
  *	CONSTANTS
  **************************************************************************/
 #define LZ4_MAX_INPUT_SIZE	0x7E000000 /* 2 113 929 216 bytes */
+
+/* Upstream's lz4.h defines the same macro, and lib/lz4's entry points and
+ * the pre-boot lib/decompress_unlz4.c see both headers.
+ */
+#ifndef LZ4_COMPRESSBOUND
 #define LZ4_COMPRESSBOUND(isize)	(\
 	(unsigned int)(isize) > (unsigned int)LZ4_MAX_INPUT_SIZE \
 	? 0 \
 	: (isize) + ((isize)/255) + 16)
+#endif
 
 #define LZ4_ACCELERATION_DEFAULT 1
 
@@ -67,12 +73,8 @@
 #define LZ4HC_CLAMP_CLEVEL			10
 
 /*-************************************************************************
- *	STREAMING CONSTANTS AND STRUCTURES
+ *	STREAMING STRUCTURES
  **************************************************************************/
-#define LZ4_STREAMDECODESIZE_U64	4
-#define LZ4_STREAMDECODESIZE		 (LZ4_STREAMDECODESIZE_U64 * \
-	sizeof(unsigned long long))
-
 /*
  * LZ4_stream_t - an LZ4 stream.  Incomplete: lib/lz4 owns the layout.
  * Allocate LZ4_MEM_COMPRESS bytes and cast, do not sizeof().
@@ -86,21 +88,11 @@ typedef union LZ4_stream_u LZ4_stream_t;
 typedef union LZ4_streamHC_u LZ4_streamHC_t;
 
 /*
- * LZ4_streamDecode_t - information structure to track an
- *	LZ4 stream during decompression.
- *
- * init this structure using LZ4_setStreamDecode (or memset()) before first use
+ * LZ4_streamDecode_t - an LZ4 stream during decompression.  Incomplete:
+ * lib/lz4 owns the layout.  Allocate LZ4_MEM_DECOMPRESS bytes and cast, do
+ * not sizeof().  Init with LZ4_setStreamDecode() (or zero it) before use.
  */
-typedef struct {
-	const uint8_t *externalDict;
-	size_t extDictSize;
-	const uint8_t *prefixEnd;
-	size_t prefixSize;
-} LZ4_streamDecode_t_internal;
-typedef union {
-	unsigned long long table[LZ4_STREAMDECODESIZE_U64];
-	LZ4_streamDecode_t_internal internal_donotuse;
-} LZ4_streamDecode_t;
+typedef union LZ4_streamDecode_u LZ4_streamDecode_t;
 
 /*-************************************************************************
  *	SIZE OF STATE
@@ -112,23 +104,12 @@ typedef union {
  */
 #define LZ4_MEM_COMPRESS	16416
 #define LZ4HC_MEM_COMPRESS	262200
+#define LZ4_MEM_DECOMPRESS	32
 
 /*-************************************************************************
  *	Compression Functions
  **************************************************************************/
 
-/**
- * LZ4_compressBound() - Max. output size in worst case szenarios
- * @isize: Size of the input data
- *
- * Return: Max. size LZ4 may output in a "worst case" szenario
- * (data not compressible)
- */
-static inline int LZ4_compressBound(size_t isize)
-{
-	return LZ4_COMPRESSBOUND(isize);
-}
-
 /**
  * LZ4_compress_default() - Compress data from source to dest
  * @source: source address of the original data
@@ -142,7 +123,7 @@ static inline int LZ4_compressBound(size_t isize)
  * Compresses 'sourceSize' bytes from buffer 'source'
  * into already allocated 'dest' buffer of size 'maxOutputSize'.
  * Compression is guaranteed to succeed if
- * 'maxOutputSize' >= LZ4_compressBound(inputSize).
+ * 'maxOutputSize' >= LZ4_COMPRESSBOUND(inputSize).
  * It also runs faster, so it's a recommended setting.
  * If the function cannot compress 'source' into a more limited 'dest' budget,
  * compression stops *immediately*, and the function result is zero.
@@ -234,7 +215,7 @@ int LZ4_decompress_safe(const char *source, char *dest, int compressedSize,
  * @source: source address of the compressed data
  * @dest: output buffer address of the decompressed data which must be
  *	already allocated
- * @compressedSize: is the precise full size of the compressed block.
+ * @compressedSize: size of the compressed data at 'source'
  * @targetOutputSize: the decompression operation will try
  *	to stop as soon as 'targetOutputSize' has been reached
  * @maxDecompressedSize: is the size of destination buffer
@@ -249,8 +230,11 @@ int LZ4_decompress_safe(const char *source, char *dest, int compressedSize,
  * It is therefore protected against malicious data packets.
  *
  * Return: the number of bytes decoded in the destination buffer
- *	(necessarily <= maxDecompressedSize)
- *	or a negative result in case of error
+ *	(necessarily <= targetOutputSize and maxDecompressedSize)
+ *	or a negative result in case of error.  For a block cut short the
+ *	result is either the bytes decoded so far, below targetOutputSize,
+ *	or an error, depending on where the cut falls; check it against the
+ *	size you expect.
  *
  */
 int LZ4_decompress_safe_partial(const char *source, char *dest,
@@ -275,7 +259,7 @@ int LZ4_decompress_safe_partial(const char *source, char *dest,
  *
  * Compress data from 'src' into 'dst', using the more powerful
  * but slower "HC" algorithm. Compression is guaranteed to succeed if
- * `dstCapacity >= LZ4_compressBound(srcSize)
+ * `dstCapacity >= LZ4_COMPRESSBOUND(srcSize)
  *
  * Return : the number of bytes written into 'dst' or 0 if compression fails.
  */
@@ -339,7 +323,7 @@ int	LZ4_loadDictHC(LZ4_streamHC_t *streamHCPtr, const char *dictionary,
  * (including initial dictionary when present) must remain accessible
  * and unmodified during compression.
  * 'dst' buffer should be sized to handle worst case scenarios, using
- *  LZ4_compressBound(), to ensure operation success.
+ *  LZ4_COMPRESSBOUND(), to ensure operation success.
  *  If, for any reason, previous data blocks can't be preserved unmodified
  *  in memory during next compression block,
  *  you must save it to a safer memory space, using LZ4_saveDictHC().
@@ -435,7 +419,7 @@ int LZ4_saveDict(LZ4_stream_t *streamPtr, char *safeBuffer, int dictSize);
  * as dictionary to improve compression ratio.
  * Important : Previous data blocks are assumed to still
  * be present and unmodified !
- * If maxDstSize >= LZ4_compressBound(srcSize),
+ * If maxDstSize >= LZ4_COMPRESSBOUND(srcSize),
  * compression is guaranteed to succeed, and runs faster.
  *
  * Return: Number of bytes written into buffer 'dst'  or 0 if compression fails
diff --git a/lib/decompress_unlz4.c b/lib/decompress_unlz4.c
index dc20c22e087900e21da49f9fcb267e6b511e1072..87eea1fbdea063e241dd98f994d7fee078b7a910 100644
--- a/lib/decompress_unlz4.c
+++ b/lib/decompress_unlz4.c
@@ -69,7 +69,7 @@ STATIC inline int INIT unlz4(u8 *input, long in_len,
 		error("NULL input pointer and missing fill function");
 		goto exit_1;
 	} else {
-		inp = large_malloc(LZ4_compressBound(uncomp_chunksize));
+		inp = large_malloc(LZ4_COMPRESSBOUND(uncomp_chunksize));
 		if (!inp) {
 			error("Could not allocate input buffer");
 			goto exit_1;
@@ -140,7 +140,7 @@ STATIC inline int INIT unlz4(u8 *input, long in_len,
 			inp += 4;
 			size -= 4;
 		} else {
-			if (chunksize > LZ4_compressBound(uncomp_chunksize)) {
+			if (chunksize > LZ4_COMPRESSBOUND(uncomp_chunksize)) {
 				error("chunk length is longer than allocated");
 				goto exit_2;
 			}
diff --git a/lib/lz4/lz4_decompress.c b/lib/lz4/lz4_decompress.c
index f89b0e547f2156990c16770f69f89e0d8b46156d..60b57862a2e387aa89fa26f6f8e2b5485b5d9995 100644
--- a/lib/lz4/lz4_decompress.c
+++ b/lib/lz4/lz4_decompress.c
@@ -1,626 +1,76 @@
+// SPDX-License-Identifier: GPL-2.0-only OR BSD-2-Clause
 /*
- * LZ4 - Fast LZ compression algorithm
  * Copyright (C) 2011 - 2016, Yann Collet.
- * BSD 2 - Clause License (http://www.opensource.org/licenses/bsd - license.php)
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions are
- * met:
- *	* Redistributions of source code must retain the above copyright
- *	  notice, this list of conditions and the following disclaimer.
- *	* Redistributions in binary form must reproduce the above
- * copyright notice, this list of conditions and the following disclaimer
- * in the documentation and/or other materials provided with the
- * distribution.
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
- * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
- * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
- * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
- * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
- * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
- * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
- * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
- * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- * You can contact the author at :
- *	- LZ4 homepage : http://www.lz4.org
- *	- LZ4 source repository : https://github.com/lz4/lz4
+ * Copyright (C) 2016, Sven Schmidt <4sschmid@informatik.uni-hamburg.de>
+ * Copyright (c) 2026 Samsung Electronics Co., Ltd.
+ * Author: Michal Wilczynski <m.wilczynski@samsung.com>
  *
- *	Changed for kernel usage by:
- *	Sven Schmidt <4sschmid@informatik.uni-hamburg.de>
+ * LZ4 decompressor -- kernel entry points
+ *
+ * The decompressor is upstream's, in the verbatim lz4.c included below.  The
+ * API is signature-compatible, so every export is a plain forwarder.  This
+ * file is also included by lib/decompress_unlz4.c for the pre-boot
+ * decompressor, which defines STATIC.
  */
 
-/*-************************************
- *	Dependencies
- **************************************/
-#include "lz4defs.h"
-#include <linux/init.h>
-#include <linux/module.h>
-#include <linux/kernel.h>
-#include <linux/unaligned.h>
+#include "lz4_deps.h"
+#include "upstream/lz4.c"
 
-/*-*****************************
- *	Decompression functions
- *******************************/
+/* Upstream's short helper macros; MIN and MAX clash with <linux/minmax.h>. */
+#undef MIN
+#undef MAX
+#undef KB
+#undef MB
+#undef GB
 
-#define DEBUGLOG(l, ...) {}	/* disabled */
+#include "lz4_kernel_api.h"
 
-#ifndef assert
-#define assert(condition) ((void)0)
+#ifndef STATIC
+#include <linux/export.h>
+#include <linux/module.h>
 #endif
 
-/*
- * LZ4_decompress_generic() :
- * This generic decompression function covers all use cases.
- * It shall be instantiated several times, using different sets of directives.
- * Note that it is important for performance that this function really get inlined,
- * in order to remove useless branches during compilation optimization.
- */
-static FORCE_INLINE int LZ4_decompress_generic(
-	 const char * const src,
-	 char * const dst,
-	 int srcSize,
-		/*
-		 * If endOnInput == endOnInputSize,
-		 * this value is `dstCapacity`
-		 */
-	 int outputSize,
-	 /* endOnOutputSize, endOnInputSize */
-	 endCondition_directive endOnInput,
-	 /* full, partial */
-	 earlyEnd_directive partialDecoding,
-	 /* noDict, withPrefix64k, usingExtDict */
-	 dict_directive dict,
-	 /* always <= dst, == dst when no prefix */
-	 const BYTE * const lowPrefix,
-	 /* only if dict == usingExtDict */
-	 const BYTE * const dictStart,
-	 /* note : = 0 if noDict */
-	 const size_t dictSize
-	 )
-{
-	const BYTE *ip = (const BYTE *) src;
-	const BYTE * const iend = ip + srcSize;
-
-	BYTE *op = (BYTE *) dst;
-	BYTE * const oend = op + outputSize;
-	BYTE *cpy;
-
-	const BYTE * const dictEnd = (const BYTE *)dictStart + dictSize;
-	static const unsigned int inc32table[8] = {0, 1, 2, 1, 0, 4, 4, 4};
-	static const int dec64table[8] = {0, 0, 0, -1, -4, 1, 2, 3};
-
-	const int safeDecode = (endOnInput == endOnInputSize);
-	const int checkOffset = ((safeDecode) && (dictSize < (int)(64 * KB)));
-
-	/* Set up the "end" pointers for the shortcut. */
-	const BYTE *const shortiend = iend -
-		(endOnInput ? 14 : 8) /*maxLL*/ - 2 /*offset*/;
-	const BYTE *const shortoend = oend -
-		(endOnInput ? 14 : 8) /*maxLL*/ - 18 /*maxML*/;
-
-	DEBUGLOG(5, "%s (srcSize:%i, dstSize:%i)", __func__,
-		 srcSize, outputSize);
-
-	/* Special cases */
-	assert(lowPrefix <= op);
-	assert(src != NULL);
-
-	/* Empty output buffer */
-	if ((endOnInput) && (unlikely(outputSize == 0)))
-		return ((srcSize == 1) && (*ip == 0)) ? 0 : -1;
-
-	if ((!endOnInput) && (unlikely(outputSize == 0)))
-		return (*ip == 0 ? 1 : -1);
-
-	if ((endOnInput) && unlikely(srcSize == 0))
-		return -1;
-
-	/* Main Loop : decode sequences */
-	while (1) {
-		size_t length;
-		const BYTE *match;
-		size_t offset;
-
-		/* get literal length */
-		unsigned int const token = *ip++;
-		length = token>>ML_BITS;
-
-		/* ip < iend before the increment */
-		assert(!endOnInput || ip <= iend);
-
-		/*
-		 * A two-stage shortcut for the most common case:
-		 * 1) If the literal length is 0..14, and there is enough
-		 * space, enter the shortcut and copy 16 bytes on behalf
-		 * of the literals (in the fast mode, only 8 bytes can be
-		 * safely copied this way).
-		 * 2) Further if the match length is 4..18, copy 18 bytes
-		 * in a similar manner; but we ensure that there's enough
-		 * space in the output for those 18 bytes earlier, upon
-		 * entering the shortcut (in other words, there is a
-		 * combined check for both stages).
-		 *
-		 * The & in the likely() below is intentionally not && so that
-		 * some compilers can produce better parallelized runtime code
-		 */
-		if ((endOnInput ? length != RUN_MASK : length <= 8)
-		   /*
-		    * strictly "less than" on input, to re-enter
-		    * the loop with at least one byte
-		    */
-		   && likely((endOnInput ? ip < shortiend : 1) &
-			     (op <= shortoend))) {
-			/* Copy the literals */
-			LZ4_memcpy(op, ip, endOnInput ? 16 : 8);
-			op += length; ip += length;
-
-			/*
-			 * The second stage:
-			 * prepare for match copying, decode full info.
-			 * If it doesn't work out, the info won't be wasted.
-			 */
-			length = token & ML_MASK; /* match length */
-			offset = LZ4_readLE16(ip);
-			ip += 2;
-			match = op - offset;
-			assert(match <= op); /* check overflow */
-
-			/* Do not deal with overlapping matches. */
-			if ((length != ML_MASK) &&
-			    (offset >= 8) &&
-			    (dict == withPrefix64k || match >= lowPrefix)) {
-				/* Copy the match. */
-				LZ4_memcpy(op + 0, match + 0, 8);
-				LZ4_memcpy(op + 8, match + 8, 8);
-				LZ4_memcpy(op + 16, match + 16, 2);
-				op += length + MINMATCH;
-				/* Both stages worked, load the next token. */
-				continue;
-			}
-
-			/*
-			 * The second stage didn't work out, but the info
-			 * is ready. Propel it right to the point of match
-			 * copying.
-			 */
-			goto _copy_match;
-		}
-
-		/* decode literal length */
-		if (length == RUN_MASK) {
-			unsigned int s;
-
-			if (unlikely(endOnInput ? ip >= iend - RUN_MASK : 0)) {
-				/* overflow detection */
-				goto _output_error;
-			}
-			do {
-				s = *ip++;
-				length += s;
-			} while (likely(endOnInput
-				? ip < iend - RUN_MASK
-				: 1) & (s == 255));
-
-			if ((safeDecode)
-			    && unlikely((uptrval)(op) +
-					length < (uptrval)(op))) {
-				/* overflow detection */
-				goto _output_error;
-			}
-			if ((safeDecode)
-			    && unlikely((uptrval)(ip) +
-					length < (uptrval)(ip))) {
-				/* overflow detection */
-				goto _output_error;
-			}
-		}
-
-		/* copy literals */
-		cpy = op + length;
-		LZ4_STATIC_ASSERT(MFLIMIT >= WILDCOPYLENGTH);
-
-		if (((endOnInput) && ((cpy > oend - MFLIMIT)
-			|| (ip + length > iend - (2 + 1 + LASTLITERALS))))
-			|| ((!endOnInput) && (cpy > oend - WILDCOPYLENGTH))) {
-			if (partialDecoding) {
-				if (cpy > oend) {
-					/*
-					 * Partial decoding :
-					 * stop in the middle of literal segment
-					 */
-					cpy = oend;
-					length = oend - op;
-				}
-				if ((endOnInput)
-					&& (ip + length > iend)) {
-					/*
-					 * Error :
-					 * read attempt beyond
-					 * end of input buffer
-					 */
-					goto _output_error;
-				}
-			} else {
-				if ((!endOnInput)
-					&& (cpy != oend)) {
-					/*
-					 * Error :
-					 * block decoding must
-					 * stop exactly there
-					 */
-					goto _output_error;
-				}
-				if ((endOnInput)
-					&& ((ip + length != iend)
-					|| (cpy > oend))) {
-					/*
-					 * Error :
-					 * input must be consumed
-					 */
-					goto _output_error;
-				}
-			}
-
-			/*
-			 * supports overlapping memory regions; only matters
-			 * for in-place decompression scenarios
-			 */
-			LZ4_memmove(op, ip, length);
-			ip += length;
-			op += length;
-
-			/* Necessarily EOF when !partialDecoding.
-			 * When partialDecoding, it is EOF if we've either
-			 * filled the output buffer or
-			 * can't proceed with reading an offset for following match.
-			 */
-			if (!partialDecoding || (cpy == oend) || (ip >= (iend - 2)))
-				break;
-		} else {
-			/* may overwrite up to WILDCOPYLENGTH beyond cpy */
-			LZ4_wildCopy(op, ip, cpy);
-			ip += length;
-			op = cpy;
-		}
-
-		/* get offset */
-		offset = LZ4_readLE16(ip);
-		ip += 2;
-		match = op - offset;
-
-		/* get matchlength */
-		length = token & ML_MASK;
-
-_copy_match:
-		if ((checkOffset) && (unlikely(match + dictSize < lowPrefix))) {
-			/* Error : offset outside buffers */
-			goto _output_error;
-		}
-
-		/* costs ~1%; silence an msan warning when offset == 0 */
-		/*
-		 * note : when partialDecoding, there is no guarantee that
-		 * at least 4 bytes remain available in output buffer
-		 */
-		if (!partialDecoding) {
-			assert(oend > op);
-			assert(oend - op >= 4);
-
-			LZ4_write32(op, (U32)offset);
-		}
-
-		if (length == ML_MASK) {
-			unsigned int s;
-
-			do {
-				s = *ip++;
-
-				if ((endOnInput) && (ip > iend - LASTLITERALS))
-					goto _output_error;
-
-				length += s;
-			} while (s == 255);
+static_assert(sizeof(LZ4_streamDecode_t) == LZ4_STREAMDECODE_MINSIZE);
+static_assert(LZ4_STREAMDECODE_MINSIZE == 32); /* LZ4_MEM_DECOMPRESS */
 
-			if ((safeDecode)
-				&& unlikely(
-					(uptrval)(op) + length < (uptrval)op)) {
-				/* overflow detection */
-				goto _output_error;
-			}
-		}
-
-		length += MINMATCH;
-
-		/* match starting within external dictionary */
-		if ((dict == usingExtDict) && (match < lowPrefix)) {
-			if (unlikely(op + length > oend - LASTLITERALS)) {
-				/* doesn't respect parsing restriction */
-				if (!partialDecoding)
-					goto _output_error;
-				length = min(length, (size_t)(oend - op));
-			}
-
-			if (length <= (size_t)(lowPrefix - match)) {
-				/*
-				 * match fits entirely within external
-				 * dictionary : just copy
-				 */
-				memmove(op, dictEnd - (lowPrefix - match),
-					length);
-				op += length;
-			} else {
-				/*
-				 * match stretches into both external
-				 * dictionary and current block
-				 */
-				size_t const copySize = (size_t)(lowPrefix - match);
-				size_t const restSize = length - copySize;
-
-				LZ4_memcpy(op, dictEnd - copySize, copySize);
-				op += copySize;
-				if (restSize > (size_t)(op - lowPrefix)) {
-					/* overlap copy */
-					BYTE * const endOfMatch = op + restSize;
-					const BYTE *copyFrom = lowPrefix;
-
-					while (op < endOfMatch)
-						*op++ = *copyFrom++;
-				} else {
-					LZ4_memcpy(op, lowPrefix, restSize);
-					op += restSize;
-				}
-			}
-			continue;
-		}
-
-		/* copy match within block */
-		cpy = op + length;
-
-		/*
-		 * partialDecoding :
-		 * may not respect endBlock parsing restrictions
-		 */
-		assert(op <= oend);
-		if (partialDecoding &&
-		    (cpy > oend - MATCH_SAFEGUARD_DISTANCE)) {
-			size_t const mlen = min(length, (size_t)(oend - op));
-			const BYTE * const matchEnd = match + mlen;
-			BYTE * const copyEnd = op + mlen;
-
-			if (matchEnd > op) {
-				/* overlap copy */
-				while (op < copyEnd)
-					*op++ = *match++;
-			} else {
-				LZ4_memcpy(op, match, mlen);
-			}
-			op = copyEnd;
-			if (op == oend)
-				break;
-			continue;
-		}
-
-		if (unlikely(offset < 8)) {
-			op[0] = match[0];
-			op[1] = match[1];
-			op[2] = match[2];
-			op[3] = match[3];
-			match += inc32table[offset];
-			LZ4_memcpy(op + 4, match, 4);
-			match -= dec64table[offset];
-		} else {
-			LZ4_copy8(op, match);
-			match += 8;
-		}
-
-		op += 8;
-
-		if (unlikely(cpy > oend - MATCH_SAFEGUARD_DISTANCE)) {
-			BYTE * const oCopyLimit = oend - (WILDCOPYLENGTH - 1);
-
-			if (cpy > oend - LASTLITERALS) {
-				/*
-				 * Error : last LASTLITERALS bytes
-				 * must be literals (uncompressed)
-				 */
-				goto _output_error;
-			}
-
-			if (op < oCopyLimit) {
-				LZ4_wildCopy(op, match, oCopyLimit);
-				match += oCopyLimit - op;
-				op = oCopyLimit;
-			}
-			while (op < cpy)
-				*op++ = *match++;
-		} else {
-			LZ4_copy8(op, match);
-			if (length > 16)
-				LZ4_wildCopy(op + 8, match + 8, cpy);
-		}
-		op = cpy; /* wildcopy correction */
-	}
-
-	/* end of decoding */
-	if (endOnInput) {
-		/* Nb of output bytes decoded */
-		return (int) (((char *)op) - dst);
-	} else {
-		/* Nb of input bytes read */
-		return (int) (((const char *)ip) - src);
-	}
-
-	/* Overflow error detected */
-_output_error:
-	return (int) (-(((const char *)ip) - src)) - 1;
-}
-
-int LZ4_decompress_safe(const char *source, char *dest,
-	int compressedSize, int maxDecompressedSize)
-{
-	return LZ4_decompress_generic(source, dest,
-				      compressedSize, maxDecompressedSize,
-				      endOnInputSize, decode_full_block,
-				      noDict, (BYTE *)dest, NULL, 0);
-}
-
-int LZ4_decompress_safe_partial(const char *src, char *dst,
-	int compressedSize, int targetOutputSize, int dstCapacity)
-{
-	dstCapacity = min(targetOutputSize, dstCapacity);
-	return LZ4_decompress_generic(src, dst, compressedSize, dstCapacity,
-				      endOnInputSize, partial_decode,
-				      noDict, (BYTE *)dst, NULL, 0);
-}
-
-/* ===== Instantiate a few more decoding cases, used more than once. ===== */
-
-static int LZ4_decompress_safe_withPrefix64k(const char *source, char *dest,
-				      int compressedSize, int maxOutputSize)
+int LZ4_decompress_safe(const char *source, char *dest, int compressedSize,
+			int maxDecompressedSize)
 {
-	return LZ4_decompress_generic(source, dest,
-				      compressedSize, maxOutputSize,
-				      endOnInputSize, decode_full_block,
-				      withPrefix64k,
-				      (BYTE *)dest - 64 * KB, NULL, 0);
+	return __lz4_decompress_safe(source, dest, compressedSize,
+				     maxDecompressedSize);
 }
 
-static int LZ4_decompress_safe_withSmallPrefix(const char *source, char *dest,
-					       int compressedSize,
-					       int maxOutputSize,
-					       size_t prefixSize)
+int LZ4_decompress_safe_partial(const char *source, char *dest,
+				int compressedSize, int targetOutputSize,
+				int maxDecompressedSize)
 {
-	return LZ4_decompress_generic(source, dest,
-				      compressedSize, maxOutputSize,
-				      endOnInputSize, decode_full_block,
-				      noDict,
-				      (BYTE *)dest - prefixSize, NULL, 0);
+	return __lz4_decompress_safe_partial(source, dest, compressedSize,
+					     targetOutputSize,
+					     maxDecompressedSize);
 }
 
-static int LZ4_decompress_safe_forceExtDict(const char *source, char *dest,
-					    int compressedSize, int maxOutputSize,
-					    const void *dictStart, size_t dictSize)
-{
-	return LZ4_decompress_generic(source, dest,
-				      compressedSize, maxOutputSize,
-				      endOnInputSize, decode_full_block,
-				      usingExtDict, (BYTE *)dest,
-				      (const BYTE *)dictStart, dictSize);
-}
-
-/*
- * The "double dictionary" mode, for use with e.g. ring buffers: the first part
- * of the dictionary is passed as prefix, and the second via dictStart + dictSize.
- * These routines are used only once, in LZ4_decompress_*_continue().
- */
-static FORCE_INLINE
-int LZ4_decompress_safe_doubleDict(const char *source, char *dest,
-				   int compressedSize, int maxOutputSize,
-				   size_t prefixSize,
-				   const void *dictStart, size_t dictSize)
-{
-	return LZ4_decompress_generic(source, dest,
-				      compressedSize, maxOutputSize,
-				      endOnInputSize, decode_full_block,
-				      usingExtDict, (BYTE *)dest - prefixSize,
-				      (const BYTE *)dictStart, dictSize);
-}
-
-/* ===== streaming decompression functions ===== */
-
 int LZ4_setStreamDecode(LZ4_streamDecode_t *LZ4_streamDecode,
-	const char *dictionary, int dictSize)
+			const char *dictionary, int dictSize)
 {
-	LZ4_streamDecode_t_internal *lz4sd =
-		&LZ4_streamDecode->internal_donotuse;
-
-	lz4sd->prefixSize = (size_t) dictSize;
-	lz4sd->prefixEnd = (const BYTE *) dictionary + dictSize;
-	lz4sd->externalDict = NULL;
-	lz4sd->extDictSize	= 0;
-	return 1;
+	return __lz4_setStreamDecode(LZ4_streamDecode, dictionary, dictSize);
 }
 
-/*
- * *_continue() :
- * These decoding functions allow decompression of multiple blocks
- * in "streaming" mode.
- * Previously decoded blocks must still be available at the memory
- * position where they were decoded.
- * If it's not possible, save the relevant part of
- * decoded data into a safe buffer,
- * and indicate where it stands using LZ4_setStreamDecode()
- */
 int LZ4_decompress_safe_continue(LZ4_streamDecode_t *LZ4_streamDecode,
-	const char *source, char *dest, int compressedSize, int maxOutputSize)
+				 const char *source, char *dest,
+				 int compressedSize, int maxDecompressedSize)
 {
-	LZ4_streamDecode_t_internal *lz4sd =
-		&LZ4_streamDecode->internal_donotuse;
-	int result;
-
-	if (lz4sd->prefixSize == 0) {
-		/* The first call, no dictionary yet. */
-		assert(lz4sd->extDictSize == 0);
-		result = LZ4_decompress_safe(source, dest,
-			compressedSize, maxOutputSize);
-		if (result <= 0)
-			return result;
-		lz4sd->prefixSize = result;
-		lz4sd->prefixEnd = (BYTE *)dest + result;
-	} else if (lz4sd->prefixEnd == (BYTE *)dest) {
-		/* They're rolling the current segment. */
-		if (lz4sd->prefixSize >= 64 * KB - 1)
-			result = LZ4_decompress_safe_withPrefix64k(source, dest,
-				compressedSize, maxOutputSize);
-		else if (lz4sd->extDictSize == 0)
-			result = LZ4_decompress_safe_withSmallPrefix(source,
-				dest, compressedSize, maxOutputSize,
-				lz4sd->prefixSize);
-		else
-			result = LZ4_decompress_safe_doubleDict(source, dest,
-				compressedSize, maxOutputSize,
-				lz4sd->prefixSize,
-				lz4sd->externalDict, lz4sd->extDictSize);
-		if (result <= 0)
-			return result;
-		lz4sd->prefixSize += result;
-		lz4sd->prefixEnd  += result;
-	} else {
-		/*
-		 * The buffer wraps around, or they're
-		 * switching to another buffer.
-		 */
-		lz4sd->extDictSize = lz4sd->prefixSize;
-		lz4sd->externalDict = lz4sd->prefixEnd - lz4sd->extDictSize;
-		result = LZ4_decompress_safe_forceExtDict(source, dest,
-			compressedSize, maxOutputSize,
-			lz4sd->externalDict, lz4sd->extDictSize);
-		if (result <= 0)
-			return result;
-		lz4sd->prefixSize = result;
-		lz4sd->prefixEnd  = (BYTE *)dest + result;
-	}
-
-	return result;
+	return __lz4_decompress_safe_continue(LZ4_streamDecode, source, dest,
+					      compressedSize,
+					      maxDecompressedSize);
 }
 
 int LZ4_decompress_safe_usingDict(const char *source, char *dest,
-				  int compressedSize, int maxOutputSize,
+				  int compressedSize, int maxDecompressedSize,
 				  const char *dictStart, int dictSize)
 {
-	if (dictSize == 0)
-		return LZ4_decompress_safe(source, dest,
-					   compressedSize, maxOutputSize);
-	if (dictStart+dictSize == dest) {
-		if (dictSize >= 64 * KB - 1)
-			return LZ4_decompress_safe_withPrefix64k(source, dest,
-				compressedSize, maxOutputSize);
-		return LZ4_decompress_safe_withSmallPrefix(source, dest,
-			compressedSize, maxOutputSize, dictSize);
-	}
-	return LZ4_decompress_safe_forceExtDict(source, dest,
-		compressedSize, maxOutputSize, dictStart, dictSize);
+	return __lz4_decompress_safe_usingDict(source, dest, compressedSize,
+					       maxDecompressedSize, dictStart,
+					       dictSize);
 }
 
 #ifndef STATIC
diff --git a/lib/lz4/lz4defs.h b/lib/lz4/lz4defs.h
deleted file mode 100644
index 17277ec16919f3a554afc7d91ad515c493d745ca..0000000000000000000000000000000000000000
--- a/lib/lz4/lz4defs.h
+++ /dev/null
@@ -1,247 +0,0 @@
-#ifndef __LZ4DEFS_H__
-#define __LZ4DEFS_H__
-
-/*
- * lz4defs.h -- common and architecture specific defines for the kernel usage
-
- * LZ4 - Fast LZ compression algorithm
- * Copyright (C) 2011-2016, Yann Collet.
- * BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions are
- * met:
- *	* Redistributions of source code must retain the above copyright
- *	  notice, this list of conditions and the following disclaimer.
- *	* Redistributions in binary form must reproduce the above
- * copyright notice, this list of conditions and the following disclaimer
- * in the documentation and/or other materials provided with the
- * distribution.
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
- * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
- * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
- * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
- * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
- * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
- * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
- * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
- * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- * You can contact the author at :
- *	- LZ4 homepage : http://www.lz4.org
- *	- LZ4 source repository : https://github.com/lz4/lz4
- *
- *	Changed for kernel usage by:
- *	Sven Schmidt <4sschmid@informatik.uni-hamburg.de>
- */
-
-#include <linux/unaligned.h>
-
-#include <linux/bitops.h>
-#include <linux/string.h>	 /* memset, memcpy */
-#include <linux/lz4.h>
-
-#define FORCE_INLINE __always_inline
-
-/*-************************************
- *	Basic Types
- **************************************/
-#include <linux/types.h>
-
-typedef	uint8_t BYTE;
-typedef uint16_t U16;
-typedef uint32_t U32;
-typedef	int32_t S32;
-typedef uint64_t U64;
-typedef uintptr_t uptrval;
-
-/*-************************************
- *	Architecture specifics
- **************************************/
-#if defined(CONFIG_64BIT)
-#define LZ4_ARCH64 1
-#else
-#define LZ4_ARCH64 0
-#endif
-
-#if defined(__LITTLE_ENDIAN)
-#define LZ4_LITTLE_ENDIAN 1
-#else
-#define LZ4_LITTLE_ENDIAN 0
-#endif
-
-/*-************************************
- *	Constants
- **************************************/
-#define MINMATCH 4
-
-#define WILDCOPYLENGTH 8
-#define LASTLITERALS 5
-#define MFLIMIT (WILDCOPYLENGTH + MINMATCH)
-/*
- * ensure it's possible to write 2 x wildcopyLength
- * without overflowing output buffer
- */
-#define MATCH_SAFEGUARD_DISTANCE  ((2 * WILDCOPYLENGTH) - MINMATCH)
-
-/* Increase this value ==> compression run slower on incompressible data */
-#define LZ4_SKIPTRIGGER 6
-
-#define HASH_UNIT sizeof(size_t)
-
-#define KB (1 << 10)
-#define MB (1 << 20)
-#define GB (1U << 30)
-
-#define MAX_DISTANCE LZ4_DISTANCE_MAX
-#define STEPSIZE sizeof(size_t)
-
-#define ML_BITS	4
-#define ML_MASK	((1U << ML_BITS) - 1)
-#define RUN_BITS (8 - ML_BITS)
-#define RUN_MASK ((1U << RUN_BITS) - 1)
-
-/*-************************************
- *	Reading and writing into memory
- **************************************/
-static FORCE_INLINE U16 LZ4_read16(const void *ptr)
-{
-	return get_unaligned((const U16 *)ptr);
-}
-
-static FORCE_INLINE U32 LZ4_read32(const void *ptr)
-{
-	return get_unaligned((const U32 *)ptr);
-}
-
-static FORCE_INLINE size_t LZ4_read_ARCH(const void *ptr)
-{
-	return get_unaligned((const size_t *)ptr);
-}
-
-static FORCE_INLINE void LZ4_write16(void *memPtr, U16 value)
-{
-	put_unaligned(value, (U16 *)memPtr);
-}
-
-static FORCE_INLINE void LZ4_write32(void *memPtr, U32 value)
-{
-	put_unaligned(value, (U32 *)memPtr);
-}
-
-static FORCE_INLINE U16 LZ4_readLE16(const void *memPtr)
-{
-	return get_unaligned_le16(memPtr);
-}
-
-static FORCE_INLINE void LZ4_writeLE16(void *memPtr, U16 value)
-{
-	return put_unaligned_le16(value, memPtr);
-}
-
-/*
- * LZ4 relies on memcpy with a constant size being inlined. In freestanding
- * environments, the compiler can't assume the implementation of memcpy() is
- * standard compliant, so apply its specialized memcpy() inlining logic. When
- * possible, use __builtin_memcpy() to tell the compiler to analyze memcpy()
- * as-if it were standard compliant, so it can inline it in freestanding
- * environments. This is needed when decompressing the Linux Kernel, for example.
- */
-#define LZ4_memcpy(dst, src, size) __builtin_memcpy(dst, src, size)
-#define LZ4_memmove(dst, src, size) __builtin_memmove(dst, src, size)
-
-static FORCE_INLINE void LZ4_copy8(void *dst, const void *src)
-{
-#if LZ4_ARCH64
-	U64 a = get_unaligned((const U64 *)src);
-
-	put_unaligned(a, (U64 *)dst);
-#else
-	U32 a = get_unaligned((const U32 *)src);
-	U32 b = get_unaligned((const U32 *)src + 1);
-
-	put_unaligned(a, (U32 *)dst);
-	put_unaligned(b, (U32 *)dst + 1);
-#endif
-}
-
-/*
- * customized variant of memcpy,
- * which can overwrite up to 7 bytes beyond dstEnd
- */
-static FORCE_INLINE void LZ4_wildCopy(void *dstPtr,
-	const void *srcPtr, void *dstEnd)
-{
-	BYTE *d = (BYTE *)dstPtr;
-	const BYTE *s = (const BYTE *)srcPtr;
-	BYTE *const e = (BYTE *)dstEnd;
-
-	do {
-		LZ4_copy8(d, s);
-		d += 8;
-		s += 8;
-	} while (d < e);
-}
-
-static FORCE_INLINE unsigned int LZ4_NbCommonBytes(register size_t val)
-{
-#if LZ4_LITTLE_ENDIAN
-	return __ffs(val) >> 3;
-#else
-	return (BITS_PER_LONG - 1 - __fls(val)) >> 3;
-#endif
-}
-
-static FORCE_INLINE unsigned int LZ4_count(
-	const BYTE *pIn,
-	const BYTE *pMatch,
-	const BYTE *pInLimit)
-{
-	const BYTE *const pStart = pIn;
-
-	while (likely(pIn < pInLimit - (STEPSIZE - 1))) {
-		size_t const diff = LZ4_read_ARCH(pMatch) ^ LZ4_read_ARCH(pIn);
-
-		if (!diff) {
-			pIn += STEPSIZE;
-			pMatch += STEPSIZE;
-			continue;
-		}
-
-		pIn += LZ4_NbCommonBytes(diff);
-
-		return (unsigned int)(pIn - pStart);
-	}
-
-#if LZ4_ARCH64
-	if ((pIn < (pInLimit - 3))
-		&& (LZ4_read32(pMatch) == LZ4_read32(pIn))) {
-		pIn += 4;
-		pMatch += 4;
-	}
-#endif
-
-	if ((pIn < (pInLimit - 1))
-		&& (LZ4_read16(pMatch) == LZ4_read16(pIn))) {
-		pIn += 2;
-		pMatch += 2;
-	}
-
-	if ((pIn < pInLimit) && (*pMatch == *pIn))
-		pIn++;
-
-	return (unsigned int)(pIn - pStart);
-}
-
-typedef enum { noLimit = 0, limitedOutput = 1 } limitedOutput_directive;
-typedef enum { byPtr, byU32, byU16 } tableType_t;
-
-typedef enum { noDict = 0, withPrefix64k, usingExtDict } dict_directive;
-typedef enum { noDictIssue = 0, dictSmall } dictIssue_directive;
-
-typedef enum { endOnOutputSize = 0, endOnInputSize = 1 } endCondition_directive;
-typedef enum { decode_full_block = 0, partial_decode = 1 } earlyEnd_directive;
-
-#define LZ4_STATIC_ASSERT(c)	BUILD_BUG_ON(!(c))
-
-#endif

-- 
2.34.1

Re: [PATCH v2 13/15] lib/lz4: switch the decompressor to the vendored sources

[sashiko-bot]

> Replace the forked decompressor with thin entry points over the vendored
> lz4.c; the API is signature-compatible, so every export is a plain
> forwarder. lz4defs.h has no users left and goes.
> 
> lib/decompress_unlz4.c includes this file before boot, together with
> <linux/lz4.h>, so LZ4_streamDecode_t becomes incomplete as upstream
> declares it, and LZ4_MEM_DECOMPRESS (32) replaces the
> LZ4_STREAMDECODESIZE macros. LZ4_COMPRESSBOUND is defined by both
> headers and gets a guard; the LZ4_compressBound() wrapper goes, as
> lib/decompress_unlz4.c now uses the macro.
> 
> LZ4_decompress_safe_partial() on a truncated block can now return the
> bytes decoded so far rather than an error, depending on where the cut
> falls; the kernel-doc says so.
> 
> [ ... ]
>   upstream c5d6f8a8be39 ("fix #783"), v1.9.3
> commit 2d8867f3e083 ("lib: make LZ4_decompress_safe_forceExtDict() static")
>   the static forward declaration in lz4_deps.h
> 
> Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>

Sashiko has reviewed this patch and found no issues.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261106-lz4-vendor-upstream-v2-0-ee04c3f0d7dc@samsung.com?part=13

Re: [PATCH v2 13/15] lib/lz4: switch the decompressor to the vendored sources

[Sergey Senozhatsky]

On (26/11/06 15:42), Michal Wilczynski wrote:
[..]
> diff --git a/drivers/block/zram/backend_lz4.c b/drivers/block/zram/backend_lz4.c
> index 50265e3ce256490994cb6dae32a45da376e408e0..b9b8a5f678c155442fa3fd2c1984e45182e6fc38 100644
> --- a/drivers/block/zram/backend_lz4.c
> +++ b/drivers/block/zram/backend_lz4.c
> @@ -84,7 +84,7 @@ static int lz4_create(struct zcomp_params *params, struct zcomp_ctx *ctx)
>  		if (!zctx->mem)
>  			goto error;
>  	} else {
> -		zctx->dstrm = kzalloc_obj(*zctx->dstrm);
> +		zctx->dstrm = kzalloc(LZ4_MEM_DECOMPRESS, GFP_KERNEL);
>  		if (!zctx->dstrm)
>  			goto error;
>  
> diff --git a/drivers/block/zram/backend_lz4hc.c b/drivers/block/zram/backend_lz4hc.c
> index e7ee3740219852f079d1e0866da6b213a3f3a583..1e09c93016264259c149e2913b2819571b56fab6 100644
> --- a/drivers/block/zram/backend_lz4hc.c
> +++ b/drivers/block/zram/backend_lz4hc.c
> @@ -65,7 +65,7 @@ static int lz4hc_create(struct zcomp_params *params, struct zcomp_ctx *ctx)
>  		if (!zctx->mem)
>  			goto error;
>  	} else {
> -		zctx->dstrm = kzalloc_obj(*zctx->dstrm);
> +		zctx->dstrm = kzalloc(LZ4_MEM_DECOMPRESS, GFP_KERNEL);
>  		if (!zctx->dstrm)
>  			goto error;

Ditto, outdated zram code.

My ask would be to not squash zram changes with lz4 changes, when (iff)
possible.

Re: [f2fs-dev] [PATCH v2 13/15] lib/lz4: switch the decompressor to the vendored sources

[Sergey Senozhatsky]

On (26/11/06 15:42), Michal Wilczynski wrote:
[..]
> diff --git a/drivers/block/zram/backend_lz4.c b/drivers/block/zram/backend_lz4.c
> index 50265e3ce256490994cb6dae32a45da376e408e0..b9b8a5f678c155442fa3fd2c1984e45182e6fc38 100644
> --- a/drivers/block/zram/backend_lz4.c
> +++ b/drivers/block/zram/backend_lz4.c
> @@ -84,7 +84,7 @@ static int lz4_create(struct zcomp_params *params, struct zcomp_ctx *ctx)
>  		if (!zctx->mem)
>  			goto error;
>  	} else {
> -		zctx->dstrm = kzalloc_obj(*zctx->dstrm);
> +		zctx->dstrm = kzalloc(LZ4_MEM_DECOMPRESS, GFP_KERNEL);
>  		if (!zctx->dstrm)
>  			goto error;
>  
> diff --git a/drivers/block/zram/backend_lz4hc.c b/drivers/block/zram/backend_lz4hc.c
> index e7ee3740219852f079d1e0866da6b213a3f3a583..1e09c93016264259c149e2913b2819571b56fab6 100644
> --- a/drivers/block/zram/backend_lz4hc.c
> +++ b/drivers/block/zram/backend_lz4hc.c
> @@ -65,7 +65,7 @@ static int lz4hc_create(struct zcomp_params *params, struct zcomp_ctx *ctx)
>  		if (!zctx->mem)
>  			goto error;
>  	} else {
> -		zctx->dstrm = kzalloc_obj(*zctx->dstrm);
> +		zctx->dstrm = kzalloc(LZ4_MEM_DECOMPRESS, GFP_KERNEL);
>  		if (!zctx->dstrm)
>  			goto error;

Ditto, outdated zram code.

My ask would be to not squash zram changes with lz4 changes, when (iff)
possible.


_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

Re: [PATCH v2 13/15] lib/lz4: switch the decompressor to the vendored sources

[Michal Wilczynski]

On 11/10/26 06:39, Sergey Senozhatsky wrote:
> On (26/11/06 15:42), Michal Wilczynski wrote:
> [..]
>> diff --git a/drivers/block/zram/backend_lz4.c b/drivers/block/zram/backend_lz4.c
>> index 50265e3ce256490994cb6dae32a45da376e408e0..b9b8a5f678c155442fa3fd2c1984e45182e6fc38 100644
>> --- a/drivers/block/zram/backend_lz4.c
>> +++ b/drivers/block/zram/backend_lz4.c
>> @@ -84,7 +84,7 @@ static int lz4_create(struct zcomp_params *params, struct zcomp_ctx *ctx)
>>  		if (!zctx->mem)
>>  			goto error;
>>  	} else {
>> -		zctx->dstrm = kzalloc_obj(*zctx->dstrm);
>> +		zctx->dstrm = kzalloc(LZ4_MEM_DECOMPRESS, GFP_KERNEL);
>>  		if (!zctx->dstrm)
>>  			goto error;
>>  
>> diff --git a/drivers/block/zram/backend_lz4hc.c b/drivers/block/zram/backend_lz4hc.c
>> index e7ee3740219852f079d1e0866da6b213a3f3a583..1e09c93016264259c149e2913b2819571b56fab6 100644
>> --- a/drivers/block/zram/backend_lz4hc.c
>> +++ b/drivers/block/zram/backend_lz4hc.c
>> @@ -65,7 +65,7 @@ static int lz4hc_create(struct zcomp_params *params, struct zcomp_ctx *ctx)
>>  		if (!zctx->mem)
>>  			goto error;
>>  	} else {
>> -		zctx->dstrm = kzalloc_obj(*zctx->dstrm);
>> +		zctx->dstrm = kzalloc(LZ4_MEM_DECOMPRESS, GFP_KERNEL);
>>  		if (!zctx->dstrm)
>>  			goto error;
> 
> Ditto, outdated zram code.
> 
> My ask would be to not squash zram changes with lz4 changes, when (iff)
> possible.
> 

Will do.
The zram changes will go into a separate zram patch ahead of the lib/lz4
ones.

Best regards,
-- 
Michal Wilczynski <m.wilczynski@samsung.com>

Re: [f2fs-dev] [PATCH v2 13/15] lib/lz4: switch the decompressor to the vendored sources

[Michal Wilczynski]

On 11/10/26 06:39, Sergey Senozhatsky wrote:
> On (26/11/06 15:42), Michal Wilczynski wrote:
> [..]
>> diff --git a/drivers/block/zram/backend_lz4.c b/drivers/block/zram/backend_lz4.c
>> index 50265e3ce256490994cb6dae32a45da376e408e0..b9b8a5f678c155442fa3fd2c1984e45182e6fc38 100644
>> --- a/drivers/block/zram/backend_lz4.c
>> +++ b/drivers/block/zram/backend_lz4.c
>> @@ -84,7 +84,7 @@ static int lz4_create(struct zcomp_params *params, struct zcomp_ctx *ctx)
>>  		if (!zctx->mem)
>>  			goto error;
>>  	} else {
>> -		zctx->dstrm = kzalloc_obj(*zctx->dstrm);
>> +		zctx->dstrm = kzalloc(LZ4_MEM_DECOMPRESS, GFP_KERNEL);
>>  		if (!zctx->dstrm)
>>  			goto error;
>>  
>> diff --git a/drivers/block/zram/backend_lz4hc.c b/drivers/block/zram/backend_lz4hc.c
>> index e7ee3740219852f079d1e0866da6b213a3f3a583..1e09c93016264259c149e2913b2819571b56fab6 100644
>> --- a/drivers/block/zram/backend_lz4hc.c
>> +++ b/drivers/block/zram/backend_lz4hc.c
>> @@ -65,7 +65,7 @@ static int lz4hc_create(struct zcomp_params *params, struct zcomp_ctx *ctx)
>>  		if (!zctx->mem)
>>  			goto error;
>>  	} else {
>> -		zctx->dstrm = kzalloc_obj(*zctx->dstrm);
>> +		zctx->dstrm = kzalloc(LZ4_MEM_DECOMPRESS, GFP_KERNEL);
>>  		if (!zctx->dstrm)
>>  			goto error;
> 
> Ditto, outdated zram code.
> 
> My ask would be to not squash zram changes with lz4 changes, when (iff)
> possible.
> 

Will do.
The zram changes will go into a separate zram patch ahead of the lib/lz4
ones.

Best regards,
-- 
Michal Wilczynski <m.wilczynski@samsung.com>


_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

[PATCH v2 14/15] lib/lz4: replace lz4_kernel_api.h's prototypes with <linux/lz4.h>

[Michal Wilczynski]

lz4_kernel_api.h carried its own copy of the twenty public prototypes
because it could not include <linux/lz4.h>, which defined the stream
types that upstream's headers also define.

The types are incomplete now, so it can. Two prototype lists for one
ABI is a drift hazard: the entry points were built against one and their
callers against the other, so no compiler ever saw both, and MODVERSIONS
would not catch a mismatch either, since it hashes the exporter's view
only. Now there is one list.

It also puts the public constants in scope of the entry points, so the
size assertions can name them, and the HC level clamp can use
LZ4HC_CLAMP_CLEVEL, static_asserted equal to upstream's
LZ4HC_CLEVEL_OPT_MIN.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 lib/lz4/lz4_compress.c   |  2 +-
 lib/lz4/lz4_decompress.c |  2 +-
 lib/lz4/lz4_kernel_api.h | 64 +++---------------------------------------------
 lib/lz4/lz4hc_compress.c |  8 +++---
 4 files changed, 11 insertions(+), 65 deletions(-)

diff --git a/lib/lz4/lz4_compress.c b/lib/lz4/lz4_compress.c
index 941e32a19ac972aef543c9a03b447a8a98bf5215..3e272584ea0c905b48b6bf0b7e77acfbad131bd1 100644
--- a/lib/lz4/lz4_compress.c
+++ b/lib/lz4/lz4_compress.c
@@ -30,7 +30,7 @@
 
 /* Catch any divergence from upstream's layout at build time. */
 static_assert(sizeof(LZ4_stream_t) == LZ4_STREAM_MINSIZE);
-static_assert(LZ4_STREAM_MINSIZE == 16416); /* LZ4_MEM_COMPRESS */
+static_assert(LZ4_STREAM_MINSIZE == LZ4_MEM_COMPRESS);
 
 /* LZ4_initStream() returns NULL for a misaligned state, which
  * LZ4_compress_fast_extState() dereferences and
diff --git a/lib/lz4/lz4_decompress.c b/lib/lz4/lz4_decompress.c
index 60b57862a2e387aa89fa26f6f8e2b5485b5d9995..5d68a6f59a62ddb3701dd3531cf606c05ef4e607 100644
--- a/lib/lz4/lz4_decompress.c
+++ b/lib/lz4/lz4_decompress.c
@@ -31,7 +31,7 @@
 #endif
 
 static_assert(sizeof(LZ4_streamDecode_t) == LZ4_STREAMDECODE_MINSIZE);
-static_assert(LZ4_STREAMDECODE_MINSIZE == 32); /* LZ4_MEM_DECOMPRESS */
+static_assert(LZ4_STREAMDECODE_MINSIZE == LZ4_MEM_DECOMPRESS);
 
 int LZ4_decompress_safe(const char *source, char *dest, int compressedSize,
 			int maxDecompressedSize)
diff --git a/lib/lz4/lz4_kernel_api.h b/lib/lz4/lz4_kernel_api.h
index de75e5d976eb9dd318da5318a26b91d7aed81042..43605a0e4d7f543662af83251184968f0d1b70dd 100644
--- a/lib/lz4/lz4_kernel_api.h
+++ b/lib/lz4/lz4_kernel_api.h
@@ -3,19 +3,10 @@
  * Copyright (c) 2026 Samsung Electronics Co., Ltd.
  * Author: Michal Wilczynski <m.wilczynski@samsung.com>
  *
- * lz4_kernel_api.h -- the LZ4 API this directory exports
+ * lz4_kernel_api.h -- hand the LZ4 names back to the kernel
  *
- * lz4_deps.h renames upstream's entry points to __lz4_* so that the kernel's
- * own definitions can use the real names; this header undoes that renaming
- * and the optimize() override, and declares what lib/lz4 exports.
- *
- * The same functions are declared to callers in <linux/lz4.h>, in terms of
- * opaque stream types.  We cannot include that header here -- it and
- * upstream's lz4.h describe the same library and collide -- so the
- * declarations are repeated, expressed in upstream's own types.  Keep the two
- * in step; <linux/lz4.h> is the one callers compile against.
- *
- * Include after upstream/lz4.c or upstream/lz4hc.c.
+ * Undo lz4_deps.h's renaming and its optimize() override, and pull in
+ * <linux/lz4.h>.  Include after upstream/lz4.c or upstream/lz4hc.c.
  */
 
 #undef LZ4_compress_fast
@@ -40,51 +31,4 @@
 
 #undef optimize
 
-/*
- * The objects that do not build lz4hc.c never see this type, and it is only
- * ever used through a pointer here.  Upstream declares it the same way.
- */
-typedef union LZ4_streamHC_u LZ4_streamHC_t;
-
-/* Compression.  wrkmem is LZ4_MEM_COMPRESS bytes, supplied by the caller. */
-int LZ4_compress_default(const char *source, char *dest, int inputSize,
-			 int maxOutputSize, void *wrkmem);
-int LZ4_compress_fast(const char *source, char *dest, int inputSize,
-		      int maxOutputSize, int acceleration, void *wrkmem);
-int LZ4_compress_destSize(const char *source, char *dest, int *sourceSizePtr,
-			  int targetDestSize, void *wrkmem);
-
-/* Streaming compression. */
-void LZ4_resetStream(LZ4_stream_t *LZ4_stream);
-int LZ4_loadDict(LZ4_stream_t *streamPtr, const char *dictionary,
-		 int dictSize);
-int LZ4_saveDict(LZ4_stream_t *streamPtr, char *safeBuffer, int dictSize);
-int LZ4_compress_fast_continue(LZ4_stream_t *streamPtr, const char *src,
-			       char *dst, int srcSize, int maxDstSize,
-			       int acceleration);
-
-/* Decompression. */
-int LZ4_decompress_safe(const char *source, char *dest, int compressedSize,
-			int maxDecompressedSize);
-int LZ4_decompress_safe_partial(const char *source, char *dest,
-				int compressedSize, int targetOutputSize,
-				int maxDecompressedSize);
-int LZ4_setStreamDecode(LZ4_streamDecode_t *LZ4_streamDecode,
-			const char *dictionary, int dictSize);
-int LZ4_decompress_safe_continue(LZ4_streamDecode_t *LZ4_streamDecode,
-				 const char *source, char *dest,
-				 int compressedSize, int maxDecompressedSize);
-int LZ4_decompress_safe_usingDict(const char *source, char *dest,
-				  int compressedSize, int maxDecompressedSize,
-				  const char *dictStart, int dictSize);
-
-/* HC compression.  wrkmem is LZ4HC_MEM_COMPRESS bytes. */
-int LZ4_compress_HC(const char *src, char *dst, int srcSize, int dstCapacity,
-		    int compressionLevel, void *wrkmem);
-void LZ4_resetStreamHC(LZ4_streamHC_t *streamHCPtr, int compressionLevel);
-int LZ4_loadDictHC(LZ4_streamHC_t *streamHCPtr, const char *dictionary,
-		   int dictSize);
-int LZ4_compress_HC_continue(LZ4_streamHC_t *streamHCPtr, const char *src,
-			     char *dst, int srcSize, int maxDstSize);
-int LZ4_saveDictHC(LZ4_streamHC_t *streamHCPtr, char *safeBuffer,
-		   int maxDictSize);
+#include <linux/lz4.h>
diff --git a/lib/lz4/lz4hc_compress.c b/lib/lz4/lz4hc_compress.c
index 4d56d45028110805628be766b2758a8dffd56f0c..4f40a12df17cfed2bab7f5d87ac087b4d423cef7 100644
--- a/lib/lz4/lz4hc_compress.c
+++ b/lib/lz4/lz4hc_compress.c
@@ -39,7 +39,7 @@ static int LZ4_compressBound(int isize)
 
 /* Catch any divergence from upstream's layout at build time. */
 static_assert(sizeof(LZ4_streamHC_t) == LZ4_STREAMHC_MINSIZE);
-static_assert(LZ4_STREAMHC_MINSIZE == 262200); /* LZ4HC_MEM_COMPRESS */
+static_assert(LZ4_STREAMHC_MINSIZE == LZ4HC_MEM_COMPRESS);
 
 /* Levels >= LZ4HC_CLEVEL_OPT_MIN (10) reach the optimal parser, which needs
  * a ~64K opt[]: too big for a kernel stack, and in the heap mode lz4_deps.h
@@ -48,10 +48,12 @@ static_assert(LZ4_STREAMHC_MINSIZE == 262200); /* LZ4HC_MEM_COMPRESS */
  * working.  Levels 3..9 keep their meaning; 1 and 2 are upstream's lz4mid,
  * faster and weaker than the shallow hash chain the fork used for them.
  */
+static_assert(LZ4HC_CLAMP_CLEVEL == LZ4HC_CLEVEL_OPT_MIN);
+
 static int lz4hc_clamp_level(int compressionLevel)
 {
-	if (compressionLevel >= LZ4HC_CLEVEL_OPT_MIN)
-		return LZ4HC_CLEVEL_OPT_MIN - 1;
+	if (compressionLevel >= LZ4HC_CLAMP_CLEVEL)
+		return LZ4HC_CLAMP_CLEVEL - 1;
 
 	return compressionLevel;
 }

-- 
2.34.1

[f2fs-dev] [PATCH v2 14/15] lib/lz4: replace lz4_kernel_api.h's prototypes with <linux/lz4.h>

[Michal Wilczynski]

lz4_kernel_api.h carried its own copy of the twenty public prototypes
because it could not include <linux/lz4.h>, which defined the stream
types that upstream's headers also define.

The types are incomplete now, so it can. Two prototype lists for one
ABI is a drift hazard: the entry points were built against one and their
callers against the other, so no compiler ever saw both, and MODVERSIONS
would not catch a mismatch either, since it hashes the exporter's view
only. Now there is one list.

It also puts the public constants in scope of the entry points, so the
size assertions can name them, and the HC level clamp can use
LZ4HC_CLAMP_CLEVEL, static_asserted equal to upstream's
LZ4HC_CLEVEL_OPT_MIN.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 lib/lz4/lz4_compress.c   |  2 +-
 lib/lz4/lz4_decompress.c |  2 +-
 lib/lz4/lz4_kernel_api.h | 64 +++---------------------------------------------
 lib/lz4/lz4hc_compress.c |  8 +++---
 4 files changed, 11 insertions(+), 65 deletions(-)

diff --git a/lib/lz4/lz4_compress.c b/lib/lz4/lz4_compress.c
index 941e32a19ac972aef543c9a03b447a8a98bf5215..3e272584ea0c905b48b6bf0b7e77acfbad131bd1 100644
--- a/lib/lz4/lz4_compress.c
+++ b/lib/lz4/lz4_compress.c
@@ -30,7 +30,7 @@
 
 /* Catch any divergence from upstream's layout at build time. */
 static_assert(sizeof(LZ4_stream_t) == LZ4_STREAM_MINSIZE);
-static_assert(LZ4_STREAM_MINSIZE == 16416); /* LZ4_MEM_COMPRESS */
+static_assert(LZ4_STREAM_MINSIZE == LZ4_MEM_COMPRESS);
 
 /* LZ4_initStream() returns NULL for a misaligned state, which
  * LZ4_compress_fast_extState() dereferences and
diff --git a/lib/lz4/lz4_decompress.c b/lib/lz4/lz4_decompress.c
index 60b57862a2e387aa89fa26f6f8e2b5485b5d9995..5d68a6f59a62ddb3701dd3531cf606c05ef4e607 100644
--- a/lib/lz4/lz4_decompress.c
+++ b/lib/lz4/lz4_decompress.c
@@ -31,7 +31,7 @@
 #endif
 
 static_assert(sizeof(LZ4_streamDecode_t) == LZ4_STREAMDECODE_MINSIZE);
-static_assert(LZ4_STREAMDECODE_MINSIZE == 32); /* LZ4_MEM_DECOMPRESS */
+static_assert(LZ4_STREAMDECODE_MINSIZE == LZ4_MEM_DECOMPRESS);
 
 int LZ4_decompress_safe(const char *source, char *dest, int compressedSize,
 			int maxDecompressedSize)
diff --git a/lib/lz4/lz4_kernel_api.h b/lib/lz4/lz4_kernel_api.h
index de75e5d976eb9dd318da5318a26b91d7aed81042..43605a0e4d7f543662af83251184968f0d1b70dd 100644
--- a/lib/lz4/lz4_kernel_api.h
+++ b/lib/lz4/lz4_kernel_api.h
@@ -3,19 +3,10 @@
  * Copyright (c) 2026 Samsung Electronics Co., Ltd.
  * Author: Michal Wilczynski <m.wilczynski@samsung.com>
  *
- * lz4_kernel_api.h -- the LZ4 API this directory exports
+ * lz4_kernel_api.h -- hand the LZ4 names back to the kernel
  *
- * lz4_deps.h renames upstream's entry points to __lz4_* so that the kernel's
- * own definitions can use the real names; this header undoes that renaming
- * and the optimize() override, and declares what lib/lz4 exports.
- *
- * The same functions are declared to callers in <linux/lz4.h>, in terms of
- * opaque stream types.  We cannot include that header here -- it and
- * upstream's lz4.h describe the same library and collide -- so the
- * declarations are repeated, expressed in upstream's own types.  Keep the two
- * in step; <linux/lz4.h> is the one callers compile against.
- *
- * Include after upstream/lz4.c or upstream/lz4hc.c.
+ * Undo lz4_deps.h's renaming and its optimize() override, and pull in
+ * <linux/lz4.h>.  Include after upstream/lz4.c or upstream/lz4hc.c.
  */
 
 #undef LZ4_compress_fast
@@ -40,51 +31,4 @@
 
 #undef optimize
 
-/*
- * The objects that do not build lz4hc.c never see this type, and it is only
- * ever used through a pointer here.  Upstream declares it the same way.
- */
-typedef union LZ4_streamHC_u LZ4_streamHC_t;
-
-/* Compression.  wrkmem is LZ4_MEM_COMPRESS bytes, supplied by the caller. */
-int LZ4_compress_default(const char *source, char *dest, int inputSize,
-			 int maxOutputSize, void *wrkmem);
-int LZ4_compress_fast(const char *source, char *dest, int inputSize,
-		      int maxOutputSize, int acceleration, void *wrkmem);
-int LZ4_compress_destSize(const char *source, char *dest, int *sourceSizePtr,
-			  int targetDestSize, void *wrkmem);
-
-/* Streaming compression. */
-void LZ4_resetStream(LZ4_stream_t *LZ4_stream);
-int LZ4_loadDict(LZ4_stream_t *streamPtr, const char *dictionary,
-		 int dictSize);
-int LZ4_saveDict(LZ4_stream_t *streamPtr, char *safeBuffer, int dictSize);
-int LZ4_compress_fast_continue(LZ4_stream_t *streamPtr, const char *src,
-			       char *dst, int srcSize, int maxDstSize,
-			       int acceleration);
-
-/* Decompression. */
-int LZ4_decompress_safe(const char *source, char *dest, int compressedSize,
-			int maxDecompressedSize);
-int LZ4_decompress_safe_partial(const char *source, char *dest,
-				int compressedSize, int targetOutputSize,
-				int maxDecompressedSize);
-int LZ4_setStreamDecode(LZ4_streamDecode_t *LZ4_streamDecode,
-			const char *dictionary, int dictSize);
-int LZ4_decompress_safe_continue(LZ4_streamDecode_t *LZ4_streamDecode,
-				 const char *source, char *dest,
-				 int compressedSize, int maxDecompressedSize);
-int LZ4_decompress_safe_usingDict(const char *source, char *dest,
-				  int compressedSize, int maxDecompressedSize,
-				  const char *dictStart, int dictSize);
-
-/* HC compression.  wrkmem is LZ4HC_MEM_COMPRESS bytes. */
-int LZ4_compress_HC(const char *src, char *dst, int srcSize, int dstCapacity,
-		    int compressionLevel, void *wrkmem);
-void LZ4_resetStreamHC(LZ4_streamHC_t *streamHCPtr, int compressionLevel);
-int LZ4_loadDictHC(LZ4_streamHC_t *streamHCPtr, const char *dictionary,
-		   int dictSize);
-int LZ4_compress_HC_continue(LZ4_streamHC_t *streamHCPtr, const char *src,
-			     char *dst, int srcSize, int maxDstSize);
-int LZ4_saveDictHC(LZ4_streamHC_t *streamHCPtr, char *safeBuffer,
-		   int maxDictSize);
+#include <linux/lz4.h>
diff --git a/lib/lz4/lz4hc_compress.c b/lib/lz4/lz4hc_compress.c
index 4d56d45028110805628be766b2758a8dffd56f0c..4f40a12df17cfed2bab7f5d87ac087b4d423cef7 100644
--- a/lib/lz4/lz4hc_compress.c
+++ b/lib/lz4/lz4hc_compress.c
@@ -39,7 +39,7 @@ static int LZ4_compressBound(int isize)
 
 /* Catch any divergence from upstream's layout at build time. */
 static_assert(sizeof(LZ4_streamHC_t) == LZ4_STREAMHC_MINSIZE);
-static_assert(LZ4_STREAMHC_MINSIZE == 262200); /* LZ4HC_MEM_COMPRESS */
+static_assert(LZ4_STREAMHC_MINSIZE == LZ4HC_MEM_COMPRESS);
 
 /* Levels >= LZ4HC_CLEVEL_OPT_MIN (10) reach the optimal parser, which needs
  * a ~64K opt[]: too big for a kernel stack, and in the heap mode lz4_deps.h
@@ -48,10 +48,12 @@ static_assert(LZ4_STREAMHC_MINSIZE == 262200); /* LZ4HC_MEM_COMPRESS */
  * working.  Levels 3..9 keep their meaning; 1 and 2 are upstream's lz4mid,
  * faster and weaker than the shallow hash chain the fork used for them.
  */
+static_assert(LZ4HC_CLAMP_CLEVEL == LZ4HC_CLEVEL_OPT_MIN);
+
 static int lz4hc_clamp_level(int compressionLevel)
 {
-	if (compressionLevel >= LZ4HC_CLEVEL_OPT_MIN)
-		return LZ4HC_CLEVEL_OPT_MIN - 1;
+	if (compressionLevel >= LZ4HC_CLAMP_CLEVEL)
+		return LZ4HC_CLAMP_CLEVEL - 1;
 
 	return compressionLevel;
 }

-- 
2.34.1



_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

Re: [PATCH v2 14/15] lib/lz4: replace lz4_kernel_api.h's prototypes with <linux/lz4.h>

[sashiko-bot]

> lz4_kernel_api.h carried its own copy of the twenty public prototypes
> because it could not include <linux/lz4.h>, which defined the stream
> types that upstream's headers also define.
> 
> The types are incomplete now, so it can. Two prototype lists for one
> ABI is a drift hazard: the entry points were built against one and their
> callers against the other, so no compiler ever saw both, and MODVERSIONS
> would not catch a mismatch either, since it hashes the exporter's view
> only. Now there is one list.
> 
> It also puts the public constants in scope of the entry points, so the
> size assertions can name them, and the HC level clamp can use
> LZ4HC_CLAMP_CLEVEL, static_asserted equal to upstream's
> LZ4HC_CLEVEL_OPT_MIN.
> 
> Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>

Sashiko has reviewed this patch and found no issues.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261106-lz4-vendor-upstream-v2-0-ee04c3f0d7dc@samsung.com?part=14

[PATCH v2 15/15] MAINTAINERS: add an entry for the LZ4 compression library

[Michal Wilczynski]

lib/lz4/ has never had one: get_maintainer.pl returns only the LKML
fallback for every file in it, including include/linux/lz4.h, which eight
subsystems build against.

The entry also takes lib/decompress_unlz4.c and its header, which
decode LZ4 kernel images and initramfs through lib/lz4, the way XZ
EMBEDDED covers its own equivalents, and crypto/lz4.c and crypto/lz4hc.c,
the crypto API wrappers zswap uses, as ZSTD takes crypto/zstd.c.

This matters more now than it did. With the previous patches lib/lz4 is
vendored upstream sources, and someone has to bump them to new upstream
releases and decide what the kernel-side glue may do. I am willing to do
that, so add myself as its maintainer.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 MAINTAINERS | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/MAINTAINERS b/MAINTAINERS
index cc3cae2e378b34aeb705cef654c720ca9f2ba223..598d87ca772eecbb1ed169d51ff0eb8cdd2e50ad 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -15606,6 +15606,16 @@ S:	Supported
 F:	drivers/net/pcs/pcs-lynx.c
 F:	include/linux/pcs-lynx.h
 
+LZ4 COMPRESSION LIBRARY
+M:	Michal Wilczynski <m.wilczynski@samsung.com>
+S:	Maintained
+F:	crypto/lz4.c
+F:	crypto/lz4hc.c
+F:	include/linux/decompress/unlz4.h
+F:	include/linux/lz4.h
+F:	lib/decompress_unlz4.c
+F:	lib/lz4/
+
 M68K ARCHITECTURE
 M:	Geert Uytterhoeven <geert@linux-m68k.org>
 L:	linux-m68k@lists.linux-m68k.org

-- 
2.34.1

[f2fs-dev] [PATCH v2 15/15] MAINTAINERS: add an entry for the LZ4 compression library

[Michal Wilczynski]

lib/lz4/ has never had one: get_maintainer.pl returns only the LKML
fallback for every file in it, including include/linux/lz4.h, which eight
subsystems build against.

The entry also takes lib/decompress_unlz4.c and its header, which
decode LZ4 kernel images and initramfs through lib/lz4, the way XZ
EMBEDDED covers its own equivalents, and crypto/lz4.c and crypto/lz4hc.c,
the crypto API wrappers zswap uses, as ZSTD takes crypto/zstd.c.

This matters more now than it did. With the previous patches lib/lz4 is
vendored upstream sources, and someone has to bump them to new upstream
releases and decide what the kernel-side glue may do. I am willing to do
that, so add myself as its maintainer.

Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
---
 MAINTAINERS | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/MAINTAINERS b/MAINTAINERS
index cc3cae2e378b34aeb705cef654c720ca9f2ba223..598d87ca772eecbb1ed169d51ff0eb8cdd2e50ad 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -15606,6 +15606,16 @@ S:	Supported
 F:	drivers/net/pcs/pcs-lynx.c
 F:	include/linux/pcs-lynx.h
 
+LZ4 COMPRESSION LIBRARY
+M:	Michal Wilczynski <m.wilczynski@samsung.com>
+S:	Maintained
+F:	crypto/lz4.c
+F:	crypto/lz4hc.c
+F:	include/linux/decompress/unlz4.h
+F:	include/linux/lz4.h
+F:	lib/decompress_unlz4.c
+F:	lib/lz4/
+
 M68K ARCHITECTURE
 M:	Geert Uytterhoeven <geert@linux-m68k.org>
 L:	linux-m68k@lists.linux-m68k.org

-- 
2.34.1



_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

Re: [PATCH v2 00/15] lib/lz4: stop forking upstream LZ4, vendor it instead

[Michal Wilczynski]

Hi everyone,

On 11/6/26 15:42, Michal Wilczynski wrote:
> 
> The import is the v1.10.0 release tag.

It appears v1.11.0, followed by v1.11.1 were released in the
meantime. The next revision of this patchset will import from the
v1.11.1 release tag to take advantage of the newest code. It already
contains the backported commits so I will drop the backports.

I will redo all the measurements with the newr version for the v3.

> 

Best regards,
-- 
Michal Wilczynski <m.wilczynski@samsung.com>

Re: [f2fs-dev] [PATCH v2 00/15] lib/lz4: stop forking upstream LZ4, vendor it instead

[Michal Wilczynski]

Hi everyone,

On 11/6/26 15:42, Michal Wilczynski wrote:
> 
> The import is the v1.10.0 release tag.

It appears v1.11.0, followed by v1.11.1 were released in the
meantime. The next revision of this patchset will import from the
v1.11.1 release tag to take advantage of the newest code. It already
contains the backported commits so I will drop the backports.

I will redo all the measurements with the newr version for the v3.

> 

Best regards,
-- 
Michal Wilczynski <m.wilczynski@samsung.com>


_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel