qemu-devel.nongnu.org archive mirror
 help / color / mirror / Atom feed
From: Richard Henderson <richard.henderson@linaro.org>
To: qemu-devel@nongnu.org
Cc: "Philippe Mathieu-Daudé" <f4bug@amsat.org>
Subject: [PATCH 01/43] tcg: Adjust simd_desc size encoding
Date: Tue,  8 Sep 2020 17:16:05 -0700	[thread overview]
Message-ID: <20200909001647.532249-2-richard.henderson@linaro.org> (raw)
In-Reply-To: <20200909001647.532249-1-richard.henderson@linaro.org>

With larger vector sizes, it turns out oprsz == maxsz, and we only
need to represent mismatch for oprsz <= 32.  We do, however, need
to represent larger oprsz and do so without reducing SIMD_DATA_BITS.

Reduce the size of the oprsz field and increase the maxsz field.
Steal the oprsz value of 24 to indicate equality with maxsz.

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 include/tcg/tcg-gvec-desc.h | 38 ++++++++++++++++++++++++-------------
 tcg/tcg-op-gvec.c           | 35 ++++++++++++++++++++++++++--------
 2 files changed, 52 insertions(+), 21 deletions(-)

diff --git a/include/tcg/tcg-gvec-desc.h b/include/tcg/tcg-gvec-desc.h
index 0224ac3e78..704bd86454 100644
--- a/include/tcg/tcg-gvec-desc.h
+++ b/include/tcg/tcg-gvec-desc.h
@@ -20,29 +20,41 @@
 #ifndef TCG_TCG_GVEC_DESC_H
 #define TCG_TCG_GVEC_DESC_H
 
-/* ??? These bit widths are set for ARM SVE, maxing out at 256 byte vectors. */
-#define SIMD_OPRSZ_SHIFT   0
-#define SIMD_OPRSZ_BITS    5
+/*
+ * This configuration allows MAXSZ to represent 2048 bytes, and
+ * OPRSZ to match MAXSZ, or represent the smaller values 8, 16, or 32.
+ *
+ * Encode this with:
+ *   0, 1, 3 -> 8, 16, 32
+ *   2       -> maxsz
+ *
+ * This steals the input that would otherwise map to 24 to match maxsz.
+ */
+#define SIMD_MAXSZ_SHIFT   0
+#define SIMD_MAXSZ_BITS    8
 
-#define SIMD_MAXSZ_SHIFT   (SIMD_OPRSZ_SHIFT + SIMD_OPRSZ_BITS)
-#define SIMD_MAXSZ_BITS    5
+#define SIMD_OPRSZ_SHIFT   (SIMD_MAXSZ_SHIFT + SIMD_MAXSZ_BITS)
+#define SIMD_OPRSZ_BITS    2
 
-#define SIMD_DATA_SHIFT    (SIMD_MAXSZ_SHIFT + SIMD_MAXSZ_BITS)
+#define SIMD_DATA_SHIFT    (SIMD_OPRSZ_SHIFT + SIMD_OPRSZ_BITS)
 #define SIMD_DATA_BITS     (32 - SIMD_DATA_SHIFT)
 
 /* Create a descriptor from components.  */
 uint32_t simd_desc(uint32_t oprsz, uint32_t maxsz, int32_t data);
 
-/* Extract the operation size from a descriptor.  */
-static inline intptr_t simd_oprsz(uint32_t desc)
-{
-    return (extract32(desc, SIMD_OPRSZ_SHIFT, SIMD_OPRSZ_BITS) + 1) * 8;
-}
-
 /* Extract the max vector size from a descriptor.  */
 static inline intptr_t simd_maxsz(uint32_t desc)
 {
-    return (extract32(desc, SIMD_MAXSZ_SHIFT, SIMD_MAXSZ_BITS) + 1) * 8;
+    return extract32(desc, SIMD_MAXSZ_SHIFT, SIMD_MAXSZ_BITS) * 8 + 8;
+}
+
+/* Extract the operation size from a descriptor.  */
+static inline intptr_t simd_oprsz(uint32_t desc)
+{
+    uint32_t f = extract32(desc, SIMD_OPRSZ_SHIFT, SIMD_OPRSZ_BITS);
+    intptr_t o = f * 8 + 8;
+    intptr_t m = simd_maxsz(desc);
+    return f == 2 ? m : o;
 }
 
 /* Extract the operation-specific data from a descriptor.  */
diff --git a/tcg/tcg-op-gvec.c b/tcg/tcg-op-gvec.c
index 7ebd9e8298..ddbe06b71a 100644
--- a/tcg/tcg-op-gvec.c
+++ b/tcg/tcg-op-gvec.c
@@ -37,11 +37,21 @@ static const TCGOpcode vecop_list_empty[1] = { 0 };
    of the operand offsets so that we can check them all at once.  */
 static void check_size_align(uint32_t oprsz, uint32_t maxsz, uint32_t ofs)
 {
-    uint32_t opr_align = oprsz >= 16 ? 15 : 7;
-    uint32_t max_align = maxsz >= 16 || oprsz >= 16 ? 15 : 7;
-    tcg_debug_assert(oprsz > 0);
-    tcg_debug_assert(oprsz <= maxsz);
-    tcg_debug_assert((oprsz & opr_align) == 0);
+    uint32_t max_align;
+
+    switch (oprsz) {
+    case 8:
+    case 16:
+    case 32:
+        tcg_debug_assert(oprsz <= maxsz);
+        break;
+    default:
+        tcg_debug_assert(oprsz == maxsz);
+        break;
+    }
+    tcg_debug_assert(maxsz <= (8 << SIMD_MAXSZ_BITS));
+
+    max_align = maxsz >= 16 ? 15 : 7;
     tcg_debug_assert((maxsz & max_align) == 0);
     tcg_debug_assert((ofs & max_align) == 0);
 }
@@ -77,12 +87,21 @@ uint32_t simd_desc(uint32_t oprsz, uint32_t maxsz, int32_t data)
 {
     uint32_t desc = 0;
 
-    assert(oprsz % 8 == 0 && oprsz <= (8 << SIMD_OPRSZ_BITS));
-    assert(maxsz % 8 == 0 && maxsz <= (8 << SIMD_MAXSZ_BITS));
-    assert(data == sextract32(data, 0, SIMD_DATA_BITS));
+    check_size_align(oprsz, maxsz, 0);
+    tcg_debug_assert(data == sextract32(data, 0, SIMD_DATA_BITS));
 
     oprsz = (oprsz / 8) - 1;
     maxsz = (maxsz / 8) - 1;
+
+    /*
+     * We have just asserted in check_size_align that either
+     * oprsz is {8,16,32} or matches maxsz.  Encode the final
+     * case with '2', as that would otherwise map to 24.
+     */
+    if (oprsz == maxsz) {
+        oprsz = 2;
+    }
+
     desc = deposit32(desc, SIMD_OPRSZ_SHIFT, SIMD_OPRSZ_BITS, oprsz);
     desc = deposit32(desc, SIMD_MAXSZ_SHIFT, SIMD_MAXSZ_BITS, maxsz);
     desc = deposit32(desc, SIMD_DATA_SHIFT, SIMD_DATA_BITS, data);
-- 
2.25.1



  reply	other threads:[~2020-09-09  0:17 UTC|newest]

Thread overview: 48+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-09-09  0:16 [PATCH 00/43] tcg patch queue Richard Henderson
2020-09-09  0:16 ` Richard Henderson [this message]
2020-09-09  0:16 ` [PATCH 02/43] tcg: Drop union from TCGArgConstraint Richard Henderson
2020-09-09 17:43   ` Philippe Mathieu-Daudé
2020-09-09  0:16 ` [PATCH 03/43] tcg: Move sorted_args into TCGArgConstraint.sort_index Richard Henderson
2020-09-09  0:16 ` [PATCH 04/43] tcg: Remove TCG_CT_REG Richard Henderson
2020-09-09  0:16 ` [PATCH 05/43] tcg: Move some TCG_CT_* bits to TCGArgConstraint bitfields Richard Henderson
2020-09-09  0:16 ` [PATCH 06/43] tcg: Remove TCGOpDef.used Richard Henderson
2020-09-09 17:45   ` Philippe Mathieu-Daudé
2020-09-09  0:16 ` [PATCH 07/43] tcg/i386: Fix dupi for avx2 32-bit hosts Richard Henderson
2020-09-09  0:16 ` [PATCH 08/43] tcg: Fix generation of dupi_vec for 32-bit host Richard Henderson
2020-09-09  0:16 ` [PATCH 09/43] tcg/optimize: Fold dup2_vec Richard Henderson
2020-09-09  0:16 ` [PATCH 10/43] tcg: Remove TCG_TARGET_HAS_cmp_vec Richard Henderson
2020-09-09 17:47   ` Philippe Mathieu-Daudé
2020-09-09  0:16 ` [PATCH 11/43] tcg: Use tcg_out_dupi_vec from temp_load Richard Henderson
2020-09-09  0:16 ` [PATCH 12/43] tcg: Increase tcg_out_dupi_vec immediate to int64_t Richard Henderson
2020-09-09  0:16 ` [PATCH 13/43] tcg: Consolidate 3 bits into enum TCGTempKind Richard Henderson
2020-09-09 17:52   ` Philippe Mathieu-Daudé
2020-09-09  0:16 ` [PATCH 14/43] tcg: Add temp_readonly Richard Henderson
2020-09-09  0:16 ` [PATCH 15/43] tcg: Expand TCGTemp.val to 64-bits Richard Henderson
2020-09-09  0:16 ` [PATCH 16/43] tcg: Rename struct tcg_temp_info to TempOptInfo Richard Henderson
2020-09-09  0:16 ` [PATCH 17/43] tcg: Expand TempOptInfo to 64-bits Richard Henderson
2020-09-09  0:16 ` [PATCH 18/43] tcg: Introduce TYPE_CONST temporaries Richard Henderson
2020-09-09  0:16 ` [PATCH 19/43] tcg/optimize: Improve find_better_copy Richard Henderson
2020-09-09  0:16 ` [PATCH 20/43] tcg/optimize: Adjust TempOptInfo allocation Richard Henderson
2020-09-09  0:16 ` [PATCH 21/43] tcg/optimize: Use tcg_constant_internal with constant folding Richard Henderson
2020-09-09  0:16 ` [PATCH 22/43] tcg: Convert tcg_gen_dupi_vec to TCG_CONST Richard Henderson
2020-09-09  0:16 ` [PATCH 23/43] tcg: Use tcg_constant_i32 with icount expander Richard Henderson
2020-09-09  0:16 ` [PATCH 24/43] tcg: Use tcg_constant_{i32,i64} with tcg int expanders Richard Henderson
2020-09-09  0:16 ` [PATCH 25/43] tcg: Use tcg_constant_{i32,i64} with tcg plugins Richard Henderson
2020-09-09  0:16 ` [PATCH 26/43] tcg: Use tcg_constant_{i32, i64, vec} with gvec expanders Richard Henderson
2020-09-09  0:16 ` [PATCH 27/43] tcg/tci: Add special tci_movi_{i32,i64} opcodes Richard Henderson
2020-09-09  0:16 ` [PATCH 28/43] tcg: Remove movi and dupi opcodes Richard Henderson
2020-09-09  0:16 ` [PATCH 29/43] tcg: Add tcg_reg_alloc_dup2 Richard Henderson
2020-09-09  0:16 ` [PATCH 30/43] tcg/i386: Use tcg_constant_vec with tcg vec expanders Richard Henderson
2020-09-09  0:16 ` [PATCH 31/43] tcg: Remove tcg_gen_dup{8,16,32,64}i_vec Richard Henderson
2020-09-09  0:16 ` [PATCH 32/43] tcg/ppc: Use tcg_constant_vec with tcg vec expanders Richard Henderson
2020-09-09  0:16 ` [PATCH 33/43] tcg/aarch64: " Richard Henderson
2020-09-09  0:16 ` [PATCH 34/43] tcg: Add tcg-constr.c.inc Richard Henderson
2020-09-09  0:16 ` [PATCH 35/43] tcg/i386: Convert to tcg-constr.c.inc Richard Henderson
2020-09-09  0:16 ` [PATCH 36/43] tcg/aarch64: " Richard Henderson
2020-09-09  0:16 ` [PATCH 37/43] tcg/arm: " Richard Henderson
2020-09-09  0:16 ` [PATCH 38/43] tcg/mips: " Richard Henderson
2020-09-09  0:16 ` [PATCH 39/43] tcg/ppc: " Richard Henderson
2020-09-09  0:16 ` [PATCH 40/43] tcg/riscv: " Richard Henderson
2020-09-09  0:16 ` [PATCH 41/43] tcg/s390: " Richard Henderson
2020-09-09  0:16 ` [PATCH 42/43] tcg/sparc: " Richard Henderson
2020-09-09  0:16 ` [PATCH 43/43] tcg/tci: " Richard Henderson

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20200909001647.532249-2-richard.henderson@linaro.org \
    --to=richard.henderson@linaro.org \
    --cc=f4bug@amsat.org \
    --cc=qemu-devel@nongnu.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).