* [PATCH v8 1/3] read-cache: add strcmp_offset function
2017-04-10 21:14 [PATCH v8 0/3] read-cache: speed up add_index_entry git
@ 2017-04-10 21:14 ` git
2017-04-10 21:14 ` [PATCH v8 2/3] p0006-read-tree-checkout: perf test to time read-tree git
2017-04-10 21:14 ` [PATCH v8 3/3] read-cache: speed up add_index_entry during checkout git
2 siblings, 0 replies; 5+ messages in thread
From: git @ 2017-04-10 21:14 UTC (permalink / raw)
To: git; +Cc: gitster, peff, Jeff Hostetler
From: Jeff Hostetler <jeffhost@microsoft.com>
Add strcmp_offset() function to also return the offset of the
first change.
Add unit test and helper to verify.
Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
---
Makefile | 1 +
cache.h | 1 +
read-cache.c | 20 ++++++++++++++++++++
t/helper/.gitignore | 1 +
t/helper/test-strcmp-offset.c | 22 ++++++++++++++++++++++
t/t0065-strcmp-offset.sh | 21 +++++++++++++++++++++
6 files changed, 66 insertions(+)
create mode 100644 t/helper/test-strcmp-offset.c
create mode 100755 t/t0065-strcmp-offset.sh
diff --git a/Makefile b/Makefile
index 9ec6065..4c4c246 100644
--- a/Makefile
+++ b/Makefile
@@ -631,6 +631,7 @@ TEST_PROGRAMS_NEED_X += test-scrap-cache-tree
TEST_PROGRAMS_NEED_X += test-sha1
TEST_PROGRAMS_NEED_X += test-sha1-array
TEST_PROGRAMS_NEED_X += test-sigchain
+TEST_PROGRAMS_NEED_X += test-strcmp-offset
TEST_PROGRAMS_NEED_X += test-string-list
TEST_PROGRAMS_NEED_X += test-submodule-config
TEST_PROGRAMS_NEED_X += test-subprocess
diff --git a/cache.h b/cache.h
index 80b6372..3c55047 100644
--- a/cache.h
+++ b/cache.h
@@ -574,6 +574,7 @@ extern int write_locked_index(struct index_state *, struct lock_file *lock, unsi
extern int discard_index(struct index_state *);
extern int unmerged_index(const struct index_state *);
extern int verify_path(const char *path);
+extern int strcmp_offset(const char *s1, const char *s2, size_t *first_change);
extern int index_dir_exists(struct index_state *istate, const char *name, int namelen);
extern void adjust_dirname_case(struct index_state *istate, char *name);
extern struct cache_entry *index_file_exists(struct index_state *istate, const char *name, int namelen, int igncase);
diff --git a/read-cache.c b/read-cache.c
index 9054369..97f13a1 100644
--- a/read-cache.c
+++ b/read-cache.c
@@ -887,6 +887,26 @@ static int has_file_name(struct index_state *istate,
return retval;
}
+
+/*
+ * Like strcmp(), but also return the offset of the first change.
+ * If strings are equal, return the length.
+ */
+int strcmp_offset(const char *s1, const char *s2, size_t *first_change)
+{
+ size_t k;
+
+ if (!first_change)
+ return strcmp(s1, s2);
+
+ for (k = 0; s1[k] == s2[k]; k++)
+ if (s1[k] == '\0')
+ break;
+
+ *first_change = k;
+ return (unsigned char)s1[k] - (unsigned char)s2[k];
+}
+
/*
* Do we have another file with a pathname that is a proper
* subset of the name we're trying to add?
diff --git a/t/helper/.gitignore b/t/helper/.gitignore
index d6e8b36..0a89531 100644
--- a/t/helper/.gitignore
+++ b/t/helper/.gitignore
@@ -25,6 +25,7 @@
/test-sha1
/test-sha1-array
/test-sigchain
+/test-strcmp-offset
/test-string-list
/test-submodule-config
/test-subprocess
diff --git a/t/helper/test-strcmp-offset.c b/t/helper/test-strcmp-offset.c
new file mode 100644
index 0000000..4a45a54
--- /dev/null
+++ b/t/helper/test-strcmp-offset.c
@@ -0,0 +1,22 @@
+#include "cache.h"
+
+int cmd_main(int argc, const char **argv)
+{
+ int result;
+ size_t offset;
+
+ if (!argv[1] || !argv[2])
+ die("usage: %s <string1> <string2>", argv[0]);
+
+ result = strcmp_offset(argv[1], argv[2], &offset);
+
+ /*
+ * Because differnt CRTs behave differently, only rely on signs
+ * of the result values.
+ */
+ result = (result < 0 ? -1 :
+ result > 0 ? 1 :
+ 0);
+ printf("%d %"PRIuMAX"\n", result, (uintmax_t)offset);
+ return 0;
+}
diff --git a/t/t0065-strcmp-offset.sh b/t/t0065-strcmp-offset.sh
new file mode 100755
index 0000000..7d6d214
--- /dev/null
+++ b/t/t0065-strcmp-offset.sh
@@ -0,0 +1,21 @@
+#!/bin/sh
+
+test_description='Test strcmp_offset functionality'
+
+. ./test-lib.sh
+
+while read s1 s2 expect
+do
+ test_expect_success "strcmp_offset($s1, $s2)" '
+ echo "$expect" >expect &&
+ test-strcmp-offset "$s1" "$s2" >actual &&
+ test_cmp expect actual
+ '
+done <<-EOF
+abc abc 0 3
+abc def -1 0
+abc abz -1 2
+abc abcdef -1 3
+EOF
+
+test_done
--
2.9.3
^ permalink raw reply related [flat|nested] 5+ messages in thread* [PATCH v8 2/3] p0006-read-tree-checkout: perf test to time read-tree
2017-04-10 21:14 [PATCH v8 0/3] read-cache: speed up add_index_entry git
2017-04-10 21:14 ` [PATCH v8 1/3] read-cache: add strcmp_offset function git
@ 2017-04-10 21:14 ` git
2017-04-10 21:21 ` Jeff King
2017-04-10 21:14 ` [PATCH v8 3/3] read-cache: speed up add_index_entry during checkout git
2 siblings, 1 reply; 5+ messages in thread
From: git @ 2017-04-10 21:14 UTC (permalink / raw)
To: git; +Cc: gitster, peff, Jeff Hostetler
From: Jeff Hostetler <jeffhost@microsoft.com>
Created t/perf/repos/many-files.sh to generate large, but
artificial repositories.
Created t/perf/p0006-read-tree-checkout.sh to measure
performance on various read-tree, checkout, and update-index
operations.
Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
---
t/perf/p0006-read-tree-checkout.sh | 90 ++++++++++++++++++++++++++++
t/perf/repos/.gitignore | 1 +
t/perf/repos/many-files.sh | 118 +++++++++++++++++++++++++++++++++++++
3 files changed, 209 insertions(+)
create mode 100755 t/perf/p0006-read-tree-checkout.sh
create mode 100644 t/perf/repos/.gitignore
create mode 100755 t/perf/repos/many-files.sh
diff --git a/t/perf/p0006-read-tree-checkout.sh b/t/perf/p0006-read-tree-checkout.sh
new file mode 100755
index 0000000..69425ae
--- /dev/null
+++ b/t/perf/p0006-read-tree-checkout.sh
@@ -0,0 +1,90 @@
+#!/bin/sh
+##
+## This test measures the performance of various read-tree
+## and checkout operations. It is primarily interested in
+## the algorithmic costs of index operations and recursive
+## tree traversal -- and NOT disk I/O on thousands of files.
+## Therefore, it uses sparse-checkout to avoid populating
+## the ballast files.
+##
+## It expects the test repo to have certain characteristics.
+## Branches:
+## () master := an arbitrary commit.
+## () ballast := an arbitrary commit with a large number
+## of changes relative to "master".
+## () ballast-alias := a branch pointing to the same commit
+## as "ballast".
+## () ballast-1 := a commit with a 1 file difference from
+## "ballast".
+##
+## Ballast files in the repository should not appear in
+## the sparse-checkout (if present).
+##
+## See "t/perf/repos/many-files.sh" to generate some
+## synthetic data.
+##
+
+test_description="Tests performance of read-tree"
+
+. ./perf-lib.sh
+
+test_perf_default_repo
+
+test_expect_success 'setup' '
+ echo "sparse/*" >>.git/info/sparse-checkout &&
+ git config --local core.sparsecheckout 1 &&
+ git checkout ballast
+'
+
+nr_files=$(git ls-files | wc -l)
+
+test_perf "read-tree master ballast ($nr_files)" '
+ git read-tree -m master ballast -n
+'
+
+## Alternate between a commit with and without the ballast.
+test_perf "switch between master ballast ($nr_files)" '
+ git checkout master &&
+ git checkout ballast
+'
+
+## Alternate between 2 commits with the ballast and 1 change.
+test_perf "switch between ballast ballast-1 ($nr_files)" '
+ git checkout ballast-1 &&
+ git checkout ballast
+'
+
+## Alternate between 2 aliases for the same commit.
+test_perf "switch between aliases ballast ballast-alias ($nr_files)" '
+ git checkout ballast-alias &&
+ git checkout ballast
+'
+
+export nr_random=200000
+
+## Add random items in sorted order. These will effectively
+## get appended because "zzzz/" sorts after anything already
+## present in index. Then delete them in reverse order, so
+## that the last is removed each time.
+test_perf "add forward sorted items ($nr_files) ($nr_random)" '
+ awk "BEGIN { for (f = 0; f <= $nr_random; f++) printf \"r_%06d\n\", f }" |
+ sed "s|^|100644 $EMPTY_BLOB zzzz/|" |
+ git update-index --index-info &&
+ awk "BEGIN { for (f = $nr_random; f >= 0; f--) printf \"r_%06d\n\", f }" |
+ sed "s|^|000000 $EMPTY_BLOB zzzz/|" |
+ git update-index --index-info
+'
+
+## Add random items in reverse order. These will be inserted
+## NEAR the end of the index, but not at the end. Then delete
+## them in the opposite order, so that we remove from the middle.
+test_perf "add reverse sorted items ($nr_files) ($nr_random)" '
+ awk "BEGIN { for (f = $nr_random; f >= 0; f--) printf \"r_%06d\n\", f }" |
+ sed "s|^|100644 $EMPTY_BLOB zzzz/|" |
+ git update-index --index-info &&
+ awk "BEGIN { for (f = 0; f <= $nr_random; f++) printf \"r_%06d\n\", f }" |
+ sed "s|^|000000 $EMPTY_BLOB zzzz/|" |
+ git update-index --index-info
+'
+
+test_done
diff --git a/t/perf/repos/.gitignore b/t/perf/repos/.gitignore
new file mode 100644
index 0000000..72e3dc3
--- /dev/null
+++ b/t/perf/repos/.gitignore
@@ -0,0 +1 @@
+gen-*/
diff --git a/t/perf/repos/many-files.sh b/t/perf/repos/many-files.sh
new file mode 100755
index 0000000..a4c44b3
--- /dev/null
+++ b/t/perf/repos/many-files.sh
@@ -0,0 +1,118 @@
+#!/bin/sh
+## Generate test data repository "many-files.git" using the given parameters.
+## Usage: [-r repo] [-d depth] [-w width] [-f files]
+##
+## -r repo: path to the new repo to be generated
+## -d depth: the depth of sub-directories
+## -w width: the number of sub-directories at each level
+## -f files: the number of files created in each directory
+##
+## Note that all files will have the same SHA-1 and each
+## directory at a level will have the same SHA-1, so we
+## will potentially have a large index, but not a large
+## ODB.
+##
+## Ballast will be created under "ballast/". Sparse-checkout
+## will be enabled so that they will not be populated by
+## default.
+
+EMPTY_BLOB=e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
+
+set -e
+
+## (5, 10, 9) will create 999,999 ballast files.
+## (4, 10, 9) will create 99,999 ballast files.
+depth=5
+width=10
+files=9
+
+while test "$#" -ne 0
+do
+ case "$1" in
+ -r)
+ shift;
+ test "$#" -ne 0 || { echo 'error: -r requires an argument' >&2; exit 1; }
+ repo=$1;
+ shift ;;
+ -d)
+ shift;
+ test "$#" -ne 0 || { echo 'error: -d requires an argument' >&2; exit 1; }
+ depth=$1;
+ shift ;;
+ -w)
+ shift;
+ test "$#" -ne 0 || { echo 'error: -w requires an argument' >&2; exit 1; }
+ width=$1;
+ shift ;;
+ -f)
+ shift;
+ test "$#" -ne 0 || { echo 'error: -f requires an argument' >&2; exit 1; }
+ files=$1;
+ shift ;;
+ *)
+ echo "error: unknown option '$1'" >&2; exit 1 ;;
+ esac
+done
+
+## Inflate the index with thousands of empty files.
+## usage: dir depth width files
+fill_index() {
+ awk -v arg_dir=$1 -v arg_depth=$2 -v arg_width=$3 -v arg_files=$4 '
+ function make_paths(dir, depth, width, files, f, w) {
+ for (f = 1; f <= files; f++) {
+ print dir "/file" f
+ }
+ if (depth > 0) {
+ for (w = 1; w <= width; w++) {
+ make_paths(dir "/dir" w, depth - 1, width, files)
+ }
+ }
+ }
+ END { make_paths(arg_dir, arg_depth, arg_width, arg_files) }
+ ' </dev/null |
+ sed "s/^/100644 $EMPTY_BLOB /" |
+ git update-index --index-info
+ return 0
+}
+
+[ -z "$repo" ] && repo=gen-many-files-$depth.$width.$files.git
+
+mkdir $repo
+cd $repo
+git init .
+
+## Create initial commit just to define branch.
+mkdir sparse
+touch sparse/file.txt
+echo "$depth $width $files" >sparse/repo.params
+git add sparse
+git commit -q -m params
+
+## Turn on sparse-checkout so that we don't have to populate
+## the ballast when switching branches. Use reset --hard to
+## quickly checkout the new HEAD with minimum actual files.
+echo 'sparse/*' >>.git/info/sparse-checkout
+git config --local core.sparsecheckout 1
+git reset --hard
+
+## Inflate the index with thousands of empty files and commit.
+git checkout -b ballast
+fill_index "ballast" $depth $width $files
+git commit -q -m "ballast"
+
+## Create an alias for that commit.
+git branch "ballast-alias"
+
+nr_files=$(git ls-files | wc -l)
+
+## Modify 1 file and commit.
+git checkout -b "ballast-1"
+echo x >sparse/file.txt
+git add sparse
+git commit -q -m "ballast plus 1"
+
+## Checkout master to put repo in canonical state.
+git checkout master
+
+echo "Repository "$repo" ($depth, $width, $files) created. Ballast $nr_files."
+exit 0
--
2.9.3
^ permalink raw reply related [flat|nested] 5+ messages in thread* Re: [PATCH v8 2/3] p0006-read-tree-checkout: perf test to time read-tree
2017-04-10 21:14 ` [PATCH v8 2/3] p0006-read-tree-checkout: perf test to time read-tree git
@ 2017-04-10 21:21 ` Jeff King
0 siblings, 0 replies; 5+ messages in thread
From: Jeff King @ 2017-04-10 21:21 UTC (permalink / raw)
To: git; +Cc: git, gitster, Jeff Hostetler
On Mon, Apr 10, 2017 at 09:14:02PM +0000, git@jeffhostetler.com wrote:
> From: Jeff Hostetler <jeffhost@microsoft.com>
>
> Created t/perf/repos/many-files.sh to generate large, but
> artificial repositories.
I think this is a good direction. In the long run we might want some
kind of magic to pull from the "library" of repos when running perf
tests, but it's not a big deal to run the script manually and point
GIT_PERF_REPO at the result.
As a bonus, this should be faster when running perf tests, since we can
reuse the built repo when testing each version of Git.
> +## This test measures the performance of various read-tree
> +## and checkout operations. It is primarily interested in
> +## the algorithmic costs of index operations and recursive
> +## tree traversal -- and NOT disk I/O on thousands of files.
> +## Therefore, it uses sparse-checkout to avoid populating
> +## the ballast files.
> +##
> +## It expects the test repo to have certain characteristics.
> +## Branches:
> +## () master := an arbitrary commit.
> +## () ballast := an arbitrary commit with a large number
> +## of changes relative to "master".
> +## () ballast-alias := a branch pointing to the same commit
> +## as "ballast".
> +## () ballast-1 := a commit with a 1 file difference from
> +## "ballast".
I'm OK with leaving these requirements on the repo in the name of
simplicity, though it does make it harder to perf-test against a regular
repo.
I wonder if we could make reasonable guesses, like:
master => HEAD
ballast => $(git rev-list HEAD | tail -n 1)
ballast-alias => git branch $ballast
ballast-1 => HEAD^
That would approximate your conditions in a real-world repository, and
it should be easy to make your synthetic one fit the bill exactly.
I don't know if you'd want to turn on sparse checkout manually or not
when testing a real-world repo.
-Peff
^ permalink raw reply [flat|nested] 5+ messages in thread
* [PATCH v8 3/3] read-cache: speed up add_index_entry during checkout
2017-04-10 21:14 [PATCH v8 0/3] read-cache: speed up add_index_entry git
2017-04-10 21:14 ` [PATCH v8 1/3] read-cache: add strcmp_offset function git
2017-04-10 21:14 ` [PATCH v8 2/3] p0006-read-tree-checkout: perf test to time read-tree git
@ 2017-04-10 21:14 ` git
2 siblings, 0 replies; 5+ messages in thread
From: git @ 2017-04-10 21:14 UTC (permalink / raw)
To: git; +Cc: gitster, peff, Jeff Hostetler
From: Jeff Hostetler <jeffhost@microsoft.com>
Teach add_index_entry_with_check() and has_dir_name()
to see if the path of the new item is greater than the
last path in the index array before attempting to search
for it.
During checkout, merge_working_tree() populates the new
index in sorted order, so this change will save at least 2
binary lookups per file. This preserves the original
behavior but simply checks the last element before starting
the search.
This helps performance on very large repositories.
This can be seen using p0006-read-tree-checkout.sh and the
artificial repository created by t/perf/repos/many-files.sh
with parameters (5, 10, 9). (1M files in index.)
1..7
Test HEAD^ HEAD
------------------------------------------------------------------------------------------------------------
0006.2: read-tree master ballast (1000001) 4.01(2.71+1.28) 3.24(1.84+1.38) -19.2%
0006.3: switch between master ballast (1000001) 8.23(5.60+2.45) 6.73(4.20+2.35) -18.2%
0006.4: switch between ballast ballast-1 (1000001) 13.36(8.60+4.35) 11.84(7.28+4.08) -11.4%
0006.5: switch between aliases ballast ballast-alias (1000001) 13.43(8.68+4.27) 12.09(7.28+4.37) -10.0%
0006.6: add forward sorted items (1000001) (200000) 2.42(1.76+0.42) 2.40(1.66+0.37) -0.8%
0006.7: add reverse sorted items (1000001) (200000) 18.59(17.96+0.33) 18.70(18.07+0.42) +0.6%
Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com>
---
read-cache.c | 46 ++++++++++++++++++++++++++++++++++++++++++++--
1 file changed, 44 insertions(+), 2 deletions(-)
diff --git a/read-cache.c b/read-cache.c
index 97f13a1..a8ef823 100644
--- a/read-cache.c
+++ b/read-cache.c
@@ -918,9 +918,24 @@ static int has_dir_name(struct index_state *istate,
int stage = ce_stage(ce);
const char *name = ce->name;
const char *slash = name + ce_namelen(ce);
+ size_t len_eq_last;
+ int cmp_last = 0;
+
+ if (istate->cache_nr > 0) {
+ /*
+ * Compare the entry's full path with the last path in the index.
+ * If it sorts AFTER the last entry in the index and they have no
+ * common prefix, then there cannot be any F/D name conflicts.
+ */
+ cmp_last = strcmp_offset(name,
+ istate->cache[istate->cache_nr-1]->name,
+ &len_eq_last);
+ if (cmp_last > 0 && len_eq_last == 0)
+ return retval;
+ }
for (;;) {
- int len;
+ size_t len;
for (;;) {
if (*--slash == '/')
@@ -930,6 +945,24 @@ static int has_dir_name(struct index_state *istate,
}
len = slash - name;
+ if (cmp_last > 0) {
+ /*
+ * If this part of the directory prefix (including the trailing
+ * slash) already appears in the path of the last entry in the
+ * index, then we cannot also have a file with this prefix (or
+ * any parent directory prefix).
+ */
+ if (len+1 <= len_eq_last)
+ return retval;
+ /*
+ * If this part of the directory prefix (excluding the trailing
+ * slash) is longer than the known equal portions, then this part
+ * of the prefix cannot collide with a file. Go on to the parent.
+ */
+ if (len > len_eq_last)
+ continue;
+ }
+
pos = index_name_stage_pos(istate, name, len, stage);
if (pos >= 0) {
/*
@@ -1021,7 +1054,16 @@ static int add_index_entry_with_check(struct index_state *istate, struct cache_e
if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
cache_tree_invalidate_path(istate, ce->name);
- pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
+
+ /*
+ * If this entry's path sorts after the last entry in the index,
+ * we can avoid searching for it.
+ */
+ if (istate->cache_nr > 0 &&
+ strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
+ pos = -istate->cache_nr - 1;
+ else
+ pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
/* existing match? Just replace it. */
if (pos >= 0) {
--
2.9.3
^ permalink raw reply related [flat|nested] 5+ messages in thread