The branch, master has been updated
via 739fa967 Change odd-ball map_ptr() to use remainder like the others.
via d474f298 Improve performance of file_checksum()
from a863c62c More NEWS updates.
https://git.samba.org/?p=rsync.git;a=shortlog;h=master
- Log -----------------------------------------------------------------
commit 739fa967378bfa8956352c9789692a5232a67f66
Author: Wayne Davison <wa...@opencoder.net>
Date: Sun May 24 20:38:48 2020 -0700
Change odd-ball map_ptr() to use remainder like the others.
commit d474f2986e656a848780bc8b36a02297e75eb3ab
Author: Jorrit Jongma <g...@jongma.org>
Date: Mon May 25 00:30:55 2020 +0200
Improve performance of file_checksum()
Previously files were hashed in blocks of CSUM_CHUNK (64) bytes. This
causes significant overhead. The CSUM_CHUNK define cannot be changed as
md5.c depends on it, but there is no obvious reason to use it in
file_checksum(). By using CHUNK_SIZE (32 kB) instead, in some test
cases throughput more than doubles.
-----------------------------------------------------------------------
Summary of changes:
checksum.c | 20 ++++++++++----------
1 file changed, 10 insertions(+), 10 deletions(-)
Changeset truncated at 500 lines:
diff --git a/checksum.c b/checksum.c
index 2e804c47..fb771402 100644
--- a/checksum.c
+++ b/checksum.c
@@ -294,7 +294,7 @@ void file_checksum(const char *fname, const STRUCT_STAT
*st_p, char *sum)
if (fd == -1)
return;
- buf = map_file(fd, len, MAX_MAP_SIZE, CSUM_CHUNK);
+ buf = map_file(fd, len, MAX_MAP_SIZE, CHUNK_SIZE);
switch (checksum_type) {
case CSUM_MD5: {
@@ -302,8 +302,8 @@ void file_checksum(const char *fname, const STRUCT_STAT
*st_p, char *sum)
MD5_Init(&m5);
- for (i = 0; i + CSUM_CHUNK <= len; i += CSUM_CHUNK)
- MD5_Update(&m5, (uchar *)map_ptr(buf, i, CSUM_CHUNK),
CSUM_CHUNK);
+ for (i = 0; i + CHUNK_SIZE <= len; i += CHUNK_SIZE)
+ MD5_Update(&m5, (uchar *)map_ptr(buf, i, CHUNK_SIZE),
CHUNK_SIZE);
remainder = (int32)(len - i);
if (remainder > 0)
@@ -319,8 +319,8 @@ void file_checksum(const char *fname, const STRUCT_STAT
*st_p, char *sum)
MD4_Init(&m4);
- for (i = 0; i + CSUM_CHUNK <= len; i += CSUM_CHUNK)
- MD4_Update(&m4, (uchar *)map_ptr(buf, i, CSUM_CHUNK),
CSUM_CHUNK);
+ for (i = 0; i + CHUNK_SIZE <= len; i += CHUNK_SIZE)
+ MD4_Update(&m4, (uchar *)map_ptr(buf, i, CHUNK_SIZE),
CHUNK_SIZE);
remainder = (int32)(len - i);
if (remainder > 0)
@@ -337,8 +337,8 @@ void file_checksum(const char *fname, const STRUCT_STAT
*st_p, char *sum)
mdfour_begin(&m);
- for (i = 0; i + CSUM_CHUNK <= len; i += CSUM_CHUNK)
- mdfour_update(&m, (uchar *)map_ptr(buf, i, CSUM_CHUNK),
CSUM_CHUNK);
+ for (i = 0; i + CHUNK_SIZE <= len; i += CHUNK_SIZE)
+ mdfour_update(&m, (uchar *)map_ptr(buf, i, CHUNK_SIZE),
CHUNK_SIZE);
/* Prior to version 27 an incorrect MD4 checksum was computed
* by failing to call mdfour_tail() for block sizes that
@@ -362,9 +362,9 @@ void file_checksum(const char *fname, const STRUCT_STAT
*st_p, char *sum)
exit_cleanup(RERR_STREAMIO);
}
- for (i = 0; i + CSUM_CHUNK <= len; i += CSUM_CHUNK) {
+ for (i = 0; i + CHUNK_SIZE <= len; i += CHUNK_SIZE) {
XXH_errorcode const updateResult =
- XXH64_update(state, (uchar *)map_ptr(buf, i,
CSUM_CHUNK), CSUM_CHUNK);
+ XXH64_update(state, (uchar *)map_ptr(buf, i,
CHUNK_SIZE), CHUNK_SIZE);
if (updateResult == XXH_ERROR) {
rprintf(FERROR, "error computing XXH64 hash");
exit_cleanup(RERR_STREAMIO);
@@ -373,7 +373,7 @@ void file_checksum(const char *fname, const STRUCT_STAT
*st_p, char *sum)
remainder = (int32)(len - i);
if (remainder > 0)
- XXH64_update(state, (uchar *)map_ptr(buf, i,
CSUM_CHUNK), remainder);
+ XXH64_update(state, (uchar *)map_ptr(buf, i,
remainder), remainder);
SIVAL64(sum, 0, XXH64_digest(state));
XXH64_freeState(state);
--
The rsync repository.
_______________________________________________
rsync-cvs mailing list
rsync-cvs@lists.samba.org
https://lists.samba.org/mailman/listinfo/rsync-cvs
