diff options
| author | Denys Vlasenko <vda.linux@googlemail.com> | 2022-01-04 00:50:32 +0100 |
|---|---|---|
| committer | Denys Vlasenko <vda.linux@googlemail.com> | 2022-01-04 00:51:04 +0100 |
| commit | 1fc520ed286f815cae1da1e9f8014cb18a256744 (patch) | |
| tree | 285e2474f85c03b7d93a4e905276569589cee156 /coreutils | |
| parent | 7abb2bb96e0cd584f44dd8b219ad16d0232a6485 (diff) | |
| download | busybox-1fc520ed286f815cae1da1e9f8014cb18a256744.tar.gz | |
md5/shaXsum: use FEATURE_COPYBUF_KB to size the buffer instead of fixed 4k
function old new delta
md5_sha1_sum_main 536 565 +29
hash_file 419 401 -18
------------------------------------------------------------------------------
(add/remove: 0/0 grow/shrink: 1/1 up/down: 29/-18) Total: 11 bytes
In my test, for unrolled sha1, COPYBUF_KB=64 increases throughput
from 367 MB/s to 457 MB/s.
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
Diffstat (limited to 'coreutils')
| -rw-r--r-- | coreutils/md5_sha1_sum.c | 21 |
1 files changed, 14 insertions, 7 deletions
diff --git a/coreutils/md5_sha1_sum.c b/coreutils/md5_sha1_sum.c index 4efa23061..3b389cb6b 100644 --- a/coreutils/md5_sha1_sum.c +++ b/coreutils/md5_sha1_sum.c @@ -151,10 +151,12 @@ static unsigned char *hash_bin_to_hex(unsigned char *hash_value, return (unsigned char *)hex_value; } +#define BUFSZ (CONFIG_FEATURE_COPYBUF_KB < 4 ? 4096 : CONFIG_FEATURE_COPYBUF_KB * 1024) + #if !ENABLE_SHA3SUM -# define hash_file(f,w) hash_file(f) +# define hash_file(b,f,w) hash_file(b,f) #endif -static uint8_t *hash_file(const char *filename, unsigned sha3_width) +static uint8_t *hash_file(unsigned char *in_buf, const char *filename, unsigned sha3_width) { int src_fd, hash_len, count; union _ctx_ { @@ -227,8 +229,7 @@ static uint8_t *hash_file(const char *filename, unsigned sha3_width) } { - RESERVE_CONFIG_UBUFFER(in_buf, 4096); - while ((count = safe_read(src_fd, in_buf, 4096)) > 0) { + while ((count = safe_read(src_fd, in_buf, BUFSZ)) > 0) { update(&context, in_buf, count); } hash_value = NULL; @@ -238,7 +239,6 @@ static uint8_t *hash_file(const char *filename, unsigned sha3_width) final(&context, in_buf); hash_value = hash_bin_to_hex(in_buf, hash_len); } - RELEASE_CONFIG_BUFFER(in_buf); } if (src_fd != STDIN_FILENO) { @@ -251,6 +251,7 @@ static uint8_t *hash_file(const char *filename, unsigned sha3_width) int md5_sha1_sum_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE; int md5_sha1_sum_main(int argc UNUSED_PARAM, char **argv) { + unsigned char *in_buf; int return_value = EXIT_SUCCESS; unsigned flags; #if ENABLE_SHA3SUM @@ -279,6 +280,12 @@ int md5_sha1_sum_main(int argc UNUSED_PARAM, char **argv) if (!*argv) *--argv = (char*)"-"; + /* The buffer is not alloc/freed for each input file: + * for big values of COPYBUF_KB, this helps to keep its pages + * pre-faulted and possibly even fully cached on local CPU. + */ + in_buf = xmalloc(BUFSZ); + do { if (ENABLE_FEATURE_MD5_SHA1_SUM_CHECK && (flags & FLAG_CHECK)) { FILE *pre_computed_stream; @@ -310,7 +317,7 @@ int md5_sha1_sum_main(int argc UNUSED_PARAM, char **argv) *filename_ptr = '\0'; filename_ptr += 2; - hash_value = hash_file(filename_ptr, sha3_width); + hash_value = hash_file(in_buf, filename_ptr, sha3_width); if (hash_value && (strcmp((char*)hash_value, line) == 0)) { if (!(flags & FLAG_SILENT)) @@ -339,7 +346,7 @@ int md5_sha1_sum_main(int argc UNUSED_PARAM, char **argv) } fclose_if_not_stdin(pre_computed_stream); } else { - uint8_t *hash_value = hash_file(*argv, sha3_width); + uint8_t *hash_value = hash_file(in_buf, *argv, sha3_width); if (hash_value == NULL) { return_value = EXIT_FAILURE; } else { |