[FFmpeg-devel] [PATCH 2/5] fate: avoid framemd5, use framecrc its faster
Michael Niedermayer
michaelni at gmx.at
Fri May 17 19:43:04 CEST 2013
On Fri, May 17, 2013 at 03:15:59PM +0200, Giorgio Vazzana wrote:
> Hi,
>
> about a year ago I studied ffmpeg's implementation of md5, and I
> noticed it was a slow compared to GNU md5sum. So I managed to make it
> almost as fast as the GNU implementation, but I never sent a patch
> because I wanted to work on it some more... of course I never did that
> because of lack of free time. Anyway, I found my old patch, since it's
> relatively simple I thought I'd send it in, hoping you will find it
> useful, maybe as a temporary solution until someone else comes up with
> something faster.
>
> Quick benchmark:
> # create test file and compile test program
> $ dd if=/dev/zero of=data bs=1M count=500
> $ gcc -Wall avutil_md5_test.c -O2 -o avutil_md5_test -L ./libavutil/ -lavutil
>
>
> # without the patch:
> $ for i in $(seq 1 3); do time ./avutil_md5_test < data; done
> d8b61b2c0025919d5321461045c8226f
>
> real 0m2.906s
> user 0m2.680s
> sys 0m0.180s
> d8b61b2c0025919d5321461045c8226f
>
> real 0m2.936s
> user 0m2.560s
> sys 0m0.330s
> d8b61b2c0025919d5321461045c8226f
>
> real 0m2.903s
> user 0m2.690s
> sys 0m0.170s
>
>
> # with the patch:
> $ for i in $(seq 1 3); do time ./avutil_md5_test < data; done
> d8b61b2c0025919d5321461045c8226f
>
> real 0m1.896s
> user 0m1.580s
> sys 0m0.280s
> d8b61b2c0025919d5321461045c8226f
>
> real 0m1.895s
> user 0m1.650s
> sys 0m0.210s
> d8b61b2c0025919d5321461045c8226f
>
> real 0m1.891s
> user 0m1.650s
> sys 0m0.200s
>
>
> # for comparison:
> $ for i in $(seq 1 3); do time md5sum < data; done
> d8b61b2c0025919d5321461045c8226f -
>
> real 0m1.842s
> user 0m1.510s
> sys 0m0.280s
> d8b61b2c0025919d5321461045c8226f -
>
> real 0m1.803s
> user 0m1.530s
> sys 0m0.250s
> d8b61b2c0025919d5321461045c8226f -
>
> real 0m1.821s
> user 0m1.620s
> sys 0m0.160s
>
> Giorgio Vazzana
> /* compile: gcc -Wall avutil_md5_test.c -O2 -o avutil_md5_test -L ./libavutil/ -lavutil */
> #include <stdio.h>
> #include <stdlib.h>
> #include <string.h>
> #include "libavutil/md5.h"
>
> #define BUFSIZE (64*1024)
>
> static void print_md5(uint8_t *md5)
> {
> int i;
> for (i = 0; i < 16; i++)
> printf("%02x", md5[i]);
> printf("\n");
> }
>
> int main()
> {
> size_t nread;
> struct AVMD5 *ctx;
> uint8_t buf[BUFSIZE];
> uint8_t md5val[16];
>
> ctx = av_md5_alloc();
>
> av_md5_init(ctx);
> while ((nread = fread(buf, 1, BUFSIZE, stdin)) > 0)
> av_md5_update(ctx, buf, nread);
> av_md5_final(ctx, md5val);
> free(ctx);
>
> print_md5(md5val);
>
> return 0;
> }
> md5.c | 41 ++++++++++++++++++++++++++++++++++++++---
> 1 file changed, 38 insertions(+), 3 deletions(-)
> 69e18ab824e95e1279af1fa6004d68d24358c2f9 0001-md5-speed-optimizations.patch
> From f7fbb6cfd6d4e70f66c21ec732902d5431e45bbc Mon Sep 17 00:00:00 2001
> From: Giorgio Vazzana <mywing81 at gmail.com>
> Date: Fri, 17 May 2013 14:51:45 +0200
> Subject: [PATCH] md5: speed optimizations
>
> ---
> libavutil/md5.c | 41 ++++++++++++++++++++++++++++++++++++++---
> 1 files changed, 38 insertions(+), 3 deletions(-)
>
> diff --git a/libavutil/md5.c b/libavutil/md5.c
> index f8f08f1..84c01f9 100644
> --- a/libavutil/md5.c
> +++ b/libavutil/md5.c
> @@ -49,12 +49,28 @@ struct AVMD5 *av_md5_alloc(void)
> return av_mallocz(sizeof(struct AVMD5));
> }
>
> +#if CONFIG_SMALL
> static const uint8_t S[4][4] = {
> { 7, 12, 17, 22 }, /* round 1 */
> { 5, 9, 14, 20 }, /* round 2 */
> { 4, 11, 16, 23 }, /* round 3 */
> { 6, 10, 15, 21 } /* round 4 */
> };
> +#else
> +static const uint32_t G[64] = { /* i = 0..63 */
> + 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, /* i */
> + 1, 6, 11, 0, 5, 10, 15, 4, 9, 14, 3, 8, 13, 2, 7, 12, /* (5*i + 1) % 16 */
> + 5, 8, 11, 14, 1, 4, 7, 10, 13, 0, 3, 6, 9, 12, 15, 2, /* (3*i + 5) % 16 */
> + 0, 7, 14, 5, 12, 3, 10, 1, 8, 15, 6, 13, 4, 11, 2, 9 /* (7*i) % 16 */
> +};
> +
> +static const uint32_t S[64] = {
> + 7, 12, 17, 22, 7, 12, 17, 22, 7, 12, 17, 22, 7, 12, 17, 22, /* round 1 */
> + 5, 9, 14, 20, 5, 9, 14, 20, 5, 9, 14, 20, 5, 9, 14, 20, /* round 2 */
> + 4, 11, 16, 23, 4, 11, 16, 23, 4, 11, 16, 23, 4, 11, 16, 23, /* round 3 */
> + 6, 10, 15, 21, 6, 10, 15, 21, 6, 10, 15, 21, 6, 10, 15, 21 /* round 4 */
> +};
> +#endif
>
> static const uint32_t T[64] = { // T[i]= fabs(sin(i+1)<<32)
> 0xd76aa478, 0xe8c7b756, 0x242070db, 0xc1bdceee, /* round 1 */
> @@ -78,6 +94,7 @@ static const uint32_t T[64] = { // T[i]= fabs(sin(i+1)<<32)
> 0xf7537e82, 0xbd3af235, 0x2ad7d2bb, 0xeb86d391,
> };
>
> +#if CONFIG_SMALL
> #define CORE(i, a, b, c, d) do { \
> t = S[i >> 4][i & 3]; \
> a += T[i]; \
> @@ -91,6 +108,21 @@ static const uint32_t T[64] = { // T[i]= fabs(sin(i+1)<<32)
> } \
> a = b + (a << t | a >> (32 - t)); \
> } while (0)
> +#else
> +#define CORE(i, a, b, c, d) do { \
> + t = S[i]; \
> + a += T[i]; \
> + \
> + if (i < 32) { \
> + if (i < 16) a += (d ^ (b & (c ^ d))) + X[G[i]]; \
> + else a += (c ^ (d & (c ^ b))) + X[G[i]]; \
> + } else { \
> + if (i < 48) a += (b ^ c ^ d) + X[G[i]]; \
> + else a += (c ^ (b | ~d)) + X[G[i]]; \
> + } \
> + a = b + (a << t | a >> (32 - t)); \
> + } while (0)
> +#endif
i wonder why these changes make any difference at all
gcc should be optimizing both to the same resulting code
[...]
--
Michael GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
Why not whip the teacher when the pupil misbehaves? -- Diogenes of Sinope
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20130517/bc499bb1/attachment.asc>
More information about the ffmpeg-devel
mailing list