Change #263682
| Category | ffmpeg |
| Changed by | Andreas Rheinhardt <andreas.rheinhardt@outlook.com> |
| Changed at | Mon 06 Apr 2026 11:28:49 |
| Repository | https://git.ffmpeg.org/ffmpeg.git |
| Project | ffmpeg |
| Branch | master |
| Revision | ed116bab029c81ce67ed2967d94ab30ad623f6ac |
Comments
avcodec/x86/me_cmp: Use tmp reg in SUMSUB_BA if possible It allows to exchange a paddw by a movdqa. Old benchmarks: hadamard8_diff_0_c: 366.1 ( 1.00x) hadamard8_diff_0_sse2: 56.4 ( 6.49x) hadamard8_diff_0_ssse3: 53.0 ( 6.90x) hadamard8_diff_1_c: 183.0 ( 1.00x) hadamard8_diff_1_sse2: 28.0 ( 6.53x) hadamard8_diff_1_ssse3: 26.0 ( 7.03x) New benchmarks: hadamard8_diff_0_c: 371.4 ( 1.00x) hadamard8_diff_0_sse2: 55.0 ( 6.76x) hadamard8_diff_0_ssse3: 49.5 ( 7.50x) hadamard8_diff_1_c: 183.4 ( 1.00x) hadamard8_diff_1_sse2: 26.8 ( 6.85x) hadamard8_diff_1_ssse3: 23.1 ( 7.92x) Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Changed files
- libavcodec/x86/me_cmp.asm