This is an automated email from the git hooks/post-receive script. Git pushed a commit to branch master in repository ffmpeg.
commit d03788af83260044c6ef0e21a1c24245920cb5c4 Author: Zuxy Meng <[email protected]> AuthorDate: Fri May 29 22:28:24 2026 -0700 Commit: Zuxy Meng <[email protected]> CommitDate: Fri Jul 24 19:16:45 2026 -0700 avcodec/x86/h264_intrapred: Deprecate MMX from pred8x8l_down_right_8 Convert use of MMX to SSE2, also remove the SSSE3 impl. since we no longer use palignr Before: pred8x8l_down_right_8_sse2: 26.4 ( 3.26x) pred8x8l_down_right_8_ssse3: 23.0 ( 3.74x) After: pred8x8l_down_right_8_sse2: 19.7 ( 4.33x) Signed-off-by: Zuxy Meng <[email protected]> --- libavcodec/x86/h264_intrapred.asm | 146 ++++++++++++++++------------------- libavcodec/x86/h264_intrapred_init.c | 2 - 2 files changed, 66 insertions(+), 82 deletions(-) diff --git a/libavcodec/x86/h264_intrapred.asm b/libavcodec/x86/h264_intrapred.asm index 88abb7f8f2..e60c1aa474 100644 --- a/libavcodec/x86/h264_intrapred.asm +++ b/libavcodec/x86/h264_intrapred.asm @@ -1158,99 +1158,91 @@ PRED8x8L_DOWN_LEFT ; int has_topright, ptrdiff_t stride) ;----------------------------------------------------------------------------- -%macro PRED8x8L_DOWN_RIGHT 0 -cglobal pred8x8l_down_right_8, 4,5 +INIT_XMM sse2 +cglobal pred8x8l_down_right_8, 4,5,7 sub r0, r3 lea r4, [r0+r3*2] - movq mm0, [r0+r3*1-8] - punpckhbw mm0, [r0+r3*0-8] - movq mm1, [r4+r3*1-8] - punpckhbw mm1, [r0+r3*2-8] + movd xmm0, [r0+r3*1-4] + movd xmm4, [r0+r3*0-4] + punpcklbw xmm0, xmm4 + movd xmm1, [r4+r3*1-4] + movd xmm4, [r0+r3*2-4] + punpcklbw xmm1, xmm4 mov r4, r0 - punpckhwd mm1, mm0 + punpcklwd xmm1, xmm0 lea r0, [r0+r3*4] - movq mm2, [r0+r3*1-8] - punpckhbw mm2, [r0+r3*0-8] + movd xmm2, [r0+r3*1-4] + movd xmm4, [r0+r3*0-4] + punpcklbw xmm2, xmm4 lea r0, [r0+r3*2] - movq mm3, [r0+r3*1-8] - punpckhbw mm3, [r0+r3*0-8] - punpckhwd mm3, mm2 - punpckhdq mm3, mm1 + movd xmm3, [r0+r3*1-4] + movd xmm4, [r0+r3*0-4] + punpcklbw xmm3, xmm4 + punpcklwd xmm3, xmm2 + punpckhdq xmm3, xmm1 + pshufd xmm3, xmm3, 0xee lea r0, [r0+r3*2] - movq mm0, [r0+r3*0-8] - movq mm1, [r4] + movq xmm0, [r0+r3*0-8] + movq xmm1, [r4] mov r0, r4 - movq mm4, mm3 - movq mm2, mm3 - PALIGNR mm4, mm0, 7, mm0 - PALIGNR mm1, mm2, 1, mm2 + mova xmm2, xmm3 + punpcklqdq xmm0, xmm3 + psrldq xmm4, xmm0, 7 + punpcklqdq xmm2, xmm1 + psrldq xmm1, xmm2, 1 test r1d, r1d - jz .fix_lt_1 - jmp .do_left + jnz .do_left .fix_lt_1: - movq mm5, mm3 - pxor mm5, mm4 - psrlq mm5, 56 - psllq mm5, 48 - pxor mm1, mm5 - jmp .do_left + pxor xmm5, xmm3, xmm4 + psrlq xmm5, 56 + psllq xmm5, 48 + pxor xmm1, xmm5 +.do_left: + mova xmm0, xmm4 + PRED4x4_LOWPASS xmm2, xmm1, xmm4, xmm3, xmm5 + mova xmm4, xmm0 + movq xmm6, xmm2 + PRED4x4_LOWPASS xmm1, xmm3, xmm0, xmm4, xmm5 + psllq xmm1, 56 + punpcklqdq xmm1, xmm2 + psrldq xmm1, 7 + mova xmm0, xmm1 + movu xmm2, [r0-8] + movu xmm1, [r0] + mova xmm3, xmm1 + psrldq xmm2, 7 + psrldq xmm1, 1 + jnz .check_tr .fix_lt_2: - movq mm5, mm3 - pxor mm5, mm2 - psllq mm5, 56 - psrlq mm5, 56 - pxor mm2, mm5 + pxor xmm5, xmm3, xmm2 + psllq xmm5, 56 + psrlq xmm5, 56 + pxor xmm2, xmm5 +.check_tr: test r2d, r2d jnz .do_top .fix_tr_1: - movq mm5, mm3 - pxor mm5, mm1 - psrlq mm5, 56 - psllq mm5, 56 - pxor mm1, mm5 - jmp .do_top -.do_left: - movq mm0, mm4 - PRED4x4_LOWPASS mm2, mm1, mm4, mm3, mm5 - movq mm1, mm0 - movq mm7, mm2 - movq2dq xmm3, mm2 - PRED4x4_LOWPASS mm1, mm3, mm0, mm1, mm5 - psllq mm1, 56 - PALIGNR mm7, mm1, 7, mm3 - movq2dq xmm1, mm7 - movq mm0, [r0-8] - movq mm3, [r0] - movq mm1, [r0+8] - movq mm2, mm3 - movq mm4, mm3 - PALIGNR mm2, mm0, 7, mm0 - PALIGNR mm1, mm4, 1, mm4 - test r1d, r1d - jz .fix_lt_2 - test r2d, r2d - jz .fix_tr_1 + pxor xmm5, xmm3, xmm1 + psrlq xmm5, 56 + psllq xmm5, 56 + pxor xmm1, xmm5 .do_top: - PRED4x4_LOWPASS mm3, mm2, mm1, mm3, mm5 - movq2dq xmm4, mm3 + PRED4x4_LOWPASS xmm4, xmm2, xmm1, xmm3, xmm5 lea r1, [r0+r3*2] - movdqa xmm0, xmm3 + movdqa xmm1, xmm6 pslldq xmm4, 8 - por xmm3, xmm4 + por xmm6, xmm4 lea r2, [r1+r3*2] pslldq xmm4, 1 - por xmm1, xmm4 - psrldq xmm0, 7 - pslldq xmm0, 15 - psrldq xmm0, 7 - por xmm1, xmm0 + por xmm0, xmm4 + psrldq xmm1, 7 + pslldq xmm1, 15 + psrldq xmm1, 7 + por xmm0, xmm1 lea r0, [r2+r3*2] - movdqa xmm2, xmm3 - psrldq xmm2, 1 -INIT_XMM cpuname - PRED4x4_LOWPASS xmm3, xmm1, xmm2, xmm3, xmm4 - movdqa xmm1, xmm3 - psrldq xmm1, 1 + psrldq xmm2, xmm6, 1 + PRED4x4_LOWPASS xmm3, xmm0, xmm2, xmm6, xmm4 + psrldq xmm1, xmm3, 1 movq [r0+r3*2], xmm3 movq [r0+r3*1], xmm1 psrldq xmm3, 2 @@ -1266,12 +1258,6 @@ INIT_XMM cpuname movq [r4+r3*2], xmm3 movq [r4+r3*1], xmm1 RET -%endmacro - -INIT_MMX sse2 -PRED8x8L_DOWN_RIGHT -INIT_MMX ssse3 -PRED8x8L_DOWN_RIGHT ;----------------------------------------------------------------------------- ; void ff_pred8x8l_vertical_right_8(uint8_t *src, int has_topleft, diff --git a/libavcodec/x86/h264_intrapred_init.c b/libavcodec/x86/h264_intrapred_init.c index 027746a8f9..486ae0fe1b 100644 --- a/libavcodec/x86/h264_intrapred_init.c +++ b/libavcodec/x86/h264_intrapred_init.c @@ -133,7 +133,6 @@ PRED8x8L(vertical, 8, sse2) PRED8x8L(down_left, 8, sse2) PRED8x8L(down_left, 8, ssse3) PRED8x8L(down_right, 8, sse2) -PRED8x8L(down_right, 8, ssse3) PRED8x8L(vertical_right, 8, sse2) PRED8x8L(vertical_right, 8, ssse3) PRED8x8L(vertical_left, 8, sse2) @@ -229,7 +228,6 @@ av_cold void ff_h264_pred_init_x86(H264PredContext *h, int codec_id, if (chroma_format_idc <= 1) h->pred8x8 [HOR_PRED8x8 ] = ff_pred8x8_horizontal_8_ssse3; h->pred8x8l [DIAG_DOWN_LEFT_PRED ] = ff_pred8x8l_down_left_8_ssse3; - h->pred8x8l [DIAG_DOWN_RIGHT_PRED ] = ff_pred8x8l_down_right_8_ssse3; h->pred8x8l [VERT_RIGHT_PRED ] = ff_pred8x8l_vertical_right_8_ssse3; h->pred8x8l [VERT_LEFT_PRED ] = ff_pred8x8l_vertical_left_8_ssse3; h->pred8x8l [HOR_DOWN_PRED ] = ff_pred8x8l_horizontal_down_8_ssse3; _______________________________________________ ffmpeg-cvslog mailing list -- [email protected] To unsubscribe send an email to [email protected]
