This is an automated email from the git hooks/post-receive script. Git pushed a commit to branch master in repository ffmpeg.
commit c7e532e81e44c5687270dfffa8c05e393bd94830 Author: Zuxy Meng <[email protected]> AuthorDate: Tue May 26 20:47:44 2026 -0700 Commit: Zuxy Meng <[email protected]> CommitDate: Fri Jul 24 19:16:45 2026 -0700 avcodec/x86/h264_intrapred: SSE2 impl. of pred8x8l_horizontal_8 Deprecate MMX. Remove the SSSE3 impl. since we no longer use palignr. No change in performance. pred8x8l_horizontal_8_mmxext: 16.2 ( 1.44x) pred8x8l_horizontal_8_ssse3: 15.6 ( 1.50x) pred8x8l_horizontal_8_sse2: 15.8 ( 1.49x) Signed-off-by: Zuxy Meng <[email protected]> --- libavcodec/x86/h264_intrapred.asm | 101 +++++++++++++++++------------------ libavcodec/x86/h264_intrapred_init.c | 6 +-- 2 files changed, 51 insertions(+), 56 deletions(-) diff --git a/libavcodec/x86/h264_intrapred.asm b/libavcodec/x86/h264_intrapred.asm index 841ee9255e..599b147758 100644 --- a/libavcodec/x86/h264_intrapred.asm +++ b/libavcodec/x86/h264_intrapred.asm @@ -958,72 +958,69 @@ cglobal pred8x8l_dc_8, 4,5,6 ; int has_topright, ptrdiff_t stride) ;----------------------------------------------------------------------------- -%macro PRED8x8L_HORIZONTAL 0 -cglobal pred8x8l_horizontal_8, 4,4 +INIT_XMM sse2 +cglobal pred8x8l_horizontal_8, 4,4,6 sub r0, r3 lea r2, [r0+r3*2] - movq mm0, [r0+r3*1-8] + movd m0, [r0+r3*1-4] test r1d, r1d lea r1, [r0+r3] cmovnz r1, r0 - punpckhbw mm0, [r1+r3*0-8] - movq mm1, [r2+r3*1-8] - punpckhbw mm1, [r0+r3*2-8] + movd m4, [r1+r3*0-4] + punpcklbw m0, m4 + movd m1, [r2+r3*1-4] + movd m4, [r0+r3*2-4] + punpcklbw m1, m4 mov r2, r0 - punpckhwd mm1, mm0 + punpcklwd m1, m0 lea r0, [r0+r3*4] - movq mm2, [r0+r3*1-8] - punpckhbw mm2, [r0+r3*0-8] + movd m2, [r0+r3*1-4] + movd m4, [r0+r3*0-4] + punpcklbw m2, m4 lea r0, [r0+r3*2] - movq mm3, [r0+r3*1-8] - punpckhbw mm3, [r0+r3*0-8] - punpckhwd mm3, mm2 - punpckhdq mm3, mm1 + movd m3, [r0+r3*1-4] + movd m4, [r0+r3*0-4] + punpcklbw m3, m4 + punpcklwd m3, m2 + punpckhdq m3, m1 + pshufd m3, m3, 0xee lea r0, [r0+r3*2] - movq mm0, [r0+r3*0-8] - movq mm1, [r1+r3*0-8] + movq m0, [r0+r3*0-8] + movq m1, [r1+r3*0-8] mov r0, r2 - movq mm4, mm3 - movq mm2, mm3 - PALIGNR mm4, mm0, 7, mm0 - PALIGNR mm1, mm2, 1, mm2 - movq mm0, mm4 - PRED4x4_LOWPASS mm2, mm1, mm4, mm3, mm5 - movq mm4, mm0 - movq mm7, mm2 - PRED4x4_LOWPASS mm4, mm3, mm0, mm4, mm5 - psllq mm4, 56 - PALIGNR mm7, mm4, 7, mm3 - movq mm3, mm7 + mova m2, m3 + punpcklqdq m0, m3 + psrldq m0, 7 + mova m4, m0 + punpcklqdq m2, m1 + psrldq m2, 1 + PRED4x4_LOWPASS m1, m2, m4, m3, m5 + mova m4, m0 + PRED4x4_LOWPASS m2, m3, m0, m4, m5 + psllq m2, 56 + punpcklqdq m2, m1 + psrldq m2, 7 lea r1, [r0+r3*2] - movq mm7, mm3 - punpckhbw mm3, mm3 - punpcklbw mm7, mm7 - pshufw mm0, mm3, 0xff - pshufw mm1, mm3, 0xaa + punpcklbw m2, m2 + pshufhw m0, m2, 0xff + pshufhw m1, m2, 0xaa lea r2, [r1+r3*2] - pshufw mm2, mm3, 0x55 - pshufw mm3, mm3, 0x00 - pshufw mm4, mm7, 0xff - pshufw mm5, mm7, 0xaa - pshufw mm6, mm7, 0x55 - pshufw mm7, mm7, 0x00 - movq [r0+r3*1], mm0 - movq [r0+r3*2], mm1 - movq [r1+r3*1], mm2 - movq [r1+r3*2], mm3 - movq [r2+r3*1], mm4 - movq [r2+r3*2], mm5 + pshufhw m3, m2, 0x55 + pshufhw m4, m2, 0x00 + movhps [r0+r3*1], m0 + movhps [r0+r3*2], m1 + movhps [r1+r3*1], m3 + movhps [r1+r3*2], m4 + pshuflw m0, m2, 0xff + pshuflw m1, m2, 0xaa + pshuflw m3, m2, 0x55 + pshuflw m4, m2, 0x00 + movq [r2+r3*1], m0 + movq [r2+r3*2], m1 lea r0, [r2+r3*2] - movq [r0+r3*1], mm6 - movq [r0+r3*2], mm7 + movq [r0+r3*1], m3 + movq [r0+r3*2], m4 RET -%endmacro - -INIT_MMX mmxext -PRED8x8L_HORIZONTAL -INIT_MMX ssse3 -PRED8x8L_HORIZONTAL ;----------------------------------------------------------------------------- ; void ff_pred8x8l_vertical_8(uint8_t *src, int has_topleft, int has_topright, diff --git a/libavcodec/x86/h264_intrapred_init.c b/libavcodec/x86/h264_intrapred_init.c index 5cc83d17c7..bd69e192ee 100644 --- a/libavcodec/x86/h264_intrapred_init.c +++ b/libavcodec/x86/h264_intrapred_init.c @@ -128,8 +128,7 @@ PRED8x8(tm_vp8, 8, ssse3) PRED8x8L(top_dc, 8, sse2) PRED8x8L(dc, 8, sse2) -PRED8x8L(horizontal, 8, mmxext) -PRED8x8L(horizontal, 8, ssse3) +PRED8x8L(horizontal, 8, sse2) PRED8x8L(vertical, 8, sse2) PRED8x8L(down_left, 8, sse2) PRED8x8L(down_left, 8, ssse3) @@ -163,7 +162,6 @@ av_cold void ff_h264_pred_init_x86(H264PredContext *h, int codec_id, if (bit_depth == 8) { if (EXTERNAL_MMXEXT(cpu_flags)) { - h->pred8x8l [HOR_PRED ] = ff_pred8x8l_horizontal_8_mmxext; h->pred8x8l [HOR_UP_PRED ] = ff_pred8x8l_horizontal_up_8_mmxext; h->pred4x4 [DIAG_DOWN_RIGHT_PRED ] = ff_pred4x4_down_right_8_mmxext; h->pred4x4 [VERT_RIGHT_PRED ] = ff_pred4x4_vertical_right_8_mmxext; @@ -193,6 +191,7 @@ av_cold void ff_h264_pred_init_x86(H264PredContext *h, int codec_id, if (EXTERNAL_SSE2(cpu_flags)) { h->pred16x16[HOR_PRED8x8 ] = ff_pred16x16_horizontal_8_sse2; h->pred16x16[DC_PRED8x8 ] = ff_pred16x16_dc_8_sse2; + h->pred8x8l [HOR_PRED ] = ff_pred8x8l_horizontal_8_sse2; h->pred8x8l [DC_PRED ] = ff_pred8x8l_dc_8_sse2; h->pred8x8l [TOP_DC_PRED ] = ff_pred8x8l_top_dc_8_sse2; h->pred8x8l [VERT_PRED ] = ff_pred8x8l_vertical_8_sse2; @@ -230,7 +229,6 @@ av_cold void ff_h264_pred_init_x86(H264PredContext *h, int codec_id, h->pred16x16[DC_PRED8x8 ] = ff_pred16x16_dc_8_ssse3; if (chroma_format_idc <= 1) h->pred8x8 [HOR_PRED8x8 ] = ff_pred8x8_horizontal_8_ssse3; - h->pred8x8l [HOR_PRED ] = ff_pred8x8l_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; _______________________________________________ ffmpeg-cvslog mailing list -- [email protected] To unsubscribe send an email to [email protected]
