[FFmpeg-devel] [PATCH 6/6] avcodec/h264: mmx2, sse2, avx 10-bit 4:2:2 h chroma intra deblock/loop filter

[James Darnley]

Yorkfield:
 - mmx2: 1.21x faster (370 vs. 306 cycles)
 - sse2: 1.20x faster (370 vs. 308 cycles)

Skylake:
 - mmx2: 1.39x faster (240 vs. 172 cycles)
 - sse2: 1.74x faster (240 vs. 138 cycles)
 - avx:  1.78x faster (240 vs. 134 cycles)
---
 libavcodec/x86/h264_deblock_10bit.asm | 27 +++++++++++++++++++++++++++
 libavcodec/x86/h264dsp_init.c         |  3 +++
 2 files changed, 30 insertions(+)

diff --git a/libavcodec/x86/h264_deblock_10bit.asm b/libavcodec/x86/h264_deblock_10bit.asm
index fb10bfb..8832210 100644
--- a/libavcodec/x86/h264_deblock_10bit.asm
+++ b/libavcodec/x86/h264_deblock_10bit.asm
@@ -1100,6 +1100,33 @@ cglobal deblock_h_chroma_intra_10, 4, 6, 8, 2*mmsize, pix_, stride_, alpha_, bet
 %endif
 RET
 
+;-----------------------------------------------------------------------------
+; void ff_deblock_h_chroma422_intra_10(uint16_t *pix, int stride, int alpha,
+;                                      int beta)
+;-----------------------------------------------------------------------------
+cglobal deblock_h_chroma422_intra_10, 4, 7, 8, 2*mmsize, pix_, stride_, alpha_, beta_
+    shl alpha_d,  2
+    shl beta_d,   2
+
+    mov r5,       pix_q
+    lea r6,      [3*stride_q]
+    add r5,       r6
+
+    mov r4, -8
+    .loop:
+
+        CHROMA_H_LOAD r5, r6, [rsp], [rsp + mmsize]
+        LOAD_AB     m4, m5, alpha_d, beta_d
+        LOAD_MASK   m0, m1, m2, m3, m4, m5, m7, m6, m4
+        CHROMA_DEBLOCK_P0_Q0_INTRA m1, m2, m0, m3, m7, m5, m6
+        CHROMA_H_STORE r5, r6, [rsp], [rsp + mmsize]
+
+        lea pix_q, [pix_q + (mmsize/2)*stride_q]
+        lea r5,    [r5 +    (mmsize/2)*stride_q]
+        add r4, (mmsize/4)
+    jl .loop
+RET
+
 %endmacro
 
 %if ARCH_X86_64 == 0
diff --git a/libavcodec/x86/h264dsp_init.c b/libavcodec/x86/h264dsp_init.c
index 395b8c3..8154cd9 100644
--- a/libavcodec/x86/h264dsp_init.c
+++ b/libavcodec/x86/h264dsp_init.c
@@ -318,6 +318,7 @@ av_cold void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth,
                 c->h264_h_loop_filter_chroma_intra = ff_deblock_h_chroma_intra_10_mmxext;
             } else {
                 c->h264_h_loop_filter_chroma = ff_deblock_h_chroma422_10_mmxext;
+                c->h264_h_loop_filter_chroma_intra = ff_deblock_h_chroma422_intra_10_mmxext;
             }
             c->h264_v_loop_filter_luma         = ff_deblock_v_luma_10_mmxext;
             c->h264_h_loop_filter_luma         = ff_deblock_h_luma_10_mmxext;
@@ -357,6 +358,7 @@ av_cold void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth,
                 c->h264_h_loop_filter_chroma_intra = ff_deblock_h_chroma_intra_10_sse2;
             } else {
                 c->h264_h_loop_filter_chroma = ff_deblock_h_chroma422_10_sse2;
+                c->h264_h_loop_filter_chroma_intra = ff_deblock_h_chroma422_intra_10_sse2;
             }
 #if HAVE_ALIGNED_STACK
             c->h264_v_loop_filter_luma       = ff_deblock_v_luma_10_sse2;
@@ -398,6 +400,7 @@ av_cold void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth,
                 c->h264_h_loop_filter_chroma_intra = ff_deblock_h_chroma_intra_10_avx;
             } else {
                 c->h264_h_loop_filter_chroma = ff_deblock_h_chroma422_10_avx;
+                c->h264_h_loop_filter_chroma_intra = ff_deblock_h_chroma422_intra_10_avx;
             }
 #if HAVE_ALIGNED_STACK
             c->h264_v_loop_filter_luma         = ff_deblock_v_luma_10_avx;
-- 
2.10.2