avcodec/h264: add avx 8-bit 4:2:2 chroma h deblock/loop filter

~1.21x faster (68 vs. 56 cycles) compared with mmxext function

libavcodec/x86/h264_deblock.asm

@@ -1163,6 +1163,33 @@ cglobal deblock_h_chroma_8, 5, 7, 8, 0-16, pix_, stride_, alpha_, beta_, tc0_
     STORE_8_ROWS PASS8ROWS(pix_q - 2, r5 - 2, stride_q, r6)
 RET
 
+cglobal deblock_h_chroma422_8, 5, 7, 8, 0-16, pix_, stride_, alpha_, beta_, tc0_,
+    CHROMA_H_START_XMM r5, r6
+    LOAD_8_ROWS PASS8ROWS(pix_q - 2, r5 - 2, stride_q, r6)
+    TRANSPOSE_8x4B_XMM
+    movq [rsp], m0
+    movq [rsp + 8], m3
+    CHROMA_INTER_BODY_XMM 2
+    movq m0, [rsp]
+    movq m3, [rsp + 8]
+    TRANSPOSE_4x8B_XMM
+    STORE_8_ROWS PASS8ROWS(pix_q - 2, r5 - 2, stride_q, r6)
+
+    lea pix_q, [pix_q + 8*stride_q]
+    lea r5,    [r5    + 8*stride_q]
+    add tc0_q,  2
+
+    LOAD_8_ROWS PASS8ROWS(pix_q - 2, r5 - 2, stride_q, r6)
+    TRANSPOSE_8x4B_XMM
+    movq [rsp], m0
+    movq [rsp + 8], m3
+    CHROMA_INTER_BODY_XMM 2
+    movq m0, [rsp]
+    movq m3, [rsp + 8]
+    TRANSPOSE_4x8B_XMM
+    STORE_8_ROWS PASS8ROWS(pix_q - 2, r5 - 2, stride_q, r6)
+RET
+
 %endmacro ; DEBLOCK_CHROMA_XMM
 
 DEBLOCK_CHROMA_XMM avx

libavcodec/x86/h264dsp_init.c

@@ -321,6 +321,8 @@ av_cold void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth,
             c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_8_avx;
             if (chroma_format_idc <= 1) {
                 c->h264_h_loop_filter_chroma       = ff_deblock_h_chroma_8_avx;
+            } else {
+                c->h264_h_loop_filter_chroma       = ff_deblock_h_chroma422_8_avx;
             }
         }
     } else if (bit_depth == 10) {