H.264: Add optimizations to predict x86 assembly.

Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>

libavcodec/x86/h264_intrapred.asm

@@ -2611,12 +2611,11 @@ cglobal pred4x4_down_left_mmxext, 3,3
     punpckldq m1, [r1]
     movq      m2, m1
     movq      m3, m1
-    movq      m4, m1
     psllq     m1, 8
     pxor      m2, m1
     psrlq     m2, 8
-    pxor      m3, m2
+    pxor      m2, m3
-    PRED4x4_LOWPASS m0, m1, m3, m4, m5
+    PRED4x4_LOWPASS m0, m1, m2, m3, m4
     lea       r1, [r0+r2*2]
     psrlq     m0, 8
     movd      [r0+r2*1], m0

libavcodec/x86/h264_intrapred_init.c

@@ -45,7 +45,6 @@ void ff_pred8x8_ ## TYPE ## _ ## DEPTH ## _ ## OPT (uint8_t *src, int stride);
 
 PRED8x8(dc, 10, mmxext)
 PRED8x8(dc, 10, sse2)
-PRED8x8(top_dc, 10, mmxext)
 PRED8x8(top_dc, 10, sse2)
 PRED8x8(plane, 10, sse2)
 PRED8x8(vertical, 10, sse2)
@@ -55,23 +54,28 @@ PRED8x8(horizontal, 10, sse2)
 void ff_pred8x8l_ ## TYPE ## _ ## DEPTH ## _ ## OPT (uint8_t *src, int has_topleft, int has_topright, int stride);
 
 PRED8x8L(dc, 10, sse2)
-PRED8x8L(dc, 10, ssse3)
+PRED8x8L(dc, 10, avx)
 PRED8x8L(128_dc, 10, mmxext)
 PRED8x8L(128_dc, 10, sse2)
 PRED8x8L(top_dc, 10, sse2)
-PRED8x8L(top_dc, 10, ssse3)
+PRED8x8L(top_dc, 10, avx)
 PRED8x8L(vertical, 10, sse2)
-PRED8x8L(vertical, 10, ssse3)
+PRED8x8L(vertical, 10, avx)
 PRED8x8L(horizontal, 10, sse2)
 PRED8x8L(horizontal, 10, ssse3)
+PRED8x8L(horizontal, 10, avx)
 PRED8x8L(down_left, 10, sse2)
 PRED8x8L(down_left, 10, ssse3)
+PRED8x8L(down_left, 10, avx)
 PRED8x8L(down_right, 10, sse2)
 PRED8x8L(down_right, 10, ssse3)
+PRED8x8L(down_right, 10, avx)
 PRED8x8L(vertical_right, 10, sse2)
 PRED8x8L(vertical_right, 10, ssse3)
+PRED8x8L(vertical_right, 10, avx)
 PRED8x8L(horizontal_up, 10, sse2)
 PRED8x8L(horizontal_up, 10, ssse3)
+PRED8x8L(horizontal_up, 10, avx)
 
 #define PRED16x16(TYPE, DEPTH, OPT)\
 void ff_pred16x16_ ## TYPE ## _ ## DEPTH ## _ ## OPT (uint8_t *src, int stride);
@@ -298,7 +302,6 @@ void ff_h264_pred_init_x86(H264PredContext *h, int codec_id, const int bit_depth
             h->pred4x4[HOR_UP_PRED         ] = ff_pred4x4_horizontal_up_10_mmxext;
 
             h->pred8x8[DC_PRED8x8          ] = ff_pred8x8_dc_10_mmxext;
-            h->pred8x8[TOP_DC_PRED8x8      ] = ff_pred8x8_top_dc_10_mmxext;
 
             h->pred8x8l[DC_128_PRED        ] = ff_pred8x8l_128_dc_10_mmxext;
 
@@ -344,18 +347,28 @@ void ff_h264_pred_init_x86(H264PredContext *h, int codec_id, const int bit_depth
             h->pred4x4[VERT_RIGHT_PRED     ] = ff_pred4x4_vertical_right_10_ssse3;
             h->pred4x4[HOR_DOWN_PRED       ] = ff_pred4x4_horizontal_down_10_ssse3;
 
-            h->pred8x8l[VERT_PRED           ] = ff_pred8x8l_vertical_10_ssse3;
             h->pred8x8l[HOR_PRED            ] = ff_pred8x8l_horizontal_10_ssse3;
-            h->pred8x8l[DC_PRED             ] = ff_pred8x8l_dc_10_ssse3;
-            h->pred8x8l[TOP_DC_PRED         ] = ff_pred8x8l_top_dc_10_ssse3;
             h->pred8x8l[DIAG_DOWN_LEFT_PRED ] = ff_pred8x8l_down_left_10_ssse3;
+            h->pred8x8l[DIAG_DOWN_RIGHT_PRED] = ff_pred8x8l_down_right_10_ssse3;
+            h->pred8x8l[VERT_RIGHT_PRED     ] = ff_pred8x8l_vertical_right_10_ssse3;
+            h->pred8x8l[HOR_UP_PRED         ] = ff_pred8x8l_horizontal_up_10_ssse3;
         }
 #if HAVE_AVX
         if (mm_flags & AV_CPU_FLAG_AVX) {
             h->pred4x4[DIAG_DOWN_LEFT_PRED ] = ff_pred4x4_down_left_10_avx;
             h->pred4x4[DIAG_DOWN_RIGHT_PRED] = ff_pred4x4_down_right_10_avx;
+            h->pred4x4[VERT_LEFT_PRED      ] = ff_pred4x4_vertical_left_10_avx;
             h->pred4x4[VERT_RIGHT_PRED     ] = ff_pred4x4_vertical_right_10_avx;
             h->pred4x4[HOR_DOWN_PRED       ] = ff_pred4x4_horizontal_down_10_avx;
+
+            h->pred8x8l[VERT_PRED           ] = ff_pred8x8l_vertical_10_avx;
+            h->pred8x8l[HOR_PRED            ] = ff_pred8x8l_horizontal_10_avx;
+            h->pred8x8l[DC_PRED             ] = ff_pred8x8l_dc_10_avx;
+            h->pred8x8l[TOP_DC_PRED         ] = ff_pred8x8l_top_dc_10_avx;
+            h->pred8x8l[DIAG_DOWN_RIGHT_PRED] = ff_pred8x8l_down_right_10_avx;
+            h->pred8x8l[DIAG_DOWN_LEFT_PRED ] = ff_pred8x8l_down_left_10_avx;
+            h->pred8x8l[VERT_RIGHT_PRED     ] = ff_pred8x8l_vertical_right_10_avx;
+            h->pred8x8l[HOR_UP_PRED         ] = ff_pred8x8l_horizontal_up_10_avx;
         }
 #endif /* HAVE_AVX */
     }