dxva2: fix surface selection when compiled with both d3d11va and dxva2

[ffmpeg] / libavcodec / h264pred_template.c

diff --git a/libavcodec/h264pred_template.c b/libavcodec/h264pred_template.c
index 750e82c12abd952c99aaa96d74f017bf12f4d7e7..8492b2b0a2f666c81f054d4a6d3e4870e8480363 100644 (file)
--- a/libavcodec/h264pred_template.c
+++ b/libavcodec/h264pred_template.c
@@ -21,15 +21,19 @@
 
 /**
  * @file
- * H.264 / AVC / MPEG4 part10 prediction functions.
+ * H.264 / AVC / MPEG-4 part10 prediction functions.
  * @author Michael Niedermayer <michaelni@gmx.at>
  */
 
+#include "libavutil/intreadwrite.h"
+
 #include "mathops.h"
 
 #include "bit_depth_template.c"
 
-static void FUNCC(pred4x4_vertical)(uint8_t *_src, const uint8_t *topright, int _stride){
+static void FUNCC(pred4x4_vertical)(uint8_t *_src, const uint8_t *topright,
+                                    ptrdiff_t _stride)
+{
     pixel *src = (pixel*)_src;
     int stride = _stride/sizeof(pixel);
     const pixel4 a= AV_RN4PA(src-stride);
@@ -40,7 +44,9 @@ static void FUNCC(pred4x4_vertical)(uint8_t *_src, const uint8_t *topright, int
     AV_WN4PA(src+3*stride, a);
 }
 
-static void FUNCC(pred4x4_horizontal)(uint8_t *_src, const uint8_t *topright, int _stride){
+static void FUNCC(pred4x4_horizontal)(uint8_t *_src, const uint8_t *topright,
+                                      ptrdiff_t _stride)
+{
     pixel *src = (pixel*)_src;
     int stride = _stride/sizeof(pixel);
     AV_WN4PA(src+0*stride, PIXEL_SPLAT_X4(src[-1+0*stride]));
@@ -49,7 +55,9 @@ static void FUNCC(pred4x4_horizontal)(uint8_t *_src, const uint8_t *topright, in
     AV_WN4PA(src+3*stride, PIXEL_SPLAT_X4(src[-1+3*stride]));
 }
 
-static void FUNCC(pred4x4_dc)(uint8_t *_src, const uint8_t *topright, int _stride){
+static void FUNCC(pred4x4_dc)(uint8_t *_src, const uint8_t *topright,
+                              ptrdiff_t _stride)
+{
     pixel *src = (pixel*)_src;
     int stride = _stride/sizeof(pixel);
     const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
@@ -62,7 +70,9 @@ static void FUNCC(pred4x4_dc)(uint8_t *_src, const uint8_t *topright, int _strid
     AV_WN4PA(src+3*stride, a);
 }
 
-static void FUNCC(pred4x4_left_dc)(uint8_t *_src, const uint8_t *topright, int _stride){
+static void FUNCC(pred4x4_left_dc)(uint8_t *_src, const uint8_t *topright,
+                                   ptrdiff_t _stride)
+{
     pixel *src = (pixel*)_src;
     int stride = _stride/sizeof(pixel);
     const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
@@ -74,7 +84,9 @@ static void FUNCC(pred4x4_left_dc)(uint8_t *_src, const uint8_t *topright, int _
     AV_WN4PA(src+3*stride, a);
 }
 
-static void FUNCC(pred4x4_top_dc)(uint8_t *_src, const uint8_t *topright, int _stride){
+static void FUNCC(pred4x4_top_dc)(uint8_t *_src, const uint8_t *topright,
+                                  ptrdiff_t _stride)
+{
     pixel *src = (pixel*)_src;
     int stride = _stride/sizeof(pixel);
     const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
@@ -86,7 +98,9 @@ static void FUNCC(pred4x4_top_dc)(uint8_t *_src, const uint8_t *topright, int _s
     AV_WN4PA(src+3*stride, a);
 }
 
-static void FUNCC(pred4x4_128_dc)(uint8_t *_src, const uint8_t *topright, int _stride){
+static void FUNCC(pred4x4_128_dc)(uint8_t *_src, const uint8_t *topright,
+                                  ptrdiff_t _stride)
+{
     pixel *src = (pixel*)_src;
     int stride = _stride/sizeof(pixel);
     const pixel4 a = PIXEL_SPLAT_X4(1<<(BIT_DEPTH-1));
@@ -97,7 +111,9 @@ static void FUNCC(pred4x4_128_dc)(uint8_t *_src, const uint8_t *topright, int _s
     AV_WN4PA(src+3*stride, a);
 }
 
-static void FUNCC(pred4x4_127_dc)(uint8_t *_src, const uint8_t *topright, int _stride){
+static void FUNCC(pred4x4_127_dc)(uint8_t *_src, const uint8_t *topright,
+                                  ptrdiff_t _stride)
+{
     pixel *src = (pixel*)_src;
     int stride = _stride/sizeof(pixel);
     const pixel4 a = PIXEL_SPLAT_X4((1<<(BIT_DEPTH-1))-1);
@@ -108,7 +124,9 @@ static void FUNCC(pred4x4_127_dc)(uint8_t *_src, const uint8_t *topright, int _s
     AV_WN4PA(src+3*stride, a);
 }
 
-static void FUNCC(pred4x4_129_dc)(uint8_t *_src, const uint8_t *topright, int _stride){
+static void FUNCC(pred4x4_129_dc)(uint8_t *_src, const uint8_t *topright,
+                                  ptrdiff_t _stride)
+{
     pixel *src = (pixel*)_src;
     int stride = _stride/sizeof(pixel);
     const pixel4 a = PIXEL_SPLAT_X4((1<<(BIT_DEPTH-1))+1);
@@ -144,7 +162,9 @@ static void FUNCC(pred4x4_129_dc)(uint8_t *_src, const uint8_t *topright, int _s
     const unsigned av_unused t2 = src[ 2-1*stride];\
     const unsigned av_unused t3 = src[ 3-1*stride];\
 
-static void FUNCC(pred4x4_down_right)(uint8_t *_src, const uint8_t *topright, int _stride){
+static void FUNCC(pred4x4_down_right)(uint8_t *_src, const uint8_t *topright,
+                                      ptrdiff_t _stride)
+{
     pixel *src = (pixel*)_src;
     int stride = _stride/sizeof(pixel);
     const int lt= src[-1-1*stride];
@@ -169,7 +189,9 @@ static void FUNCC(pred4x4_down_right)(uint8_t *_src, const uint8_t *topright, in
     src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
 }
 
-static void FUNCC(pred4x4_down_left)(uint8_t *_src, const uint8_t *_topright, int _stride){
+static void FUNCC(pred4x4_down_left)(uint8_t *_src, const uint8_t *_topright,
+                                     ptrdiff_t _stride)
+{
     pixel *src = (pixel*)_src;
     const pixel *topright = (const pixel*)_topright;
     int stride = _stride/sizeof(pixel);
@@ -195,7 +217,10 @@ static void FUNCC(pred4x4_down_left)(uint8_t *_src, const uint8_t *_topright, in
     src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
 }
 
-static void FUNCC(pred4x4_vertical_right)(uint8_t *_src, const uint8_t *topright, int _stride){
+static void FUNCC(pred4x4_vertical_right)(uint8_t *_src,
+                                          const uint8_t *topright,
+                                          ptrdiff_t _stride)
+{
     pixel *src = (pixel*)_src;
     int stride = _stride/sizeof(pixel);
     const int lt= src[-1-1*stride];
@@ -220,7 +245,10 @@ static void FUNCC(pred4x4_vertical_right)(uint8_t *_src, const uint8_t *topright
     src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
 }
 
-static void FUNCC(pred4x4_vertical_left)(uint8_t *_src, const uint8_t *_topright, int _stride){
+static void FUNCC(pred4x4_vertical_left)(uint8_t *_src,
+                                         const uint8_t *_topright,
+                                         ptrdiff_t _stride)
+{
     pixel *src = (pixel*)_src;
     const pixel *topright = (const pixel*)_topright;
     int stride = _stride/sizeof(pixel);
@@ -245,7 +273,9 @@ static void FUNCC(pred4x4_vertical_left)(uint8_t *_src, const uint8_t *_topright
     src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
 }
 
-static void FUNCC(pred4x4_horizontal_up)(uint8_t *_src, const uint8_t *topright, int _stride){
+static void FUNCC(pred4x4_horizontal_up)(uint8_t *_src, const uint8_t *topright,
+                                         ptrdiff_t _stride)
+{
     pixel *src = (pixel*)_src;
     int stride = _stride/sizeof(pixel);
     LOAD_LEFT_EDGE
@@ -268,7 +298,10 @@ static void FUNCC(pred4x4_horizontal_up)(uint8_t *_src, const uint8_t *topright,
     src[3+3*stride]=l3;
 }
 
-static void FUNCC(pred4x4_horizontal_down)(uint8_t *_src, const uint8_t *topright, int _stride){
+static void FUNCC(pred4x4_horizontal_down)(uint8_t *_src,
+                                           const uint8_t *topright,
+                                           ptrdiff_t _stride)
+{
     pixel *src = (pixel*)_src;
     int stride = _stride/sizeof(pixel);
     const int lt= src[-1-1*stride];
@@ -293,7 +326,8 @@ static void FUNCC(pred4x4_horizontal_down)(uint8_t *_src, const uint8_t *toprigh
     src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
 }
 
-static void FUNCC(pred16x16_vertical)(uint8_t *_src, int _stride){
+static void FUNCC(pred16x16_vertical)(uint8_t *_src, ptrdiff_t _stride)
+{
     int i;
     pixel *src = (pixel*)_src;
     int stride = _stride/sizeof(pixel);
@@ -310,7 +344,8 @@ static void FUNCC(pred16x16_vertical)(uint8_t *_src, int _stride){
     }
 }
 
-static void FUNCC(pred16x16_horizontal)(uint8_t *_src, int stride){
+static void FUNCC(pred16x16_horizontal)(uint8_t *_src, ptrdiff_t stride)
+{
     int i;
     pixel *src = (pixel*)_src;
     stride /= sizeof(pixel);
@@ -334,7 +369,8 @@ static void FUNCC(pred16x16_horizontal)(uint8_t *_src, int stride){
         src += stride;\
     }
 
-static void FUNCC(pred16x16_dc)(uint8_t *_src, int stride){
+static void FUNCC(pred16x16_dc)(uint8_t *_src, ptrdiff_t stride)
+{
     int i, dc=0;
     pixel *src = (pixel*)_src;
     pixel4 dcsplat;
@@ -352,7 +388,8 @@ static void FUNCC(pred16x16_dc)(uint8_t *_src, int stride){
     PREDICT_16x16_DC(dcsplat);
 }
 
-static void FUNCC(pred16x16_left_dc)(uint8_t *_src, int stride){
+static void FUNCC(pred16x16_left_dc)(uint8_t *_src, ptrdiff_t stride)
+{
     int i, dc=0;
     pixel *src = (pixel*)_src;
     pixel4 dcsplat;
@@ -366,7 +403,8 @@ static void FUNCC(pred16x16_left_dc)(uint8_t *_src, int stride){
     PREDICT_16x16_DC(dcsplat);
 }
 
-static void FUNCC(pred16x16_top_dc)(uint8_t *_src, int stride){
+static void FUNCC(pred16x16_top_dc)(uint8_t *_src, ptrdiff_t stride)
+{
     int i, dc=0;
     pixel *src = (pixel*)_src;
     pixel4 dcsplat;
@@ -381,18 +419,23 @@ static void FUNCC(pred16x16_top_dc)(uint8_t *_src, int stride){
 }
 
 #define PRED16x16_X(n, v) \
-static void FUNCC(pred16x16_##n##_dc)(uint8_t *_src, int stride){\
+static void FUNCC(pred16x16_##n##_dc)(uint8_t *_src, ptrdiff_t stride)\
+{\
     int i;\
     pixel *src = (pixel*)_src;\
     stride /= sizeof(pixel);\
     PREDICT_16x16_DC(PIXEL_SPLAT_X4(v));\
 }
 
-PRED16x16_X(127, (1<<(BIT_DEPTH-1))-1);
-PRED16x16_X(128, (1<<(BIT_DEPTH-1))+0);
-PRED16x16_X(129, (1<<(BIT_DEPTH-1))+1);
+PRED16x16_X(127, (1<<(BIT_DEPTH-1))-1)
+PRED16x16_X(128, (1<<(BIT_DEPTH-1))+0)
+PRED16x16_X(129, (1<<(BIT_DEPTH-1))+1)
 
-static inline void FUNCC(pred16x16_plane_compat)(uint8_t *_src, int _stride, const int svq3, const int rv40){
+static inline void FUNCC(pred16x16_plane_compat)(uint8_t *_src,
+                                                 ptrdiff_t _stride,
+                                                 const int svq3,
+                                                 const int rv40)
+{
   int i, j, k;
   int a;
   INIT_CLIP
@@ -437,11 +480,13 @@ static inline void FUNCC(pred16x16_plane_compat)(uint8_t *_src, int _stride, con
   }
 }
 
-static void FUNCC(pred16x16_plane)(uint8_t *src, int stride){
+static void FUNCC(pred16x16_plane)(uint8_t *src, ptrdiff_t stride)
+{
     FUNCC(pred16x16_plane_compat)(src, stride, 0, 0);
 }
 
-static void FUNCC(pred8x8_vertical)(uint8_t *_src, int _stride){
+static void FUNCC(pred8x8_vertical)(uint8_t *_src, ptrdiff_t _stride)
+{
     int i;
     pixel *src = (pixel*)_src;
     int stride = _stride/sizeof(pixel);
@@ -454,7 +499,22 @@ static void FUNCC(pred8x8_vertical)(uint8_t *_src, int _stride){
     }
 }
 
-static void FUNCC(pred8x8_horizontal)(uint8_t *_src, int stride){
+static void FUNCC(pred8x16_vertical)(uint8_t *_src, ptrdiff_t _stride)
+{
+    int i;
+    pixel *src = (pixel*)_src;
+    int stride = _stride>>(sizeof(pixel)-1);
+    const pixel4 a= AV_RN4PA(((pixel4*)(src-stride))+0);
+    const pixel4 b= AV_RN4PA(((pixel4*)(src-stride))+1);
+
+    for(i=0; i<16; i++){
+        AV_WN4PA(((pixel4*)(src+i*stride))+0, a);
+        AV_WN4PA(((pixel4*)(src+i*stride))+1, b);
+    }
+}
+
+static void FUNCC(pred8x8_horizontal)(uint8_t *_src, ptrdiff_t stride)
+{
     int i;
     pixel *src = (pixel*)_src;
     stride /= sizeof(pixel);
@@ -466,8 +526,21 @@ static void FUNCC(pred8x8_horizontal)(uint8_t *_src, int stride){
     }
 }
 
+static void FUNCC(pred8x16_horizontal)(uint8_t *_src, ptrdiff_t stride)
+{
+    int i;
+    pixel *src = (pixel*)_src;
+    stride >>= sizeof(pixel)-1;
+    for(i=0; i<16; i++){
+        const pixel4 a = PIXEL_SPLAT_X4(src[-1+i*stride]);
+        AV_WN4PA(((pixel4*)(src+i*stride))+0, a);
+        AV_WN4PA(((pixel4*)(src+i*stride))+1, a);
+    }
+}
+
 #define PRED8x8_X(n, v)\
-static void FUNCC(pred8x8_##n##_dc)(uint8_t *_src, int stride){\
+static void FUNCC(pred8x8_##n##_dc)(uint8_t *_src, ptrdiff_t stride)\
+{\
     int i;\
     const pixel4 a = PIXEL_SPLAT_X4(v);\
     pixel *src = (pixel*)_src;\
@@ -478,11 +551,18 @@ static void FUNCC(pred8x8_##n##_dc)(uint8_t *_src, int stride){\
     }\
 }
 
-PRED8x8_X(127, (1<<(BIT_DEPTH-1))-1);
-PRED8x8_X(128, (1<<(BIT_DEPTH-1))+0);
-PRED8x8_X(129, (1<<(BIT_DEPTH-1))+1);
+PRED8x8_X(127, (1<<(BIT_DEPTH-1))-1)
+PRED8x8_X(128, (1<<(BIT_DEPTH-1))+0)
+PRED8x8_X(129, (1<<(BIT_DEPTH-1))+1)
+
+static void FUNCC(pred8x16_128_dc)(uint8_t *_src, ptrdiff_t stride)
+{
+    FUNCC(pred8x8_128_dc)(_src, stride);
+    FUNCC(pred8x8_128_dc)(_src+8*stride, stride);
+}
 
-static void FUNCC(pred8x8_left_dc)(uint8_t *_src, int stride){
+static void FUNCC(pred8x8_left_dc)(uint8_t *_src, ptrdiff_t stride)
+{
     int i;
     int dc0, dc2;
     pixel4 dc0splat, dc2splat;
@@ -507,7 +587,14 @@ static void FUNCC(pred8x8_left_dc)(uint8_t *_src, int stride){
     }
 }
 
-static void FUNCC(pred8x8_top_dc)(uint8_t *_src, int stride){
+static void FUNCC(pred8x16_left_dc)(uint8_t *_src, ptrdiff_t stride)
+{
+    FUNCC(pred8x8_left_dc)(_src, stride);
+    FUNCC(pred8x8_left_dc)(_src+8*stride, stride);
+}
+
+static void FUNCC(pred8x8_top_dc)(uint8_t *_src, ptrdiff_t stride)
+{
     int i;
     int dc0, dc1;
     pixel4 dc0splat, dc1splat;
@@ -532,7 +619,30 @@ static void FUNCC(pred8x8_top_dc)(uint8_t *_src, int stride){
     }
 }
 
-static void FUNCC(pred8x8_dc)(uint8_t *_src, int stride){
+static void FUNCC(pred8x16_top_dc)(uint8_t *_src, ptrdiff_t stride)
+{
+    int i;
+    int dc0, dc1;
+    pixel4 dc0splat, dc1splat;
+    pixel *src = (pixel*)_src;
+    stride >>= sizeof(pixel)-1;
+
+    dc0=dc1=0;
+    for(i=0;i<4; i++){
+        dc0+= src[i-stride];
+        dc1+= src[4+i-stride];
+    }
+    dc0splat = PIXEL_SPLAT_X4((dc0 + 2)>>2);
+    dc1splat = PIXEL_SPLAT_X4((dc1 + 2)>>2);
+
+    for(i=0; i<16; i++){
+        AV_WN4PA(((pixel4*)(src+i*stride))+0, dc0splat);
+        AV_WN4PA(((pixel4*)(src+i*stride))+1, dc1splat);
+    }
+}
+
+static void FUNCC(pred8x8_dc)(uint8_t *_src, ptrdiff_t stride)
+{
     int i;
     int dc0, dc1, dc2;
     pixel4 dc0splat, dc1splat, dc2splat, dc3splat;
@@ -560,30 +670,103 @@ static void FUNCC(pred8x8_dc)(uint8_t *_src, int stride){
     }
 }
 
-//the following 4 function should not be optimized!
-static void FUNC(pred8x8_mad_cow_dc_l0t)(uint8_t *src, int stride){
+static void FUNCC(pred8x16_dc)(uint8_t *_src, ptrdiff_t stride)
+{
+    int i;
+    int dc0, dc1, dc2, dc3, dc4;
+    pixel4 dc0splat, dc1splat, dc2splat, dc3splat, dc4splat, dc5splat, dc6splat, dc7splat;
+    pixel *src = (pixel*)_src;
+    stride >>= sizeof(pixel)-1;
+
+    dc0=dc1=dc2=dc3=dc4=0;
+    for(i=0;i<4; i++){
+        dc0+= src[-1+i*stride] + src[i-stride];
+        dc1+= src[4+i-stride];
+        dc2+= src[-1+(i+4)*stride];
+        dc3+= src[-1+(i+8)*stride];
+        dc4+= src[-1+(i+12)*stride];
+    }
+    dc0splat = PIXEL_SPLAT_X4((dc0 + 4)>>3);
+    dc1splat = PIXEL_SPLAT_X4((dc1 + 2)>>2);
+    dc2splat = PIXEL_SPLAT_X4((dc2 + 2)>>2);
+    dc3splat = PIXEL_SPLAT_X4((dc1 + dc2 + 4)>>3);
+    dc4splat = PIXEL_SPLAT_X4((dc3 + 2)>>2);
+    dc5splat = PIXEL_SPLAT_X4((dc1 + dc3 + 4)>>3);
+    dc6splat = PIXEL_SPLAT_X4((dc4 + 2)>>2);
+    dc7splat = PIXEL_SPLAT_X4((dc1 + dc4 + 4)>>3);
+
+    for(i=0; i<4; i++){
+        AV_WN4PA(((pixel4*)(src+i*stride))+0, dc0splat);
+        AV_WN4PA(((pixel4*)(src+i*stride))+1, dc1splat);
+    }
+    for(i=4; i<8; i++){
+        AV_WN4PA(((pixel4*)(src+i*stride))+0, dc2splat);
+        AV_WN4PA(((pixel4*)(src+i*stride))+1, dc3splat);
+    }
+    for(i=8; i<12; i++){
+        AV_WN4PA(((pixel4*)(src+i*stride))+0, dc4splat);
+        AV_WN4PA(((pixel4*)(src+i*stride))+1, dc5splat);
+    }
+    for(i=12; i<16; i++){
+        AV_WN4PA(((pixel4*)(src+i*stride))+0, dc6splat);
+        AV_WN4PA(((pixel4*)(src+i*stride))+1, dc7splat);
+    }
+}
+
+static void FUNC(pred8x8_mad_cow_dc_l0t)(uint8_t *src, ptrdiff_t stride)
+{
     FUNCC(pred8x8_top_dc)(src, stride);
     FUNCC(pred4x4_dc)(src, NULL, stride);
 }
 
-static void FUNC(pred8x8_mad_cow_dc_0lt)(uint8_t *src, int stride){
+static void FUNC(pred8x16_mad_cow_dc_l0t)(uint8_t *src, ptrdiff_t stride)
+{
+    FUNCC(pred8x16_top_dc)(src, stride);
+    FUNCC(pred4x4_dc)(src, NULL, stride);
+}
+
+static void FUNC(pred8x8_mad_cow_dc_0lt)(uint8_t *src, ptrdiff_t stride)
+{
     FUNCC(pred8x8_dc)(src, stride);
     FUNCC(pred4x4_top_dc)(src, NULL, stride);
 }
 
-static void FUNC(pred8x8_mad_cow_dc_l00)(uint8_t *src, int stride){
+static void FUNC(pred8x16_mad_cow_dc_0lt)(uint8_t *src, ptrdiff_t stride)
+{
+    FUNCC(pred8x16_dc)(src, stride);
+    FUNCC(pred4x4_top_dc)(src, NULL, stride);
+}
+
+static void FUNC(pred8x8_mad_cow_dc_l00)(uint8_t *src, ptrdiff_t stride)
+{
     FUNCC(pred8x8_left_dc)(src, stride);
     FUNCC(pred4x4_128_dc)(src + 4*stride                  , NULL, stride);
     FUNCC(pred4x4_128_dc)(src + 4*stride + 4*sizeof(pixel), NULL, stride);
 }
 
-static void FUNC(pred8x8_mad_cow_dc_0l0)(uint8_t *src, int stride){
+static void FUNC(pred8x16_mad_cow_dc_l00)(uint8_t *src, ptrdiff_t stride)
+{
+    FUNCC(pred8x16_left_dc)(src, stride);
+    FUNCC(pred4x4_128_dc)(src + 4*stride                  , NULL, stride);
+    FUNCC(pred4x4_128_dc)(src + 4*stride + 4*sizeof(pixel), NULL, stride);
+}
+
+static void FUNC(pred8x8_mad_cow_dc_0l0)(uint8_t *src, ptrdiff_t stride)
+{
     FUNCC(pred8x8_left_dc)(src, stride);
     FUNCC(pred4x4_128_dc)(src                  , NULL, stride);
     FUNCC(pred4x4_128_dc)(src + 4*sizeof(pixel), NULL, stride);
 }
 
-static void FUNCC(pred8x8_plane)(uint8_t *_src, int _stride){
+static void FUNC(pred8x16_mad_cow_dc_0l0)(uint8_t *src, ptrdiff_t stride)
+{
+    FUNCC(pred8x16_left_dc)(src, stride);
+    FUNCC(pred4x4_128_dc)(src                  , NULL, stride);
+    FUNCC(pred4x4_128_dc)(src + 4*sizeof(pixel), NULL, stride);
+}
+
+static void FUNCC(pred8x8_plane)(uint8_t *_src, ptrdiff_t _stride)
+{
   int j, k;
   int a;
   INIT_CLIP
@@ -618,6 +801,48 @@ static void FUNCC(pred8x8_plane)(uint8_t *_src, int _stride){
   }
 }
 
+static void FUNCC(pred8x16_plane)(uint8_t *_src, ptrdiff_t _stride)
+{
+  int j, k;
+  int a;
+  INIT_CLIP
+  pixel *src = (pixel*)_src;
+  int stride = _stride>>(sizeof(pixel)-1);
+  const pixel * const src0 = src +3-stride;
+  const pixel *       src1 = src +8*stride-1;
+  const pixel *       src2 = src1-2*stride;    // == src+6*stride-1;
+  int H = src0[1] - src0[-1];
+  int V = src1[0] - src2[ 0];
+
+  for (k = 2; k <= 4; ++k) {
+      src1 += stride; src2 -= stride;
+      H += k*(src0[k] - src0[-k]);
+      V += k*(src1[0] - src2[ 0]);
+  }
+  for (; k <= 8; ++k) {
+      src1 += stride; src2 -= stride;
+      V += k*(src1[0] - src2[0]);
+  }
+
+  H = (17*H+16) >> 5;
+  V = (5*V+32) >> 6;
+
+  a = 16*(src1[0] + src2[8] + 1) - 7*V - 3*H;
+  for(j=16; j>0; --j) {
+    int b = a;
+    a += V;
+    src[0] = CLIP((b    ) >> 5);
+    src[1] = CLIP((b+  H) >> 5);
+    src[2] = CLIP((b+2*H) >> 5);
+    src[3] = CLIP((b+3*H) >> 5);
+    src[4] = CLIP((b+4*H) >> 5);
+    src[5] = CLIP((b+5*H) >> 5);
+    src[6] = CLIP((b+6*H) >> 5);
+    src[7] = CLIP((b+7*H) >> 5);
+    src += stride;
+  }
+}
+
 #define SRC(x,y) src[(x)+(y)*stride]
 #define PL(y) \
     const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2;
@@ -656,14 +881,16 @@ static void FUNCC(pred8x8_plane)(uint8_t *_src, int _stride){
         src += stride; \
     }
 
-static void FUNCC(pred8x8l_128_dc)(uint8_t *_src, int has_topleft, int has_topright, int _stride)
+static void FUNCC(pred8x8l_128_dc)(uint8_t *_src, int has_topleft,
+                                   int has_topright, ptrdiff_t _stride)
 {
     pixel *src = (pixel*)_src;
     int stride = _stride/sizeof(pixel);
 
     PREDICT_8x8_DC(PIXEL_SPLAT_X4(1<<(BIT_DEPTH-1)));
 }
-static void FUNCC(pred8x8l_left_dc)(uint8_t *_src, int has_topleft, int has_topright, int _stride)
+static void FUNCC(pred8x8l_left_dc)(uint8_t *_src, int has_topleft,
+                                    int has_topright, ptrdiff_t _stride)
 {
     pixel *src = (pixel*)_src;
     int stride = _stride/sizeof(pixel);
@@ -672,7 +899,8 @@ static void FUNCC(pred8x8l_left_dc)(uint8_t *_src, int has_topleft, int has_topr
     const pixel4 dc = PIXEL_SPLAT_X4((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3);
     PREDICT_8x8_DC(dc);
 }
-static void FUNCC(pred8x8l_top_dc)(uint8_t *_src, int has_topleft, int has_topright, int _stride)
+static void FUNCC(pred8x8l_top_dc)(uint8_t *_src, int has_topleft,
+                                   int has_topright, ptrdiff_t _stride)
 {
     pixel *src = (pixel*)_src;
     int stride = _stride/sizeof(pixel);
@@ -681,7 +909,8 @@ static void FUNCC(pred8x8l_top_dc)(uint8_t *_src, int has_topleft, int has_topri
     const pixel4 dc = PIXEL_SPLAT_X4((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3);
     PREDICT_8x8_DC(dc);
 }
-static void FUNCC(pred8x8l_dc)(uint8_t *_src, int has_topleft, int has_topright, int _stride)
+static void FUNCC(pred8x8l_dc)(uint8_t *_src, int has_topleft,
+                               int has_topright, ptrdiff_t _stride)
 {
     pixel *src = (pixel*)_src;
     int stride = _stride/sizeof(pixel);
@@ -692,7 +921,8 @@ static void FUNCC(pred8x8l_dc)(uint8_t *_src, int has_topleft, int has_topright,
                                      +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4);
     PREDICT_8x8_DC(dc);
 }
-static void FUNCC(pred8x8l_horizontal)(uint8_t *_src, int has_topleft, int has_topright, int _stride)
+static void FUNCC(pred8x8l_horizontal)(uint8_t *_src, int has_topleft,
+                                       int has_topright, ptrdiff_t _stride)
 {
     pixel *src = (pixel*)_src;
     int stride = _stride/sizeof(pixel);
@@ -705,7 +935,8 @@ static void FUNCC(pred8x8l_horizontal)(uint8_t *_src, int has_topleft, int has_t
     ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);
 #undef ROW
 }
-static void FUNCC(pred8x8l_vertical)(uint8_t *_src, int has_topleft, int has_topright, int _stride)
+static void FUNCC(pred8x8l_vertical)(uint8_t *_src, int has_topleft,
+                                     int has_topright, ptrdiff_t _stride)
 {
     int y;
     pixel *src = (pixel*)_src;
@@ -728,7 +959,8 @@ static void FUNCC(pred8x8l_vertical)(uint8_t *_src, int has_topleft, int has_top
         AV_WN4PA(((pixel4*)(src+y*stride))+1, b);
     }
 }
-static void FUNCC(pred8x8l_down_left)(uint8_t *_src, int has_topleft, int has_topright, int _stride)
+static void FUNCC(pred8x8l_down_left)(uint8_t *_src, int has_topleft,
+                                      int has_topright, ptrdiff_t _stride)
 {
     pixel *src = (pixel*)_src;
     int stride = _stride/sizeof(pixel);
@@ -750,7 +982,8 @@ static void FUNCC(pred8x8l_down_left)(uint8_t *_src, int has_topleft, int has_to
     SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2;
     SRC(7,7)= (t14 + 3*t15 + 2) >> 2;
 }
-static void FUNCC(pred8x8l_down_right)(uint8_t *_src, int has_topleft, int has_topright, int _stride)
+static void FUNCC(pred8x8l_down_right)(uint8_t *_src, int has_topleft,
+                                       int has_topright, ptrdiff_t _stride)
 {
     pixel *src = (pixel*)_src;
     int stride = _stride/sizeof(pixel);
@@ -773,7 +1006,8 @@ static void FUNCC(pred8x8l_down_right)(uint8_t *_src, int has_topleft, int has_t
     SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2;
     SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2;
 }
-static void FUNCC(pred8x8l_vertical_right)(uint8_t *_src, int has_topleft, int has_topright, int _stride)
+static void FUNCC(pred8x8l_vertical_right)(uint8_t *_src, int has_topleft,
+                                           int has_topright, ptrdiff_t _stride)
 {
     pixel *src = (pixel*)_src;
     int stride = _stride/sizeof(pixel);
@@ -803,7 +1037,8 @@ static void FUNCC(pred8x8l_vertical_right)(uint8_t *_src, int has_topleft, int h
     SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2;
     SRC(7,0)= (t6 + t7 + 1) >> 1;
 }
-static void FUNCC(pred8x8l_horizontal_down)(uint8_t *_src, int has_topleft, int has_topright, int _stride)
+static void FUNCC(pred8x8l_horizontal_down)(uint8_t *_src, int has_topleft,
+                                            int has_topright, ptrdiff_t _stride)
 {
     pixel *src = (pixel*)_src;
     int stride = _stride/sizeof(pixel);
@@ -833,7 +1068,8 @@ static void FUNCC(pred8x8l_horizontal_down)(uint8_t *_src, int has_topleft, int
     SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2;
     SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2;
 }
-static void FUNCC(pred8x8l_vertical_left)(uint8_t *_src, int has_topleft, int has_topright, int _stride)
+static void FUNCC(pred8x8l_vertical_left)(uint8_t *_src, int has_topleft,
+                                          int has_topright, ptrdiff_t _stride)
 {
     pixel *src = (pixel*)_src;
     int stride = _stride/sizeof(pixel);
@@ -862,7 +1098,8 @@ static void FUNCC(pred8x8l_vertical_left)(uint8_t *_src, int has_topleft, int ha
     SRC(7,6)= (t10 + t11 + 1) >> 1;
     SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2;
 }
-static void FUNCC(pred8x8l_horizontal_up)(uint8_t *_src, int has_topleft, int has_topright, int _stride)
+static void FUNCC(pred8x8l_horizontal_up)(uint8_t *_src, int has_topleft,
+                                          int has_topright, ptrdiff_t _stride)
 {
     pixel *src = (pixel*)_src;
     int stride = _stride/sizeof(pixel);
@@ -896,7 +1133,9 @@ static void FUNCC(pred8x8l_horizontal_up)(uint8_t *_src, int has_topleft, int ha
 #undef PL
 #undef SRC
 
-static void FUNCC(pred4x4_vertical_add)(uint8_t *_pix, const DCTELEM *_block, int stride){
+static void FUNCC(pred4x4_vertical_add)(uint8_t *_pix, int16_t *_block,
+                                        ptrdiff_t stride)
+{
     int i;
     pixel *pix = (pixel*)_pix;
     const dctcoef *block = (const dctcoef*)_block;
@@ -911,9 +1150,13 @@ static void FUNCC(pred4x4_vertical_add)(uint8_t *_pix, const DCTELEM *_block, in
         pix++;
         block++;
     }
+
+    memset(_block, 0, sizeof(dctcoef) * 16);
 }
 
-static void FUNCC(pred4x4_horizontal_add)(uint8_t *_pix, const DCTELEM *_block, int stride){
+static void FUNCC(pred4x4_horizontal_add)(uint8_t *_pix, int16_t *_block,
+                                          ptrdiff_t stride)
+{
     int i;
     pixel *pix = (pixel*)_pix;
     const dctcoef *block = (const dctcoef*)_block;
@@ -927,9 +1170,13 @@ static void FUNCC(pred4x4_horizontal_add)(uint8_t *_pix, const DCTELEM *_block,
         pix+= stride;
         block+= 4;
     }
+
+    memset(_block, 0, sizeof(dctcoef) * 16);
 }
 
-static void FUNCC(pred8x8l_vertical_add)(uint8_t *_pix, const DCTELEM *_block, int stride){
+static void FUNCC(pred8x8l_vertical_add)(uint8_t *_pix, int16_t *_block,
+                                         ptrdiff_t stride)
+{
     int i;
     pixel *pix = (pixel*)_pix;
     const dctcoef *block = (const dctcoef*)_block;
@@ -948,9 +1195,13 @@ static void FUNCC(pred8x8l_vertical_add)(uint8_t *_pix, const DCTELEM *_block, i
         pix++;
         block++;
     }
+
+    memset(_block, 0, sizeof(dctcoef) * 64);
 }
 
-static void FUNCC(pred8x8l_horizontal_add)(uint8_t *_pix, const DCTELEM *_block, int stride){
+static void FUNCC(pred8x8l_horizontal_add)(uint8_t *_pix, int16_t *_block,
+                                           ptrdiff_t stride)
+{
     int i;
     pixel *pix = (pixel*)_pix;
     const dctcoef *block = (const dctcoef*)_block;
@@ -968,28 +1219,63 @@ static void FUNCC(pred8x8l_horizontal_add)(uint8_t *_pix, const DCTELEM *_block,
         pix+= stride;
         block+= 8;
     }
+
+    memset(_block, 0, sizeof(dctcoef) * 64);
 }
 
-static void FUNCC(pred16x16_vertical_add)(uint8_t *pix, const int *block_offset, const DCTELEM *block, int stride){
+static void FUNCC(pred16x16_vertical_add)(uint8_t *pix, const int *block_offset,
+                                          int16_t *block,
+                                          ptrdiff_t stride)
+{
     int i;
     for(i=0; i<16; i++)
         FUNCC(pred4x4_vertical_add)(pix + block_offset[i], block + i*16*sizeof(pixel), stride);
 }
 
-static void FUNCC(pred16x16_horizontal_add)(uint8_t *pix, const int *block_offset, const DCTELEM *block, int stride){
+static void FUNCC(pred16x16_horizontal_add)(uint8_t *pix,
+                                            const int *block_offset,
+                                            int16_t *block,
+                                            ptrdiff_t stride)
+{
     int i;
     for(i=0; i<16; i++)
         FUNCC(pred4x4_horizontal_add)(pix + block_offset[i], block + i*16*sizeof(pixel), stride);
 }
 
-static void FUNCC(pred8x8_vertical_add)(uint8_t *pix, const int *block_offset, const DCTELEM *block, int stride){
+static void FUNCC(pred8x8_vertical_add)(uint8_t *pix, const int *block_offset,
+                                        int16_t *block, ptrdiff_t stride)
+{
+    int i;
+    for(i=0; i<4; i++)
+        FUNCC(pred4x4_vertical_add)(pix + block_offset[i], block + i*16*sizeof(pixel), stride);
+}
+
+static void FUNCC(pred8x16_vertical_add)(uint8_t *pix, const int *block_offset,
+                                         int16_t *block, ptrdiff_t stride)
+{
     int i;
     for(i=0; i<4; i++)
         FUNCC(pred4x4_vertical_add)(pix + block_offset[i], block + i*16*sizeof(pixel), stride);
+    for(i=4; i<8; i++)
+        FUNCC(pred4x4_vertical_add)(pix + block_offset[i+4], block + i*16*sizeof(pixel), stride);
 }
 
-static void FUNCC(pred8x8_horizontal_add)(uint8_t *pix, const int *block_offset, const DCTELEM *block, int stride){
+static void FUNCC(pred8x8_horizontal_add)(uint8_t *pix, const int *block_offset,
+                                          int16_t *block,
+                                          ptrdiff_t stride)
+{
+    int i;
+    for(i=0; i<4; i++)
+        FUNCC(pred4x4_horizontal_add)(pix + block_offset[i], block + i*16*sizeof(pixel), stride);
+}
+
+static void FUNCC(pred8x16_horizontal_add)(uint8_t *pix,
+                                           const int *block_offset,
+                                           int16_t *block, ptrdiff_t stride)
+{
     int i;
     for(i=0; i<4; i++)
         FUNCC(pred4x4_horizontal_add)(pix + block_offset[i], block + i*16*sizeof(pixel), stride);
+    for(i=4; i<8; i++)
+        FUNCC(pred4x4_horizontal_add)(pix + block_offset[i+4], block + i*16*sizeof(pixel), stride);
 }