h264: 4:2:2 intra decoding support

Signed-off-by: Diego Biurrun <diego@biurrun.de>
Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>

1 files changed, 138 insertions, 0 deletions

diff --git a/libavcodec/h264pred_template.c b/libavcodec/h264pred_template.c
index 750e82c..d4f654e 100644
--- a/libavcodec/h264pred_template.c
+++ b/libavcodec/h264pred_template.c
@@ -454,6 +454,19 @@ static void FUNCC(pred8x8_vertical)(uint8_t *_src, int _stride){
     }
 }
 
+static void FUNCC(pred8x16_vertical)(uint8_t *_src, int _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, int stride){
     int i;
     pixel *src = (pixel*)_src;
@@ -466,6 +479,17 @@ static void FUNCC(pred8x8_horizontal)(uint8_t *_src, int stride){
     }
 }
 
+static void FUNCC(pred8x16_horizontal)(uint8_t *_src, int 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){\
     int i;\
@@ -482,6 +506,11 @@ 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, int 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){
     int i;
     int dc0, dc2;
@@ -507,6 +536,11 @@ static void FUNCC(pred8x8_left_dc)(uint8_t *_src, int stride){
     }
 }
 
+static void FUNCC(pred8x16_left_dc)(uint8_t *_src, int stride){
+    FUNCC(pred8x8_left_dc)(_src, stride);
+    FUNCC(pred8x8_left_dc)(_src+8*stride, stride);
+}
+
 static void FUNCC(pred8x8_top_dc)(uint8_t *_src, int stride){
     int i;
     int dc0, dc1;
@@ -532,6 +566,27 @@ static void FUNCC(pred8x8_top_dc)(uint8_t *_src, int stride){
     }
 }
 
+static void FUNCC(pred8x16_top_dc)(uint8_t *_src, int 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, int stride){
     int i;
     int dc0, dc1, dc2;
@@ -560,6 +615,48 @@ static void FUNCC(pred8x8_dc)(uint8_t *_src, int stride){
     }
 }
 
+static void FUNCC(pred8x16_dc)(uint8_t *_src, int 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);
+    }
+}
+
 //the following 4 function should not be optimized!
 static void FUNC(pred8x8_mad_cow_dc_l0t)(uint8_t *src, int stride){
     FUNCC(pred8x8_top_dc)(src, stride);
@@ -618,6 +715,47 @@ static void FUNCC(pred8x8_plane)(uint8_t *_src, int _stride){
   }
 }
 
+static void FUNCC(pred8x16_plane)(uint8_t *_src, int _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;