H.264/SVQ3: make chroma DC work the same way as luma DC

No speed improvement, but necessary for some future stuff.
Also opens up the possibility of asm chroma dc idct/dequant.

Originally committed as revision 26349 to svn://svn.ffmpeg.org/ffmpeg/trunk

10 files changed, 42 insertions, 99 deletions

diff --git a/libavcodec/dsputil.h b/libavcodec/dsputil.h
index 0efbad918a..157149113e 100644
--- a/libavcodec/dsputil.h
+++ b/libavcodec/dsputil.h
@@ -66,6 +66,7 @@ void ff_h264_idct_add8_c(uint8_t **dest, const int *blockoffset, DCTELEM *block,
 
 void ff_h264_luma_dc_dequant_idct_c(DCTELEM *output, DCTELEM *input, int qmul);
 void ff_svq3_luma_dc_dequant_idct_c(DCTELEM *output, DCTELEM *input, int qp);
+void ff_chroma_dc_dequant_idct_c(DCTELEM *output, DCTELEM *input, int qmul);
 void ff_svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
 
 void ff_vector_fmul_window_c(float *dst, const float *src0, const float *src1,
diff --git a/libavcodec/h264.c b/libavcodec/h264.c
index 40dc276dd6..d0ee4744c8 100644
--- a/libavcodec/h264.c
+++ b/libavcodec/h264.c
@@ -246,93 +246,6 @@ int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
     return 0;
 }
 
-#if 0
-/**
- * DCT transforms the 16 dc values.
- * @param qp quantization parameter ??? FIXME
- */
-static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
-//    const int qmul= dequant_coeff[qp][0];
-    int i;
-    int temp[16]; //FIXME check if this is a good idea
-    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
-    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
-
-    for(i=0; i<4; i++){
-        const int offset= y_offset[i];
-        const int z0= block[offset+stride*0] + block[offset+stride*4];
-        const int z1= block[offset+stride*0] - block[offset+stride*4];
-        const int z2= block[offset+stride*1] - block[offset+stride*5];
-        const int z3= block[offset+stride*1] + block[offset+stride*5];
-
-        temp[4*i+0]= z0+z3;
-        temp[4*i+1]= z1+z2;
-        temp[4*i+2]= z1-z2;
-        temp[4*i+3]= z0-z3;
-    }
-
-    for(i=0; i<4; i++){
-        const int offset= x_offset[i];
-        const int z0= temp[4*0+i] + temp[4*2+i];
-        const int z1= temp[4*0+i] - temp[4*2+i];
-        const int z2= temp[4*1+i] - temp[4*3+i];
-        const int z3= temp[4*1+i] + temp[4*3+i];
-
-        block[stride*0 +offset]= (z0 + z3)>>1;
-        block[stride*2 +offset]= (z1 + z2)>>1;
-        block[stride*8 +offset]= (z1 - z2)>>1;
-        block[stride*10+offset]= (z0 - z3)>>1;
-    }
-}
-#endif
-
-#undef xStride
-#undef stride
-
-static void chroma_dc_dequant_idct_c(DCTELEM *block, int qmul){
-    const int stride= 16*2;
-    const int xStride= 16;
-    int a,b,c,d,e;
-
-    a= block[stride*0 + xStride*0];
-    b= block[stride*0 + xStride*1];
-    c= block[stride*1 + xStride*0];
-    d= block[stride*1 + xStride*1];
-
-    e= a-b;
-    a= a+b;
-    b= c-d;
-    c= c+d;
-
-    block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
-    block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
-    block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
-    block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
-}
-
-#if 0
-static void chroma_dc_dct_c(DCTELEM *block){
-    const int stride= 16*2;
-    const int xStride= 16;
-    int a,b,c,d,e;
-
-    a= block[stride*0 + xStride*0];
-    b= block[stride*0 + xStride*1];
-    c= block[stride*1 + xStride*0];
-    d= block[stride*1 + xStride*1];
-
-    e= a-b;
-    a= a+b;
-    b= c-d;
-    c= c+d;
-
-    block[stride*0 + xStride*0]= (a+c);
-    block[stride*0 + xStride*1]= (e+b);
-    block[stride*1 + xStride*0]= (a-c);
-    block[stride*1 + xStride*1]= (e-b);
-}
-#endif
-
 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
                            int src_x_offset, int src_y_offset,
@@ -1283,17 +1196,19 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple){
                     }
                 }
             }else{
+                int chroma_qpu = h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0];
+                int chroma_qpv = h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0];
                 if(is_h264){
                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+0] ])
-                        chroma_dc_dequant_idct_c(h->mb + 16*16     , h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
+                        h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16+0*16, &h->mb_chroma_dc[0], chroma_qpu );
                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+1] ])
-                        chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
+                        h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16+4*16, &h->mb_chroma_dc[1], chroma_qpv );
                     h->h264dsp.h264_idct_add8(dest, block_offset,
                                               h->mb, uvlinesize,
                                               h->non_zero_count_cache);
                 }else{
-                    chroma_dc_dequant_idct_c(h->mb + 16*16     , h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
-                    chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
+                    h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16+0*16, &h->mb_chroma_dc[0], chroma_qpu );
+                    h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16+4*16, &h->mb_chroma_dc[1], chroma_qpv );
                     for(i=16; i<16+8; i++){
                         if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
                             uint8_t * const ptr= dest[(i&4)>>2] + block_offset[i];
diff --git a/libavcodec/h264.h b/libavcodec/h264.h
index 17fd680369..4436f6f467 100644
--- a/libavcodec/h264.h
+++ b/libavcodec/h264.h
@@ -407,6 +407,7 @@ typedef struct H264Context{
 
     DECLARE_ALIGNED(16, DCTELEM, mb)[16*24];
     DECLARE_ALIGNED(16, DCTELEM, mb_luma_dc)[16];
+    DECLARE_ALIGNED(16, DCTELEM, mb_chroma_dc)[2][4];
     DCTELEM mb_padding[256];        ///< as mb is addressed by scantable[i] and scantable is uint8_t we can either check that i is not too large or ensure that there is some unused stuff after mb
 
     /**
diff --git a/libavcodec/h264_cabac.c b/libavcodec/h264_cabac.c
index 3744095473..d73293234d 100644
--- a/libavcodec/h264_cabac.c
+++ b/libavcodec/h264_cabac.c
@@ -1680,9 +1680,10 @@ decode_intra_mb:
 
         if( cbp&0x30 ){
             int c;
+            AV_ZERO128(h->mb_chroma_dc);
             for( c = 0; c < 2; c++ ) {
                 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-DC\n",c );
-                decode_cabac_residual_dc(h, h->mb + 256 + 16*4*c, 3, CHROMA_DC_BLOCK_INDEX+c, chroma_dc_scan, 4);
+                decode_cabac_residual_dc(h, h->mb_chroma_dc[c], 3, CHROMA_DC_BLOCK_INDEX+c, chroma_dc_scan, 4);
             }
         }
 
diff --git a/libavcodec/h264_cavlc.c b/libavcodec/h264_cavlc.c
index b8bc45058c..55ff02294b 100644
--- a/libavcodec/h264_cavlc.c
+++ b/libavcodec/h264_cavlc.c
@@ -987,8 +987,9 @@ decode_intra_mb:
         }
 
         if(cbp&0x30){
+            AV_ZERO128(h->mb_chroma_dc);
             for(chroma_idx=0; chroma_idx<2; chroma_idx++)
-                if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX+chroma_idx, chroma_dc_scan, NULL, 4) < 0){
+                if( decode_residual(h, gb, h->mb_chroma_dc[chroma_idx], CHROMA_DC_BLOCK_INDEX+chroma_idx, chroma_dc_scan, NULL, 4) < 0){
                     return -1;
                 }
         }
diff --git a/libavcodec/h264data.h b/libavcodec/h264data.h
index b3631da93b..73fcc5ce38 100644
--- a/libavcodec/h264data.h
+++ b/libavcodec/h264data.h
@@ -79,8 +79,7 @@ static const uint8_t luma_dc_field_scan[16]={
 };
 
 static const uint8_t chroma_dc_scan[4]={
- (0+0*2)*16, (1+0*2)*16,
- (0+1*2)*16, (1+1*2)*16,  //FIXME
+ 0,1,2,3
 };
 
 // zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
diff --git a/libavcodec/h264dsp.c b/libavcodec/h264dsp.c
index c3ee06d9df..acb458faa1 100644
--- a/libavcodec/h264dsp.c
+++ b/libavcodec/h264dsp.c
@@ -283,6 +283,7 @@ void ff_h264dsp_init(H264DSPContext *c)
     c->h264_idct_add8      = ff_h264_idct_add8_c;
     c->h264_idct_add16intra= ff_h264_idct_add16intra_c;
     c->h264_luma_dc_dequant_idct= ff_h264_luma_dc_dequant_idct_c;
+    c->h264_chroma_dc_dequant_idct= ff_chroma_dc_dequant_idct_c;
 
     c->weight_h264_pixels_tab[0]= weight_h264_pixels16x16_c;
     c->weight_h264_pixels_tab[1]= weight_h264_pixels16x8_c;
diff --git a/libavcodec/h264dsp.h b/libavcodec/h264dsp.h
index 2491c90bbb..c790f4aa4d 100644
--- a/libavcodec/h264dsp.h
+++ b/libavcodec/h264dsp.h
@@ -68,6 +68,7 @@ typedef struct H264DSPContext{
     void (*h264_idct_add8)(uint8_t **dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
     void (*h264_idct_add16intra)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
     void (*h264_luma_dc_dequant_idct)(DCTELEM *output, DCTELEM *input/*align 16*/, int qmul);
+    void (*h264_chroma_dc_dequant_idct)(DCTELEM *output, DCTELEM *input/*align 16*/, int qmul);
 }H264DSPContext;
 
 void ff_h264dsp_init(H264DSPContext *c);
diff --git a/libavcodec/h264idct.c b/libavcodec/h264idct.c
index f5b05ac24f..53e70a22df 100644
--- a/libavcodec/h264idct.c
+++ b/libavcodec/h264idct.c
@@ -250,4 +250,26 @@ void ff_h264_luma_dc_dequant_idct_c(DCTELEM *output, DCTELEM *input, int qmul){
         output[stride* 4+offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
         output[stride* 5+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
     }
+#undef stride
+}
+
+void ff_chroma_dc_dequant_idct_c(DCTELEM *output, DCTELEM *input, int qmul){
+    const int stride= 16*2;
+    const int xStride= 16;
+    int a,b,c,d,e;
+
+    a= input[0];
+    b= input[1];
+    c= input[2];
+    d= input[3];
+
+    e= a-b;
+    a= a+b;
+    b= c-d;
+    c= c+d;
+
+    output[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
+    output[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
+    output[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
+    output[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
 }
diff --git a/libavcodec/svq3.c b/libavcodec/svq3.c
index a52744228e..62a20911a3 100644
--- a/libavcodec/svq3.c
+++ b/libavcodec/svq3.c
@@ -671,11 +671,12 @@ static int svq3_decode_mb(H264Context *h, unsigned int mb_type)
         }
 
         if ((cbp & 0x30)) {
+            AV_ZERO128(h->mb_chroma_dc);
             for (i = 0; i < 2; ++i) {
-              if (svq3_decode_block(&s->gb, &h->mb[16*(16 + 4*i)], 0, 3)){
-                av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding chroma dc block\n");
-                return -1;
-              }
+                if (svq3_decode_block(&s->gb, h->mb_chroma_dc[i], 0, 3)){
+                    av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding chroma dc block\n");
+                    return -1;
+                }
             }
 
             if ((cbp & 0x20)) {