H.264: switch to x264-style tracking of luma/chroma DC NNZ

Useful so that we don't have to run the hierarchical DC iDCT if there aren't
any coefficients.  Opens up some future opportunities for optimization as well.

Originally committed as revision 26337 to svn://svn.ffmpeg.org/ffmpeg/trunk

1 files changed, 6 insertions, 5 deletions

diff --git a/libavcodec/h264_cabac.c b/libavcodec/h264_cabac.c
index 971af37114..3744095473 100644
--- a/libavcodec/h264_cabac.c
+++ b/libavcodec/h264_cabac.c
@@ -965,6 +965,7 @@ static av_always_inline int get_cabac_cbf_ctx( H264Context *h, int cat, int idx,
             nza = h->left_cbp&0x100;
             nzb = h-> top_cbp&0x100;
         } else {
+            idx -= CHROMA_DC_BLOCK_INDEX;
             nza = (h->left_cbp>>(6+idx))&0x01;
             nzb = (h-> top_cbp>>(6+idx))&0x01;
         }
@@ -1060,8 +1061,7 @@ static av_always_inline void decode_cabac_residual_internal( H264Context *h, DCT
     /* read coded block flag */
     if( is_dc || cat != 5 ) {
         if( get_cabac( CC, &h->cabac_state[85 + get_cabac_cbf_ctx( h, cat, n, is_dc ) ] ) == 0 ) {
-            if( !is_dc )
-                h->non_zero_count_cache[scan8[n]] = 0;
+            h->non_zero_count_cache[scan8[n]] = 0;
 
 #ifdef CABAC_ON_STACK
             h->cabac.range     = cc.range     ;
@@ -1112,7 +1112,8 @@ static av_always_inline void decode_cabac_residual_internal( H264Context *h, DCT
         if( cat == 0 )
             h->cbp_table[h->mb_xy] |= 0x100;
         else
-            h->cbp_table[h->mb_xy] |= 0x40 << n;
+            h->cbp_table[h->mb_xy] |= 0x40 << (n - CHROMA_DC_BLOCK_INDEX);
+        h->non_zero_count_cache[scan8[n]] = coeff_count;
     } else {
         if( cat == 5 )
             fill_rectangle(&h->non_zero_count_cache[scan8[n]], 2, 2, 8, coeff_count, 1);
@@ -1642,7 +1643,7 @@ decode_intra_mb:
             //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 DC\n" );
             AV_ZERO128(h->mb_luma_dc+0);
             AV_ZERO128(h->mb_luma_dc+8);
-            decode_cabac_residual_dc( h, h->mb_luma_dc, 0, 0, scan, 16);
+            decode_cabac_residual_dc( h, h->mb_luma_dc, 0, LUMA_DC_BLOCK_INDEX, scan, 16);
 
             if( cbp&15 ) {
                 qmul = h->dequant4_coeff[0][s->qscale];
@@ -1681,7 +1682,7 @@ decode_intra_mb:
             int c;
             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, c, chroma_dc_scan, 4);
+                decode_cabac_residual_dc(h, h->mb + 256 + 16*4*c, 3, CHROMA_DC_BLOCK_INDEX+c, chroma_dc_scan, 4);
             }
         }