Move the qp check to skip the loop filter up.

Originally committed as revision 21274 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/h264.c

@@ -2184,7 +2184,8 @@ static void loop_filter(H264Context *h){
                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
                 }
                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, !is_complex);
-                fill_filter_caches(h, mb_type); //FIXME don't fill stuff which isn't used by filter_mb
+                if(fill_filter_caches(h, mb_type) < 0)
+                    continue;
                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.qscale_table[mb_xy]);
                 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.qscale_table[mb_xy]);
 

libavcodec/h264.h

@@ -726,7 +726,7 @@ static inline int get_chroma_qp(H264Context *h, int t, int qscale){
 
 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
 
-static av_always_inline void fill_caches(H264Context *h, int mb_type, int for_deblock){
+static av_always_inline int fill_caches(H264Context *h, int mb_type, int for_deblock){
     MpegEncContext * const s = &h->s;
     const int mb_xy= h->mb_xy;
     int topleft_xy, top_xy, topright_xy, left_xy[2];
@@ -795,6 +795,19 @@ static av_always_inline void fill_caches(H264Context *h, int mb_type, int for_de
     h->left_mb_xy[0] = left_xy[0];
     h->left_mb_xy[1] = left_xy[1];
     if(for_deblock){
+
+        //for sufficiently low qp, filtering wouldn't do anything
+        //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
+        if(!FRAME_MBAFF){
+            int qp_thresh = h->qp_thresh;
+            int qp = s->current_picture.qscale_table[mb_xy];
+            if(qp <= qp_thresh
+            && (s->mb_x == 0 || ((qp + s->current_picture.qscale_table[mb_xy-1] + 1)>>1) <= qp_thresh)
+            && (top_xy   < 0 || ((qp + s->current_picture.qscale_table[top_xy ] + 1)>>1) <= qp_thresh)){
+                return 1;
+            }
+        }
+
         *((uint64_t*)&h->non_zero_count_cache[0+8*1])= *((uint64_t*)&h->non_zero_count[mb_xy][ 0]);
         *((uint64_t*)&h->non_zero_count_cache[0+8*2])= *((uint64_t*)&h->non_zero_count[mb_xy][ 8]);
         *((uint32_t*)&h->non_zero_count_cache[0+8*5])= *((uint32_t*)&h->non_zero_count[mb_xy][16]);
@@ -1186,14 +1199,19 @@ static av_always_inline void fill_caches(H264Context *h, int mb_type, int for_de
 
     if(!for_deblock)
     h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
+    return 0;
 }
 
 static void fill_decode_caches(H264Context *h, int mb_type){
     fill_caches(h, mb_type, 0);
 }
 
-static void fill_filter_caches(H264Context *h, int mb_type){
-    fill_caches(h, mb_type, 1);
+/**
+ *
+ * @returns non zero if the loop filter can be skiped
+ */
+static int fill_filter_caches(H264Context *h, int mb_type){
+    return fill_caches(h, mb_type, 1);
 }
 
 /**

libavcodec/h264_loopfilter.c

@@ -651,17 +651,6 @@ void ff_h264_filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint
     av_unused int dir;
     int list;
 
-    //for sufficiently low qp, filtering wouldn't do anything
-    //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
-    if(!FRAME_MBAFF){
-        int qp_thresh = h->qp_thresh;
-        int qp = s->current_picture.qscale_table[mb_xy];
-        if(qp <= qp_thresh
-           && (mb_x == 0 || ((qp + s->current_picture.qscale_table[mb_xy-1] + 1)>>1) <= qp_thresh)
-           && (h->top_mb_xy < 0 || ((qp + s->current_picture.qscale_table[h->top_mb_xy] + 1)>>1) <= qp_thresh)){
-            return;
-        }
-    }
     // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
     if(!h->pps.cabac && h->pps.transform_8x8_mode){
         int top_type, left_type[2];