slightly faster loopfilter

git-svn-id: svn://svn.videolan.org/x264/trunk@498 df754926-b1dd-0310-bc7b-ec298dee348c

diff --git a/common/frame.c b/common/frame.c
index ed73bb67da94519868dd89726a7163fc6d15f1c3..60762201c3b262a9666cebb52919e19e1c1b3f10 100644 (file)
--- a/common/frame.c
+++ b/common/frame.c
@@ -505,9 +505,9 @@ void x264_frame_deblocking_filter( x264_t *h, int i_slice_type )
         const int mb_xy  = mb_y * h->mb.i_mb_stride + mb_x;
         const int mb_8x8 = 2 * s8x8 * mb_y + 2 * mb_x;
         const int mb_4x4 = 4 * s4x4 * mb_y + 4 * mb_x;
-        int i_edge;
-        int i_dir;
         const int b_8x8_transform = h->mb.mb_transform_size[mb_xy];
+        const int i_edge_end = (h->mb.type[mb_xy] == P_SKIP) ? 1 : 4;
+        int i_edge, i_dir;
 
         /* cavlc + 8x8 transform stores nnz per 16 coeffs for the purpose of
          * entropy coding, but per 64 coeffs for the purpose of deblocking */
@@ -524,19 +524,21 @@ void x264_frame_deblocking_filter( x264_t *h, int i_slice_type )
          * i_dir == 1 -> horizontal edge */
         for( i_dir = 0; i_dir < 2; i_dir++ )
         {
-            int i_start;
+            int i_start = (i_dir ? mb_y : mb_x) ? 0 : 1;
             int i_qp, i_qpn;
 
-            i_start = (( i_dir == 0 && mb_x != 0 ) || ( i_dir == 1 && mb_y != 0 ) ) ? 0 : 1;
-
-            for( i_edge = i_start; i_edge < 4; i_edge++ )
+            for( i_edge = i_start; i_edge < i_edge_end; i_edge++ )
             {
-                int mbn_xy  = i_edge > 0 ? mb_xy  : ( i_dir == 0 ? mb_xy  - 1 : mb_xy - h->mb.i_mb_stride );
-                int mbn_8x8 = i_edge > 0 ? mb_8x8 : ( i_dir == 0 ? mb_8x8 - 2 : mb_8x8 - 2 * s8x8 );
-                int mbn_4x4 = i_edge > 0 ? mb_4x4 : ( i_dir == 0 ? mb_4x4 - 4 : mb_4x4 - 4 * s4x4 );
-
+                int mbn_xy, mbn_8x8, mbn_4x4;
                 int bS[4];  /* filtering strength */
 
+                if( b_8x8_transform && (i_edge&1) )
+                    continue;
+
+                mbn_xy  = i_edge > 0 ? mb_xy  : ( i_dir == 0 ? mb_xy  - 1 : mb_xy - h->mb.i_mb_stride );
+                mbn_8x8 = i_edge > 0 ? mb_8x8 : ( i_dir == 0 ? mb_8x8 - 2 : mb_8x8 - 2 * s8x8 );
+                mbn_4x4 = i_edge > 0 ? mb_4x4 : ( i_dir == 0 ? mb_4x4 - 4 : mb_4x4 - 4 * s4x4 );
+
                 /* *** Get bS for each 4px for the current edge *** */
                 if( IS_INTRA( h->mb.type[mb_xy] ) || IS_INTRA( h->mb.type[mbn_xy] ) )
                 {
@@ -594,12 +596,9 @@ void x264_frame_deblocking_filter( x264_t *h, int i_slice_type )
                 if( i_dir == 0 )
                 {
                     /* vertical edge */
-                    if( !b_8x8_transform || !(i_edge & 1) )
-                    {
-                        deblock_edge( h, &h->fdec->plane[0][16*mb_y * h->fdec->i_stride[0] + 16*mb_x + 4*i_edge],
-                                      h->fdec->i_stride[0], bS, (i_qp+i_qpn+1) >> 1, 0,
-                                      h->loopf.deblock_h_luma, h->loopf.deblock_h_luma_intra );
-                    }
+                    deblock_edge( h, &h->fdec->plane[0][16*mb_y * h->fdec->i_stride[0] + 16*mb_x + 4*i_edge],
+                                  h->fdec->i_stride[0], bS, (i_qp+i_qpn+1) >> 1, 0,
+                                  h->loopf.deblock_h_luma, h->loopf.deblock_h_luma_intra );
                     if( !(i_edge & 1) )
                     {
                         /* U/V planes */
@@ -616,12 +615,9 @@ void x264_frame_deblocking_filter( x264_t *h, int i_slice_type )
                 else
                 {
                     /* horizontal edge */
-                    if( !b_8x8_transform || !(i_edge & 1) )
-                    {
-                        deblock_edge( h, &h->fdec->plane[0][(16*mb_y + 4*i_edge) * h->fdec->i_stride[0] + 16*mb_x],
-                                      h->fdec->i_stride[0], bS, (i_qp+i_qpn+1) >> 1, 0,
-                                      h->loopf.deblock_v_luma, h->loopf.deblock_v_luma_intra );
-                    }
+                    deblock_edge( h, &h->fdec->plane[0][(16*mb_y + 4*i_edge) * h->fdec->i_stride[0] + 16*mb_x],
+                                  h->fdec->i_stride[0], bS, (i_qp+i_qpn+1) >> 1, 0,
+                                  h->loopf.deblock_v_luma, h->loopf.deblock_v_luma_intra );
                     /* U/V planes */
                     if( !(i_edge & 1) )
                     {