changeset 11025:cd1f5f6a2e45 libavcodec

Use table to speedup access to non_zero_count in MBAFF with differing interlacing. ~4 cpu cycles speedup
author michael
date Wed, 27 Jan 2010 11:14:29 +0000
parents 5ab861519e79
children f5678fb91140
files h264_loopfilter.c
diffstat 1 files changed, 11 insertions(+), 1 deletions(-) [+]
line wrap: on
line diff
--- a/h264_loopfilter.c	Wed Jan 27 00:31:52 2010 +0000
+++ b/h264_loopfilter.c	Wed Jan 27 11:14:29 2010 +0000
@@ -632,6 +632,16 @@
             *(uint64_t*)&bS[0]=
             *(uint64_t*)&bS[4]= 0x0004000400040004ULL;
         else {
+            static const uint8_t offset[2][2][8]={
+                {
+                    {7+8*0, 7+8*0, 7+8*0, 7+8*0, 7+8*1, 7+8*1, 7+8*1, 7+8*1},
+                    {7+8*2, 7+8*2, 7+8*2, 7+8*2, 7+8*3, 7+8*3, 7+8*3, 7+8*3},
+                },{
+                    {7+8*0, 7+8*1, 7+8*2, 7+8*3, 7+8*0, 7+8*1, 7+8*2, 7+8*3},
+                    {7+8*0, 7+8*1, 7+8*2, 7+8*3, 7+8*0, 7+8*1, 7+8*2, 7+8*3},
+                }
+            };
+            const uint8_t *off= offset[MB_FIELD][mb_y&1];
             for( i = 0; i < 8; i++ ) {
                 int j= MB_FIELD ? i>>2 : i&1;
                 int mbn_xy = h->left_mb_xy[j];
@@ -644,7 +654,7 @@
                          ((!h->pps.cabac && IS_8x8DCT(mbn_type)) ?
                             (h->cbp_table[mbn_xy] & ((MB_FIELD ? (i&2) : (mb_y&1)) ? 8 : 2))
                                                                        :
-                            h->non_zero_count[mbn_xy][7+(MB_FIELD ? (i&3) : (i>>2)+(mb_y&1)*2)*8]));
+                            h->non_zero_count[mbn_xy][ off[i] ]));
                 }
             }
         }