Ticket #113: ffmpeg-cbpluma-speed.diff

libavcodec/h264.c

@@ -5761 +5761 @@
         return 3;
 }
 
-static const uint8_t block_idx_x[16] = {
-    0, 1, 0, 1, 2, 3, 2, 3, 0, 1, 0, 1, 2, 3, 2, 3
-};
-static const uint8_t block_idx_y[16] = {
-    0, 0, 1, 1, 0, 0, 1, 1, 2, 2, 3, 3, 2, 2, 3, 3
-};
-static const uint8_t block_idx_xy[4][4] = {
-    { 0, 2, 8,  10},
-    { 1, 3, 9,  11},
-    { 4, 6, 12, 14},
-    { 5, 7, 13, 15}
-};
-
 static int decode_cabac_mb_cbp_luma( H264Context *h) {
     int cbp = 0;
     int cbp_b = -1;
@@ -5789 +5776 @@
         int x, y;
         int ctx = 0;
 
-        x = block_idx_x[4*i8x8];
-        y = block_idx_y[4*i8x8];
+        x = (i8x8 & 1) * 2;
+        y = i8x8 & 2;
 
         if( x > 0 )
             cbp_a = cbp;
@@ -5802 +5789 @@
         if( y > 0 )
             cbp_b = cbp;
 
+        #define block_idx_xy(r, c) ((((r) & 2) >> 1) + ((c) & 2))
+        #define lsb_is_zero(x)     (((x) & 1) ^ 1)
+
         /* No need to test for skip as we put 0 for skip block */
         /* No need to test for IPCM as we put 1 for IPCM block */
         if( cbp_a >= 0 ) {
-            int i8x8a = block_idx_xy[(x-1)&0x03][y]/4;
-            if( ((cbp_a >> i8x8a)&0x01) == 0 )
-                ctx++;
+            int i8x8a = block_idx_xy(x-1, y);
+
+            ctx = lsb_is_zero(cbp_a >> i8x8a);
         }
 
         if( cbp_b >= 0 ) {
-            int i8x8b = block_idx_xy[x][(y-1)&0x03]/4;
-            if( ((cbp_b >> i8x8b)&0x01) == 0 )
-                ctx += 2;
+            int i8x8b = block_idx_xy(x, y-1);
+
+            ctx += lsb_is_zero(cbp_b >> i8x8b) * 2;
         }
 
+       #undef block_idx_xy
+       #undef lsb_is_zero
+
         if( get_cabac( &h->cabac, &h->cabac_state[73 + ctx] ) ) {
             cbp |= 1 << i8x8;
         }