Avoid some unneccesary memory loads in macroblock_encode

diff --git a/encoder/macroblock.c b/encoder/macroblock.c
index 1f66278a4cf4cd4d25b2f3988764c44b5f696295..351650a1d9cb3629d70fe8bfb807c2e67f6fe89a 100644 (file)
--- a/encoder/macroblock.c
+++ b/encoder/macroblock.c
@@ -157,10 +157,7 @@ static void x264_mb_encode_i16x16( x264_t *h, int p, int i_qp )
         return;
     }
 
-    M32( &h->mb.cache.non_zero_count[x264_scan8[ 0+p*16]] ) = 0;
-    M32( &h->mb.cache.non_zero_count[x264_scan8[ 2+p*16]] ) = 0;
-    M32( &h->mb.cache.non_zero_count[x264_scan8[ 8+p*16]] ) = 0;
-    M32( &h->mb.cache.non_zero_count[x264_scan8[10+p*16]] ) = 0;
+    CLEAR_16x16_NNZ( p );
 
     h->dctf.sub16x16_dct( dct4x4, p_src, p_dst );
 

diff --git a/encoder/macroblock.h b/encoder/macroblock.h
index d0ba11a9707101ee37bcb7fecc3ccfacda77010b..9ca3f3d6d416f1c990c535489f8a9aa1e9a13766 100644 (file)
--- a/encoder/macroblock.h
+++ b/encoder/macroblock.h
@@ -98,10 +98,10 @@ do\
 #define CLEAR_16x16_NNZ( p ) \
 do\
 {\
-    M32( &h->mb.cache.non_zero_count[x264_scan8[16*p+ 0]] ) = 0;\
-    M32( &h->mb.cache.non_zero_count[x264_scan8[16*p+ 2]] ) = 0;\
-    M32( &h->mb.cache.non_zero_count[x264_scan8[16*p+ 8]] ) = 0;\
-    M32( &h->mb.cache.non_zero_count[x264_scan8[16*p+10]] ) = 0;\
+    M32( &h->mb.cache.non_zero_count[x264_scan8[16*p] + 0*8] ) = 0;\
+    M32( &h->mb.cache.non_zero_count[x264_scan8[16*p] + 1*8] ) = 0;\
+    M32( &h->mb.cache.non_zero_count[x264_scan8[16*p] + 2*8] ) = 0;\
+    M32( &h->mb.cache.non_zero_count[x264_scan8[16*p] + 3*8] ) = 0;\
 } while(0)
 
 /* A special for loop that iterates branchlessly over each set