Commit 3924dac4 authored by Michael Niedermayer's avatar Michael Niedermayer
Browse files

use pointer arithmetic in mc_block(), 25% faster

Originally committed as revision 3452 to svn://svn.ffmpeg.org/ffmpeg/trunk
parent a6c01275
......@@ -1929,14 +1929,15 @@ static void decode_blocks(SnowContext *s){
static void mc_block(uint8_t *dst, uint8_t *src, uint8_t *tmp, int stride, int b_w, int b_h, int dx, int dy){
int x, y;
START_TIMER
for(y=0; y < b_h+5; y++){
for(x=0; x < b_w; x++){
int a0= src[x + y*stride];
int a1= src[x + 1 + y*stride];
int a2= src[x + 2 + y*stride];
int a3= src[x + 3 + y*stride];
int a4= src[x + 4 + y*stride];
int a5= src[x + 5 + y*stride];
int a0= src[x ];
int a1= src[x + 1];
int a2= src[x + 2];
int a3= src[x + 3];
int a4= src[x + 4];
int a5= src[x + 5];
// int am= 9*(a1+a2) - (a0+a3);
int am= 20*(a2+a3) - 5*(a1+a4) + (a0+a5);
// int am= 18*(a2+a3) - 2*(a1+a4);
......@@ -1945,23 +1946,27 @@ static void mc_block(uint8_t *dst, uint8_t *src, uint8_t *tmp, int stride, int b
// if(b_w==16) am= 8*(a1+a2);
if(dx<8) tmp[x + y*stride]= (32*a2*( 8-dx) + am* dx + 128)>>8;
else tmp[x + y*stride]= ( am*(16-dx) + 32*a3*(dx-8) + 128)>>8;
if(dx<8) tmp[x]= (32*a2*( 8-dx) + am* dx + 128)>>8;
else tmp[x]= ( am*(16-dx) + 32*a3*(dx-8) + 128)>>8;
/* if (dx< 4) tmp[x + y*stride]= (16*a1*( 4-dx) + aL* dx + 32)>>6;
else if(dx< 8) tmp[x + y*stride]= ( aL*( 8-dx) + am*(dx- 4) + 32)>>6;
else if(dx<12) tmp[x + y*stride]= ( am*(12-dx) + aR*(dx- 8) + 32)>>6;
else tmp[x + y*stride]= ( aR*(16-dx) + 16*a2*(dx-12) + 32)>>6;*/
}
tmp += stride;
src += stride;
}
tmp -= (b_h+5)*stride;
for(y=0; y < b_h; y++){
for(x=0; x < b_w; x++){
int a0= tmp[x + y *stride];
int a1= tmp[x + (y + 1)*stride];
int a2= tmp[x + (y + 2)*stride];
int a3= tmp[x + (y + 3)*stride];
int a4= tmp[x + (y + 4)*stride];
int a5= tmp[x + (y + 5)*stride];
int a0= tmp[x + 0*stride];
int a1= tmp[x + 1*stride];
int a2= tmp[x + 2*stride];
int a3= tmp[x + 3*stride];
int a4= tmp[x + 4*stride];
int a5= tmp[x + 5*stride];
int am= 20*(a2+a3) - 5*(a1+a4) + (a0+a5);
// int am= 18*(a2+a3) - 2*(a1+a4);
/* int aL= (-7*a0 + 105*a1 + 35*a2 - 5*a3)>>3;
......@@ -1969,15 +1974,18 @@ static void mc_block(uint8_t *dst, uint8_t *src, uint8_t *tmp, int stride, int b
// if(b_w==16) am= 8*(a1+a2);
if(dy<8) dst[x + y*stride]= (32*a2*( 8-dy) + am* dy + 128)>>8;
else dst[x + y*stride]= ( am*(16-dy) + 32*a3*(dy-8) + 128)>>8;
if(dy<8) dst[x]= (32*a2*( 8-dy) + am* dy + 128)>>8;
else dst[x]= ( am*(16-dy) + 32*a3*(dy-8) + 128)>>8;
/* if (dy< 4) tmp[x + y*stride]= (16*a1*( 4-dy) + aL* dy + 32)>>6;
else if(dy< 8) tmp[x + y*stride]= ( aL*( 8-dy) + am*(dy- 4) + 32)>>6;
else if(dy<12) tmp[x + y*stride]= ( am*(12-dy) + aR*(dy- 8) + 32)>>6;
else tmp[x + y*stride]= ( aR*(16-dy) + 16*a2*(dy-12) + 32)>>6;*/
}
dst += stride;
tmp += stride;
}
STOP_TIMER("mc_block")
}
#define mcb(dx,dy,b_w)\
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment