Mercurial > libavcodec.hg
comparison snow.c @ 2221:3543987dccad libavcodec
use pointer arithmetic in mc_block(), 25% faster
| author | michael |
|---|---|
| date | Sun, 12 Sep 2004 10:51:02 +0000 |
| parents | 713ad427a3c7 |
| children | b26474e72d6d |
comparison
equal
deleted
inserted
replaced
| 2220:21947e176d4d | 2221:3543987dccad |
|---|---|
| 1927 } | 1927 } |
| 1928 } | 1928 } |
| 1929 | 1929 |
| 1930 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){ | 1930 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){ |
| 1931 int x, y; | 1931 int x, y; |
| 1932 START_TIMER | |
| 1932 for(y=0; y < b_h+5; y++){ | 1933 for(y=0; y < b_h+5; y++){ |
| 1933 for(x=0; x < b_w; x++){ | 1934 for(x=0; x < b_w; x++){ |
| 1934 int a0= src[x + y*stride]; | 1935 int a0= src[x ]; |
| 1935 int a1= src[x + 1 + y*stride]; | 1936 int a1= src[x + 1]; |
| 1936 int a2= src[x + 2 + y*stride]; | 1937 int a2= src[x + 2]; |
| 1937 int a3= src[x + 3 + y*stride]; | 1938 int a3= src[x + 3]; |
| 1938 int a4= src[x + 4 + y*stride]; | 1939 int a4= src[x + 4]; |
| 1939 int a5= src[x + 5 + y*stride]; | 1940 int a5= src[x + 5]; |
| 1940 // int am= 9*(a1+a2) - (a0+a3); | 1941 // int am= 9*(a1+a2) - (a0+a3); |
| 1941 int am= 20*(a2+a3) - 5*(a1+a4) + (a0+a5); | 1942 int am= 20*(a2+a3) - 5*(a1+a4) + (a0+a5); |
| 1942 // int am= 18*(a2+a3) - 2*(a1+a4); | 1943 // int am= 18*(a2+a3) - 2*(a1+a4); |
| 1943 // int aL= (-7*a0 + 105*a1 + 35*a2 - 5*a3)>>3; | 1944 // int aL= (-7*a0 + 105*a1 + 35*a2 - 5*a3)>>3; |
| 1944 // int aR= (-7*a3 + 105*a2 + 35*a1 - 5*a0)>>3; | 1945 // int aR= (-7*a3 + 105*a2 + 35*a1 - 5*a0)>>3; |
| 1945 | 1946 |
| 1946 // if(b_w==16) am= 8*(a1+a2); | 1947 // if(b_w==16) am= 8*(a1+a2); |
| 1947 | 1948 |
| 1948 if(dx<8) tmp[x + y*stride]= (32*a2*( 8-dx) + am* dx + 128)>>8; | 1949 if(dx<8) tmp[x]= (32*a2*( 8-dx) + am* dx + 128)>>8; |
| 1949 else tmp[x + y*stride]= ( am*(16-dx) + 32*a3*(dx-8) + 128)>>8; | 1950 else tmp[x]= ( am*(16-dx) + 32*a3*(dx-8) + 128)>>8; |
| 1950 | 1951 |
| 1951 /* if (dx< 4) tmp[x + y*stride]= (16*a1*( 4-dx) + aL* dx + 32)>>6; | 1952 /* if (dx< 4) tmp[x + y*stride]= (16*a1*( 4-dx) + aL* dx + 32)>>6; |
| 1952 else if(dx< 8) tmp[x + y*stride]= ( aL*( 8-dx) + am*(dx- 4) + 32)>>6; | 1953 else if(dx< 8) tmp[x + y*stride]= ( aL*( 8-dx) + am*(dx- 4) + 32)>>6; |
| 1953 else if(dx<12) tmp[x + y*stride]= ( am*(12-dx) + aR*(dx- 8) + 32)>>6; | 1954 else if(dx<12) tmp[x + y*stride]= ( am*(12-dx) + aR*(dx- 8) + 32)>>6; |
| 1954 else tmp[x + y*stride]= ( aR*(16-dx) + 16*a2*(dx-12) + 32)>>6;*/ | 1955 else tmp[x + y*stride]= ( aR*(16-dx) + 16*a2*(dx-12) + 32)>>6;*/ |
| 1955 } | 1956 } |
| 1956 } | 1957 tmp += stride; |
| 1958 src += stride; | |
| 1959 } | |
| 1960 tmp -= (b_h+5)*stride; | |
| 1961 | |
| 1957 for(y=0; y < b_h; y++){ | 1962 for(y=0; y < b_h; y++){ |
| 1958 for(x=0; x < b_w; x++){ | 1963 for(x=0; x < b_w; x++){ |
| 1959 int a0= tmp[x + y *stride]; | 1964 int a0= tmp[x + 0*stride]; |
| 1960 int a1= tmp[x + (y + 1)*stride]; | 1965 int a1= tmp[x + 1*stride]; |
| 1961 int a2= tmp[x + (y + 2)*stride]; | 1966 int a2= tmp[x + 2*stride]; |
| 1962 int a3= tmp[x + (y + 3)*stride]; | 1967 int a3= tmp[x + 3*stride]; |
| 1963 int a4= tmp[x + (y + 4)*stride]; | 1968 int a4= tmp[x + 4*stride]; |
| 1964 int a5= tmp[x + (y + 5)*stride]; | 1969 int a5= tmp[x + 5*stride]; |
| 1965 int am= 20*(a2+a3) - 5*(a1+a4) + (a0+a5); | 1970 int am= 20*(a2+a3) - 5*(a1+a4) + (a0+a5); |
| 1966 // int am= 18*(a2+a3) - 2*(a1+a4); | 1971 // int am= 18*(a2+a3) - 2*(a1+a4); |
| 1967 /* int aL= (-7*a0 + 105*a1 + 35*a2 - 5*a3)>>3; | 1972 /* int aL= (-7*a0 + 105*a1 + 35*a2 - 5*a3)>>3; |
| 1968 int aR= (-7*a3 + 105*a2 + 35*a1 - 5*a0)>>3;*/ | 1973 int aR= (-7*a3 + 105*a2 + 35*a1 - 5*a0)>>3;*/ |
| 1969 | 1974 |
| 1970 // if(b_w==16) am= 8*(a1+a2); | 1975 // if(b_w==16) am= 8*(a1+a2); |
| 1971 | 1976 |
| 1972 if(dy<8) dst[x + y*stride]= (32*a2*( 8-dy) + am* dy + 128)>>8; | 1977 if(dy<8) dst[x]= (32*a2*( 8-dy) + am* dy + 128)>>8; |
| 1973 else dst[x + y*stride]= ( am*(16-dy) + 32*a3*(dy-8) + 128)>>8; | 1978 else dst[x]= ( am*(16-dy) + 32*a3*(dy-8) + 128)>>8; |
| 1974 | 1979 |
| 1975 /* if (dy< 4) tmp[x + y*stride]= (16*a1*( 4-dy) + aL* dy + 32)>>6; | 1980 /* if (dy< 4) tmp[x + y*stride]= (16*a1*( 4-dy) + aL* dy + 32)>>6; |
| 1976 else if(dy< 8) tmp[x + y*stride]= ( aL*( 8-dy) + am*(dy- 4) + 32)>>6; | 1981 else if(dy< 8) tmp[x + y*stride]= ( aL*( 8-dy) + am*(dy- 4) + 32)>>6; |
| 1977 else if(dy<12) tmp[x + y*stride]= ( am*(12-dy) + aR*(dy- 8) + 32)>>6; | 1982 else if(dy<12) tmp[x + y*stride]= ( am*(12-dy) + aR*(dy- 8) + 32)>>6; |
| 1978 else tmp[x + y*stride]= ( aR*(16-dy) + 16*a2*(dy-12) + 32)>>6;*/ | 1983 else tmp[x + y*stride]= ( aR*(16-dy) + 16*a2*(dy-12) + 32)>>6;*/ |
| 1979 } | 1984 } |
| 1980 } | 1985 dst += stride; |
| 1986 tmp += stride; | |
| 1987 } | |
| 1988 STOP_TIMER("mc_block") | |
| 1981 } | 1989 } |
| 1982 | 1990 |
| 1983 #define mcb(dx,dy,b_w)\ | 1991 #define mcb(dx,dy,b_w)\ |
| 1984 static void mc_block ## dx ## dy(uint8_t *dst, uint8_t *src, int stride){\ | 1992 static void mc_block ## dx ## dy(uint8_t *dst, uint8_t *src, int stride){\ |
| 1985 uint8_t tmp[stride*(b_w+5)];\ | 1993 uint8_t tmp[stride*(b_w+5)];\ |