Mercurial > libavcodec.hg

comparison x86/vp8dsp.asm @ 12209:9eef00a43280 libavcodec

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Make mmx VP8 WHT faster Avoid pextrw, since it's slow on many older CPUs. Now it doesn't require mmxext either.
author darkshikari
date Wed, 21 Jul 2010 20:51:01 +0000
parents d38e8565ba05
children baf13deed97e
comparison
equal deleted inserted replaced
12208:5d73c4b4cd37 12209:9eef00a43280
1032 1032
1033 ;----------------------------------------------------------------------------- 1033 ;-----------------------------------------------------------------------------
1034 ; void vp8_luma_dc_wht_mmxext(DCTELEM block[4][4][16], DCTELEM dc[16]) 1034 ; void vp8_luma_dc_wht_mmxext(DCTELEM block[4][4][16], DCTELEM dc[16])
1035 ;----------------------------------------------------------------------------- 1035 ;-----------------------------------------------------------------------------
1036 1036
1037 %macro SCATTER_WHT 1 1037 %macro SCATTER_WHT 3
1038 pextrw r1d, m0, %1 1038 movd r1d, m%1
1039 pextrw r2d, m1, %1 1039 movd r2d, m%2
1040 mov [r0+2*16*0], r1w 1040 mov [r0+2*16*(0+%3)], r1w
1041 mov [r0+2*16*1], r2w 1041 mov [r0+2*16*(1+%3)], r2w
1042 pextrw r1d, m2, %1 1042 shr r1d, 16
1043 pextrw r2d, m3, %1 1043 shr r2d, 16
1044 mov [r0+2*16*2], r1w 1044 psrlq m%1, 32
1045 mov [r0+2*16*3], r2w 1045 psrlq m%2, 32
1046 mov [r0+2*16*(4+%3)], r1w
1047 mov [r0+2*16*(5+%3)], r2w
1048 movd r1d, m%1
1049 movd r2d, m%2
1050 mov [r0+2*16*(8+%3)], r1w
1051 mov [r0+2*16*(9+%3)], r2w
1052 shr r1d, 16
1053 shr r2d, 16
1054 mov [r0+2*16*(12+%3)], r1w
1055 mov [r0+2*16*(13+%3)], r2w
1046 %endmacro 1056 %endmacro
1047 1057
1048 %macro HADAMARD4_1D 4 1058 %macro HADAMARD4_1D 4
1049 SUMSUB_BADC m%2, m%1, m%4, m%3 1059 SUMSUB_BADC m%2, m%1, m%4, m%3
1050 SUMSUB_BADC m%4, m%2, m%3, m%1 1060 SUMSUB_BADC m%4, m%2, m%3, m%1
1051 SWAP %1, %4, %3 1061 SWAP %1, %4, %3
1052 %endmacro 1062 %endmacro
1053 1063
1054 INIT_MMX 1064 INIT_MMX
1055 cglobal vp8_luma_dc_wht_mmxext, 2,3 1065 cglobal vp8_luma_dc_wht_mmx, 2,3
1056 movq m0, [r1] 1066 movq m0, [r1]
1057 movq m1, [r1+8] 1067 movq m1, [r1+8]
1058 movq m2, [r1+16] 1068 movq m2, [r1+16]
1059 movq m3, [r1+24] 1069 movq m3, [r1+24]
1060 HADAMARD4_1D 0, 1, 2, 3 1070 HADAMARD4_1D 0, 1, 2, 3
1063 HADAMARD4_1D 0, 1, 2, 3 1073 HADAMARD4_1D 0, 1, 2, 3
1064 psraw m0, 3 1074 psraw m0, 3
1065 psraw m1, 3 1075 psraw m1, 3
1066 psraw m2, 3 1076 psraw m2, 3
1067 psraw m3, 3 1077 psraw m3, 3
1068 SCATTER_WHT 0 1078 SCATTER_WHT 0, 1, 0
1069 add r0, 2*16*4 1079 SCATTER_WHT 2, 3, 2
1070 SCATTER_WHT 1
1071 add r0, 2*16*4
1072 SCATTER_WHT 2
1073 add r0, 2*16*4
1074 SCATTER_WHT 3
1075 RET 1080 RET
1076 1081
1077 ;----------------------------------------------------------------------------- 1082 ;-----------------------------------------------------------------------------
1078 ; void vp8_h/v_loop_filter_simple_<opt>(uint8_t *dst, int stride, int flim); 1083 ; void vp8_h/v_loop_filter_simple_<opt>(uint8_t *dst, int stride, int flim);
1079 ;----------------------------------------------------------------------------- 1084 ;-----------------------------------------------------------------------------