virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2024-12-23 12:43:46 +02:00
Code Issues Releases Activity

swresample/resample: sse float linear interpolation

Browse Source 
About two times faster

Signed-off-by: James Almer <jamrial@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
This commit is contained in:
James Almer 2014-03-23 19:05:17 -03:00 committed by Michael Niedermayer
parent fa25c4c400
commit 63dbba655e
2 changed files with 36 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
libswresample/resample_template.c
View File

@ -48,6 +48,7 @@
# define RENAME(N) N ## _float # define RENAME(N) N ## _float
# elif defined(TEMPLATE_RESAMPLE_FLT_SSE) # elif defined(TEMPLATE_RESAMPLE_FLT_SSE)
# define COMMON_CORE COMMON_CORE_FLT_SSE # define COMMON_CORE COMMON_CORE_FLT_SSE
# define LINEAR_CORE LINEAR_CORE_FLT_SSE
# define RENAME(N) N ## _float_sse # define RENAME(N) N ## _float_sse
# endif # endif

35
libswresample/x86/resample_mmx.h
View File

@ -156,3 +156,38 @@ __asm__ volatile(\
"r" (((uint8_t*)filter)-len),\ "r" (((uint8_t*)filter)-len),\
"r" (dst+dst_index)\ "r" (dst+dst_index)\
); );
#define LINEAR_CORE_FLT_SSE \
x86_reg len= -4*c->filter_length;\
__asm__ volatile(\
"xorps %%xmm0, %%xmm0 \n\t"\
"xorps %%xmm2, %%xmm2 \n\t"\
"1: \n\t"\
"movups (%3, %0), %%xmm1 \n\t"\
"movaps %%xmm1, %%xmm3 \n\t"\
"mulps (%4, %0), %%xmm1 \n\t"\
"mulps (%5, %0), %%xmm3 \n\t"\
"addps %%xmm1, %%xmm0 \n\t"\
"addps %%xmm3, %%xmm2 \n\t"\
"add $16, %0 \n\t"\
" js 1b \n\t"\
"movhlps %%xmm0, %%xmm1 \n\t"\
"movhlps %%xmm2, %%xmm3 \n\t"\
"addps %%xmm1, %%xmm0 \n\t"\
"addps %%xmm3, %%xmm2 \n\t"\
"movss %%xmm0, %%xmm1 \n\t"\
"movss %%xmm2, %%xmm3 \n\t"\
"shufps $1, %%xmm0, %%xmm0 \n\t"\
"shufps $1, %%xmm2, %%xmm2 \n\t"\
"addps %%xmm1, %%xmm0 \n\t"\
"addps %%xmm3, %%xmm2 \n\t"\
"movss %%xmm0, %1 \n\t"\
"movss %%xmm2, %2 \n\t"\
: "+r" (len),\
"=m" (val),\
"=m" (v2)\
: "r" (((uint8_t*)(src+sample_index))-len),\
"r" (((uint8_t*)filter)-len),\
"r" (((uint8_t*)(filter+c->filter_alloc))-len)\
XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3")\
);