diff options
| author | James Almer <jamrial@gmail.com> | 2014-03-23 19:05:16 -0300 |
|---|---|---|
| committer | Michael Niedermayer <michaelni@gmx.at> | 2014-03-24 02:33:16 +0100 |
| commit | fa25c4c400649bcc7693107d2d4b9d1fa137173e (patch) | |
| tree | 548982ee6e1c438be8f89881e6ccca6f704d36c8 /libswresample/x86 | |
| parent | ffd77f94a26be22b8ead3178ceec3ed39e68abc5 (diff) | |
swresample/resample: mmx2/sse2 int16 linear interpolation
About three times faster
Signed-off-by: James Almer <jamrial@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Diffstat (limited to 'libswresample/x86')
| -rw-r--r-- | libswresample/x86/resample_mmx.h | 61 |
1 files changed, 61 insertions, 0 deletions
diff --git a/libswresample/x86/resample_mmx.h b/libswresample/x86/resample_mmx.h index ba36de9e5c..28a317ce78 100644 --- a/libswresample/x86/resample_mmx.h +++ b/libswresample/x86/resample_mmx.h @@ -50,6 +50,34 @@ __asm__ volatile(\ NAMED_CONSTRAINTS_ADD(ff_resample_int16_rounder)\ ); +#define LINEAR_CORE_INT16_MMX2 \ + x86_reg len= -2*c->filter_length;\ +__asm__ volatile(\ + "pxor %%mm0, %%mm0 \n\t"\ + "pxor %%mm2, %%mm2 \n\t"\ + "1: \n\t"\ + "movq (%3, %0), %%mm1 \n\t"\ + "movq %%mm1, %%mm3 \n\t"\ + "pmaddwd (%4, %0), %%mm1 \n\t"\ + "pmaddwd (%5, %0), %%mm3 \n\t"\ + "paddd %%mm1, %%mm0 \n\t"\ + "paddd %%mm3, %%mm2 \n\t"\ + "add $8, %0 \n\t"\ + " js 1b \n\t"\ + "pshufw $0x0E, %%mm0, %%mm1 \n\t"\ + "pshufw $0x0E, %%mm2, %%mm3 \n\t"\ + "paddd %%mm1, %%mm0 \n\t"\ + "paddd %%mm3, %%mm2 \n\t"\ + "movd %%mm0, %1 \n\t"\ + "movd %%mm2, %2 \n\t"\ + : "+r" (len),\ + "=r" (val),\ + "=r" (v2)\ + : "r" (((uint8_t*)(src+sample_index))-len),\ + "r" (((uint8_t*)filter)-len),\ + "r" (((uint8_t*)(filter+c->filter_alloc))-len)\ +); + #define COMMON_CORE_INT16_SSE2 \ x86_reg len= -2*c->filter_length;\ __asm__ volatile(\ @@ -74,6 +102,39 @@ __asm__ volatile(\ NAMED_CONSTRAINTS_ADD(ff_resample_int16_rounder)\ ); +#define LINEAR_CORE_INT16_SSE2 \ + x86_reg len= -2*c->filter_length;\ +__asm__ volatile(\ + "pxor %%xmm0, %%xmm0 \n\t"\ + "pxor %%xmm2, %%xmm2 \n\t"\ + "1: \n\t"\ + "movdqu (%3, %0), %%xmm1 \n\t"\ + "movdqa %%xmm1, %%xmm3 \n\t"\ + "pmaddwd (%4, %0), %%xmm1 \n\t"\ + "pmaddwd (%5, %0), %%xmm3 \n\t"\ + "paddd %%xmm1, %%xmm0 \n\t"\ + "paddd %%xmm3, %%xmm2 \n\t"\ + "add $16, %0 \n\t"\ + " js 1b \n\t"\ + "pshufd $0x0E, %%xmm0, %%xmm1 \n\t"\ + "pshufd $0x0E, %%xmm2, %%xmm3 \n\t"\ + "paddd %%xmm1, %%xmm0 \n\t"\ + "paddd %%xmm3, %%xmm2 \n\t"\ + "pshufd $0x01, %%xmm0, %%xmm1 \n\t"\ + "pshufd $0x01, %%xmm2, %%xmm3 \n\t"\ + "paddd %%xmm1, %%xmm0 \n\t"\ + "paddd %%xmm3, %%xmm2 \n\t"\ + "movd %%xmm0, %1 \n\t"\ + "movd %%xmm2, %2 \n\t"\ + : "+r" (len),\ + "=r" (val),\ + "=r" (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")\ +); + #define COMMON_CORE_FLT_SSE \ x86_reg len= -4*c->filter_length;\ __asm__ volatile(\ |