diff options
Diffstat (limited to 'libswresample/x86/resample_mmx.h')
| -rw-r--r-- | libswresample/x86/resample_mmx.h | 110 |
1 files changed, 0 insertions, 110 deletions
diff --git a/libswresample/x86/resample_mmx.h b/libswresample/x86/resample_mmx.h index 94237b0507..b0ea496361 100644 --- a/libswresample/x86/resample_mmx.h +++ b/libswresample/x86/resample_mmx.h @@ -22,116 +22,6 @@ #include "libavutil/cpu.h" #include "libswresample/swresample_internal.h" -DECLARE_ALIGNED(16, const uint64_t, ff_resample_int16_rounder)[2] = { 0x0000000000004000ULL, 0x0000000000000000ULL}; - -#define COMMON_CORE_INT16_MMX2 \ - x86_reg len= -2*c->filter_length;\ -__asm__ volatile(\ - "movq "MANGLE(ff_resample_int16_rounder)", %%mm0 \n\t"\ - "1: \n\t"\ - "movq (%1, %0), %%mm1 \n\t"\ - "pmaddwd (%2, %0), %%mm1 \n\t"\ - "paddd %%mm1, %%mm0 \n\t"\ - "add $8, %0 \n\t"\ - " js 1b \n\t"\ - "pshufw $0x0E, %%mm0, %%mm1 \n\t"\ - "paddd %%mm1, %%mm0 \n\t"\ - "psrad $15, %%mm0 \n\t"\ - "packssdw %%mm0, %%mm0 \n\t"\ - "movd %%mm0, (%3) \n\t"\ - : "+r" (len)\ - : "r" (((uint8_t*)(src+sample_index))-len),\ - "r" (((uint8_t*)filter)-len),\ - "r" (dst+dst_index)\ - NAMED_CONSTRAINTS_ARRAY_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(\ - "movdqa "MANGLE(ff_resample_int16_rounder)", %%xmm0 \n\t"\ - "1: \n\t"\ - "movdqu (%1, %0), %%xmm1 \n\t"\ - "pmaddwd (%2, %0), %%xmm1 \n\t"\ - "paddd %%xmm1, %%xmm0 \n\t"\ - "add $16, %0 \n\t"\ - " js 1b \n\t"\ - "pshufd $0x0E, %%xmm0, %%xmm1 \n\t"\ - "paddd %%xmm1, %%xmm0 \n\t"\ - "pshufd $0x01, %%xmm0, %%xmm1 \n\t"\ - "paddd %%xmm1, %%xmm0 \n\t"\ - "psrad $15, %%xmm0 \n\t"\ - "packssdw %%xmm0, %%xmm0 \n\t"\ - "movd %%xmm0, (%3) \n\t"\ - : "+r" (len)\ - : "r" (((uint8_t*)(src+sample_index))-len),\ - "r" (((uint8_t*)filter)-len),\ - "r" (dst+dst_index)\ - NAMED_CONSTRAINTS_ARRAY_ADD(ff_resample_int16_rounder)\ - XMM_CLOBBERS_ONLY("%xmm0", "%xmm1")\ -); - -#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_DBL_SSE2 \ x86_reg len= -8*c->filter_length;\ __asm__ volatile(\ |