summaryrefslogtreecommitdiffstats
path: root/chromium/third_party/ffmpeg/libswresample/x86/resample_mmx.h
diff options
context:
space:
mode:
Diffstat (limited to 'chromium/third_party/ffmpeg/libswresample/x86/resample_mmx.h')
-rw-r--r--chromium/third_party/ffmpeg/libswresample/x86/resample_mmx.h183
1 files changed, 179 insertions, 4 deletions
diff --git a/chromium/third_party/ffmpeg/libswresample/x86/resample_mmx.h b/chromium/third_party/ffmpeg/libswresample/x86/resample_mmx.h
index d96fd5a9d2d..2bd48a92955 100644
--- a/chromium/third_party/ffmpeg/libswresample/x86/resample_mmx.h
+++ b/chromium/third_party/ffmpeg/libswresample/x86/resample_mmx.h
@@ -23,7 +23,9 @@
#include "libswresample/swresample_internal.h"
int swri_resample_int16_mmx2 (struct ResampleContext *c, int16_t *dst, const int16_t *src, int *consumed, int src_size, int dst_size, int update_ctx);
-int swri_resample_int16_ssse3(struct ResampleContext *c, int16_t *dst, const int16_t *src, int *consumed, int src_size, int dst_size, int update_ctx);
+int swri_resample_int16_sse2 (struct ResampleContext *c, int16_t *dst, const int16_t *src, int *consumed, int src_size, int dst_size, int update_ctx);
+int swri_resample_float_sse (struct ResampleContext *c, float *dst, const float *src, int *consumed, int src_size, int dst_size, int update_ctx);
+int swri_resample_double_sse2(struct ResampleContext *c, double *dst, const double *src, int *consumed, int src_size, int dst_size, int update_ctx);
DECLARE_ALIGNED(16, const uint64_t, ff_resample_int16_rounder)[2] = { 0x0000000000004000ULL, 0x0000000000000000ULL};
@@ -46,9 +48,38 @@ __asm__ volatile(\
: "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 COMMON_CORE_INT16_SSSE3 \
+#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"\
@@ -58,8 +89,10 @@ __asm__ volatile(\
"paddd %%xmm1, %%xmm0 \n\t"\
"add $16, %0 \n\t"\
" js 1b \n\t"\
- "phaddd %%xmm0, %%xmm0 \n\t"\
- "phaddd %%xmm0, %%xmm0 \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"\
@@ -67,4 +100,146 @@ __asm__ volatile(\
: "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_FLT_SSE \
+ x86_reg len= -4*c->filter_length;\
+__asm__ volatile(\
+ "xorps %%xmm0, %%xmm0 \n\t"\
+ "1: \n\t"\
+ "movups (%1, %0), %%xmm1 \n\t"\
+ "mulps (%2, %0), %%xmm1 \n\t"\
+ "addps %%xmm1, %%xmm0 \n\t"\
+ "add $16, %0 \n\t"\
+ " js 1b \n\t"\
+ "movhlps %%xmm0, %%xmm1 \n\t"\
+ "addps %%xmm1, %%xmm0 \n\t"\
+ "movss %%xmm0, %%xmm1 \n\t"\
+ "shufps $1, %%xmm0, %%xmm0 \n\t"\
+ "addps %%xmm1, %%xmm0 \n\t"\
+ "movss %%xmm0, (%3) \n\t"\
+ : "+r" (len)\
+ : "r" (((uint8_t*)(src+sample_index))-len),\
+ "r" (((uint8_t*)filter)-len),\
+ "r" (dst+dst_index)\
+ XMM_CLOBBERS_ONLY("%xmm0", "%xmm1")\
+);
+
+#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")\
+);
+
+#define COMMON_CORE_DBL_SSE2 \
+ x86_reg len= -8*c->filter_length;\
+__asm__ volatile(\
+ "xorpd %%xmm0, %%xmm0 \n\t"\
+ "1: \n\t"\
+ "movupd (%1, %0), %%xmm1 \n\t"\
+ "mulpd (%2, %0), %%xmm1 \n\t"\
+ "addpd %%xmm1, %%xmm0 \n\t"\
+ "add $16, %0 \n\t"\
+ " js 1b \n\t"\
+ "movhlps %%xmm0, %%xmm1 \n\t"\
+ "addpd %%xmm1, %%xmm0 \n\t"\
+ "movsd %%xmm0, (%3) \n\t"\
+ : "+r" (len)\
+ : "r" (((uint8_t*)(src+sample_index))-len),\
+ "r" (((uint8_t*)filter)-len),\
+ "r" (dst+dst_index)\
+ XMM_CLOBBERS_ONLY("%xmm0", "%xmm1")\
+);
+
+#define LINEAR_CORE_DBL_SSE2 \
+ x86_reg len= -8*c->filter_length;\
+__asm__ volatile(\
+ "xorpd %%xmm0, %%xmm0 \n\t"\
+ "xorpd %%xmm2, %%xmm2 \n\t"\
+ "1: \n\t"\
+ "movupd (%3, %0), %%xmm1 \n\t"\
+ "movapd %%xmm1, %%xmm3 \n\t"\
+ "mulpd (%4, %0), %%xmm1 \n\t"\
+ "mulpd (%5, %0), %%xmm3 \n\t"\
+ "addpd %%xmm1, %%xmm0 \n\t"\
+ "addpd %%xmm3, %%xmm2 \n\t"\
+ "add $16, %0 \n\t"\
+ " js 1b \n\t"\
+ "movhlps %%xmm0, %%xmm1 \n\t"\
+ "movhlps %%xmm2, %%xmm3 \n\t"\
+ "addpd %%xmm1, %%xmm0 \n\t"\
+ "addpd %%xmm3, %%xmm2 \n\t"\
+ "movsd %%xmm0, %1 \n\t"\
+ "movsd %%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")\
);
generated by cgit v1.2.3 (git 2.25.1) at 2024-05-07 22:44:55 +0000