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

swr: add int32_to_float_sse2

Browse Source 
could be done for sse/3dnow too if someone wants

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
This commit is contained in:
Michael Niedermayer 2012-04-28 17:04:42 +02:00
parent 95057b1972
commit 832c3b10d2
2 changed files with 51 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

44
libswresample/x86/audio_convert.asm
View File

@ -21,6 +21,10 @@
%include "libavutil/x86/x86inc.asm" %include "libavutil/x86/x86inc.asm"
%include "libavutil/x86/x86util.asm" %include "libavutil/x86/x86util.asm"
SECTION_RODATA
flt2pm31: times 8 dd 4.6566129e-10
SECTION .text SECTION .text
%macro INT16_TO_INT32 1 %macro INT16_TO_INT32 1
@ -55,6 +59,42 @@ int16_to_int32_u_int %+ SUFFIX
REP_RET REP_RET
%endmacro %endmacro
%macro INT32_TO_FLOAT 1
cglobal int32_to_float_%1, 3, 3, 3, dst, src, len
mov srcq, [srcq]
mov dstq, [dstq]
%ifidn %1, a
test dstq, mmsize-1
jne int32_to_float_u_int %+ SUFFIX
test srcq, mmsize-1
jne int32_to_float_u_int %+ SUFFIX
%else
int32_to_float_u_int %+ SUFFIX
%endif
add srcq, lenq
add dstq, lenq
neg lenq
mova m2, [flt2pm31]
.next:
%ifidn %1, a
cvtdq2ps m0, [ srcq+lenq]
cvtdq2ps m1, [mmsize + srcq+lenq]
%else
movu m0, [ srcq+lenq]
movu m1, [mmsize + srcq+lenq]
cvtdq2ps m0, m0
cvtdq2ps m1, m1
%endif
mulps m0, m2
mulps m1, m2
mov%1 [ dstq+lenq], m0
mov%1 [mmsize + dstq+lenq], m1
add lenq, 2*mmsize
jl .next
REP_RET
%endmacro
INIT_MMX mmx INIT_MMX mmx
INT16_TO_INT32 u INT16_TO_INT32 u
INT16_TO_INT32 a INT16_TO_INT32 a
@ -62,3 +102,7 @@ INT16_TO_INT32 a
INIT_XMM sse INIT_XMM sse
INT16_TO_INT32 u INT16_TO_INT32 u
INT16_TO_INT32 a INT16_TO_INT32 a
INIT_XMM sse2
INT32_TO_FLOAT u
INT32_TO_FLOAT a

7
libswresample/x86/swresample_x86.c
View File

@ -26,6 +26,8 @@
MULTI_CAPS_FUNC_DECL(mmx) MULTI_CAPS_FUNC_DECL(mmx)
MULTI_CAPS_FUNC_DECL(sse) MULTI_CAPS_FUNC_DECL(sse)
void ff_int32_to_float_a_sse2(uint8_t **dst, const uint8_t **src, int len);
void swri_audio_convert_init_x86(struct AudioConvert *ac, void swri_audio_convert_init_x86(struct AudioConvert *ac,
enum AVSampleFormat out_fmt, enum AVSampleFormat out_fmt,
enum AVSampleFormat in_fmt, enum AVSampleFormat in_fmt,
@ -44,4 +46,9 @@ void swri_audio_convert_init_x86(struct AudioConvert *ac,
MULTI_CAPS_FUNC(AV_CPU_FLAG_MMX, mmx) MULTI_CAPS_FUNC(AV_CPU_FLAG_MMX, mmx)
MULTI_CAPS_FUNC(AV_CPU_FLAG_SSE, sse) MULTI_CAPS_FUNC(AV_CPU_FLAG_SSE, sse)
if(mm_flags & AV_CPU_FLAG_SSE2) {
if( out_fmt == AV_SAMPLE_FMT_FLT && in_fmt == AV_SAMPLE_FMT_S32 || out_fmt == AV_SAMPLE_FMT_FLTP && in_fmt == AV_SAMPLE_FMT_S32P)
ac->simd_f = ff_int32_to_float_a_sse2;
}
} }