diff options
| author | Justin Ruggles <justin.ruggles@gmail.com> | 2012-06-08 23:20:59 -0400 |
|---|---|---|
| committer | Justin Ruggles <justin.ruggles@gmail.com> | 2012-06-18 18:01:14 -0400 |
| commit | 82b2df979069063beb14be340350501c8340f9cd (patch) | |
| tree | 3c6a61d185f9fbab0d21ebce259ef232973d7219 /libavutil/x86/float_dsp.asm | |
| parent | cb5042d02c66aed68643633446f6bf623b72416e (diff) | |
float_dsp: add x86-optimized functions for vector_fmac_scalar()
Diffstat (limited to 'libavutil/x86/float_dsp.asm')
| -rw-r--r-- | libavutil/x86/float_dsp.asm | 47 |
1 files changed, 47 insertions, 0 deletions
diff --git a/libavutil/x86/float_dsp.asm b/libavutil/x86/float_dsp.asm index 53be7ab99a..66ef09398d 100644 --- a/libavutil/x86/float_dsp.asm +++ b/libavutil/x86/float_dsp.asm @@ -19,6 +19,7 @@ ;****************************************************************************** %include "x86inc.asm" +%include "x86util.asm" SECTION .text @@ -53,3 +54,49 @@ VECTOR_FMUL INIT_YMM avx VECTOR_FMUL %endif + +;------------------------------------------------------------------------------ +; void ff_vector_fmac_scalar(float *dst, const float *src, float mul, int len) +;------------------------------------------------------------------------------ + +%macro VECTOR_FMAC_SCALAR 0 +%if UNIX64 +cglobal vector_fmac_scalar, 3,3,3, dst, src, len +%else +cglobal vector_fmac_scalar, 4,4,3, dst, src, mul, len +%endif +%if WIN64 + SWAP 0, 2 +%endif +%if ARCH_X86_32 + VBROADCASTSS m0, mulm +%else + shufps xmm0, xmm0, 0 +%if cpuflag(avx) + vinsertf128 m0, m0, xmm0, 1 +%endif +%endif + lea lenq, [lend*4-2*mmsize] +.loop + mulps m1, m0, [srcq+lenq ] + mulps m2, m0, [srcq+lenq+mmsize] + addps m1, m1, [dstq+lenq ] + addps m2, m2, [dstq+lenq+mmsize] + mova [dstq+lenq ], m1 + mova [dstq+lenq+mmsize], m2 + sub lenq, 2*mmsize + jge .loop +%if mmsize == 32 + vzeroupper + RET +%else + REP_RET +%endif +%endmacro + +INIT_XMM sse +VECTOR_FMAC_SCALAR +%if HAVE_AVX +INIT_YMM avx +VECTOR_FMAC_SCALAR +%endif |