diff options
| author | Ronald S. Bultje <rsbultje@gmail.com> | 2013-01-19 22:26:58 -0800 |
|---|---|---|
| committer | Ronald S. Bultje <rsbultje@gmail.com> | 2013-01-22 11:55:42 -0800 |
| commit | 55aa03b9f8f11ebb7535424cc0e5635558590f49 (patch) | |
| tree | e13b12d715aa32cb1d81ed066735af40ea6dbb8a /libavutil/x86/float_dsp.asm | |
| parent | 0881cbf314982cce8448bd12644ce2a6e0b8c576 (diff) | |
| download | ffmpeg-55aa03b9f8f11ebb7535424cc0e5635558590f49.tar.gz | |
floatdsp: move vector_fmul_add from dsputil to avfloatdsp.
Diffstat (limited to 'libavutil/x86/float_dsp.asm')
| -rw-r--r-- | libavutil/x86/float_dsp.asm | 28 |
1 files changed, 28 insertions, 0 deletions
diff --git a/libavutil/x86/float_dsp.asm b/libavutil/x86/float_dsp.asm index 4113fd91e4..70fc1d0310 100644 --- a/libavutil/x86/float_dsp.asm +++ b/libavutil/x86/float_dsp.asm @@ -162,3 +162,31 @@ VECTOR_DMUL_SCALAR INIT_YMM avx VECTOR_DMUL_SCALAR %endif + +;----------------------------------------------------------------------------- +; vector_fmul_add(float *dst, const float *src0, const float *src1, +; const float *src2, int len) +;----------------------------------------------------------------------------- +%macro VECTOR_FMUL_ADD 0 +cglobal vector_fmul_add, 5,5,2, dst, src0, src1, src2, len + lea lenq, [lend*4 - 2*mmsize] +ALIGN 16 +.loop: + mova m0, [src0q + lenq] + mova m1, [src0q + lenq + mmsize] + mulps m0, m0, [src1q + lenq] + mulps m1, m1, [src1q + lenq + mmsize] + addps m0, m0, [src2q + lenq] + addps m1, m1, [src2q + lenq + mmsize] + mova [dstq + lenq], m0 + mova [dstq + lenq + mmsize], m1 + + sub lenq, 2*mmsize + jge .loop + REP_RET +%endmacro + +INIT_XMM sse +VECTOR_FMUL_ADD +INIT_YMM avx +VECTOR_FMUL_ADD |