Mercurial > libavcodec.hg
changeset 10646:bb14c1586891 libavcodec
slightly faster scalarproduct_and_madd_int16_ssse3 on penryn, no change on conroe
| author | lorenm |
|---|---|
| date | Sat, 05 Dec 2009 17:53:11 +0000 |
| parents | 6f958f237d7d |
| children | 0139a62827ce |
| files | x86/dsputil_yasm.asm |
| diffstat | 1 files changed, 13 insertions(+), 5 deletions(-) [+] |
line wrap: on
line diff
--- a/x86/dsputil_yasm.asm Sat Dec 05 17:51:57 2009 +0000 +++ b/x86/dsputil_yasm.asm Sat Dec 05 17:53:11 2009 +0000 @@ -202,12 +202,20 @@ mova m2, [v3q + orderq] mova m3, [v3q + orderq + mmsize] %endif - pmaddwd m0, [v1q + orderq] - pmaddwd m1, [v1q + orderq + mmsize] + %define t0 [v1q + orderq] + %define t1 [v1q + orderq + mmsize] +%ifdef ARCH_X86_64 + mova m8, t0 + mova m9, t1 + %define t0 m8 + %define t1 m9 +%endif + pmaddwd m0, t0 + pmaddwd m1, t1 pmullw m2, m7 pmullw m3, m7 - paddw m2, [v1q + orderq] - paddw m3, [v1q + orderq + mmsize] + paddw m2, t0 + paddw m3, t1 paddd m6, m0 paddd m6, m1 mova [v1q + orderq], m2 @@ -219,7 +227,7 @@ %endmacro ; int scalarproduct_and_madd_int16(int16_t *v1, int16_t *v2, int16_t *v3, int order, int mul) -cglobal scalarproduct_and_madd_int16_ssse3, 4,5,8, v1, v2, v3, order, mul +cglobal scalarproduct_and_madd_int16_ssse3, 4,5,10, v1, v2, v3, order, mul shl orderq, 1 movd m7, mulm pshuflw m7, m7, 0