On 12/01/2012 06:20 AM, Christophe Gisquet wrote: > Corrected 2 vs 4-spaces tabs and base4 constants. > > > 0007-SBR-DSP-x86-implement-SSE-qmf_deint_bfly.patch > > > From cf3326b20caea42b8654b754abb49f6186743501 Mon Sep 17 00:00:00 2001 > From: Christophe Gisquet <christophe.gisq...@gmail.com> > Date: Mon, 26 Nov 2012 23:12:03 +0100 > Subject: [PATCH 07/11] SBR DSP x86: implement SSE qmf_deint_bfly > > From 713 to 209 cycles on Penrynn. > Having a loop counter is a 7 cycle gain. > Unrolling is another 7 cycle gain. > Working in reverse scan is another 6 cycles. > --- > libavcodec/x86/sbrdsp.asm | 31 +++++++++++++++++++++++++++++++ > libavcodec/x86/sbrdsp_init.c | 2 ++ > 2 files changed, 33 insertions(+), 0 deletions(-) > > diff --git a/libavcodec/x86/sbrdsp.asm b/libavcodec/x86/sbrdsp.asm > index 2304983..9c3ea84 100644 > --- a/libavcodec/x86/sbrdsp.asm > +++ b/libavcodec/x86/sbrdsp.asm > @@ -286,3 +286,34 @@ cglobal sbr_qmf_deint_neg, 2,3,4,v,src,vrev > sub srcq, 32 > cmp vq, vrevq > jl .loop > + > +; sbr_qmf_deint_bfly(float *v, const float *src0, const float *src1) > +cglobal sbr_qmf_deint_bfly, 3,5,8, v,src0,src1,vrev,c
needs INIT_XMM sse > + mov cq, 64*4-2*mmsize > + lea vrevq, [vq + 64*4] > +.loop: > + mova m0, [src0q+cq] > + mova m1, [src1q] > + mova m4, [src0q+cq+mmsize] > + mova m5, [src1q+mmsize] > + mova m2, m0 > + mova m3, m1 > + shufps m2, m2, q0123 > + shufps m3, m3, q0123 > + mova m6, m4 > + mova m7, m5 > + shufps m6, m6, q0123 > + shufps m7, m7, q0123 shufps m2, m0, m0, q0123 shufps m3, m1, m1, q0123 shufps m6, m4, m4, q0123 shufps m7, m5, m5, q0123 An AVX version might be worth testing with that as well. > + addps m5, m2 > + subps m0, m7 > + addps m1, m6 > + subps m4, m3 > + mova [vrevq], m1 > + mova [vrevq+mmsize], m5 > + mova [vq+cq], m0 > + mova [vq+cq+mmsize], m4 > + add src1q, 2*mmsize > + add vrevq, 2*mmsize > + sub cq, 2*mmsize > + jge .loop > + REP_RET Other than that it looks ok. -Justin _______________________________________________ libav-devel mailing list libav-devel@libav.org https://lists.libav.org/mailman/listinfo/libav-devel
