From ede4ec1f8f5fd94dccd880199419a1f1b8137ab6 Mon Sep 17 00:00:00 2001 From: James Almer Date: Tue, 14 Jun 2016 12:41:23 -0300 Subject: [PATCH] x86/aacpsdsp: optimize add_squares loop Signed-off-by: James Almer --- libavcodec/x86/aacpsdsp.asm | 14 +++++++++----- 1 file changed, 9 insertions(+), 5 deletions(-) diff --git a/libavcodec/x86/aacpsdsp.asm b/libavcodec/x86/aacpsdsp.asm index d7d7a9a570..e92cbbce08 100644 --- a/libavcodec/x86/aacpsdsp.asm +++ b/libavcodec/x86/aacpsdsp.asm @@ -33,18 +33,22 @@ SECTION .text ;************************************************************************* %macro PS_ADD_SQUARES 1 cglobal ps_add_squares, 3, 3, %1, dst, src, n + shl nd, 3 + add srcq, nq + neg nq + +align 16 .loop: - movaps m0, [srcq] - movaps m1, [srcq+mmsize] + movaps m0, [srcq+nq] + movaps m1, [srcq+nq+mmsize] mulps m0, m0 mulps m1, m1 HADDPS m0, m1, m2 addps m0, [dstq] movaps [dstq], m0 add dstq, mmsize - add srcq, mmsize*2 - sub nd, mmsize/4 - jg .loop + add nq, mmsize*2 + jl .loop REP_RET %endmacro -- 2.11.0