From a806d37903f0db0081a49a440b1f3a82f5e65916 Mon Sep 17 00:00:00 2001 From: "H.J. Lu" Date: Fri, 15 Jan 2016 12:14:18 -0800 Subject: Use vmovdqu64 load with vmovntdq store --- sysdeps/x86_64/multiarch/memcpy-avx512-no-vzeroupper.S | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-) (limited to 'sysdeps') diff --git a/sysdeps/x86_64/multiarch/memcpy-avx512-no-vzeroupper.S b/sysdeps/x86_64/multiarch/memcpy-avx512-no-vzeroupper.S index 75bc836d0c..56e727bfc9 100644 --- a/sysdeps/x86_64/multiarch/memcpy-avx512-no-vzeroupper.S +++ b/sysdeps/x86_64/multiarch/memcpy-avx512-no-vzeroupper.S @@ -353,10 +353,10 @@ L(gobble_256bytes_nt_loop): prefetchnta 0x340(%rsi) prefetchnta 0x380(%rsi) prefetchnta 0x3C0(%rsi) - vmovups (%rsi), %zmm0 - vmovups 0x40(%rsi), %zmm1 - vmovups 0x80(%rsi), %zmm2 - vmovups 0xC0(%rsi), %zmm3 + vmovdqu64 (%rsi), %zmm0 + vmovdqu64 0x40(%rsi), %zmm1 + vmovdqu64 0x80(%rsi), %zmm2 + vmovdqu64 0xC0(%rsi), %zmm3 vmovntdq %zmm0, (%rdi) vmovntdq %zmm1, 0x40(%rdi) vmovntdq %zmm2, 0x80(%rdi) @@ -391,10 +391,10 @@ L(gobble_256bytes_nt_loop_bkw): prefetchnta -0x3C0(%rcx) prefetchnta -0x380(%rcx) prefetchnta -0x340(%rcx) - vmovups -0x100(%rcx), %zmm0 - vmovups -0xC0(%rcx), %zmm1 - vmovups -0x80(%rcx), %zmm2 - vmovups -0x40(%rcx), %zmm3 + vmovdqu64 -0x100(%rcx), %zmm0 + vmovdqu64 -0xC0(%rcx), %zmm1 + vmovdqu64 -0x80(%rcx), %zmm2 + vmovdqu64 -0x40(%rcx), %zmm3 vmovntdq %zmm0, -0x100(%r9) vmovntdq %zmm1, -0xC0(%r9) vmovntdq %zmm2, -0x80(%r9) -- cgit 1.4.1