diff options
| author | H.J. Lu <hjl.tools@gmail.com> | 2016-01-15 12:58:16 -0800 |
|---|---|---|
| committer | H.J. Lu <hjl.tools@gmail.com> | 2016-01-15 12:58:16 -0800 |
| commit | 28abfeeeea44baab0fc19da56ce005f14c6c98c4 (patch) | |
| tree | 3be31b788de031ef242ee934d190d81c0d0d9ce5 | |
| parent | a806d37903f0db0081a49a440b1f3a82f5e65916 (diff) | |
| download | glibc-hjl/avx512f-mem/prefetcht1.tar.gz glibc-hjl/avx512f-mem/prefetcht1.tar.xz glibc-hjl/avx512f-mem/prefetcht1.zip | |
Use prefetcht1 with non-temporal stores hjl/avx512f-mem/prefetcht1
| -rw-r--r-- | sysdeps/x86_64/multiarch/memcpy-avx512-no-vzeroupper.S | 16 |
1 files changed, 8 insertions, 8 deletions
diff --git a/sysdeps/x86_64/multiarch/memcpy-avx512-no-vzeroupper.S b/sysdeps/x86_64/multiarch/memcpy-avx512-no-vzeroupper.S index 56e727bfc9..754fce1e74 100644 --- a/sysdeps/x86_64/multiarch/memcpy-avx512-no-vzeroupper.S +++ b/sysdeps/x86_64/multiarch/memcpy-avx512-no-vzeroupper.S @@ -349,10 +349,10 @@ L(gobble_256bytes_nt_loop): prefetcht1 0x240(%rsi) prefetcht1 0x280(%rsi) prefetcht1 0x2C0(%rsi) - prefetchnta 0x300(%rsi) - prefetchnta 0x340(%rsi) - prefetchnta 0x380(%rsi) - prefetchnta 0x3C0(%rsi) + prefetcht1 0x300(%rsi) + prefetcht1 0x340(%rsi) + prefetcht1 0x380(%rsi) + prefetcht1 0x3C0(%rsi) vmovdqu64 (%rsi), %zmm0 vmovdqu64 0x40(%rsi), %zmm1 vmovdqu64 0x80(%rsi), %zmm2 @@ -387,10 +387,10 @@ L(gobble_256bytes_nt_loop_bkw): prefetcht1 -0x2C0(%rcx) prefetcht1 -0x280(%rcx) prefetcht1 -0x240(%rcx) - prefetchnta -0x400(%rcx) - prefetchnta -0x3C0(%rcx) - prefetchnta -0x380(%rcx) - prefetchnta -0x340(%rcx) + prefetcht1 -0x400(%rcx) + prefetcht1 -0x3C0(%rcx) + prefetcht1 -0x380(%rcx) + prefetcht1 -0x340(%rcx) vmovdqu64 -0x100(%rcx), %zmm0 vmovdqu64 -0xC0(%rcx), %zmm1 vmovdqu64 -0x80(%rcx), %zmm2 |