[llvm-bugs] [Bug 32744] New: [X86][AVX1] Split 256-bit vector non-temporal loads to keep it non-temporal
via llvm-bugs
llvm-bugs at lists.llvm.org
Sat Apr 22 08:29:32 PDT 2017
https://bugs.llvm.org/show_bug.cgi?id=32744
Bug ID: 32744
Summary: [X86][AVX1] Split 256-bit vector non-temporal loads to
keep it non-temporal
Product: libraries
Version: trunk
Hardware: PC
OS: All
Status: NEW
Severity: enhancement
Priority: P
Component: Backend: X86
Assignee: unassignedbugs at nondot.org
Reporter: llvm-dev at redking.me.uk
CC: llvm-bugs at lists.llvm.org, spatel+llvm at rotateright.com
SSE4.1 introduced 128-bit non-temporal loads and AVX2 introduced 256-bit
non-temporal loads.
But for AVX1 targets which have legal 256-bit vectors, we end up using 256-bit
temporal loads again - we need to split the 128-bit loads and concatenate the
results to prevent cache pollution.
llvm/test/CodeGen/X86/nontemporal-loads.ll
define <8 x float> @test_v8f32(<8 x float>* %src) {
; SSE2-LABEL: test_v8f32:
; SSE2: # BB#0:
; SSE2-NEXT: movaps (%rdi), %xmm0
; SSE2-NEXT: movaps 16(%rdi), %xmm1
; SSE2-NEXT: retq
;
; SSE41-LABEL: test_v8f32:
; SSE41: # BB#0:
; SSE41-NEXT: movntdqa (%rdi), %xmm0
; SSE41-NEXT: movntdqa 16(%rdi), %xmm1
; SSE41-NEXT: retq
;
; AVX1-LABEL: test_v8f32:
; AVX1: # BB#0:
; AVX1-NEXT: vmovaps (%rdi), %ymm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: test_v8f32:
; AVX2: # BB#0:
; AVX2-NEXT: vmovntdqa (%rdi), %ymm0
; AVX2-NEXT: retq
;
; AVX512-LABEL: test_v8f32:
; AVX512: # BB#0:
; AVX512-NEXT: vmovntdqa (%rdi), %ymm0
; AVX512-NEXT: retq
%1 = load <8 x float>, <8 x float>* %src, align 32, !nontemporal !1
ret <8 x float> %1
}
!1 = !{i32 1}
--
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20170422/2318646c/attachment.html>
More information about the llvm-bugs
mailing list