[LLVMbugs] [Bug 22359] New: [X86][AVX] extract_subvector shuffle disaster
bugzilla-daemon at llvm.org
bugzilla-daemon at llvm.org
Tue Jan 27 13:21:10 PST 2015
http://llvm.org/bugs/show_bug.cgi?id=22359
Bug ID: 22359
Summary: [X86][AVX] extract_subvector shuffle disaster
Product: libraries
Version: trunk
Hardware: PC
OS: All
Status: NEW
Severity: normal
Priority: P
Component: Backend: X86
Assignee: unassignedbugs at nondot.org
Reporter: spatel+llvm at rotateright.com
CC: llvmbugs at cs.uiuc.edu
Classification: Unclassified
This is a testcase I was considering using in a patch for bug 21711, but
something is causing the AVX1/AVX2 codegen to go wrong. This doesn't happen
with SSE.
We're trying to store half of a 256-bit vector here:
define void @f7(<8 x float> %v, <2 x float>* %ptr) {
%idx2 = getelementptr inbounds <2 x float>* %ptr, i64 1
%shuffle0 = shufflevector <8 x float> %v, <8 x float> undef, <2 x i32> <i32
0, i32 1>
%shuffle1 = shufflevector <8 x float> %v, <8 x float> undef, <2 x i32> <i32
2, i32 3>
store <2 x float> %shuffle0, <2 x float>* %ptr, align 8
store <2 x float> %shuffle1, <2 x float>* %idx2, align 8
ret void
}
$ ./llc extract.ll -mattr=sse -o -
...
movups %xmm0, (%rdi)
retq
$ ./llc extract.ll -mattr=avx -o -
...
vpermilps $231, %xmm0, %xmm1 ## xmm1 = xmm0[3,1,2,3]
vpermilpd $1, %xmm0, %xmm2 ## xmm2 = xmm0[1,0]
vmovshdup %xmm0, %xmm3 ## xmm3 = xmm0[1,1,3,3]
vinsertps $16, %xmm3, %xmm0, %xmm0 ## xmm0 = xmm0[0],xmm3[0],xmm0[2,3]
vinsertps $32, %xmm2, %xmm0, %xmm0 ## xmm0 = xmm0[0,1],xmm2[0],xmm0[3]
vinsertps $48, %xmm1, %xmm0, %xmm0 ## xmm0 = xmm0[0,1,2],xmm1[0]
vmovups %xmm0, (%rdi)
vzeroupper
retq
--
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/20150127/3002b38f/attachment.html>
More information about the llvm-bugs
mailing list