squiid/llvm
1
0
Fork 0
Code Issues Pull requests Activity

[X86] Limit X86InterleavedAccessGroup to handle the same type case only

Browse source 
The current implementation assumes the destination type of shuffle is the same as the decomposed ones. Add the check to avoid crush when the condition is not satisfied.

This fixes PR37616.

Reviewed By: RKSimon

Differential Revision: https://reviews.llvm.org/D102751
This commit is contained in:
Wang, Pengfei 2021-05-19 18:01:11 +08:00
parent 81b2fcf26f
commit ca23a38e37
2 changed files with 22 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

3
llvm/lib/Target/X86/X86InterleavedAccess.cpp
View file

@ -733,6 +733,9 @@ bool X86InterleavedAccessGroup::lowerIntoOptimizedSequence() {
// results by generating some sort of (optimized) target-specific
// instructions.
if (ShuffleTy->getNumElements() != NumSubVecElems)
return false;
switch (NumSubVecElems) {
default:
return false;

19
llvm/test/CodeGen/X86/x86-interleaved-access.ll
View file

@ -1930,3 +1930,22 @@ define void @splat4_v4i64_load_store(<4 x i64>* %s, <16 x i64>* %d) {
store <16 x i64> %r, <16 x i64>* %d, align 8
ret void
}
define <2 x i64> @PR37616(<16 x i64>* %a0) {
; AVX1-LABEL: PR37616:
; AVX1: # %bb.0:
; AVX1-NEXT: vmovaps 16(%rdi), %xmm0
; AVX1-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm0[0],mem[0]
; AVX1-NEXT: retq
;
; AVX2OR512-LABEL: PR37616:
; AVX2OR512: # %bb.0:
; AVX2OR512-NEXT: vmovaps (%rdi), %ymm0
; AVX2OR512-NEXT: vunpcklpd {{.*#+}} ymm0 = ymm0[0],mem[0],ymm0[2],mem[2]
; AVX2OR512-NEXT: vextractf128 $1, %ymm0, %xmm0
; AVX2OR512-NEXT: vzeroupper
; AVX2OR512-NEXT: retq
%load = load <16 x i64>, <16 x i64>* %a0, align 128
%shuffle = shufflevector <16 x i64> %load, <16 x i64> undef, <2 x i32> <i32 2, i32 6>
ret <2 x i64> %shuffle
}