[X86] Improvements for InstructionCost X86TTIImpl::getScalarizationOverhead
Open
backend:X86
llvm:analysis
- Dominant language
- LLVM
- Stars
- 40.5k
- Forks
- 18.7k
- PR merge metrics
- PR metrics pending
Description
We see good improvements in perf with this change in X86 cost model in X86TTIImpl::getScalarizationOverhead:
```
if (Insert) {
if ((MScalarTy == MVT::i16 && ST->hasSSE2()) ||
(MScalarTy.isInteger() && ST->hasSSE41()) ||
- (MScalarTy == MVT::f32 && ST->hasSSE41())) {
+ (MScalarTy == MVT::f32 && ST->hasSSE41()) ||
+ (MScalarTy == MVT::f64 && ST->hasSSE2())) {
```
@RKSimon could you try to evaluate this?
Contributor guide
Assessment
This issue has not been assessed yet.