This fixes regressions exposed in existing AMDGPU tests in a
future commit when all loads are custom lowered.
Details
Details
- Reviewers
ab
Diff Detail
Diff Detail
Event Timeline
Comment Actions
If I understand correctly, ExpandUnalignedStore was correct and the changes are NFC, but ExpandUnalignedLoad wasn't, right?
If so, LGTM.
Otherwise I'm not sure I understand why TLI.getPointerTy(DL, AS) isn't sufficient.
Comment Actions
No, both have the same problem.
I think it's cleaner to get the existing type rather than re-figuring it out from the address space.
Comment Actions
LGTM.
I had a closer look; I see that in ExpandUnalignedStore, there's a problem when StackPtrVT and PtrVT are different.
I think it's cleaner to get the existing type rather than re-figuring it out from the address space.
Agreed.