[PATCH] D85584: AMDGPU/GlobalISel: Legalize odd sized loads with widening
Jay Foad via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Mon Aug 17 08:05:27 PDT 2020
foad added inline comments.
================
Comment at: llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp:2303-2308
+ if (Ty.isVector()) {
+ const LLT EltTy = Ty.getElementType();
+ return LLT::vector(PowerOf2Ceil(Ty.getNumElements()), EltTy);
+ }
+
+ return LLT::scalar(PowerOf2Ceil(Ty.getSizeInBits()));
----------------
arsenm wrote:
> foad wrote:
> > `Ty.changeElementSize(PowerOf2Ceil(Ty.getScalarSizeInBits()))`?
> I think that should only be used to handle cases that could be a vector or not, but it’s known scalar here so doesn’t save code
I meant as a replacement for all of lines 2303-2308, but now I see that in the vector case you're increasing the number of elements, not the size of each element.
Could use `Ty.changeNumElements(PowerOf2Ceil(Ty.getNumElements()))` in the vector case.
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D85584/new/
https://reviews.llvm.org/D85584
More information about the llvm-commits
mailing list