2 Commits

Author SHA1 Message Date
Matt Arsenault
7a84624079 AMDGPU: Make various vector undefs legal
Surprisingly these were getting legalized to something
zero initialized.

This fixes an infinite loop when combining some vector types.
Also fixes zero initializing some undef values.

SimplifyDemandedVectorElts / SimplifyDemandedBits are not checking
for the legality of the output undefs they are replacing unused
operations with. This resulted in turning vectors into undefs
that were later re-legalized back into zero vectors.
2022-09-28 10:48:52 -04:00
Stanislav Mekhanoshin
6e0fa292c2 [AMDGPU] Change register type for v32 vectors
When it is AReg_1024 this results in unnecessary copying into
AGPRs of a 32 element vectors even though they are not intended
for an mfma instruction.

Differential Revision: https://reviews.llvm.org/D64815

llvm-svn: 366252
2019-07-16 20:06:00 +00:00