[llvm] [AMDGPU] Enable "amdgpu-uniform-intrinsic-combine" pass in pipeline. (PR #128687)
Christudasan Devadasan via llvm-commits
llvm-commits at lists.llvm.org
Tue Feb 25 22:01:44 PST 2025
================
@@ -1,3 +1,4 @@
+; XFAIL: *
----------------
cdevadas wrote:
> > **$sgpr0 = COPY %4:vgpr_32**
> > SI_RETURN_TO_EPILOG $sgpr0
This COPY shouldn't be here in the first place. ISel should insert readfirstlane for this cross bank register move. We have seen similar problems earlier in certain shader test cases at the return.
https://github.com/llvm/llvm-project/pull/128687
More information about the llvm-commits
mailing list