[PATCH] D94645: [AMDGPU] Fix llvm.amdgcn.init.exec and frame materialization

[Ruiling, Song via Phabricator via llvm-commits]

ruiling added a comment.

Thanks for the patch. Basically LGTM with some minor comments.

> Fix this by moving lowering of llvm.amdgcn.init.exec post-RA.

The commit message need slightly updated as this is not post-RA anymore.



================
Comment at: llvm/lib/Target/AMDGPU/SILowerControlFlow.cpp:746
+  LIS->removeInterval(InputReg);
+  LIS->createAndComputeVirtRegInterval(InputReg);
+  LIS->createAndComputeVirtRegInterval(CountReg);
----------------
I am not sure whether you like to put this under an `if (InputReg.isVirtual())` condition. I guess we may only meet issue when moving this piece of code or the pass around.


================
Comment at: llvm/test/CodeGen/AMDGPU/llvm.amdgcn.init.exec.ll:171-172
+endif:
+  %array0 = alloca [1024 x i32], align 16, addrspace(5)
+  %array1 = alloca [20 x i32], align 16, addrspace(5)
+
----------------
There are some testing failure for Windows, seems dynamic `alloca` not correctly handled for GlobalISel path under Windows. I think we can simply move these two `alloca`s to entry block. The case will be used to only check `llvm.amdgcn.init.exec.from.input` works correctly when placing in non-entry block. Do you have any other idea?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D94645/new/

https://reviews.llvm.org/D94645