[llvm] [AMDGPU] Invalidate uniformity info after intrinsic combine (PR #222357)

[Yaxun Liu via llvm-commits]

================
@@ -0,0 +1,72 @@
+; REQUIRES: asserts
+; RUN: opt -mtriple=amdgpu10.10-amd-amdhsa -passes=amdgpu-uniform-intrinsic-combine -S < %s | FileCheck %s
+; RUN: opt -mtriple=amdgpu10.10-amd-amdhsa -passes='amdgpu-uniform-intrinsic-combine,reassociate' -verify-analysis-invalidation -disable-output < %s
+
+; Reassociate must not use uniformity information that refers to invalidated
+; cycle information after the combine changes the IR.
+
+; The pass must report a change when erasing an already unused ballot.
+define amdgpu_kernel void @erase_unused_ballot() {
----------------
yxsamliu wrote:

Done. `update_test_checks.py` generated the checks from the first `RUN:` line and skipped the `-disable-output` line as expected.

https://github.com/llvm/llvm-project/pull/222357