[llvm] [FunctionAttrs] Improve handling of alias-preserving intrinsic calls (PR #68453)

[Krzysztof Drewniak via llvm-commits]

================
@@ -0,0 +1,47 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --function-signature --check-attributes
+; RUN: opt -passes=function-attrs -S < %s | FileCheck --check-prefixes=COMMON,FNATTRS %s
+; RUN: opt -passes=attributor-light -S < %s | FileCheck --check-prefixes=COMMON,ATTRIBUTOR %s
+
+target triple = "amdgcn-amd-amdhsa"
----------------
krzysz00 wrote:

Huh, then yeah, we can probably make this not target-specific, but I vaguely remember running into issues with AMDGPU-specific tests before that made me cautious.

https://github.com/llvm/llvm-project/pull/68453