[llvm-branch-commits] [llvm] [AMDGPU] Insert wmma coexec aware anti-hints rules (PR #226397)

[Quentin Colombet via llvm-branch-commits]

================
@@ -0,0 +1,54 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 6
+# RUN: llc -mtriple=amdgpu12.50-amd-amdhsa -passes=amdgpu-pre-ra-optimizations,two-address-instruction,greedy,virt-reg-rewriter,post-RA-hazard-rec -amdgpu-anti-hints-rules=swmmac-war-index %s -o - | FileCheck %s --check-prefix=CHECK
+# RUN: llc -mtriple=amdgpu12.50-amd-amdhsa -passes=amdgpu-pre-ra-optimizations,two-address-instruction,greedy,virt-reg-rewriter,post-RA-hazard-rec -amdgpu-anti-hints-rules=none %s -o - | FileCheck %s --check-prefix=NOAH
+
+--- |
+  define amdgpu_kernel void @war_swmmac_index_src2() #0 { ret void }
+
+  attributes #0 = { nounwind "amdgpu-agpr-alloc"="0" "frame-pointer"="none" }
+...
+
+---
----------------
qcolombet wrote:

Could you add a comment for each new test on what differences we should expect between the with and without the anti-hint as well a brief explanation of why?

https://github.com/llvm/llvm-project/pull/226397