[llvm-branch-commits] [llvm] [AMDGPU][UnifyDivergentExitNodes][StructurizeCFG] Add support for callbr instruction with inline-asm (PR #152161)

[Matt Arsenault via llvm-branch-commits]

================
@@ -1397,13 +1396,18 @@ bool StructurizeCFG::makeUniformRegion(Region *R, UniformityInfo &UA) {
 /// Run the transformation for each region found
 bool StructurizeCFG::run(Region *R, DominatorTree *DT,
                          const TargetTransformInfo *TTI) {
-  if (R->isTopLevelRegion())
+  // CallBr and its corresponding direct target blocks are for now ignored by
+  // this pass. This is not a limitation for the currently intended uses cases
+  // of callbr in the AMDGPU backend (cf.
+  // https://discourse.llvm.org/t/rfc-add-callbr-intrinsic-support/86087).
----------------
arsenm wrote:

Probably shouldn't be linking to discourse in the sources 

https://github.com/llvm/llvm-project/pull/152161