[llvm] [AMDGPU] Fix undivergent parameter pass using vgpr (PR #178198)

Matt Arsenault via llvm-commits llvm-commits at lists.llvm.org
Tue Jan 27 09:25:50 PST 2026


================
@@ -134,6 +134,9 @@ void TargetLoweringBase::ArgListEntry::setAttributes(const CallBase *Call,
   IsSwiftError = Call->paramHasAttr(ArgIdx, Attribute::SwiftError);
   Alignment = Call->getParamStackAlign(ArgIdx);
   IndirectType = nullptr;
+  Value *Op = Call->getArgOperand(ArgIdx);
+  if (const Argument *Arg = dyn_cast<Argument>(Op))
+    IsArgInReg = Arg->hasInRegAttr();
----------------
arsenm wrote:

Is this hacking around ABI mismatch between caller and callee? Which test needs this? I don't think this is the case in any of the functions here 

https://github.com/llvm/llvm-project/pull/178198


More information about the llvm-commits mailing list