[llvm] [AMDGPU] Fix SIFoldOperands miscompiling values that leave a divergent loop (PR #203256)

[via llvm-commits]

================
@@ -976,6 +988,22 @@ bool SIFoldOperandsImpl::isUseSafeToFold(const MachineInstr &MI,
   return !TII->isSDWA(MI);
 }
 
+// An SGPR->VGPR copy inside a divergent loop latches each lane value as it
+// exits. Folding its scalar source into a use after the loop would make every
+// lane read the same reconverged value, so do not fold across the loop exit.
+bool SIFoldOperandsImpl::isTemporallyDivergentUse(
+    const FoldableDef &OpToFold, const MachineInstr &UseMI) const {
+  if (!OpToFold.isReg())
+    return false;
+  const MachineInstr *DefMI = OpToFold.DefMI;
+  if (!DefMI || !DefMI->isCopy() ||
+      TRI->isSGPRReg(*MRI, DefMI->getOperand(0).getReg()) ||
+      !TRI->isSGPRReg(*MRI, OpToFold.getReg()))
+    return false;
+  const MachineLoop *DefLoop = MLI->getLoopFor(DefMI->getParent());
----------------
LU-JOHN wrote:

Should we also check that DefLoop modifies EXEC?

https://github.com/llvm/llvm-project/pull/203256