[llvm] [AMDGPU][PromoteAlloca] Correctly handle a variable vector index (PR #83597)

[Matt Arsenault via llvm-commits]

================
@@ -401,7 +401,8 @@ static Value *promoteAllocaUserToVector(
     // We're loading the full vector.
     Type *AccessTy = Inst->getType();
     TypeSize AccessSize = DL.getTypeStoreSize(AccessTy);
-    if (AccessSize == VecStoreSize && cast<Constant>(Index)->isZeroValue()) {
+    if (AccessSize == VecStoreSize && isa<Constant>(Index) &&
+        cast<Constant>(Index)->isZeroValue()) {
----------------
arsenm wrote:

dyn_cast instead of isa+cast?

https://github.com/llvm/llvm-project/pull/83597