[llvm] [SPIRV] Fix code quality issues. (PR #152005)

[Marcos Maronas via llvm-commits]

================
@@ -893,9 +893,11 @@ Type *SPIRVEmitIntrinsics::deduceNestedTypeHelper(
     bool Change = false;
     for (unsigned i = 0; i < U->getNumOperands(); ++i) {
       Value *Op = U->getOperand(i);
-      Type *OpTy = Op->getType();
-      Type *Ty = OpTy;
+      Type *OpTy = nullptr;
+      Type *Ty = nullptr;
       if (Op) {
+        OpTy = Op->getType();
+        Ty = OpTy;
----------------
maarquitos14 wrote:

The problem here, from the point of view of the tool, is that we're deref'ing `Op` just before checking if it is `nullptr`. The chain of thought of the tool is: if `Op` is nullptr-checked, it can be nullptr, so there's risk that you're deref'ing a nullptr. To be fair, it makes sense. `assert(OpTy)` wouldn't prevent the risk of deref'ing `Op` if it's nullptr. I also agree with your point of view, so I'd say we drop `if (Op) {`, and add `assert(Op && "Operands should not be null.");`. Does that work for you?

https://github.com/llvm/llvm-project/pull/152005