[llvm] [AMDGPU] Infer amdgpu-no-flat-scratch-init attribute in AMDGPUAttributor (PR #94647)

[Matt Arsenault via llvm-commits]

================
@@ -677,6 +693,33 @@ struct AAAMDAttributesFunction : public AAAMDAttributes {
     return !A.checkForAllCallLikeInstructions(DoesNotRetrieve, *this,
                                               UsedAssumedInformation);
   }
+
+  // Returns true if FlatScratchInit is needed, i.e., no-flat-scratch-init is
+  // not to be set.
+  bool needFlatScratchInit(Attributor &A) {
+    // This is called on each callee; false means callee shouldn't have
+    // no-flat-scratch-init.
+    auto CheckForNoFlatScratchInit = [&](Instruction &I) {
+      const auto &CB = cast<CallBase>(I);
+      const Value *CalleeOp = CB.getCalledOperand();
+      const Function *Callee = dyn_cast<Function>(CalleeOp);
+      if (!Callee) // indirect call
+        return CB.isInlineAsm();
+
+      if (Callee->isIntrinsic())
+        return true;
+
+      const auto *CalleeInfo = A.getAAFor<AAAMDAttributes>(
+          *this, IRPosition::function(*Callee), DepClassTy::REQUIRED);
+      return CalleeInfo && CalleeInfo->isAssumed(FLAT_SCRATCH_INIT);
+    };
+
+    bool UsedAssumedInformation = false;
+    // If any callee is false (i.e. need FlatScratchInit),
+    // checkForAllCallLikeInstructions returns false
+    return !A.checkForAllCallLikeInstructions(CheckForNoFlatScratchInit, *this,
----------------
arsenm wrote:

This should look more like how the queue pointer is handled. This is just a slightly more complicated version of checkForQueuePtr. The instruction walk you put in initialize should be handled by checkForAllInstructions looking for addrspacecast 

https://github.com/llvm/llvm-project/pull/94647