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

[Jun Wang 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();
----------------
jwanggit86 wrote:

@ssahasra I fully agree with your description of how it works if `checkForAllCallees()` is used. Based on this, I'm not sure I see the advantage of using `checkForAllCallees()`. I think you still need to check all call-like instructions, and all it does is to move the real work (check for intrinsics, indirect calls etc) to this CallSite attribute. Furthermore, don't you need both a CallSite attribute and a Function attribute, because when you set up the CallSite attribute you need to retrieve some attribute from the callee? Besides, I want to mention that I was not able to find any use of `checkForAllCallees()` or any definition of CallSite attributes in the existing code.
@arsenm @jdoerfert Your thoughts?

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