[llvm] [AMDGPU][Attributor] Infer `inreg` attribute in `AMDGPUAttributor` (PR #101609)

[Shilei Tian via llvm-commits]

================
@@ -1014,6 +1015,100 @@ struct AAAMDGPUNoAGPR
 
 const char AAAMDGPUNoAGPR::ID = 0;
 
+struct AAAMDGPUInreg
+    : public IRAttribute<Attribute::InReg,
+                         StateWrapper<BooleanState, AbstractAttribute>,
+                         AAAMDGPUInreg> {
+  AAAMDGPUInreg(const IRPosition &IRP, Attributor &A) : IRAttribute(IRP) {}
+
+  /// Create an abstract attribute view for the position \p IRP.
+  static AAAMDGPUInreg &createForPosition(const IRPosition &IRP, Attributor &A);
+
+  /// See AbstractAttribute::getName()
+  const std::string getName() const override { return "AAAMDGPUInreg"; }
+
+  const std::string getAsStr(Attributor *A) const override {
+    return getAssumed() ? "inreg" : "non-inreg";
+  }
+
+  void trackStatistics() const override {}
+
+  /// See AbstractAttribute::getIdAddr()
+  const char *getIdAddr() const override { return &ID; }
+
+  /// This function should return true if the type of the \p AA is AAAMDGPUInreg
+  static bool classof(const AbstractAttribute *AA) {
+    return (AA->getIdAddr() == &ID);
+  }
+
+  /// Unique ID (due to the unique address)
+  static const char ID;
+};
+
+const char AAAMDGPUInreg::ID = 0;
+
+namespace {
+
+struct AAAMDGPUInregArgument : public AAAMDGPUInreg {
+  AAAMDGPUInregArgument(const IRPosition &IRP, Attributor &A)
+      : AAAMDGPUInreg(IRP, A) {}
+
+  void initialize(Attributor &A) override {
+    assert(getAssociatedFunction()->getCallingConv() !=
+           CallingConv::AMDGPU_KERNEL);
+    if (getAssociatedArgument()->hasAttribute(Attribute::InReg))
+      indicateOptimisticFixpoint();
+  }
+
+  ChangeStatus updateImpl(Attributor &A) override {
+    unsigned ArgNo = getAssociatedArgument()->getArgNo();
+
+    auto isUniform = [&](AbstractCallSite ACS) -> bool {
+      CallBase *CB = ACS.getInstruction();
+      Value *V = CB->getArgOperandUse(ArgNo);
+      if (isa<Constant>(V))
+        return true;
----------------
shiltian wrote:

> How can the simplified version of a value, if it is a single other value, not be as uniform as the original?

There might be some confusion here. In my experiment, I only used simplified value when UA says it is not uniform. If UA already tells us it is uniform, we don't need to do this. The idea was to find more opportunities when UA is being conservative, especially when it comes to inter-procedural analysis, which is the advance the attributor framework has.

https://github.com/llvm/llvm-project/pull/101609