[clang] [llvm] [AMDGPU] Add alignment attr & propagate alignment through make.buffer.rsrc inst (PR #145278)

[Shilei Tian via llvm-commits]

================
@@ -5501,7 +5505,32 @@ struct AAAlignCallSiteReturned final
   using Base = AACalleeToCallSite<AAAlign, AAAlignImpl>;
   AAAlignCallSiteReturned(const IRPosition &IRP, Attributor &A)
       : Base(IRP, A) {}
+  ChangeStatus updateImpl(Attributor &A) override {
+    SmallVector<AA::ValueAndContext> Values;
+    SmallVector<Attributor::AlignmentCallbackTy, 1> AligmentCBs =
+        A.getAlignmentCallback(getIRPosition());
+
+    for (Attributor::AlignmentCallbackTy CB : AligmentCBs)
+      CB(getIRPosition(), this, Values);
+
+    if (!Values.empty()) {
+      StateType T;
+      for (AA::ValueAndContext &VAC : Values) {
+        const AAAlign *AA = A.getAAFor<AAAlign>(
+            *this, IRPosition::value(*VAC.getValue()), DepClassTy::REQUIRED);
+        if (AA && this != AA) {
----------------
shiltian wrote:

Not sure if that's a good idea to compare AA directly instead of comparing the associated value.

https://github.com/llvm/llvm-project/pull/145278