[llvm-branch-commits] [X86][NewPM] Port lower-amx-intrinsics to NewPM (PR #165113)

[Aiden Grossman via llvm-branch-commits]

================
@@ -179,7 +179,18 @@ FunctionPass *createX86LowerAMXTypeLegacyPass();
 
 /// The pass transforms amx intrinsics to scalar operation if the function has
 /// optnone attribute or it is O0.
-FunctionPass *createX86LowerAMXIntrinsicsPass();
+class X86LowerAMXIntrinsicsPass
+    : public PassInfoMixin<X86LowerAMXIntrinsicsPass> {
+private:
+  const TargetMachine *TM;
+
+public:
+  X86LowerAMXIntrinsicsPass(const TargetMachine *TM) : TM(TM) {}
+  PreservedAnalyses run(Function &F, FunctionAnalysisManager &FAM);
+  static bool isRequired() { return true; }
----------------
boomanaiden154 wrote:

I'm not sure we should be using a backend pass enabled at O0 to remove redundant stack load/stores if those are supposed to be cleaned up in the middle end.

Either way, this patch just intends to port the pass, not fix any latent issues or clean up any latent tech debt of this sort.

https://github.com/llvm/llvm-project/pull/165113