[AMDGPU][NPM] Port AMDGPUSetWavePriority to NPM (#130064)

Author: optimisan

llvm/lib/Target/AMDGPU/AMDGPU.h

@@ -399,6 +399,13 @@ class SILateBranchLoweringPass
   static bool isRequired() { return true; }
 };
 
+class AMDGPUSetWavePriorityPass
+    : public PassInfoMixin<AMDGPUSetWavePriorityPass> {
+public:
+  PreservedAnalyses run(MachineFunction &MF,
+                        MachineFunctionAnalysisManager &MFAM);
+};
+
 FunctionPass *createAMDGPUAnnotateUniformValuesLegacy();
 
 ModulePass *createAMDGPUPrintfRuntimeBinding();
@@ -511,7 +518,7 @@ void initializeGCNPreRAOptimizationsLegacyPass(PassRegistry &);
 extern char &GCNPreRAOptimizationsID;
 
 FunctionPass *createAMDGPUSetWavePriorityPass();
-void initializeAMDGPUSetWavePriorityPass(PassRegistry &);
+void initializeAMDGPUSetWavePriorityLegacyPass(PassRegistry &);
 
 void initializeGCNRewritePartialRegUsesLegacyPass(llvm::PassRegistry &);
 extern char &GCNRewritePartialRegUsesID;

llvm/lib/Target/AMDGPU/AMDGPUPassRegistry.def

@@ -102,6 +102,7 @@ MACHINE_FUNCTION_PASS("amdgpu-mark-last-scratch-load", AMDGPUMarkLastScratchLoad
 MACHINE_FUNCTION_PASS("amdgpu-pre-ra-long-branch-reg", GCNPreRALongBranchRegPass())
 MACHINE_FUNCTION_PASS("amdgpu-reserve-wwm-regs", AMDGPUReserveWWMRegsPass())
 MACHINE_FUNCTION_PASS("amdgpu-rewrite-partial-reg-uses", GCNRewritePartialRegUsesPass())
+MACHINE_FUNCTION_PASS("amdgpu-set-wave-priority", AMDGPUSetWavePriorityPass())
 MACHINE_FUNCTION_PASS("amdgpu-pre-ra-optimizations", GCNPreRAOptimizationsPass())
 MACHINE_FUNCTION_PASS("amdgpu-nsa-reassign", GCNNSAReassignPass())
 MACHINE_FUNCTION_PASS("gcn-create-vopd", GCNCreateVOPDPass())
@@ -133,7 +134,6 @@ MACHINE_FUNCTION_PASS("si-wqm", SIWholeQuadModePass())
 #define DUMMY_MACHINE_FUNCTION_PASS(NAME, CREATE_PASS)
 DUMMY_MACHINE_FUNCTION_PASS("amdgpu-pre-ra-optimizations", GCNPreRAOptimizationsPass())
 DUMMY_MACHINE_FUNCTION_PASS("amdgpu-rewrite-partial-reg-uses", GCNRewritePartialRegUsesPass())
-DUMMY_MACHINE_FUNCTION_PASS("amdgpu-set-wave-priority", AMDGPUSetWavePriorityPass())
 
 DUMMY_MACHINE_FUNCTION_PASS("si-pre-emit-peephole", SIPreEmitPeepholePass())
 // TODO: Move amdgpu-preload-kern-arg-prolog to MACHINE_FUNCTION_PASS since it

llvm/lib/Target/AMDGPU/AMDGPUSetWavePriority.cpp

@@ -19,6 +19,7 @@
 #include "SIInstrInfo.h"
 #include "llvm/ADT/PostOrderIterator.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachinePassManager.h"
 
 using namespace llvm;
 
@@ -40,15 +41,11 @@ struct MBBInfo {
 
 using MBBInfoSet = DenseMap<const MachineBasicBlock *, MBBInfo>;
 
-class AMDGPUSetWavePriority : public MachineFunctionPass {
+class AMDGPUSetWavePriority {
 public:
   static char ID;
 
-  AMDGPUSetWavePriority() : MachineFunctionPass(ID) {}
-
-  StringRef getPassName() const override { return "Set wave priority"; }
-
-  bool runOnMachineFunction(MachineFunction &MF) override;
+  bool run(MachineFunction &MF);
 
 private:
   MachineInstr *BuildSetprioMI(MachineBasicBlock &MBB,
@@ -58,15 +55,31 @@ class AMDGPUSetWavePriority : public MachineFunctionPass {
   const SIInstrInfo *TII;
 };
 
+class AMDGPUSetWavePriorityLegacy : public MachineFunctionPass {
+public:
+  static char ID;
+
+  AMDGPUSetWavePriorityLegacy() : MachineFunctionPass(ID) {}
+
+  StringRef getPassName() const override { return "Set wave priority"; }
+
+  bool runOnMachineFunction(MachineFunction &MF) override {
+    if (skipFunction(MF.getFunction()))
+      return false;
+
+    return AMDGPUSetWavePriority().run(MF);
+  }
+};
+
 } // End anonymous namespace.
 
-INITIALIZE_PASS(AMDGPUSetWavePriority, DEBUG_TYPE, "Set wave priority", false,
-                false)
+INITIALIZE_PASS(AMDGPUSetWavePriorityLegacy, DEBUG_TYPE, "Set wave priority",
+                false, false)
 
-char AMDGPUSetWavePriority::ID = 0;
+char AMDGPUSetWavePriorityLegacy::ID = 0;
 
 FunctionPass *llvm::createAMDGPUSetWavePriorityPass() {
-  return new AMDGPUSetWavePriority();
+  return new AMDGPUSetWavePriorityLegacy();
 }
 
 MachineInstr *
@@ -96,12 +109,21 @@ static bool isVMEMLoad(const MachineInstr &MI) {
   return SIInstrInfo::isVMEM(MI) && MI.mayLoad();
 }
 
-bool AMDGPUSetWavePriority::runOnMachineFunction(MachineFunction &MF) {
+PreservedAnalyses
+llvm::AMDGPUSetWavePriorityPass::run(MachineFunction &MF,
+                                     MachineFunctionAnalysisManager &MFAM) {
+  if (!AMDGPUSetWavePriority().run(MF))
+    return PreservedAnalyses::all();
+
+  return getMachineFunctionPassPreservedAnalyses();
+}
+
+bool AMDGPUSetWavePriority::run(MachineFunction &MF) {
   const unsigned HighPriority = 3;
   const unsigned LowPriority = 0;
 
   Function &F = MF.getFunction();
-  if (skipFunction(F) || !AMDGPU::isEntryFunctionCC(F.getCallingConv()))
+  if (!AMDGPU::isEntryFunctionCC(F.getCallingConv()))
     return false;
 
   const GCNSubtarget &ST = MF.getSubtarget<GCNSubtarget>();

llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp

@@ -2168,9 +2168,8 @@ void AMDGPUCodeGenPassBuilder::addPreEmitPass(AddMachinePass &addPass) const {
 
   addPass(SILateBranchLoweringPass());
 
-  if (isPassEnabled(EnableSetWavePriority, CodeGenOptLevel::Less)) {
-    // TODO: addPass(AMDGPUSetWavePriorityPass());
-  }
+  if (isPassEnabled(EnableSetWavePriority, CodeGenOptLevel::Less))
+    addPass(AMDGPUSetWavePriorityPass());
 
   if (TM.getOptLevel() > CodeGenOptLevel::None) {
     // TODO: addPass(SIPreEmitPeepholePass());

llvm/test/CodeGen/AMDGPU/set-wave-priority.ll

@@ -1,6 +1,11 @@
 ; RUN: llc -mtriple=amdgcn -amdgpu-set-wave-priority=true -o - %s | \
 ; RUN:   FileCheck %s
 
+; RUN: llc -mtriple=amdgcn -stop-after=si-late-branch-lowering -o - %s | \
+; RUN:   llc -x mir -mtriple=amdgcn -passes=amdgpu-set-wave-priority -o - | \
+; RUN:   llc -x mir -mtriple=amdgcn -start-after=si-late-branch-lowering -o - | \
+; RUN:   FileCheck %s
+
 ; CHECK-LABEL: no_setprio:
 ; CHECK-NOT:       s_setprio
 ; CHECK:           ; return to shader part epilog