CodeGen][NewPM] Port PostRAScheduler to NPM. (#125798)

llvm/include/llvm/CodeGen/PostRASchedulerList.h

@@ -0,0 +1,32 @@
+//===- llvm/CodeGen/PostRASchedulerList.h ------------------------*- C++-*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_CODEGEN_POSTRASCHEDULERLIST_H
+#define LLVM_CODEGEN_POSTRASCHEDULERLIST_H
+
+#include "llvm/CodeGen/MachinePassManager.h"
+
+namespace llvm {
+
+class PostRASchedulerPass : public PassInfoMixin<PostRASchedulerPass> {
+  const TargetMachine *TM;
+
+public:
+  PostRASchedulerPass(const TargetMachine *TM) : TM(TM) {}
+  PreservedAnalyses run(MachineFunction &MF,
+                        MachineFunctionAnalysisManager &MFAM);
+
+  MachineFunctionProperties getRequiredProperties() const {
+    return MachineFunctionProperties().set(
+        MachineFunctionProperties::Property::NoVRegs);
+  }
+};
+
+} // namespace llvm
+
+#endif // LLVM_CODEGEN_POSTRASCHEDULERLIST_H

llvm/include/llvm/InitializePasses.h

@@ -241,7 +241,7 @@ void initializePostInlineEntryExitInstrumenterPass(PassRegistry &);
 void initializePostMachineSchedulerLegacyPass(PassRegistry &);
 void initializePostRAHazardRecognizerPass(PassRegistry &);
 void initializePostRAMachineSinkingPass(PassRegistry &);
-void initializePostRASchedulerPass(PassRegistry &);
+void initializePostRASchedulerLegacyPass(PassRegistry &);
 void initializePreISelIntrinsicLoweringLegacyPassPass(PassRegistry &);
 void initializePrintFunctionPassWrapperPass(PassRegistry &);
 void initializePrintModulePassWrapperPass(PassRegistry &);

llvm/include/llvm/Passes/CodeGenPassBuilder.h

@@ -55,6 +55,7 @@
 #include "llvm/CodeGen/OptimizePHIs.h"
 #include "llvm/CodeGen/PHIElimination.h"
 #include "llvm/CodeGen/PeepholeOptimizer.h"
+#include "llvm/CodeGen/PostRASchedulerList.h"
 #include "llvm/CodeGen/PreISelIntrinsicLowering.h"
 #include "llvm/CodeGen/RegAllocFast.h"
 #include "llvm/CodeGen/RegUsageInfoCollector.h"
@@ -960,7 +961,7 @@ Error CodeGenPassBuilder<Derived, TargetMachineT>::addMachinePasses(
     if (Opt.MISchedPostRA)
       addPass(PostMachineSchedulerPass(&TM));
     else
-      addPass(PostRASchedulerPass());
+      addPass(PostRASchedulerPass(&TM));
   }
 
   // GC

llvm/include/llvm/Passes/MachinePassRegistry.def

@@ -149,6 +149,7 @@ MACHINE_FUNCTION_PASS("opt-phis", OptimizePHIsPass())
 MACHINE_FUNCTION_PASS("peephole-opt", PeepholeOptimizerPass())
 MACHINE_FUNCTION_PASS("phi-node-elimination", PHIEliminationPass())
 MACHINE_FUNCTION_PASS("postmisched", PostMachineSchedulerPass(TM))
+MACHINE_FUNCTION_PASS("post-RA-sched", PostRASchedulerPass(TM))
 MACHINE_FUNCTION_PASS("print", PrintMIRPass())
 MACHINE_FUNCTION_PASS("print<livedebugvars>", LiveDebugVariablesPrinterPass(errs()))
 MACHINE_FUNCTION_PASS("print<live-intervals>", LiveIntervalsPrinterPass(errs()))
@@ -247,7 +248,6 @@ DUMMY_MACHINE_FUNCTION_PASS("machine-uniformity", MachineUniformityInfoWrapperPa
 DUMMY_MACHINE_FUNCTION_PASS("machineinstr-printer", MachineFunctionPrinterPass)
 DUMMY_MACHINE_FUNCTION_PASS("mirfs-discriminators", MIRAddFSDiscriminatorsPass)
 DUMMY_MACHINE_FUNCTION_PASS("patchable-function", PatchableFunctionPass)
-DUMMY_MACHINE_FUNCTION_PASS("post-RA-sched", PostRASchedulerPass)
 DUMMY_MACHINE_FUNCTION_PASS("postra-machine-sink", PostRAMachineSinkingPass)
 DUMMY_MACHINE_FUNCTION_PASS("postrapseudos", ExpandPostRAPseudosPass)
 DUMMY_MACHINE_FUNCTION_PASS("print-machine-cycles", MachineCycleInfoPrinterPass)

llvm/lib/CodeGen/CodeGen.cpp

@@ -108,7 +108,7 @@ void llvm::initializeCodeGen(PassRegistry &Registry) {
   initializePostMachineSchedulerLegacyPass(Registry);
   initializePostRAHazardRecognizerPass(Registry);
   initializePostRAMachineSinkingPass(Registry);
-  initializePostRASchedulerPass(Registry);
+  initializePostRASchedulerLegacyPass(Registry);
   initializePreISelIntrinsicLoweringLegacyPassPass(Registry);
   initializeProcessImplicitDefsPass(Registry);
   initializeRABasicPass(Registry);

llvm/lib/CodeGen/PostRASchedulerList.cpp

@@ -17,6 +17,7 @@
 //
 //===----------------------------------------------------------------------===//
 
+#include "llvm/CodeGen/PostRASchedulerList.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/Analysis/AliasAnalysis.h"
 #include "llvm/CodeGen/AntiDepBreaker.h"
@@ -39,6 +40,7 @@
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetMachine.h"
 using namespace llvm;
 
 #define DEBUG_TYPE "post-RA-sched"
@@ -73,13 +75,24 @@ DebugMod("postra-sched-debugmod",
 AntiDepBreaker::~AntiDepBreaker() = default;
 
 namespace {
-  class PostRAScheduler : public MachineFunctionPass {
+class PostRAScheduler {
   const TargetInstrInfo *TII = nullptr;
+  MachineLoopInfo *MLI = nullptr;
+  AliasAnalysis *AA = nullptr;
+  const TargetMachine *TM = nullptr;
   RegisterClassInfo RegClassInfo;
 
+public:
+  PostRAScheduler(MachineFunction &MF, MachineLoopInfo *MLI, AliasAnalysis *AA,
+                  const TargetMachine *TM)
+      : TII(MF.getSubtarget().getInstrInfo()), MLI(MLI), AA(AA), TM(TM) {}
+  bool run(MachineFunction &MF);
+};
+
+class PostRASchedulerLegacy : public MachineFunctionPass {
 public:
   static char ID;
-    PostRAScheduler() : MachineFunctionPass(ID) {}
+  PostRASchedulerLegacy() : MachineFunctionPass(ID) {}
 
   void getAnalysisUsage(AnalysisUsage &AU) const override {
     AU.setPreservesCFG();
@@ -99,7 +112,7 @@ namespace {
 
   bool runOnMachineFunction(MachineFunction &Fn) override;
 };
-  char PostRAScheduler::ID = 0;
+char PostRASchedulerLegacy::ID = 0;
 
 class SchedulePostRATDList : public ScheduleDAGInstrs {
   /// AvailableQueue - The priority queue to use for the available SUnits.
@@ -151,8 +164,7 @@ namespace {
   void setEndIndex(unsigned EndIdx) { EndIndex = EndIdx; }
 
   /// Initialize the scheduler state for the next scheduling region.
-    void enterRegion(MachineBasicBlock *bb,
+  void enterRegion(MachineBasicBlock *bb, MachineBasicBlock::iterator begin,
-                     MachineBasicBlock::iterator begin,
                    MachineBasicBlock::iterator end,
                    unsigned regioninstrs) override;
 
@@ -186,11 +198,11 @@ namespace {
   void dumpSchedule() const;
   void emitNoop(unsigned CurCycle);
 };
-}
+} // namespace
 
-char &llvm::PostRASchedulerID = PostRAScheduler::ID;
+char &llvm::PostRASchedulerID = PostRASchedulerLegacy::ID;
 
-INITIALIZE_PASS(PostRAScheduler, DEBUG_TYPE,
+INITIALIZE_PASS(PostRASchedulerLegacy, DEBUG_TYPE,
                 "Post RA top-down list latency scheduler", false, false)
 
 SchedulePostRATDList::SchedulePostRATDList(
@@ -263,19 +275,12 @@ static bool enablePostRAScheduler(const TargetSubtargetInfo &ST,
          OptLevel >= ST.getOptLevelToEnablePostRAScheduler();
 }
 
-bool PostRAScheduler::runOnMachineFunction(MachineFunction &Fn) {
+bool PostRAScheduler::run(MachineFunction &MF) {
-  if (skipFunction(Fn.getFunction()))
+  const auto &Subtarget = MF.getSubtarget();
-    return false;
-
-  const auto &Subtarget = Fn.getSubtarget();
-  TargetPassConfig *PassConfig = &getAnalysis<TargetPassConfig>();
   // Check that post-RA scheduling is enabled for this target.
-  if (!enablePostRAScheduler(Subtarget, PassConfig->getOptLevel()))
+  if (!enablePostRAScheduler(Subtarget, TM->getOptLevel()))
     return false;
 
-  TII = Subtarget.getInstrInfo();
-  MachineLoopInfo &MLI = getAnalysis<MachineLoopInfoWrapperPass>().getLI();
-  AliasAnalysis *AA = &getAnalysis<AAResultsWrapperPass>().getAAResults();
   TargetSubtargetInfo::AntiDepBreakMode AntiDepMode =
       Subtarget.getAntiDepBreakMode();
   if (EnableAntiDepBreaking.getPosition() > 0) {
@@ -287,22 +292,22 @@ bool PostRAScheduler::runOnMachineFunction(MachineFunction &Fn) {
   }
   SmallVector<const TargetRegisterClass *, 4> CriticalPathRCs;
   Subtarget.getCriticalPathRCs(CriticalPathRCs);
-  RegClassInfo.runOnMachineFunction(Fn);
+  RegClassInfo.runOnMachineFunction(MF);
 
   LLVM_DEBUG(dbgs() << "PostRAScheduler\n");
 
-  SchedulePostRATDList Scheduler(Fn, MLI, AA, RegClassInfo, AntiDepMode,
+  SchedulePostRATDList Scheduler(MF, *MLI, AA, RegClassInfo, AntiDepMode,
                                  CriticalPathRCs);
 
   // Loop over all of the basic blocks
-  for (auto &MBB : Fn) {
+  for (auto &MBB : MF) {
 #ifndef NDEBUG
     // If DebugDiv > 0 then only schedule MBB with (ID % DebugDiv) == DebugMod
     if (DebugDiv > 0) {
       static int bbcnt = 0;
       if (bbcnt++ % DebugDiv != DebugMod)
         continue;
-      dbgs() << "*** DEBUG scheduling " << Fn.getName() << ":"
+      dbgs() << "*** DEBUG scheduling " << MF.getName() << ":"
              << printMBBReference(MBB) << " ***\n";
     }
 #endif
@@ -320,7 +325,7 @@ bool PostRAScheduler::runOnMachineFunction(MachineFunction &Fn) {
       // Calls are not scheduling boundaries before register allocation, but
       // post-ra we don't gain anything by scheduling across calls since we
       // don't need to worry about register pressure.
-      if (MI.isCall() || TII->isSchedulingBoundary(MI, &MBB, Fn)) {
+      if (MI.isCall() || TII->isSchedulingBoundary(MI, &MBB, MF)) {
         Scheduler.enterRegion(&MBB, I, Current, CurrentCount - Count);
         Scheduler.setEndIndex(CurrentCount);
         Scheduler.schedule();
@@ -353,6 +358,39 @@ bool PostRAScheduler::runOnMachineFunction(MachineFunction &Fn) {
   return true;
 }
 
+bool PostRASchedulerLegacy::runOnMachineFunction(MachineFunction &MF) {
+  if (skipFunction(MF.getFunction()))
+    return false;
+
+  MachineLoopInfo *MLI = &getAnalysis<MachineLoopInfoWrapperPass>().getLI();
+  AliasAnalysis *AA = &getAnalysis<AAResultsWrapperPass>().getAAResults();
+  const TargetMachine *TM =
+      &getAnalysis<TargetPassConfig>().getTM<TargetMachine>();
+  PostRAScheduler Impl(MF, MLI, AA, TM);
+  return Impl.run(MF);
+}
+
+PreservedAnalyses
+PostRASchedulerPass::run(MachineFunction &MF,
+                         MachineFunctionAnalysisManager &MFAM) {
+  MFPropsModifier _(*this, MF);
+
+  MachineLoopInfo *MLI = &MFAM.getResult<MachineLoopAnalysis>(MF);
+  auto &FAM = MFAM.getResult<FunctionAnalysisManagerMachineFunctionProxy>(MF)
+                  .getManager();
+  AliasAnalysis *AA = &FAM.getResult<AAManager>(MF.getFunction());
+  PostRAScheduler Impl(MF, MLI, AA, TM);
+  bool Changed = Impl.run(MF);
+  if (!Changed)
+    return PreservedAnalyses::all();
+
+  PreservedAnalyses PA = getMachineFunctionPassPreservedAnalyses();
+  PA.preserveSet<CFGAnalyses>();
+  PA.preserve<MachineDominatorTreeAnalysis>();
+  PA.preserve<MachineLoopAnalysis>();
+  return PA;
+}
+
 /// StartBlock - Initialize register live-range state for scheduling in
 /// this block.
 ///

llvm/lib/Passes/PassBuilder.cpp

@@ -125,6 +125,7 @@
 #include "llvm/CodeGen/OptimizePHIs.h"
 #include "llvm/CodeGen/PHIElimination.h"
 #include "llvm/CodeGen/PeepholeOptimizer.h"
+#include "llvm/CodeGen/PostRASchedulerList.h"
 #include "llvm/CodeGen/PreISelIntrinsicLowering.h"
 #include "llvm/CodeGen/RegAllocFast.h"
 #include "llvm/CodeGen/RegUsageInfoCollector.h"

llvm/test/CodeGen/AMDGPU/bundle-latency.mir

@@ -1,5 +1,6 @@
 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
 # RUN: llc -mtriple=amdgcn -mcpu=gfx900 -verify-machineinstrs -run-pass=post-RA-sched %s -o - | FileCheck -check-prefix=GCN %s
+# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -passes=post-RA-sched %s -o - | FileCheck -check-prefix=GCN %s
 
 # Check that we move consumer further from producer, even if one of them is in a bundle.
 

llvm/test/CodeGen/AMDGPU/cluster-flat-loads-postra.mir

@@ -1,4 +1,5 @@
 # RUN: llc -mtriple=amdgcn -mcpu=tonga -mattr=-xnack -run-pass post-RA-sched -verify-machineinstrs -o - %s | FileCheck -check-prefix=GCN %s
+# RUN: llc -mtriple=amdgcn -mcpu=tonga -mattr=-xnack -passes=post-RA-sched -o - %s | FileCheck -check-prefix=GCN %s
 
 # GCN:      FLAT_LOAD_DWORD
 # GCN-NEXT: FLAT_LOAD_DWORD

llvm/test/CodeGen/AMDGPU/hazard-kill.mir

@@ -1,4 +1,5 @@
 # RUN: llc -mtriple=amdgcn -mcpu=gfx900 -verify-machineinstrs -run-pass post-RA-sched %s -o - | FileCheck -check-prefix=GFX90 %s
+# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -passes=post-RA-sched %s -o - | FileCheck -check-prefix=GFX90 %s
 
 # This tests that a KILL isn't considered as a valid instruction for a hazard
 # slot (e.g. m0 def followed by V_INTERP for gfx9)

llvm/test/CodeGen/AMDGPU/misched-killflags.mir

@@ -1,4 +1,5 @@
 # RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=fiji -verify-machineinstrs -run-pass=post-RA-sched -o - %s | FileCheck %s
+# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=fiji -passes=post-RA-sched -o - %s | FileCheck %s
 # Make sure ScheduleDAGInstrs::fixupKills does not produce invalid kill flags.
 ---
 name: func0

llvm/test/CodeGen/AMDGPU/movrels-bug.mir

@@ -1,4 +1,5 @@
 # RUN: llc -mtriple=amdgcn -verify-machineinstrs -run-pass post-RA-sched  %s -o - | FileCheck %s
+# RUN: llc -mtriple=amdgcn -passes=post-RA-sched  %s -o - | FileCheck %s
 
 # This tests a situation where a sub-register of a killed super-register operand
 # of V_MOVRELS happens to have an undef use later on. This leads to the post RA

llvm/test/CodeGen/AMDGPU/post-ra-sched-kill-bundle-use-inst.mir

@@ -1,5 +1,6 @@
 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
 # RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -run-pass=post-RA-sched -verify-machineinstrs -o -  %s | FileCheck %s
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -passes=post-RA-sched -o -  %s | FileCheck %s
 
 # The scheduler was not inspecting the first instruction in the bundle
 # when adding kill flags, so it would incorrectly mark the first use

llvm/test/CodeGen/ARM/vldmia-sched.mir

@@ -1,4 +1,5 @@
 # RUN: llc -run-pass=post-RA-sched %s -o - | FileCheck %s
+# RUN: llc -passes=post-RA-sched %s -o - | FileCheck %s
 # CHECK: VLDMDIA
 --- |
   target triple = "thumbv7-w64-windows-gnu"

llvm/test/CodeGen/Hexagon/bank-conflict-load.mir

@@ -1,4 +1,5 @@
 # RUN: llc -mtriple=hexagon -run-pass post-RA-sched %s -o - | FileCheck %s
+# RUN: llc -mtriple=hexagon -passes=post-RA-sched %s -o - | FileCheck %s
 
 # The two loads from %a ($r0) can cause a bank conflict. Check that they
 # are not scheduled next to each other.

llvm/test/CodeGen/Hexagon/bank-conflict.mir

@@ -1,4 +1,5 @@
 # RUN: llc -mtriple=hexagon -run-pass post-RA-sched %s -o - | FileCheck %s
+# RUN: llc -mtriple=hexagon -passes=post-RA-sched %s -o - | FileCheck %s
 
 # Test that the Post RA scheduler does not schedule back-to-back loads
 # when there is another instruction to schedule. The scheduler avoids

llvm/test/CodeGen/X86/post-ra-sched-with-debug.mir

@@ -1,5 +1,7 @@
 # RUN: llc -mtriple=x86_64-unknown-unknown -mcpu=btver2 -run-pass=post-RA-sched -o - %s | FileCheck %s
+# RUN: llc -mtriple=x86_64-unknown-unknown -mcpu=btver2 -passes=post-RA-sched -o - %s | FileCheck %s
 # RUN: llc -mtriple=x86_64-unknown-unknown -mcpu=btver2 -run-pass=post-RA-sched -o - %s -experimental-debug-variable-locations| FileCheck %s
+# RUN: llc -mtriple=x86_64-unknown-unknown -mcpu=btver2 -passes=post-RA-sched -o - %s -experimental-debug-variable-locations| FileCheck %s
 
 # Test that multiple DBG_VALUE's and DBG_PHIs following an instruction whose
 # register needs # to be changed during the post-RA scheduler pass are updated

llvm/test/CodeGen/X86/pr27681.mir

@@ -1,4 +1,5 @@
 # RUN: llc -mtriple=i386-unknown-linux-gnu -mcpu=slm -run-pass post-RA-sched -o - %s | FileCheck %s
+# RUN: llc -mtriple=i386-unknown-linux-gnu -mcpu=slm -passes=post-RA-sched -o - %s | FileCheck %s
 #
 # Verify that the critical antidependence breaker does not consider
 # a high byte register as available as a replacement register