[CodeGen] Add user interface for DetectDeadLanes

[BeMg]

Summary

Let other class could reuse the used/def of subregister from DetectDeadLanes pass.

[craig.topper]

llvm/lib/CodeGen/DetectDeadLanes.cpp

499

Do we need to skip the "Mark operands as dead/unused." part of runOnce for this usage?

[arsenm]

llvm/include/llvm/CodeGen/DetectDeadLanes.h

106–107

Keep member variables together below here

llvm/lib/CodeGen/DetectDeadLanes.cpp

483–505

Why is this all repeated here?

[BeMg]

1. let runOnMachineFunction reuse getSubRegisterLaneBitInfo
2. Move the variable

[BeMg]

llvm/lib/CodeGen/DetectDeadLanes.cpp

499

I'm not sure whether "Mark operands as dead" is part of computing DeadLanes algorithm. And we remove "Mark operands as dead" may get an unaccomplished result?

[BeMg]

rebase

[BeMg]

Add a parameter to control whether set Operand as undef/def

[craig.topper]

llvm/lib/CodeGen/DetectDeadLanes.cpp

440

Can we split this function into two functions above and below this point and check this flag in the while loop?

484

This debug message doesn't make sense for the call from the RISC-V pass.

516

This comment belongs to the if (!MRI->subRegLivenessEnabled()) that got moved.

[craig.topper]

llvm/lib/CodeGen/DetectDeadLanes.cpp

440

let me rephrase

Can we split this function into two functions above and below this point? Have the while loop in the caller only call the second half when NeedSetUndef is true. The first function can just return void since it doesn't modify anything.

[BeMg]

Split runOnce function into ModifySubRegisterOperandStatus and computeSubRegisterLaneBitInfo

[craig.topper]

llvm/lib/CodeGen/DetectDeadLanes.cpp

439

Function names should be start with a lower case letter

503

We don't need to call computeSubRegisterLaneBitInfo if ModifySubRegisterOperandStatus didn't change anything

[arsenm]

llvm/include/llvm/CodeGen/DetectDeadLanes.h

97

Needs comment, I don't know what the two return values mean.

llvm/lib/CodeGen/DetectDeadLanes.cpp

470

I don't really understand what again is supposed to mean (but I guess this was the case before too)

[BeMg]

1. ModifySubRegisterOperandStatus -> modifySubRegisterOperandStatus
2. if not change anything, skip computeSubRegisterLaneBitInfo
3. Add comment

[craig.topper]

llvm/lib/CodeGen/DetectDeadLanes.cpp

498

Change the loop back to do while and just break in the middle like:

do {
  computeSubRegisterLaneBitInfo();
  if (!NeedSetUndef)
    break;

  bool LocalChanged;
  std::tie(LocalChanged, Again) = runOnce(MF);
  Changed |= LocalChanged;
} while(Again);

[BeMg]

Use do while

[craig.topper]

llvm/include/llvm/CodeGen/DetectDeadLanes.h

42

We shouldn't have "using namespace llvm" in a header file

[BeMg]

D141993 solved it.

llvm/include/llvm/CodeGen/DetectDeadLanes.h

+//===- DetectDeadLanes.h - SubRegister Lane Usage Analysis --*- 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
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// Analysis that tracks defined/used subregister lanes across COPY instructions
+/// and instructions that get lowered to a COPY (PHI, REG_SEQUENCE,
+/// INSERT_SUBREG, EXTRACT_SUBREG).
+/// The information is used to detect dead definitions and the usage of
+/// (completely) undefined values and mark the operands as such.
+/// This pass is necessary because the dead/undef status is not obvious anymore
+/// when subregisters are involved.
+///
+/// Example:
+///    %0 = some definition
+///    %1 = IMPLICIT_DEF
+///    %2 = REG_SEQUENCE %0, sub0, %1, sub1
+///    %3 = EXTRACT_SUBREG %2, sub1
+///       = use %3
+/// The %0 definition is dead and %3 contains an undefined value.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_CODEGEN_DETECTDEADLANES_H
+#define LLVM_CODEGEN_DETECTDEADLANES_H
+
+#include "llvm/ADT/BitVector.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/TargetRegisterInfo.h"
+#include "llvm/InitializePasses.h"
+#include "llvm/Pass.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/raw_ostream.h"
+#include <deque>
+
+using namespace llvm;
+
+/// Contains a bitmask of which lanes of a given virtual register are
+/// defined and which ones are actually used.
+struct VRegInfo {
+  LaneBitmask UsedLanes;
+  LaneBitmask DefinedLanes;
+};
+
+class DetectDeadLanes : public MachineFunctionPass {
+public:
+  bool runOnMachineFunction(MachineFunction &MF) override;
+
+  static char ID;
+  DetectDeadLanes() : MachineFunctionPass(ID) {}
+
+  StringRef getPassName() const override { return "Detect Dead Lanes"; }
+
+  void getAnalysisUsage(AnalysisUsage &AU) const override {
+    AU.setPreservesCFG();
+    MachineFunctionPass::getAnalysisUsage(AU);
+  }
+
+  // Provide the Used/undef LaneBit status for all virtual reigster.
+  bool getSubRegisterLaneBitInfo(MachineFunction &MF, VRegInfo *Info);
+
+private:
+  /// Add used lane bits on the register used by operand \p MO. This translates
+  /// the bitmask based on the operands subregister, and puts the register into
+  /// the worklist if any new bits were added.
+  void addUsedLanesOnOperand(const MachineOperand &MO, LaneBitmask UsedLanes);
+
+  /// Given a bitmask \p UsedLanes for the used lanes on a def output of a
+  /// COPY-like instruction determine the lanes used on the use operands
+  /// and call addUsedLanesOnOperand() for them.
+  void transferUsedLanesStep(const MachineInstr &MI, LaneBitmask UsedLanes);
+
+  /// Given a use regiser operand \p Use and a mask of defined lanes, check
+  /// if the operand belongs to a lowersToCopies() instruction, transfer the
+  /// mask to the def and put the instruction into the worklist.
+  void transferDefinedLanesStep(const MachineOperand &Use,
+                                LaneBitmask DefinedLanes);
+
+  /// Given a mask \p DefinedLanes of lanes defined at operand \p OpNum
+  /// of COPY-like instruction, determine which lanes are defined at the output
+  /// operand \p Def.
+  LaneBitmask transferDefinedLanes(const MachineOperand &Def, unsigned OpNum,
+                                   LaneBitmask DefinedLanes) const;
+
+  /// Given a mask \p UsedLanes used from the output of instruction \p MI
+  /// determine which lanes are used from operand \p MO of this instruction.
+  LaneBitmask transferUsedLanes(const MachineInstr &MI, LaneBitmask UsedLanes,
+                                const MachineOperand &MO) const;
+
+  std::pair<bool, bool> runOnce(MachineFunction &MF);
+
+  LaneBitmask determineInitialDefinedLanes(unsigned Reg);
+  LaneBitmask determineInitialUsedLanes(unsigned Reg);
+
+  bool isUndefRegAtInput(const MachineOperand &MO,
+                         const VRegInfo &RegInfo) const;
+
+  bool isUndefInput(const MachineOperand &MO, bool *CrossCopy) const;
+
+  void PutInWorklist(unsigned RegIdx) {
+    if (WorklistMembers.test(RegIdx))
+      return;
+    WorklistMembers.set(RegIdx);
+    Worklist.push_back(RegIdx);
+  }
+
+  const MachineRegisterInfo *MRI;
+  const TargetRegisterInfo *TRI;
+  VRegInfo *VRegInfos;
+  /// Worklist containing virtreg indexes.
+  std::deque<unsigned> Worklist;
+  BitVector WorklistMembers;
+  /// This bitvector is set for each vreg index where the vreg is defined
+  /// by an instruction where lowersToCopies()==true.
+  BitVector DefinedByCopy;
+};
+
+#endif // LLVM_CODEGEN_DETECTDEADLANES_H
 No newline at end of file

llvm/lib/CodeGen/DetectDeadLanes.cpp

 ///    %1 = IMPLICIT_DEF
 ///    %2 = REG_SEQUENCE %0, sub0, %1, sub1
 ///    %3 = EXTRACT_SUBREG %2, sub1
 ///       = use %3
 /// The %0 definition is dead and %3 contains an undefined value.
 //
 //===----------------------------------------------------------------------===//
 
-#include "llvm/ADT/BitVector.h"
-#include "llvm/CodeGen/MachineFunctionPass.h"
-#include "llvm/CodeGen/MachineRegisterInfo.h"
-#include "llvm/CodeGen/TargetRegisterInfo.h"
-#include "llvm/InitializePasses.h"
-#include "llvm/Pass.h"
-#include "llvm/Support/Debug.h"
-#include "llvm/Support/raw_ostream.h"
-#include <deque>
+#include "llvm/CodeGen/DetectDeadLanes.h"
 
 using namespace llvm;
 
 #define DEBUG_TYPE "detect-dead-lanes"
 
-namespace {
-
-/// Contains a bitmask of which lanes of a given virtual register are
-/// defined and which ones are actually used.
-struct VRegInfo {
-  LaneBitmask UsedLanes;
-  LaneBitmask DefinedLanes;
-};
-
-class DetectDeadLanes : public MachineFunctionPass {
-public:
-  bool runOnMachineFunction(MachineFunction &MF) override;
-
-  static char ID;
-  DetectDeadLanes() : MachineFunctionPass(ID) {}
-
-  StringRef getPassName() const override { return "Detect Dead Lanes"; }
-
-  void getAnalysisUsage(AnalysisUsage &AU) const override {
-    AU.setPreservesCFG();
-    MachineFunctionPass::getAnalysisUsage(AU);
-  }
-
-private:
-  /// Add used lane bits on the register used by operand \p MO. This translates
-  /// the bitmask based on the operands subregister, and puts the register into
-  /// the worklist if any new bits were added.
-  void addUsedLanesOnOperand(const MachineOperand &MO, LaneBitmask UsedLanes);
-
-  /// Given a bitmask \p UsedLanes for the used lanes on a def output of a
-  /// COPY-like instruction determine the lanes used on the use operands
-  /// and call addUsedLanesOnOperand() for them.
-  void transferUsedLanesStep(const MachineInstr &MI, LaneBitmask UsedLanes);
-
-  /// Given a use regiser operand \p Use and a mask of defined lanes, check
-  /// if the operand belongs to a lowersToCopies() instruction, transfer the
-  /// mask to the def and put the instruction into the worklist.
-  void transferDefinedLanesStep(const MachineOperand &Use,
-                                LaneBitmask DefinedLanes);
-
-  /// Given a mask \p DefinedLanes of lanes defined at operand \p OpNum
-  /// of COPY-like instruction, determine which lanes are defined at the output
-  /// operand \p Def.
-  LaneBitmask transferDefinedLanes(const MachineOperand &Def, unsigned OpNum,
-                                   LaneBitmask DefinedLanes) const;
-
-  /// Given a mask \p UsedLanes used from the output of instruction \p MI
-  /// determine which lanes are used from operand \p MO of this instruction.
-  LaneBitmask transferUsedLanes(const MachineInstr &MI, LaneBitmask UsedLanes,
-                                const MachineOperand &MO) const;
-
-  std::pair<bool, bool> runOnce(MachineFunction &MF);
-
-  LaneBitmask determineInitialDefinedLanes(unsigned Reg);
-  LaneBitmask determineInitialUsedLanes(unsigned Reg);
-
-  bool isUndefRegAtInput(const MachineOperand &MO,
-                         const VRegInfo &RegInfo) const;
-
-  bool isUndefInput(const MachineOperand &MO, bool *CrossCopy) const;
-
-  const MachineRegisterInfo *MRI;
-  const TargetRegisterInfo *TRI;
-
-  void PutInWorklist(unsigned RegIdx) {
-    if (WorklistMembers.test(RegIdx))
-      return;
-    WorklistMembers.set(RegIdx);
-    Worklist.push_back(RegIdx);
-  }
-
-  VRegInfo *VRegInfos;
-  /// Worklist containing virtreg indexes.
-  std::deque<unsigned> Worklist;
-  BitVector WorklistMembers;
-  /// This bitvector is set for each vreg index where the vreg is defined
-  /// by an instruction where lowersToCopies()==true.
-  BitVector DefinedByCopy;
-};
-
-} // end anonymous namespace
-
 char DetectDeadLanes::ID = 0;
 char &llvm::DetectDeadLanesID = DetectDeadLanes::ID;
 
 INITIALIZE_PASS(DetectDeadLanes, DEBUG_TYPE, "Detect Dead Lanes", false, false)
 
 /// Returns true if \p MI will get lowered to a series of COPY instructions.
 /// We call this a COPY-like instruction.
 static bool lowersToCopies(const MachineInstr &MI) {
for (unsigned RegIdx = 0; RegIdx < NumVirtRegs; ++RegIdx) {
       dbgs() << printReg(Reg, nullptr)
              << " Used: " << PrintLaneMask(Info.UsedLanes)
              << " Def: " << PrintLaneMask(Info.DefinedLanes) << '\n';
     }
     dbgs() << "\n";
   });
 
   bool Changed = false;
   bool Again = false;
   // Mark operands as dead/unused.
   for (MachineBasicBlock &MBB : MF) {
     for (MachineInstr &MI : MBB) {
       for (MachineOperand &MO : MI.operands()) {
         if (!MO.isReg())
           continue;
         Register Reg = MO.getReg();
         if (!Register::isVirtualRegister(Reg))
           continue;
for (MachineInstr &MI : MBB) {
             MO.setIsUndef();
             Changed = true;
           } else if (isUndefInput(MO, &CrossCopy)) {
             LLVM_DEBUG(dbgs()
                        << "Marking operand '" << MO << "' as undef in " << MI);
             MO.setIsUndef();
             Changed = true;
             if (CrossCopy)
               Again = true;
           }
         }
       }
     }
   }
 
   return std::make_pair(Changed, Again);
 }
 
-bool DetectDeadLanes::runOnMachineFunction(MachineFunction &MF) {
-  // Don't bother if we won't track subregister liveness later.  This pass is
-  // required for correctness if subregister liveness is enabled because the
-  // register coalescer cannot deal with hidden dead defs. However without
-  // subregister liveness enabled, the expected benefits of this pass are small
-  // so we safe the compile time.
+bool DetectDeadLanes::getSubRegisterLaneBitInfo(MachineFunction &MF,
+                                                VRegInfo *Info) {
   MRI = &MF.getRegInfo();
   if (!MRI->subRegLivenessEnabled()) {
     LLVM_DEBUG(dbgs() << "Skipping Detect dead lanes pass\n");
     return false;
   }
 
   TRI = MRI->getTargetRegisterInfo();
 
   unsigned NumVirtRegs = MRI->getNumVirtRegs();
+  if (Info == nullptr)
     VRegInfos = new VRegInfo[NumVirtRegs];
+  else
+    VRegInfos = Info;
   WorklistMembers.resize(NumVirtRegs);
   DefinedByCopy.resize(NumVirtRegs);
 
   bool Changed = false;
   bool Again;
   do {
     bool LocalChanged;
     std::tie(LocalChanged, Again) = runOnce(MF);
     Changed |= LocalChanged;
-  } while(Again);
+  } while (Again);
 
   DefinedByCopy.clear();
   WorklistMembers.clear();
+
+  if (Info == nullptr)
     delete[] VRegInfos;
+
   return Changed;
 }
+
+bool DetectDeadLanes::runOnMachineFunction(MachineFunction &MF) {
+  // Don't bother if we won't track subregister liveness later.  This pass is
+  // required for correctness if subregister liveness is enabled because the
+  // register coalescer cannot deal with hidden dead defs. However without
+  // subregister liveness enabled, the expected benefits of this pass are small
+  // so we safe the compile time.
+  // bool Changed = false;
+  return getSubRegisterLaneBitInfo(MF, nullptr);
+}