llvm-project/llvm/lib/Target/AArch64/AArch64CondBrTuning.cpp

339 lines
10 KiB
C++

//===-- AArch64CondBrTuning.cpp --- Conditional branch tuning for AArch64 -===//
//
// 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
/// This file contains a pass that transforms CBZ/CBNZ/TBZ/TBNZ instructions
/// into a conditional branch (B.cond), when the NZCV flags can be set for
/// "free". This is preferred on targets that have more flexibility when
/// scheduling B.cond instructions as compared to CBZ/CBNZ/TBZ/TBNZ (assuming
/// all other variables are equal). This can also reduce register pressure.
///
/// A few examples:
///
/// 1) add w8, w0, w1 -> cmn w0, w1 ; CMN is an alias of ADDS.
/// cbz w8, .LBB_2 -> b.eq .LBB0_2
///
/// 2) add w8, w0, w1 -> adds w8, w0, w1 ; w8 has multiple uses.
/// cbz w8, .LBB1_2 -> b.eq .LBB1_2
///
/// 3) sub w8, w0, w1 -> subs w8, w0, w1 ; w8 has multiple uses.
/// tbz w8, #31, .LBB6_2 -> b.pl .LBB6_2
///
//===----------------------------------------------------------------------===//
#include "AArch64.h"
#include "AArch64Subtarget.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineFunctionPass.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/Passes.h"
#include "llvm/CodeGen/TargetInstrInfo.h"
#include "llvm/CodeGen/TargetRegisterInfo.h"
#include "llvm/CodeGen/TargetSubtargetInfo.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
using namespace llvm;
#define DEBUG_TYPE "aarch64-cond-br-tuning"
#define AARCH64_CONDBR_TUNING_NAME "AArch64 Conditional Branch Tuning"
namespace {
class AArch64CondBrTuning : public MachineFunctionPass {
const AArch64InstrInfo *TII;
const TargetRegisterInfo *TRI;
MachineRegisterInfo *MRI;
public:
static char ID;
AArch64CondBrTuning() : MachineFunctionPass(ID) {
initializeAArch64CondBrTuningPass(*PassRegistry::getPassRegistry());
}
void getAnalysisUsage(AnalysisUsage &AU) const override;
bool runOnMachineFunction(MachineFunction &MF) override;
StringRef getPassName() const override { return AARCH64_CONDBR_TUNING_NAME; }
private:
MachineInstr *getOperandDef(const MachineOperand &MO);
MachineInstr *convertToFlagSetting(MachineInstr &MI, bool IsFlagSetting);
MachineInstr *convertToCondBr(MachineInstr &MI);
bool tryToTuneBranch(MachineInstr &MI, MachineInstr &DefMI);
};
} // end anonymous namespace
char AArch64CondBrTuning::ID = 0;
INITIALIZE_PASS(AArch64CondBrTuning, "aarch64-cond-br-tuning",
AARCH64_CONDBR_TUNING_NAME, false, false)
void AArch64CondBrTuning::getAnalysisUsage(AnalysisUsage &AU) const {
AU.setPreservesCFG();
MachineFunctionPass::getAnalysisUsage(AU);
}
MachineInstr *AArch64CondBrTuning::getOperandDef(const MachineOperand &MO) {
if (!Register::isVirtualRegister(MO.getReg()))
return nullptr;
return MRI->getUniqueVRegDef(MO.getReg());
}
MachineInstr *AArch64CondBrTuning::convertToFlagSetting(MachineInstr &MI,
bool IsFlagSetting) {
// If this is already the flag setting version of the instruction (e.g., SUBS)
// just make sure the implicit-def of NZCV isn't marked dead.
if (IsFlagSetting) {
for (unsigned I = MI.getNumExplicitOperands(), E = MI.getNumOperands();
I != E; ++I) {
MachineOperand &MO = MI.getOperand(I);
if (MO.isReg() && MO.isDead() && MO.getReg() == AArch64::NZCV)
MO.setIsDead(false);
}
return &MI;
}
bool Is64Bit;
unsigned NewOpc = TII->convertToFlagSettingOpc(MI.getOpcode(), Is64Bit);
Register NewDestReg = MI.getOperand(0).getReg();
if (MRI->hasOneNonDBGUse(MI.getOperand(0).getReg()))
NewDestReg = Is64Bit ? AArch64::XZR : AArch64::WZR;
MachineInstrBuilder MIB = BuildMI(*MI.getParent(), MI, MI.getDebugLoc(),
TII->get(NewOpc), NewDestReg);
for (unsigned I = 1, E = MI.getNumOperands(); I != E; ++I)
MIB.add(MI.getOperand(I));
return MIB;
}
MachineInstr *AArch64CondBrTuning::convertToCondBr(MachineInstr &MI) {
AArch64CC::CondCode CC;
MachineBasicBlock *TargetMBB = TII->getBranchDestBlock(MI);
switch (MI.getOpcode()) {
default:
llvm_unreachable("Unexpected opcode!");
case AArch64::CBZW:
case AArch64::CBZX:
CC = AArch64CC::EQ;
break;
case AArch64::CBNZW:
case AArch64::CBNZX:
CC = AArch64CC::NE;
break;
case AArch64::TBZW:
case AArch64::TBZX:
CC = AArch64CC::PL;
break;
case AArch64::TBNZW:
case AArch64::TBNZX:
CC = AArch64CC::MI;
break;
}
return BuildMI(*MI.getParent(), MI, MI.getDebugLoc(), TII->get(AArch64::Bcc))
.addImm(CC)
.addMBB(TargetMBB);
}
bool AArch64CondBrTuning::tryToTuneBranch(MachineInstr &MI,
MachineInstr &DefMI) {
// We don't want NZCV bits live across blocks.
if (MI.getParent() != DefMI.getParent())
return false;
bool IsFlagSetting = true;
unsigned MIOpc = MI.getOpcode();
MachineInstr *NewCmp = nullptr, *NewBr = nullptr;
switch (DefMI.getOpcode()) {
default:
return false;
case AArch64::ADDWri:
case AArch64::ADDWrr:
case AArch64::ADDWrs:
case AArch64::ADDWrx:
case AArch64::ANDWri:
case AArch64::ANDWrr:
case AArch64::ANDWrs:
case AArch64::BICWrr:
case AArch64::BICWrs:
case AArch64::SUBWri:
case AArch64::SUBWrr:
case AArch64::SUBWrs:
case AArch64::SUBWrx:
IsFlagSetting = false;
LLVM_FALLTHROUGH;
case AArch64::ADDSWri:
case AArch64::ADDSWrr:
case AArch64::ADDSWrs:
case AArch64::ADDSWrx:
case AArch64::ANDSWri:
case AArch64::ANDSWrr:
case AArch64::ANDSWrs:
case AArch64::BICSWrr:
case AArch64::BICSWrs:
case AArch64::SUBSWri:
case AArch64::SUBSWrr:
case AArch64::SUBSWrs:
case AArch64::SUBSWrx:
switch (MIOpc) {
default:
llvm_unreachable("Unexpected opcode!");
case AArch64::CBZW:
case AArch64::CBNZW:
case AArch64::TBZW:
case AArch64::TBNZW:
// Check to see if the TBZ/TBNZ is checking the sign bit.
if ((MIOpc == AArch64::TBZW || MIOpc == AArch64::TBNZW) &&
MI.getOperand(1).getImm() != 31)
return false;
// There must not be any instruction between DefMI and MI that clobbers or
// reads NZCV.
MachineBasicBlock::iterator I(DefMI), E(MI);
for (I = std::next(I); I != E; ++I) {
if (I->modifiesRegister(AArch64::NZCV, TRI) ||
I->readsRegister(AArch64::NZCV, TRI))
return false;
}
LLVM_DEBUG(dbgs() << " Replacing instructions:\n ");
LLVM_DEBUG(DefMI.print(dbgs()));
LLVM_DEBUG(dbgs() << " ");
LLVM_DEBUG(MI.print(dbgs()));
NewCmp = convertToFlagSetting(DefMI, IsFlagSetting);
NewBr = convertToCondBr(MI);
break;
}
break;
case AArch64::ADDXri:
case AArch64::ADDXrr:
case AArch64::ADDXrs:
case AArch64::ADDXrx:
case AArch64::ANDXri:
case AArch64::ANDXrr:
case AArch64::ANDXrs:
case AArch64::BICXrr:
case AArch64::BICXrs:
case AArch64::SUBXri:
case AArch64::SUBXrr:
case AArch64::SUBXrs:
case AArch64::SUBXrx:
IsFlagSetting = false;
LLVM_FALLTHROUGH;
case AArch64::ADDSXri:
case AArch64::ADDSXrr:
case AArch64::ADDSXrs:
case AArch64::ADDSXrx:
case AArch64::ANDSXri:
case AArch64::ANDSXrr:
case AArch64::ANDSXrs:
case AArch64::BICSXrr:
case AArch64::BICSXrs:
case AArch64::SUBSXri:
case AArch64::SUBSXrr:
case AArch64::SUBSXrs:
case AArch64::SUBSXrx:
switch (MIOpc) {
default:
llvm_unreachable("Unexpected opcode!");
case AArch64::CBZX:
case AArch64::CBNZX:
case AArch64::TBZX:
case AArch64::TBNZX: {
// Check to see if the TBZ/TBNZ is checking the sign bit.
if ((MIOpc == AArch64::TBZX || MIOpc == AArch64::TBNZX) &&
MI.getOperand(1).getImm() != 63)
return false;
// There must not be any instruction between DefMI and MI that clobbers or
// reads NZCV.
MachineBasicBlock::iterator I(DefMI), E(MI);
for (I = std::next(I); I != E; ++I) {
if (I->modifiesRegister(AArch64::NZCV, TRI) ||
I->readsRegister(AArch64::NZCV, TRI))
return false;
}
LLVM_DEBUG(dbgs() << " Replacing instructions:\n ");
LLVM_DEBUG(DefMI.print(dbgs()));
LLVM_DEBUG(dbgs() << " ");
LLVM_DEBUG(MI.print(dbgs()));
NewCmp = convertToFlagSetting(DefMI, IsFlagSetting);
NewBr = convertToCondBr(MI);
break;
}
}
break;
}
(void)NewCmp; (void)NewBr;
assert(NewCmp && NewBr && "Expected new instructions.");
LLVM_DEBUG(dbgs() << " with instruction:\n ");
LLVM_DEBUG(NewCmp->print(dbgs()));
LLVM_DEBUG(dbgs() << " ");
LLVM_DEBUG(NewBr->print(dbgs()));
// If this was a flag setting version of the instruction, we use the original
// instruction by just clearing the dead marked on the implicit-def of NCZV.
// Therefore, we should not erase this instruction.
if (!IsFlagSetting)
DefMI.eraseFromParent();
MI.eraseFromParent();
return true;
}
bool AArch64CondBrTuning::runOnMachineFunction(MachineFunction &MF) {
if (skipFunction(MF.getFunction()))
return false;
LLVM_DEBUG(
dbgs() << "********** AArch64 Conditional Branch Tuning **********\n"
<< "********** Function: " << MF.getName() << '\n');
TII = static_cast<const AArch64InstrInfo *>(MF.getSubtarget().getInstrInfo());
TRI = MF.getSubtarget().getRegisterInfo();
MRI = &MF.getRegInfo();
bool Changed = false;
for (MachineBasicBlock &MBB : MF) {
bool LocalChange = false;
for (MachineBasicBlock::iterator I = MBB.getFirstTerminator(),
E = MBB.end();
I != E; ++I) {
MachineInstr &MI = *I;
switch (MI.getOpcode()) {
default:
break;
case AArch64::CBZW:
case AArch64::CBZX:
case AArch64::CBNZW:
case AArch64::CBNZX:
case AArch64::TBZW:
case AArch64::TBZX:
case AArch64::TBNZW:
case AArch64::TBNZX:
MachineInstr *DefMI = getOperandDef(MI.getOperand(0));
LocalChange = (DefMI && tryToTuneBranch(MI, *DefMI));
break;
}
// If the optimization was successful, we can't optimize any other
// branches because doing so would clobber the NZCV flags.
if (LocalChange) {
Changed = true;
break;
}
}
}
return Changed;
}
FunctionPass *llvm::createAArch64CondBrTuning() {
return new AArch64CondBrTuning();
}