Google Git
Sign in
llvm / llvm-project / a87782c34d667d1c1a18fe82a9a7abfc72fd345b / . / llvm / lib / Target / PowerPC / PPCInstrInfo.cpp
blob: a0fd2111de11d28277314f2eabd60751060b6fbb [file] [log] [blame]
//===-- PPCInstrInfo.cpp - PowerPC Instruction Information ----------------===//
//
// 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
//
//===----------------------------------------------------------------------===//
//
// This file contains the PowerPC implementation of the TargetInstrInfo class.
//
//===----------------------------------------------------------------------===//
#include "PPCInstrInfo.h"
#include "MCTargetDesc/PPCPredicates.h"
#include "PPC.h"
#include "PPCHazardRecognizers.h"
#include "PPCInstrBuilder.h"
#include "PPCMachineFunctionInfo.h"
#include "PPCTargetMachine.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/CodeGen/LiveIntervals.h"
#include "llvm/CodeGen/MachineConstantPool.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineFunctionPass.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineMemOperand.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/PseudoSourceValue.h"
#include "llvm/CodeGen/RegisterClassInfo.h"
#include "llvm/CodeGen/RegisterPressure.h"
#include "llvm/CodeGen/ScheduleDAG.h"
#include "llvm/CodeGen/SlotIndexes.h"
#include "llvm/CodeGen/StackMaps.h"
#include "llvm/MC/MCAsmInfo.h"
#include "llvm/MC/MCInst.h"
#include "llvm/MC/TargetRegistry.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/raw_ostream.h"
using namespace llvm;
#define DEBUG_TYPE "ppc-instr-info"
#define GET_INSTRMAP_INFO
#define GET_INSTRINFO_CTOR_DTOR
#include "PPCGenInstrInfo.inc"
STATISTIC(NumStoreSPILLVSRRCAsVec,
"Number of spillvsrrc spilled to stack as vec");
STATISTIC(NumStoreSPILLVSRRCAsGpr,
"Number of spillvsrrc spilled to stack as gpr");
STATISTIC(NumGPRtoVSRSpill, "Number of gpr spills to spillvsrrc");
STATISTIC(CmpIselsConverted,
"Number of ISELs that depend on comparison of constants converted");
STATISTIC(MissedConvertibleImmediateInstrs,
"Number of compare-immediate instructions fed by constants");
STATISTIC(NumRcRotatesConvertedToRcAnd,
"Number of record-form rotates converted to record-form andi");
static cl::
opt<bool> DisableCTRLoopAnal("disable-ppc-ctrloop-analysis", cl::Hidden,
cl::desc("Disable analysis for CTR loops"));
static cl::opt<bool> DisableCmpOpt("disable-ppc-cmp-opt",
cl::desc("Disable compare instruction optimization"), cl::Hidden);
static cl::opt<bool> VSXSelfCopyCrash("crash-on-ppc-vsx-self-copy",
cl::desc("Causes the backend to crash instead of generating a nop VSX copy"),
cl::Hidden);
static cl::opt<bool>
UseOldLatencyCalc("ppc-old-latency-calc", cl::Hidden,
cl::desc("Use the old (incorrect) instruction latency calculation"));
static cl::opt<float>
FMARPFactor("ppc-fma-rp-factor", cl::Hidden, cl::init(1.5),
cl::desc("register pressure factor for the transformations."));
static cl::opt<bool> EnableFMARegPressureReduction(
"ppc-fma-rp-reduction", cl::Hidden, cl::init(true),
cl::desc("enable register pressure reduce in machine combiner pass."));
// Pin the vtable to this file.
void PPCInstrInfo::anchor() {}
PPCInstrInfo::PPCInstrInfo(PPCSubtarget &STI)
: PPCGenInstrInfo(PPC::ADJCALLSTACKDOWN, PPC::ADJCALLSTACKUP,
/* CatchRetOpcode */ -1,
STI.isPPC64() ? PPC::BLR8 : PPC::BLR),
Subtarget(STI), RI(STI.getTargetMachine()) {}
/// CreateTargetHazardRecognizer - Return the hazard recognizer to use for
/// this target when scheduling the DAG.
ScheduleHazardRecognizer *
PPCInstrInfo::CreateTargetHazardRecognizer(const TargetSubtargetInfo *STI,
const ScheduleDAG *DAG) const {
unsigned Directive =
static_cast<const PPCSubtarget *>(STI)->getCPUDirective();
if (Directive == PPC::DIR_440 || Directive == PPC::DIR_A2 ||
Directive == PPC::DIR_E500mc || Directive == PPC::DIR_E5500) {
const InstrItineraryData *II =
static_cast<const PPCSubtarget *>(STI)->getInstrItineraryData();
return new ScoreboardHazardRecognizer(II, DAG);
}
return TargetInstrInfo::CreateTargetHazardRecognizer(STI, DAG);
}
/// CreateTargetPostRAHazardRecognizer - Return the postRA hazard recognizer
/// to use for this target when scheduling the DAG.
ScheduleHazardRecognizer *
PPCInstrInfo::CreateTargetPostRAHazardRecognizer(const InstrItineraryData *II,
const ScheduleDAG *DAG) const {
unsigned Directive =
DAG->MF.getSubtarget<PPCSubtarget>().getCPUDirective();
// FIXME: Leaving this as-is until we have POWER9 scheduling info
if (Directive == PPC::DIR_PWR7 || Directive == PPC::DIR_PWR8)
return new PPCDispatchGroupSBHazardRecognizer(II, DAG);
// Most subtargets use a PPC970 recognizer.
if (Directive != PPC::DIR_440 && Directive != PPC::DIR_A2 &&
Directive != PPC::DIR_E500mc && Directive != PPC::DIR_E5500) {
assert(DAG->TII && "No InstrInfo?");
return new PPCHazardRecognizer970(*DAG);
}
return new ScoreboardHazardRecognizer(II, DAG);
}
unsigned PPCInstrInfo::getInstrLatency(const InstrItineraryData *ItinData,
const MachineInstr &MI,
unsigned *PredCost) const {
if (!ItinData || UseOldLatencyCalc)
return PPCGenInstrInfo::getInstrLatency(ItinData, MI, PredCost);
// The default implementation of getInstrLatency calls getStageLatency, but
// getStageLatency does not do the right thing for us. While we have
// itinerary, most cores are fully pipelined, and so the itineraries only
// express the first part of the pipeline, not every stage. Instead, we need
// to use the listed output operand cycle number (using operand 0 here, which
// is an output).
unsigned Latency = 1;
unsigned DefClass = MI.getDesc().getSchedClass();
for (unsigned i = 0, e = MI.getNumOperands(); i != e; ++i) {
const MachineOperand &MO = MI.getOperand(i);
if (!MO.isReg() || !MO.isDef() || MO.isImplicit())
continue;
int Cycle = ItinData->getOperandCycle(DefClass, i);
if (Cycle < 0)
continue;
Latency = std::max(Latency, (unsigned) Cycle);
}
return Latency;
}
int PPCInstrInfo::getOperandLatency(const InstrItineraryData *ItinData,
const MachineInstr &DefMI, unsigned DefIdx,
const MachineInstr &UseMI,
unsigned UseIdx) const {
int Latency = PPCGenInstrInfo::getOperandLatency(ItinData, DefMI, DefIdx,
UseMI, UseIdx);
if (!DefMI.getParent())
return Latency;
const MachineOperand &DefMO = DefMI.getOperand(DefIdx);
Register Reg = DefMO.getReg();
bool IsRegCR;
if (Register::isVirtualRegister(Reg)) {
const MachineRegisterInfo *MRI =
&DefMI.getParent()->getParent()->getRegInfo();
IsRegCR = MRI->getRegClass(Reg)->hasSuperClassEq(&PPC::CRRCRegClass) ||
MRI->getRegClass(Reg)->hasSuperClassEq(&PPC::CRBITRCRegClass);
} else {
IsRegCR = PPC::CRRCRegClass.contains(Reg) ||
PPC::CRBITRCRegClass.contains(Reg);
}
if (UseMI.isBranch() && IsRegCR) {
if (Latency < 0)
Latency = getInstrLatency(ItinData, DefMI);
// On some cores, there is an additional delay between writing to a condition
// register, and using it from a branch.
unsigned Directive = Subtarget.getCPUDirective();
switch (Directive) {
default: break;
case PPC::DIR_7400:
case PPC::DIR_750:
case PPC::DIR_970:
case PPC::DIR_E5500:
case PPC::DIR_PWR4:
case PPC::DIR_PWR5:
case PPC::DIR_PWR5X:
case PPC::DIR_PWR6:
case PPC::DIR_PWR6X:
case PPC::DIR_PWR7:
case PPC::DIR_PWR8:
// FIXME: Is this needed for POWER9?
Latency += 2;
break;
}
}
return Latency;
}
/// This is an architecture-specific helper function of reassociateOps.
/// Set special operand attributes for new instructions after reassociation.
void PPCInstrInfo::setSpecialOperandAttr(MachineInstr &OldMI1,
MachineInstr &OldMI2,
MachineInstr &NewMI1,
MachineInstr &NewMI2) const {
// Propagate FP flags from the original instructions.
// But clear poison-generating flags because those may not be valid now.
uint16_t IntersectedFlags = OldMI1.getFlags() & OldMI2.getFlags();
NewMI1.setFlags(IntersectedFlags);
NewMI1.clearFlag(MachineInstr::MIFlag::NoSWrap);
NewMI1.clearFlag(MachineInstr::MIFlag::NoUWrap);
NewMI1.clearFlag(MachineInstr::MIFlag::IsExact);
NewMI2.setFlags(IntersectedFlags);
NewMI2.clearFlag(MachineInstr::MIFlag::NoSWrap);
NewMI2.clearFlag(MachineInstr::MIFlag::NoUWrap);
NewMI2.clearFlag(MachineInstr::MIFlag::IsExact);
}
void PPCInstrInfo::setSpecialOperandAttr(MachineInstr &MI,
uint16_t Flags) const {
MI.setFlags(Flags);
MI.clearFlag(MachineInstr::MIFlag::NoSWrap);
MI.clearFlag(MachineInstr::MIFlag::NoUWrap);
MI.clearFlag(MachineInstr::MIFlag::IsExact);
}
// This function does not list all associative and commutative operations, but
// only those worth feeding through the machine combiner in an attempt to
// reduce the critical path. Mostly, this means floating-point operations,
// because they have high latencies(>=5) (compared to other operations, such as
// and/or, which are also associative and commutative, but have low latencies).
bool PPCInstrInfo::isAssociativeAndCommutative(const MachineInstr &Inst) const {
switch (Inst.getOpcode()) {
// Floating point:
// FP Add:
case PPC::FADD:
case PPC::FADDS:
// FP Multiply:
case PPC::FMUL:
case PPC::FMULS:
// Altivec Add:
case PPC::VADDFP:
// VSX Add:
case PPC::XSADDDP:
case PPC::XVADDDP:
case PPC::XVADDSP:
case PPC::XSADDSP:
// VSX Multiply:
case PPC::XSMULDP:
case PPC::XVMULDP:
case PPC::XVMULSP:
case PPC::XSMULSP:
return Inst.getFlag(MachineInstr::MIFlag::FmReassoc) &&
Inst.getFlag(MachineInstr::MIFlag::FmNsz);
// Fixed point:
// Multiply:
case PPC::MULHD:
case PPC::MULLD:
case PPC::MULHW:
case PPC::MULLW:
return true;
default:
return false;
}
}
#define InfoArrayIdxFMAInst 0
#define InfoArrayIdxFAddInst 1
#define InfoArrayIdxFMULInst 2
#define InfoArrayIdxAddOpIdx 3
#define InfoArrayIdxMULOpIdx 4
#define InfoArrayIdxFSubInst 5
// Array keeps info for FMA instructions:
// Index 0(InfoArrayIdxFMAInst): FMA instruction;
// Index 1(InfoArrayIdxFAddInst): ADD instruction associated with FMA;
// Index 2(InfoArrayIdxFMULInst): MUL instruction associated with FMA;
// Index 3(InfoArrayIdxAddOpIdx): ADD operand index in FMA operands;
// Index 4(InfoArrayIdxMULOpIdx): first MUL operand index in FMA operands;
// second MUL operand index is plus 1;
// Index 5(InfoArrayIdxFSubInst): SUB instruction associated with FMA.
static const uint16_t FMAOpIdxInfo[][6] = {
// FIXME: Add more FMA instructions like XSNMADDADP and so on.
{PPC::XSMADDADP, PPC::XSADDDP, PPC::XSMULDP, 1, 2, PPC::XSSUBDP},
{PPC::XSMADDASP, PPC::XSADDSP, PPC::XSMULSP, 1, 2, PPC::XSSUBSP},
{PPC::XVMADDADP, PPC::XVADDDP, PPC::XVMULDP, 1, 2, PPC::XVSUBDP},
{PPC::XVMADDASP, PPC::XVADDSP, PPC::XVMULSP, 1, 2, PPC::XVSUBSP},
{PPC::FMADD, PPC::FADD, PPC::FMUL, 3, 1, PPC::FSUB},
{PPC::FMADDS, PPC::FADDS, PPC::FMULS, 3, 1, PPC::FSUBS}};
// Check if an opcode is a FMA instruction. If it is, return the index in array
// FMAOpIdxInfo. Otherwise, return -1.
int16_t PPCInstrInfo::getFMAOpIdxInfo(unsigned Opcode) const {
for (unsigned I = 0; I < array_lengthof(FMAOpIdxInfo); I++)
if (FMAOpIdxInfo[I][InfoArrayIdxFMAInst] == Opcode)
return I;
return -1;
}
// On PowerPC target, we have two kinds of patterns related to FMA:
// 1: Improve ILP.
// Try to reassociate FMA chains like below:
//
// Pattern 1:
// A = FADD X, Y (Leaf)
// B = FMA A, M21, M22 (Prev)
// C = FMA B, M31, M32 (Root)
// -->
// A = FMA X, M21, M22
// B = FMA Y, M31, M32
// C = FADD A, B
//
// Pattern 2:
// A = FMA X, M11, M12 (Leaf)
// B = FMA A, M21, M22 (Prev)
// C = FMA B, M31, M32 (Root)
// -->
// A = FMUL M11, M12
// B = FMA X, M21, M22
// D = FMA A, M31, M32
// C = FADD B, D
//
// breaking the dependency between A and B, allowing FMA to be executed in
// parallel (or back-to-back in a pipeline) instead of depending on each other.
//
// 2: Reduce register pressure.
// Try to reassociate FMA with FSUB and a constant like below:
// C is a floating point const.
//
// Pattern 1:
// A = FSUB X, Y (Leaf)
// D = FMA B, C, A (Root)
// -->
// A = FMA B, Y, -C
// D = FMA A, X, C
//
// Pattern 2:
// A = FSUB X, Y (Leaf)
// D = FMA B, A, C (Root)
// -->
// A = FMA B, Y, -C
// D = FMA A, X, C
//
// Before the transformation, A must be assigned with different hardware
// register with D. After the transformation, A and D must be assigned with
// same hardware register due to TIE attribute of FMA instructions.
//
bool PPCInstrInfo::getFMAPatterns(
MachineInstr &Root, SmallVectorImpl<MachineCombinerPattern> &Patterns,
bool DoRegPressureReduce) const {
MachineBasicBlock *MBB = Root.getParent();
const MachineRegisterInfo *MRI = &MBB->getParent()->getRegInfo();
const TargetRegisterInfo *TRI = &getRegisterInfo();
auto IsAllOpsVirtualReg = [](const MachineInstr &Instr) {
for (const auto &MO : Instr.explicit_operands())
if (!(MO.isReg() && Register::isVirtualRegister(MO.getReg())))
return false;
return true;
};
auto IsReassociableAddOrSub = [&](const MachineInstr &Instr,
unsigned OpType) {
if (Instr.getOpcode() !=
FMAOpIdxInfo[getFMAOpIdxInfo(Root.getOpcode())][OpType])
return false;
// Instruction can be reassociated.
// fast math flags may prohibit reassociation.
if (!(Instr.getFlag(MachineInstr::MIFlag::FmReassoc) &&
Instr.getFlag(MachineInstr::MIFlag::FmNsz)))
return false;
// Instruction operands are virtual registers for reassociation.
if (!IsAllOpsVirtualReg(Instr))
return false;
// For register pressure reassociation, the FSub must have only one use as
// we want to delete the sub to save its def.
if (OpType == InfoArrayIdxFSubInst &&
!MRI->hasOneNonDBGUse(Instr.getOperand(0).getReg()))
return false;
return true;
};
auto IsReassociableFMA = [&](const MachineInstr &Instr, int16_t &AddOpIdx,
int16_t &MulOpIdx, bool IsLeaf) {
int16_t Idx = getFMAOpIdxInfo(Instr.getOpcode());
if (Idx < 0)
return false;
// Instruction can be reassociated.
// fast math flags may prohibit reassociation.
if (!(Instr.getFlag(MachineInstr::MIFlag::FmReassoc) &&
Instr.getFlag(MachineInstr::MIFlag::FmNsz)))
return false;
// Instruction operands are virtual registers for reassociation.
if (!IsAllOpsVirtualReg(Instr))
return false;
MulOpIdx = FMAOpIdxInfo[Idx][InfoArrayIdxMULOpIdx];
if (IsLeaf)
return true;
AddOpIdx = FMAOpIdxInfo[Idx][InfoArrayIdxAddOpIdx];
const MachineOperand &OpAdd = Instr.getOperand(AddOpIdx);
MachineInstr *MIAdd = MRI->getUniqueVRegDef(OpAdd.getReg());
// If 'add' operand's def is not in current block, don't do ILP related opt.
if (!MIAdd || MIAdd->getParent() != MBB)
return false;
// If this is not Leaf FMA Instr, its 'add' operand should only have one use
// as this fma will be changed later.
return IsLeaf ? true : MRI->hasOneNonDBGUse(OpAdd.getReg());
};
int16_t AddOpIdx = -1;
int16_t MulOpIdx = -1;
bool IsUsedOnceL = false;
bool IsUsedOnceR = false;
MachineInstr *MULInstrL = nullptr;
MachineInstr *MULInstrR = nullptr;
auto IsRPReductionCandidate = [&]() {
// Currently, we only support float and double.
// FIXME: add support for other types.
unsigned Opcode = Root.getOpcode();
if (Opcode != PPC::XSMADDASP && Opcode != PPC::XSMADDADP)
return false;
// Root must be a valid FMA like instruction.
// Treat it as leaf as we don't care its add operand.
if (IsReassociableFMA(Root, AddOpIdx, MulOpIdx, true)) {
assert((MulOpIdx >= 0) && "mul operand index not right!");
Register MULRegL = TRI->lookThruSingleUseCopyChain(
Root.getOperand(MulOpIdx).getReg(), MRI);
Register MULRegR = TRI->lookThruSingleUseCopyChain(
Root.getOperand(MulOpIdx + 1).getReg(), MRI);
if (!MULRegL && !MULRegR)
return false;
if (MULRegL && !MULRegR) {
MULRegR =
TRI->lookThruCopyLike(Root.getOperand(MulOpIdx + 1).getReg(), MRI);
IsUsedOnceL = true;
} else if (!MULRegL && MULRegR) {
MULRegL =
TRI->lookThruCopyLike(Root.getOperand(MulOpIdx).getReg(), MRI);
IsUsedOnceR = true;
} else {
IsUsedOnceL = true;
IsUsedOnceR = true;
}
if (!Register::isVirtualRegister(MULRegL) ||
!Register::isVirtualRegister(MULRegR))
return false;
MULInstrL = MRI->getVRegDef(MULRegL);
MULInstrR = MRI->getVRegDef(MULRegR);
return true;
}
return false;
};
// Register pressure fma reassociation patterns.
if (DoRegPressureReduce && IsRPReductionCandidate()) {
assert((MULInstrL && MULInstrR) && "wrong register preduction candidate!");
// Register pressure pattern 1
if (isLoadFromConstantPool(MULInstrL) && IsUsedOnceR &&
IsReassociableAddOrSub(*MULInstrR, InfoArrayIdxFSubInst)) {
LLVM_DEBUG(dbgs() << "add pattern REASSOC_XY_BCA\n");
Patterns.push_back(MachineCombinerPattern::REASSOC_XY_BCA);
return true;
}
// Register pressure pattern 2
if ((isLoadFromConstantPool(MULInstrR) && IsUsedOnceL &&
IsReassociableAddOrSub(*MULInstrL, InfoArrayIdxFSubInst))) {
LLVM_DEBUG(dbgs() << "add pattern REASSOC_XY_BAC\n");
Patterns.push_back(MachineCombinerPattern::REASSOC_XY_BAC);
return true;
}
}
// ILP fma reassociation patterns.
// Root must be a valid FMA like instruction.
AddOpIdx = -1;
if (!IsReassociableFMA(Root, AddOpIdx, MulOpIdx, false))
return false;
assert((AddOpIdx >= 0) && "add operand index not right!");
Register RegB = Root.getOperand(AddOpIdx).getReg();
MachineInstr *Prev = MRI->getUniqueVRegDef(RegB);
// Prev must be a valid FMA like instruction.
AddOpIdx = -1;
if (!IsReassociableFMA(*Prev, AddOpIdx, MulOpIdx, false))
return false;
assert((AddOpIdx >= 0) && "add operand index not right!");
Register RegA = Prev->getOperand(AddOpIdx).getReg();
MachineInstr *Leaf = MRI->getUniqueVRegDef(RegA);
AddOpIdx = -1;
if (IsReassociableFMA(*Leaf, AddOpIdx, MulOpIdx, true)) {
Patterns.push_back(MachineCombinerPattern::REASSOC_XMM_AMM_BMM);
LLVM_DEBUG(dbgs() << "add pattern REASSOC_XMM_AMM_BMM\n");
return true;
}
if (IsReassociableAddOrSub(*Leaf, InfoArrayIdxFAddInst)) {
Patterns.push_back(MachineCombinerPattern::REASSOC_XY_AMM_BMM);
LLVM_DEBUG(dbgs() << "add pattern REASSOC_XY_AMM_BMM\n");
return true;
}
return false;
}
void PPCInstrInfo::finalizeInsInstrs(
MachineInstr &Root, MachineCombinerPattern &P,
SmallVectorImpl<MachineInstr *> &InsInstrs) const {
assert(!InsInstrs.empty() && "Instructions set to be inserted is empty!");
MachineFunction *MF = Root.getMF();
MachineRegisterInfo *MRI = &MF->getRegInfo();
const TargetRegisterInfo *TRI = &getRegisterInfo();
MachineConstantPool *MCP = MF->getConstantPool();
int16_t Idx = getFMAOpIdxInfo(Root.getOpcode());
if (Idx < 0)
return;
uint16_t FirstMulOpIdx = FMAOpIdxInfo[Idx][InfoArrayIdxMULOpIdx];
// For now we only need to fix up placeholder for register pressure reduce
// patterns.
Register ConstReg = 0;
switch (P) {
case MachineCombinerPattern::REASSOC_XY_BCA:
ConstReg =
TRI->lookThruCopyLike(Root.getOperand(FirstMulOpIdx).getReg(), MRI);
break;
case MachineCombinerPattern::REASSOC_XY_BAC:
ConstReg =
TRI->lookThruCopyLike(Root.getOperand(FirstMulOpIdx + 1).getReg(), MRI);
break;
default:
// Not register pressure reduce patterns.
return;
}
MachineInstr *ConstDefInstr = MRI->getVRegDef(ConstReg);
// Get const value from const pool.
const Constant *C = getConstantFromConstantPool(ConstDefInstr);
assert(isa<llvm::ConstantFP>(C) && "not a valid constant!");
// Get negative fp const.
APFloat F1((dyn_cast<ConstantFP>(C))->getValueAPF());
F1.changeSign();
Constant *NegC = ConstantFP::get(dyn_cast<ConstantFP>(C)->getContext(), F1);
Align Alignment = MF->getDataLayout().getPrefTypeAlign(C->getType());
// Put negative fp const into constant pool.
unsigned ConstPoolIdx = MCP->getConstantPoolIndex(NegC, Alignment);
MachineOperand *Placeholder = nullptr;
// Record the placeholder PPC::ZERO8 we add in reassociateFMA.
for (auto *Inst : InsInstrs) {
for (MachineOperand &Operand : Inst->explicit_operands()) {
assert(Operand.isReg() && "Invalid instruction in InsInstrs!");
if (Operand.getReg() == PPC::ZERO8) {
Placeholder = &Operand;
break;
}
}
}
assert(Placeholder && "Placeholder does not exist!");
// Generate instructions to load the const fp from constant pool.
// We only support PPC64 and medium code model.
Register LoadNewConst =
generateLoadForNewConst(ConstPoolIdx, &Root, C->getType(), InsInstrs);
// Fill the placeholder with the new load from constant pool.
Placeholder->setReg(LoadNewConst);
}
bool PPCInstrInfo::shouldReduceRegisterPressure(
MachineBasicBlock *MBB, RegisterClassInfo *RegClassInfo) const {
if (!EnableFMARegPressureReduction)
return false;
// Currently, we only enable register pressure reducing in machine combiner
// for: 1: PPC64; 2: Code Model is Medium; 3: Power9 which also has vector
// support.
//
// So we need following instructions to access a TOC entry:
//
// %6:g8rc_and_g8rc_nox0 = ADDIStocHA8 $x2, %const.0
// %7:vssrc = DFLOADf32 target-flags(ppc-toc-lo) %const.0,
// killed %6:g8rc_and_g8rc_nox0, implicit $x2 :: (load 4 from constant-pool)
//
// FIXME: add more supported targets, like Small and Large code model, PPC32,
// AIX.
if (!(Subtarget.isPPC64() && Subtarget.hasP9Vector() &&
Subtarget.getTargetMachine().getCodeModel() == CodeModel::Medium))
return false;
const TargetRegisterInfo *TRI = &getRegisterInfo();
MachineFunction *MF = MBB->getParent();
MachineRegisterInfo *MRI = &MF->getRegInfo();
auto GetMBBPressure = [&](MachineBasicBlock *MBB) -> std::vector<unsigned> {
RegionPressure Pressure;
RegPressureTracker RPTracker(Pressure);
// Initialize the register pressure tracker.
RPTracker.init(MBB->getParent(), RegClassInfo, nullptr, MBB, MBB->end(),
/*TrackLaneMasks*/ false, /*TrackUntiedDefs=*/true);
for (MachineBasicBlock::iterator MII = MBB->instr_end(),
MIE = MBB->instr_begin();
MII != MIE; --MII) {
MachineInstr &MI = *std::prev(MII);
if (MI.isDebugValue() || MI.isDebugLabel())
continue;
RegisterOperands RegOpers;
RegOpers.collect(MI, *TRI, *MRI, false, false);
RPTracker.recedeSkipDebugValues();
assert(&*RPTracker.getPos() == &MI && "RPTracker sync error!");
RPTracker.recede(RegOpers);
}
// Close the RPTracker to finalize live ins.
RPTracker.closeRegion();
return RPTracker.getPressure().MaxSetPressure;
};
// For now we only care about float and double type fma.
unsigned VSSRCLimit = TRI->getRegPressureSetLimit(
*MBB->getParent(), PPC::RegisterPressureSets::VSSRC);
// Only reduce register pressure when pressure is high.
return GetMBBPressure(MBB)[PPC::RegisterPressureSets::VSSRC] >
(float)VSSRCLimit * FMARPFactor;
}
bool PPCInstrInfo::isLoadFromConstantPool(MachineInstr *I) const {
// I has only one memory operand which is load from constant pool.
if (!I->hasOneMemOperand())
return false;
MachineMemOperand *Op = I->memoperands()[0];
return Op->isLoad() && Op->getPseudoValue() &&
Op->getPseudoValue()->kind() == PseudoSourceValue::ConstantPool;
}
Register PPCInstrInfo::generateLoadForNewConst(
unsigned Idx, MachineInstr *MI, Type *Ty,
SmallVectorImpl<MachineInstr *> &InsInstrs) const {
// Now we only support PPC64, Medium code model and P9 with vector.
// We have immutable pattern to access const pool. See function
// shouldReduceRegisterPressure.
assert((Subtarget.isPPC64() && Subtarget.hasP9Vector() &&
Subtarget.getTargetMachine().getCodeModel() == CodeModel::Medium) &&
"Target not supported!\n");
MachineFunction *MF = MI->getMF();
MachineRegisterInfo *MRI = &MF->getRegInfo();
// Generate ADDIStocHA8
Register VReg1 = MRI->createVirtualRegister(&PPC::G8RC_and_G8RC_NOX0RegClass);
MachineInstrBuilder TOCOffset =
BuildMI(*MF, MI->getDebugLoc(), get(PPC::ADDIStocHA8), VReg1)
.addReg(PPC::X2)
.addConstantPoolIndex(Idx);
assert((Ty->isFloatTy() || Ty->isDoubleTy()) &&
"Only float and double are supported!");
unsigned LoadOpcode;
// Should be float type or double type.
if (Ty->isFloatTy())
LoadOpcode = PPC::DFLOADf32;
else
LoadOpcode = PPC::DFLOADf64;
const TargetRegisterClass *RC = MRI->getRegClass(MI->getOperand(0).getReg());
Register VReg2 = MRI->createVirtualRegister(RC);
MachineMemOperand *MMO = MF->getMachineMemOperand(
MachinePointerInfo::getConstantPool(*MF), MachineMemOperand::MOLoad,
Ty->getScalarSizeInBits() / 8, MF->getDataLayout().getPrefTypeAlign(Ty));
// Generate Load from constant pool.
MachineInstrBuilder Load =
BuildMI(*MF, MI->getDebugLoc(), get(LoadOpcode), VReg2)
.addConstantPoolIndex(Idx)
.addReg(VReg1, getKillRegState(true))
.addMemOperand(MMO);
Load->getOperand(1).setTargetFlags(PPCII::MO_TOC_LO);
// Insert the toc load instructions into InsInstrs.
InsInstrs.insert(InsInstrs.begin(), Load);
InsInstrs.insert(InsInstrs.begin(), TOCOffset);
return VReg2;
}
// This function returns the const value in constant pool if the \p I is a load
// from constant pool.
const Constant *
PPCInstrInfo::getConstantFromConstantPool(MachineInstr *I) const {
MachineFunction *MF = I->getMF();
MachineRegisterInfo *MRI = &MF->getRegInfo();
MachineConstantPool *MCP = MF->getConstantPool();
assert(I->mayLoad() && "Should be a load instruction.\n");
for (auto MO : I->uses()) {
if (!MO.isReg())
continue;
Register Reg = MO.getReg();
if (Reg == 0 || !Register::isVirtualRegister(Reg))
continue;
// Find the toc address.
MachineInstr *DefMI = MRI->getVRegDef(Reg);
for (auto MO2 : DefMI->uses())
if (MO2.isCPI())
return (MCP->getConstants())[MO2.getIndex()].Val.ConstVal;
}
return nullptr;
}
bool PPCInstrInfo::getMachineCombinerPatterns(
MachineInstr &Root, SmallVectorImpl<MachineCombinerPattern> &Patterns,
bool DoRegPressureReduce) const {
// Using the machine combiner in this way is potentially expensive, so
// restrict to when aggressive optimizations are desired.
if (Subtarget.getTargetMachine().getOptLevel() != CodeGenOpt::Aggressive)
return false;
if (getFMAPatterns(Root, Patterns, DoRegPressureReduce))
return true;
return TargetInstrInfo::getMachineCombinerPatterns(Root, Patterns,
DoRegPressureReduce);
}
void PPCInstrInfo::genAlternativeCodeSequence(
MachineInstr &Root, MachineCombinerPattern Pattern,
SmallVectorImpl<MachineInstr *> &InsInstrs,
SmallVectorImpl<MachineInstr *> &DelInstrs,
DenseMap<unsigned, unsigned> &InstrIdxForVirtReg) const {
switch (Pattern) {
case MachineCombinerPattern::REASSOC_XY_AMM_BMM:
case MachineCombinerPattern::REASSOC_XMM_AMM_BMM:
case MachineCombinerPattern::REASSOC_XY_BCA:
case MachineCombinerPattern::REASSOC_XY_BAC:
reassociateFMA(Root, Pattern, InsInstrs, DelInstrs, InstrIdxForVirtReg);
break;
default:
// Reassociate default patterns.
TargetInstrInfo::genAlternativeCodeSequence(Root, Pattern, InsInstrs,
DelInstrs, InstrIdxForVirtReg);
break;
}
}
void PPCInstrInfo::reassociateFMA(
MachineInstr &Root, MachineCombinerPattern Pattern,
SmallVectorImpl<MachineInstr *> &InsInstrs,
SmallVectorImpl<MachineInstr *> &DelInstrs,
DenseMap<unsigned, unsigned> &InstrIdxForVirtReg) const {
MachineFunction *MF = Root.getMF();
MachineRegisterInfo &MRI = MF->getRegInfo();
const TargetRegisterInfo *TRI = &getRegisterInfo();
MachineOperand &OpC = Root.getOperand(0);
Register RegC = OpC.getReg();
const TargetRegisterClass *RC = MRI.getRegClass(RegC);
MRI.constrainRegClass(RegC, RC);
unsigned FmaOp = Root.getOpcode();
int16_t Idx = getFMAOpIdxInfo(FmaOp);
assert(Idx >= 0 && "Root must be a FMA instruction");
bool IsILPReassociate =
(Pattern == MachineCombinerPattern::REASSOC_XY_AMM_BMM) ||
(Pattern == MachineCombinerPattern::REASSOC_XMM_AMM_BMM);
uint16_t AddOpIdx = FMAOpIdxInfo[Idx][InfoArrayIdxAddOpIdx];
uint16_t FirstMulOpIdx = FMAOpIdxInfo[Idx][InfoArrayIdxMULOpIdx];
MachineInstr *Prev = nullptr;
MachineInstr *Leaf = nullptr;
switch (Pattern) {
default:
llvm_unreachable("not recognized pattern!");
case MachineCombinerPattern::REASSOC_XY_AMM_BMM:
case MachineCombinerPattern::REASSOC_XMM_AMM_BMM:
Prev = MRI.getUniqueVRegDef(Root.getOperand(AddOpIdx).getReg());
Leaf = MRI.getUniqueVRegDef(Prev->getOperand(AddOpIdx).getReg());
break;
case MachineCombinerPattern::REASSOC_XY_BAC: {
Register MULReg =
TRI->lookThruCopyLike(Root.getOperand(FirstMulOpIdx).getReg(), &MRI);
Leaf = MRI.getVRegDef(MULReg);
break;
}
case MachineCombinerPattern::REASSOC_XY_BCA: {
Register MULReg = TRI->lookThruCopyLike(
Root.getOperand(FirstMulOpIdx + 1).getReg(), &MRI);
Leaf = MRI.getVRegDef(MULReg);
break;
}
}
uint16_t IntersectedFlags = 0;
if (IsILPReassociate)
IntersectedFlags = Root.getFlags() & Prev->getFlags() & Leaf->getFlags();
else
IntersectedFlags = Root.getFlags() & Leaf->getFlags();
auto GetOperandInfo = [&](const MachineOperand &Operand, Register &Reg,
bool &KillFlag) {
Reg = Operand.getReg();
MRI.constrainRegClass(Reg, RC);
KillFlag = Operand.isKill();
};
auto GetFMAInstrInfo = [&](const MachineInstr &Instr, Register &MulOp1,
Register &MulOp2, Register &AddOp,
bool &MulOp1KillFlag, bool &MulOp2KillFlag,
bool &AddOpKillFlag) {
GetOperandInfo(Instr.getOperand(FirstMulOpIdx), MulOp1, MulOp1KillFlag);
GetOperandInfo(Instr.getOperand(FirstMulOpIdx + 1), MulOp2, MulOp2KillFlag);
GetOperandInfo(Instr.getOperand(AddOpIdx), AddOp, AddOpKillFlag);
};
Register RegM11, RegM12, RegX, RegY, RegM21, RegM22, RegM31, RegM32, RegA11,
RegA21, RegB;
bool KillX = false, KillY = false, KillM11 = false, KillM12 = false,
KillM21 = false, KillM22 = false, KillM31 = false, KillM32 = false,
KillA11 = false, KillA21 = false, KillB = false;
GetFMAInstrInfo(Root, RegM31, RegM32, RegB, KillM31, KillM32, KillB);
if (IsILPReassociate)
GetFMAInstrInfo(*Prev, RegM21, RegM22, RegA21, KillM21, KillM22, KillA21);
if (Pattern == MachineCombinerPattern::REASSOC_XMM_AMM_BMM) {
GetFMAInstrInfo(*Leaf, RegM11, RegM12, RegA11, KillM11, KillM12, KillA11);
GetOperandInfo(Leaf->getOperand(AddOpIdx), RegX, KillX);
} else if (Pattern == MachineCombinerPattern::REASSOC_XY_AMM_BMM) {
GetOperandInfo(Leaf->getOperand(1), RegX, KillX);
GetOperandInfo(Leaf->getOperand(2), RegY, KillY);
} else {
// Get FSUB instruction info.
GetOperandInfo(Leaf->getOperand(1), RegX, KillX);
GetOperandInfo(Leaf->getOperand(2), RegY, KillY);
}
// Create new virtual registers for the new results instead of
// recycling legacy ones because the MachineCombiner's computation of the
// critical path requires a new register definition rather than an existing
// one.
// For register pressure reassociation, we only need create one virtual
// register for the new fma.
Register NewVRA = MRI.createVirtualRegister(RC);
InstrIdxForVirtReg.insert(std::make_pair(NewVRA, 0));
Register NewVRB = 0;
if (IsILPReassociate) {
NewVRB = MRI.createVirtualRegister(RC);
InstrIdxForVirtReg.insert(std::make_pair(NewVRB, 1));
}
Register NewVRD = 0;
if (Pattern == MachineCombinerPattern::REASSOC_XMM_AMM_BMM) {
NewVRD = MRI.createVirtualRegister(RC);
InstrIdxForVirtReg.insert(std::make_pair(NewVRD, 2));
}
auto AdjustOperandOrder = [&](MachineInstr *MI, Register RegAdd, bool KillAdd,
Register RegMul1, bool KillRegMul1,
Register RegMul2, bool KillRegMul2) {
MI->getOperand(AddOpIdx).setReg(RegAdd);
MI->getOperand(AddOpIdx).setIsKill(KillAdd);
MI->getOperand(FirstMulOpIdx).setReg(RegMul1);
MI->getOperand(FirstMulOpIdx).setIsKill(KillRegMul1);
MI->getOperand(FirstMulOpIdx + 1).setReg(RegMul2);
MI->getOperand(FirstMulOpIdx + 1).setIsKill(KillRegMul2);
};
MachineInstrBuilder NewARegPressure, NewCRegPressure;
switch (Pattern) {
default:
llvm_unreachable("not recognized pattern!");
case MachineCombinerPattern::REASSOC_XY_AMM_BMM: {
// Create new instructions for insertion.
MachineInstrBuilder MINewB =
BuildMI(*MF, Prev->getDebugLoc(), get(FmaOp), NewVRB)
.addReg(RegX, getKillRegState(KillX))
.addReg(RegM21, getKillRegState(KillM21))
.addReg(RegM22, getKillRegState(KillM22));
MachineInstrBuilder MINewA =
BuildMI(*MF, Root.getDebugLoc(), get(FmaOp), NewVRA)
.addReg(RegY, getKillRegState(KillY))
.addReg(RegM31, getKillRegState(KillM31))
.addReg(RegM32, getKillRegState(KillM32));
// If AddOpIdx is not 1, adjust the order.
if (AddOpIdx != 1) {
AdjustOperandOrder(MINewB, RegX, KillX, RegM21, KillM21, RegM22, KillM22);
AdjustOperandOrder(MINewA, RegY, KillY, RegM31, KillM31, RegM32, KillM32);
}
MachineInstrBuilder MINewC =
BuildMI(*MF, Root.getDebugLoc(),
get(FMAOpIdxInfo[Idx][InfoArrayIdxFAddInst]), RegC)
.addReg(NewVRB, getKillRegState(true))
.addReg(NewVRA, getKillRegState(true));
// Update flags for newly created instructions.
setSpecialOperandAttr(*MINewA, IntersectedFlags);
setSpecialOperandAttr(*MINewB, IntersectedFlags);
setSpecialOperandAttr(*MINewC, IntersectedFlags);
// Record new instructions for insertion.
InsInstrs.push_back(MINewA);
InsInstrs.push_back(MINewB);
InsInstrs.push_back(MINewC);
break;
}
case MachineCombinerPattern::REASSOC_XMM_AMM_BMM: {
assert(NewVRD && "new FMA register not created!");
// Create new instructions for insertion.
MachineInstrBuilder MINewA =
BuildMI(*MF, Leaf->getDebugLoc(),
get(FMAOpIdxInfo[Idx][InfoArrayIdxFMULInst]), NewVRA)
.addReg(RegM11, getKillRegState(KillM11))
.addReg(RegM12, getKillRegState(KillM12));
MachineInstrBuilder MINewB =
BuildMI(*MF, Prev->getDebugLoc(), get(FmaOp), NewVRB)
.addReg(RegX, getKillRegState(KillX))
.addReg(RegM21, getKillRegState(KillM21))
.addReg(RegM22, getKillRegState(KillM22));
MachineInstrBuilder MINewD =
BuildMI(*MF, Root.getDebugLoc(), get(FmaOp), NewVRD)
.addReg(NewVRA, getKillRegState(true))
.addReg(RegM31, getKillRegState(KillM31))
.addReg(RegM32, getKillRegState(KillM32));
// If AddOpIdx is not 1, adjust the order.
if (AddOpIdx != 1) {
AdjustOperandOrder(MINewB, RegX, KillX, RegM21, KillM21, RegM22, KillM22);
AdjustOperandOrder(MINewD, NewVRA, true, RegM31, KillM31, RegM32,
KillM32);
}
MachineInstrBuilder MINewC =
BuildMI(*MF, Root.getDebugLoc(),
get(FMAOpIdxInfo[Idx][InfoArrayIdxFAddInst]), RegC)
.addReg(NewVRB, getKillRegState(true))
.addReg(NewVRD, getKillRegState(true));
// Update flags for newly created instructions.
setSpecialOperandAttr(*MINewA, IntersectedFlags);
setSpecialOperandAttr(*MINewB, IntersectedFlags);
setSpecialOperandAttr(*MINewD, IntersectedFlags);
setSpecialOperandAttr(*MINewC, IntersectedFlags);
// Record new instructions for insertion.
InsInstrs.push_back(MINewA);
InsInstrs.push_back(MINewB);
InsInstrs.push_back(MINewD);
InsInstrs.push_back(MINewC);
break;
}
case MachineCombinerPattern::REASSOC_XY_BAC:
case MachineCombinerPattern::REASSOC_XY_BCA: {
Register VarReg;
bool KillVarReg = false;
if (Pattern == MachineCombinerPattern::REASSOC_XY_BCA) {
VarReg = RegM31;
KillVarReg = KillM31;
} else {
VarReg = RegM32;
KillVarReg = KillM32;
}
// We don't want to get negative const from memory pool too early, as the
// created entry will not be deleted even if it has no users. Since all
// operand of Leaf and Root are virtual register, we use zero register
// here as a placeholder. When the InsInstrs is selected in
// MachineCombiner, we call finalizeInsInstrs to replace the zero register
// with a virtual register which is a load from constant pool.
NewARegPressure = BuildMI(*MF, Root.getDebugLoc(), get(FmaOp), NewVRA)
.addReg(RegB, getKillRegState(RegB))
.addReg(RegY, getKillRegState(KillY))
.addReg(PPC::ZERO8);
NewCRegPressure = BuildMI(*MF, Root.getDebugLoc(), get(FmaOp), RegC)
.addReg(NewVRA, getKillRegState(true))
.addReg(RegX, getKillRegState(KillX))
.addReg(VarReg, getKillRegState(KillVarReg));
// For now, we only support xsmaddadp/xsmaddasp, their add operand are
// both at index 1, no need to adjust.
// FIXME: when add more fma instructions support, like fma/fmas, adjust
// the operand index here.
break;
}
}
if (!IsILPReassociate) {
setSpecialOperandAttr(*NewARegPressure, IntersectedFlags);
setSpecialOperandAttr(*NewCRegPressure, IntersectedFlags);
InsInstrs.push_back(NewARegPressure);
InsInstrs.push_back(NewCRegPressure);
}
assert(!InsInstrs.empty() &&
"Insertion instructions set should not be empty!");
// Record old instructions for deletion.
DelInstrs.push_back(Leaf);
if (IsILPReassociate)
DelInstrs.push_back(Prev);
DelInstrs.push_back(&Root);
}
// Detect 32 -> 64-bit extensions where we may reuse the low sub-register.
bool PPCInstrInfo::isCoalescableExtInstr(const MachineInstr &MI,
Register &SrcReg, Register &DstReg,
unsigned &SubIdx) const {
switch (MI.getOpcode()) {
default: return false;
case PPC::EXTSW:
case PPC::EXTSW_32:
case PPC::EXTSW_32_64:
SrcReg = MI.getOperand(1).getReg();
DstReg = MI.getOperand(0).getReg();
SubIdx = PPC::sub_32;
return true;
}
}
unsigned PPCInstrInfo::isLoadFromStackSlot(const MachineInstr &MI,
int &FrameIndex) const {
unsigned Opcode = MI.getOpcode();
const unsigned *OpcodesForSpill = getLoadOpcodesForSpillArray();
const unsigned *End = OpcodesForSpill + SOK_LastOpcodeSpill;
if (End != std::find(OpcodesForSpill, End, Opcode)) {
// Check for the operands added by addFrameReference (the immediate is the
// offset which defaults to 0).
if (MI.getOperand(1).isImm() && !MI.getOperand(1).getImm() &&
MI.getOperand(2).isFI()) {
FrameIndex = MI.getOperand(2).getIndex();
return MI.getOperand(0).getReg();
}
}
return 0;
}
// For opcodes with the ReMaterializable flag set, this function is called to
// verify the instruction is really rematable.
bool PPCInstrInfo::isReallyTriviallyReMaterializable(const MachineInstr &MI,
AliasAnalysis *AA) const {
switch (MI.getOpcode()) {
default:
// This function should only be called for opcodes with the ReMaterializable
// flag set.
llvm_unreachable("Unknown rematerializable operation!");
break;
case PPC::LI:
case PPC::LI8:
case PPC::PLI:
case PPC::PLI8:
case PPC::LIS:
case PPC::LIS8:
case PPC::ADDIStocHA:
case PPC::ADDIStocHA8:
case PPC::ADDItocL:
case PPC::LOAD_STACK_GUARD:
case PPC::XXLXORz:
case PPC::XXLXORspz:
case PPC::XXLXORdpz:
case PPC::XXLEQVOnes:
case PPC::XXSPLTI32DX:
case PPC::XXSPLTIW:
case PPC::XXSPLTIDP:
case PPC::V_SET0B:
case PPC::V_SET0H:
case PPC::V_SET0:
case PPC::V_SETALLONESB:
case PPC::V_SETALLONESH:
case PPC::V_SETALLONES:
case PPC::CRSET:
case PPC::CRUNSET:
case PPC::XXSETACCZ:
return true;
}
return false;
}
unsigned PPCInstrInfo::isStoreToStackSlot(const MachineInstr &MI,
int &FrameIndex) const {
unsigned Opcode = MI.getOpcode();
const unsigned *OpcodesForSpill = getStoreOpcodesForSpillArray();
const unsigned *End = OpcodesForSpill + SOK_LastOpcodeSpill;
if (End != std::find(OpcodesForSpill, End, Opcode)) {
if (MI.getOperand(1).isImm() && !MI.getOperand(1).getImm() &&
MI.getOperand(2).isFI()) {
FrameIndex = MI.getOperand(2).getIndex();
return MI.getOperand(0).getReg();
}
}
return 0;
}
MachineInstr *PPCInstrInfo::commuteInstructionImpl(MachineInstr &MI, bool NewMI,
unsigned OpIdx1,
unsigned OpIdx2) const {
MachineFunction &MF = *MI.getParent()->getParent();
// Normal instructions can be commuted the obvious way.
if (MI.getOpcode() != PPC::RLWIMI && MI.getOpcode() != PPC::RLWIMI_rec)
return TargetInstrInfo::commuteInstructionImpl(MI, NewMI, OpIdx1, OpIdx2);
// Note that RLWIMI can be commuted as a 32-bit instruction, but not as a
// 64-bit instruction (so we don't handle PPC::RLWIMI8 here), because
// changing the relative order of the mask operands might change what happens
// to the high-bits of the mask (and, thus, the result).
// Cannot commute if it has a non-zero rotate count.
if (MI.getOperand(3).getImm() != 0)
return nullptr;
// If we have a zero rotate count, we have:
// M = mask(MB,ME)
// Op0 = (Op1 & ~M) | (Op2 & M)
// Change this to:
// M = mask((ME+1)&31, (MB-1)&31)
// Op0 = (Op2 & ~M) | (Op1 & M)
// Swap op1/op2
assert(((OpIdx1 == 1 && OpIdx2 == 2) || (OpIdx1 == 2 && OpIdx2 == 1)) &&
"Only the operands 1 and 2 can be swapped in RLSIMI/RLWIMI_rec.");
Register Reg0 = MI.getOperand(0).getReg();
Register Reg1 = MI.getOperand(1).getReg();
Register Reg2 = MI.getOperand(2).getReg();
unsigned SubReg1 = MI.getOperand(1).getSubReg();
unsigned SubReg2 = MI.getOperand(2).getSubReg();
bool Reg1IsKill = MI.getOperand(1).isKill();
bool Reg2IsKill = MI.getOperand(2).isKill();
bool ChangeReg0 = false;
// If machine instrs are no longer in two-address forms, update
// destination register as well.
if (Reg0 == Reg1) {
// Must be two address instruction!
assert(MI.getDesc().getOperandConstraint(0, MCOI::TIED_TO) &&
"Expecting a two-address instruction!");
assert(MI.getOperand(0).getSubReg() == SubReg1 && "Tied subreg mismatch");
Reg2IsKill = false;
ChangeReg0 = true;
}
// Masks.
unsigned MB = MI.getOperand(4).getImm();
unsigned ME = MI.getOperand(5).getImm();
// We can't commute a trivial mask (there is no way to represent an all-zero
// mask).
if (MB == 0 && ME == 31)
return nullptr;
if (NewMI) {
// Create a new instruction.
Register Reg0 = ChangeReg0 ? Reg2 : MI.getOperand(0).getReg();
bool Reg0IsDead = MI.getOperand(0).isDead();
return BuildMI(MF, MI.getDebugLoc(), MI.getDesc())
.addReg(Reg0, RegState::Define | getDeadRegState(Reg0IsDead))
.addReg(Reg2, getKillRegState(Reg2IsKill))
.addReg(Reg1, getKillRegState(Reg1IsKill))
.addImm((ME + 1) & 31)
.addImm((MB - 1) & 31);
}
if (ChangeReg0) {
MI.getOperand(0).setReg(Reg2);
MI.getOperand(0).setSubReg(SubReg2);
}
MI.getOperand(2).setReg(Reg1);
MI.getOperand(1).setReg(Reg2);
MI.getOperand(2).setSubReg(SubReg1);
MI.getOperand(1).setSubReg(SubReg2);
MI.getOperand(2).setIsKill(Reg1IsKill);
MI.getOperand(1).setIsKill(Reg2IsKill);
// Swap the mask around.
MI.getOperand(4).setImm((ME + 1) & 31);
MI.getOperand(5).setImm((MB - 1) & 31);
return &MI;
}
bool PPCInstrInfo::findCommutedOpIndices(const MachineInstr &MI,
unsigned &SrcOpIdx1,
unsigned &SrcOpIdx2) const {
// For VSX A-Type FMA instructions, it is the first two operands that can be
// commuted, however, because the non-encoded tied input operand is listed
// first, the operands to swap are actually the second and third.
int AltOpc = PPC::getAltVSXFMAOpcode(MI.getOpcode());
if (AltOpc == -1)
return TargetInstrInfo::findCommutedOpIndices(MI, SrcOpIdx1, SrcOpIdx2);
// The commutable operand indices are 2 and 3. Return them in SrcOpIdx1
// and SrcOpIdx2.
return fixCommutedOpIndices(SrcOpIdx1, SrcOpIdx2, 2, 3);
}
void PPCInstrInfo::insertNoop(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI) const {
// This function is used for scheduling, and the nop wanted here is the type
// that terminates dispatch groups on the POWER cores.
unsigned Directive = Subtarget.getCPUDirective();
unsigned Opcode;
switch (Directive) {
default: Opcode = PPC::NOP; break;
case PPC::DIR_PWR6: Opcode = PPC::NOP_GT_PWR6; break;
case PPC::DIR_PWR7: Opcode = PPC::NOP_GT_PWR7; break;
case PPC::DIR_PWR8: Opcode = PPC::NOP_GT_PWR7; break; /* FIXME: Update when P8 InstrScheduling model is ready */
// FIXME: Update when POWER9 scheduling model is ready.
case PPC::DIR_PWR9: Opcode = PPC::NOP_GT_PWR7; break;
}
DebugLoc DL;
BuildMI(MBB, MI, DL, get(Opcode));
}
/// Return the noop instruction to use for a noop.
MCInst PPCInstrInfo::getNop() const {
MCInst Nop;
Nop.setOpcode(PPC::NOP);
return Nop;
}
// Branch analysis.
// Note: If the condition register is set to CTR or CTR8 then this is a
// BDNZ (imm == 1) or BDZ (imm == 0) branch.
bool PPCInstrInfo::analyzeBranch(MachineBasicBlock &MBB,
MachineBasicBlock *&TBB,
MachineBasicBlock *&FBB,
SmallVectorImpl<MachineOperand> &Cond,
bool AllowModify) const {
bool isPPC64 = Subtarget.isPPC64();
// If the block has no terminators, it just falls into the block after it.
MachineBasicBlock::iterator I = MBB.getLastNonDebugInstr();
if (I == MBB.end())
return false;
if (!isUnpredicatedTerminator(*I))
return false;
if (AllowModify) {
// If the BB ends with an unconditional branch to the fallthrough BB,
// we eliminate the branch instruction.
if (I->getOpcode() == PPC::B &&
MBB.isLayoutSuccessor(I->getOperand(0).getMBB())) {
I->eraseFromParent();
// We update iterator after deleting the last branch.
I = MBB.getLastNonDebugInstr();
if (I == MBB.end() || !isUnpredicatedTerminator(*I))
return false;
}
}
// Get the last instruction in the block.
MachineInstr &LastInst = *I;
// If there is only one terminator instruction, process it.
if (I == MBB.begin() || !isUnpredicatedTerminator(*--I)) {
if (LastInst.getOpcode() == PPC::B) {
if (!LastInst.getOperand(0).isMBB())
return true;
TBB = LastInst.getOperand(0).getMBB();
return false;
} else if (LastInst.getOpcode() == PPC::BCC) {
if (!LastInst.getOperand(2).isMBB())
return true;
// Block ends with fall-through condbranch.
TBB = LastInst.getOperand(2).getMBB();
Cond.push_back(LastInst.getOperand(0));
Cond.push_back(LastInst.getOperand(1));
return false;
} else if (LastInst.getOpcode() == PPC::BC) {
if (!LastInst.getOperand(1).isMBB())
return true;
// Block ends with fall-through condbranch.
TBB = LastInst.getOperand(1).getMBB();
Cond.push_back(MachineOperand::CreateImm(PPC::PRED_BIT_SET));
Cond.push_back(LastInst.getOperand(0));
return false;
} else if (LastInst.getOpcode() == PPC::BCn) {
if (!LastInst.getOperand(1).isMBB())
return true;
// Block ends with fall-through condbranch.
TBB = LastInst.getOperand(1).getMBB();
Cond.push_back(MachineOperand::CreateImm(PPC::PRED_BIT_UNSET));
Cond.push_back(LastInst.getOperand(0));
return false;
} else if (LastInst.getOpcode() == PPC::BDNZ8 ||
LastInst.getOpcode() == PPC::BDNZ) {
if (!LastInst.getOperand(0).isMBB())
return true;
if (DisableCTRLoopAnal)
return true;
TBB = LastInst.getOperand(0).getMBB();
Cond.push_back(MachineOperand::CreateImm(1));
Cond.push_back(MachineOperand::CreateReg(isPPC64 ? PPC::CTR8 : PPC::CTR,
true));
return false;
} else if (LastInst.getOpcode() == PPC::BDZ8 ||
LastInst.getOpcode() == PPC::BDZ) {
if (!LastInst.getOperand(0).isMBB())
return true;
if (DisableCTRLoopAnal)
return true;
TBB = LastInst.getOperand(0).getMBB();
Cond.push_back(MachineOperand::CreateImm(0));
Cond.push_back(MachineOperand::CreateReg(isPPC64 ? PPC::CTR8 : PPC::CTR,
true));
return false;
}
// Otherwise, don't know what this is.
return true;
}
// Get the instruction before it if it's a terminator.
MachineInstr &SecondLastInst = *I;
// If there are three terminators, we don't know what sort of block this is.
if (I != MBB.begin() && isUnpredicatedTerminator(*--I))
return true;
// If the block ends with PPC::B and PPC:BCC, handle it.
if (SecondLastInst.getOpcode() == PPC::BCC &&
LastInst.getOpcode() == PPC::B) {
if (!SecondLastInst.getOperand(2).isMBB() ||
!LastInst.getOperand(0).isMBB())
return true;
TBB = SecondLastInst.getOperand(2).getMBB();
Cond.push_back(SecondLastInst.getOperand(0));
Cond.push_back(SecondLastInst.getOperand(1));
FBB = LastInst.getOperand(0).getMBB();
return false;
} else if (SecondLastInst.getOpcode() == PPC::BC &&
LastInst.getOpcode() == PPC::B) {
if (!SecondLastInst.getOperand(1).isMBB() ||
!LastInst.getOperand(0).isMBB())
return true;
TBB = SecondLastInst.getOperand(1).getMBB();
Cond.push_back(MachineOperand::CreateImm(PPC::PRED_BIT_SET));
Cond.push_back(SecondLastInst.getOperand(0));
FBB = LastInst.getOperand(0).getMBB();
return false;
} else if (SecondLastInst.getOpcode() == PPC::BCn &&
LastInst.getOpcode() == PPC::B) {
if (!SecondLastInst.getOperand(1).isMBB() ||
!LastInst.getOperand(0).isMBB())
return true;
TBB = SecondLastInst.getOperand(1).getMBB();
Cond.push_back(MachineOperand::CreateImm(PPC::PRED_BIT_UNSET));
Cond.push_back(SecondLastInst.getOperand(0));
FBB = LastInst.getOperand(0).getMBB();
return false;
} else if ((SecondLastInst.getOpcode() == PPC::BDNZ8 ||
SecondLastInst.getOpcode() == PPC::BDNZ) &&
LastInst.getOpcode() == PPC::B) {
if (!SecondLastInst.getOperand(0).isMBB() ||
!LastInst.getOperand(0).isMBB())
return true;
if (DisableCTRLoopAnal)
return true;
TBB = SecondLastInst.getOperand(0).getMBB();
Cond.push_back(MachineOperand::CreateImm(1));
Cond.push_back(MachineOperand::CreateReg(isPPC64 ? PPC::CTR8 : PPC::CTR,
true));
FBB = LastInst.getOperand(0).getMBB();
return false;
} else if ((SecondLastInst.getOpcode() == PPC::BDZ8 ||
SecondLastInst.getOpcode() == PPC::BDZ) &&
LastInst.getOpcode() == PPC::B) {
if (!SecondLastInst.getOperand(0).isMBB() ||
!LastInst.getOperand(0).isMBB())
return true;
if (DisableCTRLoopAnal)
return true;
TBB = SecondLastInst.getOperand(0).getMBB();
Cond.push_back(MachineOperand::CreateImm(0));
Cond.push_back(MachineOperand::CreateReg(isPPC64 ? PPC::CTR8 : PPC::CTR,
true));
FBB = LastInst.getOperand(0).getMBB();
return false;
}
// If the block ends with two PPC:Bs, handle it. The second one is not
// executed, so remove it.
if (SecondLastInst.getOpcode() == PPC::B && LastInst.getOpcode() == PPC::B) {
if (!SecondLastInst.getOperand(0).isMBB())
return true;
TBB = SecondLastInst.getOperand(0).getMBB();
I = LastInst;
if (AllowModify)
I->eraseFromParent();
return false;
}
// Otherwise, can't handle this.
return true;
}
unsigned PPCInstrInfo::removeBranch(MachineBasicBlock &MBB,
int *BytesRemoved) const {
assert(!BytesRemoved && "code size not handled");
MachineBasicBlock::iterator I = MBB.getLastNonDebugInstr();
if (I == MBB.end())
return 0;
if (I->getOpcode() != PPC::B && I->getOpcode() != PPC::BCC &&
I->getOpcode() != PPC::BC && I->getOpcode() != PPC::BCn &&
I->getOpcode() != PPC::BDNZ8 && I->getOpcode() != PPC::BDNZ &&
I->getOpcode() != PPC::BDZ8 && I->getOpcode() != PPC::BDZ)
return 0;
// Remove the branch.
I->eraseFromParent();
I = MBB.end();
if (I == MBB.begin()) return 1;
--I;
if (I->getOpcode() != PPC::BCC &&
I->getOpcode() != PPC::BC && I->getOpcode() != PPC::BCn &&
I->getOpcode() != PPC::BDNZ8 && I->getOpcode() != PPC::BDNZ &&
I->getOpcode() != PPC::BDZ8 && I->getOpcode() != PPC::BDZ)
return 1;
// Remove the branch.
I->eraseFromParent();
return 2;
}
unsigned PPCInstrInfo::insertBranch(MachineBasicBlock &MBB,
MachineBasicBlock *TBB,
MachineBasicBlock *FBB,
ArrayRef<MachineOperand> Cond,
const DebugLoc &DL,
int *BytesAdded) const {
// Shouldn't be a fall through.
assert(TBB && "insertBranch must not be told to insert a fallthrough");
assert((Cond.size() == 2 || Cond.size() == 0) &&
"PPC branch conditions have two components!");
assert(!BytesAdded && "code size not handled");
bool isPPC64 = Subtarget.isPPC64();
// One-way branch.
if (!FBB) {
if (Cond.empty()) // Unconditional branch
BuildMI(&MBB, DL, get(PPC::B)).addMBB(TBB);
else if (Cond[1].getReg() == PPC::CTR || Cond[1].getReg() == PPC::CTR8)
BuildMI(&MBB, DL, get(Cond[0].getImm() ?
(isPPC64 ? PPC::BDNZ8 : PPC::BDNZ) :
(isPPC64 ? PPC::BDZ8 : PPC::BDZ))).addMBB(TBB);
else if (Cond[0].getImm() == PPC::PRED_BIT_SET)
BuildMI(&MBB, DL, get(PPC::BC)).add(Cond[1]).addMBB(TBB);
else if (Cond[0].getImm() == PPC::PRED_BIT_UNSET)
BuildMI(&MBB, DL, get(PPC::BCn)).add(Cond[1]).addMBB(TBB);
else // Conditional branch
BuildMI(&MBB, DL, get(PPC::BCC))
.addImm(Cond[0].getImm())
.add(Cond[1])
.addMBB(TBB);
return 1;
}
// Two-way Conditional Branch.
if (Cond[1].getReg() == PPC::CTR || Cond[1].getReg() == PPC::CTR8)
BuildMI(&MBB, DL, get(Cond[0].getImm() ?
(isPPC64 ? PPC::BDNZ8 : PPC::BDNZ) :
(isPPC64 ? PPC::BDZ8 : PPC::BDZ))).addMBB(TBB);
else if (Cond[0].getImm() == PPC::PRED_BIT_SET)
BuildMI(&MBB, DL, get(PPC::BC)).add(Cond[1]).addMBB(TBB);
else if (Cond[0].getImm() == PPC::PRED_BIT_UNSET)
BuildMI(&MBB, DL, get(PPC::BCn)).add(Cond[1]).addMBB(TBB);
else
BuildMI(&MBB, DL, get(PPC::BCC))
.addImm(Cond[0].getImm())
.add(Cond[1])
.addMBB(TBB);
BuildMI(&MBB, DL, get(PPC::B)).addMBB(FBB);
return 2;
}
// Select analysis.
bool PPCInstrInfo::canInsertSelect(const MachineBasicBlock &MBB,
ArrayRef<MachineOperand> Cond,
Register DstReg, Register TrueReg,
Register FalseReg, int &CondCycles,
int &TrueCycles, int &FalseCycles) const {
if (Cond.size() != 2)
return false;
// If this is really a bdnz-like condition, then it cannot be turned into a
// select.
if (Cond[1].getReg() == PPC::CTR || Cond[1].getReg() == PPC::CTR8)
return false;
// If the conditional branch uses a physical register, then it cannot be
// turned into a select.
if (Register::isPhysicalRegister(Cond[1].getReg()))
return false;
// Check register classes.
const MachineRegisterInfo &MRI = MBB.getParent()->getRegInfo();
const TargetRegisterClass *RC =
RI.getCommonSubClass(MRI.getRegClass(TrueReg), MRI.getRegClass(FalseReg));
if (!RC)
return false;
// isel is for regular integer GPRs only.
if (!PPC::GPRCRegClass.hasSubClassEq(RC) &&
!PPC::GPRC_NOR0RegClass.hasSubClassEq(RC) &&
!PPC::G8RCRegClass.hasSubClassEq(RC) &&
!PPC::G8RC_NOX0RegClass.hasSubClassEq(RC))
return false;
// FIXME: These numbers are for the A2, how well they work for other cores is
// an open question. On the A2, the isel instruction has a 2-cycle latency
// but single-cycle throughput. These numbers are used in combination with
// the MispredictPenalty setting from the active SchedMachineModel.
CondCycles = 1;
TrueCycles = 1;
FalseCycles = 1;
return true;
}
void PPCInstrInfo::insertSelect(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI,
const DebugLoc &dl, Register DestReg,
ArrayRef<MachineOperand> Cond, Register TrueReg,
Register FalseReg) const {
assert(Cond.size() == 2 &&
"PPC branch conditions have two components!");
// Get the register classes.
MachineRegisterInfo &MRI = MBB.getParent()->getRegInfo();
const TargetRegisterClass *RC =
RI.getCommonSubClass(MRI.getRegClass(TrueReg), MRI.getRegClass(FalseReg));
assert(RC && "TrueReg and FalseReg must have overlapping register classes");
bool Is64Bit = PPC::G8RCRegClass.hasSubClassEq(RC) ||
PPC::G8RC_NOX0RegClass.hasSubClassEq(RC);
assert((Is64Bit ||
PPC::GPRCRegClass.hasSubClassEq(RC) ||
PPC::GPRC_NOR0RegClass.hasSubClassEq(RC)) &&
"isel is for regular integer GPRs only");
unsigned OpCode = Is64Bit ? PPC::ISEL8 : PPC::ISEL;
auto SelectPred = static_cast<PPC::Predicate>(Cond[0].getImm());
unsigned SubIdx = 0;
bool SwapOps = false;
switch (SelectPred) {
case PPC::PRED_EQ:
case PPC::PRED_EQ_MINUS:
case PPC::PRED_EQ_PLUS:
SubIdx = PPC::sub_eq; SwapOps = false; break;
case PPC::PRED_NE:
case PPC::PRED_NE_MINUS:
case PPC::PRED_NE_PLUS:
SubIdx = PPC::sub_eq; SwapOps = true; break;
case PPC::PRED_LT:
case PPC::PRED_LT_MINUS:
case PPC::PRED_LT_PLUS:
SubIdx = PPC::sub_lt; SwapOps = false; break;
case PPC::PRED_GE:
case PPC::PRED_GE_MINUS:
case PPC::PRED_GE_PLUS:
SubIdx = PPC::sub_lt; SwapOps = true; break;
case PPC::PRED_GT:
case PPC::PRED_GT_MINUS:
case PPC::PRED_GT_PLUS:
SubIdx = PPC::sub_gt; SwapOps = false; break;
case PPC::PRED_LE:
case PPC::PRED_LE_MINUS:
case PPC::PRED_LE_PLUS:
SubIdx = PPC::sub_gt; SwapOps = true; break;
case PPC::PRED_UN:
case PPC::PRED_UN_MINUS:
case PPC::PRED_UN_PLUS:
SubIdx = PPC::sub_un; SwapOps = false; break;
case PPC::PRED_NU:
case PPC::PRED_NU_MINUS:
case PPC::PRED_NU_PLUS:
SubIdx = PPC::sub_un; SwapOps = true; break;
case PPC::PRED_BIT_SET: SubIdx = 0; SwapOps = false; break;
case PPC::PRED_BIT_UNSET: SubIdx = 0; SwapOps = true; break;
}
Register FirstReg = SwapOps ? FalseReg : TrueReg,
SecondReg = SwapOps ? TrueReg : FalseReg;
// The first input register of isel cannot be r0. If it is a member
// of a register class that can be r0, then copy it first (the
// register allocator should eliminate the copy).
if (MRI.getRegClass(FirstReg)->contains(PPC::R0) ||
MRI.getRegClass(FirstReg)->contains(PPC::X0)) {
const TargetRegisterClass *FirstRC =
MRI.getRegClass(FirstReg)->contains(PPC::X0) ?
&PPC::G8RC_NOX0RegClass : &PPC::GPRC_NOR0RegClass;
Register OldFirstReg = FirstReg;
FirstReg = MRI.createVirtualRegister(FirstRC);
BuildMI(MBB, MI, dl, get(TargetOpcode::COPY), FirstReg)
.addReg(OldFirstReg);
}
BuildMI(MBB, MI, dl, get(OpCode), DestReg)
.addReg(FirstReg).addReg(SecondReg)
.addReg(Cond[1].getReg(), 0, SubIdx);
}
static unsigned getCRBitValue(unsigned CRBit) {
unsigned Ret = 4;
if (CRBit == PPC::CR0LT || CRBit == PPC::CR1LT ||
CRBit == PPC::CR2LT || CRBit == PPC::CR3LT ||
CRBit == PPC::CR4LT || CRBit == PPC::CR5LT ||
CRBit == PPC::CR6LT || CRBit == PPC::CR7LT)
Ret = 3;
if (CRBit == PPC::CR0GT || CRBit == PPC::CR1GT ||
CRBit == PPC::CR2GT || CRBit == PPC::CR3GT ||
CRBit == PPC::CR4GT || CRBit == PPC::CR5GT ||
CRBit == PPC::CR6GT || CRBit == PPC::CR7GT)
Ret = 2;
if (CRBit == PPC::CR0EQ || CRBit == PPC::CR1EQ ||
CRBit == PPC::CR2EQ || CRBit == PPC::CR3EQ ||
CRBit == PPC::CR4EQ || CRBit == PPC::CR5EQ ||
CRBit == PPC::CR6EQ || CRBit == PPC::CR7EQ)
Ret = 1;
if (CRBit == PPC::CR0UN || CRBit == PPC::CR1UN ||
CRBit == PPC::CR2UN || CRBit == PPC::CR3UN ||
CRBit == PPC::CR4UN || CRBit == PPC::CR5UN ||
CRBit == PPC::CR6UN || CRBit == PPC::CR7UN)
Ret = 0;
assert(Ret != 4 && "Invalid CR bit register");
return Ret;
}
void PPCInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
MachineBasicBlock::iterator I,
const DebugLoc &DL, MCRegister DestReg,
MCRegister SrcReg, bool KillSrc) const {
// We can end up with self copies and similar things as a result of VSX copy
// legalization. Promote them here.
const TargetRegisterInfo *TRI = &getRegisterInfo();
if (PPC::F8RCRegClass.contains(DestReg) &&
PPC::VSRCRegClass.contains(SrcReg)) {
MCRegister SuperReg =
TRI->getMatchingSuperReg(DestReg, PPC::sub_64, &PPC::VSRCRegClass);
if (VSXSelfCopyCrash && SrcReg == SuperReg)
llvm_unreachable("nop VSX copy");
DestReg = SuperReg;
} else if (PPC::F8RCRegClass.contains(SrcReg) &&
PPC::VSRCRegClass.contains(DestReg)) {
MCRegister SuperReg =
TRI->getMatchingSuperReg(SrcReg, PPC::sub_64, &PPC::VSRCRegClass);
if (VSXSelfCopyCrash && DestReg == SuperReg)
llvm_unreachable("nop VSX copy");
SrcReg = SuperReg;
}
// Different class register copy
if (PPC::CRBITRCRegClass.contains(SrcReg) &&
PPC::GPRCRegClass.contains(DestReg)) {
MCRegister CRReg = getCRFromCRBit(SrcReg);
BuildMI(MBB, I, DL, get(PPC::MFOCRF), DestReg).addReg(CRReg);
getKillRegState(KillSrc);
// Rotate the CR bit in the CR fields to be the least significant bit and
// then mask with 0x1 (MB = ME = 31).
BuildMI(MBB, I, DL, get(PPC::RLWINM), DestReg)
.addReg(DestReg, RegState::Kill)
.addImm(TRI->getEncodingValue(CRReg) * 4 + (4 - getCRBitValue(SrcReg)))
.addImm(31)
.addImm(31);
return;
} else if (PPC::CRRCRegClass.contains(SrcReg) &&
(PPC::G8RCRegClass.contains(DestReg) ||
PPC::GPRCRegClass.contains(DestReg))) {
bool Is64Bit = PPC::G8RCRegClass.contains(DestReg);
unsigned MvCode = Is64Bit ? PPC::MFOCRF8 : PPC::MFOCRF;
unsigned ShCode = Is64Bit ? PPC::RLWINM8 : PPC::RLWINM;
unsigned CRNum = TRI->getEncodingValue(SrcReg);
BuildMI(MBB, I, DL, get(MvCode), DestReg).addReg(SrcReg);
getKillRegState(KillSrc);
if (CRNum == 7)
return;
// Shift the CR bits to make the CR field in the lowest 4 bits of GRC.
BuildMI(MBB, I, DL, get(ShCode), DestReg)
.addReg(DestReg, RegState::Kill)
.addImm(CRNum * 4 + 4)
.addImm(28)
.addImm(31);
return;
} else if (PPC::G8RCRegClass.contains(SrcReg) &&
PPC::VSFRCRegClass.contains(DestReg)) {
assert(Subtarget.hasDirectMove() &&
"Subtarget doesn't support directmove, don't know how to copy.");
BuildMI(MBB, I, DL, get(PPC::MTVSRD), DestReg).addReg(SrcReg);
NumGPRtoVSRSpill++;
getKillRegState(KillSrc);
return;
} else if (PPC::VSFRCRegClass.contains(SrcReg) &&
PPC::G8RCRegClass.contains(DestReg)) {
assert(Subtarget.hasDirectMove() &&
"Subtarget doesn't support directmove, don't know how to copy.");
BuildMI(MBB, I, DL, get(PPC::MFVSRD), DestReg).addReg(SrcReg);
getKillRegState(KillSrc);
return;
} else if (PPC::SPERCRegClass.contains(SrcReg) &&
PPC::GPRCRegClass.contains(DestReg)) {
BuildMI(MBB, I, DL, get(PPC::EFSCFD), DestReg).addReg(SrcReg);
getKillRegState(KillSrc);
return;
} else if (PPC::GPRCRegClass.contains(SrcReg) &&
PPC::SPERCRegClass.contains(DestReg)) {
BuildMI(MBB, I, DL, get(PPC::EFDCFS), DestReg).addReg(SrcReg);
getKillRegState(KillSrc);
return;
}
unsigned Opc;
if (PPC::GPRCRegClass.contains(DestReg, SrcReg))
Opc = PPC::OR;
else if (PPC::G8RCRegClass.contains(DestReg, SrcReg))
Opc = PPC::OR8;
else if (PPC::F4RCRegClass.contains(DestReg, SrcReg))
Opc = PPC::FMR;
else if (PPC::CRRCRegClass.contains(DestReg, SrcReg))
Opc = PPC::MCRF;
else if (PPC::VRRCRegClass.contains(DestReg, SrcReg))
Opc = PPC::VOR;
else if (PPC::VSRCRegClass.contains(DestReg, SrcReg))
// There are two different ways this can be done:
// 1. xxlor : This has lower latency (on the P7), 2 cycles, but can only
// issue in VSU pipeline 0.
// 2. xmovdp/xmovsp: This has higher latency (on the P7), 6 cycles, but
// can go to either pipeline.
// We'll always use xxlor here, because in practically all cases where
// copies are generated, they are close enough to some use that the
// lower-latency form is preferable.
Opc = PPC::XXLOR;
else if (PPC::VSFRCRegClass.contains(DestReg, SrcReg) ||
PPC::VSSRCRegClass.contains(DestReg, SrcReg))
Opc = (Subtarget.hasP9Vector()) ? PPC::XSCPSGNDP : PPC::XXLORf;
else if (Subtarget.pairedVectorMemops() &&
PPC::VSRpRCRegClass.contains(DestReg, SrcReg)) {
if (SrcReg > PPC::VSRp15)
SrcReg = PPC::V0 + (SrcReg - PPC::VSRp16) * 2;
else
SrcReg = PPC::VSL0 + (SrcReg - PPC::VSRp0) * 2;
if (DestReg > PPC::VSRp15)
DestReg = PPC::V0 + (DestReg - PPC::VSRp16) * 2;
else
DestReg = PPC::VSL0 + (DestReg - PPC::VSRp0) * 2;
BuildMI(MBB, I, DL, get(PPC::XXLOR), DestReg).
addReg(SrcReg).addReg(SrcReg, getKillRegState(KillSrc));
BuildMI(MBB, I, DL, get(PPC::XXLOR), DestReg + 1).
addReg(SrcReg + 1).addReg(SrcReg + 1, getKillRegState(KillSrc));
return;
}
else if (PPC::CRBITRCRegClass.contains(DestReg, SrcReg))
Opc = PPC::CROR;
else if (PPC::SPERCRegClass.contains(DestReg, SrcReg))
Opc = PPC::EVOR;
else if ((PPC::ACCRCRegClass.contains(DestReg) ||
PPC::UACCRCRegClass.contains(DestReg)) &&
(PPC::ACCRCRegClass.contains(SrcReg) ||
PPC::UACCRCRegClass.contains(SrcReg))) {
// If primed, de-prime the source register, copy the individual registers
// and prime the destination if needed. The vector subregisters are
// vs[(u)acc * 4] - vs[(u)acc * 4 + 3]. If the copy is not a kill and the
// source is primed, we need to re-prime it after the copy as well.
PPCRegisterInfo::emitAccCopyInfo(MBB, DestReg, SrcReg);
bool DestPrimed = PPC::ACCRCRegClass.contains(DestReg);
bool SrcPrimed = PPC::ACCRCRegClass.contains(SrcReg);
MCRegister VSLSrcReg =
PPC::VSL0 + (SrcReg - (SrcPrimed ? PPC::ACC0 : PPC::UACC0)) * 4;
MCRegister VSLDestReg =
PPC::VSL0 + (DestReg - (DestPrimed ? PPC::ACC0 : PPC::UACC0)) * 4;
if (SrcPrimed)
BuildMI(MBB, I, DL, get(PPC::XXMFACC), SrcReg).addReg(SrcReg);
for (unsigned Idx = 0; Idx < 4; Idx++)
BuildMI(MBB, I, DL, get(PPC::XXLOR), VSLDestReg + Idx)
.addReg(VSLSrcReg + Idx)
.addReg(VSLSrcReg + Idx, getKillRegState(KillSrc));
if (DestPrimed)
BuildMI(MBB, I, DL, get(PPC::XXMTACC), DestReg).addReg(DestReg);
if (SrcPrimed && !KillSrc)
BuildMI(MBB, I, DL, get(PPC::XXMTACC), SrcReg).addReg(SrcReg);
return;
} else if (PPC::G8pRCRegClass.contains(DestReg) &&
PPC::G8pRCRegClass.contains(SrcReg)) {
// TODO: Handle G8RC to G8pRC (and vice versa) copy.
unsigned DestRegIdx = DestReg - PPC::G8p0;
MCRegister DestRegSub0 = PPC::X0 + 2 * DestRegIdx;
MCRegister DestRegSub1 = PPC::X0 + 2 * DestRegIdx + 1;
unsigned SrcRegIdx = SrcReg - PPC::G8p0;
MCRegister SrcRegSub0 = PPC::X0 + 2 * SrcRegIdx;
MCRegister SrcRegSub1 = PPC::X0 + 2 * SrcRegIdx + 1;
BuildMI(MBB, I, DL, get(PPC::OR8), DestRegSub0)
.addReg(SrcRegSub0)
.addReg(SrcRegSub0, getKillRegState(KillSrc));
BuildMI(MBB, I, DL, get(PPC::OR8), DestRegSub1)
.addReg(SrcRegSub1)
.addReg(SrcRegSub1, getKillRegState(KillSrc));
return;
} else
llvm_unreachable("Impossible reg-to-reg copy");
const MCInstrDesc &MCID = get(Opc);
if (MCID.getNumOperands() == 3)
BuildMI(MBB, I, DL, MCID, DestReg)
.addReg(SrcReg).addReg(SrcReg, getKillRegState(KillSrc));
else
BuildMI(MBB, I, DL, MCID, DestReg).addReg(SrcReg, getKillRegState(KillSrc));
}
unsigned PPCInstrInfo::getSpillIndex(const TargetRegisterClass *RC) const {
int OpcodeIndex = 0;
if (PPC::GPRCRegClass.hasSubClassEq(RC) ||
PPC::GPRC_NOR0RegClass.hasSubClassEq(RC)) {
OpcodeIndex = SOK_Int4Spill;
} else if (PPC::G8RCRegClass.hasSubClassEq(RC) ||
PPC::G8RC_NOX0RegClass.hasSubClassEq(RC)) {
OpcodeIndex = SOK_Int8Spill;
} else if (PPC::F8RCRegClass.hasSubClassEq(RC)) {
OpcodeIndex = SOK_Float8Spill;
} else if (PPC::F4RCRegClass.hasSubClassEq(RC)) {
OpcodeIndex = SOK_Float4Spill;
} else if (PPC::SPERCRegClass.hasSubClassEq(RC)) {
OpcodeIndex = SOK_SPESpill;
} else if (PPC::CRRCRegClass.hasSubClassEq(RC)) {
OpcodeIndex = SOK_CRSpill;
} else if (PPC::CRBITRCRegClass.hasSubClassEq(RC)) {
OpcodeIndex = SOK_CRBitSpill;
} else if (PPC::VRRCRegClass.hasSubClassEq(RC)) {
OpcodeIndex = SOK_VRVectorSpill;
} else if (PPC::VSRCRegClass.hasSubClassEq(RC)) {
OpcodeIndex = SOK_VSXVectorSpill;
} else if (PPC::VSFRCRegClass.hasSubClassEq(RC)) {
OpcodeIndex = SOK_VectorFloat8Spill;
} else if (PPC::VSSRCRegClass.hasSubClassEq(RC)) {
OpcodeIndex = SOK_VectorFloat4Spill;
} else if (PPC::SPILLTOVSRRCRegClass.hasSubClassEq(RC)) {
OpcodeIndex = SOK_SpillToVSR;
} else if (PPC::ACCRCRegClass.hasSubClassEq(RC)) {
assert(Subtarget.pairedVectorMemops() &&
"Register unexpected when paired memops are disabled.");
OpcodeIndex = SOK_AccumulatorSpill;
} else if (PPC::UACCRCRegClass.hasSubClassEq(RC)) {
assert(Subtarget.pairedVectorMemops() &&
"Register unexpected when paired memops are disabled.");
OpcodeIndex = SOK_UAccumulatorSpill;
} else if (PPC::VSRpRCRegClass.hasSubClassEq(RC)) {
assert(Subtarget.pairedVectorMemops() &&
"Register unexpected when paired memops are disabled.");
OpcodeIndex = SOK_PairedVecSpill;
} else if (PPC::G8pRCRegClass.hasSubClassEq(RC)) {
OpcodeIndex = SOK_PairedG8Spill;
} else {
llvm_unreachable("Unknown regclass!");
}
return OpcodeIndex;
}
unsigned
PPCInstrInfo::getStoreOpcodeForSpill(const TargetRegisterClass *RC) const {
const unsigned *OpcodesForSpill = getStoreOpcodesForSpillArray();
return OpcodesForSpill[getSpillIndex(RC)];
}
unsigned
PPCInstrInfo::getLoadOpcodeForSpill(const TargetRegisterClass *RC) const {
const unsigned *OpcodesForSpill = getLoadOpcodesForSpillArray();
return OpcodesForSpill[getSpillIndex(RC)];
}
void PPCInstrInfo::StoreRegToStackSlot(
MachineFunction &MF, unsigned SrcReg, bool isKill, int FrameIdx,
const TargetRegisterClass *RC,
SmallVectorImpl<MachineInstr *> &NewMIs) const {
unsigned Opcode = getStoreOpcodeForSpill(RC);
DebugLoc DL;
PPCFunctionInfo *FuncInfo = MF.getInfo<PPCFunctionInfo>();
FuncInfo->setHasSpills();
NewMIs.push_back(addFrameReference(
BuildMI(MF, DL, get(Opcode)).addReg(SrcReg, getKillRegState(isKill)),
FrameIdx));
if (PPC::CRRCRegClass.hasSubClassEq(RC) ||
PPC::CRBITRCRegClass.hasSubClassEq(RC))
FuncInfo->setSpillsCR();
if (isXFormMemOp(Opcode))
FuncInfo->setHasNonRISpills();
}
void PPCInstrInfo::storeRegToStackSlotNoUpd(
MachineBasicBlock &MBB, MachineBasicBlock::iterator MI, unsigned SrcReg,
bool isKill, int FrameIdx, const TargetRegisterClass *RC,
const TargetRegisterInfo *TRI) const {
MachineFunction &MF = *MBB.getParent();
SmallVector<MachineInstr *, 4> NewMIs;
StoreRegToStackSlot(MF, SrcReg, isKill, FrameIdx, RC, NewMIs);
for (unsigned i = 0, e = NewMIs.size(); i != e; ++i)
MBB.insert(MI, NewMIs[i]);
const MachineFrameInfo &MFI = MF.getFrameInfo();
MachineMemOperand *MMO = MF.getMachineMemOperand(
MachinePointerInfo::getFixedStack(MF, FrameIdx),
MachineMemOperand::MOStore, MFI.getObjectSize(FrameIdx),
MFI.getObjectAlign(FrameIdx));
NewMIs.back()->addMemOperand(MF, MMO);
}
void PPCInstrInfo::storeRegToStackSlot(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI,
Register SrcReg, bool isKill,
int FrameIdx,
const TargetRegisterClass *RC,
const TargetRegisterInfo *TRI) const {
// We need to avoid a situation in which the value from a VRRC register is
// spilled using an Altivec instruction and reloaded into a VSRC register
// using a VSX instruction. The issue with this is that the VSX
// load/store instructions swap the doublewords in the vector and the Altivec
// ones don't. The register classes on the spill/reload may be different if
// the register is defined using an Altivec instruction and is then used by a
// VSX instruction.
RC = updatedRC(RC);
storeRegToStackSlotNoUpd(MBB, MI, SrcReg, isKill, FrameIdx, RC, TRI);
}
void PPCInstrInfo::LoadRegFromStackSlot(MachineFunction &MF, const DebugLoc &DL,
unsigned DestReg, int FrameIdx,
const TargetRegisterClass *RC,
SmallVectorImpl<MachineInstr *> &NewMIs)
const {
unsigned Opcode = getLoadOpcodeForSpill(RC);
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(Opcode), DestReg),
FrameIdx));
PPCFunctionInfo *FuncInfo = MF.getInfo<PPCFunctionInfo>();
if (PPC::CRRCRegClass.hasSubClassEq(RC) ||
PPC::CRBITRCRegClass.hasSubClassEq(RC))
FuncInfo->setSpillsCR();
if (isXFormMemOp(Opcode))
FuncInfo->setHasNonRISpills();
}
void PPCInstrInfo::loadRegFromStackSlotNoUpd(
MachineBasicBlock &MBB, MachineBasicBlock::iterator MI, unsigned DestReg,
int FrameIdx, const TargetRegisterClass *RC,
const TargetRegisterInfo *TRI) const {
MachineFunction &MF = *MBB.getParent();
SmallVector<MachineInstr*, 4> NewMIs;
DebugLoc DL;
if (MI != MBB.end()) DL = MI->getDebugLoc();
PPCFunctionInfo *FuncInfo = MF.getInfo<PPCFunctionInfo>();
FuncInfo->setHasSpills();
LoadRegFromStackSlot(MF, DL, DestReg, FrameIdx, RC, NewMIs);
for (unsigned i = 0, e = NewMIs.size(); i != e; ++i)
MBB.insert(MI, NewMIs[i]);
const MachineFrameInfo &MFI = MF.getFrameInfo();
MachineMemOperand *MMO = MF.getMachineMemOperand(
MachinePointerInfo::getFixedStack(MF, FrameIdx),
MachineMemOperand::MOLoad, MFI.getObjectSize(FrameIdx),
MFI.getObjectAlign(FrameIdx));
NewMIs.back()->addMemOperand(MF, MMO);
}
void PPCInstrInfo::loadRegFromStackSlot(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI,
Register DestReg, int FrameIdx,
const TargetRegisterClass *RC,
const TargetRegisterInfo *TRI) const {
// We need to avoid a situation in which the value from a VRRC register is
// spilled using an Altivec instruction and reloaded into a VSRC register
// using a VSX instruction. The issue with this is that the VSX
// load/store instructions swap the doublewords in the vector and the Altivec
// ones don't. The register classes on the spill/reload may be different if
// the register is defined using an Altivec instruction and is then used by a
// VSX instruction.
RC = updatedRC(RC);
loadRegFromStackSlotNoUpd(MBB, MI, DestReg, FrameIdx, RC, TRI);
}
bool PPCInstrInfo::
reverseBranchCondition(SmallVectorImpl<MachineOperand> &Cond) const {
assert(Cond.size() == 2 && "Invalid PPC branch opcode!");
if (Cond[1].getReg() == PPC::CTR8 || Cond[1].getReg() == PPC::CTR)
Cond[0].setImm(Cond[0].getImm() == 0 ? 1 : 0);
else
// Leave the CR# the same, but invert the condition.
Cond[0].setImm(PPC::InvertPredicate((PPC::Predicate)Cond[0].getImm()));
return false;
}
// For some instructions, it is legal to fold ZERO into the RA register field.
// This function performs that fold by replacing the operand with PPC::ZERO,
// it does not consider whether the load immediate zero is no longer in use.
bool PPCInstrInfo::onlyFoldImmediate(MachineInstr &UseMI, MachineInstr &DefMI,
Register Reg) const {
// A zero immediate should always be loaded with a single li.
unsigned DefOpc = DefMI.getOpcode();
if (DefOpc != PPC::LI && DefOpc != PPC::LI8)
return false;
if (!DefMI.getOperand(1).isImm())
return false;
if (DefMI.getOperand(1).getImm() != 0)
return false;
// Note that we cannot here invert the arguments of an isel in order to fold
// a ZERO into what is presented as the second argument. All we have here
// is the condition bit, and that might come from a CR-logical bit operation.
const MCInstrDesc &UseMCID = UseMI.getDesc();
// Only fold into real machine instructions.
if (UseMCID.isPseudo())
return false;
// We need to find which of the User's operands is to be folded, that will be
// the operand that matches the given register ID.
unsigned UseIdx;
for (UseIdx = 0; UseIdx < UseMI.getNumOperands(); ++UseIdx)
if (UseMI.getOperand(UseIdx).isReg() &&
UseMI.getOperand(UseIdx).getReg() == Reg)
break;
assert(UseIdx < UseMI.getNumOperands() && "Cannot find Reg in UseMI");
assert(UseIdx < UseMCID.getNumOperands() && "No operand description for Reg");
const MCOperandInfo *UseInfo = &UseMCID.OpInfo[UseIdx];
// We can fold the zero if this register requires a GPRC_NOR0/G8RC_NOX0
// register (which might also be specified as a pointer class kind).
if (UseInfo->isLookupPtrRegClass()) {
if (UseInfo->RegClass /* Kind */ != 1)
return false;
} else {
if (UseInfo->RegClass != PPC::GPRC_NOR0RegClassID &&
UseInfo->RegClass != PPC::G8RC_NOX0RegClassID)
return false;
}
// Make sure this is not tied to an output register (or otherwise
// constrained). This is true for ST?UX registers, for example, which
// are tied to their output registers.
if (UseInfo->Constraints != 0)
return false;
MCRegister ZeroReg;
if (UseInfo->isLookupPtrRegClass()) {
bool isPPC64 = Subtarget.isPPC64();
ZeroReg = isPPC64 ? PPC::ZERO8 : PPC::ZERO;
} else {
ZeroReg = UseInfo->RegClass == PPC::G8RC_NOX0RegClassID ?
PPC::ZERO8 : PPC::ZERO;
}
UseMI.getOperand(UseIdx).setReg(ZeroReg);
return true;
}
// Folds zero into instructions which have a load immediate zero as an operand
// but also recognize zero as immediate zero. If the definition of the load
// has no more users it is deleted.
bool PPCInstrInfo::FoldImmediate(MachineInstr &UseMI, MachineInstr &DefMI,
Register Reg, MachineRegisterInfo *MRI) const {
bool Changed = onlyFoldImmediate(UseMI, DefMI, Reg);
if (MRI->use_nodbg_empty(Reg))
DefMI.eraseFromParent();
return Changed;
}
static bool MBBDefinesCTR(MachineBasicBlock &MBB) {
for (MachineInstr &MI : MBB)
if (MI.definesRegister(PPC::CTR) || MI.definesRegister(PPC::CTR8))
return true;
return false;
}
// We should make sure that, if we're going to predicate both sides of a
// condition (a diamond), that both sides don't define the counter register. We
// can predicate counter-decrement-based branches, but while that predicates
// the branching, it does not predicate the counter decrement. If we tried to
// merge the triangle into one predicated block, we'd decrement the counter
// twice.
bool PPCInstrInfo::isProfitableToIfCvt(MachineBasicBlock &TMBB,
unsigned NumT, unsigned ExtraT,
MachineBasicBlock &FMBB,
unsigned NumF, unsigned ExtraF,
BranchProbability Probability) const {
return !(MBBDefinesCTR(TMBB) && MBBDefinesCTR(FMBB));
}
bool PPCInstrInfo::isPredicated(const MachineInstr &MI) const {
// The predicated branches are identified by their type, not really by the
// explicit presence of a predicate. Furthermore, some of them can be
// predicated more than once. Because if conversion won't try to predicate
// any instruction which already claims to be predicated (by returning true
// here), always return false. In doing so, we let isPredicable() be the
// final word on whether not the instruction can be (further) predicated.
return false;
}
bool PPCInstrInfo::isSchedulingBoundary(const MachineInstr &MI,
const MachineBasicBlock *MBB,
const MachineFunction &MF) const {
// Set MFFS and MTFSF as scheduling boundary to avoid unexpected code motion
// across them, since some FP operations may change content of FPSCR.
// TODO: Model FPSCR in PPC instruction definitions and remove the workaround
if (MI.getOpcode() == PPC::MFFS || MI.getOpcode() == PPC::MTFSF)
return true;
return TargetInstrInfo::isSchedulingBoundary(MI, MBB, MF);
}
bool PPCInstrInfo::PredicateInstruction(MachineInstr &MI,
ArrayRef<MachineOperand> Pred) const {
unsigned OpC = MI.getOpcode();
if (OpC == PPC::BLR || OpC == PPC::BLR8) {
if (Pred[1].getReg() == PPC::CTR8 || Pred[1].getReg() == PPC::CTR) {
bool isPPC64 = Subtarget.isPPC64();
MI.setDesc(get(Pred[0].getImm() ? (isPPC64 ? PPC::BDNZLR8 : PPC::BDNZLR)
: (isPPC64 ? PPC::BDZLR8 : PPC::BDZLR)));
// Need add Def and Use for CTR implicit operand.
MachineInstrBuilder(*MI.getParent()->getParent(), MI)
.addReg(Pred[1].getReg(), RegState::Implicit)
.addReg(Pred[1].getReg(), RegState::ImplicitDefine);
} else if (Pred[0].getImm() == PPC::PRED_BIT_SET) {
MI.setDesc(get(PPC::BCLR));
MachineInstrBuilder(*MI.getParent()->getParent(), MI).add(Pred[1]);
} else if (Pred[0].getImm() == PPC::PRED_BIT_UNSET) {
MI.setDesc(get(PPC::BCLRn));
MachineInstrBuilder(*MI.getParent()->getParent(), MI).add(Pred[1]);
} else {
MI.setDesc(get(PPC::BCCLR));
MachineInstrBuilder(*MI.getParent()->getParent(), MI)
.addImm(Pred[0].getImm())
.add(Pred[1]);
}
return true;
} else if (OpC == PPC::B) {
if (Pred[1].getReg() == PPC::CTR8 || Pred[1].getReg() == PPC::CTR) {
bool isPPC64 = Subtarget.isPPC64();
MI.setDesc(get(Pred[0].getImm() ? (isPPC64 ? PPC::BDNZ8 : PPC::BDNZ)
: (isPPC64 ? PPC::BDZ8 : PPC::BDZ)));
// Need add Def and Use for CTR implicit operand.
MachineInstrBuilder(*MI.getParent()->getParent(), MI)
.addReg(Pred[1].getReg(), RegState::Implicit)
.addReg(Pred[1].getReg(), RegState::ImplicitDefine);
} else if (Pred[0].getImm() == PPC::PRED_BIT_SET) {
MachineBasicBlock *MBB = MI.getOperand(0).getMBB();
MI.RemoveOperand(0);
MI.setDesc(get(PPC::BC));
MachineInstrBuilder(*MI.getParent()->getParent(), MI)
.add(Pred[1])
.addMBB(MBB);
} else if (Pred[0].getImm() == PPC::PRED_BIT_UNSET) {
MachineBasicBlock *MBB = MI.getOperand(0).getMBB();
MI.RemoveOperand(0);
MI.setDesc(get(PPC::BCn));
MachineInstrBuilder(*MI.getParent()->getParent(), MI)
.add(Pred[1])
.addMBB(MBB);
} else {
MachineBasicBlock *MBB = MI.getOperand(0).getMBB();
MI.RemoveOperand(0);
MI.setDesc(get(PPC::BCC));
MachineInstrBuilder(*MI.getParent()->getParent(), MI)
.addImm(Pred[0].getImm())
.add(Pred[1])
.addMBB(MBB);
}
return true;
} else if (OpC == PPC::BCTR || OpC == PPC::BCTR8 || OpC == PPC::BCTRL ||
OpC == PPC::BCTRL8 || OpC == PPC::BCTRL_RM ||
OpC == PPC::BCTRL8_RM) {
if (Pred[1].getReg() == PPC::CTR8 || Pred[1].getReg() == PPC::CTR)
llvm_unreachable("Cannot predicate bctr[l] on the ctr register");
bool setLR = OpC == PPC::BCTRL || OpC == PPC::BCTRL8 ||
OpC == PPC::BCTRL_RM || OpC == PPC::BCTRL8_RM;
bool isPPC64 = Subtarget.isPPC64();
if (Pred[0].getImm() == PPC::PRED_BIT_SET) {
MI.setDesc(get(isPPC64 ? (setLR ? PPC::BCCTRL8 : PPC::BCCTR8)
: (setLR ? PPC::BCCTRL : PPC::BCCTR)));
MachineInstrBuilder(*MI.getParent()->getParent(), MI).add(Pred[1]);
} else if (Pred[0].getImm() == PPC::PRED_BIT_UNSET) {
MI.setDesc(get(isPPC64 ? (setLR ? PPC::BCCTRL8n : PPC::BCCTR8n)
: (setLR ? PPC::BCCTRLn : PPC::BCCTRn)));
MachineInstrBuilder(*MI.getParent()->getParent(), MI).add(Pred[1]);
} else {
MI.setDesc(get(isPPC64 ? (setLR ? PPC::BCCCTRL8 : PPC::BCCCTR8)
: (setLR ? PPC::BCCCTRL : PPC::BCCCTR)));
MachineInstrBuilder(*MI.getParent()->getParent(), MI)
.addImm(Pred[0].getImm())
.add(Pred[1]);
}
// Need add Def and Use for LR implicit operand.
if (setLR)
MachineInstrBuilder(*MI.getParent()->getParent(), MI)
.addReg(isPPC64 ? PPC::LR8 : PPC::LR, RegState::Implicit)
.addReg(isPPC64 ? PPC::LR8 : PPC::LR, RegState::ImplicitDefine);
if (OpC == PPC::BCTRL_RM || OpC == PPC::BCTRL8_RM)
MachineInstrBuilder(*MI.getParent()->getParent(), MI)
.addReg(PPC::RM, RegState::ImplicitDefine);
return true;
}
return false;
}
bool PPCInstrInfo::SubsumesPredicate(ArrayRef<MachineOperand> Pred1,
ArrayRef<MachineOperand> Pred2) const {
assert(Pred1.size() == 2 && "Invalid PPC first predicate");
assert(Pred2.size() == 2 && "Invalid PPC second predicate");
if (Pred1[1].getReg() == PPC::CTR8 || Pred1[1].getReg() == PPC::CTR)
return false;
if (Pred2[1].getReg() == PPC::CTR8 || Pred2[1].getReg() == PPC::CTR)
return false;
// P1 can only subsume P2 if they test the same condition register.
if (Pred1[1].getReg() != Pred2[1].getReg())
return false;
PPC::Predicate P1 = (PPC::Predicate) Pred1[0].getImm();
PPC::Predicate P2 = (PPC::Predicate) Pred2[0].getImm();
if (P1 == P2)
return true;
// Does P1 subsume P2, e.g. GE subsumes GT.
if (P1 == PPC::PRED_LE &&
(P2 == PPC::PRED_LT || P2 == PPC::PRED_EQ))
return true;
if (P1 == PPC::PRED_GE &&
(P2 == PPC::PRED_GT || P2 == PPC::PRED_EQ))
return true;
return false;
}
bool PPCInstrInfo::ClobbersPredicate(MachineInstr &MI,
std::vector<MachineOperand> &Pred,
bool SkipDead) const {
// Note: At the present time, the contents of Pred from this function is
// unused by IfConversion. This implementation follows ARM by pushing the
// CR-defining operand. Because the 'DZ' and 'DNZ' count as types of
// predicate, instructions defining CTR or CTR8 are also included as
// predicate-defining instructions.
const TargetRegisterClass *RCs[] =
{ &PPC::CRRCRegClass, &PPC::CRBITRCRegClass,
&PPC::CTRRCRegClass, &PPC::CTRRC8RegClass };
bool Found = false;
for (const MachineOperand &MO : MI.operands()) {
for (unsigned c = 0; c < array_lengthof(RCs) && !Found; ++c) {
const TargetRegisterClass *RC = RCs[c];
if (MO.isReg()) {
if (MO.isDef() && RC->contains(MO.getReg())) {
Pred.push_back(MO);
Found = true;
}
} else if (MO.isRegMask()) {
for (TargetRegisterClass::iterator I = RC->begin(),
IE = RC->end(); I != IE; ++I)
if (MO.clobbersPhysReg(*I)) {
Pred.push_back(MO);
Found = true;
}
}
}
}
return Found;
}
bool PPCInstrInfo::analyzeCompare(const MachineInstr &MI, Register &SrcReg,
Register &SrcReg2, int64_t &Mask,
int64_t &Value) const {
unsigned Opc = MI.getOpcode();
switch (Opc) {
default: return false;
case PPC::CMPWI:
case PPC::CMPLWI:
case PPC::CMPDI:
case PPC::CMPLDI:
SrcReg = MI.getOperand(1).getReg();
SrcReg2 = 0;
Value = MI.getOperand(2).getImm();
Mask = 0xFFFF;
return true;
case PPC::CMPW:
case PPC::CMPLW:
case PPC::CMPD:
case PPC::CMPLD:
case PPC::FCMPUS:
case PPC::FCMPUD:
SrcReg = MI.getOperand(1).getReg();
SrcReg2 = MI.getOperand(2).getReg();
Value = 0;
Mask = 0;
return true;
}
}
bool PPCInstrInfo::optimizeCompareInstr(MachineInstr &CmpInstr, Register SrcReg,
Register SrcReg2, int64_t Mask,
int64_t Value,
const MachineRegisterInfo *MRI) const {
if (DisableCmpOpt)
return false;
int OpC = CmpInstr.getOpcode();
Register CRReg = CmpInstr.getOperand(0).getReg();
// FP record forms set CR1 based on the exception status bits, not a
// comparison with zero.
if (OpC == PPC::FCMPUS || OpC == PPC::FCMPUD)
return false;
const TargetRegisterInfo *TRI = &getRegisterInfo();
// The record forms set the condition register based on a signed comparison
// with zero (so says the ISA manual). This is not as straightforward as it
// seems, however, because this is always a 64-bit comparison on PPC64, even
// for instructions that are 32-bit in nature (like slw for example).
// So, on PPC32, for unsigned comparisons, we can use the record forms only
// for equality checks (as those don't depend on the sign). On PPC64,
// we are restricted to equality for unsigned 64-bit comparisons and for
// signed 32-bit comparisons the applicability is more restricted.
bool isPPC64 = Subtarget.isPPC64();
bool is32BitSignedCompare = OpC == PPC::CMPWI || OpC == PPC::CMPW;
bool is32BitUnsignedCompare = OpC == PPC::CMPLWI || OpC == PPC::CMPLW;
bool is64BitUnsignedCompare = OpC == PPC::CMPLDI || OpC == PPC::CMPLD;
// Look through copies unless that gets us to a physical register.
Register ActualSrc = TRI->lookThruCopyLike(SrcReg, MRI);
if (ActualSrc.isVirtual())
SrcReg = ActualSrc;
// Get the unique definition of SrcReg.
MachineInstr *MI = MRI->getUniqueVRegDef(SrcReg);
if (!MI) return false;
bool equalityOnly = false;
bool noSub = false;
if (isPPC64) {
if (is32BitSignedCompare) {
// We can perform this optimization only if MI is sign-extending.
if (isSignExtended(*MI))
noSub = true;
else
return false;
} else if (is32BitUnsignedCompare) {
// We can perform this optimization, equality only, if MI is
// zero-extending.
if (isZeroExtended(*MI)) {
noSub = true;
equalityOnly = true;
} else
return false;
} else
equalityOnly = is64BitUnsignedCompare;
} else
equalityOnly = is32BitUnsignedCompare;
if (equalityOnly) {
// We need to check the uses of the condition register in order to reject
// non-equality comparisons.
for (MachineRegisterInfo::use_instr_iterator
I = MRI->use_instr_begin(CRReg), IE = MRI->use_instr_end();
I != IE; ++I) {
MachineInstr *UseMI = &*I;
if (UseMI->getOpcode() == PPC::BCC) {
PPC::Predicate Pred = (PPC::Predicate)UseMI->getOperand(0).getImm();
unsigned PredCond = PPC::getPredicateCondition(Pred);
// We ignore hint bits when checking for non-equality comparisons.
if (PredCond != PPC::PRED_EQ && PredCond != PPC::PRED_NE)
return false;
} else if (UseMI->getOpcode() == PPC::ISEL ||
UseMI->getOpcode() == PPC::ISEL8) {
unsigned SubIdx = UseMI->getOperand(3).getSubReg();
if (SubIdx != PPC::sub_eq)
return false;
} else
return false;
}
}
MachineBasicBlock::iterator I = CmpInstr;
// Scan forward to find the first use of the compare.
for (MachineBasicBlock::iterator EL = CmpInstr.getParent()->end(); I != EL;
++I) {
bool FoundUse = false;
for (MachineRegisterInfo::use_instr_iterator
J = MRI->use_instr_begin(CRReg), JE = MRI->use_instr_end();
J != JE; ++J)
if (&*J == &*I) {
FoundUse = true;
break;
}
if (FoundUse)
break;
}
SmallVector<std::pair<MachineOperand*, PPC::Predicate>, 4> PredsToUpdate;
SmallVector<std::pair<MachineOperand*, unsigned>, 4> SubRegsToUpdate;
// There are two possible candidates which can be changed to set CR[01].
// One is MI, the other is a SUB instruction.
// For CMPrr(r1,r2), we are looking for SUB(r1,r2) or SUB(r2,r1).
MachineInstr *Sub = nullptr;
if (SrcReg2 != 0)
// MI is not a candidate for CMPrr.
MI = nullptr;
// FIXME: Conservatively refuse to convert an instruction which isn't in the
// same BB as the comparison. This is to allow the check below to avoid calls
// (and other explicit clobbers); instead we should really check for these
// more explicitly (in at least a few predecessors).
else if (MI->getParent() != CmpInstr.getParent())
return false;
else if (Value != 0) {
// The record-form instructions set CR bit based on signed comparison
// against 0. We try to convert a compare against 1 or -1 into a compare
// against 0 to exploit record-form instructions. For example, we change
// the condition "greater than -1" into "greater than or equal to 0"
// and "less than 1" into "less than or equal to 0".