#include "llvm/CodeGen/PseudoSourceValue.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/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
static cl::opt<bool> DisableCmpOpt("disable-ppc-cmp-opt",
cl::desc("Disable compare instruction optimization"), cl::Hidden);
-static cl::opt<bool> DisableVSXFMAMutate("disable-ppc-vsx-fma-mutation",
-cl::desc("Disable VSX FMA instruction mutation"), 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"));
+
// Pin the vtable to this file.
void PPCInstrInfo::anchor() {}
PPCInstrInfo::PPCInstrInfo(PPCSubtarget &STI)
: PPCGenInstrInfo(PPC::ADJCALLSTACKDOWN, PPC::ADJCALLSTACKUP),
- Subtarget(STI), RI(STI) {}
+ Subtarget(STI), RI(STI.getTargetMachine()) {}
/// CreateTargetHazardRecognizer - Return the hazard recognizer to use for
/// this target when scheduling the DAG.
-ScheduleHazardRecognizer *PPCInstrInfo::CreateTargetHazardRecognizer(
- const TargetMachine *TM,
- const ScheduleDAG *DAG) const {
- unsigned Directive = TM->getSubtarget<PPCSubtarget>().getDarwinDirective();
+ScheduleHazardRecognizer *
+PPCInstrInfo::CreateTargetHazardRecognizer(const TargetSubtargetInfo *STI,
+ const ScheduleDAG *DAG) const {
+ unsigned Directive =
+ static_cast<const PPCSubtarget *>(STI)->getDarwinDirective();
if (Directive == PPC::DIR_440 || Directive == PPC::DIR_A2 ||
Directive == PPC::DIR_E500mc || Directive == PPC::DIR_E5500) {
- const InstrItineraryData *II = TM->getInstrItineraryData();
+ const InstrItineraryData *II =
+ static_cast<const PPCSubtarget *>(STI)->getInstrItineraryData();
return new ScoreboardHazardRecognizer(II, DAG);
}
- return TargetInstrInfo::CreateTargetHazardRecognizer(TM, 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 {
+ScheduleHazardRecognizer *
+PPCInstrInfo::CreateTargetPostRAHazardRecognizer(const InstrItineraryData *II,
+ const ScheduleDAG *DAG) const {
unsigned Directive =
- DAG->
TM.getSubtarget<PPCSubtarget>().getDarwinDirective();
+ DAG->
MF.getSubtarget<PPCSubtarget>().getDarwinDirective();
- if (Directive == PPC::DIR_PWR7)
+ if (Directive == PPC::DIR_PWR7 || Directive == PPC::DIR_PWR8)
return new PPCDispatchGroupSBHazardRecognizer(II, DAG);
// Most subtargets use a PPC970 recognizer.
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,
int Latency = PPCGenInstrInfo::getOperandLatency(ItinData, DefMI, DefIdx,
UseMI, UseIdx);
+ if (!DefMI->getParent())
+ return Latency;
+
const MachineOperand &DefMO = DefMI->getOperand(DefIdx);
unsigned Reg = DefMO.getReg();
- const TargetRegisterInfo *TRI = &getRegisterInfo();
bool IsRegCR;
- if (TRI->isVirtualRegister(Reg)) {
+ if (TargetRegisterInfo::isVirtualRegister(Reg)) {
const MachineRegisterInfo *MRI =
&DefMI->getParent()->getParent()->getRegInfo();
IsRegCR = MRI->getRegClass(Reg)->hasSuperClassEq(&PPC::CRRCRegClass) ||
case PPC::DIR_PWR6:
case PPC::DIR_PWR6X:
case PPC::DIR_PWR7:
+ case PPC::DIR_PWR8:
Latency += 2;
break;
}
return Latency;
}
+static bool hasVirtualRegDefsInBasicBlock(const MachineInstr &Inst,
+ const MachineBasicBlock *MBB) {
+ const MachineOperand &Op1 = Inst.getOperand(1);
+ const MachineOperand &Op2 = Inst.getOperand(2);
+ const MachineRegisterInfo &MRI = MBB->getParent()->getRegInfo();
+
+ // We need virtual register definitions.
+ MachineInstr *MI1 = nullptr;
+ MachineInstr *MI2 = nullptr;
+ if (Op1.isReg() && TargetRegisterInfo::isVirtualRegister(Op1.getReg()))
+ MI1 = MRI.getUniqueVRegDef(Op1.getReg());
+ if (Op2.isReg() && TargetRegisterInfo::isVirtualRegister(Op2.getReg()))
+ MI2 = MRI.getUniqueVRegDef(Op2.getReg());
+
+ // And they need to be in the trace (otherwise, they won't have a depth).
+ if (MI1 && MI2 && MI1->getParent() == MBB && MI2->getParent() == MBB)
+ return true;
+
+ return false;
+}
+
+static bool hasReassocSibling(const MachineInstr &Inst, bool &Commuted) {
+ const MachineBasicBlock *MBB = Inst.getParent();
+ const MachineRegisterInfo &MRI = MBB->getParent()->getRegInfo();
+ MachineInstr *MI1 = MRI.getUniqueVRegDef(Inst.getOperand(1).getReg());
+ MachineInstr *MI2 = MRI.getUniqueVRegDef(Inst.getOperand(2).getReg());
+ unsigned AssocOpcode = Inst.getOpcode();
+
+ // If only one operand has the same opcode and it's the second source operand,
+ // the operands must be commuted.
+ Commuted = MI1->getOpcode() != AssocOpcode && MI2->getOpcode() == AssocOpcode;
+ if (Commuted)
+ std::swap(MI1, MI2);
+
+ // 1. The previous instruction must be the same type as Inst.
+ // 2. The previous instruction must have virtual register definitions for its
+ // operands in the same basic block as Inst.
+ // 3. The previous instruction's result must only be used by Inst.
+ if (MI1->getOpcode() == AssocOpcode &&
+ hasVirtualRegDefsInBasicBlock(*MI1, MBB) &&
+ MRI.hasOneNonDBGUse(MI1->getOperand(0).getReg()))
+ return true;
+
+ return false;
+}
+
+// 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 (compared to other operations, such and
+// and/or, which are also associative and commutative, but have low latencies).
+//
+// The concept is that these operations can benefit from this kind of
+// transformation:
+//
+// A = ? op ?
+// B = A op X
+// C = B op Y
+// -->
+// A = ? op ?
+// B = X op Y
+// C = A op B
+//
+// breaking the dependency between A and B, allowing them to be executed in
+// parallel (or back-to-back in a pipeline) instead of depending on each other.
+static bool isAssociativeAndCommutative(unsigned Opcode) {
+ switch (Opcode) {
+ // 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:
+ // QPX Add:
+ case PPC::QVFADD:
+ case PPC::QVFADDS:
+ case PPC::QVFADDSs:
+ // QPX Multiply:
+ case PPC::QVFMUL:
+ case PPC::QVFMULS:
+ case PPC::QVFMULSs:
+ return true;
+ default:
+ return false;
+ }
+}
+
+/// Return true if the input instruction is part of a chain of dependent ops
+/// that are suitable for reassociation, otherwise return false.
+/// If the instruction's operands must be commuted to have a previous
+/// instruction of the same type define the first source operand, Commuted will
+/// be set to true.
+static bool isReassocCandidate(const MachineInstr &Inst, bool &Commuted) {
+ // 1. The operation must be associative and commutative.
+ // 2. The instruction must have virtual register definitions for its
+ // operands in the same basic block.
+ // 3. The instruction must have a reassociable sibling.
+ if (isAssociativeAndCommutative(Inst.getOpcode()) &&
+ hasVirtualRegDefsInBasicBlock(Inst, Inst.getParent()) &&
+ hasReassocSibling(Inst, Commuted))
+ return true;
+
+ return false;
+}
+
+bool PPCInstrInfo::getMachineCombinerPatterns(MachineInstr &Root,
+ SmallVectorImpl<MachineCombinerPattern::MC_PATTERN> &Patterns) 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;
+
+ // FP reassociation is only legal when we don't need strict IEEE semantics.
+ if (!Root.getParent()->getParent()->getTarget().Options.UnsafeFPMath)
+ return false;
+
+ // Look for this reassociation pattern:
+ // B = A op X (Prev)
+ // C = B op Y (Root)
+
+ // FIXME: We should also match FMA operations here, where we consider the
+ // 'part' of the FMA, either the addition or the multiplication, paired with
+ // an actual addition or multiplication.
+
+ bool Commute;
+ if (isReassocCandidate(Root, Commute)) {
+ // We found a sequence of instructions that may be suitable for a
+ // reassociation of operands to increase ILP. Specify each commutation
+ // possibility for the Prev instruction in the sequence and let the
+ // machine combiner decide if changing the operands is worthwhile.
+ if (Commute) {
+ Patterns.push_back(MachineCombinerPattern::MC_REASSOC_AX_YB);
+ Patterns.push_back(MachineCombinerPattern::MC_REASSOC_XA_YB);
+ } else {
+ Patterns.push_back(MachineCombinerPattern::MC_REASSOC_AX_BY);
+ Patterns.push_back(MachineCombinerPattern::MC_REASSOC_XA_BY);
+ }
+ return true;
+ }
+
+ return false;
+}
+
+/// Attempt the following reassociation to reduce critical path length:
+/// B = A op X (Prev)
+/// C = B op Y (Root)
+/// ===>
+/// B = X op Y
+/// C = A op B
+static void reassociateOps(MachineInstr &Root, MachineInstr &Prev,
+ MachineCombinerPattern::MC_PATTERN Pattern,
+ SmallVectorImpl<MachineInstr *> &InsInstrs,
+ SmallVectorImpl<MachineInstr *> &DelInstrs,
+ DenseMap<unsigned, unsigned> &InstrIdxForVirtReg) {
+ MachineFunction *MF = Root.getParent()->getParent();
+ MachineRegisterInfo &MRI = MF->getRegInfo();
+ const TargetInstrInfo *TII = MF->getSubtarget().getInstrInfo();
+ const TargetRegisterInfo *TRI = MF->getSubtarget().getRegisterInfo();
+ const TargetRegisterClass *RC = Root.getRegClassConstraint(0, TII, TRI);
+
+ // This array encodes the operand index for each parameter because the
+ // operands may be commuted. Each row corresponds to a pattern value,
+ // and each column specifies the index of A, B, X, Y.
+ unsigned OpIdx[4][4] = {
+ { 1, 1, 2, 2 },
+ { 1, 2, 2, 1 },
+ { 2, 1, 1, 2 },
+ { 2, 2, 1, 1 }
+ };
+
+ MachineOperand &OpA = Prev.getOperand(OpIdx[Pattern][0]);
+ MachineOperand &OpB = Root.getOperand(OpIdx[Pattern][1]);
+ MachineOperand &OpX = Prev.getOperand(OpIdx[Pattern][2]);
+ MachineOperand &OpY = Root.getOperand(OpIdx[Pattern][3]);
+ MachineOperand &OpC = Root.getOperand(0);
+
+ unsigned RegA = OpA.getReg();
+ unsigned RegB = OpB.getReg();
+ unsigned RegX = OpX.getReg();
+ unsigned RegY = OpY.getReg();
+ unsigned RegC = OpC.getReg();
+
+ if (TargetRegisterInfo::isVirtualRegister(RegA))
+ MRI.constrainRegClass(RegA, RC);
+ if (TargetRegisterInfo::isVirtualRegister(RegB))
+ MRI.constrainRegClass(RegB, RC);
+ if (TargetRegisterInfo::isVirtualRegister(RegX))
+ MRI.constrainRegClass(RegX, RC);
+ if (TargetRegisterInfo::isVirtualRegister(RegY))
+ MRI.constrainRegClass(RegY, RC);
+ if (TargetRegisterInfo::isVirtualRegister(RegC))
+ MRI.constrainRegClass(RegC, RC);
+
+ // Create a new virtual register for the result of (X op Y) instead of
+ // recycling RegB because the MachineCombiner's computation of the critical
+ // path requires a new register definition rather than an existing one.
+ unsigned NewVR = MRI.createVirtualRegister(RC);
+ InstrIdxForVirtReg.insert(std::make_pair(NewVR, 0));
+
+ unsigned Opcode = Root.getOpcode();
+ bool KillA = OpA.isKill();
+ bool KillX = OpX.isKill();
+ bool KillY = OpY.isKill();
+
+ // Create new instructions for insertion.
+ MachineInstrBuilder MIB1 =
+ BuildMI(*MF, Prev.getDebugLoc(), TII->get(Opcode), NewVR)
+ .addReg(RegX, getKillRegState(KillX))
+ .addReg(RegY, getKillRegState(KillY));
+ InsInstrs.push_back(MIB1);
+
+ MachineInstrBuilder MIB2 =
+ BuildMI(*MF, Root.getDebugLoc(), TII->get(Opcode), RegC)
+ .addReg(RegA, getKillRegState(KillA))
+ .addReg(NewVR, getKillRegState(true));
+ InsInstrs.push_back(MIB2);
+
+ // Record old instructions for deletion.
+ DelInstrs.push_back(&Prev);
+ DelInstrs.push_back(&Root);
+}
+
+void PPCInstrInfo::genAlternativeCodeSequence(
+ MachineInstr &Root,
+ MachineCombinerPattern::MC_PATTERN Pattern,
+ SmallVectorImpl<MachineInstr *> &InsInstrs,
+ SmallVectorImpl<MachineInstr *> &DelInstrs,
+ DenseMap<unsigned, unsigned> &InstIdxForVirtReg) const {
+ MachineRegisterInfo &MRI = Root.getParent()->getParent()->getRegInfo();
+
+ // Select the previous instruction in the sequence based on the input pattern.
+ MachineInstr *Prev = nullptr;
+ switch (Pattern) {
+ case MachineCombinerPattern::MC_REASSOC_AX_BY:
+ case MachineCombinerPattern::MC_REASSOC_XA_BY:
+ Prev = MRI.getUniqueVRegDef(Root.getOperand(1).getReg());
+ break;
+ case MachineCombinerPattern::MC_REASSOC_AX_YB:
+ case MachineCombinerPattern::MC_REASSOC_XA_YB:
+ Prev = MRI.getUniqueVRegDef(Root.getOperand(2).getReg());
+ }
+ assert(Prev && "Unknown pattern for machine combiner");
+
+ reassociateOps(Root, *Prev, Pattern, InsInstrs, DelInstrs, InstIdxForVirtReg);
+ return;
+}
+
// Detect 32 -> 64-bit extensions where we may reuse the low sub-register.
bool PPCInstrInfo::isCoalescableExtInstr(const MachineInstr &MI,
unsigned &SrcReg, unsigned &DstReg,
case PPC::RESTORE_CRBIT:
case PPC::LVX:
case PPC::LXVD2X:
+ case PPC::QVLFDX:
+ case PPC::QVLFSXs:
+ case PPC::QVLFDXb:
case PPC::RESTORE_VRSAVE:
// Check for the operands added by addFrameReference (the immediate is the
// offset which defaults to 0).
case PPC::SPILL_CRBIT:
case PPC::STVX:
case PPC::STXVD2X:
+ case PPC::QVSTFDX:
+ case PPC::QVSTFSXs:
+ case PPC::QVSTFDXb:
case PPC::SPILL_VRSAVE:
// Check for the operands added by addFrameReference (the immediate is the
// offset which defaults to 0).
// Normal instructions can be commuted the obvious way.
if (MI->getOpcode() != PPC::RLWIMI &&
- MI->getOpcode() != PPC::RLWIMIo &&
- MI->getOpcode() != PPC::RLWIMI8 &&
- MI->getOpcode() != PPC::RLWIMI8o)
+ MI->getOpcode() != PPC::RLWIMIo)
return TargetInstrInfo::commuteInstruction(MI, NewMI);
+ // 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)
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.
unsigned Reg0 = ChangeReg0 ? Reg2 : MI->getOperand(0).getReg();
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 */
}
DebugLoc DL;
BuildMI(MBB, MI, DL, get(Opcode));
}
+/// getNoopForMachoTarget - Return the noop instruction to use for a noop.
+void PPCInstrInfo::getNoopForMachoTarget(MCInst &NopInst) const {
+ NopInst.setOpcode(PPC::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 isPPC64 = Subtarget.isPPC64();
// If the block has no terminators, it just falls into the block after it.
- MachineBasicBlock::iterator I = MBB.end();
- if (I == MBB.begin())
+ MachineBasicBlock::iterator I = MBB.getLastNonDebugInstr();
+ if (I == MBB.end())
return false;
- --I;
- while (I->isDebugValue()) {
- if (I == MBB.begin())
- return false;
- --I;
- }
+
if (!isUnpredicatedTerminator(I))
return false;
}
unsigned PPCInstrInfo::RemoveBranch(MachineBasicBlock &MBB) const {
- MachineBasicBlock::iterator I = MBB.end();
- if (I == MBB.begin()) return 0;
- --I;
- while (I->isDebugValue()) {
- if (I == MBB.begin())
- return 0;
- --I;
- }
+ 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 &&
unsigned
PPCInstrInfo::InsertBranch(MachineBasicBlock &MBB, MachineBasicBlock *TBB,
MachineBasicBlock *FBB,
- const SmallVectorImpl<MachineOperand> &Cond,
+ ArrayRef<MachineOperand> Cond,
DebugLoc DL) const {
// Shouldn't be a fall through.
assert(TBB && "InsertBranch must not be told to insert a fallthrough");
// Select analysis.
bool PPCInstrInfo::canInsertSelect(const MachineBasicBlock &MBB,
- const SmallVectorImpl<MachineOperand> &Cond,
+ ArrayRef<MachineOperand> Cond,
unsigned TrueReg, unsigned FalseReg,
int &CondCycles, int &TrueCycles, int &FalseCycles) const {
if (!Subtarget.hasISEL())
void PPCInstrInfo::insertSelect(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI, DebugLoc dl,
- unsigned DestReg,
- const SmallVectorImpl<MachineOperand> &Cond,
+ unsigned DestReg, ArrayRef<MachineOperand> Cond,
unsigned TrueReg, unsigned FalseReg) const {
assert(Cond.size() == 2 &&
"PPC branch conditions have two components!");
.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, DebugLoc DL,
unsigned DestReg, unsigned SrcReg,
// legalization. Promote them here.
const TargetRegisterInfo *TRI = &getRegisterInfo();
if (PPC::F8RCRegClass.contains(DestReg) &&
- PPC::VSLRCRegClass.contains(SrcReg)) {
+ PPC::VSRCRegClass.contains(SrcReg)) {
unsigned SuperReg =
TRI->getMatchingSuperReg(DestReg, PPC::sub_64, &PPC::VSRCRegClass);
DestReg = SuperReg;
} else if (PPC::VRRCRegClass.contains(DestReg) &&
- PPC::VSHRCRegClass.contains(SrcReg)) {
+ PPC::VSRCRegClass.contains(SrcReg)) {
unsigned SuperReg =
TRI->getMatchingSuperReg(DestReg, PPC::sub_128, &PPC::VSRCRegClass);
DestReg = SuperReg;
} else if (PPC::F8RCRegClass.contains(SrcReg) &&
- PPC::VSLRCRegClass.contains(DestReg)) {
+ PPC::VSRCRegClass.contains(DestReg)) {
unsigned SuperReg =
TRI->getMatchingSuperReg(SrcReg, PPC::sub_64, &PPC::VSRCRegClass);
SrcReg = SuperReg;
} else if (PPC::VRRCRegClass.contains(SrcReg) &&
- PPC::VSHRCRegClass.contains(DestReg)) {
+ PPC::VSRCRegClass.contains(DestReg)) {
unsigned SuperReg =
TRI->getMatchingSuperReg(SrcReg, PPC::sub_128, &PPC::VSRCRegClass);
SrcReg = SuperReg;
}
+ // Different class register copy
+ if (PPC::CRBITRCRegClass.contains(SrcReg) &&
+ PPC::GPRCRegClass.contains(DestReg)) {
+ unsigned 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)) {
+ BuildMI(MBB, I, DL, get(PPC::MFOCRF8), DestReg)
+ .addReg(SrcReg), getKillRegState(KillSrc);
+ return;
+ } else if (PPC::CRRCRegClass.contains(SrcReg) &&
+ PPC::GPRCRegClass.contains(DestReg)) {
+ BuildMI(MBB, I, DL, get(PPC::MFOCRF), DestReg)
+ .addReg(SrcReg), getKillRegState(KillSrc);
+ return;
+ }
+
unsigned Opc;
if (PPC::GPRCRegClass.contains(DestReg, SrcReg))
Opc = PPC::OR;
// 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))
+ else if (PPC::VSFRCRegClass.contains(DestReg, SrcReg) ||
+ PPC::VSSRCRegClass.contains(DestReg, SrcReg))
Opc = PPC::XXLORf;
+ else if (PPC::QFRCRegClass.contains(DestReg, SrcReg))
+ Opc = PPC::QVFMR;
+ else if (PPC::QSRCRegClass.contains(DestReg, SrcReg))
+ Opc = PPC::QVFMRs;
+ else if (PPC::QBRCRegClass.contains(DestReg, SrcReg))
+ Opc = PPC::QVFMRb;
else if (PPC::CRBITRCRegClass.contains(DestReg, SrcReg))
Opc = PPC::CROR;
else
getKillRegState(isKill)),
FrameIdx));
NonRI = true;
+ } else if (PPC::VSSRCRegClass.hasSubClassEq(RC)) {
+ NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::STXSSPX))
+ .addReg(SrcReg,
+ getKillRegState(isKill)),
+ FrameIdx));
+ NonRI = true;
} else if (PPC::VRSAVERCRegClass.hasSubClassEq(RC)) {
assert(Subtarget.isDarwin() &&
"VRSAVE only needs spill/restore on Darwin");
getKillRegState(isKill)),
FrameIdx));
SpillsVRS = true;
+ } else if (PPC::QFRCRegClass.hasSubClassEq(RC)) {
+ NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::QVSTFDX))
+ .addReg(SrcReg,
+ getKillRegState(isKill)),
+ FrameIdx));
+ NonRI = true;
+ } else if (PPC::QSRCRegClass.hasSubClassEq(RC)) {
+ NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::QVSTFSXs))
+ .addReg(SrcReg,
+ getKillRegState(isKill)),
+ FrameIdx));
+ NonRI = true;
+ } else if (PPC::QBRCRegClass.hasSubClassEq(RC)) {
+ NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::QVSTFDXb))
+ .addReg(SrcReg,
+ getKillRegState(isKill)),
+ FrameIdx));
+ NonRI = true;
} else {
llvm_unreachable("Unknown regclass!");
}
MBB.insert(MI, NewMIs[i]);
const MachineFrameInfo &MFI = *MF.getFrameInfo();
- MachineMemOperand *MMO =
- MF.getMachineMemOperand(MachinePointerInfo::getFixedStack(FrameIdx),
- MachineMemOperand::MOStore,
- MFI.getObjectSize(FrameIdx),
- MFI.getObjectAlignment(FrameIdx));
+ MachineMemOperand *MMO = MF.getMachineMemOperand(
+ MachinePointerInfo::getFixedStack(MF, FrameIdx),
+ MachineMemOperand::MOStore, MFI.getObjectSize(FrameIdx),
+ MFI.getObjectAlignment(FrameIdx));
NewMIs.back()->addMemOperand(MF, MMO);
}
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LXSDX), DestReg),
FrameIdx));
NonRI = true;
+ } else if (PPC::VSSRCRegClass.hasSubClassEq(RC)) {
+ NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LXSSPX), DestReg),
+ FrameIdx));
+ NonRI = true;
} else if (PPC::VRSAVERCRegClass.hasSubClassEq(RC)) {
assert(Subtarget.isDarwin() &&
"VRSAVE only needs spill/restore on Darwin");
DestReg),
FrameIdx));
SpillsVRS = true;
+ } else if (PPC::QFRCRegClass.hasSubClassEq(RC)) {
+ NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::QVLFDX), DestReg),
+ FrameIdx));
+ NonRI = true;
+ } else if (PPC::QSRCRegClass.hasSubClassEq(RC)) {
+ NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::QVLFSXs), DestReg),
+ FrameIdx));
+ NonRI = true;
+ } else if (PPC::QBRCRegClass.hasSubClassEq(RC)) {
+ NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::QVLFDXb), DestReg),
+ FrameIdx));
+ NonRI = true;
} else {
llvm_unreachable("Unknown regclass!");
}
MBB.insert(MI, NewMIs[i]);
const MachineFrameInfo &MFI = *MF.getFrameInfo();
- MachineMemOperand *MMO =
- MF.getMachineMemOperand(MachinePointerInfo::getFixedStack(FrameIdx),
- MachineMemOperand::MOLoad,
- MFI.getObjectSize(FrameIdx),
- MFI.getObjectAlignment(FrameIdx));
+ MachineMemOperand *MMO = MF.getMachineMemOperand(
+ MachinePointerInfo::getFixedStack(MF, FrameIdx),
+ MachineMemOperand::MOLoad, MFI.getObjectSize(FrameIdx),
+ MFI.getObjectAlignment(FrameIdx));
NewMIs.back()->addMemOperand(MF, MMO);
}
return !isPredicated(MI);
}
-bool PPCInstrInfo::PredicateInstruction(
- MachineInstr *MI,
- const SmallVectorImpl<MachineOperand> &Pred) const {
+bool PPCInstrInfo::PredicateInstruction(MachineInstr *MI,
+ ArrayRef<MachineOperand> Pred) const {
unsigned OpC = MI->getOpcode();
- if (OpC == PPC::BLR) {
+ 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() ?
return false;
}
-bool PPCInstrInfo::SubsumesPredicate(
- const SmallVectorImpl<MachineOperand> &Pred1,
- const SmallVectorImpl<MachineOperand> &Pred2) const {
+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");
return false;
case PPC::B:
case PPC::BLR:
+ case PPC::BLR8:
case PPC::BCTR:
case PPC::BCTR8:
case PPC::BCTRL:
const MachineFunction *MF = MI->getParent()->getParent();
const char *AsmStr = MI->getOperand(0).getSymbolName();
return getInlineAsmLength(AsmStr, *MF->getTarget().getMCAsmInfo());
+ } else if (Opcode == TargetOpcode::STACKMAP) {
+ return MI->getOperand(1).getImm();
+ } else if (Opcode == TargetOpcode::PATCHPOINT) {
+ PatchPointOpers Opers(MI);
+ return Opers.getMetaOper(PatchPointOpers::NBytesPos).getImm();
} else {
const MCInstrDesc &Desc = get(Opcode);
return Desc.getSize();
}
}
-#undef DEBUG_TYPE
-#define DEBUG_TYPE "ppc-vsx-fma-mutate"
-
-namespace {
- // PPCVSXFMAMutate pass - For copies between VSX registers and non-VSX registers
- // (Altivec and scalar floating-point registers), we need to transform the
- // copies into subregister copies with other restrictions.
- struct PPCVSXFMAMutate : public MachineFunctionPass {
- static char ID;
- PPCVSXFMAMutate() : MachineFunctionPass(ID) {
- initializePPCVSXFMAMutatePass(*PassRegistry::getPassRegistry());
- }
-
- LiveIntervals *LIS;
-
- const PPCTargetMachine *TM;
- const PPCInstrInfo *TII;
-
-protected:
- bool processBlock(MachineBasicBlock &MBB) {
- bool Changed = false;
-
- MachineRegisterInfo &MRI = MBB.getParent()->getRegInfo();
- for (MachineBasicBlock::iterator I = MBB.begin(), IE = MBB.end();
- I != IE; ++I) {
- MachineInstr *MI = I;
-
- // The default (A-type) VSX FMA form kills the addend (it is taken from
- // the target register, which is then updated to reflect the result of
- // the FMA). If the instruction, however, kills one of the registers
- // used for the product, then we can use the M-form instruction (which
- // will take that value from the to-be-defined register).
-
- int AltOpc = PPC::getAltVSXFMAOpcode(MI->getOpcode());
- if (AltOpc == -1)
- continue;
-
- // This pass is run after register coalescing, and so we're looking for
- // a situation like this:
- // ...
- // %vreg5<def> = COPY %vreg9; VSLRC:%vreg5,%vreg9
- // %vreg5<def,tied1> = XSMADDADP %vreg5<tied0>, %vreg17, %vreg16,
- // %RM<imp-use>; VSLRC:%vreg5,%vreg17,%vreg16
- // ...
- // %vreg9<def,tied1> = XSMADDADP %vreg9<tied0>, %vreg17, %vreg19,
- // %RM<imp-use>; VSLRC:%vreg9,%vreg17,%vreg19
- // ...
- // Where we can eliminate the copy by changing from the A-type to the
- // M-type instruction. Specifically, for this example, this means:
- // %vreg5<def,tied1> = XSMADDADP %vreg5<tied0>, %vreg17, %vreg16,
- // %RM<imp-use>; VSLRC:%vreg5,%vreg17,%vreg16
- // is replaced by:
- // %vreg16<def,tied1> = XSMADDMDP %vreg16<tied0>, %vreg18, %vreg9,
- // %RM<imp-use>; VSLRC:%vreg16,%vreg18,%vreg9
- // and we remove: %vreg5<def> = COPY %vreg9; VSLRC:%vreg5,%vreg9
-
- SlotIndex FMAIdx = LIS->getInstructionIndex(MI);
-
- VNInfo *AddendValNo =
- LIS->getInterval(MI->getOperand(1).getReg()).Query(FMAIdx).valueIn();
- MachineInstr *AddendMI = LIS->getInstructionFromIndex(AddendValNo->def);
-
- // The addend and this instruction must be in the same block.
-
- if (!AddendMI || AddendMI->getParent() != MI->getParent())
- continue;
-
- // The addend must be a full copy within the same register class.
-
- if (!AddendMI->isFullCopy())
- continue;
-
- unsigned AddendSrcReg = AddendMI->getOperand(1).getReg();
- if (TargetRegisterInfo::isVirtualRegister(AddendSrcReg)) {
- if (MRI.getRegClass(AddendMI->getOperand(0).getReg()) !=
- MRI.getRegClass(AddendSrcReg))
- continue;
- } else {
- // If AddendSrcReg is a physical register, make sure the destination
- // register class contains it.
- if (!MRI.getRegClass(AddendMI->getOperand(0).getReg())
- ->contains(AddendSrcReg))
- continue;
- }
-
- // In theory, there could be other uses of the addend copy before this
- // fma. We could deal with this, but that would require additional
- // logic below and I suspect it will not occur in any relevant
- // situations.
- bool OtherUsers = false;
- for (auto J = std::prev(I), JE = MachineBasicBlock::iterator(AddendMI);
- J != JE; --J)
- if (J->readsVirtualRegister(AddendMI->getOperand(0).getReg())) {
- OtherUsers = true;
- break;
- }
-
- if (OtherUsers)
- continue;
-
- // Find one of the product operands that is killed by this instruction.
-
- unsigned KilledProdOp = 0, OtherProdOp = 0;
- if (LIS->getInterval(MI->getOperand(2).getReg())
- .Query(FMAIdx).isKill()) {
- KilledProdOp = 2;
- OtherProdOp = 3;
- } else if (LIS->getInterval(MI->getOperand(3).getReg())
- .Query(FMAIdx).isKill()) {
- KilledProdOp = 3;
- OtherProdOp = 2;
- }
-
- // If there are no killed product operands, then this transformation is
- // likely not profitable.
- if (!KilledProdOp)
- continue;
-
- // In order to replace the addend here with the source of the copy,
- // it must still be live here.
- if (!LIS->getInterval(AddendMI->getOperand(1).getReg()).liveAt(FMAIdx))
- continue;
-
- // Transform: (O2 * O3) + O1 -> (O2 * O1) + O3.
-
- unsigned AddReg = AddendMI->getOperand(1).getReg();
- unsigned KilledProdReg = MI->getOperand(KilledProdOp).getReg();
- unsigned OtherProdReg = MI->getOperand(OtherProdOp).getReg();
-
- unsigned AddSubReg = AddendMI->getOperand(1).getSubReg();
- unsigned KilledProdSubReg = MI->getOperand(KilledProdOp).getSubReg();
- unsigned OtherProdSubReg = MI->getOperand(OtherProdOp).getSubReg();
-
- bool AddRegKill = AddendMI->getOperand(1).isKill();
- bool KilledProdRegKill = MI->getOperand(KilledProdOp).isKill();
- bool OtherProdRegKill = MI->getOperand(OtherProdOp).isKill();
-
- bool AddRegUndef = AddendMI->getOperand(1).isUndef();
- bool KilledProdRegUndef = MI->getOperand(KilledProdOp).isUndef();
- bool OtherProdRegUndef = MI->getOperand(OtherProdOp).isUndef();
-
- unsigned OldFMAReg = MI->getOperand(0).getReg();
-
- assert(OldFMAReg == AddendMI->getOperand(0).getReg() &&
- "Addend copy not tied to old FMA output!");
-
- DEBUG(dbgs() << "VSX FMA Mutation:\n " << *MI;);
-
- MI->getOperand(0).setReg(KilledProdReg);
- MI->getOperand(1).setReg(KilledProdReg);
- MI->getOperand(3).setReg(AddReg);
- MI->getOperand(2).setReg(OtherProdReg);
-
- MI->getOperand(0).setSubReg(KilledProdSubReg);
- MI->getOperand(1).setSubReg(KilledProdSubReg);
- MI->getOperand(3).setSubReg(AddSubReg);
- MI->getOperand(2).setSubReg(OtherProdSubReg);
-
- MI->getOperand(1).setIsKill(KilledProdRegKill);
- MI->getOperand(3).setIsKill(AddRegKill);
- MI->getOperand(2).setIsKill(OtherProdRegKill);
-
- MI->getOperand(1).setIsUndef(KilledProdRegUndef);
- MI->getOperand(3).setIsUndef(AddRegUndef);
- MI->getOperand(2).setIsUndef(OtherProdRegUndef);
-
- MI->setDesc(TII->get(AltOpc));
-
- DEBUG(dbgs() << " -> " << *MI);
-
- // The killed product operand was killed here, so we can reuse it now
- // for the result of the fma.
-
- LiveInterval &FMAInt = LIS->getInterval(OldFMAReg);
- VNInfo *FMAValNo = FMAInt.getVNInfoAt(FMAIdx.getRegSlot());
- for (auto UI = MRI.reg_nodbg_begin(OldFMAReg), UE = MRI.reg_nodbg_end();
- UI != UE;) {
- MachineOperand &UseMO = *UI;
- MachineInstr *UseMI = UseMO.getParent();
- ++UI;
-
- // Don't replace the result register of the copy we're about to erase.
- if (UseMI == AddendMI)
- continue;
-
- UseMO.setReg(KilledProdReg);
- UseMO.setSubReg(KilledProdSubReg);
- }
-
- // Extend the live intervals of the killed product operand to hold the
- // fma result.
-
- LiveInterval &NewFMAInt = LIS->getInterval(KilledProdReg);
- for (LiveInterval::iterator AI = FMAInt.begin(), AE = FMAInt.end();
- AI != AE; ++AI) {
- // Don't add the segment that corresponds to the original copy.
- if (AI->valno == AddendValNo)
- continue;
-
- VNInfo *NewFMAValNo =
- NewFMAInt.getNextValue(AI->start,
- LIS->getVNInfoAllocator());
-
- NewFMAInt.addSegment(LiveInterval::Segment(AI->start, AI->end,
- NewFMAValNo));
- }
- DEBUG(dbgs() << " extended: " << NewFMAInt << '\n');
-
- FMAInt.removeValNo(FMAValNo);
- DEBUG(dbgs() << " trimmed: " << FMAInt << '\n');
-
- // Remove the (now unused) copy.
-
- DEBUG(dbgs() << " removing: " << *AddendMI << '\n');
- LIS->RemoveMachineInstrFromMaps(AddendMI);
- AddendMI->eraseFromParent();
-
- Changed = true;
- }
-
- return Changed;
- }
-
-public:
- bool runOnMachineFunction(MachineFunction &MF) override {
- TM = static_cast<const PPCTargetMachine *>(&MF.getTarget());
- // If we don't have VSX then go ahead and return without doing
- // anything.
- if (!TM->getSubtargetImpl()->hasVSX())
- return false;
-
- LIS = &getAnalysis<LiveIntervals>();
-
- TII = TM->getInstrInfo();
-
- bool Changed = false;
-
- if (DisableVSXFMAMutate)
- return Changed;
-
- for (MachineFunction::iterator I = MF.begin(); I != MF.end();) {
- MachineBasicBlock &B = *I++;
- if (processBlock(B))
- Changed = true;
- }
-
- return Changed;
- }
-
- void getAnalysisUsage(AnalysisUsage &AU) const override {
- AU.addRequired<LiveIntervals>();
- AU.addPreserved<LiveIntervals>();
- AU.addRequired<SlotIndexes>();
- AU.addPreserved<SlotIndexes>();
- MachineFunctionPass::getAnalysisUsage(AU);
- }
- };
-}
-
-INITIALIZE_PASS_BEGIN(PPCVSXFMAMutate, DEBUG_TYPE,
- "PowerPC VSX FMA Mutation", false, false)
-INITIALIZE_PASS_DEPENDENCY(LiveIntervals)
-INITIALIZE_PASS_DEPENDENCY(SlotIndexes)
-INITIALIZE_PASS_END(PPCVSXFMAMutate, DEBUG_TYPE,
- "PowerPC VSX FMA Mutation", false, false)
-
-char &llvm::PPCVSXFMAMutateID = PPCVSXFMAMutate::ID;
-
-char PPCVSXFMAMutate::ID = 0;
-FunctionPass*
-llvm::createPPCVSXFMAMutatePass() { return new PPCVSXFMAMutate(); }
-
-#undef DEBUG_TYPE
-#define DEBUG_TYPE "ppc-vsx-copy"
-
-namespace llvm {
- void initializePPCVSXCopyPass(PassRegistry&);
+std::pair<unsigned, unsigned>
+PPCInstrInfo::decomposeMachineOperandsTargetFlags(unsigned TF) const {
+ const unsigned Mask = PPCII::MO_ACCESS_MASK;
+ return std::make_pair(TF & Mask, TF & ~Mask);
}
-namespace {
- // PPCVSXCopy pass - For copies between VSX registers and non-VSX registers
- // (Altivec and scalar floating-point registers), we need to transform the
- // copies into subregister copies with other restrictions.
- struct PPCVSXCopy : public MachineFunctionPass {
- static char ID;
- PPCVSXCopy() : MachineFunctionPass(ID) {
- initializePPCVSXCopyPass(*PassRegistry::getPassRegistry());
- }
-
- const PPCTargetMachine *TM;
- const PPCInstrInfo *TII;
-
- bool IsRegInClass(unsigned Reg, const TargetRegisterClass *RC,
- MachineRegisterInfo &MRI) {
- if (TargetRegisterInfo::isVirtualRegister(Reg)) {
- return RC->hasSubClassEq(MRI.getRegClass(Reg));
- } else if (RC->contains(Reg)) {
- return true;
- }
-
- return false;
- }
-
- bool IsVSReg(unsigned Reg, MachineRegisterInfo &MRI) {
- return IsRegInClass(Reg, &PPC::VSRCRegClass, MRI);
- }
-
- bool IsVRReg(unsigned Reg, MachineRegisterInfo &MRI) {
- return IsRegInClass(Reg, &PPC::VRRCRegClass, MRI);
- }
-
- bool IsF8Reg(unsigned Reg, MachineRegisterInfo &MRI) {
- return IsRegInClass(Reg, &PPC::F8RCRegClass, MRI);
- }
-
-protected:
- bool processBlock(MachineBasicBlock &MBB) {
- bool Changed = false;
-
- MachineRegisterInfo &MRI = MBB.getParent()->getRegInfo();
- for (MachineBasicBlock::iterator I = MBB.begin(), IE = MBB.end();
- I != IE; ++I) {
- MachineInstr *MI = I;
- if (!MI->isFullCopy())
- continue;
-
- MachineOperand &DstMO = MI->getOperand(0);
- MachineOperand &SrcMO = MI->getOperand(1);
-
- if ( IsVSReg(DstMO.getReg(), MRI) &&
- !IsVSReg(SrcMO.getReg(), MRI)) {
- // This is a copy *to* a VSX register from a non-VSX register.
- Changed = true;
-
- const TargetRegisterClass *SrcRC =
- IsVRReg(SrcMO.getReg(), MRI) ? &PPC::VSHRCRegClass :
- &PPC::VSLRCRegClass;
- assert((IsF8Reg(SrcMO.getReg(), MRI) ||
- IsVRReg(SrcMO.getReg(), MRI)) &&
- "Unknown source for a VSX copy");
-
- unsigned NewVReg = MRI.createVirtualRegister(SrcRC);
- BuildMI(MBB, MI, MI->getDebugLoc(),
- TII->get(TargetOpcode::SUBREG_TO_REG), NewVReg)
- .addImm(1) // add 1, not 0, because there is no implicit clearing
- // of the high bits.
- .addOperand(SrcMO)
- .addImm(IsVRReg(SrcMO.getReg(), MRI) ? PPC::sub_128 :
- PPC::sub_64);
-
- // The source of the original copy is now the new virtual register.
- SrcMO.setReg(NewVReg);
- } else if (!IsVSReg(DstMO.getReg(), MRI) &&
- IsVSReg(SrcMO.getReg(), MRI)) {
- // This is a copy *from* a VSX register to a non-VSX register.
- Changed = true;
-
- const TargetRegisterClass *DstRC =
- IsVRReg(DstMO.getReg(), MRI) ? &PPC::VSHRCRegClass :
- &PPC::VSLRCRegClass;
- assert((IsF8Reg(DstMO.getReg(), MRI) ||
- IsVRReg(DstMO.getReg(), MRI)) &&
- "Unknown destination for a VSX copy");
-
- // Copy the VSX value into a new VSX register of the correct subclass.
- unsigned NewVReg = MRI.createVirtualRegister(DstRC);
- BuildMI(MBB, MI, MI->getDebugLoc(),
- TII->get(TargetOpcode::COPY), NewVReg)
- .addOperand(SrcMO);
-
- // Transform the original copy into a subregister extraction copy.
- SrcMO.setReg(NewVReg);
- SrcMO.setSubReg(IsVRReg(DstMO.getReg(), MRI) ? PPC::sub_128 :
- PPC::sub_64);
- }
- }
-
- return Changed;
- }
-
-public:
- bool runOnMachineFunction(MachineFunction &MF) override {
- TM = static_cast<const PPCTargetMachine *>(&MF.getTarget());
- // If we don't have VSX on the subtarget, don't do anything.
- if (!TM->getSubtargetImpl()->hasVSX())
- return false;
- TII = TM->getInstrInfo();
-
- bool Changed = false;
-
- for (MachineFunction::iterator I = MF.begin(); I != MF.end();) {
- MachineBasicBlock &B = *I++;
- if (processBlock(B))
- Changed = true;
- }
-
- return Changed;
- }
-
- void getAnalysisUsage(AnalysisUsage &AU) const override {
- MachineFunctionPass::getAnalysisUsage(AU);
- }
- };
-}
-
-INITIALIZE_PASS(PPCVSXCopy, DEBUG_TYPE,
- "PowerPC VSX Copy Legalization", false, false)
-
-char PPCVSXCopy::ID = 0;
-FunctionPass*
-llvm::createPPCVSXCopyPass() { return new PPCVSXCopy(); }
-
-#undef DEBUG_TYPE
-#define DEBUG_TYPE "ppc-vsx-copy-cleanup"
-
-namespace llvm {
- void initializePPCVSXCopyCleanupPass(PassRegistry&);
-}
-
-namespace {
- // PPCVSXCopyCleanup pass - We sometimes end up generating self copies of VSX
- // registers (mostly because the ABI code still places all values into the
- // "traditional" floating-point and vector registers). Remove them here.
- struct PPCVSXCopyCleanup : public MachineFunctionPass {
- static char ID;
- PPCVSXCopyCleanup() : MachineFunctionPass(ID) {
- initializePPCVSXCopyCleanupPass(*PassRegistry::getPassRegistry());
- }
-
- const PPCTargetMachine *TM;
- const PPCInstrInfo *TII;
-
-protected:
- bool processBlock(MachineBasicBlock &MBB) {
- bool Changed = false;
-
- SmallVector<MachineInstr *, 4> ToDelete;
- for (MachineBasicBlock::iterator I = MBB.begin(), IE = MBB.end();
- I != IE; ++I) {
- MachineInstr *MI = I;
- if (MI->getOpcode() == PPC::XXLOR &&
- MI->getOperand(0).getReg() == MI->getOperand(1).getReg() &&
- MI->getOperand(0).getReg() == MI->getOperand(2).getReg())
- ToDelete.push_back(MI);
- }
-
- if (!ToDelete.empty())
- Changed = true;
-
- for (unsigned i = 0, ie = ToDelete.size(); i != ie; ++i) {
- DEBUG(dbgs() << "Removing VSX self-copy: " << *ToDelete[i]);
- ToDelete[i]->eraseFromParent();
- }
-
- return Changed;
- }
-
-public:
- bool runOnMachineFunction(MachineFunction &MF) override {
- TM = static_cast<const PPCTargetMachine *>(&MF.getTarget());
- // If we don't have VSX don't bother doing anything here.
- if (!TM->getSubtargetImpl()->hasVSX())
- return false;
- TII = TM->getInstrInfo();
-
- bool Changed = false;
-
- for (MachineFunction::iterator I = MF.begin(); I != MF.end();) {
- MachineBasicBlock &B = *I++;
- if (processBlock(B))
- Changed = true;
- }
-
- return Changed;
- }
-
- void getAnalysisUsage(AnalysisUsage &AU) const override {
- MachineFunctionPass::getAnalysisUsage(AU);
- }
- };
+ArrayRef<std::pair<unsigned, const char *>>
+PPCInstrInfo::getSerializableDirectMachineOperandTargetFlags() const {
+ using namespace PPCII;
+ static const std::pair<unsigned, const char *> TargetFlags[] = {
+ {MO_LO, "ppc-lo"},
+ {MO_HA, "ppc-ha"},
+ {MO_TPREL_LO, "ppc-tprel-lo"},
+ {MO_TPREL_HA, "ppc-tprel-ha"},
+ {MO_DTPREL_LO, "ppc-dtprel-lo"},
+ {MO_TLSLD_LO, "ppc-tlsld-lo"},
+ {MO_TOC_LO, "ppc-toc-lo"},
+ {MO_TLS, "ppc-tls"}};
+ return makeArrayRef(TargetFlags);
}
-INITIALIZE_PASS(PPCVSXCopyCleanup, DEBUG_TYPE,
- "PowerPC VSX Copy Cleanup", false, false)
-
-char PPCVSXCopyCleanup::ID = 0;
-FunctionPass*
-llvm::createPPCVSXCopyCleanupPass() { return new PPCVSXCopyCleanup(); }
-
-#undef DEBUG_TYPE
-#define DEBUG_TYPE "ppc-early-ret"
-STATISTIC(NumBCLR, "Number of early conditional returns");
-STATISTIC(NumBLR, "Number of early returns");
-
-namespace llvm {
- void initializePPCEarlyReturnPass(PassRegistry&);
-}
-
-namespace {
- // PPCEarlyReturn pass - For simple functions without epilogue code, move
- // returns up, and create conditional returns, to avoid unnecessary
- // branch-to-blr sequences.
- struct PPCEarlyReturn : public MachineFunctionPass {
- static char ID;
- PPCEarlyReturn() : MachineFunctionPass(ID) {
- initializePPCEarlyReturnPass(*PassRegistry::getPassRegistry());
- }
-
- const PPCTargetMachine *TM;
- const PPCInstrInfo *TII;
-
-protected:
- bool processBlock(MachineBasicBlock &ReturnMBB) {
- bool Changed = false;
-
- MachineBasicBlock::iterator I = ReturnMBB.begin();
- I = ReturnMBB.SkipPHIsAndLabels(I);
-
- // The block must be essentially empty except for the blr.
- if (I == ReturnMBB.end() || I->getOpcode() != PPC::BLR ||
- I != ReturnMBB.getLastNonDebugInstr())
- return Changed;
-
- SmallVector<MachineBasicBlock*, 8> PredToRemove;
- for (MachineBasicBlock::pred_iterator PI = ReturnMBB.pred_begin(),
- PIE = ReturnMBB.pred_end(); PI != PIE; ++PI) {
- bool OtherReference = false, BlockChanged = false;
- for (MachineBasicBlock::iterator J = (*PI)->getLastNonDebugInstr();;) {
- if (J->getOpcode() == PPC::B) {
- if (J->getOperand(0).getMBB() == &ReturnMBB) {
- // This is an unconditional branch to the return. Replace the
- // branch with a blr.
- BuildMI(**PI, J, J->getDebugLoc(), TII->get(PPC::BLR));
- MachineBasicBlock::iterator K = J--;
- K->eraseFromParent();
- BlockChanged = true;
- ++NumBLR;
- continue;
- }
- } else if (J->getOpcode() == PPC::BCC) {
- if (J->getOperand(2).getMBB() == &ReturnMBB) {
- // This is a conditional branch to the return. Replace the branch
- // with a bclr.
- BuildMI(**PI, J, J->getDebugLoc(), TII->get(PPC::BCCLR))
- .addImm(J->getOperand(0).getImm())
- .addReg(J->getOperand(1).getReg());
- MachineBasicBlock::iterator K = J--;
- K->eraseFromParent();
- BlockChanged = true;
- ++NumBCLR;
- continue;
- }
- } else if (J->getOpcode() == PPC::BC || J->getOpcode() == PPC::BCn) {
- if (J->getOperand(1).getMBB() == &ReturnMBB) {
- // This is a conditional branch to the return. Replace the branch
- // with a bclr.
- BuildMI(**PI, J, J->getDebugLoc(),
- TII->get(J->getOpcode() == PPC::BC ?
- PPC::BCLR : PPC::BCLRn))
- .addReg(J->getOperand(0).getReg());
- MachineBasicBlock::iterator K = J--;
- K->eraseFromParent();
- BlockChanged = true;
- ++NumBCLR;
- continue;
- }
- } else if (J->isBranch()) {
- if (J->isIndirectBranch()) {
- if (ReturnMBB.hasAddressTaken())
- OtherReference = true;
- } else
- for (unsigned i = 0; i < J->getNumOperands(); ++i)
- if (J->getOperand(i).isMBB() &&
- J->getOperand(i).getMBB() == &ReturnMBB)
- OtherReference = true;
- } else if (!J->isTerminator() && !J->isDebugValue())
- break;
-
- if (J == (*PI)->begin())
- break;
-
- --J;
- }
-
- if ((*PI)->canFallThrough() && (*PI)->isLayoutSuccessor(&ReturnMBB))
- OtherReference = true;
-
- // Predecessors are stored in a vector and can't be removed here.
- if (!OtherReference && BlockChanged) {
- PredToRemove.push_back(*PI);
- }
-
- if (BlockChanged)
- Changed = true;
- }
-
- for (unsigned i = 0, ie = PredToRemove.size(); i != ie; ++i)
- PredToRemove[i]->removeSuccessor(&ReturnMBB);
-
- if (Changed && !ReturnMBB.hasAddressTaken()) {
- // We now might be able to merge this blr-only block into its
- // by-layout predecessor.
- if (ReturnMBB.pred_size() == 1 &&
- (*ReturnMBB.pred_begin())->isLayoutSuccessor(&ReturnMBB)) {
- // Move the blr into the preceding block.
- MachineBasicBlock &PrevMBB = **ReturnMBB.pred_begin();
- PrevMBB.splice(PrevMBB.end(), &ReturnMBB, I);
- PrevMBB.removeSuccessor(&ReturnMBB);
- }
-
- if (ReturnMBB.pred_empty())
- ReturnMBB.eraseFromParent();
- }
-
- return Changed;
- }
-
-public:
- bool runOnMachineFunction(MachineFunction &MF) override {
- TM = static_cast<const PPCTargetMachine *>(&MF.getTarget());
- TII = TM->getInstrInfo();
-
- bool Changed = false;
-
- // If the function does not have at least two blocks, then there is
- // nothing to do.
- if (MF.size() < 2)
- return Changed;
-
- for (MachineFunction::iterator I = MF.begin(); I != MF.end();) {
- MachineBasicBlock &B = *I++;
- if (processBlock(B))
- Changed = true;
- }
-
- return Changed;
- }
-
- void getAnalysisUsage(AnalysisUsage &AU) const override {
- MachineFunctionPass::getAnalysisUsage(AU);
- }
- };
+ArrayRef<std::pair<unsigned, const char *>>
+PPCInstrInfo::getSerializableBitmaskMachineOperandTargetFlags() const {
+ using namespace PPCII;
+ static const std::pair<unsigned, const char *> TargetFlags[] = {
+ {MO_PLT_OR_STUB, "ppc-plt-or-stub"},
+ {MO_PIC_FLAG, "ppc-pic"},
+ {MO_NLP_FLAG, "ppc-nlp"},
+ {MO_NLP_HIDDEN_FLAG, "ppc-nlp-hidden"}};
+ return makeArrayRef(TargetFlags);
}
-INITIALIZE_PASS(PPCEarlyReturn, DEBUG_TYPE,
- "PowerPC Early-Return Creation", false, false)
-
-char PPCEarlyReturn::ID = 0;
-FunctionPass*
-llvm::createPPCEarlyReturnPass() { return new PPCEarlyReturn(); }
projects / oota-llvm.git / blobdiff