#ifndef LLVM_TARGET_TARGETINSTRINFO_H
#define LLVM_TARGET_TARGETINSTRINFO_H
-#include "llvm/Target/TargetInstrDesc.h"
+#include "llvm/MC/MCInstrInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
namespace llvm {
-class CalleeSavedInfo;
class InstrItineraryData;
class LiveVariables;
class MCAsmInfo;
class MachineMemOperand;
+class MachineRegisterInfo;
class MDNode;
class MCInst;
class SDNode;
class ScheduleHazardRecognizer;
class SelectionDAG;
+class ScheduleDAG;
class TargetRegisterClass;
class TargetRegisterInfo;
+class BranchProbability;
template<class T> class SmallVectorImpl;
///
/// TargetInstrInfo - Interface to description of machine instruction set
///
-class TargetInstrInfo {
- const TargetInstrDesc *Descriptors; // Raw array to allow static init'n
- unsigned NumOpcodes; // Number of entries in the desc array
-
+class TargetInstrInfo : public MCInstrInfo {
TargetInstrInfo(const TargetInstrInfo &); // DO NOT IMPLEMENT
void operator=(const TargetInstrInfo &); // DO NOT IMPLEMENT
public:
- TargetInstrInfo(const TargetInstrDesc *desc, unsigned NumOpcodes);
- virtual ~TargetInstrInfo();
+ TargetInstrInfo(int CFSetupOpcode = -1, int CFDestroyOpcode = -1)
+ : CallFrameSetupOpcode(CFSetupOpcode),
+ CallFrameDestroyOpcode(CFDestroyOpcode) {
+ }
- unsigned getNumOpcodes() const { return NumOpcodes; }
+ virtual ~TargetInstrInfo();
- /// get - Return the machine instruction descriptor that corresponds to the
- /// specified instruction opcode.
- ///
- const TargetInstrDesc &get(unsigned Opcode) const {
- assert(Opcode < NumOpcodes && "Invalid opcode!");
- return Descriptors[Opcode];
- }
+ /// getRegClass - Givem a machine instruction descriptor, returns the register
+ /// class constraint for OpNum, or NULL.
+ const TargetRegisterClass *getRegClass(const MCInstrDesc &TID,
+ unsigned OpNum,
+ const TargetRegisterInfo *TRI) const;
/// isTriviallyReMaterializable - Return true if the instruction is trivially
/// rematerializable, meaning it has no side effects and requires no operands
AliasAnalysis *AA) const;
public:
+ /// getCallFrameSetup/DestroyOpcode - These methods return the opcode of the
+ /// frame setup/destroy instructions if they exist (-1 otherwise). Some
+ /// targets use pseudo instructions in order to abstract away the difference
+ /// between operating with a frame pointer and operating without, through the
+ /// use of these two instructions.
+ ///
+ int getCallFrameSetupOpcode() const { return CallFrameSetupOpcode; }
+ int getCallFrameDestroyOpcode() const { return CallFrameDestroyOpcode; }
+
/// isCoalescableExtInstr - Return true if the instruction is a "coalescable"
/// extension instruction. That is, it's like a copy where it's legal for the
/// source to overlap the destination. e.g. X86::MOVSX64rr32. If this returns
int &FrameIndex) const {
return 0;
}
-
+
/// isStoreToStackSlot - If the specified machine instruction is a direct
/// store to a stack slot, return the virtual or physical register number of
/// the source reg along with the FrameIndex of the loaded stack slot. If
/// produceSameValue - Return true if two machine instructions would produce
/// identical values. By default, this is only true when the two instructions
- /// are deemed identical except for defs.
+ /// are deemed identical except for defs. If this function is called when the
+ /// IR is still in SSA form, the caller can pass the MachineRegisterInfo for
+ /// aggressive checks.
virtual bool produceSameValue(const MachineInstr *MI0,
- const MachineInstr *MI1) const = 0;
+ const MachineInstr *MI1,
+ const MachineRegisterInfo *MRI = 0) const = 0;
/// AnalyzeBranch - Analyze the branching code at the end of MBB, returning
/// true if it cannot be understood (e.g. it's a switch dispatch or isn't
/// This is only invoked in cases where AnalyzeBranch returns success. It
/// returns the number of instructions that were removed.
virtual unsigned RemoveBranch(MachineBasicBlock &MBB) const {
- assert(0 && "Target didn't implement TargetInstrInfo::RemoveBranch!");
+ assert(0 && "Target didn't implement TargetInstrInfo::RemoveBranch!");
return 0;
}
MachineBasicBlock *FBB,
const SmallVectorImpl<MachineOperand> &Cond,
DebugLoc DL) const {
- assert(0 && "Target didn't implement TargetInstrInfo::InsertBranch!");
+ assert(0 && "Target didn't implement TargetInstrInfo::InsertBranch!");
return 0;
}
return true;
}
- /// isProfitableToIfCvt - Return true if it's profitable to first "NumInstrs"
- /// of the specified basic block.
+ /// isProfitableToIfCvt - Return true if it's profitable to predicate
+ /// instructions with accumulated instruction latency of "NumCycles"
+ /// of the specified basic block, where the probability of the instructions
+ /// being executed is given by Probability, and Confidence is a measure
+ /// of our confidence that it will be properly predicted.
virtual
- bool isProfitableToIfCvt(MachineBasicBlock &MBB, unsigned NumInstrs) const {
+ bool isProfitableToIfCvt(MachineBasicBlock &MBB, unsigned NumCyles,
+ unsigned ExtraPredCycles,
+ const BranchProbability &Probability) const {
return false;
}
-
+
/// isProfitableToIfCvt - Second variant of isProfitableToIfCvt, this one
/// checks for the case where two basic blocks from true and false path
/// of a if-then-else (diamond) are predicated on mutally exclusive
- /// predicates.
+ /// predicates, where the probability of the true path being taken is given
+ /// by Probability, and Confidence is a measure of our confidence that it
+ /// will be properly predicted.
virtual bool
- isProfitableToIfCvt(MachineBasicBlock &TMBB, unsigned NumTInstrs,
- MachineBasicBlock &FMBB, unsigned NumFInstrs) const {
+ isProfitableToIfCvt(MachineBasicBlock &TMBB,
+ unsigned NumTCycles, unsigned ExtraTCycles,
+ MachineBasicBlock &FMBB,
+ unsigned NumFCycles, unsigned ExtraFCycles,
+ const BranchProbability &Probability) const {
return false;
}
/// isProfitableToDupForIfCvt - Return true if it's profitable for
- /// if-converter to duplicate a specific number of instructions in the
- /// specified MBB to enable if-conversion.
+ /// if-converter to duplicate instructions of specified accumulated
+ /// instruction latencies in the specified MBB to enable if-conversion.
+ /// The probability of the instructions being executed is given by
+ /// Probability, and Confidence is a measure of our confidence that it
+ /// will be properly predicted.
virtual bool
- isProfitableToDupForIfCvt(MachineBasicBlock &MBB,unsigned NumInstrs) const {
+ isProfitableToDupForIfCvt(MachineBasicBlock &MBB, unsigned NumCyles,
+ const BranchProbability &Probability) const {
+ return false;
+ }
+
+ /// isProfitableToUnpredicate - Return true if it's profitable to unpredicate
+ /// one side of a 'diamond', i.e. two sides of if-else predicated on mutually
+ /// exclusive predicates.
+ /// e.g.
+ /// subeq r0, r1, #1
+ /// addne r0, r1, #1
+ /// =>
+ /// sub r0, r1, #1
+ /// addne r0, r1, #1
+ ///
+ /// This may be profitable is conditional instructions are always executed.
+ virtual bool isProfitableToUnpredicate(MachineBasicBlock &TMBB,
+ MachineBasicBlock &FMBB) const {
return false;
}
-
+
/// copyPhysReg - Emit instructions to copy a pair of physical registers.
virtual void copyPhysReg(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI, DebugLoc DL,
const TargetRegisterInfo *TRI) const {
assert(0 && "Target didn't implement TargetInstrInfo::loadRegFromStackSlot!");
}
-
- /// spillCalleeSavedRegisters - Issues instruction(s) to spill all callee
- /// saved registers and returns true if it isn't possible / profitable to do
- /// so by issuing a series of store instructions via
- /// storeRegToStackSlot(). Returns false otherwise.
- virtual bool spillCalleeSavedRegisters(MachineBasicBlock &MBB,
- MachineBasicBlock::iterator MI,
- const std::vector<CalleeSavedInfo> &CSI,
- const TargetRegisterInfo *TRI) const {
- return false;
- }
- /// restoreCalleeSavedRegisters - Issues instruction(s) to restore all callee
- /// saved registers and returns true if it isn't possible / profitable to do
- /// so by issuing a series of load instructions via loadRegToStackSlot().
- /// Returns false otherwise.
- virtual bool restoreCalleeSavedRegisters(MachineBasicBlock &MBB,
- MachineBasicBlock::iterator MI,
- const std::vector<CalleeSavedInfo> &CSI,
- const TargetRegisterInfo *TRI) const {
+ /// expandPostRAPseudo - This function is called for all pseudo instructions
+ /// that remain after register allocation. Many pseudo instructions are
+ /// created to help register allocation. This is the place to convert them
+ /// into real instructions. The target can edit MI in place, or it can insert
+ /// new instructions and erase MI. The function should return true if
+ /// anything was changed.
+ virtual bool expandPostRAPseudo(MachineBasicBlock::iterator MI) const {
return false;
}
-
+
/// emitFrameIndexDebugValue - Emit a target-dependent form of
/// DBG_VALUE encoding the address of a frame index. Addresses would
/// normally be lowered the same way as other addresses on the target,
}
/// shouldScheduleLoadsNear - This is a used by the pre-regalloc scheduler to
- /// determine (in conjuction with areLoadsFromSameBasePtr) if two loads should
+ /// determine (in conjunction with areLoadsFromSameBasePtr) if two loads should
/// be scheduled togther. On some targets if two loads are loading from
/// addresses in the same cache line, it's better if they are scheduled
/// together. This function takes two integers that represent the load offsets
unsigned NumLoads) const {
return false;
}
-
+
/// ReverseBranchCondition - Reverses the branch condition of the specified
/// condition list, returning false on success and true if it cannot be
/// reversed.
bool ReverseBranchCondition(SmallVectorImpl<MachineOperand> &Cond) const {
return true;
}
-
+
/// insertNoop - Insert a noop into the instruction stream at the specified
/// point.
- virtual void insertNoop(MachineBasicBlock &MBB,
+ virtual void insertNoop(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI) const;
-
-
+
+
/// getNoopForMachoTarget - Return the noop instruction to use for a noop.
virtual void getNoopForMachoTarget(MCInst &NopInst) const {
// Default to just using 'nop' string.
}
-
-
+
+
/// isPredicated - Returns true if the instruction is already predicated.
///
virtual bool isPredicated(const MachineInstr *MI) const {
/// isUnpredicatedTerminator - Returns true if the instruction is a
/// terminator instruction that has not been predicated.
- virtual bool isUnpredicatedTerminator(const MachineInstr *MI) const;
+ virtual bool isUnpredicatedTerminator(const MachineInstr *MI) const = 0;
/// PredicateInstruction - Convert the instruction into a predicated
/// instruction. It returns true if the operation was successful.
const MachineBasicBlock *MBB,
const MachineFunction &MF) const = 0;
- /// GetInstSize - Returns the size of the specified Instruction.
- ///
- virtual unsigned GetInstSizeInBytes(const MachineInstr *MI) const {
- assert(0 && "Target didn't implement TargetInstrInfo::GetInstSize!");
- return 0;
- }
-
/// Measure the specified inline asm to determine an approximation of its
/// length.
virtual unsigned getInlineAsmLength(const char *Str,
const MCAsmInfo &MAI) const;
- /// CreateTargetHazardRecognizer - Allocate and return a hazard recognizer
- /// to use for this target when scheduling the machine instructions after
+ /// CreateTargetHazardRecognizer - Allocate and return a hazard recognizer to
+ /// use for this target when scheduling the machine instructions before
/// register allocation.
virtual ScheduleHazardRecognizer*
- CreateTargetPostRAHazardRecognizer(const InstrItineraryData&) const = 0;
+ CreateTargetHazardRecognizer(const TargetMachine *TM,
+ const ScheduleDAG *DAG) const = 0;
+
+ /// CreateTargetPostRAHazardRecognizer - Allocate and return a hazard
+ /// recognizer to use for this target when scheduling the machine instructions
+ /// after register allocation.
+ virtual ScheduleHazardRecognizer*
+ CreateTargetPostRAHazardRecognizer(const InstrItineraryData*,
+ const ScheduleDAG *DAG) const = 0;
+
+ /// AnalyzeCompare - For a comparison instruction, return the source register
+ /// in SrcReg and the value it compares against in CmpValue. Return true if
+ /// the comparison instruction can be analyzed.
+ virtual bool AnalyzeCompare(const MachineInstr *MI,
+ unsigned &SrcReg, int &Mask, int &Value) const {
+ return false;
+ }
+
+ /// OptimizeCompareInstr - See if the comparison instruction can be converted
+ /// into something more efficient. E.g., on ARM most instructions can set the
+ /// flags register, obviating the need for a separate CMP.
+ virtual bool OptimizeCompareInstr(MachineInstr *CmpInstr,
+ unsigned SrcReg, int Mask, int Value,
+ const MachineRegisterInfo *MRI) const {
+ return false;
+ }
+
+ /// FoldImmediate - 'Reg' is known to be defined by a move immediate
+ /// instruction, try to fold the immediate into the use instruction.
+ virtual bool FoldImmediate(MachineInstr *UseMI, MachineInstr *DefMI,
+ unsigned Reg, MachineRegisterInfo *MRI) const {
+ return false;
+ }
+
+ /// getNumMicroOps - Return the number of u-operations the given machine
+ /// instruction will be decoded to on the target cpu.
+ virtual unsigned getNumMicroOps(const InstrItineraryData *ItinData,
+ const MachineInstr *MI) const;
+
+ /// isZeroCost - Return true for pseudo instructions that don't consume any
+ /// machine resources in their current form. These are common cases that the
+ /// scheduler should consider free, rather than conservatively handling them
+ /// as instructions with no itinerary.
+ bool isZeroCost(unsigned Opcode) const {
+ return Opcode <= TargetOpcode::COPY;
+ }
+
+ /// getOperandLatency - Compute and return the use operand latency of a given
+ /// pair of def and use.
+ /// In most cases, the static scheduling itinerary was enough to determine the
+ /// operand latency. But it may not be possible for instructions with variable
+ /// number of defs / uses.
+ virtual int getOperandLatency(const InstrItineraryData *ItinData,
+ const MachineInstr *DefMI, unsigned DefIdx,
+ const MachineInstr *UseMI, unsigned UseIdx) const;
+
+ virtual int getOperandLatency(const InstrItineraryData *ItinData,
+ SDNode *DefNode, unsigned DefIdx,
+ SDNode *UseNode, unsigned UseIdx) const = 0;
+
+ /// getOutputLatency - Compute and return the output dependency latency of a
+ /// a given pair of defs which both target the same register. This is usually
+ /// one.
+ virtual unsigned getOutputLatency(const InstrItineraryData *ItinData,
+ const MachineInstr *DefMI, unsigned DefIdx,
+ const MachineInstr *DepMI) const {
+ return 1;
+ }
+
+ /// getInstrLatency - Compute the instruction latency of a given instruction.
+ /// If the instruction has higher cost when predicated, it's returned via
+ /// PredCost.
+ virtual int getInstrLatency(const InstrItineraryData *ItinData,
+ const MachineInstr *MI,
+ unsigned *PredCost = 0) const;
+
+ virtual int getInstrLatency(const InstrItineraryData *ItinData,
+ SDNode *Node) const = 0;
+
+ /// isHighLatencyDef - Return true if this opcode has high latency to its
+ /// result.
+ virtual bool isHighLatencyDef(int opc) const { return false; }
+
+ /// hasHighOperandLatency - Compute operand latency between a def of 'Reg'
+ /// and an use in the current loop, return true if the target considered
+ /// it 'high'. This is used by optimization passes such as machine LICM to
+ /// determine whether it makes sense to hoist an instruction out even in
+ /// high register pressure situation.
+ virtual
+ bool hasHighOperandLatency(const InstrItineraryData *ItinData,
+ const MachineRegisterInfo *MRI,
+ const MachineInstr *DefMI, unsigned DefIdx,
+ const MachineInstr *UseMI, unsigned UseIdx) const {
+ return false;
+ }
+
+ /// hasLowDefLatency - Compute operand latency of a def of 'Reg', return true
+ /// if the target considered it 'low'.
+ virtual
+ bool hasLowDefLatency(const InstrItineraryData *ItinData,
+ const MachineInstr *DefMI, unsigned DefIdx) const;
+
+ /// verifyInstruction - Perform target specific instruction verification.
+ virtual
+ bool verifyInstruction(const MachineInstr *MI, StringRef &ErrInfo) const {
+ return true;
+ }
+
+ /// getExecutionDomain - Return the current execution domain and bit mask of
+ /// possible domains for instruction.
+ ///
+ /// Some micro-architectures have multiple execution domains, and multiple
+ /// opcodes that perform the same operation in different domains. For
+ /// example, the x86 architecture provides the por, orps, and orpd
+ /// instructions that all do the same thing. There is a latency penalty if a
+ /// register is written in one domain and read in another.
+ ///
+ /// This function returns a pair (domain, mask) containing the execution
+ /// domain of MI, and a bit mask of possible domains. The setExecutionDomain
+ /// function can be used to change the opcode to one of the domains in the
+ /// bit mask. Instructions whose execution domain can't be changed should
+ /// return a 0 mask.
+ ///
+ /// The execution domain numbers don't have any special meaning except domain
+ /// 0 is used for instructions that are not associated with any interesting
+ /// execution domain.
+ ///
+ virtual std::pair<uint16_t, uint16_t>
+ getExecutionDomain(const MachineInstr *MI) const {
+ return std::make_pair(0, 0);
+ }
+
+ /// setExecutionDomain - Change the opcode of MI to execute in Domain.
+ ///
+ /// The bit (1 << Domain) must be set in the mask returned from
+ /// getExecutionDomain(MI).
+ ///
+ virtual void setExecutionDomain(MachineInstr *MI, unsigned Domain) const {}
+
+
+ /// getPartialRegUpdateClearance - Returns the preferred minimum clearance
+ /// before an instruction with an unwanted partial register update.
+ ///
+ /// Some instructions only write part of a register, and implicitly need to
+ /// read the other parts of the register. This may cause unwanted stalls
+ /// preventing otherwise unrelated instructions from executing in parallel in
+ /// an out-of-order CPU.
+ ///
+ /// For example, the x86 instruction cvtsi2ss writes its result to bits
+ /// [31:0] of the destination xmm register. Bits [127:32] are unaffected, so
+ /// the instruction needs to wait for the old value of the register to become
+ /// available:
+ ///
+ /// addps %xmm1, %xmm0
+ /// movaps %xmm0, (%rax)
+ /// cvtsi2ss %rbx, %xmm0
+ ///
+ /// In the code above, the cvtsi2ss instruction needs to wait for the addps
+ /// instruction before it can issue, even though the high bits of %xmm0
+ /// probably aren't needed.
+ ///
+ /// This hook returns the preferred clearance before MI, measured in
+ /// instructions. Other defs of MI's operand OpNum are avoided in the last N
+ /// instructions before MI. It should only return a positive value for
+ /// unwanted dependencies. If the old bits of the defined register have
+ /// useful values, or if MI is determined to otherwise read the dependency,
+ /// the hook should return 0.
+ ///
+ /// The unwanted dependency may be handled by:
+ ///
+ /// 1. Allocating the same register for an MI def and use. That makes the
+ /// unwanted dependency identical to a required dependency.
+ ///
+ /// 2. Allocating a register for the def that has no defs in the previous N
+ /// instructions.
+ ///
+ /// 3. Calling breakPartialRegDependency() with the same arguments. This
+ /// allows the target to insert a dependency breaking instruction.
+ ///
+ virtual unsigned
+ getPartialRegUpdateClearance(const MachineInstr *MI, unsigned OpNum,
+ const TargetRegisterInfo *TRI) const {
+ // The default implementation returns 0 for no partial register dependency.
+ return 0;
+ }
+
+ /// breakPartialRegDependency - Insert a dependency-breaking instruction
+ /// before MI to eliminate an unwanted dependency on OpNum.
+ ///
+ /// If it wasn't possible to avoid a def in the last N instructions before MI
+ /// (see getPartialRegUpdateClearance), this hook will be called to break the
+ /// unwanted dependency.
+ ///
+ /// On x86, an xorps instruction can be used as a dependency breaker:
+ ///
+ /// addps %xmm1, %xmm0
+ /// movaps %xmm0, (%rax)
+ /// xorps %xmm0, %xmm0
+ /// cvtsi2ss %rbx, %xmm0
+ ///
+ /// An <imp-kill> operand should be added to MI if an instruction was
+ /// inserted. This ties the instructions together in the post-ra scheduler.
+ ///
+ virtual void
+ breakPartialRegDependency(MachineBasicBlock::iterator MI, unsigned OpNum,
+ const TargetRegisterInfo *TRI) const {}
+
+private:
+ int CallFrameSetupOpcode, CallFrameDestroyOpcode;
};
/// TargetInstrInfoImpl - This is the default implementation of
/// libcodegen, not in libtarget.
class TargetInstrInfoImpl : public TargetInstrInfo {
protected:
- TargetInstrInfoImpl(const TargetInstrDesc *desc, unsigned NumOpcodes)
- : TargetInstrInfo(desc, NumOpcodes) {}
+ TargetInstrInfoImpl(int CallFrameSetupOpcode = -1,
+ int CallFrameDestroyOpcode = -1)
+ : TargetInstrInfo(CallFrameSetupOpcode, CallFrameDestroyOpcode) {}
public:
virtual void ReplaceTailWithBranchTo(MachineBasicBlock::iterator OldInst,
MachineBasicBlock *NewDest) const;
unsigned &SrcOpIdx2) const;
virtual bool canFoldMemoryOperand(const MachineInstr *MI,
const SmallVectorImpl<unsigned> &Ops) const;
+ virtual bool hasLoadFromStackSlot(const MachineInstr *MI,
+ const MachineMemOperand *&MMO,
+ int &FrameIndex) const;
+ virtual bool hasStoreToStackSlot(const MachineInstr *MI,
+ const MachineMemOperand *&MMO,
+ int &FrameIndex) const;
+ virtual bool isUnpredicatedTerminator(const MachineInstr *MI) const;
virtual bool PredicateInstruction(MachineInstr *MI,
const SmallVectorImpl<MachineOperand> &Pred) const;
virtual void reMaterialize(MachineBasicBlock &MBB,
virtual MachineInstr *duplicate(MachineInstr *Orig,
MachineFunction &MF) const;
virtual bool produceSameValue(const MachineInstr *MI0,
- const MachineInstr *MI1) const;
+ const MachineInstr *MI1,
+ const MachineRegisterInfo *MRI) const;
virtual bool isSchedulingBoundary(const MachineInstr *MI,
const MachineBasicBlock *MBB,
const MachineFunction &MF) const;
+ using TargetInstrInfo::getOperandLatency;
+ virtual int getOperandLatency(const InstrItineraryData *ItinData,
+ SDNode *DefNode, unsigned DefIdx,
+ SDNode *UseNode, unsigned UseIdx) const;
+ using TargetInstrInfo::getInstrLatency;
+ virtual int getInstrLatency(const InstrItineraryData *ItinData,
+ SDNode *Node) const;
+
+ bool usePreRAHazardRecognizer() const;
+
+ virtual ScheduleHazardRecognizer *
+ CreateTargetHazardRecognizer(const TargetMachine*, const ScheduleDAG*) const;
virtual ScheduleHazardRecognizer *
- CreateTargetPostRAHazardRecognizer(const InstrItineraryData&) const;
+ CreateTargetPostRAHazardRecognizer(const InstrItineraryData*,
+ const ScheduleDAG*) const;
};
} // End llvm namespace
projects / oota-llvm.git / blobdiff