#ifndef LLVM_TARGET_TARGETINSTRINFO_H
#define LLVM_TARGET_TARGETINSTRINFO_H
+#include "llvm/ADT/SmallSet.h"
#include "llvm/MC/MCInstrInfo.h"
#include "llvm/CodeGen/DFAPacketizer.h"
#include "llvm/CodeGen/MachineFunction.h"
class MachineRegisterInfo;
class MDNode;
class MCInst;
+class MCSchedModel;
class SDNode;
class ScheduleHazardRecognizer;
class SelectionDAG;
/// TargetInstrInfo - Interface to description of machine instruction set
///
class TargetInstrInfo : public MCInstrInfo {
- TargetInstrInfo(const TargetInstrInfo &); // DO NOT IMPLEMENT
- void operator=(const TargetInstrInfo &); // DO NOT IMPLEMENT
+ TargetInstrInfo(const TargetInstrInfo &) LLVM_DELETED_FUNCTION;
+ void operator=(const TargetInstrInfo &) LLVM_DELETED_FUNCTION;
public:
TargetInstrInfo(int CFSetupOpcode = -1, int CFDestroyOpcode = -1)
: CallFrameSetupOpcode(CFSetupOpcode),
const MachineInstr *Orig,
const TargetRegisterInfo &TRI) const = 0;
- /// scheduleTwoAddrSource - Schedule the copy / re-mat of the source of the
- /// two-addrss instruction inserted by two-address pass.
- virtual void scheduleTwoAddrSource(MachineInstr *SrcMI,
- MachineInstr *UseMI,
- const TargetRegisterInfo &TRI) const {
- // Do nothing.
- }
-
/// duplicate - Create a duplicate of the Orig instruction in MF. This is like
/// MachineFunction::CloneMachineInstr(), but the target may update operands
/// that are required to be unique.
return false;
}
+ /// canInsertSelect - Return true if it is possible to insert a select
+ /// instruction that chooses between TrueReg and FalseReg based on the
+ /// condition code in Cond.
+ ///
+ /// When successful, also return the latency in cycles from TrueReg,
+ /// FalseReg, and Cond to the destination register. The Cond latency should
+ /// compensate for a conditional branch being removed. For example, if a
+ /// conditional branch has a 3 cycle latency from the condition code read,
+ /// and a cmov instruction has a 2 cycle latency from the condition code
+ /// read, CondCycles should be returned as -1.
+ ///
+ /// @param MBB Block where select instruction would be inserted.
+ /// @param Cond Condition returned by AnalyzeBranch.
+ /// @param TrueReg Virtual register to select when Cond is true.
+ /// @param FalseReg Virtual register to select when Cond is false.
+ /// @param CondCycles Latency from Cond+Branch to select output.
+ /// @param TrueCycles Latency from TrueReg to select output.
+ /// @param FalseCycles Latency from FalseReg to select output.
+ virtual bool canInsertSelect(const MachineBasicBlock &MBB,
+ const SmallVectorImpl<MachineOperand> &Cond,
+ unsigned TrueReg, unsigned FalseReg,
+ int &CondCycles,
+ int &TrueCycles, int &FalseCycles) const {
+ return false;
+ }
+
+ /// insertSelect - Insert a select instruction into MBB before I that will
+ /// copy TrueReg to DstReg when Cond is true, and FalseReg to DstReg when
+ /// Cond is false.
+ ///
+ /// This function can only be called after canInsertSelect() returned true.
+ /// The condition in Cond comes from AnalyzeBranch, and it can be assumed
+ /// that the same flags or registers required by Cond are available at the
+ /// insertion point.
+ ///
+ /// @param MBB Block where select instruction should be inserted.
+ /// @param I Insertion point.
+ /// @param DL Source location for debugging.
+ /// @param DstReg Virtual register to be defined by select instruction.
+ /// @param Cond Condition as computed by AnalyzeBranch.
+ /// @param TrueReg Virtual register to copy when Cond is true.
+ /// @param FalseReg Virtual register to copy when Cons is false.
+ virtual void insertSelect(MachineBasicBlock &MBB,
+ MachineBasicBlock::iterator I, DebugLoc DL,
+ unsigned DstReg,
+ const SmallVectorImpl<MachineOperand> &Cond,
+ unsigned TrueReg, unsigned FalseReg) const {
+ llvm_unreachable("Target didn't implement TargetInstrInfo::insertSelect!");
+ }
+
+ /// analyzeSelect - Analyze the given select instruction, returning true if
+ /// it cannot be understood. It is assumed that MI->isSelect() is true.
+ ///
+ /// When successful, return the controlling condition and the operands that
+ /// determine the true and false result values.
+ ///
+ /// Result = SELECT Cond, TrueOp, FalseOp
+ ///
+ /// Some targets can optimize select instructions, for example by predicating
+ /// the instruction defining one of the operands. Such targets should set
+ /// Optimizable.
+ ///
+ /// @param MI Select instruction to analyze.
+ /// @param Cond Condition controlling the select.
+ /// @param TrueOp Operand number of the value selected when Cond is true.
+ /// @param FalseOp Operand number of the value selected when Cond is false.
+ /// @param Optimizable Returned as true if MI is optimizable.
+ /// @returns False on success.
+ virtual bool analyzeSelect(const MachineInstr *MI,
+ SmallVectorImpl<MachineOperand> &Cond,
+ unsigned &TrueOp, unsigned &FalseOp,
+ bool &Optimizable) const {
+ assert(MI && MI->isSelect() && "MI must be a select instruction");
+ return true;
+ }
+
+ /// optimizeSelect - Given a select instruction that was understood by
+ /// analyzeSelect and returned Optimizable = true, attempt to optimize MI by
+ /// merging it with one of its operands. Returns NULL on failure.
+ ///
+ /// When successful, returns the new select instruction. The client is
+ /// responsible for deleting MI.
+ ///
+ /// If both sides of the select can be optimized, PreferFalse is used to pick
+ /// a side.
+ ///
+ /// @param MI Optimizable select instruction.
+ /// @param PreferFalse Try to optimize FalseOp instead of TrueOp.
+ /// @returns Optimized instruction or NULL.
+ virtual MachineInstr *optimizeSelect(MachineInstr *MI,
+ bool PreferFalse = false) const {
+ // This function must be implemented if Optimizable is ever set.
+ llvm_unreachable("Target must implement TargetInstrInfo::optimizeSelect!");
+ }
+
/// copyPhysReg - Emit instructions to copy a pair of physical registers.
+ ///
+ /// This function should support copies within any legal register class as
+ /// well as any cross-class copies created during instruction selection.
+ ///
+ /// The source and destination registers may overlap, which may require a
+ /// careful implementation when multiple copy instructions are required for
+ /// large registers. See for example the ARM target.
virtual void copyPhysReg(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI, DebugLoc DL,
unsigned DestReg, unsigned SrcReg,
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 {
+ /// analyzeCompare - For a comparison instruction, return the source registers
+ /// in SrcReg and SrcReg2 if having two register operands, 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, unsigned &SrcReg2,
+ int &Mask, int &Value) const {
return false;
}
- /// OptimizeCompareInstr - See if the comparison instruction can be converted
+ /// 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,
+ virtual bool optimizeCompareInstr(MachineInstr *CmpInstr,
+ unsigned SrcReg, unsigned SrcReg2,
+ int Mask, int Value,
const MachineRegisterInfo *MRI) const {
return false;
}
+ /// optimizeLoadInstr - Try to remove the load by folding it to a register
+ /// operand at the use. We fold the load instructions if and only if the
+ /// def and use are in the same BB. We only look at one load and see
+ /// whether it can be folded into MI. FoldAsLoadDefReg is the virtual register
+ /// defined by the load we are trying to fold. DefMI returns the machine
+ /// instruction that defines FoldAsLoadDefReg, and the function returns
+ /// the machine instruction generated due to folding.
+ virtual MachineInstr* optimizeLoadInstr(MachineInstr *MI,
+ const MachineRegisterInfo *MRI,
+ unsigned &FoldAsLoadDefReg,
+ MachineInstr *&DefMI) const {
+ return 0;
+ }
+
/// 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,
}
/// getNumMicroOps - Return the number of u-operations the given machine
- /// instruction will be decoded to on the target cpu.
+ /// instruction will be decoded to on the target cpu. The itinerary's
+ /// IssueWidth is the number of microops that can be dispatched each
+ /// cycle. An instruction with zero microops takes no dispatch resources.
virtual unsigned getNumMicroOps(const InstrItineraryData *ItinData,
const MachineInstr *MI) const = 0;
const MachineInstr *UseMI, unsigned UseIdx,
bool FindMin = false) const;
- /// computeOperandLatency - Compute and return the latency of the given data
- /// dependent def and use. DefMI must be a valid def. UseMI may be NULL for
- /// an unknown use. If the subtarget allows, this may or may not need to call
- /// getOperandLatency().
- ///
- /// FindMin may be set to get the minimum vs. expected latency. Minimum
- /// latency is used for scheduling groups, while expected latency is for
- /// instruction cost and critical path.
- unsigned computeOperandLatency(const InstrItineraryData *ItinData,
- const TargetRegisterInfo *TRI,
- const MachineInstr *DefMI,
- const MachineInstr *UseMI,
- unsigned Reg, bool FindMin) const;
-
- /// 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.
SDNode *Node) const = 0;
/// Return the default expected latency for a def based on it's opcode.
- unsigned defaultDefLatency(const InstrItineraryData *ItinData,
+ unsigned defaultDefLatency(const MCSchedModel *SchedModel,
const MachineInstr *DefMI) const;
+ int computeDefOperandLatency(const InstrItineraryData *ItinData,
+ const MachineInstr *DefMI, bool FindMin) const;
+
/// isHighLatencyDef - Return true if this opcode has high latency to its
/// result.
virtual bool isHighLatencyDef(int opc) const { return false; }