const MachineMemOperand *&MMO,
int &FrameIndex) const;
+ /// isStackSlotCopy - Return true if the specified machine instruction
+ /// is a copy of one stack slot to another and has no other effect.
+ /// Provide the identity of the two frame indices.
+ virtual bool isStackSlotCopy(const MachineInstr *MI, int &DestFrameIndex,
+ int &SrcFrameIndex) const {
+ return false;
+ }
+
+ /// Compute the size in bytes and offset within a stack slot of a spilled
+ /// register or subregister.
+ ///
+ /// \param [out] Size in bytes of the spilled value.
+ /// \param [out] Offset in bytes within the stack slot.
+ /// \returns true if both Size and Offset are successfully computed.
+ ///
+ /// Not all subregisters have computable spill slots. For example,
+ /// subregisters registers may not be byte-sized, and a pair of discontiguous
+ /// subregisters has no single offset.
+ ///
+ /// Targets with nontrivial bigendian implementations may need to override
+ /// this, particularly to support spilled vector registers.
+ virtual bool getStackSlotRange(const TargetRegisterClass *RC, unsigned SubIdx,
+ unsigned &Size, unsigned &Offset,
+ const TargetMachine *TM) const;
+
/// reMaterialize - Re-issue the specified 'original' instruction at the
/// specific location targeting a new destination register.
/// The register in Orig->getOperand(0).getReg() will be substituted by
/// 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.
+ /// FalseReg, and Cond to the destination register. In most cases, a select
+ /// instruction will be 1 cycle, so CondCycles = TrueCycles = FalseCycles = 1
+ ///
+ /// Some x86 implementations have 2-cycle cmov instructions.
///
/// @param MBB Block where select instruction would be inserted.
/// @param Cond Condition returned by AnalyzeBranch.
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,
- /// e.g. in load instructions. For targets that do not support this
- /// the debug info is simply lost.
- /// If you add this for a target you should handle this DBG_VALUE in the
- /// target-specific AsmPrinter code as well; you will probably get invalid
- /// assembly output if you don't.
- virtual MachineInstr *emitFrameIndexDebugValue(MachineFunction &MF,
- int FrameIx,
- uint64_t Offset,
- const MDNode *MDPtr,
- DebugLoc dl) const {
- return 0;
- }
-
/// foldMemoryOperand - Attempt to fold a load or store of the specified stack
/// slot into the specified machine instruction for the specified operand(s).
/// If this is possible, a new instruction is returned with the specified
return false;
}
+ virtual bool enableClusterLoads() const { return false; }
+
virtual bool shouldClusterLoads(MachineInstr *FirstLdSt,
MachineInstr *SecondLdSt,
unsigned NumLoads) const {
/// FoldImmediate - 'Reg' is known to be defined by a move immediate
/// instruction, try to fold the immediate into the use instruction.
+ /// If MRI->hasOneNonDBGUse(Reg) is true, and this function returns true,
+ /// then the caller may assume that DefMI has been erased from its parent
+ /// block. The caller may assume that it will not be erased by this
+ /// function otherwise.
virtual bool FoldImmediate(MachineInstr *UseMI, MachineInstr *DefMI,
unsigned Reg, MachineRegisterInfo *MRI) const {
return false;
/// computeOperandLatency - Compute and return the latency of the given data
/// dependent def and use when the operand indices are already known.
- ///
- /// FindMin may be set to get the minimum vs. expected latency.
unsigned computeOperandLatency(const InstrItineraryData *ItinData,
const MachineInstr *DefMI, unsigned DefIdx,
- const MachineInstr *UseMI, unsigned UseIdx,
- bool FindMin = false) const;
+ const MachineInstr *UseMI, unsigned UseIdx)
+ const;
/// getInstrLatency - Compute the instruction latency of a given instruction.
/// If the instruction has higher cost when predicated, it's returned via
const MachineInstr *MI,
unsigned *PredCost = 0) const;
+ virtual unsigned getPredicationCost(const MachineInstr *MI) const;
+
virtual int getInstrLatency(const InstrItineraryData *ItinData,
SDNode *Node) const;
const MachineInstr *DefMI) const;
int computeDefOperandLatency(const InstrItineraryData *ItinData,
- const MachineInstr *DefMI, bool FindMin) const;
+ const MachineInstr *DefMI) const;
/// isHighLatencyDef - Return true if this opcode has high latency to its
/// result.
return 0;
}
+ /// \brief Return the minimum clearance before an instruction that reads an
+ /// unused register.
+ ///
+ /// For example, AVX instructions may copy part of an register operand into
+ /// the unused high bits of the destination register.
+ ///
+ /// vcvtsi2sdq %rax, %xmm0<undef>, %xmm14
+ ///
+ /// In the code above, vcvtsi2sdq copies %xmm0[127:64] into %xmm14 creating a
+ /// false dependence on any previous write to %xmm0.
+ ///
+ /// This hook works similarly to getPartialRegUpdateClearance, except that it
+ /// does not take an operand index. Instead sets \p OpNum to the index of the
+ /// unused register.
+ virtual unsigned getUndefRegClearance(const MachineInstr *MI, unsigned &OpNum,
+ const TargetRegisterInfo *TRI) const {
+ // The default implementation returns 0 for no undef register dependency.
+ return 0;
+ }
+
/// breakPartialRegDependency - Insert a dependency-breaking instruction
/// before MI to eliminate an unwanted dependency on OpNum.
///
projects / oota-llvm.git / blobdiff