LLVM  mainline
Classes | Public Member Functions
llvm::TargetTransformInfo Class Reference

This pass provides access to the codegen interfaces that are needed for IR-level transformations. More...

#include <TargetTransformInfo.h>

List of all members.

Classes

class  Concept
struct  UnrollingPreferences
 Parameters that control the generic loop unrolling transformation. More...

Public Member Functions

template<typename T >
 TargetTransformInfo (T Impl)
 Construct a TTI object using a type implementing the Concept API below.
 TargetTransformInfo (const DataLayout *DL)
 Construct a baseline TTI object using a minimal implementation of the Concept API below.
 TargetTransformInfo (TargetTransformInfo &&Arg)
TargetTransformInfooperator= (TargetTransformInfo &&RHS)
 ~TargetTransformInfo ()
bool invalidate (Function &, const PreservedAnalyses &)
 Handle the invalidation of this information.

Generic Target Information

enum  TargetCostConstants { TCC_Free = 0, TCC_Basic = 1, TCC_Expensive = 4 }
 Underlying constants for 'cost' values in this interface. More...
unsigned getOperationCost (unsigned Opcode, Type *Ty, Type *OpTy=nullptr) const
 Estimate the cost of a specific operation when lowered.
unsigned getGEPCost (const Value *Ptr, ArrayRef< const Value * > Operands) const
 Estimate the cost of a GEP operation when lowered.
unsigned getCallCost (FunctionType *FTy, int NumArgs=-1) const
 Estimate the cost of a function call when lowered.
unsigned getCallCost (const Function *F, int NumArgs=-1) const
 Estimate the cost of calling a specific function when lowered.
unsigned getCallCost (const Function *F, ArrayRef< const Value * > Arguments) const
 Estimate the cost of calling a specific function when lowered.
unsigned getIntrinsicCost (Intrinsic::ID IID, Type *RetTy, ArrayRef< Type * > ParamTys) const
 Estimate the cost of an intrinsic when lowered.
unsigned getIntrinsicCost (Intrinsic::ID IID, Type *RetTy, ArrayRef< const Value * > Arguments) const
 Estimate the cost of an intrinsic when lowered.
unsigned getUserCost (const User *U) const
 Estimate the cost of a given IR user when lowered.
bool hasBranchDivergence () const
 Return true if branch divergence exists.
bool isSourceOfDivergence (const Value *V) const
 Returns whether V is a source of divergence.
bool isLoweredToCall (const Function *F) const
 Test whether calls to a function lower to actual program function calls.
void getUnrollingPreferences (Loop *L, UnrollingPreferences &UP) const
 Get target-customized preferences for the generic loop unrolling transformation. The caller will initialize UP with the current target-independent defaults.

Scalar Target Information

enum  PopcntSupportKind { PSK_Software, PSK_SlowHardware, PSK_FastHardware }
 Flags indicating the kind of support for population count. More...
bool isLegalAddImmediate (int64_t Imm) const
 Return true if the specified immediate is legal add immediate, that is the target has add instructions which can add a register with the immediate without having to materialize the immediate into a register.
bool isLegalICmpImmediate (int64_t Imm) const
 Return true if the specified immediate is legal icmp immediate, that is the target has icmp instructions which can compare a register against the immediate without having to materialize the immediate into a register.
bool isLegalAddressingMode (Type *Ty, GlobalValue *BaseGV, int64_t BaseOffset, bool HasBaseReg, int64_t Scale) const
 Return true if the addressing mode represented by AM is legal for this target, for a load/store of the specified type. The type may be VoidTy, in which case only return true if the addressing mode is legal for a load/store of any legal type. TODO: Handle pre/postinc as well.
bool isLegalMaskedStore (Type *DataType, int Consecutive) const
 Return true if the target works with masked instruction AVX2 allows masks for consecutive load and store for i32 and i64 elements. AVX-512 architecture will also allow masks for non-consecutive memory accesses.
bool isLegalMaskedLoad (Type *DataType, int Consecutive) const
int getScalingFactorCost (Type *Ty, GlobalValue *BaseGV, int64_t BaseOffset, bool HasBaseReg, int64_t Scale) const
 Return the cost of the scaling factor used in the addressing mode represented by AM for this target, for a load/store of the specified type. If the AM is supported, the return value must be >= 0. If the AM is not supported, it returns a negative value. TODO: Handle pre/postinc as well.
bool isTruncateFree (Type *Ty1, Type *Ty2) const
 Return true if it's free to truncate a value of type Ty1 to type Ty2. e.g. On x86 it's free to truncate a i32 value in register EAX to i16 by referencing its sub-register AX.
bool isProfitableToHoist (Instruction *I) const
 Return true if it is profitable to hoist instruction in the then/else to before if.
bool isTypeLegal (Type *Ty) const
 Return true if this type is legal.
unsigned getJumpBufAlignment () const
 Returns the target's jmp_buf alignment in bytes.
unsigned getJumpBufSize () const
 Returns the target's jmp_buf size in bytes.
bool shouldBuildLookupTables () const
 Return true if switches should be turned into lookup tables for the target.
bool enableAggressiveInterleaving (bool LoopHasReductions) const
 Don't restrict interleaved unrolling to small loops.
PopcntSupportKind getPopcntSupport (unsigned IntTyWidthInBit) const
 Return hardware support for population count.
bool haveFastSqrt (Type *Ty) const
 Return true if the hardware has a fast square-root instruction.
unsigned getFPOpCost (Type *Ty) const
 Return the expected cost of supporting the floating point operation of the specified type.
unsigned getIntImmCost (const APInt &Imm, Type *Ty) const
 Return the expected cost of materializing for the given integer immediate of the specified type.
unsigned getIntImmCost (unsigned Opc, unsigned Idx, const APInt &Imm, Type *Ty) const
 Return the expected cost of materialization for the given integer immediate of the specified type for a given instruction. The cost can be zero if the immediate can be folded into the specified instruction.
unsigned getIntImmCost (Intrinsic::ID IID, unsigned Idx, const APInt &Imm, Type *Ty) const

Vector Target Information

enum  ShuffleKind {
  SK_Broadcast, SK_Reverse, SK_Alternate, SK_InsertSubvector,
  SK_ExtractSubvector
}
 The various kinds of shuffle patterns for vector queries. More...
enum  OperandValueKind { OK_AnyValue, OK_UniformValue, OK_UniformConstantValue, OK_NonUniformConstantValue }
 Additional information about an operand's possible values. More...
enum  OperandValueProperties { OP_None = 0, OP_PowerOf2 = 1 }
 Additional properties of an operand's values. More...
unsigned getNumberOfRegisters (bool Vector) const
unsigned getRegisterBitWidth (bool Vector) const
unsigned getMaxInterleaveFactor () const
unsigned getArithmeticInstrCost (unsigned Opcode, Type *Ty, OperandValueKind Opd1Info=OK_AnyValue, OperandValueKind Opd2Info=OK_AnyValue, OperandValueProperties Opd1PropInfo=OP_None, OperandValueProperties Opd2PropInfo=OP_None) const
unsigned getShuffleCost (ShuffleKind Kind, Type *Tp, int Index=0, Type *SubTp=nullptr) const
unsigned getCastInstrCost (unsigned Opcode, Type *Dst, Type *Src) const
unsigned getCFInstrCost (unsigned Opcode) const
unsigned getCmpSelInstrCost (unsigned Opcode, Type *ValTy, Type *CondTy=nullptr) const
unsigned getVectorInstrCost (unsigned Opcode, Type *Val, unsigned Index=-1) const
unsigned getMemoryOpCost (unsigned Opcode, Type *Src, unsigned Alignment, unsigned AddressSpace) const
unsigned getMaskedMemoryOpCost (unsigned Opcode, Type *Src, unsigned Alignment, unsigned AddressSpace) const
unsigned getReductionCost (unsigned Opcode, Type *Ty, bool IsPairwiseForm) const
 Calculate the cost of performing a vector reduction.
unsigned getIntrinsicInstrCost (Intrinsic::ID ID, Type *RetTy, ArrayRef< Type * > Tys) const
unsigned getCallInstrCost (Function *F, Type *RetTy, ArrayRef< Type * > Tys) const
unsigned getNumberOfParts (Type *Tp) const
unsigned getAddressComputationCost (Type *Ty, bool IsComplex=false) const
unsigned getCostOfKeepingLiveOverCall (ArrayRef< Type * > Tys) const
bool getTgtMemIntrinsic (IntrinsicInst *Inst, MemIntrinsicInfo &Info) const
ValuegetOrCreateResultFromMemIntrinsic (IntrinsicInst *Inst, Type *ExpectedType) const

Detailed Description

This pass provides access to the codegen interfaces that are needed for IR-level transformations.

Definition at line 58 of file TargetTransformInfo.h.


Member Enumeration Documentation

Additional information about an operand's possible values.

Enumerator:
OK_AnyValue 
OK_UniformValue 
OK_UniformConstantValue 
OK_NonUniformConstantValue 

Definition at line 385 of file TargetTransformInfo.h.

Additional properties of an operand's values.

Enumerator:
OP_None 
OP_PowerOf2 

Definition at line 393 of file TargetTransformInfo.h.

Flags indicating the kind of support for population count.

Compared to the SW implementation, HW support is supposed to significantly boost the performance when the population is dense, and it may or may not degrade performance if the population is sparse. A HW support is considered as "Fast" if it can outperform, or is on a par with, SW implementation when the population is sparse; otherwise, it is considered as "Slow".

Enumerator:
PSK_Software 
PSK_SlowHardware 
PSK_FastHardware 

Definition at line 287 of file TargetTransformInfo.h.

The various kinds of shuffle patterns for vector queries.

Enumerator:
SK_Broadcast 

Broadcast element 0 to all other elements.

SK_Reverse 

Reverse the order of the vector.

SK_Alternate 

Choose alternate elements from vector.

SK_InsertSubvector 

InsertSubvector. Index indicates start offset.

SK_ExtractSubvector 

ExtractSubvector Index indicates start offset.

Definition at line 376 of file TargetTransformInfo.h.

Underlying constants for 'cost' values in this interface.

Many APIs in this interface return a cost. This enum defines the fundamental values that should be used to interpret (and produce) those costs. The costs are returned as an unsigned rather than a member of this enumeration because it is expected that the cost of one IR instruction may have a multiplicative factor to it or otherwise won't fit directly into the enum. Moreover, it is common to sum or average costs which works better as simple integral values. Thus this enum only provides constants.

Note that these costs should usually reflect the intersection of code-size cost and execution cost. A free instruction is typically one that folds into another instruction. For example, reg-to-reg moves can often be skipped by renaming the registers in the CPU, but they still are encoded and thus wouldn't be considered 'free' here.

Enumerator:
TCC_Free 

Expected to fold away in lowering.

TCC_Basic 

The cost of a typical 'add' instruction.

TCC_Expensive 

The cost of a 'div' instruction on x86.

Definition at line 111 of file TargetTransformInfo.h.


Constructor & Destructor Documentation

template<typename T >
llvm::TargetTransformInfo::TargetTransformInfo ( T  Impl)

Construct a TTI object using a type implementing the Concept API below.

This is used by targets to construct a TTI wrapping their target-specific implementaion that encodes appropriate costs for their target.

Definition at line 775 of file TargetTransformInfo.h.

Construct a baseline TTI object using a minimal implementation of the Concept API below.

The TTI implementation will reflect the information in the DataLayout provided if non-null.

Definition at line 36 of file TargetTransformInfo.cpp.

Definition at line 41 of file TargetTransformInfo.cpp.

Definition at line 39 of file TargetTransformInfo.cpp.


Member Function Documentation

Don't restrict interleaved unrolling to small loops.

Definition at line 150 of file TargetTransformInfo.cpp.

Returns:
The cost of the address computation. For most targets this can be merged into the instruction indexing mode. Some targets might want to distinguish between address computation for memory operations on vector types and scalar types. Such targets should override this function. The 'IsComplex' parameter is a hint that the address computation is likely to involve multiple instructions and as such unlikely to be merged into the address indexing mode.

Definition at line 253 of file TargetTransformInfo.cpp.

Returns:
The expected cost of arithmetic ops, such as mul, xor, fsub, etc.

Definition at line 193 of file TargetTransformInfo.cpp.

Referenced by visitIVCast().

unsigned TargetTransformInfo::getCallCost ( FunctionType FTy,
int  NumArgs = -1 
) const

Estimate the cost of a function call when lowered.

The contract for this is the same as getOperationCost except that it supports an interface that provides extra information specific to call instructions.

This is the most basic query for estimating call cost: it only knows the function type and (potentially) the number of arguments at the call site. The latter is only interesting for varargs function types.

Definition at line 54 of file TargetTransformInfo.cpp.

Estimate the cost of calling a specific function when lowered.

This overload adds the ability to reason about the particular function being called in the event it is a library call with special lowering.

Estimate the cost of calling a specific function when lowered.

This overload allows specifying a set of candidate argument values.

Definition at line 60 of file TargetTransformInfo.cpp.

unsigned TargetTransformInfo::getCallInstrCost ( Function F,
Type RetTy,
ArrayRef< Type * >  Tys 
) const
Returns:
The cost of Call instructions.

Definition at line 244 of file TargetTransformInfo.cpp.

Referenced by getVectorCallCost().

unsigned TargetTransformInfo::getCastInstrCost ( unsigned  Opcode,
Type Dst,
Type Src 
) const
Returns:
The expected cost of cast instructions, such as bitcast, trunc, zext, etc.

Definition at line 206 of file TargetTransformInfo.cpp.

Returns:
The expected cost of control-flow related instructions such as Phi, Ret, Br.

Definition at line 211 of file TargetTransformInfo.cpp.

unsigned TargetTransformInfo::getCmpSelInstrCost ( unsigned  Opcode,
Type ValTy,
Type CondTy = nullptr 
) const
Returns:
The expected cost of compare and select instructions.

Definition at line 215 of file TargetTransformInfo.cpp.

Returns:
The cost, if any, of keeping values of the given types alive over a callsite.

Some types may require the use of register classes that do not have any callee-saved registers, so would require a spill and fill.

Definition at line 264 of file TargetTransformInfo.cpp.

Return the expected cost of supporting the floating point operation of the specified type.

Definition at line 163 of file TargetTransformInfo.cpp.

Estimate the cost of a GEP operation when lowered.

The contract for this function is the same as getOperationCost except that it supports an interface that provides extra information specific to the GEP operation.

Return the expected cost of materializing for the given integer immediate of the specified type.

Definition at line 167 of file TargetTransformInfo.cpp.

unsigned TargetTransformInfo::getIntImmCost ( unsigned  Opc,
unsigned  Idx,
const APInt Imm,
Type Ty 
) const

Return the expected cost of materialization for the given integer immediate of the specified type for a given instruction. The cost can be zero if the immediate can be folded into the specified instruction.

Definition at line 171 of file TargetTransformInfo.cpp.

Definition at line 176 of file TargetTransformInfo.cpp.

Estimate the cost of an intrinsic when lowered.

Mirrors the getCallCost method but uses an intrinsic identifier.

unsigned TargetTransformInfo::getIntrinsicCost ( Intrinsic::ID  IID,
Type RetTy,
ArrayRef< const Value * >  Arguments 
) const

Estimate the cost of an intrinsic when lowered.

Mirrors the getCallCost method but uses an intrinsic identifier.

Definition at line 66 of file TargetTransformInfo.cpp.

Returns:
The cost of Intrinsic instructions.

Definition at line 239 of file TargetTransformInfo.cpp.

Referenced by getVectorIntrinsicCost().

Returns the target's jmp_buf alignment in bytes.

Definition at line 138 of file TargetTransformInfo.cpp.

Returns the target's jmp_buf size in bytes.

Definition at line 142 of file TargetTransformInfo.cpp.

unsigned TargetTransformInfo::getMaskedMemoryOpCost ( unsigned  Opcode,
Type Src,
unsigned  Alignment,
unsigned  AddressSpace 
) const
Returns:
The cost of masked Load and Store instructions.

Definition at line 232 of file TargetTransformInfo.cpp.

Returns:
The maximum interleave factor that any transform should try to perform for this target. This number depends on the level of parallelism and the number of execution units in the CPU.

Definition at line 189 of file TargetTransformInfo.cpp.

unsigned TargetTransformInfo::getMemoryOpCost ( unsigned  Opcode,
Type Src,
unsigned  Alignment,
unsigned  AddressSpace 
) const
Returns:
The cost of Load and Store instructions.

Definition at line 225 of file TargetTransformInfo.cpp.

Returns:
The number of pieces into which the provided type must be split during legalization. Zero is returned when the answer is unknown.

Definition at line 249 of file TargetTransformInfo.cpp.

Returns:
The number of scalar or vector registers that the target has. If 'Vectors' is true, it returns the number of vector registers. If it is set to false, it returns the number of scalar registers.

Definition at line 181 of file TargetTransformInfo.cpp.

unsigned TargetTransformInfo::getOperationCost ( unsigned  Opcode,
Type Ty,
Type OpTy = nullptr 
) const

Estimate the cost of a specific operation when lowered.

Note that this is designed to work on an arbitrary synthetic opcode, and thus work for hypothetical queries before an instruction has even been formed. However, this does *not* work for GEPs, and must not be called for a GEP instruction. Instead, use the dedicated getGEPCost interface as analyzing a GEP's cost required more information.

Typically only the result type is required, and the operand type can be omitted. However, if the opcode is one of the cast instructions, the operand type is required.

The returned cost is defined in terms of TargetCostConstants, see its comments for a detailed explanation of the cost values.

Definition at line 49 of file TargetTransformInfo.cpp.

Returns:
A value which is the result of the given memory intrinsic. New instructions may be created to extract the result from the given intrinsic memory operation. Returns nullptr if the target cannot create a result from the given intrinsic.

Definition at line 273 of file TargetTransformInfo.cpp.

Return hardware support for population count.

Definition at line 155 of file TargetTransformInfo.cpp.

unsigned TargetTransformInfo::getReductionCost ( unsigned  Opcode,
Type Ty,
bool  IsPairwiseForm 
) const

Calculate the cost of performing a vector reduction.

This is the cost of reducing the vector value of type Ty to a scalar value using the operation denoted by Opcode. The form of the reduction can either be a pairwise reduction or a reduction that splits the vector at every reduction level.

Pairwise: (v0, v1, v2, v3) ((v0+v1), (v2, v3), undef, undef) Split: (v0, v1, v2, v3) ((v0+v2), (v1+v3), undef, undef)

Definition at line 258 of file TargetTransformInfo.cpp.

Returns:
The width of the largest scalar or vector register type.

Definition at line 185 of file TargetTransformInfo.cpp.

int TargetTransformInfo::getScalingFactorCost ( Type Ty,
GlobalValue BaseGV,
int64_t  BaseOffset,
bool  HasBaseReg,
int64_t  Scale 
) const

Return the cost of the scaling factor used in the addressing mode represented by AM for this target, for a load/store of the specified type. If the AM is supported, the return value must be >= 0. If the AM is not supported, it returns a negative value. TODO: Handle pre/postinc as well.

Definition at line 118 of file TargetTransformInfo.cpp.

Referenced by getScalingFactorCost().

unsigned TargetTransformInfo::getShuffleCost ( ShuffleKind  Kind,
Type Tp,
int  Index = 0,
Type SubTp = nullptr 
) const
Returns:
The cost of a shuffle instruction of kind Kind and of type Tp. The index and subtype parameters are used by the subvector insertion and extraction shuffle kinds.

Definition at line 201 of file TargetTransformInfo.cpp.

Returns:
True if the intrinsic is a supported memory intrinsic. Info will contain additional information - whether the intrinsic may write or read to memory, volatility and the pointer. Info is undefined if false is returned.

Definition at line 268 of file TargetTransformInfo.cpp.

Get target-customized preferences for the generic loop unrolling transformation. The caller will initialize UP with the current target-independent defaults.

Definition at line 87 of file TargetTransformInfo.cpp.

Estimate the cost of a given IR user when lowered.

This can estimate the cost of either a ConstantExpr or Instruction when lowered. It has two primary advantages over the getOperationCost and getGEPCost above, and one significant disadvantage: it can only be used when the IR construct has already been formed.

The advantages are that it can inspect the SSA use graph to reason more accurately about the cost. For example, all-constant-GEPs can often be folded into a load or other instruction, but if they are used in some other context they may not be folded. This routine can distinguish such cases.

The returned cost is defined in terms of TargetCostConstants, see its comments for a detailed explanation of the cost values.

Definition at line 71 of file TargetTransformInfo.cpp.

Referenced by llvm::CodeMetrics::analyzeBasicBlock(), and ComputeSpeculationCost().

unsigned TargetTransformInfo::getVectorInstrCost ( unsigned  Opcode,
Type Val,
unsigned  Index = -1 
) const
Returns:
The expected cost of vector Insert and Extract. Use -1 to indicate that there is no information on the index value.

Definition at line 220 of file TargetTransformInfo.cpp.

Referenced by getScalarizationOverhead().

Return true if branch divergence exists.

Branch divergence has a significantly negative impact on GPU performance when threads in the same wavefront take different paths due to conditional branches.

Definition at line 75 of file TargetTransformInfo.cpp.

Return true if the hardware has a fast square-root instruction.

Definition at line 159 of file TargetTransformInfo.cpp.

Handle the invalidation of this information.

When used as a result of TargetIRAnalysis this method will be called when the function this was computed for changes. When it returns false, the information is preserved across those changes.

Definition at line 87 of file TargetTransformInfo.h.

Return true if the specified immediate is legal add immediate, that is the target has add instructions which can add a register with the immediate without having to materialize the immediate into a register.

Definition at line 92 of file TargetTransformInfo.cpp.

bool TargetTransformInfo::isLegalAddressingMode ( Type Ty,
GlobalValue BaseGV,
int64_t  BaseOffset,
bool  HasBaseReg,
int64_t  Scale 
) const

Return true if the addressing mode represented by AM is legal for this target, for a load/store of the specified type. The type may be VoidTy, in which case only return true if the addressing mode is legal for a load/store of any legal type. TODO: Handle pre/postinc as well.

Definition at line 100 of file TargetTransformInfo.cpp.

Referenced by isAddFoldable(), isAMCompletelyFolded(), and isGEPFoldable().

Return true if the specified immediate is legal icmp immediate, that is the target has icmp instructions which can compare a register against the immediate without having to materialize the immediate into a register.

Definition at line 96 of file TargetTransformInfo.cpp.

Referenced by isAMCompletelyFolded().

bool TargetTransformInfo::isLegalMaskedLoad ( Type DataType,
int  Consecutive 
) const

Definition at line 113 of file TargetTransformInfo.cpp.

bool TargetTransformInfo::isLegalMaskedStore ( Type DataType,
int  Consecutive 
) const

Return true if the target works with masked instruction AVX2 allows masks for consecutive load and store for i32 and i64 elements. AVX-512 architecture will also allow masks for non-consecutive memory accesses.

Definition at line 108 of file TargetTransformInfo.cpp.

Test whether calls to a function lower to actual program function calls.

The idea is to test whether the program is likely to require a 'call' instruction or equivalent in order to call the given function.

FIXME: It's not clear that this is a good or useful query API. Client's should probably move to simpler cost metrics using the above. Alternatively, we could split the cost interface into distinct code-size and execution-speed costs. This would allow modelling the core of this query more accurately as a call is a single small instruction, but incurs significant execution cost.

Definition at line 83 of file TargetTransformInfo.cpp.

Referenced by llvm::CodeMetrics::analyzeBasicBlock().

Return true if it is profitable to hoist instruction in the then/else to before if.

Definition at line 130 of file TargetTransformInfo.cpp.

Referenced by HoistThenElseCodeToIf().

Returns whether V is a source of divergence.

This function provides the target-dependent information for the target-independent DivergenceAnalysis. DivergenceAnalysis first builds the dependency graph, and then runs the reachability algorithm starting with the sources of divergence.

Definition at line 79 of file TargetTransformInfo.cpp.

bool TargetTransformInfo::isTruncateFree ( Type Ty1,
Type Ty2 
) const

Return true if it's free to truncate a value of type Ty1 to type Ty2. e.g. On x86 it's free to truncate a i32 value in register EAX to i16 by referencing its sub-register AX.

Definition at line 126 of file TargetTransformInfo.cpp.

Referenced by llvm::SCEVExpander::replaceCongruentIVs().

Return true if this type is legal.

Definition at line 134 of file TargetTransformInfo.cpp.

Referenced by ShouldBuildLookupTable().

TargetTransformInfo & TargetTransformInfo::operator= ( TargetTransformInfo &&  RHS)

Definition at line 44 of file TargetTransformInfo.cpp.

Return true if switches should be turned into lookup tables for the target.

Definition at line 146 of file TargetTransformInfo.cpp.

Referenced by SwitchToLookupTable().


The documentation for this class was generated from the following files: