llvm::LegalizerHelper

Helper that rewrites illegal Generic MIR into legal sequences.

Synopsis

Declared in <llvm/CodeGen/GlobalISel/LegalizerHelper.h>

class LegalizerHelper;

Types

NameDescription
ShiftParams Cached registers used while emitting multi-way shift parts.

Enums

NameDescription
LegalizeResult Result of attempting to legalize a single instruction.

Member Functions

NameDescription
LegalizerHelper [constructor]Constructors
bitcast Legalize an instruction by replacing the value type.
bitcastConcatVector Bitcast-legalize a vector concatenation.
bitcastDst Insert a G_BITCAST after a def operand of MI.
bitcastExtractSubvector Bitcast-legalize an extract-subvector.
bitcastExtractVectorElt Perform Bitcast legalize action on G_EXTRACT_VECTOR_ELT.
bitcastInsertSubvector Bitcast-legalize an insert-subvector.
bitcastInsertVectorElt Perform Bitcast legalize action on G_INSERT_VECTOR_ELT.
bitcastShuffleVector Bitcast-legalize a vector shuffle.
bitcastSrc Insert a G_BITCAST before a use operand of MI.
buildConstantShiftPart Generate one output part for a constant multi-way shift.
buildVariableShiftPart Generate one shift part with carry for a variable multi-way shift.
coerceToInteger Cast Val to an integer LLT of the same size when needed.
conversionLibcall Lower a conversion instruction by emitting a conversion libcall.
createAtomicLibcall Lower an atomic instruction by emitting an atomic libcall.
createLibcall createLibcall overloads
createMemLibcall Create a libcall to memcpy et al.
createStackStoreLoad Store Val to a stack temporary and reload it as Res.
createStackTemporary Create a stack temporary of Bytes bytes with Alignment.
equalizeVectorShuffleLengths Equalize source and destination vector sizes of G_SHUFFLE_VECTOR.
fewerElementsBitcast Split a bitcast while keeping source and destination sizes equal.
fewerElementsVector Legalize a vector by splitting it into fewer-element components.
fewerElementsVectorExtractInsertVectorElt Split extract/insert vector element into fewer-element pieces.
fewerElementsVectorMerge Split a merge into fewer-element merge pieces.
fewerElementsVectorMultiEltType Split MI into fewer-element operations of size NumElts.
fewerElementsVectorPhi Split a vector PHI into fewer-element PHI pieces.
fewerElementsVectorReductions Split a vector reduction into fewer-element reductions.
fewerElementsVectorSeqReductions Split a sequential vector reduction into fewer-element pieces.
fewerElementsVectorShuffle Split a shuffle into fewer-element shuffle pieces.
fewerElementsVectorUnmergeValues Split an unmerge into fewer-element unmerge pieces.
getDynStackAllocTargetPtr Compute the target pointer for a dynamic stack allocation.
getLegalizerInfo Expose LegalizerInfo so the clients can re-use.
getLibcallLoweringInfo Return the optional libcall lowering info, if configured.
getStackTemporaryAlignment Return the alignment to use for a stack temporary of type Type.
getTargetLowering Return the TargetLowering used for target-specific hooks.
getValueTracking Return the optional value-tracking analysis, if configured.
getVectorElementPointer Get a pointer to vector element Index in memory.
legalizeInstrStep Replace MI by a sequence of legal instructions.
libcall Legalize an instruction by emitting a runtime library call instead.
lower Legalize an instruction by splitting it into simpler parts.
lowerAbsDiffToMinMax Lower absolute difference using min and max.
lowerAbsDiffToSelect Lower absolute difference using a select sequence.
lowerAbsToAddXor Lower absolute value using add and xor.
lowerAbsToCNeg Lower absolute value using conditional negation.
lowerAbsToMaxNeg Lower absolute value using max and negation.
lowerAddSubSatToAddoSubo Lower saturating add/sub using overflow arithmetic.
lowerAddSubSatToMinMax Lower saturating add/sub using min/max clamping.
lowerBitCount Lower a bit-counting instruction into simpler operations.
lowerBitcast Lower G_BITCAST into equivalent legal operations.
lowerBitreverse Lower G_BITREVERSE into target-legal operations.
lowerBswap Lower G_BSWAP into target-legal operations.
lowerConstant Lower G_CONSTANT into target-legal materialization.
lowerDIVREM Lower a divrem into separate divide and remainder.
lowerDynStackAlloc Lower a dynamic stack allocation into stack-pointer updates.
lowerEXT Lower an integer extend into target-legal operations.
lowerExtract Lower G_EXTRACT into target-legal operations.
lowerExtractInsertVectorElt Lower extract/insert vector element into legal operations.
lowerFAbs Lower G_FABS into target-legal operations.
lowerFConstant Lower G_FCONSTANT into target-legal materialization.
lowerFCopySign Lower G_FCOPYSIGN into target-legal bit operations.
lowerFFloor Lower G_FFLOOR into target-legal operations.
lowerFMODF Lower G_FMODF into target-legal operations.
lowerFMad Lower G_FMAD into a multiply and add sequence.
lowerFMinNumMaxNum Lower G_FMINNUM/G_FMAXNUM into target-legal operations.
lowerFMinimumMaximum Lower G_FMINIMUM/G_FMAXIMUM into target-legal operations.
lowerFPEXT Lower G_FPEXT into target-legal operations.
lowerFPEXT_BF16 Lower a bfloat16 floating-point extend.
lowerFPExtAndTruncMem Lower float extend/truncate via memory round-trip when needed.
lowerFPOWI Lower G_FPOWI into target-legal operations.
lowerFPTOINT_SAT Lower a saturating float-to-integer conversion.
lowerFPTOSI Lower G_FPTOSI into target-legal operations.
lowerFPTOUI Lower G_FPTOUI into target-legal operations.
lowerFPTRUNC Lower G_FPTRUNC into target-legal operations.
lowerFPTRUNC_F32_TO_BF16 Lower f32-to-bf16 floating-point truncation.
lowerFPTRUNC_F64_TO_BF16 Lower f64-to-bf16 floating-point truncation.
lowerFPTRUNC_F64_TO_F16 Lower f64-to-f16 floating-point truncation.
lowerFunnelShift Lower a funnel shift into target-legal operations.
lowerFunnelShiftAsShifts Lower a funnel shift into a sequence of ordinary shifts.
lowerFunnelShiftWithInverse Lower a funnel shift using the inverse funnel-shift opcode.
lowerISFPCLASS Lower G_ISFPCLASS into target-legal comparisons.
lowerInsert Lower G_INSERT into target-legal operations.
lowerIntrinsicRound Lower an intrinsic round into target-legal operations.
lowerLoad Lower a generic load into legal memory operations.
lowerMemCpyFamily lowerMemCpyFamily overloads
lowerMergeValues Lower G_MERGE_VALUES into target-legal operations.
lowerMinMax Lower integer min/max into compares and selects.
lowerMulfix Lower a fixed-point multiply into extend, multiply, and shift.
lowerReadWriteRegister Lower read/write register intrinsics into legal operations.
lowerRotate Lower a rotate into target-legal operations.
lowerRotateWithReverseRotate Lower a rotate using the reverse rotate direction.
lowerRoundInexactToOdd Round Op inexactly to odd and return a value of ResultTy.
lowerSADDE Lower signed add-with-carry-in into legal operations.
lowerSADDO_SSUBO Lower signed add/sub-with-overflow into legal operations.
lowerSITOFP Lower G_SITOFP into target-legal operations.
lowerSMULH_UMULH Lower high-half multiply into wider multiply and extract.
lowerSSUBE Lower signed sub-with-carry-in into legal operations.
lowerSelect Lower G_SELECT into target-legal operations.
lowerShlSat Lower a saturating left shift into legal operations.
lowerShuffleVector Lower G_SHUFFLE_VECTOR into target-legal operations.
lowerStackRestore Lower G_STACKRESTORE into a stack-pointer restore.
lowerStackSave Lower G_STACKSAVE into a stack-pointer copy.
lowerStore Lower a generic store into legal memory operations.
lowerTRUNC Lower an integer truncate into target-legal operations.
lowerThreewayCompare Lower a three-way compare into target-legal comparisons.
lowerTruncSat Lower a saturating truncate into legal operations.
lowerU64ToF32BitOps Lower uint64-to-float conversion using bit operations to f32.
lowerU64ToF32WithSITOFP Lower uint64-to-float conversion via signed integer-to-float.
lowerU64ToF64BitFloatOps Lower uint64-to-double conversion using floating-point bit ops.
lowerUITOFP Lower G_UITOFP into target-legal operations.
lowerUnmergeValues Lower G_UNMERGE_VALUES into target-legal operations.
lowerVAArg Lower G_VAARG into target-legal operations.
lowerVECTOR_COMPRESS Lower G_VECTOR_COMPRESS into target-legal operations.
lowerVectorReduction Lower a vector reduction into target-legal operations.
moreElementsVector Legalize a vector by increasing the number of elements.
moreElementsVectorDst Produce a wider-vector def and extract the original elements in place.
moreElementsVectorPhi Widen a vector PHI to a type with more elements.
moreElementsVectorShuffle Widen a shuffle to operate on vectors with more elements.
moreElementsVectorSrc Pad a use operand of MI with undefined high vector elements.
narrowScalar Legalize an instruction by reducing the width of the underlying scalar type.
narrowScalarAddSub Narrow a scalar add or subtract into smaller-type operations.
narrowScalarBasic Narrow a basic scalar arithmetic or bitwise instruction.
narrowScalarCTLS Narrow a scalar CTLZ/CTTZ-style count into smaller-type operations.
narrowScalarCTLZ Narrow a scalar CTLZ into smaller-type operations.
narrowScalarCTPOP Narrow a scalar CTPOP into smaller-type operations.
narrowScalarCTTZ Narrow a scalar CTTZ into smaller-type operations.
narrowScalarDst Narrow a def operand of MI and extend the result back in place.
narrowScalarExt Narrow a scalar extend into smaller-type operations.
narrowScalarExtract Narrow a scalar extract into smaller-type operations.
narrowScalarFLDEXP Narrow a scalar FLDEXP into smaller-type operations.
narrowScalarFPTOI Narrow a floating-point to integer conversion.
narrowScalarInsert Narrow a scalar insert into smaller-type operations.
narrowScalarMul Narrow a scalar multiply into smaller-type operations.
narrowScalarSelect Narrow a scalar select into smaller-type operations.
narrowScalarShift Narrow a scalar shift into operations on a smaller type.
narrowScalarShiftByConstant Narrow a scalar shift by a known constant amount.
narrowScalarShiftByConstantMultiway Narrow a constant shift with direct multi-way indexing.
narrowScalarShiftMultiway Split a wide shift into target-sized parts in one step.
narrowScalarSrc Truncate a use operand of MI to NarrowTy in place.
reduceLoadStoreWidth Reduce the data width of a load or store instruction.
scalarizeVectorBooleanStore Scalarize a store of a boolean vector.
simpleLibcall Emit a simple libcall where all operands share the same type.
widenScalar Legalize an instruction by performing it on a wider scalar type.
widenScalarDst Widen a def operand of MI and truncate the result back in place.
widenScalarDstUsingFPTrunc Widen a def operand of MI and truncate it with G_FPTRUNC.
widenScalarSrc Extend a use operand of MI to WideTy in place.
widenScalarSrcUsingFPExt Extend a use operand of MI with G_FPEXT in place.

Data Members

NameDescription
MIRBuilder Expose MIRBuilder so clients can set their own RecordInsertInstruction functions
Observer To keep track of changes made by the LegalizerHelper.