NVPTXISelLowering.cpp File Reference

#include "NVPTXISelLowering.h"
#include "MCTargetDesc/NVPTXBaseInfo.h"
#include "NVPTX.h"
#include "NVPTXISelDAGToDAG.h"
#include "NVPTXSelectionDAGInfo.h"
#include "NVPTXSubtarget.h"
#include "NVPTXTargetMachine.h"
#include "NVPTXTargetObjectFile.h"
#include "NVPTXUtilities.h"
#include "llvm/ADT/APFloat.h"
#include "llvm/ADT/APInt.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/StringRef.h"
#include "llvm/CodeGen/Analysis.h"
#include "llvm/CodeGen/ISDOpcodes.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineJumpTableInfo.h"
#include "llvm/CodeGen/MachineMemOperand.h"
#include "llvm/CodeGen/SDPatternMatch.h"
#include "llvm/CodeGen/SelectionDAG.h"
#include "llvm/CodeGen/SelectionDAGNodes.h"
#include "llvm/CodeGen/TargetCallingConv.h"
#include "llvm/CodeGen/TargetLowering.h"
#include "llvm/CodeGen/ValueTypes.h"
#include "llvm/CodeGenTypes/MachineValueType.h"
#include "llvm/IR/Argument.h"
#include "llvm/IR/Attributes.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/DiagnosticInfo.h"
#include "llvm/IR/FPEnv.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/GlobalValue.h"
#include "llvm/IR/IRBuilder.h"
#include "llvm/IR/Instruction.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/IntrinsicsNVPTX.h"
#include "llvm/IR/Module.h"
#include "llvm/IR/Type.h"
#include "llvm/IR/Value.h"
#include "llvm/Support/Alignment.h"
#include "llvm/Support/AtomicOrdering.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/CodeGen.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/KnownBits.h"
#include "llvm/Support/NVPTXAddrSpace.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetOptions.h"
#include <algorithm>
#include <cassert>
#include <cmath>
#include <cstdint>
#include <iterator>
#include <optional>
#include <string>
#include <tuple>
#include <utility>
#include <vector>

Go to the source code of this file.

Macros
#define	DEBUG_TYPE   "nvptx-lower"
#define	TCGEN05_LD_RED_INTR(SHAPE, NUM, TYPE)
#define	TCGEN05_LD_RED_INST(SHAPE, NUM, TYPE)

Enumerations
enum	OperandSignedness { Signed = 0 , Unsigned , Unknown }

Functions
static bool	IsPTXVectorType (MVT VT)
static std::optional< std::pair< unsigned int, MVT > >	getVectorLoweringShape (EVT VectorEVT, const NVPTXSubtarget &STI, unsigned AddressSpace)
static void	ComputePTXValueVTs (const TargetLowering &TLI, const DataLayout &DL, LLVMContext &Ctx, CallingConv::ID CallConv, Type *Ty, SmallVectorImpl< EVT > &ValueVTs, SmallVectorImpl< uint64_t > &Offsets, uint64_t StartingOffset=0)
	ComputePTXValueVTs - For the given Type Ty, returns the set of primitive legal-ish MVTs that compose it.
static EVT	getVectorizedVT (EVT VT, unsigned N, LLVMContext &C)
static SDValue	getExtractVectorizedValue (SDValue V, unsigned I, EVT VT, const SDLoc &dl, SelectionDAG &DAG)
template<typename T>
static SDValue	getBuildVectorizedValue (unsigned N, const SDLoc &dl, SelectionDAG &DAG, T GetElement)
static EVT	promoteScalarIntegerPTX (const EVT VT)
	PromoteScalarIntegerPTX Used to make sure the arguments/returns are suitable for passing and promote them to a larger size if they're not.
template<typename T>
static unsigned	canMergeParamLoadStoresStartingAt (unsigned Idx, uint32_t AccessSize, const SmallVectorImpl< EVT > &ValueVTs, const SmallVectorImpl< T > &Offsets, Align ParamAlignment)
template<typename T>
static SmallVector< unsigned, 16 >	VectorizePTXValueVTs (const SmallVectorImpl< EVT > &ValueVTs, const SmallVectorImpl< T > &Offsets, Align ParamAlignment, bool IsVAArg=false)
static bool	shouldConvertToIndirectCall (const CallBase *CB, const GlobalAddressSDNode *Func)
static MachinePointerInfo	refinePtrAS (SDValue &Ptr, SelectionDAG &DAG, const DataLayout &DL, const TargetLowering &TL)
static ISD::NodeType	getExtOpcode (const ISD::ArgFlagsTy &Flags)
static SDValue	correctParamType (SDValue V, EVT ExpectedVT, ISD::ArgFlagsTy Flags, SelectionDAG &DAG, SDLoc dl)
static SDValue	getPRMT (SDValue A, SDValue B, SDValue Selector, SDLoc DL, SelectionDAG &DAG, unsigned Mode=NVPTX::PTXPrmtMode::NONE)
static SDValue	getPRMT (SDValue A, SDValue B, uint64_t Selector, SDLoc DL, SelectionDAG &DAG, unsigned Mode=NVPTX::PTXPrmtMode::NONE)
static SDValue	buildTreeReduction (const SmallVector< SDValue > &Elements, EVT EltTy, ArrayRef< std::pair< unsigned, unsigned > > Ops, const SDLoc &DL, const SDNodeFlags Flags, SelectionDAG &DAG)
	Reduces the elements using the scalar operations provided.
static ISD::NodeType	getScalarOpcodeForReduction (unsigned ReductionOpcode)
static std::optional< unsigned >	getScalar3OpcodeForReduction (unsigned ReductionOpcode)
	Get 3-input scalar reduction opcode.
static SDValue	PromoteBinOpToF32 (SDNode *N, SelectionDAG &DAG)
static SDValue	LowerVectorArith (SDValue Op, SelectionDAG &DAG)
static SDValue	lowerTcgen05St (SDValue Op, SelectionDAG &DAG, bool hasOffset=false)
static SDValue	lowerBSWAP (SDValue Op, SelectionDAG &DAG)
static unsigned	getTcgen05MMADisableOutputLane (unsigned IID)
static SDValue	LowerTcgen05MMADisableOutputLane (SDValue Op, SelectionDAG &DAG)
static std::optional< std::pair< SDValue, SDValue > >	lowerTcgen05Ld (SDNode *N, SelectionDAG &DAG, bool HasOffset=false)
static SDValue	reportInvalidTensormapReplaceUsage (SDValue Op, SelectionDAG &DAG, unsigned Val)
static SDValue	lowerTensormapReplaceElemtype (SDValue Op, SelectionDAG &DAG)
static SDValue	lowerTensormapReplaceSwizzleMode (SDValue Op, SelectionDAG &DAG)
static SDValue	lowerIntrinsicVoid (SDValue Op, SelectionDAG &DAG)
static SDValue	LowerClusterLaunchControlQueryCancel (SDValue Op, SelectionDAG &DAG)
static SDValue	lowerCvtRSIntrinsics (SDValue Op, SelectionDAG &DAG)
static SDValue	lowerPrmtIntrinsic (SDValue Op, SelectionDAG &DAG)
static unsigned	getTcgen05LdRedID (Intrinsic::ID IID)
static std::optional< std::tuple< SDValue, SDValue, SDValue > >	lowerTcgen05LdRed (SDNode *N, SelectionDAG &DAG)
static SDValue	lowerIntrinsicWChain (SDValue Op, SelectionDAG &DAG)
static SDValue	lowerIntrinsicWOChain (SDValue Op, SelectionDAG &DAG)
static SDValue	lowerCTLZCTPOP (SDValue Op, SelectionDAG &DAG)
static SDValue	expandFSH64 (SDValue A, SDValue B, SDValue ShiftAmount, SDLoc DL, unsigned Opcode, SelectionDAG &DAG)
static SDValue	lowerFSH (SDValue Op, SelectionDAG &DAG)
static SDValue	lowerROT (SDValue Op, SelectionDAG &DAG)
static SDValue	lowerFREM (SDValue Op, SelectionDAG &DAG)
static SDValue	lowerSELECT (SDValue Op, SelectionDAG &DAG)
static SDValue	lowerMSTORE (SDValue Op, SelectionDAG &DAG)
static std::pair< MemSDNode *, uint32_t >	convertMLOADToLoadWithUsedBytesMask (MemSDNode *N, SelectionDAG &DAG, const NVPTXSubtarget &STI)
static std::optional< std::pair< SDValue, SDValue > >	replaceLoadVector (SDNode *N, SelectionDAG &DAG, const NVPTXSubtarget &STI)
	replaceLoadVector - Convert vector loads into multi-output scalar loads.
static void	replaceLoadVector (SDNode *N, SelectionDAG &DAG, SmallVectorImpl< SDValue > &Results, const NVPTXSubtarget &STI)
static SDValue	lowerLoadVector (SDNode *N, SelectionDAG &DAG, const NVPTXSubtarget &STI)
static SDValue	lowerLOADi1 (LoadSDNode *LD, SelectionDAG &DAG)
static SDValue	lowerSTOREVector (SDValue Op, SelectionDAG &DAG, const NVPTXSubtarget &STI)
static bool	isConstZero (const SDValue &Operand)
static SDValue	PerformADDCombineWithOperands (SDNode *N, SDValue N0, SDValue N1, TargetLowering::DAGCombinerInfo &DCI)
	PerformADDCombineWithOperands - Try DAG combinations for an ADD with operands N0 and N1.
static SDValue	PerformFADDCombineWithOperands (SDNode *N, SDValue N0, SDValue N1, TargetLowering::DAGCombinerInfo &DCI, CodeGenOptLevel OptLevel)
static SDValue	combineUnpackingMovIntoLoad (SDNode *N, TargetLowering::DAGCombinerInfo &DCI)
	Fold unpacking movs into a load by increasing the number of return values.
static SDValue	combinePackingMovIntoStore (SDNode *N, TargetLowering::DAGCombinerInfo &DCI, unsigned Front, unsigned Back)
	Fold packing movs into a store.
static SDValue	combineSTORE (SDNode *N, TargetLowering::DAGCombinerInfo &DCI, const NVPTXSubtarget &STI)
static SDValue	combineLOAD (SDNode *N, TargetLowering::DAGCombinerInfo &DCI, const NVPTXSubtarget &STI)
static SDValue	PerformADDCombine (SDNode *N, TargetLowering::DAGCombinerInfo &DCI, CodeGenOptLevel OptLevel)
	PerformADDCombine - Target-specific dag combine xforms for ISD::ADD.
static bool	isNonCoalescableBuildVector (const SDValue &BV)
	Check if a v2f32 BUILD_VECTOR provably packs values from non-adjacent register pairs (non-coalescable).
static SDValue	PerformScalarizeV2F32Op (SDNode *N, TargetLowering::DAGCombinerInfo &DCI)
	Scalarize a v2f32 arithmetic node (FADD, FMUL, FSUB, FMA) when at least one operand is a BUILD_VECTOR that repacks values from non-adjacent register pairs.
static SDValue	PerformFADDCombine (SDNode *N, TargetLowering::DAGCombinerInfo &DCI, CodeGenOptLevel OptLevel)
	PerformFADDCombine - Target-specific dag combine xforms for ISD::FADD.
static unsigned	getMinMax3Opcode (unsigned MinMax2Opcode)
	Get 3-input version of a 2-input min/max opcode.
static SDValue	PerformFMinMaxCombine (SDNode *N, TargetLowering::DAGCombinerInfo &DCI, unsigned PTXVersion, unsigned SmVersion)
	PerformFMinMaxCombine - Combine (fmaxnum (fmaxnum a, b), c) into (fmaxnum3 a, b, c).
static SDValue	PerformREMCombine (SDNode *N, TargetLowering::DAGCombinerInfo &DCI, CodeGenOptLevel OptLevel)
static SDValue	combineMulWide (SDNode *N, TargetLowering::DAGCombinerInfo &DCI, CodeGenOptLevel OptLevel)
static bool	IsMulWideOperandDemotable (SDValue Op, unsigned OptSize, OperandSignedness &S)
	IsMulWideOperandDemotable - Checks if the provided DAG node is an operand that can be demoted to OptSize bits without loss of information.
static bool	AreMulWideOperandsDemotable (SDValue LHS, SDValue RHS, unsigned OptSize, bool &IsSigned)
	AreMulWideOperandsDemotable - Checks if the given LHS and RHS operands can be demoted to OptSize bits without loss of information.
static SDValue	TryMULWIDECombine (SDNode *N, TargetLowering::DAGCombinerInfo &DCI)
	TryMULWIDECombine - Attempt to replace a multiply of M bits with a multiply of M/2 bits that produces an M-bit result (i.e.
static bool	isConstOne (const SDValue &Operand)
static SDValue	matchMADConstOnePattern (SDValue Add)
static SDValue	combineMADConstOne (SDValue X, SDValue Add, EVT VT, SDLoc DL, TargetLowering::DAGCombinerInfo &DCI)
static SDValue	combineMulSelectConstOne (SDValue X, SDValue Select, EVT VT, SDLoc DL, TargetLowering::DAGCombinerInfo &DCI)
static SDValue	PerformMULCombineWithOperands (SDNode *N, SDValue N0, SDValue N1, TargetLowering::DAGCombinerInfo &DCI)
static SDValue	PerformMULCombine (SDNode *N, TargetLowering::DAGCombinerInfo &DCI, CodeGenOptLevel OptLevel)
	PerformMULCombine - Runs PTX-specific DAG combine patterns on MUL nodes.
static SDValue	PerformSHLCombine (SDNode *N, TargetLowering::DAGCombinerInfo &DCI, CodeGenOptLevel OptLevel)
	PerformSHLCombine - Runs PTX-specific DAG combine patterns on SHL nodes.
static SDValue	PerformSETCCCombine (SDNode *N, TargetLowering::DAGCombinerInfo &DCI, unsigned int SmVersion)
static SDValue	PerformEXTRACTCombine (SDNode *N, TargetLowering::DAGCombinerInfo &DCI)
static SDValue	PerformSELECTShiftCombine (SDNode *N, TargetLowering::DAGCombinerInfo &DCI)
	Transform patterns like: (select (ugt shift_amt, BitWidth-1), 0, (srl/shl x, shift_amt)) (select (ult shift_amt, BitWidth), (srl/shl x, shift_amt), 0) Into: (NVPTXISD::SRL_CLAMP x, shift_amt) or (NVPTXISD::SHL_CLAMP x, shift_amt)
static SDValue	PerformVSELECTCombine (SDNode *N, TargetLowering::DAGCombinerInfo &DCI)
static SDValue	PerformBUILD_VECTORCombine (SDNode *N, TargetLowering::DAGCombinerInfo &DCI)
static SDValue	combineADDRSPACECAST (SDNode *N, TargetLowering::DAGCombinerInfo &DCI)
static APInt	getPRMTSelector (const APInt &Selector, unsigned Mode)
static APInt	computePRMT (APInt A, APInt B, APInt Selector, unsigned Mode)
static SDValue	combinePRMT (SDNode *N, TargetLowering::DAGCombinerInfo &DCI, CodeGenOptLevel OptLevel)
static SDValue	sinkProxyReg (SDValue R, SDValue Chain, TargetLowering::DAGCombinerInfo &DCI)
static unsigned	getF16SubOpc (Intrinsic::ID AddIntrinsicID)
static SDValue	combineF16AddWithNeg (SDNode *N, SelectionDAG &DAG, Intrinsic::ID AddIntrinsicID)
static SDValue	combineIntrinsicWOChain (SDNode *N, TargetLowering::DAGCombinerInfo &DCI, const NVPTXSubtarget &STI)
static SDValue	combineProxyReg (SDNode *N, TargetLowering::DAGCombinerInfo &DCI)
static void	ReplaceBITCAST (SDNode *Node, SelectionDAG &DAG, SmallVectorImpl< SDValue > &Results)
static void	ReplaceINTRINSIC_W_CHAIN (SDNode *N, SelectionDAG &DAG, SmallVectorImpl< SDValue > &Results)
static void	ReplaceCopyFromReg_128 (SDNode *N, SelectionDAG &DAG, SmallVectorImpl< SDValue > &Results)
static void	replaceProxyReg (SDNode *N, SelectionDAG &DAG, const TargetLowering &TLI, SmallVectorImpl< SDValue > &Results)
static void	replaceAtomicSwap128 (SDNode *N, SelectionDAG &DAG, const NVPTXSubtarget &STI, SmallVectorImpl< SDValue > &Results)
static void	computeKnownBitsForPRMT (const SDValue Op, KnownBits &Known, const SelectionDAG &DAG, unsigned Depth)
static void	computeKnownBitsForLoadV (const SDValue Op, KnownBits &Known)
static std::pair< APInt, APInt >	getPRMTDemandedBits (const APInt &SelectorVal, const APInt &DemandedBits)
static SDValue	canonicalizePRMTInput (SDValue Op, SelectionDAG &DAG)
static SDValue	simplifyDemandedBitsForPRMT (SDValue PRMT, const APInt &DemandedBits, SelectionDAG &DAG, const TargetLowering &TLI, unsigned Depth)

Variables
static cl::opt< bool >	sched4reg ("nvptx-sched4reg", cl::desc("NVPTX Specific: schedule for register pressue"), cl::init(false))
static cl::opt< unsigned >	FMAContractLevelOpt ("nvptx-fma-level", cl::Hidden, cl::desc("NVPTX Specific: FMA contraction (0: don't do it" " 1: do it 2: do it aggressively"), cl::init(2))
static cl::opt< NVPTX::DivPrecisionLevel >	UsePrecDivF32 ("nvptx-prec-divf32", cl::Hidden, cl::desc("NVPTX Specific: Override the precision of the lowering for f32 fdiv"), cl::values(clEnumValN(NVPTX::DivPrecisionLevel::Approx, "0", "Use div.approx"), clEnumValN(NVPTX::DivPrecisionLevel::Full, "1", "Use div.full"), clEnumValN(NVPTX::DivPrecisionLevel::IEEE754, "2", "Use IEEE Compliant F32 div.rnd if available (default)"), clEnumValN(NVPTX::DivPrecisionLevel::IEEE754_NoFTZ, "3", "Use IEEE Compliant F32 div.rnd if available, no FTZ")), cl::init(NVPTX::DivPrecisionLevel::IEEE754))
static cl::opt< bool >	UsePrecSqrtF32 ("nvptx-prec-sqrtf32", cl::Hidden, cl::desc("NVPTX Specific: 0 use sqrt.approx, 1 use sqrt.rn."), cl::init(true))
static cl::opt< bool >	UseApproxLog2F32 ("nvptx-approx-log2f32", cl::desc("NVPTX Specific: whether to use lg2.approx for log2"), cl::init(false))
	Whereas CUDA's implementation (see libdevice) uses ex2.approx for exp2(), it does NOT use lg2.approx for log2, so this is disabled by default.
static cl::opt< bool >	ForceMinByValParamAlign ("nvptx-force-min-byval-param-align", cl::Hidden, cl::desc("NVPTX Specific: force 4-byte minimal alignment for byval" " params of device functions."), cl::init(false))

Macro Definition Documentation

DEBUG_TYPE

#define DEBUG_TYPE   "nvptx-lower"

Definition at line 78 of file NVPTXISelLowering.cpp.

TCGEN05_LD_RED_INST

#define TCGEN05_LD_RED_INST	(		SHAPE,
			NUM,
			TYPE )

Value:

  NVPTXISD::TCGEN05_LD_RED_##SHAPE##_X##NUM##_##TYPE

Definition at line 3034 of file NVPTXISelLowering.cpp.

Referenced by getTcgen05LdRedID().

TCGEN05_LD_RED_INTR

#define TCGEN05_LD_RED_INTR	(		SHAPE,
			NUM,
			TYPE )

Value:

  Intrinsic::nvvm_tcgen05_ld_red_##SHAPE##_x##NUM##_##TYPE

Definition at line 3031 of file NVPTXISelLowering.cpp.

Referenced by getTcgen05LdRedID().

Enumeration Type Documentation

OperandSignedness

enum OperandSignedness

Enumerator
Signed
Unsigned
Unknown

Definition at line 6394 of file NVPTXISelLowering.cpp.

Function Documentation

AreMulWideOperandsDemotable()

bool AreMulWideOperandsDemotable ( SDValue LHS, SDValue RHS, unsigned OptSize, bool & IsSigned )

static

AreMulWideOperandsDemotable - Checks if the given LHS and RHS operands can be demoted to OptSize bits without loss of information.

If the operands contain a constant, it should appear as the RHS operand. The signedness of the operands is placed in IsSigned.

Definition at line 6430 of file NVPTXISelLowering.cpp.

References llvm::dyn_cast(), llvm::APInt::isIntN(), IsMulWideOperandDemotable(), llvm::APInt::isSignedIntN(), LHS, RHS, Signed, Unknown, and Unsigned.

Referenced by TryMULWIDECombine().

buildTreeReduction()

SDValue buildTreeReduction ( const SmallVector< SDValue > & Elements, EVT EltTy, ArrayRef< std::pair< unsigned, unsigned > > Ops, const SDLoc & DL, const SDNodeFlags Flags, SelectionDAG & DAG )

static

Reduces the elements using the scalar operations provided.

The operations are sorted descending in number of inputs they take. The flags on the original reduction operation will be propagated to each scalar operation. Nearby elements are grouped in tree reduction, unlike the shuffle reduction used in ExpandReductions and SelectionDAG.

Definition at line 1908 of file NVPTXISelLowering.cpp.

References AbstractManglingParser< Derived, Alloc >::Ops, assert(), DL, E(), llvm::SmallVectorTemplateCommon< T, typename >::empty(), llvm::SelectionDAG::getNode(), I, OpIdx, llvm::SmallVectorTemplateBase< T, bool >::push_back(), and llvm::ArrayRef< T >::slice().

canMergeParamLoadStoresStartingAt()

template<typename T>

unsigned canMergeParamLoadStoresStartingAt ( unsigned Idx, uint32_t AccessSize, const SmallVectorImpl< EVT > & ValueVTs, const SmallVectorImpl< T > & Offsets, Align ParamAlignment )

static

Definition at line 427 of file NVPTXISelLowering.cpp.

References llvm::EVT::getStoreSize(), and llvm::SmallVectorTemplateCommon< T, typename >::size().

Referenced by VectorizePTXValueVTs().

canonicalizePRMTInput()

SDValue canonicalizePRMTInput ( SDValue Op, SelectionDAG & DAG )

static

Definition at line 7800 of file NVPTXISelLowering.cpp.

References llvm::SelectionDAG::getConstant(), and SDValue().

Referenced by simplifyDemandedBitsForPRMT().

combineADDRSPACECAST()

SDValue combineADDRSPACECAST ( SDNode * N, TargetLowering::DAGCombinerInfo & DCI )

static

Definition at line 6874 of file NVPTXISelLowering.cpp.

References assert(), llvm::cast(), llvm::dyn_cast(), N, and SDValue().

combineF16AddWithNeg()

SDValue combineF16AddWithNeg ( SDNode * N, SelectionDAG & DAG, Intrinsic::ID AddIntrinsicID )

static

Definition at line 7041 of file NVPTXISelLowering.cpp.

References DL, llvm::ISD::FNEG, getF16SubOpc(), llvm::SelectionDAG::getNode(), llvm::SDValue::getOpcode(), llvm::SDValue::getOperand(), N, and SDValue().

Referenced by combineIntrinsicWOChain().

combineIntrinsicWOChain()

SDValue combineIntrinsicWOChain ( SDNode * N, TargetLowering::DAGCombinerInfo & DCI, const NVPTXSubtarget & STI )

static

Definition at line 7063 of file NVPTXISelLowering.cpp.

References combineF16AddWithNeg(), llvm::TargetLowering::DAGCombinerInfo::DAG, N, and SDValue().

combineLOAD()

SDValue combineLOAD ( SDNode * N, TargetLowering::DAGCombinerInfo & DCI, const NVPTXSubtarget & STI )

static

Definition at line 6108 of file NVPTXISelLowering.cpp.

References combineUnpackingMovIntoLoad(), llvm::TargetLowering::DAGCombinerInfo::DAG, llvm::TargetLowering::DAGCombinerInfo::isBeforeLegalize(), llvm::ISD::LOAD, lowerLoadVector(), and N.

combineMADConstOne()

SDValue combineMADConstOne ( SDValue X, SDValue Add, EVT VT, SDLoc DL, TargetLowering::DAGCombinerInfo & DCI )

static

Definition at line 6550 of file NVPTXISelLowering.cpp.

References llvm::ISD::ADD, llvm::Add, llvm::TargetLowering::DAGCombinerInfo::DAG, DL, llvm::SelectionDAG::getNode(), matchMADConstOnePattern(), llvm::ISD::MUL, Mul, SDValue(), X, and Y.

Referenced by PerformMULCombineWithOperands().

combineMulSelectConstOne()

SDValue combineMulSelectConstOne ( SDValue X, SDValue Select, EVT VT, SDLoc DL, TargetLowering::DAGCombinerInfo & DCI )

static

Definition at line 6561 of file NVPTXISelLowering.cpp.

References Cond, llvm::TargetLowering::DAGCombinerInfo::DAG, DL, llvm::SelectionDAG::getNode(), isConstOne(), matchMADConstOnePattern(), llvm::ISD::MUL, SDValue(), llvm::ISD::SELECT, Select, X, and Y.

Referenced by PerformMULCombineWithOperands().

combineMulWide()

SDValue combineMulWide ( SDNode * N, TargetLowering::DAGCombinerInfo & DCI, CodeGenOptLevel OptLevel )

static

Definition at line 6359 of file NVPTXISelLowering.cpp.

References llvm::TargetLowering::DAGCombinerInfo::DAG, DL, llvm::SelectionDAG::getConstant(), llvm::SelectionDAG::getNode(), llvm::EVT::getSizeInBits(), llvm::isa(), llvm::ISD::MUL, N, llvm::None, RHS, SDValue(), llvm::ISD::SHL, llvm::ISD::SIGN_EXTEND, and llvm::ISD::ZERO_EXTEND.

combinePackingMovIntoStore()

SDValue combinePackingMovIntoStore ( SDNode * N, TargetLowering::DAGCombinerInfo & DCI, unsigned Front, unsigned Back )

static

Fold packing movs into a store.

ex: v1: v2f16 = BUILD_VECTOR a:f16, b:f16 v2: v2f16 = BUILD_VECTOR c:f16, d:f16 StoreV2 v1, v2

...is turned into...

StoreV4 a, b, c, d

Definition at line 6013 of file NVPTXISelLowering.cpp.

References llvm::SmallVectorImpl< T >::append(), llvm::ISD::BITCAST, llvm::ISD::BUILD_VECTOR, llvm::cast(), llvm::TargetLowering::DAGCombinerInfo::DAG, llvm::ISD::FP_ROUND, llvm::SelectionDAG::getMemIntrinsicNode(), llvm::TargetLowering::DAGCombinerInfo::isAfterLegalizeDAG(), llvm::NVPTX::isPackedVectorTy(), llvm_unreachable, N, SDValue(), llvm::ISD::STORE, llvm::NVPTXISD::StoreV2, llvm::NVPTXISD::StoreV4, llvm::NVPTXISD::StoreV8, and llvm::ISD::TRUNCATE.

Referenced by combineSTORE().

combinePRMT()

SDValue combinePRMT ( SDNode * N, TargetLowering::DAGCombinerInfo & DCI, CodeGenOptLevel OptLevel )

static

Definition at line 6945 of file NVPTXISelLowering.cpp.

References computePRMT(), llvm::TargetLowering::DAGCombinerInfo::DAG, llvm::SelectionDAG::getConstant(), llvm::isa(), N, llvm::None, and SDValue().

combineProxyReg()

SDValue combineProxyReg ( SDNode * N, TargetLowering::DAGCombinerInfo & DCI )

static

Definition at line 7080 of file NVPTXISelLowering.cpp.

References llvm::ISD::LOAD, N, Reg, SDValue(), and sinkProxyReg().

combineSTORE()

SDValue combineSTORE ( SDNode * N, TargetLowering::DAGCombinerInfo & DCI, const NVPTXSubtarget & STI )

static

Definition at line 6093 of file NVPTXISelLowering.cpp.

References llvm::cast(), combinePackingMovIntoStore(), llvm::TargetLowering::DAGCombinerInfo::DAG, llvm::TargetLowering::DAGCombinerInfo::isBeforeLegalize(), lowerSTOREVector(), N, SDValue(), and llvm::ISD::STORE.

combineUnpackingMovIntoLoad()

SDValue combineUnpackingMovIntoLoad ( SDNode * N, TargetLowering::DAGCombinerInfo & DCI )

static

Fold unpacking movs into a load by increasing the number of return values.

ex: L: v2f16,ch = load

a: f16 = extractelt L:0, 0 b: f16 = extractelt L:0, 1 use(a, b)

...is turned into...

L: f16,f16,ch = LoadV2

use(L:0, L:1)

Definition at line 5899 of file NVPTXISelLowering.cpp.

References llvm::all_of(), llvm::TargetLowering::DAGCombinerInfo::isAfterLegalizeDAG(), llvm::NVPTX::isPackedVectorTy(), N, and SDValue().

Referenced by combineLOAD().

computeKnownBitsForLoadV()

void computeKnownBitsForLoadV ( const SDValue Op, KnownBits & Known )

static

Definition at line 7737 of file NVPTXISelLowering.cpp.

References assert(), llvm::cast(), llvm::KnownBits::getBitWidth(), llvm::NVPTXDAGToDAGISel::getFromTypeWidthForLoad(), llvm::APInt::setHighBits(), llvm::ISD::SEXTLOAD, and llvm::KnownBits::Zero.

Referenced by llvm::NVPTXTargetLowering::computeKnownBitsForTargetNode().

computeKnownBitsForPRMT()

void computeKnownBitsForPRMT ( const SDValue Op, KnownBits & Known, const SelectionDAG & DAG, unsigned Depth )

static

Definition at line 7706 of file NVPTXISelLowering.cpp.

References A(), llvm::KnownBits::ashr(), assert(), B(), llvm::SelectionDAG::computeKnownBits(), llvm::KnownBits::concat(), llvm::Depth, llvm::dyn_cast(), llvm::APInt::extractBits(), llvm::ConstantSDNode::getAPIntValue(), llvm::KnownBits::getBitWidth(), llvm::APInt::getHiBits(), llvm::APInt::getLoBits(), getPRMTSelector(), llvm::APInt::getZExtValue(), I, llvm::KnownBits::insertBits(), Mode, and llvm::seq().

Referenced by llvm::NVPTXTargetLowering::computeKnownBitsForTargetNode().

computePRMT()

APInt computePRMT ( APInt A, APInt B, APInt Selector, unsigned Mode )

static

Definition at line 6926 of file NVPTXISelLowering.cpp.

References A(), assert(), B(), llvm::APInt::extractBits(), llvm::APInt::getBitWidth(), llvm::APInt::getHiBits(), llvm::APInt::getLoBits(), getPRMTSelector(), llvm::APInt::getZExtValue(), I, Mode, and llvm::seq().

Referenced by combinePRMT().

ComputePTXValueVTs()

void ComputePTXValueVTs ( const TargetLowering & TLI, const DataLayout & DL, LLVMContext & Ctx, CallingConv::ID CallConv, Type * Ty, SmallVectorImpl< EVT > & ValueVTs, SmallVectorImpl< uint64_t > & Offsets, uint64_t StartingOffset = 0 )

static

ComputePTXValueVTs - For the given Type Ty, returns the set of primitive legal-ish MVTs that compose it.

Unlike ComputeValueVTs, this will legalize the types as required by the calling convention (with special handling for i8s). NOTE: This is a band-aid for code that expects ComputeValueVTs to return the same number of types as the Ins/Outs arrays in LowerFormalArguments, LowerCall, and LowerReturn.

Definition at line 303 of file NVPTXISelLowering.cpp.

References assert(), llvm::ComputeValueVTs(), DL, llvm::TargetLoweringBase::getNumRegistersForCallingConv(), llvm::TargetLoweringBase::getRegisterTypeForCallingConv(), llvm::MVT::getStoreSize(), I, llvm::SmallVectorTemplateBase< T, bool >::push_back(), llvm::seq(), and llvm::zip().

Referenced by llvm::NVPTXTargetLowering::LowerCall(), llvm::NVPTXTargetLowering::LowerFormalArguments(), and llvm::NVPTXTargetLowering::LowerReturn().

convertMLOADToLoadWithUsedBytesMask()

std::pair< MemSDNode *, uint32_t > convertMLOADToLoadWithUsedBytesMask ( MemSDNode * N, SelectionDAG & DAG, const NVPTXSubtarget & STI )

static

Definition at line 3626 of file NVPTXISelLowering.cpp.

References assert(), llvm::cast(), DL, llvm::SelectionDAG::getLoad(), llvm::SDValue::getNode(), llvm::SDValue::getOpcode(), llvm::EVT::getSizeInBits(), llvm::EVT::getVectorElementType(), llvm::NVPTXSubtarget::hasUsedBytesMaskPragma(), llvm::EVT::isVector(), N, llvm::ISD::POISON, llvm::reverse(), and llvm::ISD::UNDEF.

Referenced by replaceLoadVector().

correctParamType()

SDValue correctParamType ( SDValue V, EVT ExpectedVT, ISD::ArgFlagsTy Flags, SelectionDAG & DAG, SDLoc dl )

static

Definition at line 1376 of file NVPTXISelLowering.cpp.

References assert(), llvm::EVT::bitsGT(), llvm::EVT::bitsLT(), getExtOpcode(), llvm::SelectionDAG::getNode(), llvm::EVT::isInteger(), and llvm::ISD::TRUNCATE.

Referenced by llvm::NVPTXTargetLowering::LowerCall(), llvm::NVPTXTargetLowering::LowerFormalArguments(), and llvm::NVPTXTargetLowering::LowerReturn().

expandFSH64()

SDValue expandFSH64 ( SDValue A, SDValue B, SDValue ShiftAmount, SDLoc DL, unsigned Opcode, SelectionDAG & DAG )

static

Definition at line 3222 of file NVPTXISelLowering.cpp.

References A(), assert(), B(), DL, llvm::dyn_cast(), llvm::ISD::FSHL, llvm::SelectionDAG::getConstant(), llvm::SelectionDAG::getNode(), llvm::SDValue::getValue(), High, llvm::Low, SDValue(), and llvm::NVPTXISD::UNPACK_VECTOR.

Referenced by lowerFSH(), and lowerROT().

getBuildVectorizedValue()

template<typename T>

SDValue getBuildVectorizedValue ( unsigned N, const SDLoc & dl, SelectionDAG & DAG, T GetElement )

inlinestatic

Definition at line 369 of file NVPTXISelLowering.cpp.

References llvm::SelectionDAG::ExtractVectorElements(), llvm::SelectionDAG::getBuildVector(), llvm::SelectionDAG::getContext(), llvm::SDValue::getValueType(), llvm::EVT::getVectorVT(), I, llvm::EVT::isVector(), N, llvm::SmallVectorTemplateBase< T, bool >::push_back(), llvm::seq(), llvm::SmallVectorTemplateCommon< T, typename >::size(), and T.

Referenced by llvm::NVPTXTargetLowering::LowerCall(), and llvm::NVPTXTargetLowering::LowerReturn().

getExtOpcode()

ISD::NodeType getExtOpcode ( const ISD::ArgFlagsTy & Flags )

static

Definition at line 1368 of file NVPTXISelLowering.cpp.

References llvm::ISD::ANY_EXTEND, llvm::ISD::SIGN_EXTEND, and llvm::ISD::ZERO_EXTEND.

Referenced by correctParamType().

getExtractVectorizedValue()

SDValue getExtractVectorizedValue ( SDValue V, unsigned I, EVT VT, const SDLoc & dl, SelectionDAG & DAG )

static

Definition at line 352 of file NVPTXISelLowering.cpp.

References assert(), llvm::ISD::EXTRACT_SUBVECTOR, llvm::ISD::EXTRACT_VECTOR_ELT, llvm::SelectionDAG::getNode(), llvm::SelectionDAG::getVectorIdxConstant(), llvm::EVT::getVectorNumElements(), I, and llvm::EVT::isVector().

Referenced by llvm::NVPTXTargetLowering::LowerCall(), and llvm::NVPTXTargetLowering::LowerFormalArguments().

getF16SubOpc()

unsigned getF16SubOpc ( Intrinsic::ID AddIntrinsicID )

static

Definition at line 7027 of file NVPTXISelLowering.cpp.

References llvm_unreachable.

Referenced by combineF16AddWithNeg().

getMinMax3Opcode()

unsigned getMinMax3Opcode ( unsigned MinMax2Opcode )

static

Get 3-input version of a 2-input min/max opcode.

Definition at line 6278 of file NVPTXISelLowering.cpp.

References llvm::ISD::FMAXIMUM, llvm::ISD::FMAXIMUMNUM, llvm::ISD::FMAXNUM, llvm::ISD::FMINIMUM, llvm::ISD::FMINIMUMNUM, llvm::ISD::FMINNUM, and llvm_unreachable.

Referenced by PerformFMinMaxCombine().

getPRMT() [1/2]

SDValue getPRMT ( SDValue A, SDValue B, SDValue Selector, SDLoc DL, SelectionDAG & DAG, unsigned Mode = NVPTX::PTXPrmtMode::NONE )

static

Definition at line 1888 of file NVPTXISelLowering.cpp.

References A(), assert(), B(), DL, llvm::SelectionDAG::getConstant(), llvm::SelectionDAG::getNode(), llvm::SDValue::getValueType(), Mode, and llvm::NVPTX::PTXPrmtMode::NONE.

Referenced by getPRMT(), lowerBSWAP(), lowerPrmtIntrinsic(), PerformBUILD_VECTORCombine(), and simplifyDemandedBitsForPRMT().

getPRMT() [2/2]

SDValue getPRMT ( SDValue A, SDValue B, uint64_t Selector, SDLoc DL, SelectionDAG & DAG, unsigned Mode = NVPTX::PTXPrmtMode::NONE )

static

Definition at line 1897 of file NVPTXISelLowering.cpp.

References A(), B(), DL, llvm::SelectionDAG::getConstant(), getPRMT(), Mode, and llvm::NVPTX::PTXPrmtMode::NONE.

getPRMTDemandedBits()

std::pair< APInt, APInt > getPRMTDemandedBits ( const APInt & SelectorVal, const APInt & DemandedBits )

static

Definition at line 7774 of file NVPTXISelLowering.cpp.

References llvm::APInt::extractBits(), llvm::APInt::getHiBits(), llvm::APInt::getLoBits(), llvm::APInt::getZExtValue(), I, llvm::seq(), and llvm::APInt::setBit().

Referenced by simplifyDemandedBitsForPRMT().

getPRMTSelector()

APInt getPRMTSelector ( const APInt & Selector, unsigned Mode )

static

Definition at line 6893 of file NVPTXISelLowering.cpp.

References assert(), llvm::NVPTX::PTXPrmtMode::B4E, llvm::NVPTX::PTXPrmtMode::ECL, llvm::NVPTX::PTXPrmtMode::ECR, llvm::NVPTX::PTXPrmtMode::F4E, llvm::APInt::getBitWidth(), llvm::APInt::getZExtValue(), llvm_unreachable, Mode, llvm::NVPTX::PTXPrmtMode::NONE, llvm::NVPTX::PTXPrmtMode::RC16, llvm::NVPTX::PTXPrmtMode::RC8, S1, and llvm::APInt::trunc().

Referenced by computeKnownBitsForPRMT(), computePRMT(), and simplifyDemandedBitsForPRMT().

getScalar3OpcodeForReduction()

std::optional< unsigned > getScalar3OpcodeForReduction ( unsigned ReductionOpcode )

static

Get 3-input scalar reduction opcode.

Definition at line 1972 of file NVPTXISelLowering.cpp.

References llvm::ISD::VECREDUCE_FMAX, llvm::ISD::VECREDUCE_FMAXIMUM, llvm::ISD::VECREDUCE_FMIN, and llvm::ISD::VECREDUCE_FMINIMUM.

getScalarOpcodeForReduction()

ISD::NodeType getScalarOpcodeForReduction ( unsigned ReductionOpcode )

static

Definition at line 1955 of file NVPTXISelLowering.cpp.

References llvm::ISD::FMAXIMUM, llvm::ISD::FMAXNUM, llvm::ISD::FMINIMUM, llvm::ISD::FMINNUM, llvm_unreachable, llvm::ISD::VECREDUCE_FMAX, llvm::ISD::VECREDUCE_FMAXIMUM, llvm::ISD::VECREDUCE_FMIN, and llvm::ISD::VECREDUCE_FMINIMUM.

getTcgen05LdRedID()

unsigned getTcgen05LdRedID ( Intrinsic::ID IID )

static

Definition at line 3037 of file NVPTXISelLowering.cpp.

References F32, llvm_unreachable, TCGEN05_LD_RED_INST, and TCGEN05_LD_RED_INTR.

Referenced by lowerTcgen05LdRed().

getTcgen05MMADisableOutputLane()

unsigned getTcgen05MMADisableOutputLane ( unsigned IID )

static

Definition at line 2640 of file NVPTXISelLowering.cpp.

References llvm_unreachable.

Referenced by LowerTcgen05MMADisableOutputLane().

getVectorizedVT()

EVT getVectorizedVT ( EVT VT, unsigned N, LLVMContext & C )

static

Definition at line 343 of file NVPTXISelLowering.cpp.

References llvm::CallingConv::C, llvm::EVT::getScalarType(), llvm::EVT::getVectorNumElements(), llvm::EVT::getVectorVT(), llvm::EVT::isVector(), and N.

Referenced by llvm::NVPTXTargetLowering::LowerCall(), and llvm::NVPTXTargetLowering::LowerFormalArguments().

getVectorLoweringShape()

std::optional< std::pair< unsigned int, MVT > > getVectorLoweringShape ( EVT VectorEVT, const NVPTXSubtarget & STI, unsigned AddressSpace )

static

Definition at line 201 of file NVPTXISelLowering.cpp.

References llvm::EVT::getSimpleVT(), llvm::EVT::getSizeInBits(), llvm::MVT::getSizeInBits(), llvm::MVT::getVectorElementType(), llvm::MVT::getVectorNumElements(), llvm::MVT::getVectorVT(), llvm::NVPTXSubtarget::has256BitVectorLoadStore(), llvm::NVPTXSubtarget::hasF32x2Instructions(), llvm::EVT::isScalarInteger(), llvm::EVT::isSimple(), llvm::MVT::isVector(), and llvm::MVT::SimpleTy.

Referenced by lowerSTOREVector(), and replaceLoadVector().

isConstOne()

bool isConstOne ( const SDValue & Operand )

static

Definition at line 6532 of file NVPTXISelLowering.cpp.

References llvm::dyn_cast().

Referenced by combineMulSelectConstOne(), and matchMADConstOnePattern().

isConstZero()

bool isConstZero ( const SDValue & Operand )

static

Definition at line 5758 of file NVPTXISelLowering.cpp.

References llvm::dyn_cast().

Referenced by PerformADDCombineWithOperands().

IsMulWideOperandDemotable()

bool IsMulWideOperandDemotable ( SDValue Op, unsigned OptSize, OperandSignedness & S )

static

IsMulWideOperandDemotable - Checks if the provided DAG node is an operand that can be demoted to OptSize bits without loss of information.

The signedness of the operand, if determinable, is placed in S.

Definition at line 6403 of file NVPTXISelLowering.cpp.

References llvm::EVT::getFixedSizeInBits(), llvm::ISD::SIGN_EXTEND, llvm::ISD::SIGN_EXTEND_INREG, Signed, Unknown, Unsigned, and llvm::ISD::ZERO_EXTEND.

Referenced by AreMulWideOperandsDemotable().

isNonCoalescableBuildVector()

bool isNonCoalescableBuildVector ( const SDValue & BV )

static

Check if a v2f32 BUILD_VECTOR provably packs values from non-adjacent register pairs (non-coalescable).

Definition at line 6147 of file NVPTXISelLowering.cpp.

References llvm::ISD::BUILD_VECTOR, llvm::dyn_cast(), llvm::ISD::EXTRACT_VECTOR_ELT, llvm::SDValue::getOpcode(), llvm::SDValue::getOperand(), and llvm::SDValue::getValueType().

Referenced by PerformScalarizeV2F32Op().

IsPTXVectorType()

bool IsPTXVectorType ( MVT VT )

static

Definition at line 156 of file NVPTXISelLowering.cpp.

References llvm::MVT::SimpleTy.

Referenced by llvm::NVPTXTargetLowering::NVPTXTargetLowering(), and PerformEXTRACTCombine().

lowerBSWAP()

SDValue lowerBSWAP ( SDValue Op, SelectionDAG & DAG )

static

Definition at line 2602 of file NVPTXISelLowering.cpp.

References llvm::ISD::ANY_EXTEND, llvm::ISD::BITCAST, DL, llvm::SelectionDAG::getBitcast(), llvm::SelectionDAG::getConstant(), llvm::SelectionDAG::getNode(), getPRMT(), llvm::EVT::getSimpleVT(), llvm::SDValue::getValue(), llvm_unreachable, llvm::MVT::SimpleTy, llvm::ISD::TRUNCATE, and llvm::NVPTXISD::UNPACK_VECTOR.

Referenced by llvm::NVPTXTargetLowering::LowerOperation().

LowerClusterLaunchControlQueryCancel()

SDValue LowerClusterLaunchControlQueryCancel ( SDValue Op, SelectionDAG & DAG )

static

Definition at line 2907 of file NVPTXISelLowering.cpp.

References llvm::ISD::BITCAST, llvm::cast(), DL, llvm::ISD::EXTRACT_VECTOR_ELT, llvm::SelectionDAG::getIntPtrConstant(), llvm::SelectionDAG::getNode(), llvm_unreachable, N, and SDValue().

Referenced by lowerIntrinsicWOChain().

lowerCTLZCTPOP()

SDValue lowerCTLZCTPOP ( SDValue Op, SelectionDAG & DAG )

static

Definition at line 3212 of file NVPTXISelLowering.cpp.

References assert(), DL, llvm::SelectionDAG::getNode(), llvm::SDNodeFlags::NonNeg, and llvm::ISD::ZERO_EXTEND.

Referenced by llvm::NVPTXTargetLowering::LowerOperation().

lowerCvtRSIntrinsics()

SDValue lowerCvtRSIntrinsics ( SDValue Op, SelectionDAG & DAG )

static

Definition at line 2947 of file NVPTXISelLowering.cpp.

References AbstractManglingParser< Derived, Alloc >::Ops, DL, llvm::ISD::EXTRACT_VECTOR_ELT, llvm::SelectionDAG::getConstant(), llvm::SelectionDAG::getIntPtrConstant(), llvm::SelectionDAG::getNode(), llvm_unreachable, and N.

Referenced by lowerIntrinsicWOChain().

lowerFREM()

SDValue lowerFREM ( SDValue Op, SelectionDAG & DAG )

static

Definition at line 3277 of file NVPTXISelLowering.cpp.

References llvm::SDNodeFlags::AllowContract, DL, llvm::ISD::FABS, llvm::ISD::FDIV, llvm::ISD::FMUL, llvm::ISD::FSUB, llvm::ISD::FTRUNC, llvm::SelectionDAG::getConstantFP(), llvm::APFloat::getInf(), llvm::SelectionDAG::getNode(), llvm::SelectionDAG::getSelect(), llvm::SelectionDAG::getSetCC(), Mul, llvm::ISD::SETEQ, llvm::Sub, X, and Y.

Referenced by llvm::NVPTXTargetLowering::LowerOperation().

lowerFSH()

SDValue lowerFSH ( SDValue Op, SelectionDAG & DAG )

static

Definition at line 3266 of file NVPTXISelLowering.cpp.

References expandFSH64().

Referenced by llvm::NVPTXTargetLowering::LowerOperation().

lowerIntrinsicVoid()

SDValue lowerIntrinsicVoid ( SDValue Op, SelectionDAG & DAG )

static

Definition at line 2825 of file NVPTXISelLowering.cpp.

References llvm::cast(), llvm::SDValue::getNode(), LowerTcgen05MMADisableOutputLane(), lowerTcgen05St(), lowerTensormapReplaceElemtype(), lowerTensormapReplaceSwizzleMode(), and N.

Referenced by llvm::NVPTXTargetLowering::LowerOperation().

lowerIntrinsicWChain()

SDValue lowerIntrinsicWChain ( SDValue Op, SelectionDAG & DAG )

static

Definition at line 3146 of file NVPTXISelLowering.cpp.

References llvm::SelectionDAG::getMergeValues(), lowerTcgen05Ld(), lowerTcgen05LdRed(), and SDValue().

Referenced by llvm::NVPTXTargetLowering::LowerOperation().

lowerIntrinsicWOChain()

SDValue lowerIntrinsicWOChain ( SDValue Op, SelectionDAG & DAG )

static

Definition at line 3176 of file NVPTXISelLowering.cpp.

References LowerClusterLaunchControlQueryCancel(), lowerCvtRSIntrinsics(), and lowerPrmtIntrinsic().

Referenced by llvm::NVPTXTargetLowering::LowerOperation().

lowerLOADi1()

SDValue lowerLOADi1 ( LoadSDNode * LD, SelectionDAG & DAG )

static

Definition at line 3800 of file NVPTXISelLowering.cpp.

References assert(), llvm::SelectionDAG::getExtLoad(), llvm::SelectionDAG::getMergeValues(), llvm::SelectionDAG::getNode(), llvm::ISD::NON_EXTLOAD, llvm::ISD::TRUNCATE, and llvm::ISD::ZEXTLOAD.

lowerLoadVector()

SDValue lowerLoadVector ( SDNode * N, SelectionDAG & DAG, const NVPTXSubtarget & STI )

static

Definition at line 3789 of file NVPTXISelLowering.cpp.

References llvm::SelectionDAG::getMergeValues(), N, replaceLoadVector(), and SDValue().

Referenced by combineLOAD().

lowerMSTORE()

SDValue lowerMSTORE ( SDValue Op, SelectionDAG & DAG )

static

Definition at line 3339 of file NVPTXISelLowering.cpp.

References AbstractManglingParser< Derived, Alloc >::Ops, assert(), llvm::ISD::BUILD_VECTOR, llvm::cast(), DL, llvm::enumerate(), llvm::ISD::EXTRACT_VECTOR_ELT, llvm::MemSDNode::getAlign(), llvm::SelectionDAG::getEVTAlign(), llvm::SelectionDAG::getIntPtrConstant(), llvm::SelectionDAG::getMemIntrinsicNode(), llvm::MemSDNode::getMemOperand(), llvm::MemSDNode::getMemoryVT(), llvm::SelectionDAG::getNode(), llvm::SelectionDAG::getRegister(), llvm::EVT::getSimpleVT(), llvm::SDValue::getValueType(), llvm::EVT::getVectorElementType(), llvm::EVT::getVectorNumElements(), llvm::SelectionDAG::getVTList(), I, llvm::EVT::isVector(), llvm_unreachable, N, llvm::MCRegister::NoRegister, llvm::Offset, llvm::MVT::SimpleTy, llvm::NVPTXISD::StoreV4, llvm::NVPTXISD::StoreV8, and llvm::ISD::UNDEF.

Referenced by llvm::NVPTXTargetLowering::LowerOperation().

lowerPrmtIntrinsic()

SDValue lowerPrmtIntrinsic ( SDValue Op, SelectionDAG & DAG )

static

Definition at line 3002 of file NVPTXISelLowering.cpp.

References A(), B(), llvm::NVPTX::PTXPrmtMode::B4E, DL, llvm::NVPTX::PTXPrmtMode::ECL, llvm::NVPTX::PTXPrmtMode::ECR, llvm::NVPTX::PTXPrmtMode::F4E, llvm::SelectionDAG::getConstant(), getPRMT(), llvm_unreachable, Mode, llvm::NVPTX::PTXPrmtMode::NONE, llvm::NVPTX::PTXPrmtMode::RC16, and llvm::NVPTX::PTXPrmtMode::RC8.

Referenced by lowerIntrinsicWOChain().

lowerROT()

SDValue lowerROT ( SDValue Op, SelectionDAG & DAG )

static

Definition at line 3271 of file NVPTXISelLowering.cpp.

References expandFSH64(), llvm::ISD::FSHL, llvm::ISD::FSHR, and llvm::ISD::ROTL.

Referenced by llvm::NVPTXTargetLowering::LowerOperation().

lowerSELECT()

SDValue lowerSELECT ( SDValue Op, SelectionDAG & DAG )

static

Definition at line 3305 of file NVPTXISelLowering.cpp.

References llvm::ISD::AND, assert(), Cond, DL, llvm::SelectionDAG::getAnyExtOrTrunc(), llvm::SelectionDAG::getFreeze(), llvm::SelectionDAG::getNode(), llvm::SelectionDAG::getNOT(), llvm::SelectionDAG::getSelect(), llvm::ISD::OR, Select, and llvm::ISD::TRUNCATE.

Referenced by llvm::NVPTXTargetLowering::LowerOperation().

lowerSTOREVector()

SDValue lowerSTOREVector ( SDValue Op, SelectionDAG & DAG, const NVPTXSubtarget & STI )

static

Definition at line 3871 of file NVPTXISelLowering.cpp.

References AbstractManglingParser< Derived, Alloc >::Ops, llvm::ISD::ANY_EXTEND, assert(), llvm::cast(), DL, llvm::ISD::EXTRACT_VECTOR_ELT, llvm::SelectionDAG::ExtractVectorElements(), llvm::SelectionDAG::getBitcast(), llvm::SelectionDAG::getBuildVector(), llvm::SelectionDAG::getContext(), llvm::SelectionDAG::getDataLayout(), llvm::SelectionDAG::getIntPtrConstant(), llvm::SelectionDAG::getMemIntrinsicNode(), llvm::SelectionDAG::getNode(), llvm::DataLayout::getPrefTypeAlign(), llvm::EVT::getSizeInBits(), llvm::EVT::getTypeForEVT(), llvm::SDValue::getValueType(), llvm::EVT::getVectorElementType(), getVectorLoweringShape(), llvm::EVT::getVectorNumElements(), llvm::MVT::getVectorVT(), llvm::SelectionDAG::getVTList(), I, llvm::EVT::isVector(), N, SDValue(), llvm::seq(), llvm::NVPTXISD::StoreV2, llvm::NVPTXISD::StoreV4, and llvm::NVPTXISD::StoreV8.

Referenced by combineSTORE().

lowerTcgen05Ld()

std::optional< std::pair< SDValue, SDValue > > lowerTcgen05Ld ( SDNode * N, SelectionDAG & DAG, bool HasOffset = false )

static

Definition at line 2731 of file NVPTXISelLowering.cpp.

References AbstractManglingParser< Derived, Alloc >::Ops, llvm::ISD::BUILD_VECTOR, llvm::cast(), DL, llvm::SelectionDAG::getMemIntrinsicNode(), llvm::MemSDNode::getMemOperand(), llvm::MemSDNode::getMemoryVT(), llvm::SelectionDAG::getNode(), llvm::SDValue::getValue(), llvm::EVT::getVectorNumElements(), llvm::SelectionDAG::getVTList(), llvm::ISD::INTRINSIC_W_CHAIN, llvm::EVT::isVector(), N, and llvm::SmallVectorTemplateBase< T, bool >::push_back().

Referenced by lowerIntrinsicWChain(), and ReplaceINTRINSIC_W_CHAIN().

lowerTcgen05LdRed()

std::optional< std::tuple< SDValue, SDValue, SDValue > > lowerTcgen05LdRed ( SDNode * N, SelectionDAG & DAG )

static

Definition at line 3102 of file NVPTXISelLowering.cpp.

References AbstractManglingParser< Derived, Alloc >::Ops, llvm::ISD::BUILD_VECTOR, llvm::cast(), DL, llvm::SelectionDAG::getMemIntrinsicNode(), llvm::MemSDNode::getMemOperand(), llvm::MemSDNode::getMemoryVT(), llvm::SelectionDAG::getNode(), getTcgen05LdRedID(), llvm::SDValue::getValue(), llvm::EVT::getVectorElementType(), llvm::EVT::getVectorNumElements(), llvm::SelectionDAG::getVTList(), llvm::EVT::isFloatingPoint(), llvm::EVT::isVector(), N, and llvm::SmallVectorTemplateBase< T, bool >::push_back().

Referenced by lowerIntrinsicWChain(), and ReplaceINTRINSIC_W_CHAIN().

LowerTcgen05MMADisableOutputLane()

SDValue LowerTcgen05MMADisableOutputLane ( SDValue Op, SelectionDAG & DAG )

static

Definition at line 2700 of file NVPTXISelLowering.cpp.

References AbstractManglingParser< Derived, Alloc >::Ops, llvm::cast(), DL, llvm::ISD::EXTRACT_VECTOR_ELT, llvm::SelectionDAG::getIntPtrConstant(), llvm::SelectionDAG::getMemIntrinsicNode(), llvm::MemSDNode::getMemOperand(), llvm::MemSDNode::getMemoryVT(), llvm::SelectionDAG::getNode(), getTcgen05MMADisableOutputLane(), llvm::SDValue::getValueType(), llvm::EVT::getVectorElementType(), llvm::EVT::getVectorNumElements(), I, llvm::EVT::isVector(), and N.

Referenced by lowerIntrinsicVoid().

lowerTcgen05St()

SDValue lowerTcgen05St ( SDValue Op, SelectionDAG & DAG, bool hasOffset = false )

static

Definition at line 2571 of file NVPTXISelLowering.cpp.

References AbstractManglingParser< Derived, Alloc >::Ops, llvm::cast(), DL, llvm::ISD::EXTRACT_VECTOR_ELT, llvm::SelectionDAG::getIntPtrConstant(), llvm::SelectionDAG::getMemIntrinsicNode(), llvm::MemSDNode::getMemOperand(), llvm::MemSDNode::getMemoryVT(), llvm::SelectionDAG::getNode(), llvm::SDValue::getValueType(), llvm::EVT::getVectorElementType(), llvm::EVT::getVectorNumElements(), I, llvm::ISD::INTRINSIC_VOID, llvm::EVT::isVector(), and N.

Referenced by lowerIntrinsicVoid().

lowerTensormapReplaceElemtype()

SDValue lowerTensormapReplaceElemtype ( SDValue Op, SelectionDAG & DAG )

static

Definition at line 2797 of file NVPTXISelLowering.cpp.

References DL, llvm::SelectionDAG::getSubtarget(), llvm::NVPTXSubtarget::hasTensormapReplaceElemtypeSupport(), N, and reportInvalidTensormapReplaceUsage().

Referenced by lowerIntrinsicVoid().

lowerTensormapReplaceSwizzleMode()

SDValue lowerTensormapReplaceSwizzleMode ( SDValue Op, SelectionDAG & DAG )

static

Definition at line 2811 of file NVPTXISelLowering.cpp.

References DL, llvm::SelectionDAG::getSubtarget(), llvm::NVPTXSubtarget::hasTensormapReplaceSwizzleModeSupport(), N, and reportInvalidTensormapReplaceUsage().

Referenced by lowerIntrinsicVoid().

LowerVectorArith()

SDValue LowerVectorArith ( SDValue Op, SelectionDAG & DAG )

static

Definition at line 2551 of file NVPTXISelLowering.cpp.

References llvm::ISD::BUILD_VECTOR, DL, E(), llvm::SelectionDAG::getNode(), I, llvm::SmallVectorTemplateBase< T, bool >::push_back(), and llvm::transform().

Referenced by llvm::NVPTXTargetLowering::LowerOperation().

matchMADConstOnePattern()

SDValue matchMADConstOnePattern ( SDValue Add )

static

Definition at line 6537 of file NVPTXISelLowering.cpp.

References llvm::ISD::ADD, llvm::Add, isConstOne(), and SDValue().

Referenced by combineMADConstOne(), and combineMulSelectConstOne().

PerformADDCombine()

SDValue PerformADDCombine ( SDNode * N, TargetLowering::DAGCombinerInfo & DCI, CodeGenOptLevel OptLevel )

static

PerformADDCombine - Target-specific dag combine xforms for ISD::ADD.

Definition at line 6123 of file NVPTXISelLowering.cpp.

References llvm::SDValue::getValueType(), llvm::EVT::isVector(), N, llvm::None, PerformADDCombineWithOperands(), and SDValue().

PerformADDCombineWithOperands()

SDValue PerformADDCombineWithOperands ( SDNode * N, SDValue N0, SDValue N1, TargetLowering::DAGCombinerInfo & DCI )

static

PerformADDCombineWithOperands - Try DAG combinations for an ADD with operands N0 and N1.

This is a helper for PerformADDCombine that is called with the default operands, and if that fails, with commuted operands.

Definition at line 5768 of file NVPTXISelLowering.cpp.

References llvm::ISD::ADD, llvm::TargetLowering::DAGCombinerInfo::DAG, DL, llvm::SDValue::getNode(), llvm::SelectionDAG::getNode(), llvm::SDValue::getOpcode(), llvm::SDNode::getOperand(), llvm::SelectionDAG::getSelect(), llvm::SDValue::getValueType(), llvm::SDNode::hasOneUse(), isConstZero(), llvm::ISD::MUL, Mul, N, SDValue(), and llvm::ISD::SELECT.

PerformBUILD_VECTORCombine()

SDValue PerformBUILD_VECTORCombine ( SDNode * N, TargetLowering::DAGCombinerInfo & DCI )

static

Definition at line 6813 of file NVPTXISelLowering.cpp.

References assert(), llvm::ISD::BITCAST, llvm::cast(), llvm::TargetLowering::DAGCombinerInfo::DAG, DL, llvm::SelectionDAG::getBitcast(), getPRMT(), llvm::EVT::getVectorNumElements(), llvm::EVT::is32BitVector(), llvm::isa(), llvm::TargetLowering::DAGCombinerInfo::isAfterLegalizeDAG(), llvm::NVPTX::isPackedVectorTy(), N, SDValue(), llvm::ISD::SRL, and llvm::ISD::TRUNCATE.

PerformEXTRACTCombine()

SDValue PerformEXTRACTCombine ( SDNode * N, TargetLowering::DAGCombinerInfo & DCI )

static

Definition at line 6673 of file NVPTXISelLowering.cpp.

References llvm::ISD::allOperandsUndef(), llvm::ISD::ANY_EXTEND, llvm::ISD::BITCAST, llvm::EVT::changeTypeToInteger(), llvm::TargetLowering::DAGCombinerInfo::DAG, DL, llvm::dyn_cast(), llvm::ISD::FREEZE, llvm::SelectionDAG::getConstant(), llvm::MVT::getIntegerVT(), llvm::SelectionDAG::getNode(), llvm::EVT::getScalarSizeInBits(), llvm::EVT::getSimpleVT(), llvm::EVT::getSizeInBits(), llvm::EVT::getVectorElementType(), llvm::EVT::getVectorNumElements(), llvm::NVPTX::isPackedVectorTy(), IsPTXVectorType(), llvm::EVT::isSimple(), llvm::ISD::LOAD, N, SDValue(), llvm::ISD::SRA, llvm::ISD::TRUNCATE, and llvm::Vector.

PerformFADDCombine()

SDValue PerformFADDCombine ( SDNode * N, TargetLowering::DAGCombinerInfo & DCI, CodeGenOptLevel OptLevel )

static

PerformFADDCombine - Target-specific dag combine xforms for ISD::FADD.

Definition at line 6256 of file NVPTXISelLowering.cpp.

References llvm::SDValue::getValueType(), llvm::EVT::isVector(), N, PerformFADDCombineWithOperands(), PerformScalarizeV2F32Op(), and SDValue().

PerformFADDCombineWithOperands()

SDValue PerformFADDCombineWithOperands ( SDNode * N, SDValue N0, SDValue N1, TargetLowering::DAGCombinerInfo & DCI, CodeGenOptLevel OptLevel )

static

Definition at line 5808 of file NVPTXISelLowering.cpp.

References llvm::TargetLowering::DAGCombinerInfo::DAG, llvm::ISD::FADD, llvm::ISD::FMA, llvm::ISD::FMUL, llvm::SDNode::getFlags(), llvm::SDNode::getIROrder(), llvm::SelectionDAG::getMachineFunction(), llvm::SDValue::getNode(), llvm::SelectionDAG::getNode(), llvm::SDValue::getOpcode(), llvm::SDValue::getOperand(), llvm::SelectionDAG::getTargetLoweringInfo(), llvm::SDValue::getValueType(), llvm::SDNodeFlags::hasAllowContract(), llvm::isa(), N, SDValue(), and llvm::SDNode::users().

Referenced by PerformFADDCombine().

PerformFMinMaxCombine()

SDValue PerformFMinMaxCombine ( SDNode * N, TargetLowering::DAGCombinerInfo & DCI, unsigned PTXVersion, unsigned SmVersion )

static

PerformFMinMaxCombine - Combine (fmaxnum (fmaxnum a, b), c) into (fmaxnum3 a, b, c).

Also covers other llvm min/max intrinsics.

Definition at line 6297 of file NVPTXISelLowering.cpp.

References A(), B(), llvm::CallingConv::C, llvm::TargetLowering::DAGCombinerInfo::DAG, getMinMax3Opcode(), llvm::SelectionDAG::getNode(), llvm::SDValue::getOpcode(), llvm::SDValue::getOperand(), llvm::SDValue::hasOneUse(), N, and SDValue().

PerformMULCombine()

SDValue PerformMULCombine ( SDNode * N, TargetLowering::DAGCombinerInfo & DCI, CodeGenOptLevel OptLevel )

static

PerformMULCombine - Runs PTX-specific DAG combine patterns on MUL nodes.

Definition at line 6619 of file NVPTXISelLowering.cpp.

References N, llvm::None, PerformMULCombineWithOperands(), SDValue(), and TryMULWIDECombine().

PerformMULCombineWithOperands()

SDValue PerformMULCombineWithOperands ( SDNode * N, SDValue N0, SDValue N1, TargetLowering::DAGCombinerInfo & DCI )

static

Definition at line 6591 of file NVPTXISelLowering.cpp.

References combineMADConstOne(), combineMulSelectConstOne(), DL, llvm::SDValue::getValueType(), llvm::EVT::isVector(), N, and SDValue().

Referenced by PerformMULCombine().

PerformREMCombine()

SDValue PerformREMCombine ( SDNode * N, TargetLowering::DAGCombinerInfo & DCI, CodeGenOptLevel OptLevel )

static

Definition at line 6327 of file NVPTXISelLowering.cpp.

References assert(), llvm::TargetLowering::DAGCombinerInfo::DAG, llvm::Default, DL, llvm::SelectionDAG::getNode(), llvm::SDNode::getOperand(), llvm::ISD::MUL, N, llvm::ISD::SDIV, SDValue(), llvm::ISD::SREM, llvm::ISD::SUB, llvm::ISD::UDIV, llvm::ISD::UREM, and llvm::SDNode::users().

PerformScalarizeV2F32Op()

SDValue PerformScalarizeV2F32Op ( SDNode * N, TargetLowering::DAGCombinerInfo & DCI )

static

Scalarize a v2f32 arithmetic node (FADD, FMUL, FSUB, FMA) when at least one operand is a BUILD_VECTOR that repacks values from non-adjacent register pairs.

Without this combine the BUILD_VECTOR forces allocation of a temporary 64-bit register, increasing register pressure.

Example - before: t0: v2f32,v2f32,ch = LoadV2 ... t1: f32 = extract_vector_elt t0, 0 t2: f32 = extract_vector_elt t0:1, 0 t3: v2f32 = BUILD_VECTOR t1, t2 ;; non-coalescable repack t4: v2f32 = fma t_a, t3, t_c

After: t0: v2f32,v2f32,ch = LoadV2 ... t1: f32 = extract_vector_elt t0, 0 t2: f32 = extract_vector_elt t0:1, 0 a0: f32 = extract_vector_elt t_a, 0 a1: f32 = extract_vector_elt t_a, 1 c0: f32 = extract_vector_elt t_c, 0 c1: f32 = extract_vector_elt t_c, 1 r0: f32 = fma a0, t1, c0 r1: f32 = fma a1, t2, c1 t4: v2f32 = BUILD_VECTOR r0, r1

Definition at line 6215 of file NVPTXISelLowering.cpp.

References llvm::ISD::BUILD_VECTOR, llvm::TargetLowering::DAGCombinerInfo::DAG, DL, llvm::ISD::EXTRACT_VECTOR_ELT, llvm::SelectionDAG::getNode(), llvm::EVT::getVectorElementType(), llvm::SelectionDAG::getVectorIdxConstant(), isNonCoalescableBuildVector(), N, llvm::none_of(), Opc, llvm::SmallVectorTemplateBase< T, bool >::push_back(), and SDValue().

Referenced by PerformFADDCombine().

PerformSELECTShiftCombine()

SDValue PerformSELECTShiftCombine ( SDNode * N, TargetLowering::DAGCombinerInfo & DCI )

static

Transform patterns like: (select (ugt shift_amt, BitWidth-1), 0, (srl/shl x, shift_amt)) (select (ult shift_amt, BitWidth), (srl/shl x, shift_amt), 0) Into: (NVPTXISD::SRL_CLAMP x, shift_amt) or (NVPTXISD::SHL_CLAMP x, shift_amt)

These patterns arise from C/C++ code like shift >= 32 ? 0 : x >> shift which guards against undefined behavior. PTX shr/shl instructions clamp shift amounts >= BitWidth to produce 0 for logical shifts, making the guard redundant.

Note: We only handle SRL and SHL, not SRA, because arithmetic right shifts could produce 0 or -1 when shift >= BitWidth. Note: We don't handle uge or ule. These don't appear because of canonicalization.

Definition at line 6740 of file NVPTXISelLowering.cpp.

References llvm::BitWidth, llvm::TargetLowering::DAGCombinerInfo::DAG, llvm::SelectionDAG::getNode(), llvm::SDValue::getOpcode(), llvm::SDValue::getOperand(), llvm::SDValue::getValueType(), llvm::TargetLowering::DAGCombinerInfo::isAfterLegalizeDAG(), llvm::PatternMatch::m_Deferred(), llvm::PatternMatch::m_Select(), llvm::PatternMatch::m_Shl(), llvm::PatternMatch::m_SpecificInt(), llvm::PatternMatch::m_TruncOrSelf(), llvm::PatternMatch::m_Value(), llvm::PatternMatch::m_Zero(), N, SDValue(), llvm::ISD::SETUGT, llvm::ISD::SETULT, and llvm::ISD::SRL.

PerformSETCCCombine()

SDValue PerformSETCCCombine ( SDNode * N, TargetLowering::DAGCombinerInfo & DCI, unsigned int SmVersion )

static

Definition at line 6646 of file NVPTXISelLowering.cpp.

References A(), B(), llvm::ISD::BUILD_VECTOR, llvm::TargetLowering::DAGCombinerInfo::DAG, DL, llvm::SelectionDAG::getNode(), llvm::SDValue::getValue(), llvm::SelectionDAG::getVTList(), N, SDValue(), llvm::NVPTXISD::SETP_BF16X2, and llvm::NVPTXISD::SETP_F16X2.

PerformSHLCombine()

SDValue PerformSHLCombine ( SDNode * N, TargetLowering::DAGCombinerInfo & DCI, CodeGenOptLevel OptLevel )

static

PerformSHLCombine - Runs PTX-specific DAG combine patterns on SHL nodes.

Definition at line 6634 of file NVPTXISelLowering.cpp.

References N, llvm::None, SDValue(), and TryMULWIDECombine().

PerformVSELECTCombine()

SDValue PerformVSELECTCombine ( SDNode * N, TargetLowering::DAGCombinerInfo & DCI )

static

Definition at line 6780 of file NVPTXISelLowering.cpp.

References llvm::ISD::BUILD_VECTOR, llvm::CallingConv::C, llvm::TargetLowering::DAGCombinerInfo::DAG, DL, E(), llvm::ISD::EXTRACT_VECTOR_ELT, llvm::SelectionDAG::getAnyExtOrTrunc(), llvm::SelectionDAG::getConstant(), llvm::SelectionDAG::getNode(), llvm::SDValue::getValueType(), I, N, SDValue(), and llvm::ISD::SELECT.

PromoteBinOpToF32()

SDValue PromoteBinOpToF32 ( SDNode * N, SelectionDAG & DAG )

static

Definition at line 2437 of file NVPTXISelLowering.cpp.

References DL, llvm::SelectionDAG::getContext(), llvm::SelectionDAG::getFPExtendOrRound(), llvm::SelectionDAG::getNode(), llvm::EVT::getVectorElementCount(), llvm::EVT::getVectorVT(), llvm::EVT::isVector(), and N.

promoteScalarIntegerPTX()

EVT promoteScalarIntegerPTX ( const EVT VT )

static

PromoteScalarIntegerPTX Used to make sure the arguments/returns are suitable for passing and promote them to a larger size if they're not.

The promoted type is placed in PromoteVT if the function returns true.

Definition at line 393 of file NVPTXISelLowering.cpp.

References llvm::EVT::getFixedSizeInBits(), llvm::EVT::isScalarInteger(), llvm_unreachable, and llvm::PowerOf2Ceil().

Referenced by llvm::NVPTXTargetLowering::LowerCall(), and llvm::NVPTXTargetLowering::LowerReturn().

refinePtrAS()

MachinePointerInfo refinePtrAS ( SDValue & Ptr, SelectionDAG & DAG, const DataLayout & DL, const TargetLowering & TL )

static

Definition at line 1344 of file NVPTXISelLowering.cpp.

References llvm::NVPTXAS::ADDRESS_SPACE_GENERIC, llvm::NVPTXAS::ADDRESS_SPACE_LOCAL, llvm::ISD::ADDRSPACECAST, llvm::cast(), DL, llvm::ISD::FrameIndex, llvm::SelectionDAG::getAddrSpaceCast(), llvm::SDNode::getOpcode(), llvm::SDNode::getOperand(), and llvm::TargetLoweringBase::getPointerTy().

Referenced by llvm::NVPTXTargetLowering::LowerCall().

replaceAtomicSwap128()

void replaceAtomicSwap128 ( SDNode * N, SelectionDAG & DAG, const NVPTXSubtarget & STI, SmallVectorImpl< SDValue > & Results )

static

Definition at line 7403 of file NVPTXISelLowering.cpp.

References AbstractManglingParser< Derived, Alloc >::Ops, assert(), llvm::NVPTXISD::ATOMIC_CMP_SWAP_B128, llvm::ISD::ATOMIC_SWAP, llvm::NVPTXISD::ATOMIC_SWAP_B128, llvm::ISD::BUILD_PAIR, llvm::cast(), llvm::LLVMContext::diagnose(), llvm::ISD::EXTRACT_ELEMENT, llvm::SelectionDAG::getContext(), llvm::SDLoc::getDebugLoc(), llvm::MachineFunction::getFunction(), llvm::SelectionDAG::getIntPtrConstant(), llvm::SelectionDAG::getMachineFunction(), llvm::SelectionDAG::getMemIntrinsicNode(), llvm::MemSDNode::getMemOperand(), llvm::SelectionDAG::getNode(), llvm::SDNode::getOperand(), llvm::SelectionDAG::getUNDEF(), llvm::SelectionDAG::getVTList(), llvm::NVPTXSubtarget::hasAtomSwap128(), N, llvm::SDNode::ops(), and Results.

ReplaceBITCAST()

void ReplaceBITCAST ( SDNode * Node, SelectionDAG & DAG, SmallVectorImpl< SDValue > & Results )

static

Definition at line 7158 of file NVPTXISelLowering.cpp.

References llvm::ISD::BUILD_VECTOR, DL, llvm::SelectionDAG::getBitcast(), llvm::SelectionDAG::getConstant(), llvm::SelectionDAG::getNode(), Results, llvm::ISD::SRL, and llvm::ISD::TRUNCATE.

ReplaceCopyFromReg_128()

void ReplaceCopyFromReg_128 ( SDNode * N, SelectionDAG & DAG, SmallVectorImpl< SDValue > & Results )

static

Definition at line 7363 of file NVPTXISelLowering.cpp.

References assert(), llvm::ISD::BUILD_PAIR, llvm::ISD::CopyFromReg, DL, llvm::SelectionDAG::getNode(), llvm::SDValue::getValue(), N, Reg, and Results.

ReplaceINTRINSIC_W_CHAIN()

void ReplaceINTRINSIC_W_CHAIN ( SDNode * N, SelectionDAG & DAG, SmallVectorImpl< SDValue > & Results )

static

Definition at line 7180 of file NVPTXISelLowering.cpp.

References AbstractManglingParser< Derived, Alloc >::Ops, llvm::SmallVectorImpl< T >::append(), assert(), llvm::cast(), DL, llvm::SDNode::getAsZExtVal(), llvm::SelectionDAG::getBuildVector(), llvm::SelectionDAG::getMemIntrinsicNode(), llvm::MemSDNode::getMemOperand(), llvm::MemSDNode::getMemoryVT(), llvm::SDValue::getNode(), llvm::SelectionDAG::getNode(), llvm::EVT::getSimpleVT(), llvm::EVT::getSizeInBits(), llvm::SDValue::getValue(), llvm::EVT::getVectorElementType(), llvm::EVT::getVectorNumElements(), llvm::SelectionDAG::getVTList(), llvm::ISD::INTRINSIC_W_CHAIN, llvm::EVT::isSimple(), llvm::EVT::isVector(), llvm::NVPTXISD::LDUV2, llvm::NVPTXISD::LDUV4, lowerTcgen05Ld(), lowerTcgen05LdRed(), N, llvm::SmallVectorTemplateBase< T, bool >::push_back(), Results, llvm::MVT::SimpleTy, and llvm::ISD::TRUNCATE.

replaceLoadVector() [1/2]

std::optional< std::pair< SDValue, SDValue > > replaceLoadVector ( SDNode * N, SelectionDAG & DAG, const NVPTXSubtarget & STI )

static

replaceLoadVector - Convert vector loads into multi-output scalar loads.

Definition at line 3679 of file NVPTXISelLowering.cpp.

References assert(), llvm::cast(), convertMLOADToLoadWithUsedBytesMask(), DL, llvm::SelectionDAG::ExtractVectorElements(), llvm::SelectionDAG::getBitcast(), llvm::SelectionDAG::getBuildVector(), llvm::SelectionDAG::getConstant(), llvm::SelectionDAG::getContext(), llvm::SelectionDAG::getDataLayout(), llvm::SelectionDAG::getIntPtrConstant(), llvm::SelectionDAG::getMemIntrinsicNode(), llvm::SelectionDAG::getNode(), llvm::EVT::getScalarType(), llvm::EVT::getSizeInBits(), llvm::EVT::getTypeForEVT(), llvm::SDValue::getValue(), llvm::EVT::getVectorElementType(), getVectorLoweringShape(), llvm::EVT::getVectorNumElements(), llvm::MVT::getVectorVT(), llvm::SelectionDAG::getVTList(), I, llvm::EVT::isVector(), llvm::NVPTXISD::LoadV2, llvm::NVPTXISD::LoadV4, llvm::NVPTXISD::LoadV8, llvm::ISD::MLOAD, N, llvm::SmallVectorTemplateBase< T, bool >::push_back(), llvm::seq(), llvm::SmallVectorTemplateCommon< T, typename >::size(), llvm::SmallVector, and llvm::ISD::TRUNCATE.

Referenced by lowerLoadVector(), and replaceLoadVector().

replaceLoadVector() [2/2]

void replaceLoadVector ( SDNode * N, SelectionDAG & DAG, SmallVectorImpl< SDValue > & Results, const NVPTXSubtarget & STI )

static

Definition at line 3782 of file NVPTXISelLowering.cpp.

References N, replaceLoadVector(), and Results.

replaceProxyReg()

void replaceProxyReg ( SDNode * N, SelectionDAG & DAG, const TargetLowering & TLI, SmallVectorImpl< SDValue > & Results )

static

Definition at line 7387 of file NVPTXISelLowering.cpp.

References llvm::SelectionDAG::getAnyExtOrTrunc(), llvm::SelectionDAG::getContext(), llvm::SelectionDAG::getNode(), llvm::TargetLoweringBase::getRegisterType(), N, Reg, and Results.

reportInvalidTensormapReplaceUsage()

SDValue reportInvalidTensormapReplaceUsage ( SDValue Op, SelectionDAG & DAG, unsigned Val )

static

Definition at line 2774 of file NVPTXISelLowering.cpp.

References llvm::LLVMContext::diagnose(), DL, llvm::dyn_cast(), llvm::PointerType::get(), llvm::SelectionDAG::getContext(), llvm::MachineFunction::getFunction(), llvm::SelectionDAG::getMachineFunction(), llvm::Intrinsic::getName(), llvm::GlobalValue::getParent(), and N.

Referenced by lowerTensormapReplaceElemtype(), and lowerTensormapReplaceSwizzleMode().

shouldConvertToIndirectCall()

bool shouldConvertToIndirectCall ( const CallBase * CB, const GlobalAddressSDNode * Func )

static

Definition at line 1335 of file NVPTXISelLowering.cpp.

References llvm::dyn_cast(), and llvm::CallBase::getFunctionType().

Referenced by llvm::NVPTXTargetLowering::LowerCall().

simplifyDemandedBitsForPRMT()

SDValue simplifyDemandedBitsForPRMT ( SDValue PRMT, const APInt & DemandedBits, SelectionDAG & DAG, const TargetLowering & TLI, unsigned Depth )

static

Definition at line 7808 of file NVPTXISelLowering.cpp.

References assert(), canonicalizePRMTInput(), llvm::Depth, llvm::dyn_cast(), llvm::SDValue::getConstantOperandVal(), llvm::APInt::getLoBits(), llvm::SDValue::getOpcode(), llvm::SDValue::getOperand(), getPRMT(), getPRMTDemandedBits(), getPRMTSelector(), llvm::APInt::getZExtValue(), Mode, SDValue(), and llvm::TargetLowering::SimplifyMultipleUseDemandedBits().

Referenced by llvm::NVPTXTargetLowering::SimplifyDemandedBitsForTargetNode().

sinkProxyReg()

SDValue sinkProxyReg ( SDValue R, SDValue Chain, TargetLowering::DAGCombinerInfo & DCI )

static

Definition at line 6971 of file NVPTXISelLowering.cpp.

References A(), AbstractManglingParser< Derived, Alloc >::Ops, llvm::ISD::ANY_EXTEND, B(), llvm::ISD::BITCAST, llvm::ISD::BUILD_VECTOR, llvm::ISD::Constant, llvm::TargetLowering::DAGCombinerInfo::DAG, llvm::ISD::EXTRACT_VECTOR_ELT, llvm::SelectionDAG::getNode(), llvm::TargetLowering::DAGCombinerInfo::isBeforeLegalize(), llvm::ISD::LOAD, llvm::NVPTXISD::LoadV2, llvm::NVPTXISD::LoadV4, llvm::ISD::OR, SDValue(), llvm::ISD::SHL, llvm::ISD::SIGN_EXTEND, sinkProxyReg(), llvm::ISD::SRA, llvm::ISD::SRL, llvm::ISD::TRUNCATE, and llvm::ISD::ZERO_EXTEND.

Referenced by combineProxyReg(), and sinkProxyReg().

TryMULWIDECombine()

SDValue TryMULWIDECombine ( SDNode * N, TargetLowering::DAGCombinerInfo & DCI )

static

TryMULWIDECombine - Attempt to replace a multiply of M bits with a multiply of M/2 bits that produces an M-bit result (i.e.

mul.wide). This transform works on both multiply DAG nodes and SHL DAG nodes with a constant shift amount.

Definition at line 6466 of file NVPTXISelLowering.cpp.

References AreMulWideOperandsDemotable(), llvm::BitWidth, llvm::TargetLowering::DAGCombinerInfo::DAG, DL, llvm::dyn_cast(), llvm::ConstantSDNode::getAPIntValue(), llvm::SelectionDAG::getConstant(), llvm::SelectionDAG::getNode(), llvm::EVT::getSizeInBits(), llvm::isa(), LHS, llvm::ISD::MUL, N, Opc, RHS, SDValue(), llvm::APInt::sge(), llvm::ISD::SHL, Signed, llvm::APInt::slt(), std::swap(), and llvm::ISD::TRUNCATE.

Referenced by PerformMULCombine(), and PerformSHLCombine().

VectorizePTXValueVTs()

template<typename T>

SmallVector< unsigned, 16 > VectorizePTXValueVTs ( const SmallVectorImpl< EVT > & ValueVTs, const SmallVectorImpl< T > & Offsets, Align ParamAlignment, bool IsVAArg = false )

static

Definition at line 481 of file NVPTXISelLowering.cpp.

References assert(), canMergeParamLoadStoresStartingAt(), E(), I, and llvm::SmallVectorTemplateCommon< T, typename >::size().

Referenced by llvm::NVPTXTargetLowering::LowerCall(), llvm::NVPTXTargetLowering::LowerFormalArguments(), and llvm::NVPTXTargetLowering::LowerReturn().

Variable Documentation

FMAContractLevelOpt

cl::opt< unsigned > FMAContractLevelOpt("nvptx-fma-level", cl::Hidden, cl::desc("NVPTX Specific: FMA contraction (0: don't do it" " 1: do it 2: do it aggressively"), cl::init(2)) ( "nvptx-fma-level" , cl::Hidden  )

static

Referenced by llvm::NVPTXTargetLowering::allowFMA().

ForceMinByValParamAlign

cl::opt< bool > ForceMinByValParamAlign("nvptx-force-min-byval-param-align", cl::Hidden, cl::desc("NVPTX Specific: force 4-byte minimal alignment for byval" " params of device functions."), cl::init(false)) ( "nvptx-force-min-byval-param-align" , cl::Hidden , cl::desc("NVPTX Specific: force 4-byte minimal alignment for byval" " params of device functions.") , cl::init(false)  )

static

Referenced by llvm::NVPTXTargetLowering::getFunctionByValParamAlign().

sched4reg

cl::opt< bool > sched4reg("nvptx-sched4reg", cl::desc("NVPTX Specific: schedule for register pressue"), cl::init(false)) ( "nvptx-sched4reg" , cl::desc("NVPTX Specific: schedule for register pressue") , cl::init(false)  )

static

Referenced by llvm::NVPTXTargetLowering::NVPTXTargetLowering().

UseApproxLog2F32

cl::opt< bool > UseApproxLog2F32("nvptx-approx-log2f32", cl::desc("NVPTX Specific: whether to use lg2.approx for log2"), cl::init(false)) ( "nvptx-approx-log2f32" , cl::desc("NVPTX Specific: whether to use lg2.approx for log2") , cl::init(false)  )

static

Whereas CUDA's implementation (see libdevice) uses ex2.approx for exp2(), it does NOT use lg2.approx for log2, so this is disabled by default.

Referenced by llvm::NVPTXTargetLowering::NVPTXTargetLowering().

UsePrecDivF32

cl::opt< NVPTX::DivPrecisionLevel > UsePrecDivF32("nvptx-prec-divf32", cl::Hidden, cl::desc( "NVPTX Specific: Override the precision of the lowering for f32 fdiv"), cl::values( clEnumValN(NVPTX::DivPrecisionLevel::Approx, "0", "Use div.approx"), clEnumValN(NVPTX::DivPrecisionLevel::Full, "1", "Use div.full"), clEnumValN(NVPTX::DivPrecisionLevel::IEEE754, "2", "Use IEEE Compliant F32 div.rnd if available (default)"), clEnumValN(NVPTX::DivPrecisionLevel::IEEE754_NoFTZ, "3", "Use IEEE Compliant F32 div.rnd if available, no FTZ")), cl::init(NVPTX::DivPrecisionLevel::IEEE754)) ( "nvptx-prec-divf32" , cl::Hidden , cl::desc( "NVPTX Specific: Override the precision of the lowering for f32 fdiv") , cl::values( clEnumValN(NVPTX::DivPrecisionLevel::Approx, "0", "Use div.approx"), clEnumValN(NVPTX::DivPrecisionLevel::Full, "1", "Use div.full"), clEnumValN(NVPTX::DivPrecisionLevel::IEEE754, "2", "Use IEEE Compliant F32 div.rnd if available (default)"), clEnumValN(NVPTX::DivPrecisionLevel::IEEE754_NoFTZ, "3", "Use IEEE Compliant F32 div.rnd if available, no FTZ")) , cl::init(NVPTX::DivPrecisionLevel::IEEE754)  )

static

Referenced by llvm::NVPTXTargetLowering::getDivF32Level().

UsePrecSqrtF32

cl::opt< bool > UsePrecSqrtF32("nvptx-prec-sqrtf32", cl::Hidden, cl::desc("NVPTX Specific: 0 use sqrt.approx, 1 use sqrt.rn."), cl::init(true)) ( "nvptx-prec-sqrtf32" , cl::Hidden , cl::desc("NVPTX Specific: 0 use sqrt.approx, 1 use sqrt.rn.") , cl::init(true)  )

static

Referenced by llvm::NVPTXTargetLowering::usePrecSqrtF32().