#include "Target/AMDGPU/AMDGPUMachineFunction.h"

Inheritance diagram for llvm::AMDGPUMachineFunction:

[legend]

Public Member Functions
	AMDGPUMachineFunction (const Function &F, const AMDGPUSubtarget &ST)

uint64_t	getExplicitKernArgSize () const

Align	getMaxKernArgAlign () const

uint32_t	getLDSSize () const

uint32_t	getGDSSize () const

bool	isEntryFunction () const

bool	isModuleEntryFunction () const

bool	hasNoSignedZerosFPMath () const

bool	isMemoryBound () const

bool	needsWaveLimiter () const

unsigned	allocateLDSGlobal (const DataLayout &DL, const GlobalVariable &GV)

unsigned	allocateLDSGlobal (const DataLayout &DL, const GlobalVariable &GV, Align Trailing)

void	allocateKnownAddressLDSGlobal (const Function &F)

Align	getDynLDSAlign () const

void	setDynLDSAlign (const Function &F, const GlobalVariable &GV)

Public Member Functions inherited from llvm::MachineFunctionInfo
virtual	~MachineFunctionInfo ()

virtual MachineFunctionInfo *	clone (BumpPtrAllocator &Allocator, MachineFunction &DestMF, const DenseMap< MachineBasicBlock , MachineBasicBlock > &Src2DstMBB) const
	Make a functionally equivalent copy of this MachineFunctionInfo in `MF`.

Static Public Member Functions
static std::optional< uint32_t >	getLDSKernelIdMetadata (const Function &F)

static std::optional< uint32_t >	getLDSAbsoluteAddress (const GlobalValue &GV)

Static Public Member Functions inherited from llvm::MachineFunctionInfo
template<typename FuncInfoTy , typename SubtargetTy = TargetSubtargetInfo>
static FuncInfoTy *	create (BumpPtrAllocator &Allocator, const Function &F, const SubtargetTy *STI)
	Factory function: default behavior is to call new using the supplied allocator.

template<typename Ty >
static Ty *	create (BumpPtrAllocator &Allocator, const Ty &MFI)

Protected Attributes
uint64_t	ExplicitKernArgSize = 0

Align	MaxKernArgAlign

uint32_t	LDSSize = 0
	Number of bytes in the LDS that are being used.

uint32_t	GDSSize = 0

uint32_t	StaticLDSSize = 0
	Number of bytes in the LDS allocated statically.

uint32_t	StaticGDSSize = 0

Align	DynLDSAlign
	Align for dynamic shared memory if any.

bool	IsEntryFunction = false

bool	IsModuleEntryFunction = false

bool	NoSignedZerosFPMath = false

bool	MemoryBound = false

bool	WaveLimiter = false

Detailed Description

Definition at line 25 of file AMDGPUMachineFunction.h.

Constructor & Destructor Documentation

◆ AMDGPUMachineFunction()

AMDGPUMachineFunction::AMDGPUMachineFunction	(	const Function &	F,
		const AMDGPUSubtarget &	ST
	)

Definition at line 21 of file AMDGPUMachineFunction.cpp.

References llvm::CallingConv::AMDGPU_KERNEL, CC, llvm::StringRef::consumeInteger(), llvm::StringRef::empty(), ExplicitKernArgSize, F, GDSSize, llvm::Attribute::getValueAsBool(), llvm::Attribute::getValueAsString(), llvm::Attribute::isStringAttribute(), MaxKernArgAlign, MemoryBound, NoSignedZerosFPMath, llvm::CallingConv::SPIR_KERNEL, StaticGDSSize, and WaveLimiter.

Member Function Documentation

◆ allocateKnownAddressLDSGlobal()

void AMDGPUMachineFunction::allocateKnownAddressLDSGlobal ( const Function & F )

Definition at line 116 of file AMDGPUMachineFunction.cpp.

References allocateLDSGlobal(), assert(), canElideModuleLDS(), F, getDynLDSAlign(), getKernelDynLDSGlobalFromFunction(), getKernelLDSGlobalFromFunction(), getLDSAbsoluteAddress(), isModuleEntryFunction(), LDSSize, ModuleLDSName, llvm::Offset, llvm::report_fatal_error(), and setDynLDSAlign().

◆ allocateLDSGlobal() [1/2]

unsigned llvm::AMDGPUMachineFunction::allocateLDSGlobal	(	const DataLayout &	DL,
		const GlobalVariable &	GV
	)

inline

Definition at line 100 of file AMDGPUMachineFunction.h.

References allocateLDSGlobal(), DL, and DynLDSAlign.

Referenced by allocateKnownAddressLDSGlobal(), allocateLDSGlobal(), llvm::AMDGPULegalizerInfo::legalizeGlobalValue(), and llvm::AMDGPUTargetLowering::LowerGlobalAddress().

◆ allocateLDSGlobal() [2/2]

unsigned AMDGPUMachineFunction::allocateLDSGlobal	(	const DataLayout &	DL,
		const GlobalVariable &	GV,
		Align	Trailing
	)

TODO: We should sort these to minimize wasted space due to alignment padding. Currently the padding is decided by the first encountered use during lowering.

Definition at line 56 of file AMDGPUMachineFunction.cpp.

References llvm::alignTo(), assert(), DL, GDSSize, llvm::GlobalValue::getAddressSpace(), llvm::GlobalObject::getAlign(), llvm::GlobalValue::getValueType(), LDSSize, llvm::AMDGPUAS::LOCAL_ADDRESS, llvm::Offset, llvm::AMDGPUAS::REGION_ADDRESS, StaticGDSSize, and StaticLDSSize.

◆ getDynLDSAlign()

Align llvm::AMDGPUMachineFunction::getDynLDSAlign ( ) const

inline

Definition at line 112 of file AMDGPUMachineFunction.h.

References DynLDSAlign.

Referenced by allocateKnownAddressLDSGlobal().

◆ getExplicitKernArgSize()

uint64_t llvm::AMDGPUMachineFunction::getExplicitKernArgSize ( ) const

inline

Definition at line 68 of file AMDGPUMachineFunction.h.

References ExplicitKernArgSize.

Referenced by llvm::AMDGPUTargetLowering::getImplicitParameterOffset().

◆ getGDSSize()

uint32_t llvm::AMDGPUMachineFunction::getGDSSize ( ) const

inline

Definition at line 78 of file AMDGPUMachineFunction.h.

References GDSSize.

◆ getLDSAbsoluteAddress()

std::optional< uint32_t > AMDGPUMachineFunction::getLDSAbsoluteAddress ( const GlobalValue & GV )

static

Definition at line 195 of file AMDGPUMachineFunction.cpp.

References llvm::GlobalValue::getAbsoluteSymbolRange(), llvm::GlobalValue::getAddressSpace(), and llvm::AMDGPUAS::LOCAL_ADDRESS.

Referenced by allocateKnownAddressLDSGlobal(), llvm::AMDGPUAsmPrinter::lowerConstant(), llvm::AMDGPUTargetLowering::LowerGlobalAddress(), and setDynLDSAlign().

◆ getLDSKernelIdMetadata()

std::optional< uint32_t > AMDGPUMachineFunction::getLDSKernelIdMetadata ( const Function & F )

static

Definition at line 179 of file AMDGPUMachineFunction.cpp.

References F, llvm::MDNode::getNumOperands(), and llvm::MDNode::getOperand().

Referenced by llvm::AMDGPULegalizerInfo::getLDSKernelId(), llvm::SITargetLowering::passSpecialInputs(), and llvm::AMDGPUCallLowering::passSpecialInputs().

◆ getLDSSize()

uint32_t llvm::AMDGPUMachineFunction::getLDSSize ( ) const

inline

Definition at line 74 of file AMDGPUMachineFunction.h.

References LDSSize.

Referenced by llvm::SITargetLowering::EmitInstrWithCustomInserter(), llvm::SIRegisterInfo::getRegPressureLimit(), llvm::PreRARematStage::initGCNSchedStage(), llvm::SIMachineFunctionInfo::limitOccupancy(), llvm::GCNTargetMachine::parseMachineFunctionInfo(), and llvm::SIMachineFunctionInfo::SIMachineFunctionInfo().

◆ getMaxKernArgAlign()

Align llvm::AMDGPUMachineFunction::getMaxKernArgAlign ( ) const

inline

Definition at line 72 of file AMDGPUMachineFunction.h.

References MaxKernArgAlign.

◆ hasNoSignedZerosFPMath()

bool llvm::AMDGPUMachineFunction::hasNoSignedZerosFPMath ( ) const

inline

Definition at line 88 of file AMDGPUMachineFunction.h.

References NoSignedZerosFPMath.

◆ isEntryFunction()

bool llvm::AMDGPUMachineFunction::isEntryFunction ( ) const

inline

Definition at line 82 of file AMDGPUMachineFunction.h.

References IsEntryFunction.

Referenced by llvm::SIMachineFunctionInfo::allocateWWMSpill(), llvm::SIFrameLowering::determineCalleeSaves(), llvm::SIFrameLowering::determineCalleeSavesSGPR(), llvm::SIRegisterInfo::eliminateFrameIndex(), llvm::SIFrameLowering::emitEntryFunctionPrologue(), llvm::AMDGPUAsmPrinter::emitFunctionBodyEnd(), llvm::AMDGPUAsmPrinter::emitFunctionBodyStart(), llvm::AMDGPUAsmPrinter::emitFunctionEntryLabel(), llvm::SIMachineFunctionInfo::getScavengeFI(), llvm::SIFrameLowering::hasFP(), llvm::AMDGPULegalizerInfo::legalizeImplicitArgPtr(), llvm::AMDGPULegalizerInfo::legalizeLDSKernelId(), llvm::SIFrameLowering::requiresStackPointerReference(), llvm::AMDGPUAsmPrinter::runOnMachineFunction(), and llvm::SIMachineFunctionInfo::SIMachineFunctionInfo().

◆ isMemoryBound()

bool llvm::AMDGPUMachineFunction::isMemoryBound ( ) const

inline

Definition at line 92 of file AMDGPUMachineFunction.h.

References MemoryBound.

Referenced by llvm::SIMachineFunctionInfo::getMinAllowedOccupancy().

◆ isModuleEntryFunction()

bool llvm::AMDGPUMachineFunction::isModuleEntryFunction ( ) const

inline

Definition at line 86 of file AMDGPUMachineFunction.h.

References IsModuleEntryFunction.

Referenced by allocateKnownAddressLDSGlobal(), llvm::AMDGPULegalizerInfo::legalizeGlobalValue(), llvm::AMDGPUTargetLowering::LowerGlobalAddress(), and llvm::AMDGPUAsmPrinter::runOnMachineFunction().

◆ needsWaveLimiter()

bool llvm::AMDGPUMachineFunction::needsWaveLimiter ( ) const

inline

Definition at line 96 of file AMDGPUMachineFunction.h.

References WaveLimiter.

Referenced by llvm::SIMachineFunctionInfo::getMinAllowedOccupancy(), and llvm::AMDGPUAsmPrinter::runOnMachineFunction().

◆ setDynLDSAlign()

void AMDGPUMachineFunction::setDynLDSAlign	(	const Function &	F,
		const GlobalVariable &	GV
	)

Definition at line 213 of file AMDGPUMachineFunction.cpp.

References llvm::alignTo(), assert(), DL, DynLDSAlign, F, llvm::GlobalObject::getAlign(), getKernelDynLDSGlobalFromFunction(), getLDSAbsoluteAddress(), llvm::GlobalValue::getValueType(), LDSSize, llvm::Offset, llvm::report_fatal_error(), and StaticLDSSize.

Referenced by allocateKnownAddressLDSGlobal(), and llvm::AMDGPULegalizerInfo::legalizeGlobalValue().

Member Data Documentation

◆ DynLDSAlign

Align llvm::AMDGPUMachineFunction::DynLDSAlign

protected

Align for dynamic shared memory if any.

Dynamic shared memory is allocated directly after the static one, i.e., LDSSize. Need to pad LDSSize to ensure that dynamic one is aligned accordingly. The maximal alignment is updated during IR translation or lowering stages.

Definition at line 48 of file AMDGPUMachineFunction.h.

Referenced by allocateLDSGlobal(), getDynLDSAlign(), and setDynLDSAlign().

◆ ExplicitKernArgSize

uint64_t llvm::AMDGPUMachineFunction::ExplicitKernArgSize = 0

protected

Definition at line 31 of file AMDGPUMachineFunction.h.

Referenced by AMDGPUMachineFunction(), and getExplicitKernArgSize().

◆ GDSSize

uint32_t llvm::AMDGPUMachineFunction::GDSSize = 0

protected

Definition at line 36 of file AMDGPUMachineFunction.h.

Referenced by allocateLDSGlobal(), AMDGPUMachineFunction(), and getGDSSize().

◆ IsEntryFunction

bool llvm::AMDGPUMachineFunction::IsEntryFunction = false

protected

Definition at line 52 of file AMDGPUMachineFunction.h.

Referenced by isEntryFunction().

◆ IsModuleEntryFunction

bool llvm::AMDGPUMachineFunction::IsModuleEntryFunction = false

protected

Definition at line 55 of file AMDGPUMachineFunction.h.

Referenced by isModuleEntryFunction().

◆ LDSSize

uint32_t llvm::AMDGPUMachineFunction::LDSSize = 0

protected

Number of bytes in the LDS that are being used.

Definition at line 35 of file AMDGPUMachineFunction.h.

Referenced by allocateKnownAddressLDSGlobal(), allocateLDSGlobal(), getLDSSize(), and setDynLDSAlign().

◆ MaxKernArgAlign

Align llvm::AMDGPUMachineFunction::MaxKernArgAlign

protected

Definition at line 32 of file AMDGPUMachineFunction.h.

Referenced by AMDGPUMachineFunction(), getMaxKernArgAlign(), and llvm::SIMachineFunctionInfo::SIMachineFunctionInfo().

◆ MemoryBound

bool llvm::AMDGPUMachineFunction::MemoryBound = false

protected

Definition at line 60 of file AMDGPUMachineFunction.h.

Referenced by AMDGPUMachineFunction(), and isMemoryBound().

◆ NoSignedZerosFPMath

bool llvm::AMDGPUMachineFunction::NoSignedZerosFPMath = false

protected

Definition at line 57 of file AMDGPUMachineFunction.h.

Referenced by AMDGPUMachineFunction(), and hasNoSignedZerosFPMath().

◆ StaticGDSSize

uint32_t llvm::AMDGPUMachineFunction::StaticGDSSize = 0

protected

Definition at line 41 of file AMDGPUMachineFunction.h.

Referenced by allocateLDSGlobal(), and AMDGPUMachineFunction().

◆ StaticLDSSize

uint32_t llvm::AMDGPUMachineFunction::StaticLDSSize = 0

protected

Number of bytes in the LDS allocated statically.

This field is only used in the instruction selector and not part of the machine function info.

Definition at line 40 of file AMDGPUMachineFunction.h.

Referenced by allocateLDSGlobal(), and setDynLDSAlign().

◆ WaveLimiter

bool llvm::AMDGPUMachineFunction::WaveLimiter = false

protected

Definition at line 63 of file AMDGPUMachineFunction.h.

Referenced by AMDGPUMachineFunction(), and needsWaveLimiter().

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

lib/Target/AMDGPU/AMDGPUMachineFunction.h
lib/Target/AMDGPU/AMDGPUMachineFunction.cpp

Public Member Functions

Static Public Member Functions

Protected Attributes

Detailed Description

Constructor & Destructor Documentation

◆ AMDGPUMachineFunction()

Member Function Documentation

◆ allocateKnownAddressLDSGlobal()

◆ allocateLDSGlobal() [1/2]

◆ allocateLDSGlobal() [2/2]

◆ getDynLDSAlign()

◆ getExplicitKernArgSize()

◆ getGDSSize()

◆ getLDSAbsoluteAddress()

◆ getLDSKernelIdMetadata()

◆ getLDSSize()

◆ getMaxKernArgAlign()

◆ hasNoSignedZerosFPMath()

◆ isEntryFunction()

◆ isMemoryBound()

◆ isModuleEntryFunction()

◆ needsWaveLimiter()

◆ setDynLDSAlign()

Member Data Documentation

◆ DynLDSAlign

◆ ExplicitKernArgSize

◆ GDSSize

◆ IsEntryFunction

◆ IsModuleEntryFunction

◆ LDSSize

◆ MaxKernArgAlign

◆ MemoryBound

◆ NoSignedZerosFPMath

◆ StaticGDSSize

◆ StaticLDSSize

◆ WaveLimiter