#include "Target/AMDGPU/AMDGPUMachineFunctionInfo.h"

Inheritance diagram for llvm::AMDGPUMachineFunctionInfo:

Public Member Functions
	AMDGPUMachineFunctionInfo (const Function &F, const AMDGPUSubtarget &ST)
uint64_t	getExplicitKernArgSize () const
Align	getMaxKernArgAlign () const
uint32_t	getLDSSize () const
uint32_t	getGDSSize () const
void	recordNumNamedBarriers (uint32_t GVAddr, unsigned BarCnt)
uint32_t	getNumNamedBarriers () const
bool	isEntryFunction () const
bool	isModuleEntryFunction () const
bool	isChainFunction () const
bool	isBottomOfStack () const
bool	isMemoryBound () const
bool	needsWaveLimiter () const
bool	hasInitWholeWave () const
void	setInitWholeWave ()
unsigned	allocateLDSGlobal (const DataLayout &DL, const GlobalVariable &GV)
unsigned	allocateLDSGlobal (const DataLayout &DL, const GlobalVariable &GV, Align Trailing)
Align	getDynLDSAlign () const
void	setDynLDSAlign (const Function &F, const GlobalVariable &GV)
void	setUsesDynamicLDS (bool DynLDS)
bool	isDynamicLDSUsed () const
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	UsesDynamicLDS = false
uint32_t	NumNamedBarriers = 0
bool	IsEntryFunction = false
bool	IsModuleEntryFunction = false
bool	IsChainFunction = false
bool	MemoryBound = false
bool	WaveLimiter = false
bool	HasInitWholeWave = false

Detailed Description

Definition at line 24 of file AMDGPUMachineFunctionInfo.h.

Constructor & Destructor Documentation

◆ AMDGPUMachineFunctionInfo()

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

Member Function Documentation

◆ allocateLDSGlobal() [1/2]

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

inline

Definition at line 105 of file AMDGPUMachineFunctionInfo.h.

References allocateLDSGlobal(), DL, and DynLDSAlign.

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

◆ allocateLDSGlobal() [2/2]

unsigned AMDGPUMachineFunctionInfo::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 88 of file AMDGPUMachineFunctionInfo.cpp.

References llvm::alignTo(), assert(), DL, GDSSize, llvm::GlobalValue::getAddressSpace(), llvm::GlobalVariable::getAlign(), llvm::GlobalVariable::getGlobalSize(), getLDSAbsoluteAddress(), llvm::GlobalValue::getValueType(), isModuleEntryFunction(), llvm::AMDGPU::isNamedBarrier(), LDSSize, llvm_unreachable, llvm::AMDGPUAS::LOCAL_ADDRESS, llvm::Offset, recordNumNamedBarriers(), llvm::AMDGPUAS::REGION_ADDRESS, llvm::report_fatal_error(), StaticGDSSize, and StaticLDSSize.

◆ getDynLDSAlign()

Align llvm::AMDGPUMachineFunctionInfo::getDynLDSAlign ( ) const

inline

Definition at line 115 of file AMDGPUMachineFunctionInfo.h.

References DynLDSAlign.

◆ getExplicitKernArgSize()

uint64_t llvm::AMDGPUMachineFunctionInfo::getExplicitKernArgSize ( ) const

inline

Definition at line 75 of file AMDGPUMachineFunctionInfo.h.

References ExplicitKernArgSize.

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

◆ getGDSSize()

uint32_t llvm::AMDGPUMachineFunctionInfo::getGDSSize ( ) const

inline

Definition at line 81 of file AMDGPUMachineFunctionInfo.h.

References GDSSize.

◆ getLDSAbsoluteAddress()

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

static

Definition at line 185 of file AMDGPUMachineFunctionInfo.cpp.

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

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

◆ getLDSKernelIdMetadata()

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

static

Definition at line 169 of file AMDGPUMachineFunctionInfo.cpp.

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

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

◆ getLDSSize()

uint32_t llvm::AMDGPUMachineFunctionInfo::getLDSSize ( ) const

inline

Definition at line 79 of file AMDGPUMachineFunctionInfo.h.

References LDSSize.

Referenced by llvm::SITargetLowering::EmitInstrWithCustomInserter(), and llvm::SIMachineFunctionInfo::SIMachineFunctionInfo().

◆ getMaxKernArgAlign()

Align llvm::AMDGPUMachineFunctionInfo::getMaxKernArgAlign ( ) const

inline

Definition at line 77 of file AMDGPUMachineFunctionInfo.h.

References MaxKernArgAlign.

◆ getNumNamedBarriers()

uint32_t llvm::AMDGPUMachineFunctionInfo::getNumNamedBarriers ( ) const

inline

Definition at line 87 of file AMDGPUMachineFunctionInfo.h.

References NumNamedBarriers.

Referenced by llvm::AMDGPUResourceUsageAnalysisImpl::analyzeResourceUsage().

◆ hasInitWholeWave()

bool llvm::AMDGPUMachineFunctionInfo::hasInitWholeWave ( ) const

inline

Definition at line 102 of file AMDGPUMachineFunctionInfo.h.

References HasInitWholeWave.

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

◆ isBottomOfStack()

bool llvm::AMDGPUMachineFunctionInfo::isBottomOfStack ( ) const

inline

Definition at line 96 of file AMDGPUMachineFunctionInfo.h.

References isEntryFunction().

Referenced by llvm::SIRegisterInfo::buildSpillLoadStore(), llvm::SIRegisterInfo::eliminateFrameIndex(), llvm::SIRegisterInfo::getFrameRegister(), and llvm::AMDGPULegalizerInfo::legalizeIntrinsic().

◆ isChainFunction()

bool llvm::AMDGPUMachineFunctionInfo::isChainFunction ( ) const

inline

Definition at line 93 of file AMDGPUMachineFunctionInfo.h.

References IsChainFunction.

Referenced by llvm::SIMachineFunctionInfo::allocateWWMSpill(), llvm::SIFrameLowering::determineCalleeSaves(), llvm::SIFrameLowering::determinePrologEpilogSGPRSaves(), llvm::SIFrameLowering::emitEpilogue(), and llvm::SIFrameLowering::emitPrologue().

◆ isDynamicLDSUsed()

bool AMDGPUMachineFunctionInfo::isDynamicLDSUsed ( ) const

Definition at line 235 of file AMDGPUMachineFunctionInfo.cpp.

References UsesDynamicLDS.

Referenced by llvm::AMDGPU::HSAMD::MetadataStreamerMsgPackV5::emitHiddenKernelArgs().

◆ isEntryFunction()

bool llvm::AMDGPUMachineFunctionInfo::isEntryFunction ( ) const

inline

Definition at line 89 of file AMDGPUMachineFunctionInfo.h.

References IsEntryFunction.

Referenced by llvm::SIMachineFunctionInfo::allocateWWMSpill(), llvm::SIFrameLowering::determineCalleeSaves(), llvm::SIFrameLowering::determineCalleeSavesSGPR(), llvm::SIFrameLowering::emitEntryFunctionPrologue(), llvm::SIFrameLowering::emitEpilogue(), llvm::AMDGPUAsmPrinter::emitFunctionBodyEnd(), llvm::AMDGPUAsmPrinter::emitFunctionBodyStart(), llvm::AMDGPUAsmPrinter::emitFunctionEntryLabel(), llvm::SIFrameLowering::emitPrologue(), llvm::SIFrameLowering::hasFPImpl(), isBottomOfStack(), llvm::AMDGPULegalizerInfo::legalizeImplicitArgPtr(), llvm::AMDGPULegalizerInfo::legalizeLDSKernelId(), llvm::SIFrameLowering::requiresStackPointerReference(), llvm::AMDGPUAsmPrinter::runOnMachineFunction(), llvm::SIMachineFunctionInfo::SIMachineFunctionInfo(), and llvm::SITargetLowering::supportSplitCSR().

◆ isMemoryBound()

bool llvm::AMDGPUMachineFunctionInfo::isMemoryBound ( ) const

inline

Definition at line 98 of file AMDGPUMachineFunctionInfo.h.

References MemoryBound.

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

◆ isModuleEntryFunction()

bool llvm::AMDGPUMachineFunctionInfo::isModuleEntryFunction ( ) const

inline

Definition at line 91 of file AMDGPUMachineFunctionInfo.h.

References IsModuleEntryFunction.

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

◆ needsWaveLimiter()

bool llvm::AMDGPUMachineFunctionInfo::needsWaveLimiter ( ) const

inline

Definition at line 100 of file AMDGPUMachineFunctionInfo.h.

References WaveLimiter.

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

◆ recordNumNamedBarriers()

void llvm::AMDGPUMachineFunctionInfo::recordNumNamedBarriers	(	uint32_t	GVAddr,
		unsigned	BarCnt )

inline

Definition at line 83 of file AMDGPUMachineFunctionInfo.h.

References NumNamedBarriers.

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

◆ setDynLDSAlign()

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

Definition at line 203 of file AMDGPUMachineFunctionInfo.cpp.

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

Referenced by llvm::AMDGPULegalizerInfo::legalizeGlobalValue().

◆ setInitWholeWave()

void llvm::AMDGPUMachineFunctionInfo::setInitWholeWave ( )

inline

Definition at line 103 of file AMDGPUMachineFunctionInfo.h.

References HasInitWholeWave.

Referenced by llvm::GCNTargetMachine::parseMachineFunctionInfo().

◆ setUsesDynamicLDS()

void AMDGPUMachineFunctionInfo::setUsesDynamicLDS ( bool DynLDS )

Definition at line 231 of file AMDGPUMachineFunctionInfo.cpp.

References UsesDynamicLDS.

Member Data Documentation

◆ DynLDSAlign

Align llvm::AMDGPUMachineFunctionInfo::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 47 of file AMDGPUMachineFunctionInfo.h.

Referenced by allocateLDSGlobal(), getDynLDSAlign(), llvm::SIMachineFunctionInfo::initializeBaseYamlFields(), and setDynLDSAlign().