LLVM 18.0.0git
Public Member Functions | Static Public Member Functions | Protected Attributes | List of all members
llvm::AMDGPUMachineFunction Class Reference

#include "Target/AMDGPU/AMDGPUMachineFunction.h"

Inheritance diagram for llvm::AMDGPUMachineFunction:
Inheritance graph
[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 isChainFunction () 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)
 
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 IsChainFunction = 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 22 of file AMDGPUMachineFunction.cpp.

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

Member Function Documentation

◆ allocateLDSGlobal() [1/2]

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

Definition at line 105 of file AMDGPUMachineFunction.h.

References allocateLDSGlobal(), DL, and DynLDSAlign.

Referenced by 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 70 of file AMDGPUMachineFunction.cpp.

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

◆ getDynLDSAlign()

Align llvm::AMDGPUMachineFunction::getDynLDSAlign ( ) const
inline

Definition at line 115 of file AMDGPUMachineFunction.h.

References DynLDSAlign.

◆ getExplicitKernArgSize()

uint64_t llvm::AMDGPUMachineFunction::getExplicitKernArgSize ( ) const
inline

Definition at line 71 of file AMDGPUMachineFunction.h.

References ExplicitKernArgSize.

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

◆ getGDSSize()

uint32_t llvm::AMDGPUMachineFunction::getGDSSize ( ) const
inline

Definition at line 81 of file AMDGPUMachineFunction.h.

References GDSSize.

◆ getLDSAbsoluteAddress()

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

Definition at line 168 of file AMDGPUMachineFunction.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 > AMDGPUMachineFunction::getLDSKernelIdMetadata ( const Function &  F)
static

Definition at line 152 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 77 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 75 of file AMDGPUMachineFunction.h.

References MaxKernArgAlign.

◆ hasNoSignedZerosFPMath()

bool llvm::AMDGPUMachineFunction::hasNoSignedZerosFPMath ( ) const
inline

Definition at line 93 of file AMDGPUMachineFunction.h.

References NoSignedZerosFPMath.

◆ isChainFunction()

bool llvm::AMDGPUMachineFunction::isChainFunction ( ) const
inline

Definition at line 91 of file AMDGPUMachineFunction.h.

References IsChainFunction.

◆ isEntryFunction()

bool llvm::AMDGPUMachineFunction::isEntryFunction ( ) const
inline

Definition at line 85 of file AMDGPUMachineFunction.h.

References IsEntryFunction.

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

◆ isMemoryBound()

bool llvm::AMDGPUMachineFunction::isMemoryBound ( ) const
inline

Definition at line 97 of file AMDGPUMachineFunction.h.

References MemoryBound.

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

◆ isModuleEntryFunction()

bool llvm::AMDGPUMachineFunction::isModuleEntryFunction ( ) const
inline

Definition at line 89 of file AMDGPUMachineFunction.h.

References IsModuleEntryFunction.

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

◆ needsWaveLimiter()

bool llvm::AMDGPUMachineFunction::needsWaveLimiter ( ) const
inline

Definition at line 101 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 186 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 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().

◆ IsChainFunction

bool llvm::AMDGPUMachineFunction::IsChainFunction = false
protected

Definition at line 58 of file AMDGPUMachineFunction.h.

Referenced by isChainFunction().

◆ 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 allocateLDSGlobal(), AMDGPUMachineFunction(), 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 63 of file AMDGPUMachineFunction.h.

Referenced by AMDGPUMachineFunction(), and isMemoryBound().

◆ NoSignedZerosFPMath

bool llvm::AMDGPUMachineFunction::NoSignedZerosFPMath = false
protected

Definition at line 60 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(), AMDGPUMachineFunction(), and setDynLDSAlign().

◆ WaveLimiter

bool llvm::AMDGPUMachineFunction::WaveLimiter = false
protected

Definition at line 66 of file AMDGPUMachineFunction.h.

Referenced by AMDGPUMachineFunction(), and needsWaveLimiter().

The documentation for this class was generated from the following files:
Generated on Mon Sep 25 2023 19:08:00 for LLVM by doxygen 1.9.6