#include "Target/AMDGPU/AMDGPUSubtarget.h"

Inheritance diagram for llvm::AMDGPUSubtarget:

Public Types
enum	Generation { INVALID = 0 , R600 = 1 , R700 = 2 , EVERGREEN = 3 , NORTHERN_ISLANDS = 4 , SOUTHERN_ISLANDS = 5 , SEA_ISLANDS = 6 , VOLCANIC_ISLANDS = 7 , GFX9 = 8 , GFX10 = 9 , GFX11 = 10 }

Public Member Functions
	AMDGPUSubtarget (const Triple &TT)

std::pair< unsigned, unsigned >	getDefaultFlatWorkGroupSize (CallingConv::ID CC) const

std::pair< unsigned, unsigned >	getFlatWorkGroupSizes (const Function &F) const

std::pair< unsigned, unsigned >	getWavesPerEU (const Function &F) const

std::pair< unsigned, unsigned >	getWavesPerEU (const Function &F, std::pair< unsigned, unsigned > FlatWorkGroupSizes) const
	Overload which uses the specified values for the flat work group sizes, rather than querying the function itself.

unsigned	getMaxLocalMemSizeWithWaveCount (unsigned WaveCount, const Function &) const
	Return the amount of LDS that can be used that will not restrict the occupancy lower than WaveCount.

unsigned	getOccupancyWithLocalMemSize (uint32_t Bytes, const Function &) const
	Inverse of getMaxLocalMemWithWaveCount.

unsigned	getOccupancyWithLocalMemSize (const MachineFunction &MF) const

bool	isAmdHsaOS () const

bool	isAmdPalOS () const

bool	isMesa3DOS () const

bool	isMesaKernel (const Function &F) const

bool	isAmdHsaOrMesa (const Function &F) const

bool	isGCN () const

bool	isGCN3Encoding () const

bool	has16BitInsts () const

bool	hasTrue16BitInsts () const

bool	hasMadMixInsts () const

bool	hasMadMacF32Insts () const

bool	hasDsSrc2Insts () const

bool	hasSDWA () const

bool	hasVOP3PInsts () const

bool	hasMulI24 () const

bool	hasMulU24 () const

bool	hasSMulHi () const

bool	hasInv2PiInlineImm () const

bool	hasFminFmaxLegacy () const

bool	hasTrigReducedRange () const

bool	isPromoteAllocaEnabled () const

unsigned	getWavefrontSize () const

unsigned	getWavefrontSizeLog2 () const

unsigned	getLocalMemorySize () const

unsigned	getAddressableLocalMemorySize () const

unsigned	getEUsPerCU () const
	Number of SIMDs/EUs (execution units) per "CU" ("compute unit"), where the "CU" is the unit onto which workgroups are mapped.

Align	getAlignmentForImplicitArgPtr () const

unsigned	getExplicitKernelArgOffset (const Function &F) const
	Returns the offset in bytes from the start of the input buffer of the first explicit kernel argument.

virtual unsigned	getMaxWorkGroupsPerCU (unsigned FlatWorkGroupSize) const =0

virtual unsigned	getMinFlatWorkGroupSize () const =0

virtual unsigned	getMaxFlatWorkGroupSize () const =0

virtual unsigned	getWavesPerEUForWorkGroup (unsigned FlatWorkGroupSize) const =0

virtual unsigned	getMinWavesPerEU () const =0

unsigned	getMaxWavesPerEU () const

unsigned	getMaxWorkitemID (const Function &Kernel, unsigned Dimension) const
	Return the maximum workitem ID value in the function, for the given (0, 1, 2) dimension.

bool	makeLIDRangeMetadata (Instruction *I) const
	Creates value range metadata on an workitemid.* intrinsic call or load.

unsigned	getImplicitArgNumBytes (const Function &F) const

uint64_t	getExplicitKernArgSize (const Function &F, Align &MaxAlign) const

unsigned	getKernArgSegmentSize (const Function &F, Align &MaxAlign) const

AMDGPUDwarfFlavour	getAMDGPUDwarfFlavour () const

virtual	~AMDGPUSubtarget ()=default

Static Public Member Functions
static const AMDGPUSubtarget &	get (const MachineFunction &MF)

static const AMDGPUSubtarget &	get (const TargetMachine &TM, const Function &F)

Protected Attributes
bool	GCN3Encoding = false

bool	Has16BitInsts = false

bool	HasTrue16BitInsts = false

bool	HasMadMixInsts = false

bool	HasMadMacF32Insts = false

bool	HasDsSrc2Insts = false

bool	HasSDWA = false

bool	HasVOP3PInsts = false

bool	HasMulI24 = true

bool	HasMulU24 = true

bool	HasSMulHi = false

bool	HasInv2PiInlineImm = false

bool	HasFminFmaxLegacy = true

bool	EnablePromoteAlloca = false

bool	HasTrigReducedRange = false

unsigned	EUsPerCU = 4

unsigned	MaxWavesPerEU = 10

unsigned	LocalMemorySize = 0

unsigned	AddressableLocalMemorySize = 0

char	WavefrontSizeLog2 = 0

Detailed Description

Definition at line 29 of file AMDGPUSubtarget.h.

Member Enumeration Documentation

◆ Generation

enum llvm::AMDGPUSubtarget::Generation

Enumerator
INVALID
R600
R700
EVERGREEN
NORTHERN_ISLANDS
SOUTHERN_ISLANDS
SEA_ISLANDS
VOLCANIC_ISLANDS
GFX9
GFX10
GFX11

Definition at line 31 of file AMDGPUSubtarget.h.

Constructor & Destructor Documentation

◆ AMDGPUSubtarget()

AMDGPUSubtarget::AMDGPUSubtarget ( const Triple & TT )

Definition at line 167 of file AMDGPUSubtarget.cpp.

◆ ~AMDGPUSubtarget()

virtual llvm::AMDGPUSubtarget::~AMDGPUSubtarget ( )

virtualdefault

Member Function Documentation

◆ get() [1/2]

const AMDGPUSubtarget & AMDGPUSubtarget::get ( const MachineFunction & MF )

static

Definition at line 987 of file AMDGPUSubtarget.cpp.

References llvm::Triple::amdgcn, llvm::Triple::getArch(), llvm::MachineFunction::getSubtarget(), llvm::MachineFunction::getTarget(), and llvm::TargetMachine::getTargetTriple().

Referenced by llvm::AMDGPUTargetLowering::analyzeFormalArgumentsCompute(), and llvm::AMDGPUTargetLowering::getImplicitParameterOffset().

◆ get() [2/2]

const AMDGPUSubtarget & AMDGPUSubtarget::get	(	const TargetMachine &	TM,
		const Function &	F
	)

static

Definition at line 994 of file AMDGPUSubtarget.cpp.

References llvm::Triple::amdgcn, F, and TM.

◆ getAddressableLocalMemorySize()

unsigned llvm::AMDGPUSubtarget::getAddressableLocalMemorySize ( ) const

inline

Definition at line 214 of file AMDGPUSubtarget.h.

References AddressableLocalMemorySize.

◆ getAlignmentForImplicitArgPtr()

Align llvm::AMDGPUSubtarget::getAlignmentForImplicitArgPtr ( ) const

inline

Definition at line 223 of file AMDGPUSubtarget.h.

References isAmdHsaOS().

Referenced by getKernArgSegmentSize().

◆ getAMDGPUDwarfFlavour()

AMDGPUDwarfFlavour AMDGPUSubtarget::getAMDGPUDwarfFlavour ( ) const

Returns: Corresponding DWARF register number mapping flavour for the WavefrontSize.

Definition at line 593 of file AMDGPUSubtarget.cpp.

References getWavefrontSize(), llvm::Wave32, and llvm::Wave64.

◆ getDefaultFlatWorkGroupSize()

std::pair< unsigned, unsigned > AMDGPUSubtarget::getDefaultFlatWorkGroupSize ( CallingConv::ID CC ) const

Returns: Default range flat work group size for a calling convention.

Definition at line 382 of file AMDGPUSubtarget.cpp.

References llvm::CallingConv::AMDGPU_ES, llvm::CallingConv::AMDGPU_GS, llvm::CallingConv::AMDGPU_HS, llvm::CallingConv::AMDGPU_LS, llvm::CallingConv::AMDGPU_PS, llvm::CallingConv::AMDGPU_VS, CC, getMaxFlatWorkGroupSize(), and getWavefrontSize().

Referenced by getFlatWorkGroupSizes().

◆ getEUsPerCU()

unsigned llvm::AMDGPUSubtarget::getEUsPerCU ( ) const

inline

Number of SIMDs/EUs (execution units) per "CU" ("compute unit"), where the "CU" is the unit onto which workgroups are mapped.

This takes WGP mode vs. CU mode into account.

Definition at line 221 of file AMDGPUSubtarget.h.

References EUsPerCU.

Referenced by getMaxLocalMemSizeWithWaveCount(), and getOccupancyWithLocalMemSize().

◆ getExplicitKernArgSize()

uint64_t AMDGPUSubtarget::getExplicitKernArgSize	(	const Function &	F,
		Align &	MaxAlign
	)		const

Definition at line 553 of file AMDGPUSubtarget.cpp.

References llvm::alignTo(), llvm::CallingConv::AMDGPU_KERNEL, Arg, assert(), DL, F, and llvm::CallingConv::SPIR_KERNEL.

Referenced by getKernArgSegmentSize().

◆ getExplicitKernelArgOffset()

unsigned llvm::AMDGPUSubtarget::getExplicitKernelArgOffset ( const Function & F ) const

inline

Returns the offset in bytes from the start of the input buffer of the first explicit kernel argument.

Definition at line 229 of file AMDGPUSubtarget.h.

References llvm::Triple::AMDHSA, llvm::Triple::AMDPAL, llvm::Triple::getOS(), llvm_unreachable, llvm::Triple::Mesa3D, and llvm::Triple::UnknownOS.

Referenced by getKernArgSegmentSize(), and llvm::AMDGPUCallLowering::lowerFormalArgumentsKernel().

◆ getFlatWorkGroupSizes()

std::pair< unsigned, unsigned > AMDGPUSubtarget::getFlatWorkGroupSizes ( const Function & F ) const

Returns: Subtarget's default pair of minimum/maximum flat work group sizes for function F, or minimum/maximum flat work group sizes explicitly requested using "amdgpu-flat-work-group-size" attribute attached to function F.; Subtarget's default values if explicitly requested values cannot be converted to integer, or violate subtarget's specifications.

Definition at line 396 of file AMDGPUSubtarget.cpp.

References llvm::Default, F, getDefaultFlatWorkGroupSize(), llvm::AMDGPU::getIntegerPairAttribute(), getMaxFlatWorkGroupSize(), and getMinFlatWorkGroupSize().

Referenced by getMaxLocalMemSizeWithWaveCount(), getMaxWorkitemID(), getOccupancyWithLocalMemSize(), getWavesPerEU(), and makeLIDRangeMetadata().

◆ getImplicitArgNumBytes()

unsigned AMDGPUSubtarget::getImplicitArgNumBytes ( const Function & F ) const

Returns: Number of bytes of arguments that are passed to a shader or kernel in addition to the explicit ones declared for the function.

Definition at line 534 of file AMDGPUSubtarget.cpp.

References llvm::AMDGPU::AMDHSA_COV5, assert(), F, llvm::AMDGPU::getCodeObjectVersion(), llvm::AMDGPU::isKernel(), and isMesaKernel().

Referenced by getKernArgSegmentSize().

◆ getKernArgSegmentSize()

unsigned AMDGPUSubtarget::getKernArgSegmentSize	(	const Function &	F,
		Align &	MaxAlign
	)		const

Definition at line 575 of file AMDGPUSubtarget.cpp.

References llvm::alignTo(), F, getAlignmentForImplicitArgPtr(), getExplicitKernArgSize(), getExplicitKernelArgOffset(), and getImplicitArgNumBytes().

Referenced by llvm::AMDGPU::HSAMD::MetadataStreamerMsgPackV3::getHSAKernelProps().

◆ getLocalMemorySize()

unsigned llvm::AMDGPUSubtarget::getLocalMemorySize ( ) const

inline

Definition at line 210 of file AMDGPUSubtarget.h.

References LocalMemorySize.

Referenced by getMaxLocalMemSizeWithWaveCount(), and getOccupancyWithLocalMemSize().

◆ getMaxFlatWorkGroupSize()

virtual unsigned llvm::AMDGPUSubtarget::getMaxFlatWorkGroupSize ( ) const

pure virtual

Returns: Maximum flat work group size supported by the subtarget.

Implemented in llvm::GCNSubtarget, and llvm::R600Subtarget.

Referenced by getDefaultFlatWorkGroupSize(), and getFlatWorkGroupSizes().

◆ getMaxLocalMemSizeWithWaveCount()

unsigned AMDGPUSubtarget::getMaxLocalMemSizeWithWaveCount	(	unsigned	WaveCount,
		const Function &	F
	)		const

Return the amount of LDS that can be used that will not restrict the occupancy lower than WaveCount.

Definition at line 317 of file AMDGPUSubtarget.cpp.

References F, getEUsPerCU(), getFlatWorkGroupSizes(), getLocalMemorySize(), and getWavefrontSize().

◆ getMaxWavesPerEU()

unsigned llvm::AMDGPUSubtarget::getMaxWavesPerEU ( ) const

inline

Returns: Maximum number of waves per execution unit supported by the subtarget without any kind of limitation.

Definition at line 266 of file AMDGPUSubtarget.h.

References MaxWavesPerEU.

Referenced by getOccupancyWithLocalMemSize(), and getWavesPerEU().

◆ getMaxWorkGroupsPerCU()

virtual unsigned llvm::AMDGPUSubtarget::getMaxWorkGroupsPerCU ( unsigned FlatWorkGroupSize ) const

pure virtual

Returns: Maximum number of work groups per compute unit supported by the subtarget and limited by given FlatWorkGroupSize.

Implemented in llvm::GCNSubtarget, and llvm::R600Subtarget.

Referenced by getOccupancyWithLocalMemSize().

◆ getMaxWorkitemID()

unsigned AMDGPUSubtarget::getMaxWorkitemID	(	const Function &	Kernel,
		unsigned	Dimension
	)		const

Return the maximum workitem ID value in the function, for the given (0, 1, 2) dimension.

Definition at line 464 of file AMDGPUSubtarget.cpp.

References getFlatWorkGroupSizes(), and getReqdWorkGroupSize().

Referenced by llvm::AMDGPUTargetLowering::computeKnownBitsForTargetNode(), llvm::GCNTTIImpl::isAlwaysUniform(), llvm::AMDGPULegalizerInfo::legalizeWorkitemIDIntrinsic(), and llvm::SITargetLowering::passSpecialInputs().

◆ getMinFlatWorkGroupSize()

virtual unsigned llvm::AMDGPUSubtarget::getMinFlatWorkGroupSize ( ) const

pure virtual

Returns: Minimum flat work group size supported by the subtarget.

Implemented in llvm::GCNSubtarget, and llvm::R600Subtarget.

Referenced by getFlatWorkGroupSizes().

◆ getMinWavesPerEU()

virtual unsigned llvm::AMDGPUSubtarget::getMinWavesPerEU ( ) const

pure virtual

Returns: Minimum number of waves per execution unit supported by the subtarget.

Implemented in llvm::GCNSubtarget, and llvm::R600Subtarget.

Referenced by getWavesPerEU().

◆ getOccupancyWithLocalMemSize() [1/2]

unsigned AMDGPUSubtarget::getOccupancyWithLocalMemSize ( const MachineFunction & MF ) const

Definition at line 376 of file AMDGPUSubtarget.cpp.

References llvm::MachineFunction::getFunction(), llvm::MachineFunction::getInfo(), and getOccupancyWithLocalMemSize().

◆ getOccupancyWithLocalMemSize() [2/2]

unsigned AMDGPUSubtarget::getOccupancyWithLocalMemSize	(	uint32_t	Bytes,
		const Function &	F
	)		const

Inverse of getMaxLocalMemWithWaveCount.

Return the maximum wavecount if the given LDS memory size is the only constraint.

Definition at line 336 of file AMDGPUSubtarget.cpp.

References assert(), llvm::divideCeil(), F, getEUsPerCU(), getFlatWorkGroupSizes(), getLocalMemorySize(), getMaxWavesPerEU(), getMaxWorkGroupsPerCU(), and getWavefrontSize().

Referenced by llvm::GCNSchedStage::checkScheduling(), llvm::GCNSubtarget::computeOccupancy(), getOccupancyWithLocalMemSize(), and llvm::SIRegisterInfo::getRegPressureLimit().

◆ getWavefrontSize()

unsigned llvm::AMDGPUSubtarget::getWavefrontSize ( ) const

inline

Definition at line 202 of file AMDGPUSubtarget.h.

References WavefrontSizeLog2.

Referenced by llvm::SIRegisterInfo::buildSpillLoadStore(), getAMDGPUDwarfFlavour(), getDefaultFlatWorkGroupSize(), llvm::AMDGPU::HSAMD::MetadataStreamerMsgPackV3::getHSAKernelProps(), getMaxLocalMemSizeWithWaveCount(), getOccupancyWithLocalMemSize(), llvm::SITargetLowering::getRegClassFor(), llvm::SIInstrInfo::getScratchRsrcWords23(), llvm::GCNSubtarget::isWave32(), llvm::GCNSubtarget::isWave64(), llvm::AMDGPULegalizerInfo::legalizeIntrinsic(), lowerFCMPIntrinsic(), lowerICMPIntrinsic(), and llvm::SITargetLowering::requiresUniformRegister().

◆ getWavefrontSizeLog2()

unsigned llvm::AMDGPUSubtarget::getWavefrontSizeLog2 ( ) const

inline

Definition at line 206 of file AMDGPUSubtarget.h.

References WavefrontSizeLog2.

Referenced by llvm::SIRegisterInfo::eliminateFrameIndex(), llvm::GCNSubtarget::getKnownHighZeroBitsForFrameIndex(), and llvm::GCNTTIImpl::isAlwaysUniform().

◆ getWavesPerEU() [1/2]

std::pair< unsigned, unsigned > llvm::AMDGPUSubtarget::getWavesPerEU ( const Function & F ) const

inline

Returns: Subtarget's default pair of minimum/maximum number of waves per execution unit for function F, or minimum/maximum number of waves per execution unit explicitly requested using "amdgpu-waves-per-eu" attribute attached to function F.; Subtarget's default values if explicitly requested values cannot be converted to integer, violate subtarget's specifications, or are not compatible with minimum/maximum number of waves limited by flat work group size, register usage, and/or lds usage.

Definition at line 98 of file AMDGPUSubtarget.h.

References F, getFlatWorkGroupSizes(), and getWavesPerEU().

Referenced by llvm::GCNSubtarget::getMaxNumSGPRs(), llvm::GCNSubtarget::getMaxNumVGPRs(), and getWavesPerEU().