AMDGPU‐specific address‐space helpers and DWARF mappings.

Namespaces

Name

Description

DWARFAS

DWARF address‐space encodings for AMDGPU.

HSAMD

AMDGPU HSA metadata (HSAMD) definitions and verification.

PALMD

PAL (Platform Abstraction Layer) metadata definitions.

impl

Implementation details for AMDGPU address‐space mapping.

Types

Name

Description

IsaVersion

Instruction set architecture version.

TargetID

AMDGPU target identity: processor plus optional xnack/sramecc settings.

Type Aliases

Name

Description

AMDGPUFeatureBitset

Bitset of enabled AMDGPU frontend‐visible features.

Enums

Name

Description

Unnamed enum

Constants for GPUs affected by the VI SGPR initialization bug.

AMDGPUFeature

One enumerator per frontend‐visible feature bit; NUM_FEATURES is the count.

ArchFeatureKind

GFX6+ architecture feature flags needed by the frontend driver.

FeatureError

Result of validating AMDGPU target feature combinations.

FlatAddrSpace

Identifies which FLAT address‐space segment an instruction operates on.

FuncInfoFlags

Per‐function flags packed into INFO_FLAGS entries.

GPUKind

GPU kinds supported by the AMDGPU target.

InfoKind

Entry kind values for the .amdgpu.info section.

R600FeatureKind

R600 feature flags needed by the frontend driver.

TargetIDSetting

Setting of an optional TargetID feature such as xnack or sramecc.

Functions

Name

Description

fillAMDGPUFeatureMap

Fills Features with default values for the given target GPU.

fillValidArchListAMDGCN

Append the valid AMDGCN GPU names to Values.

fillValidArchListR600

Append the valid R600 GPU names to Values.

getAddressableNumSGPRs

getAddressableNumSGPRs overloads

getArchAttrAMDGCN [deprecated]

getArchAttrAMDGCN overloads

getArchAttrR600

Returns the R600FeatureKind bitfield for GPU kind AK.

getArchFamilyNameAMDGCN

Returns the architecture family name for AMDGCN GPU kind AK.

getArchNameAMDGCN

Returns the canonical AMDGCN architecture name for GPU kind AK.

getArchNameFromSubArch

Returns the canonical GPU name for an AMDGPU subarch.

getArchNameR600

Returns the canonical R600 architecture name for GPU kind AK.

getBaseArchNameAMDGCN

The canonical GPU name for a variant name.

getCanonicalArchName

Returns the canonical architecture name for Arch under triple T.

getFeatureBitset

Returns AK's feature bitset, or an empty bitset if unknown.

getFeatureNames

Appends the feature name of each bit set in Features to Names.

getGPUKindFromSubArch

Returns the representative GPUKind for subarch SubArch.

getIsaVersion

getIsaVersion overloads

getMajorSubArch

Returns the major‐family subarch for SubArch.

getMaxHWAddressableLocalMemorySize

getMaxHWAddressableLocalMemorySize overloads

getMaxWavesPerEU

Maximum number of waves per execution unit without any kind of limitation.

getMinWavesPerEU

Minimum number of waves per execution unit.

getNullPointerValue

Get the null pointer value for the given address space.

getNumWorkGroupSIMDs

Number of SIMDs a work‐group's waves run on.

getSGPRAllocGranule

getSGPRAllocGranule overloads

getSubArch

Returns the triple subarch for GPU kind AK.

getSubArchName

Returns the triple subarch name for an AMDGPU subarch, e.g. AMDGPUSubArch900 ‐> "amdgpu9.00". Returns "amdgpu" for NoSubArch.

getTotalNumSGPRs

getTotalNumSGPRs overloads

isCPUValidForSubArch

isCPUValidForSubArch overloads

isConstantAddressSpace

True if AS is an AMDGPU constant address space.

isExtendedGlobalAddrSpace

True if AS is an extended global address space.

isFlatGlobalAddrSpace

True if AS is flat, global, constant, or an extended global address space.

isPseudoTarget

isPseudoTarget overloads

isSubArchCompatible

isSubArchCompatible overloads

mapToDWARFAddrSpace

If LLVMAddrSpace has a corresponding DWARF encoding, return it; otherwise return the sentinel value ‐1 to indicate no such mapping exists.

mergeSubArch

Returns the effective triple appropriate to use when linking B into A by merging the subarches in case of inexact match.

parseArchAMDGCN

Parses an AMDGCN GPU name CPU into a GPUKind.

parseArchR600

Parses an R600 GPU name CPU into a GPUKind.

operator<<

Writes TargetID to OS.

Created with MrDocs