LLVM 22.0.0git
|
Class for storing and accessing the IR2Vec vocabulary. More...
#include "llvm/Analysis/IR2Vec.h"
Public Types | |
enum class | CanonicalTypeID : unsigned { FloatTy , VoidTy , LabelTy , MetadataTy , VectorTy , TokenTy , IntegerTy , FunctionTy , PointerTy , StructTy , ArrayTy , UnknownTy , MaxCanonicalType } |
Canonical type IDs supported by IR2Vec Vocabulary. More... | |
enum class | OperandKind : unsigned { FunctionID , PointerID , ConstantID , VariableID , MaxOperandKind } |
Operand kinds supported by IR2Vec Vocabulary. More... | |
using | const_iterator = VocabVector::const_iterator |
Const Iterator type aliases. | |
Public Member Functions | |
Vocabulary ()=default | |
LLVM_ABI | Vocabulary (VocabVector &&Vocab) |
LLVM_ABI bool | isValid () const |
LLVM_ABI unsigned | getDimension () const |
LLVM_ABI const ir2vec::Embedding & | operator[] (unsigned Opcode) const |
Accessors to get the embedding for a given entity. | |
LLVM_ABI const ir2vec::Embedding & | operator[] (Type::TypeID TypeId) const |
LLVM_ABI const ir2vec::Embedding & | operator[] (const Value &Arg) const |
const_iterator | begin () const |
const_iterator | cbegin () const |
const_iterator | end () const |
const_iterator | cend () const |
LLVM_ABI bool | invalidate (Module &M, const PreservedAnalyses &PA, ModuleAnalysisManager::Invalidator &Inv) const |
Static Public Member Functions | |
static constexpr size_t | getCanonicalSize () |
Total number of entries (opcodes + canonicalized types + operand kinds) | |
static LLVM_ABI StringRef | getVocabKeyForOpcode (unsigned Opcode) |
Function to get vocabulary key for a given Opcode. | |
static LLVM_ABI StringRef | getVocabKeyForTypeID (Type::TypeID TypeID) |
Function to get vocabulary key for a given TypeID. | |
static LLVM_ABI StringRef | getVocabKeyForOperandKind (OperandKind Kind) |
Function to get vocabulary key for a given OperandKind. | |
static LLVM_ABI OperandKind | getOperandKind (const Value *Op) |
Function to classify an operand into OperandKind. | |
static LLVM_ABI unsigned | getSlotIndex (unsigned Opcode) |
Functions to return the slot index or position of a given Opcode, TypeID, or OperandKind in the vocabulary. | |
static LLVM_ABI unsigned | getSlotIndex (Type::TypeID TypeID) |
static LLVM_ABI unsigned | getSlotIndex (const Value &Op) |
static LLVM_ABI StringRef | getStringKey (unsigned Pos) |
Returns the string key for a given index position in the vocabulary. | |
static LLVM_ABI VocabVector | createDummyVocabForTest (unsigned Dim=1) |
Create a dummy vocabulary for testing purposes. | |
Static Public Attributes | |
static constexpr unsigned | MaxTypeIDs = Type::TypeID::TargetExtTyID + 1 |
static constexpr unsigned | MaxCanonicalTypeIDs |
static constexpr unsigned | MaxOperandKinds |
Friends | |
class | llvm::IR2VecVocabAnalysis |
Class for storing and accessing the IR2Vec vocabulary.
The Vocabulary class manages seed embeddings for LLVM IR entities. The seed embeddings are the initial learned representations of the entities of LLVM IR. The IR2Vec representation for a given IR is derived from these seed embeddings.
The vocabulary contains the seed embeddings for three types of entities: instruction opcodes, types, and operands. Types are grouped/canonicalized for better learning (e.g., all float variants map to FloatTy). The vocabulary abstracts away the canonicalization effectively, the exposed APIs handle all the known LLVM IR opcodes, types and operands.
This class helps populate the seed embeddings in an internal vector-based ADT. It provides logic to map every IR entity to a specific slot index or position in this vector, enabling O(1) embedding lookup while avoiding unnecessary computations involving string based lookups while generating the embeddings.
using llvm::ir2vec::Vocabulary::const_iterator = VocabVector::const_iterator |
|
strong |
Canonical type IDs supported by IR2Vec Vocabulary.
Enumerator | |
---|---|
FloatTy | |
VoidTy | |
LabelTy | |
MetadataTy | |
VectorTy | |
TokenTy | |
IntegerTy | |
FunctionTy | |
PointerTy | |
StructTy | |
ArrayTy | |
UnknownTy | |
MaxCanonicalType |
|
strong |
Operand kinds supported by IR2Vec Vocabulary.
Enumerator | |
---|---|
FunctionID | |
PointerID | |
ConstantID | |
VariableID | |
MaxOperandKind |
|
default |
Vocabulary::Vocabulary | ( | VocabVector && | Vocab | ) |
Definition at line 263 of file IR2Vec.cpp.
|
inline |
|
inline |
|
inline |
|
static |
Create a dummy vocabulary for testing purposes.
Definition at line 369 of file IR2Vec.cpp.
References _, MaxCanonicalTypeIDs, MaxOperandKinds, and llvm::seq().
|
inline |
|
inlinestaticconstexpr |
unsigned Vocabulary::getDimension | ( | ) | const |
Definition at line 270 of file IR2Vec.cpp.
References assert().
Referenced by llvm::FunctionPropertiesInfo::getFunctionPropertiesInfo().
|
static |
Function to classify an operand into OperandKind.
Definition at line 338 of file IR2Vec.cpp.
References ConstantID, FunctionID, PointerID, and VariableID.
Referenced by getSlotIndex().
Definition at line 285 of file IR2Vec.cpp.
References assert(), getOperandKind(), MaxCanonicalTypeIDs, and MaxOperandKinds.
|
static |
Definition at line 280 of file IR2Vec.cpp.
References assert(), and MaxTypeIDs.
Functions to return the slot index or position of a given Opcode, TypeID, or OperandKind in the vocabulary.
Definition at line 275 of file IR2Vec.cpp.
References assert().
Referenced by operator[]().
Returns the string key for a given index position in the vocabulary.
This is useful for debugging or printing the vocabulary. Do not use this for embedding generation as string based lookups are inefficient.
Definition at line 348 of file IR2Vec.cpp.
References assert(), getVocabKeyForOpcode(), getVocabKeyForOperandKind(), and MaxCanonicalTypeIDs.
Function to get vocabulary key for a given Opcode.
Definition at line 303 of file IR2Vec.cpp.
References assert().
Referenced by getStringKey().
|
static |
Function to get vocabulary key for a given OperandKind.
Definition at line 331 of file IR2Vec.cpp.
References assert(), and MaxOperandKinds.
Referenced by getStringKey().
|
static |
Function to get vocabulary key for a given TypeID.
Definition at line 327 of file IR2Vec.cpp.
bool Vocabulary::invalidate | ( | Module & | M, |
const PreservedAnalyses & | PA, | ||
ModuleAnalysisManager::Invalidator & | Inv | ||
) | const |
Definition at line 363 of file IR2Vec.cpp.
References llvm::PreservedAnalyses::getChecker().
bool Vocabulary::isValid | ( | ) | const |
Definition at line 266 of file IR2Vec.cpp.
Referenced by llvm::FunctionPropertiesInfo::getFunctionPropertiesInfo(), and llvm::IR2VecPrinterPass::run().
const ir2vec::Embedding & Vocabulary::operator[] | ( | const Value & | Arg | ) | const |
Definition at line 299 of file IR2Vec.cpp.
References getSlotIndex().
const Embedding & Vocabulary::operator[] | ( | Type::TypeID | TypeId | ) | const |
Definition at line 295 of file IR2Vec.cpp.
References getSlotIndex().
Accessors to get the embedding for a given entity.
Definition at line 291 of file IR2Vec.cpp.
References getSlotIndex().
|
friend |
|
staticconstexpr |
Definition at line 201 of file IR2Vec.h.
Referenced by createDummyVocabForTest(), getSlotIndex(), and getStringKey().
|
staticconstexpr |
Definition at line 203 of file IR2Vec.h.
Referenced by createDummyVocabForTest(), getSlotIndex(), and getVocabKeyForOperandKind().
|
staticconstexpr |
Definition at line 200 of file IR2Vec.h.
Referenced by getSlotIndex().