Base class for all quantized types known to this dialect. More...
#include "mlir/Dialect/Quant/IR/QuantTypes.h"
Inheritance diagram for mlir::quant::QuantizedType:
Public Types | |
| using | ImplType = detail::QuantizedTypeStorage |
| Public Types inherited from mlir::Type | |
| template<typename ConcreteType , typename BaseType , typename StorageType , template< typename T > class... Traits> | |
| using | TypeBase = detail::StorageUserBase< ConcreteType, BaseType, StorageType, detail::TypeUniquer, Traits... > |
| Utility class for implementing types. More... | |
| using | ImplType = TypeStorage |
| using | AbstractTy = AbstractType |
Public Member Functions | |
| Type | getExpressedType () const |
| Gets the original expressed type that this quantized type approximates. More... | |
| unsigned | getFlags () const |
| Gets the flags associated with this type. More... | |
| bool | isSigned () const |
| Whether the storage type should be interpreted as a signed quantity (true) or an unsigned value (false). More... | |
| Type | getStorageType () const |
| Gets the underlying type used for to store values. More... | |
| int64_t | getStorageTypeMin () const |
| The minimum value that storageType can take. More... | |
| int64_t | getStorageTypeMax () const |
| The maximum value that storageType can take. More... | |
| bool | hasStorageTypeBounds () const |
| Return whether the storage type has explicit min or max boundaries different from the minimum and maximum representable values. More... | |
| unsigned | getStorageTypeIntegralWidth () const |
| Gets the integral bit width that the underlying storage type can exactly represent. More... | |
| bool | isCompatibleExpressedType (Type candidateExpressedType) |
| Returns whether the candidateExpressedType is a match for this QuantizedType. More... | |
| Type | castFromStorageType (Type candidateType) |
| Casts from a type based on the storageType to a corresponding type based on this type (returns nullptr if the cast is not valid). More... | |
| Type | castFromExpressedType (Type candidateType) |
| Casts from a type based on the expressedType to a corresponding type based on this type (returns nullptr if the cast is not valid). More... | |
| Type | castExpressedToStorageType (Type candidateType) |
| Casts from a type based on the expressedType to the equivalent type based on storageType by way of this QuantizedType. More... | |
| constexpr | Type ()=default |
| Type (const ImplType *impl) | |
| Type (const Type &other)=default | |
| Public Member Functions inherited from mlir::Type | |
| constexpr | Type ()=default |
| Type (const ImplType *impl) | |
| Type (const Type &other)=default | |
| Type & | operator= (const Type &other)=default |
| bool | operator== (Type other) const |
| bool | operator!= (Type other) const |
| operator bool () const | |
| bool | operator! () const |
| template<typename... Tys> | |
| bool | isa () const |
| template<typename... Tys> | |
| bool | isa_and_nonnull () const |
| template<typename U > | |
| U | dyn_cast () const |
| template<typename U > | |
| U | dyn_cast_or_null () const |
| template<typename U > | |
| U | cast () const |
| TypeID | getTypeID () |
| Return a unique identifier for the concrete type. More... | |
| MLIRContext * | getContext () const |
| Return the MLIRContext in which this type was uniqued. More... | |
| Dialect & | getDialect () const |
| Get the dialect this type is registered to. More... | |
| bool | isIndex () const |
| bool | isFloat4E2M1FN () const |
| bool | isFloat6E2M3FN () const |
| bool | isFloat6E3M2FN () const |
| bool | isFloat8E5M2 () const |
| bool | isFloat8E4M3 () const |
| bool | isFloat8E4M3FN () const |
| bool | isFloat8E5M2FNUZ () const |
| bool | isFloat8E4M3FNUZ () const |
| bool | isFloat8E4M3B11FNUZ () const |
| bool | isFloat8E3M4 () const |
| bool | isFloat8E8M0FNU () const |
| bool | isBF16 () const |
| bool | isF16 () const |
| bool | isTF32 () const |
| bool | isF32 () const |
| bool | isF64 () const |
| bool | isF80 () const |
| bool | isF128 () const |
| bool | isInteger () const |
| Return true if this is an integer type (with the specified width). More... | |
| bool | isInteger (unsigned width) const |
| Return true if this is an integer type with the specified width. More... | |
| bool | isSignlessInteger () const |
| Return true if this is a signless integer type (with the specified width). More... | |
| bool | isSignlessInteger (unsigned width) const |
| bool | isSignedInteger () const |
| Return true if this is a signed integer type (with the specified width). More... | |
| bool | isSignedInteger (unsigned width) const |
| bool | isUnsignedInteger () const |
| Return true if this is an unsigned integer type (with the specified width). More... | |
| bool | isUnsignedInteger (unsigned width) const |
| unsigned | getIntOrFloatBitWidth () const |
| Return the bit width of an integer or a float type, assert failure on other types. More... | |
| bool | isSignlessIntOrIndex () const |
| Return true if this is a signless integer or index type. More... | |
| bool | isSignlessIntOrIndexOrFloat () const |
| Return true if this is a signless integer, index, or float type. More... | |
| bool | isSignlessIntOrFloat () const |
| Return true of this is a signless integer or a float type. More... | |
| bool | isIntOrIndex () const |
| Return true if this is an integer (of any signedness) or an index type. More... | |
| bool | isIntOrFloat () const |
| Return true if this is an integer (of any signedness) or a float type. More... | |
| bool | isIntOrIndexOrFloat () const |
| Return true if this is an integer (of any signedness), index, or float type. More... | |
| void | print (raw_ostream &os) const |
| Print the current type. More... | |
| void | print (raw_ostream &os, AsmState &state) const |
| void | dump () const |
| const void * | getAsOpaquePointer () const |
| Methods for supporting PointerLikeTypeTraits. More... | |
| template<typename InterfaceT > | |
| bool | hasPromiseOrImplementsInterface () |
Returns true if InterfaceT has been promised by the dialect or implemented. More... | |
| template<template< typename T > class Trait> | |
| bool | hasTrait () |
| Returns true if the type was registered with a particular trait. More... | |
| const AbstractTy & | getAbstractType () const |
| Return the abstract type descriptor for this type. More... | |
| ImplType * | getImpl () const |
| Return the Type implementation. More... | |
| void | walkImmediateSubElements (function_ref< void(Attribute)> walkAttrsFn, function_ref< void(Type)> walkTypesFn) const |
| Walk all of the immediately nested sub-attributes and sub-types. More... | |
| auto | replaceImmediateSubElements (ArrayRef< Attribute > replAttrs, ArrayRef< Type > replTypes) const |
| Replace the immediately nested sub-attributes and sub-types with those provided. More... | |
| template<WalkOrder Order = WalkOrder::PostOrder, typename... WalkFns> | |
| auto | walk (WalkFns &&...walkFns) |
| Walk this type and all attibutes/types nested within using the provided walk functions. More... | |
| template<typename... ReplacementFns> | |
| auto | replace (ReplacementFns &&...replacementFns) |
| Recursively replace all of the nested sub-attributes and sub-types using the provided map functions. More... | |
Static Public Member Functions | |
| static LogicalResult | verifyInvariants (function_ref< InFlightDiagnostic()> emitError, unsigned flags, Type storageType, Type expressedType, int64_t storageTypeMin, int64_t storageTypeMax) |
| static bool | classof (Type type) |
| Support method to enable LLVM-style type casting. More... | |
| static int64_t | getDefaultMinimumForInteger (bool isSigned, unsigned integralWidth) |
| Gets the minimum possible stored by a storageType. More... | |
| static int64_t | getDefaultMaximumForInteger (bool isSigned, unsigned integralWidth) |
| Gets the maximum possible stored by a storageType. More... | |
| static QuantizedType | getQuantizedElementType (Type primitiveOrContainerType) |
| Returns the element type as a QuantizedType or nullptr if it is not a quantized type. More... | |
| static Type | castToStorageType (Type quantizedType) |
| Casts from a type based on a QuantizedType to a corresponding type based on the storageType (returns nullptr if the cast is not valid). More... | |
| static Type | castToExpressedType (Type quantizedType) |
| Casts from a type based on QuantizedType to a corresponding type based on the expressedType (returns nullptr if the cast is not valid). More... | |
| Static Public Member Functions inherited from mlir::Type | |
| static Type | getFromOpaquePointer (const void *pointer) |
Static Public Attributes | |
| static constexpr unsigned | MaxStorageBits = 32 |
| The maximum number of bits supported for storage types. More... | |
Additional Inherited Members | |
| Protected Attributes inherited from mlir::Type | |
| ImplType * | impl {nullptr} |
Detailed Description
Base class for all quantized types known to this dialect.
All quantized types have:
- storageType: The (narrower) numeric type that is being used to approximate some expressed type.
- expressedType: The type that is being approximated.
The base class provides generic support for manipulating the types based on these fields.
Definition at line 49 of file QuantTypes.h.
Member Typedef Documentation
◆ ImplType
Definition at line 51 of file QuantTypes.h.
Member Function Documentation
◆ castExpressedToStorageType()
Casts from a type based on the expressedType to the equivalent type based on storageType by way of this QuantizedType.
Equivalent to: QuantizedType::castToStorageType(castFromExpressedType(candidateType)) (but with validity checks). Example (for this = !quant.uniform<i8:f32, 1.0>): tensor<4xf32> -> tensor<4xi8>
Definition at line 237 of file QuantTypes.cpp.
References castToStorageType().
◆ castFromExpressedType()
Casts from a type based on the expressedType to a corresponding type based on this type (returns nullptr if the cast is not valid).
Examples: f32 -> !quant.uniform<i8:f32, 1.0> tensor<4xf32> -> tensor<4x!quant.uniform<i8:f32, 1.0>> vector<4xf32> -> vector<4x!quant.uniform<i8:f32, 1.0>>
Definition at line 182 of file QuantTypes.cpp.
References mlir::get().
◆ castFromStorageType()
Casts from a type based on the storageType to a corresponding type based on this type (returns nullptr if the cast is not valid).
Examples: i8 -> !quant.uniform<i8:f32, 1.0> tensor<4xi8> -> tensor<4x!quant.uniform<i8:f32, 1.0}>> vector<4xi8> -> vector<4x!quant.uniform<i8:f32, 1.0>>
Definition at line 131 of file QuantTypes.cpp.
References mlir::get(), and getShape().
◆ castToExpressedType()
Casts from a type based on QuantizedType to a corresponding type based on the expressedType (returns nullptr if the cast is not valid).
This is the inverse of castFromExpressedType.
Definition at line 210 of file QuantTypes.cpp.
References mlir::get().
Referenced by mlirQuantizedTypeCastToExpressedType().
◆ castToStorageType()
Casts from a type based on a QuantizedType to a corresponding type based on the storageType (returns nullptr if the cast is not valid).
This is the inverse of castFromStorageType().
Definition at line 155 of file QuantTypes.cpp.
References mlir::get().
Referenced by castExpressedToStorageType(), and mlirQuantizedTypeCastToStorageType().
◆ classof()
|
static |
Support method to enable LLVM-style type casting.
Definition at line 43 of file QuantTypes.cpp.
References mlir::Type::getDialect().
◆ getDefaultMaximumForInteger()
|
inlinestatic |
Gets the maximum possible stored by a storageType.
storageTypeMax must be less than or equal to this value.
Definition at line 77 of file QuantTypes.h.
References isSigned().
Referenced by mlirQuantizedTypeGetDefaultMaximumForInteger(), and parseStorageRange().
◆ getDefaultMinimumForInteger()
|
inlinestatic |
Gets the minimum possible stored by a storageType.
storageTypeMin must be greater than or equal to this value.
Definition at line 67 of file QuantTypes.h.
References isSigned().
Referenced by mlirQuantizedTypeGetDefaultMinimumForInteger(), and parseStorageRange().
◆ getExpressedType()
| Type QuantizedType::getExpressedType | ( | ) | const |
Gets the original expressed type that this quantized type approximates.
Note that this presumes that the quantized type was always derived from a floating point type, which in the broadest definition, is not true (i.e. it could be some form of integral, fixed type or affine type in its own right); however, at the high level, no examples of such usage are presently known and the restriction serves some useful purposes (such as always being able to reverse a transformation or measure error). In most cases, this will be f32.
Definition at line 109 of file QuantTypes.cpp.
Referenced by mlir::quant::ExpressedToQuantizedConverter::convert().
◆ getFlags()
| unsigned QuantizedType::getFlags | ( | ) | const |
Gets the flags associated with this type.
Typically a more specific accessor is appropriate.
Definition at line 39 of file QuantTypes.cpp.
Referenced by isSigned().
◆ getQuantizedElementType()
|
static |
Returns the element type as a QuantizedType or nullptr if it is not a quantized type.
If the type is primitive, returns that. If it is a container (vector/tensor), return the element type. Examples: !quant.uniform<i8:f32, 1.0> -> !quant.uniform<i8:f32, 1.0> tensor<4x!quant.uniform<i8:f32, 1.0> -> quant.uniform<i8:f32, 1.0>
Definition at line 122 of file QuantTypes.cpp.
Referenced by mlirQuantizedTypeGetQuantizedElementType().
◆ getStorageType()
| Type QuantizedType::getStorageType | ( | ) | const |
Gets the underlying type used for to store values.
Note that this may be signed or unsigned. Use the isSigned() accessor to differentiate.
Definition at line 80 of file QuantTypes.cpp.
◆ getStorageTypeIntegralWidth()
| unsigned QuantizedType::getStorageTypeIntegralWidth | ( | ) | const |
Gets the integral bit width that the underlying storage type can exactly represent.
For integral storage types, this will just be their width.
Definition at line 103 of file QuantTypes.cpp.
Referenced by printStorageType().
◆ getStorageTypeMax()
| int64_t QuantizedType::getStorageTypeMax | ( | ) | const |
The maximum value that storageType can take.
Definition at line 88 of file QuantTypes.cpp.
Referenced by printStorageType().
◆ getStorageTypeMin()
| int64_t QuantizedType::getStorageTypeMin | ( | ) | const |
The minimum value that storageType can take.
Definition at line 84 of file QuantTypes.cpp.
Referenced by printStorageType().
◆ hasStorageTypeBounds()
| bool QuantizedType::hasStorageTypeBounds | ( | ) | const |
Return whether the storage type has explicit min or max boundaries different from the minimum and maximum representable values.
Definition at line 92 of file QuantTypes.cpp.
Referenced by printStorageType().
◆ isCompatibleExpressedType()
| bool QuantizedType::isCompatibleExpressedType | ( | Type | candidateExpressedType | ) |
Returns whether the candidateExpressedType is a match for this QuantizedType.
This will be true if the candidate type is either a primitive type or a container type whose element type equals this QuantizedType's expressed type. Examples of compatible candidateExpressedType: !quant.uniform<i8:f32, 1.0> =~ f32 !quant.uniform<i8:f32, 1.0> =~ tensor<4xf32>
Definition at line 113 of file QuantTypes.cpp.
◆ isSigned()
|
inline |
Whether the storage type should be interpreted as a signed quantity (true) or an unsigned value (false).
Definition at line 102 of file QuantTypes.h.
References getFlags(), and mlir::quant::QuantizationFlags::Signed.
Referenced by getDefaultMaximumForInteger(), getDefaultMinimumForInteger(), and printStorageType().
◆ Type() [1/3]
|
constexprdefault |
◆ Type() [2/3]
◆ Type() [3/3]
|
default |
◆ verifyInvariants()
|
static |
Definition at line 48 of file QuantTypes.cpp.
References mlir::emitError(), and mlir::quant::QuantizationFlags::Signed.
Referenced by mlir::quant::UniformQuantizedPerAxisType::verifyInvariants(), mlir::quant::UniformQuantizedType::verifyInvariants(), and mlir::quant::AnyQuantizedType::verifyInvariants().
Member Data Documentation
◆ MaxStorageBits
|
staticconstexpr |
The maximum number of bits supported for storage types.
Definition at line 55 of file QuantTypes.h.
Referenced by parseStorageType().
The documentation for this class was generated from the following files:
- include/mlir/Dialect/Quant/IR/QuantTypes.h
- lib/Dialect/Quant/IR/QuantTypes.cpp
