MLIR  21.0.0git
Public Types | Public Member Functions | Static Public Member Functions | Static Public Attributes | List of all members
mlir::quant::UniformQuantizedSubChannelType Class Reference

Represents sub-channel (also known as blockwise quantization). More...

#include "mlir/Dialect/Quant/IR/QuantTypes.h"

+ Inheritance diagram for mlir::quant::UniformQuantizedSubChannelType:

Public Types

using Base = StorageUserBase< ConcreteT, BaseT, StorageT, UniquerT, Traits... >
 Utility declarations for the concrete attribute class. More...
 
- Public Types inherited from mlir::detail::StorageUserBase< ConcreteT, BaseT, StorageT, UniquerT, Traits >
using Base = StorageUserBase< ConcreteT, BaseT, StorageT, UniquerT, Traits... >
 Utility declarations for the concrete attribute class. More...
 
using ImplType = StorageT
 
using HasTraitFn = bool(*)(TypeID)
 

Public Member Functions

DenseElementsAttr getScales () const
 Gets the quantization scales. More...
 
DenseElementsAttr getZeroPoints () const
 Gets the quantization zero-points. More...
 
ArrayRef< int32_t > getQuantizedDimensions () const
 Gets the quantized dimensions. More...
 
ArrayRef< int64_t > getBlockSizes () const
 Gets the block sizes for the quantized dimensions. More...
 
const SmallVector< std::pair< int32_t, int64_t > > getBlockSizeInfo () const
 Gets the block size information. More...
 
- Public Member Functions inherited from mlir::detail::StorageUserBase< ConcreteT, BaseT, StorageT, UniquerT, Traits >
ImplTypegetImpl () const
 Utility for easy access to the storage instance. More...
 

Static Public Member Functions

static UniformQuantizedSubChannelType get (unsigned flags, Type storageType, Type expressedType, DenseElementsAttr scales, DenseElementsAttr zeroPoints, ArrayRef< int32_t > quantizedDimensions, ArrayRef< int64_t > blockSizes, int64_t storageTypeMin, int64_t storageTypeMax)
 Gets an instance of the type with all parameters specified but not checked. More...
 
static UniformQuantizedSubChannelType getChecked (function_ref< InFlightDiagnostic()> emitError, unsigned flags, Type storageType, Type expressedType, DenseElementsAttr scales, DenseElementsAttr zeroPoints, ArrayRef< int32_t > quantizedDimensions, ArrayRef< int64_t > blockSizes, int64_t storageTypeMin, int64_t storageTypeMax)
 Gets an instance of the type with all specified parameters checked. More...
 
static LogicalResult verifyInvariants (function_ref< InFlightDiagnostic()> emitError, unsigned flags, Type storageType, Type expressedType, DenseElementsAttr scales, DenseElementsAttr zeroPoints, ArrayRef< int32_t > quantizedDimensions, ArrayRef< int64_t > blockSizes, int64_t storageTypeMin, int64_t storageTypeMax)
 Verifies construction invariants and issues errors/warnings. More...
 
template<typename... Args>
static ConcreteT getChecked (const Location &loc, Args &&...args)
 Get or create a new ConcreteT instance within the ctx, defined at the given, potentially unknown, location. More...
 
template<typename... Args>
static ConcreteT getChecked (function_ref< InFlightDiagnostic()> emitErrorFn, MLIRContext *ctx, Args... args)
 Get or create a new ConcreteT instance within the ctx. More...
 
- Static Public Member Functions inherited from mlir::detail::StorageUserBase< ConcreteT, BaseT, StorageT, UniquerT, Traits >
static TypeID getTypeID ()
 Return a unique identifier for the concrete type. More...
 
template<typename T >
static bool classof (T val)
 Provide an implementation of 'classof' that compares the type id of the provided value with that of the concrete type. More...
 
static detail::InterfaceMap getInterfaceMap ()
 Returns an interface map for the interfaces registered to this storage user. More...
 
static HasTraitFn getHasTraitFn ()
 Returns the function that returns true if the given Trait ID matches the IDs of any of the traits defined by the storage user. More...
 
static auto getWalkImmediateSubElementsFn ()
 Returns a function that walks immediate sub elements of a given instance of the storage user. More...
 
static auto getReplaceImmediateSubElementsFn ()
 Returns a function that replaces immediate sub elements of a given instance of the storage user. More...
 
template<typename... IfaceModels>
static void attachInterface (MLIRContext &context)
 Attach the given models as implementations of the corresponding interfaces for the concrete storage user class. More...
 
template<typename... Args>
static ConcreteT get (MLIRContext *ctx, Args &&...args)
 Get or create a new ConcreteT instance within the ctx. More...
 
template<typename... Args>
static ConcreteT getChecked (const Location &loc, Args &&...args)
 Get or create a new ConcreteT instance within the ctx, defined at the given, potentially unknown, location. More...
 
template<typename... Args>
static ConcreteT getChecked (function_ref< InFlightDiagnostic()> emitErrorFn, MLIRContext *ctx, Args... args)
 Get or create a new ConcreteT instance within the ctx. More...
 
static ConcreteT getFromOpaquePointer (const void *ptr)
 Get an instance of the concrete type from a void pointer. More...
 

Static Public Attributes

static constexpr StringLiteral name = "quant.uniform_sub_channel"
 

Additional Inherited Members

- Protected Member Functions inherited from mlir::detail::StorageUserBase< ConcreteT, BaseT, StorageT, UniquerT, Traits >
template<typename... Args>
LogicalResult mutate (Args &&...args)
 Mutate the current storage instance. More...
 
- Static Protected Member Functions inherited from mlir::detail::StorageUserBase< ConcreteT, BaseT, StorageT, UniquerT, Traits >
template<typename... Args>
static LogicalResult verifyInvariants (Args... args)
 Default implementation that just returns success. More...
 

Detailed Description

Represents sub-channel (also known as blockwise quantization).

Syntax synopsis: UniformQuantizedSubChannelType ::= '!quant.uniform' '<' storageType ('<' storageMin ':' storageMax '>')? ':' expressedType ':' BlockSizeInfo ',' ScaleZeroTensor '>' BlockSizeInfo: '{' '}' | '{' AxisBlock (',' AxisBlock)* '}' AxisBlock ::= AxisSpec ':' BlockSizeSpec ScaleZeroTensor ::= ScaleZeroDenseExp | ScaleZeroList ScaleZeroDenseExp ::= '{' ScaleZeroTensor (',' ScaleZeroTensor)* '}' ScaleZeroList ::= ScaleZero (',' ScaleZero)* ScaleZero ::= Scale (':' ZeroPoint)?

StorageType: 'i'|'u' NumBits ExpressedType: 'f16', 'f32', 'bf16', 'f64' AxisSpec: An integer value BlockSizeSpec: An integer value Scale: An attribute (usually floating-point value) ZeroPoint: An attribute (usually integer value)

Definition at line 405 of file QuantTypes.h.

Member Typedef Documentation

◆ Base

using mlir::detail::StorageUserBase< ConcreteT, BaseT, StorageT, UniquerT, Traits >::Base = StorageUserBase<ConcreteT, BaseT, StorageT, UniquerT, Traits...>

Utility declarations for the concrete attribute class.

Definition at line 100 of file StorageUniquerSupport.h.

Member Function Documentation

◆ get()

UniformQuantizedSubChannelType UniformQuantizedSubChannelType::get ( unsigned  flags,
Type  storageType,
Type  expressedType,
DenseElementsAttr  scales,
DenseElementsAttr  zeroPoints,
ArrayRef< int32_t >  quantizedDimensions,
ArrayRef< int64_t >  blockSizes,
int64_t  storageTypeMin,
int64_t  storageTypeMax 
)
static

Gets an instance of the type with all parameters specified but not checked.

Definition at line 413 of file QuantTypes.cpp.

References mlir::get(), and mlir::Type::getContext().

Referenced by mlirUniformQuantizedSubChannelTypeGet().

◆ getBlockSizeInfo()

const SmallVector< std::pair< int32_t, int64_t > > UniformQuantizedSubChannelType::getBlockSizeInfo ( ) const

Gets the block size information.

This returns a list of pairs, where each pair represents a quantized dimension and its corresponding block size.

For example, for the type: tensor<8x4x!quant.uniform<i8:f32:{1:2, 0:8}, {{2.0, 3.0}}>

This method returns: [(1, 2), (0, 8)]

This list indicates that axis 1 has a block size of 2, and axis 0 has a block size of 8.

Definition at line 518 of file QuantTypes.cpp.

Referenced by printUniformQuantizedSubChannelType().

◆ getBlockSizes()

ArrayRef< int64_t > UniformQuantizedSubChannelType::getBlockSizes ( ) const

Gets the block sizes for the quantized dimensions.

The i-th element in the returned list corresponds to the block size for the i-th dimension in the list returned by getQuantizedDimensions().

See getQuantizedDimensions() for more details and examples.

Definition at line 513 of file QuantTypes.cpp.

◆ getChecked() [1/3]

template<typename... Args>
static ConcreteT mlir::detail::StorageUserBase< ConcreteT, BaseT, StorageT, UniquerT, Traits >::getChecked ( typename...  Args)
inlinestatic

Get or create a new ConcreteT instance within the ctx, defined at the given, potentially unknown, location.

If the arguments provided are invalid, errors are emitted using the provided location and a null object is returned.

Definition at line 189 of file StorageUniquerSupport.h.

◆ getChecked() [2/3]

UniformQuantizedSubChannelType UniformQuantizedSubChannelType::getChecked ( function_ref< InFlightDiagnostic()>  emitError,
unsigned  flags,
Type  storageType,
Type  expressedType,
DenseElementsAttr  scales,
DenseElementsAttr  zeroPoints,
ArrayRef< int32_t >  quantizedDimensions,
ArrayRef< int64_t >  blockSizes,
int64_t  storageTypeMin,
int64_t  storageTypeMax 
)
static

Gets an instance of the type with all specified parameters checked.

Returns a nullptr convertible type on failure.

Definition at line 423 of file QuantTypes.cpp.

References mlir::emitError(), and mlir::Type::getContext().

◆ getChecked() [3/3]

template<typename... Args>
static ConcreteT mlir::detail::StorageUserBase< ConcreteT, BaseT, StorageT, UniquerT, Traits >::getChecked ( typename...  Args)
inlinestatic

Get or create a new ConcreteT instance within the ctx.

If the arguments provided are invalid, errors are emitted using the provided emitError and a null object is returned.

Definition at line 198 of file StorageUniquerSupport.h.

◆ getQuantizedDimensions()

ArrayRef< int32_t > UniformQuantizedSubChannelType::getQuantizedDimensions ( ) const

Gets the quantized dimensions.

Each element in the returned list represents an axis of the quantized data tensor that has a specified block size. The order of elements corresponds to the order of block sizes returned by getBlockSizes().

It means that the data tensor is quantized along the i-th dimension in the returned list using the i-th block size from getBlockSizes().

Note that the type expression does not have to specify the block size for all axes in the data tensor. Any unspecified block size for an axis i defaults to the tensor dimension size of that axis.

For example, for a quantized type: tensor<8x4x2x!quant.uniform<i8:f32:{1:2, 0:8}, {{1.0, 2.0}, {3.0, 4.0}}>

getQuantizedDimensions() returns [1, 0]. getBlockSizes() returns [2, 8].

This indicates that:

  • Axis 1 (second dimension) is quantized with a block size of 2.
  • Axis 0 (first dimension) is quantized with a block size of 8. Since axis 2 is not specified, it implicitly has a block size equal to the size of the third dimension (which is 2 in this case).

Definition at line 509 of file QuantTypes.cpp.

◆ getScales()

DenseElementsAttr UniformQuantizedSubChannelType::getScales ( ) const

Gets the quantization scales.

The scales are organized in a multi-dimensional tensor. The size of each dimension in the scales tensor is determined by the number of blocks along the corresponding dimension in the quantized data tensor.

For example, if the quantized data tensor has shape [X0, X1, ..., XR-1] and the block sizes are [B0, B1, ..., BR-1], then the scales tensor will have shape [X0/B0, X1/B1, ..., XR-1/BR-1].

The scale value for a specific element in the quantized data tensor at position [i0, i1, ..., iR-1] is determined by accessing the corresponding element in the scales tensor at position [i0/B0, i1/B1, ..., iR-1/BR-1].

Definition at line 500 of file QuantTypes.cpp.

Referenced by printUniformQuantizedSubChannelType().

◆ getZeroPoints()

DenseElementsAttr UniformQuantizedSubChannelType::getZeroPoints ( ) const

Gets the quantization zero-points.

The zero-points are organized in a multi-dimensional tensor. The size of each dimension in the zero-point tensor is determined by the number of blocks along the corresponding dimension in the quantized data tensor.

For example, if the quantized data tensor has shape [X0, X1, ..., XR-1] and the block sizes are [B0, B1, ..., BR-1], then the zero-point tensor will have shape [X0/B0, X1/B1, ..., XR-1/BR-1].

The zero-point value for a specific element in the quantized data tensor at position [i0, i1, ..., iR-1] is determined by accessing the corresponding element in the zero-point tensor at position [i0/B0, i1/B1, ..., iR-1/BR-1].

Definition at line 504 of file QuantTypes.cpp.

Referenced by printUniformQuantizedSubChannelType().

◆ verifyInvariants()

LogicalResult UniformQuantizedSubChannelType::verifyInvariants ( function_ref< InFlightDiagnostic()>  emitError,
unsigned  flags,
Type  storageType,
Type  expressedType,
DenseElementsAttr  scales,
DenseElementsAttr  zeroPoints,
ArrayRef< int32_t >  quantizedDimensions,
ArrayRef< int64_t >  blockSizes,
int64_t  storageTypeMin,
int64_t  storageTypeMax 
)
static

Verifies construction invariants and issues errors/warnings.

Definition at line 435 of file QuantTypes.cpp.

References mlir::emitError(), mlir::DenseElementsAttr::getType(), mlir::DenseElementsAttr::size(), and mlir::quant::QuantizedType::verifyInvariants().

Member Data Documentation

◆ name

constexpr StringLiteral mlir::quant::UniformQuantizedSubChannelType::name = "quant.uniform_sub_channel"
staticconstexpr

Definition at line 412 of file QuantTypes.h.


The documentation for this class was generated from the following files: