My Project
|
#include <QuantTypes.h>
Public Member Functions | |
double | getScale () const |
int64_t | getZeroPoint () const |
bool | isFixedPoint () const |
Static Public Member Functions | |
static UniformQuantizedType | get (unsigned flags, Type storageType, Type expressedType, double scale, int64_t zeroPoint, int64_t storageTypeMin, int64_t storageTypeMax) |
static UniformQuantizedType | getChecked (unsigned flags, Type storageType, Type expressedType, double scale, int64_t zeroPoint, int64_t storageTypeMin, int64_t storageTypeMax, Location location) |
static LogicalResult | verifyConstructionInvariants (Optional< Location > loc, MLIRContext *context, unsigned flags, Type storageType, Type expressedType, double scale, int64_t zeroPoint, int64_t storageTypeMin, int64_t storageTypeMax) |
Verifies construction invariants and issues errors/warnings. More... | |
static bool | kindof (unsigned kind) |
Support method to enable LLVM-style type casting. More... | |
![]() | |
static ClassID * | getClassID () |
Return a unique identifier for the concrete type. More... | |
template<typename T > | |
static bool | classof (T val) |
Additional Inherited Members | |
![]() | |
using | Base = StorageUserBase< ConcreteT, BaseT, StorageT, UniquerT > |
Utility declarations for the concrete attribute class. More... | |
using | ImplType = StorageT |
![]() | |
ImplType * | getImpl () const |
Utility for easy access to the storage instance. More... | |
![]() | |
template<typename... Args> | |
static ConcreteT | get (MLIRContext *ctx, unsigned kind, Args... args) |
template<typename... Args> | |
static ConcreteT | getChecked (const Location &loc, MLIRContext *ctx, unsigned kind, Args... args) |
template<typename... Args> | |
static LogicalResult | verifyConstructionInvariants (Args... args) |
Default implementation that just returns success. More... | |
Represents a family of uniform, quantized types.
Each instance of this type expresses a mapping between real values (most often expressed in floating point f32) and quantized values (either fixed point or affine).
The relationship is: real_value = scale * (quantized_value - zero_point)
It is used as part of high level graph transformations that have the goal of re-expressing parts of a computation in terms of this common form for more efficient execution at runtime. In addition, it is designed to be expressive enough to facilitate lowering to precise types and operations in target hardware.
As a high-level type, focused on intermediate passes, this type holds opinions consistent with high-level usage. If lowering math kernels below the high level arithmetic ops (i.e. to LLVM IR or hardware specific instruction sets), it is expected that the information expressed here will be used to drive low level codegen and target specific type selection, but this type will likely be erased in the process.
Syntax synopsis: Per-layer, all parameters expressed: !quant<uniform[StorageType:ExpressedType]{Scale:ZeroPoint}> Per-layer, optional parameters omitted: !quant<uniform[StorageType]{Scale}>
StorageType: 'i'|'u' NumBits ExpressedType: 'f16', 'f32', 'bf16', 'f64' Scale: A legal double value ZeroPoint: An integer value
|
static |
Gets an instance of the type with all parameters specified but not checked.
|
static |
Gets an instance of the type with all specified parameters checked. Returns a nullptr convertible type on failure.
double UniformQuantizedType::getScale | ( | ) | const |
Gets the scale term. The scale designates the difference between the real values corresponding to consecutive quantized values differing by 1.
int64_t UniformQuantizedType::getZeroPoint | ( | ) | const |
Gets the storage value corresponding to the real value 0 in the affine equation.
|
inline |
|
inlinestatic |
Support method to enable LLVM-style type casting.
|
static |
Verifies construction invariants and issues errors/warnings.