'tfl' диалект

Диалект TensorFlow Lite.

Этот диалект соответствует операциям TensorFlow Lite.

Инварианты:

• Все значения имеют тип Tensor (в частности, скаляры представляются нульмерными тензорами);

Определение операции

tfl.abs (::mlir::TFL::AbsOp)

Оператор абсолютного значения

Учитывая тензор x , эта операция возвращает тензор, содержащий абсолютное значение каждого элемента в x . Например, если x — входной элемент, а y — выходной элемент, эта операция вычисляет \(y = |x|\).

Черты: AlwaysSpeculatableImplTrait, QuantizableResult, SameOperandsAndResultShape

Интерфейсы: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Операнды:

Полученные результаты:

tfl.add_n (::mlir::TFL::AddNOp)

оператор add_n

Складывает все входные тензоры поэлементно.

Черты: AlwaysSpeculatableImplTrait, Коммутативный

Интерфейсы: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflArithmeticCountOpInterface, TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Операнды:

Полученные результаты:

tfl.add (::mlir::TFL::AddOp)

Оператор сложения

Поэлементная операция сложения.

Черты: ::mlir::OpTrait::TFLRuntimeOpTrait, AlwaysSpeculatableImplTrait, Commutative, QuantizableResult, ResultsBroadcastableShape

Интерфейсы: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflArithmeticCountOpInterface, TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Атрибуты:

Операнды:

Полученные результаты:

tfl.arg_max (::mlir::TFL::ArgMaxOp)

Оператор ArgMax

Возвращает индекс с наибольшим значением по измерениям тензора.

Черты: AlwaysSpeculatableImplTrait, QuantizableResult

Интерфейсы: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Атрибуты:

Операнды:

Полученные результаты:

tfl.arg_min (::mlir::TFL::ArgMinOp)

Оператор ArgMin

Возвращает индекс с наименьшим значением по измерениям тензора. a = [1, 10, 26,9, 2,8, 166,32, 62,3] b = tf.math.argmin(input = a) c = tf.keras.backend.eval(b)

Черты: AlwaysSpeculatableImplTrait, QuantizableResult

Интерфейсы: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Атрибуты:

Операнды:

Полученные результаты:

tfl.assign_variable (::mlir::TFL::AssignVariableOp)

Присваивает новое значение переменной.

Любой ReadVariableOp с управляющей зависимостью от этой операции гарантированно возвращает это значение или последующее более новое значение переменной.

Интерфейсы: TflRuntimeVerifyOpInterface

Операнды:

tfl.atan2 (::mlir::TFL::Atan2Op)

Операция Атан2

Операция "atan2" вычисляет арктангенс y/x поэлементно, учитывая знаки аргументов.

Черты: AlwaysSpeculatableImplTrait, SameOperandsAndResultElementType, SameOperandsAndResultShape

Интерфейсы: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Операнды:

Полученные результаты:

tfl.average_pool_2d (::mlir::TFL::AveragePool2DOp)

Оператор Average_pool_2d

Выполняет операцию объединения средних значений на входе.

Черты: AlwaysSpeculatableImplTrait, QuantizableResult

Интерфейсы: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflArithmeticCountOpInterface, TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Атрибуты:

Операнды:

Полученные результаты:

tfl.basic_lstm (::mlir::TFL::BasicLSTMOp)

Основной оператор lstm

базовый оператор сотовой связи LSTM.

Черты: AlwaysSpeculatableImplTrait, QuantizableResult

Интерфейсы: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Атрибуты:

Операнды:

Полученные результаты:

tfl.batch_matmul (::mlir::TFL::BatchMatMulOp)

Пакетный оператор умножения матриц

Выполняет пакетное матричное умножение на входах. Соответствует соглашениям TensorFlow BatchMatMulV2 с поддержкой неизвестных измерений в пакетных измерениях и широковещании.

Inputs:
  `inputs[0]`: required: input LHS
  `inputs[1]`: required: input RHS
  `adjoint_lhs`: optional: Transpose LHS (default false)
  `adjoint_lhs`: optional: Transpose LHS (default false)

Особенности: ::mlir::OpTrait::TFLRuntimeOpTrait, AlwaysSpeculatableImplTrait, QuantizableResult

Интерфейсы: ConditionallySpeculatable, DynamicRangeQuantizedOpInterface, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Атрибуты:

Операнды:

Полученные результаты:

tfl.batch_to_space_nd (::mlir::TFL::BatchToSpaceNdOp)

Оператор BatchToSpaceNd

Эта операция преобразует "пакетное" измерение 0 в пространственные измерения.

Черты: AlwaysSpeculatableImplTrait, QuantizableResult

Интерфейсы: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Операнды:

Полученные результаты:

tfl.bidirectional_sequence_lstm (::mlir::TFL::BidirectionalSequenceLSTMOp)

Оператор двунаправленной последовательности lstm

Двунаправленный lstm — это, по сути, два lstms, один работает вперед, а другой — назад. А на выходе получается конкатенация двух lstms.

Черты: QuantizableResult

Интерфейсы: DynamicRangeQuantizedOpInterface, TFL_StatefulOp, TflRuntimeVerifyOpInterface

Атрибуты:

Операнды:

Полученные результаты:

tfl.broadcast_args (::mlir::TFL::BroadcastArgsOp)

Верните форму s0 op s1 с трансляцией.

Учитывая s0 и s1 , тензоры, которые представляют формы, вычисляют r0 , переданную форму. s0 , s1 и r0 — целочисленные векторы.

Черты: AlwaysSpeculatableImplTrait

Интерфейсы: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Операнды:

Полученные результаты:

tfl.broadcast_to (::mlir::TFL::BroadcastToOp)

Передайте массив для совместимой формы.

Вещание — это процесс создания массивов, имеющих совместимые формы для арифметических операций. Две фигуры совместимы, если для каждой пары измерений они либо равны, либо одна из них равна единице. При попытке передать тензор фигуре он начинает с конечных измерений и продвигается вперед.

Например,

x = tf.constant([1, 2, 3]) y = tf.broadcast_to(x, [3, 3]) print(y) tf.Tensor([[1 2 3] [1 2 3] [1 2 3]], форма=(3, 3), dtype=int32)

В приведенном выше примере входной Tensor с формой [1, 3] транслируется на выходной Tensor с формой [3, 3] .

При выполнении широковещательных операций, таких как умножение тензора на скаляр, широковещательная передача (обычно) дает некоторое преимущество во времени или пространстве, поскольку широковещательный тензор никогда не материализуется.

Тем не менее, broadcast_to не несет в себе таких преимуществ. Вновь созданный тензор берет всю память переданной формы. (Однако в контексте графа broadcast_to может быть объединена с последующей операцией, а затем оптимизирована.)

Черты: AlwaysSpeculatableImplTrait, QuantizableResult

Интерфейсы: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Операнды:

Полученные результаты:

tfl.bucketize (::mlir::TFL::BucketizeOp)

Разделяет «входные данные» на основе «границ».

Пример:

Если входными данными являются boundaries = [0, 10, 100] и input = [[-5, 10000][150, 10][5, 100]] , то вывод будет output = [[0, 3][3, 2][1, 3]] .

Черты: AlwaysSpeculatableImplTrait, SameOperandsAndResultShape

Интерфейсы: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Атрибуты:

Операнды:

Полученные результаты:

tfl.call_once (::mlir::TFL::CallOnceOp)

Вызывает функцию инициализации

Эта операция вызывает данную функцию инициализации для инициализатора сеанса на диалекте сохраненной модели.

Интерфейсы: TflRuntimeVerifyOpInterface

Атрибуты:

tfl.cast (::mlir::TFL::CastOp)

Приведение оператора

Преобразует ввод из типа ввода в тип вывода.

Черты: AlwaysSpeculatableImplTrait, SameOperandsAndResultShape

Интерфейсы: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Операнды:

Полученные результаты:

tfl.ceil (::mlir::TFL::CeilOp)

Потолочный оператор

Возвращает поэлементное значение ceil ввода.

Признаки: AlwaysSpeculatableImplTrait, InferTensorType, TF::SameOperandsAndResultTypeResolveRef

Интерфейсы: ConditionallySpeculatable, InferShapedTypeOpInterface, InferTypeOpInterface, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Операнды:

Полученные результаты:

tfl.complex_abs (::mlir::TFL::ComplexAbsOp)

Вычисляет комплексное абсолютное значение тензора.

Учитывая тензор комплексных чисел x , эта операция возвращает тензор типа float или double , который является абсолютным значением каждого элемента в x . Все элементы в x должны быть комплексными числами формы \(a + bj\). Абсолютное значение вычисляется как \( \sqrt{a^2 + b^2}\).

Черты: AlwaysSpeculatableImplTrait, SameOperandsAndResultShape

Интерфейсы: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Операнды:

Полученные результаты:

tfl.concatenation (::mlir::TFL::ConcatenationOp)

Оператор конкатенации

Объединяет тензоры по одному измерению

Черты: AlwaysSpeculatableImplTrait, QuantizableResult

Интерфейсы: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Атрибуты:

Операнды:

Полученные результаты:

tfl.pseudo_const (::mlir::TFL::ConstOp)

Постоянный псевдо ор.

Представляет постоянное значение на диалекте TensorFlow Lite. Это не фактическая операция, и вместо этого она будет понижена до буфера.

Оператору разрешено иметь все атрибуты того же типа, что и tf.Const (например, разрешены непрозрачные атрибуты TF).

Черты: AlwaysSpeculatableImplTrait, ConstantLike, FirstAttrDerivedResultType, QuantizableResult

Интерфейсы: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Атрибуты:

Полученные результаты:

tfl.control_node (::mlir::TFL::ControlNodeOp)

The `TFL.control_node` operation wraps single-block operations in order to attach control edges.

Это используется для переноса областей и прикрепления к ним управляющих зависимостей. Как правило, это происходит на одном из последних шагов перед созданием модели плоского буфера, чтобы обеспечить оптимизацию, основанную на фиксированном порядке операций (например, повторную материализацию). Средство экспорта плоского буфера разворачивает обернутую область и аннотирует сгенерированную модель метаданными. таким образом, что любые переупорядочения во время выполнения будут соблюдать порядок, заданный зависимостями элементов управления.

Особенности: HasParent mlir::func::FuncOp , RecursiveMemoryEffects, SingleBlockImplicitTerminator

Операнды:

Полученные результаты:

tfl.conv_2d (::mlir::TFL::Conv2DOp)

Оператор свертки

Выполняет операцию свертки над входными данными.

Входные данные: inputs[0] : обязательно: тензор активации ввода inputs[1] : обязательно: тензор веса фильтра inputs[2] : необязательно: тензор смещения

Черты: AlwaysSpeculatableImplTrait, QuantizableResult, quant::AccumulatorUniformScale<2, 0, 1>, quant::AffineOpCoefficient<0, 1>

Интерфейсы: AffineQuantizedOpInterface, ConditionallySpeculatable, DynamicRangeQuantizedOpInterface, InferTypeOpInterface, NoMemoryEffect (MemoryEffectOpInterface), TFL_SparseOp, TflArithmeticCountOpInterface, TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Атрибуты:

Операнды:

Полученные результаты:

tfl.conv_3d (::mlir::TFL::Conv3DOp)

3D-оператор свертки

Выполняет операцию свертки на 3D-входах. Входные данные: inputs[0] : обязательно: тензор активации ввода inputs[1] : обязательно: тензор веса фильтра inputs[2] : необязательно: тензор смещения

Особенности: AlwaysSpeculatableImplTrait, quant::AccumulatorUniformScale<2, 0, 1>

Интерфейсы: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Атрибуты:

Операнды:

Полученные результаты:

tfl.conv_3d_transpose (::mlir::TFL::Conv3DTransposeOp)

3D-оператор транспонированной свертки

Выполняет транспонированную операцию свертки на трехмерных входных данных. Входные данные: inputs[0] : обязательно: форма выходного тензора inputs[1] : обязательно: тензор веса фильтра inputs[2] : обязательно: тензор активации ввода inputs[3] : необязательный: тензор смещения

Особенности: AlwaysSpeculatableImplTrait, quant::AccumulatorUniformScale<2, 0, 1>

Интерфейсы: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Атрибуты:

Операнды:

Полученные результаты:

tfl.cos (::mlir::TFL::CosOp)

Косинус оператор

Поэлементно вычисляет косинус ввода

Признаки: AlwaysSpeculatableImplTrait, InferTensorType, TF::SameOperandsAndResultTypeResolveRef

Интерфейсы: ConditionallySpeculatable, InferShapedTypeOpInterface, InferTypeOpInterface, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Операнды:

Полученные результаты:

tfl.cumsum (::mlir::TFL::CumsumOp)

Суммарный оператор

Вычислите кумулятивную сумму тензора x вдоль оси.

Черты: AlwaysSpeculatableImplTrait

Интерфейсы: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Атрибуты:

Операнды:

Полученные результаты:

tfl.custom (::mlir::TFL::CustomOp)

Пользовательская операция

Общая операция для любой пользовательской операции TFLite.

ввод: список вводов в исходной операции. custom_code: строка, используемая для определения того, какой именно является эта операция, которая соответствует operator_codes.custom_code в плоском буфере. custom_option: держатель для сохранения атрибутов операции в байтах. вывод: список выходов в исходной операции.

Интерфейсы: TflRuntimeVerifyOpInterface

Атрибуты:

Операнды:

Полученные результаты:

tfl.custom_tf (::mlir::TFL::CustomTfOp)

Обертка для пользовательских операций TF.

Операция-оболочка вокруг любой операции Custom TF. К ним относятся операции, определенные с помощью custom_opdefs или связанные, которые не определены в диалекте TF. Эта операция просто оборачивает пользовательскую операцию внутри региона. Примечание № 1. Эта операция не будет включать пользовательские операции TF Lite, определенные с помощью CustomOp. Примечание № 2. Эта операция является просто внутренним представлением внутри конвертера и не отображается/не экспортируется при экспорте модели в Flatbuffer.

Особенности: IsolatedFromAbove, RecursiveMemoryEffects, SingleBlockImplicitTerminator

Интерфейсы: InferTypeOpInterface, TflRuntimeVerifyOpInterface

Операнды:

Полученные результаты:

tfl.densify (::mlir::TFL::DensifyOp)

Оператор уплотнения

Преобразует разреженный тензор в плотный формат.

Черты: AlwaysSpeculatableImplTrait

Интерфейсы: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Операнды:

Полученные результаты:

tfl.depth_to_space (::mlir::TFL::DepthToSpaceOp)

Оператор DepthToSpace

Переупорядочивает данные из глубины в блоки пространственных данных. Это обратное преобразование SpaceToDepth. В частности, эта операция выводит копию входного тензора, где значения из измерения depth перемещаются в пространственных блоках в измерения height и width . Атрибут block_size указывает размер входного блока и способ перемещения данных.

Черты: AlwaysSpeculatableImplTrait, QuantizableResult

Интерфейсы: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Атрибуты:

Операнды:

Полученные результаты:

tfl.depthwise_conv_2d (::mlir::TFL::DepthwiseConv2DOp)

Глубоко отделимый оператор свертки

Выполняет операцию свертки над входными данными.

Входные данные: inputs[0] : обязательно: тензор активации ввода inputs[1] : обязательно: тензор веса фильтра inputs[2] : необязательно: тензор смещения

Черты: AlwaysSpeculatableImplTrait, QuantizableResult, quant::AccumulatorUniformScale<2, 0, 1>, quant::AffineOpCoefficient<3, 1>

Интерфейсы: AffineQuantizedOpInterface, ConditionallySpeculatable, DynamicRangeQuantizedOpInterface, NoMemoryEffect (MemoryEffectOpInterface), TFL_SparseOp, TflArithmeticCountOpInterface, TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Атрибуты:

Операнды:

Полученные результаты:

tfl.dequantize (::mlir::TFL::DequantizeOp)

Оператор деквантования

Преобразует квантованный массив целых чисел в числа с плавающей запятой в соответствии с параметрами квантования.

Интерфейсы: NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Операнды:

Полученные результаты:

tfl.div (::mlir::TFL::DivOp)

Оператор отдела

Поэлементная операция деления.

Особенности: ::mlir::OpTrait::TFLRuntimeOpTrait, AlwaysSpeculatableImplTrait, ResultsBroadcastableShape

Интерфейсы: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflArithmeticCountOpInterface, TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Атрибуты:

Операнды:

Полученные результаты:

tfl.dynamic_update_slice (::mlir::TFL::DynamicUpdateSliceOp)

Динамическое обновление слайса.

Оператор DynamicUpdateSlice, имеющий ту же семантику, что и XLA DynamicUpdateSlice. Генерирует результат, который является значением операнда входного массива, с обновлением среза, перезаписываемым в start_indices.

См. https://www.tensorflow.org/xla/operation_semantics#dynamicupdateslice .

Черты: AlwaysSpeculatableImplTrait

Интерфейсы: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Эффекты: MemoryEffects::Effect{}

Операнды:

Results:

tfl.elu (::mlir::TFL::EluOp)

Exponential Linear Unit operator

Computes the exponential linear f(x) -> exp(x) - 1 for x < 0, x for x >= 0. element-wise.

Traits: AlwaysSpeculatableImplTrait, SameOperandsAndResultShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.embedding_lookup (::mlir::TFL::EmbeddingLookupOp)

Embedding lookup operator

Looks up ids in a list of embedding tensors.

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, DynamicRangeQuantizedOpInterface, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.equal (::mlir::TFL::EqualOp)

Equal operator

Returns the truth element of x == y element-wise

Traits: ::mlir::OpTrait::TFLRuntimeOpTrait, AlwaysSpeculatableImplTrait, Commutative, QuantizableResult, ResultsBroadcastableShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.exp (::mlir::TFL::ExpOp)

Natural exponentiation operator

Performs element-wise natural exponentiation operation on input.

Traits: AlwaysSpeculatableImplTrait, InferTensorType, TF::SameOperandsAndResultTypeResolveRef

Interfaces: ConditionallySpeculatable, InferShapedTypeOpInterface, InferTypeOpInterface, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.expand_dims (::mlir::TFL::ExpandDimsOp)

Inserts a dimension of 1 into a tensor's shape.

Given a tensor input , this operation inserts a dimension of 1 at the dimension index axis of input 's shape. The dimension index axis starts at zero; if you specify a negative number for axis it is counted backward from the end.

This operation is useful if you want to add a batch dimension to a single element. For example, if you have a single image of shape [height, width, channels] , you can make it a batch of 1 image with expand_dims(image, 0) , which will make the shape [1, height, width, channels] .

Other examples:

# 't' is a tensor of shape [2]
shape(expand_dims(t, 0)) ==> [1, 2]
shape(expand_dims(t, 1)) ==> [2, 1]
shape(expand_dims(t, -1)) ==> [2, 1]

# 't2' is a tensor of shape [2, 3, 5]
shape(expand_dims(t2, 0)) ==> [1, 2, 3, 5]
shape(expand_dims(t2, 2)) ==> [2, 3, 1, 5]
shape(expand_dims(t2, 3)) ==> [2, 3, 5, 1]

This operation requires that:

-1-input.dims() <= dim <= input.dims()

This operation is related to squeeze() , which removes dimensions of size 1.

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.external_const (::mlir::TFL::ExternalConstOp)

External const op.

External const op holds a buffer_index which points to a constant in the flatbuffer.

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Results:

tfl.fake_quant (::mlir::TFL::FakeQuantOp)

FakeQuant operator

Fake-quantize the 'inputs' tensor of type float via float scalars min and max to 'outputs' tensor of same shape as inputs.

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.fill (::mlir::TFL::FillOp)

Fill the tensor with given value.

Fill the tensor with given value.

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.floor_div (::mlir::TFL::FloorDivOp)

Floor div operator

Element-wise floor div operation.

Traits: ::mlir::OpTrait::TFLRuntimeOpTrait, AlwaysSpeculatableImplTrait, ResultsBroadcastableShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.floor_mod (::mlir::TFL::FloorModOp)

Division reminder

Element-wise division reminder operation.

Traits: ::mlir::OpTrait::TFLRuntimeOpTrait, AlwaysSpeculatableImplTrait, ResultsBroadcastableShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.floor (::mlir::TFL::FloorOp)

Floor operator

Returns element-wise floor value of the input.

Traits: AlwaysSpeculatableImplTrait, InferTensorType, TF::SameOperandsAndResultTypeResolveRef

Interfaces: ConditionallySpeculatable, InferShapedTypeOpInterface, InferTypeOpInterface, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.fully_connected (::mlir::TFL::FullyConnectedOp)

Fully connected op

Traits: AlwaysSpeculatableImplTrait, QuantizableResult, quant::AccumulatorUniformScale<2, 0, 1>, quant::AffineOpCoefficient<-1, 1>

Interfaces: AffineQuantizedOpInterface, ConditionallySpeculatable, DynamicRangeQuantizedOpInterface, NoMemoryEffect (MemoryEffectOpInterface), TFL_SparseOp, TflArithmeticCountOpInterface, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.gather_nd (::mlir::TFL::GatherNdOp)

Gather_nd operator

Gather slices from params into a Tensor with shape specified by indices .

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.gather (::mlir::TFL::GatherOp)

Gather operator

Gather slices from params axis axis according to indices .

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, DynamicRangeQuantizedOpInterface, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.gelu (::mlir::TFL::GeluOp)

GELU activation function.

Computes GELU activation function element-wise.

Traits: AlwaysSpeculatableImplTrait, QuantizableResult, SameOperandsAndResultShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.greater_equal (::mlir::TFL::GreaterEqualOp)

Greater_equal operator

Element-wise greater_equal operation.

Traits: ::mlir::OpTrait::TFLRuntimeOpTrait, AlwaysSpeculatableImplTrait, QuantizableResult, ResultsBroadcastableShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.greater (::mlir::TFL::GreaterOp)

Greater operator

Element-wise greater operation.

Traits: ::mlir::OpTrait::TFLRuntimeOpTrait, AlwaysSpeculatableImplTrait, QuantizableResult, ResultsBroadcastableShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.hard_swish (::mlir::TFL::HardSwishOp)

Hardswish activation function.

Computes hard-swish activation function f(x) -> (x * relu6(x+3))/6 element-wise.

Traits: AlwaysSpeculatableImplTrait, QuantizableResult, SameOperandsAndResultShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.hashtable_find (::mlir::TFL::HashtableFindOp)

Looks up keys in a table, outputs the corresponding values.

The tensor keys must of the same type as the keys of the table. The output values is of the type of the table values.

The scalar default_value is the value output for keys not present in the table. It must also be of the same type as the table values.

Interfaces: TflRuntimeVerifyOpInterface

Operands:

Results:

tfl.hashtable_import (::mlir::TFL::HashtableImportOp)

Replaces the contents of the table with the specified keys and values.

The tensor keys must be of the same type as the keys of the table. The tensor values must be of the type of the table values.

Interfaces: TflRuntimeVerifyOpInterface

Operands:

tfl.hashtable (::mlir::TFL::HashtableOp)

Creates a non-initialized hash table.

This op creates a hash table, specifying the type of its keys and values. Before using the table you will have to initialize it. After initialization the table will be immutable.

Interfaces: TflRuntimeVerifyOpInterface

Attributes:

Results:

tfl.hashtable_size (::mlir::TFL::HashtableSizeOp)

Computes the number of elements in the given table.

Interfaces: TflRuntimeVerifyOpInterface

Operands:

Results:

tfl.if (::mlir::TFL::IfOp)

if-then-else operation

The tfl.if operation represents an if-then-else construct for conditionally executing two regions of code. The operand to an if operation is a boolean value. For example:

tfl.if %b  {
  ...
} else {
  ...
}

tfl.if may also return results that are defined in its regions. The values defined are determined by which execution path is taken.

Example:

%x, %y = tfl.if %b -> (tensor<f32>, tensor<f32>) {
  %x_true = ...
  %y_true = ...
  tfl.yield %x_true, %y_true : tensor<f32>, tensor<f32>
} else {
  %x_false = ...
  %y_false = ...
  tfl.yield %x_false, %y_false : tensor<f32>, tensor<f32>
}

tfl.if regions are always terminated with "tfl.yield". If "tfl.if" defines no values, the "tfl.yield" can be left out, and will be inserted implicitly. Otherwise, it must be explicit. Also, if "tfl.if" defines one or more values, the 'else' block cannot be omitted.

Example:

tfl.if %b  {
  ...
}

Traits: NoRegionArguments, RecursiveMemoryEffects, SingleBlockImplicitTerminator

Interfaces: RegionBranchOpInterface, TflRuntimeVerifyOpInterface

Operands:

Results:

tfl.imag (::mlir::TFL::ImagOp)

Returns the imaginary part of a complex number.

Given a tensor input of complex numbers, this operation returns a tensor of type float that is the imaginary part of each element in input . All elements in input must be complex numbers of the form \(a + bj\), where a is the real part and b is the imaginary part returned by this operation.

Traits: AlwaysSpeculatableImplTrait, SameOperandsAndResultShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.l2_normalization (::mlir::TFL::L2NormalizationOp)

L2 Normalize Operator

L2Normalization Op

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, FixedOutputRangeInterface, NoMemoryEffect (MemoryEffectOpInterface), TflArithmeticCountOpInterface, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.lstm (::mlir::TFL::LSTMOp)

The full lstm operator

Long short-term memory unit (LSTM) recurrent network layer. The default non-peephole implementation is based on: http://deeplearning.cs.cmu.edu/pdfs/Hochreiter97_lstm.pdf S. Hochreiter and J. Schmidhuber. 'Long Short-Term Memory'. Neural Computation, 9(8):1735-1780, 1997. The peephole implementation is based on: https://research.google.com/pubs/archive/43905.pdf Hasim Sak, Andrew Senior, and Francoise Beaufays. 'Long short-term memory recurrent neural network architectures for large scale acoustic modeling.' INTERSPEECH, 2014. The coupling of input and forget gate (CIFG) is based on: http://arxiv.org/pdf/1503.04069.pdf Greff et al. 'LSTM: A Search Space Odyssey' The layer normalization is based on: https://arxiv.org/pdf/1607.06450.pdf Ba et al. 'Layer Normalization'

Traits: QuantizableResult

Interfaces: DynamicRangeQuantizedOpInterface, TFL_StatefulOp, TflRuntimeVerifyOpInterface

Attributes:

Operands:

Results:

tfl.leaky_relu (::mlir::TFL::LeakyReluOp)

Leaky Relu operator

Element-wise Leaky ReLU operator x -> x >= 0 ? x : (alpha * x)

Traits: AlwaysSpeculatableImplTrait, QuantizableResult, SameOperandsAndResultShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.less_equal (::mlir::TFL::LessEqualOp)

Less_equal operator

Element-wise less_equal operation.

Traits: ::mlir::OpTrait::TFLRuntimeOpTrait, AlwaysSpeculatableImplTrait, QuantizableResult, ResultsBroadcastableShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.less (::mlir::TFL::LessOp)

Less operator

Element-wise less operation.

Traits: ::mlir::OpTrait::TFLRuntimeOpTrait, AlwaysSpeculatableImplTrait, QuantizableResult, ResultsBroadcastableShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.local_response_normalization (::mlir::TFL::LocalResponseNormalizationOp)

Local Response Normalization.

The 4-D input tensor is treated as a 3-D array of 1-D vectors (along the last dimension), and each vector is normalized independently. Within a given vector, each component is divided by the weighted, squared sum of inputs within depth_radius . In detail,

sqr_sum[a, b, c, d] =
    sum(input[a, b, c, d - depth_radius : d + depth_radius + 1] ** 2)
output = input / (bias + alpha * sqr_sum) ** beta

For details, see Krizhevsky et al., ImageNet classification with deep convolutional neural networks (NIPS 2012) .

Traits: AlwaysSpeculatableImplTrait, InferTensorType, TF::SameOperandsAndResultTypeResolveRef

Interfaces: ConditionallySpeculatable, InferShapedTypeOpInterface, InferTypeOpInterface, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.log (::mlir::TFL::LogOp)

Natural logarithm operator

Performs element-wise natural logarithm operation on input.

Traits: AlwaysSpeculatableImplTrait, InferTensorType, TF::SameOperandsAndResultTypeResolveRef

Interfaces: ConditionallySpeculatable, InferShapedTypeOpInterface, InferTypeOpInterface, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.log_softmax (::mlir::TFL::LogSoftmaxOp)

Log softmax operator

Computes element-wise log softmax activations with the following formula

input - log(reduce_sum(exp(input), dim))

Traits: AlwaysSpeculatableImplTrait, QuantizableResult, SameOperandsAndResultShape

Interfaces: ConditionallySpeculatable, FixedOutputRangeInterface, NoMemoryEffect (MemoryEffectOpInterface), TflArithmeticCountOpInterface, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.logical_and (::mlir::TFL::LogicalAndOp)

Logical AND operator

Element-wise logical AND operation.

Traits: AlwaysSpeculatableImplTrait

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.logical_not (::mlir::TFL::LogicalNotOp)

Logical NOT operator

Element-wise logical NOT operation.

Traits: AlwaysSpeculatableImplTrait, SameOperandsAndResultShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.logical_or (::mlir::TFL::LogicalOrOp)

Logical OR operator

Element-wise logical OR operation.

Traits: AlwaysSpeculatableImplTrait

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.logistic (::mlir::TFL::LogisticOp)

Logistic operator

Computes element-wise Sigmoid of input

Traits: AlwaysSpeculatableImplTrait, QuantizableResult, SameOperandsAndResultShape

Interfaces: ConditionallySpeculatable, FixedOutputRangeInterface, NoMemoryEffect (MemoryEffectOpInterface), TflArithmeticCountOpInterface, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.matrix_diag (::mlir::TFL::MatrixDiagOp)

Returns a tensor with the provided diagonal and everything else padded with zeros.

Given a diagonal, returns a tensor with the diagonal and everything else padded with zeros. Assume diagonal has k dimensions [I, J, K, ..., N] , then the output is a tensor of rank k+1 with dimensions [I, J, K, ..., N, N] where: output[i, j, k, ..., m, n] = 1{m=n} * diagonal[i, j, k, ..., n].

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.matrix_set_diag (::mlir::TFL::MatrixSetDiagOp)

Returns a batched matrix tensor with new batched diagonal values.

Given input and diagonal , this operation returns a tensor with the same shape and values as input , except for the main diagonal of the innermost matrices. These will be overwritten by the values in diagonal .

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.max_pool_2d (::mlir::TFL::MaxPool2DOp)

Max Pool 2D op

Performs max pool 2D on input.

Inputs: inputs[0] : required: the input tensor

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflArithmeticCountOpInterface, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.maximum (::mlir::TFL::MaximumOp)

Max operator

Element-wise max operation.

Traits: ::mlir::OpTrait::TFLRuntimeOpTrait, AlwaysSpeculatableImplTrait, Commutative, QuantizableResult, ResultsBroadcastableShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.mean (::mlir::TFL::MeanOp)

Mean operator

Computes the mean of elements across dimensions of a tensor. Reduces input_tensor along the dimensions given in axis. Unless keepdims is true, the rank of the tensor is reduced by 1 for each entry in axis. If keepdims is true, the reduced dimensions are retained with length 1.

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.minimum (::mlir::TFL::MinimumOp)

Min operator

Element-wise min operation.

Traits: ::mlir::OpTrait::TFLRuntimeOpTrait, AlwaysSpeculatableImplTrait, Commutative, QuantizableResult, ResultsBroadcastableShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.mirror_pad (::mlir::TFL::MirrorPadOp)

MirrorPad Operator. Pads a tensor with mirrored values.

This operation pads a input with mirrored values according to the paddings you specify. paddings is an integer tensor with shape [n, 2], where n is the rank of input. For each dimension D of input, paddings[D, 0] indicates how many values to add before the contents of input in that dimension, and paddings[D, 1] indicates how many values to add after the contents of input in that dimension.

Both paddings[D, 0] and paddings[D, 1] must be no greater than input.dim_size(D) (or input.dim_size(D) - 1) if copy_border is true (if false, respectively).

The padded size of each dimension D of the output is:

paddings(D, 0) + input.dim_size(D) + paddings(D, 1)

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.mul (::mlir::TFL::MulOp)

Multiplication operator

Element-wise multiplication operation.

Traits: ::mlir::OpTrait::TFLRuntimeOpTrait, AlwaysSpeculatableImplTrait, Commutative, QuantizableResult, ResultsBroadcastableShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflArithmeticCountOpInterface, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.multinomial (::mlir::TFL::MultinomialOp)

Draws samples from a categorical distribution.

The generated values will have a categorical distribution based on the logits or unnormalized log-probabilities provided for all classes.

Interfaces: TflRuntimeVerifyOpInterface

Attributes:

Operands:

Results:

tfl.neg (::mlir::TFL::NegOp)

Negation operator

Computes element-wise negation of input

Traits: AlwaysSpeculatableImplTrait, InferTensorType, TF::SameOperandsAndResultTypeResolveRef

Interfaces: ConditionallySpeculatable, InferShapedTypeOpInterface, InferTypeOpInterface, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.no_value (::mlir::TFL::NoValueOp)

constant representing no value.

No value constant op.

Traits: AlwaysSpeculatableImplTrait, ConstantLike

Interfaces: ConditionallySpeculatable, InferTypeOpInterface, NoMemoryEffect (MemoryEffectOpInterface)

Effects: MemoryEffects::Effect{}

Attributes:

Results:

tfl.non_max_suppression_v4 (::mlir::TFL::NonMaxSuppressionV4Op)

Greedily selects a subset of bounding boxes in descending order of score,

pruning away boxes that have high intersection-over-union (IOU) overlap with previously selected boxes. Bounding boxes with score less than score_threshold are removed. Bounding boxes are supplied as [y1, x1, y2, x2], where (y1, x1) and (y2, x2) are the coordinates of any diagonal pair of box corners and the coordinates can be provided as normalized (ie, lying in the interval [0, 1]) or absolute. Note that this algorithm is agnostic to where the origin is in the coordinate system and more generally is invariant to orthogonal transformations and translations of the coordinate system; thus translating or reflections of the coordinate system result in the same boxes being selected by the algorithm. The output of this operation is a set of integers indexing into the input collection of bounding boxes representing the selected boxes. The bounding box coordinates corresponding to the selected indices can then be obtained using the tf.gather operation . For example: selected_indices = tf.image.non_max_suppression_v2( boxes, scores, max_output_size, iou_threshold, score_threshold) selected_boxes = tf.gather(boxes, selected_indices)

Traits: AlwaysSpeculatableImplTrait

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.non_max_suppression_v5 (::mlir::TFL::NonMaxSuppressionV5Op)

Greedily selects a subset of bounding boxes in descending order of score,

pruning away boxes that have high intersection-over-union (IOU) overlap with previously selected boxes. Bounding boxes with score less than score_threshold are removed. Bounding boxes are supplied as [y1, x1, y2, x2], where (y1, x1) and (y2, x2) are the coordinates of any diagonal pair of box corners and the coordinates can be provided as normalized (ie, lying in the interval [0, 1]) or absolute. Note that this algorithm is agnostic to where the origin is in the coordinate system and more generally is invariant to orthogonal transformations and translations of the coordinate system; thus translating or reflections of the coordinate system result in the same boxes being selected by the algorithm. The output of this operation is a set of integers indexing into the input collection of bounding boxes representing the selected boxes. The bounding box coordinates corresponding to the selected indices can then be obtained using the tf.gather operation . For example: selected_indices = tf.image.non_max_suppression_v2( boxes, scores, max_output_size, iou_threshold, score_threshold) selected_boxes = tf.gather(boxes, selected_indices) This op also supports a Soft-NMS (with Gaussian weighting) mode (cf Bodla et al, https://arxiv.org/abs/1704.04503 ) where boxes reduce the score of other overlapping boxes instead of directly causing them to be pruned. To enable this Soft-NMS mode, set the soft_nms_sigma parameter to be larger than 0.

Traits: AlwaysSpeculatableImplTrait

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.not_equal (::mlir::TFL::NotEqualOp)

Not_equal operator

Element-wise not_equal operation.

Traits: ::mlir::OpTrait::TFLRuntimeOpTrait, AlwaysSpeculatableImplTrait, Commutative, ResultsBroadcastableShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.NumericVerify (::mlir::TFL::NumericVerifyOp)

Verifies the numericals of the two operands

The NumericVerify op is a debugging op to verify the numericals of the two activations. It is a custom op in TFLite. If log_if_failed is true, the NumericVerify op calculates statistics on differences between float and quantized activations, output logs, set differences to the output tensors, and throws an error if errors above tolerance exist. If log_if_failed = false, then it doesn't care about errors.

Traits: QuantizableResult, SameOperandsShape

Interfaces: TflRuntimeVerifyOpInterface

Attributes:

Operands:

Results:

tfl.one_hot (::mlir::TFL::OneHotOp)

OneHot operator

Returns a one-hot tensor.The locations represented by indices in indices take value on_value , while all other locations take value off_value .

If the input indices is rank N , the output will have rank N+1 , The new axis is created at dimension axis (default: the new axis is appended at the end).

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.prelu (::mlir::TFL::PReluOp)

Parameterized Relu operator

Parameterized Relu operator x -> x >= 0 ? x : (alpha * x) where alpha is a trainable tensor. input and alpha should be the same size as input or be broadcastable.

Traits: ::mlir::OpTrait::TFLRuntimeOpTrait, AlwaysSpeculatableImplTrait, QuantizableResult, ResultsBroadcastableShape, quant::AffineOpCoefficient<-1, 1>

Interfaces: AffineQuantizedOpInterface, ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.pack (::mlir::TFL::PackOp)

Packs a list of tensors along a dimension into one tensor

Packs a list of values_count rank- R tensors into one rank- (R+1) tensor.

Packs the values_count tensors in values into a tensor with rank one higher than each tensor in values , by packing them along the axis dimension.

Given a list of tensors of shape (A, B, C) ;

if axis == 0 then the output tensor will have the shape (N, A, B, C) . if axis == 1 then the output tensor will have the shape (A, N, B, C) . Etc.

For example:

# 'x' is [1, 4]
# 'y' is [2, 5]
# 'z' is [3, 6]
pack([x, y, z]) => [[1, 4], [2, 5], [3, 6]]  # Pack along first dim.
pack([x, y, z], axis=1) => [[1, 2, 3], [4, 5, 6]]

This is the opposite of unpack .

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.pad (::mlir::TFL::PadOp)

Padding operator

This operation pads a input with zeros according to the paddings you specify. paddings is an integer tensor with shape [Dn, 2] , where n is the rank of input . For each dimension D of input , paddings[D, 0] indicates how many zeros to add before the contents of input in that dimension, and paddings[D, 1] indicates how many zeros to add after the contents of input in that dimension.

The padded size of each dimension D of the output is:

paddings(D, 0) + input.dim_size(D) + paddings(D, 1)

For example:

# 't' is [[1, 1], [2, 2]]
# 'paddings' is [[1, 1], [2, 2]]
# rank of 't' is 2
pad(t, paddings) ==> [[0, 0, 0, 0, 0, 0]
                      [0, 0, 1, 1, 0, 0]
                      [0, 0, 2, 2, 0, 0]
                      [0, 0, 0, 0, 0, 0]]

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.padv2 (::mlir::TFL::PadV2Op)

Padding operator v2

This operation pads a input according to the paddings and constant_values you specify. paddings is an integer tensor with shape [Dn, 2] , where n is the rank of input . For each dimension D of input , paddings[D, 0] indicates how many zeros to add before the contents of input in that dimension, and paddings[D, 1] indicates how many zeros to add after the contents of input in that dimension. constant_values is a scalar tensor of the same type as input that indicates the value to use for padding input .

The padded size of each dimension D of the output is:

paddings(D, 0) + input.dim_size(D) + paddings(D, 1)

For example:

# 't' is [[1, 1], [2, 2]]
# 'paddings' is [[1, 1], [2, 2]]
# rank of 't' is 2
pad(t, paddings) ==> [[0, 0, 0, 0, 0, 0]
                      [0, 0, 1, 1, 0, 0]
                      [0, 0, 2, 2, 0, 0]
                      [0, 0, 0, 0, 0, 0]]

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.poly_call (::mlir::TFL::PolyCallOp)

Poly call

Have multiple function bodies for the same computation. This allows a program compiler/interpreter to choose one of the available options to execute the program based on which one is most suitable for the target backend.

input: A list of input tensors whose types are T. output: A list of output tensors whose types are T.

call: Multiple regions, each of which encapsulates the same semantic computation but in different forms.

Traits: SingleBlockImplicitTerminator

Interfaces: RegionBranchOpInterface

Operands:

Results:

tfl.pow (::mlir::TFL::PowOp)

Power operator

Element-wise power operation.

Traits: ::mlir::OpTrait::TFLRuntimeOpTrait, AlwaysSpeculatableImplTrait, ResultsBroadcastableShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.pseudo_qconst (::mlir::TFL::QConstOp)

Quantized constant pseudo op

Represents a quantized constant value in TensorFlow Lite dialect. This is not an actual operation and it will be lowered to buffer instead. The quantization parameters are stored as a type attribute in this constant.

Traits: AlwaysSpeculatableImplTrait, FirstAttrDerivedResultType

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface)

Effects: MemoryEffects::Effect{}

Attributes:

Results:

tfl.quantize (::mlir::TFL::QuantizeOp)

Quantize operator

Converts floating point tensors to quantized integer tensors according to the quantization parameters defined in the type attribute.

Traits: FirstAttrDerivedResultType, SameOperandsAndResultShape

Interfaces: NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.rfft2d (::mlir::TFL::RFFT2dOp)

2D real-valued fast Fourier transform.

Computes the 2-dimensional discrete Fourier transform of a real-valued signal over the inner-most 2 dimensions of input .

Since the DFT of a real signal is Hermitian-symmetric, RFFT2D only returns the fft_length / 2 + 1 unique components of the FFT for the inner-most dimension of output : the zero-frequency term, followed by the fft_length / 2 positive-frequency terms.

Along each axis RFFT2D is computed on, if fft_length is smaller than the corresponding dimension of input , the dimension is cropped. If it is larger, the dimension is padded with zeros.

Traits: AlwaysSpeculatableImplTrait

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.random_standard_normal (::mlir::TFL::RandomStandardNormalOp)

Outputs random values from a normal distribution.

The generated values will have mean 0 and standard deviation 1.

Interfaces: TflRuntimeVerifyOpInterface

Attributes:

Operands:

Results:

tfl.random_uniform (::mlir::TFL::RandomUniformOp)

Outputs random values from a uniform distribution.

The generated values follow a uniform distribution in the range [0, 1) . The lower bound 0 is included in the range, while the upper bound 1 is excluded.

Interfaces: TflRuntimeVerifyOpInterface

Attributes:

Operands:

Results:

tfl.range (::mlir::TFL::RangeOp)

Range operator

Returns a 1D tensor defined by a sequence from start to limit with a given delta .

Traits: AlwaysSpeculatableImplTrait

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.rank (::mlir::TFL::RankOp)

Rank operator.

Returns the rank of a tensor.

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.read_variable (::mlir::TFL::ReadVariableOp)

Reads variable value.

Read variable data identified by 'resource_id'.

Interfaces: TflRuntimeVerifyOpInterface

Operands:

Results:

tfl.real (::mlir::TFL::RealOp)

Returns the real part of a complex number.

Given a tensor input of complex numbers, this operation returns a tensor of type float that is the real part of each element in input . All elements in input must be complex numbers of the form \(a + bj\), where a is the real part returned by this operation and b is the imaginary part.

Traits: AlwaysSpeculatableImplTrait, SameOperandsAndResultShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.reduce_all (::mlir::TFL::ReduceAllOp)

Computes the "logical and" of elements across dimensions of a tensor.

Reduces input along the dimensions given in axis . Unless keep_dims is true, the rank of the tensor is reduced by 1 for each entry in axis . If keep_dims is true, the reduced dimensions are retained with length 1.

Traits: AlwaysSpeculatableImplTrait

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.reduce_any (::mlir::TFL::ReduceAnyOp)

Computes the "logical or" of elements across dimensions of a tensor.

Reduces input along the dimensions given in axis . Unless keep_dims is true, the rank of the tensor is reduced by 1 for each entry in axis . If keep_dims is true, the reduced dimensions are retained with length 1.

Traits: AlwaysSpeculatableImplTrait

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.reduce_max (::mlir::TFL::ReduceMaxOp)

Max-reduction operator

Computes the max reduction along the specified axes

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.reduce_min (::mlir::TFL::ReduceMinOp)

Min-reduction operator

Computes the min reduction along the specified axes

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.reduce_prod (::mlir::TFL::ReduceProdOp)

Prod-reduction operator

Computes the product along the specified axes

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.relu_0_to_1 (::mlir::TFL::Relu0To1Op)

Relu0To1 operator

Element-wise Relu0To1 operator x -> max(0, min(1, x))

Traits: AlwaysSpeculatableImplTrait, QuantizableResult, SameOperandsAndResultShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.relu_n1_to_1 (::mlir::TFL::Relu1Op)

Relu1 operator

Element-wise Relu1 operator x -> max(-1, min(1, x))

Traits: AlwaysSpeculatableImplTrait, QuantizableResult, SameOperandsAndResultShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.relu6 (::mlir::TFL::Relu6Op)

Relu6 operator

Element-wise Relu6 operator x -> max(0, min(6, x))

Traits: AlwaysSpeculatableImplTrait, QuantizableResult, SameOperandsAndResultShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.relu (::mlir::TFL::ReluOp)

Relu operator

Element-wise Relu operator x -> max(0, x)

Traits: AlwaysSpeculatableImplTrait, QuantizableResult, SameOperandsAndResultShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.reshape (::mlir::TFL::ReshapeOp)

Reshape operator

Produces a tensor with the same values but different static shape defined by the output type.

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, InferTypeOpInterface, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.resize_bilinear (::mlir::TFL::ResizeBilinearOp)

ResizeBilinear Op

Resize images to size using bilinear interpolation.

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.resize_nearest_neighbor (::mlir::TFL::ResizeNearestNeighborOp)

ResizeNearestNeighbor Op

Resize images to size using nearest neighbor interpolation.

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.reverse_sequence (::mlir::TFL::ReverseSequenceOp)

Reverses variable length slices.

This op first slices input along the dimension batch_dim , and for each slice i , reverses the first seq_lengths[i] elements along the dimension seq_dim .

The elements of seq_lengths must obey seq_lengths[i] <= input.dims[seq_dim] , and seq_lengths must be a vector of length input.dims[batch_dim] .

The output slice i along dimension batch_dim is then given by input slice i , with the first seq_lengths[i] slices along dimension seq_dim reversed.

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.reverse_v2 (::mlir::TFL::ReverseV2Op)

ReverseV2 Operator

Reverses specific dimensions of a tensor.

Given a tensor, and a int32/int64 tensor axis representing the set of dimensions of tensor to reverse. This operation reverses each dimension i for which there exists j st axis[j] == i.

Args: tensor: A Tensor. Must be one of the following types: uint8, int8, int16, int32, int64, float32, bool Up to 8-D.

axis: A Tensor. Must be one of the following types: int32, int64. with only 1 element which is the axis index. TODO: Add support for multiple elements.

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.round (::mlir::TFL::RoundOp)

Round operator

Rounds the values of a tensor to the nearest integer, element-wise.

Traits: AlwaysSpeculatableImplTrait, InferTensorType, TF::SameOperandsAndResultTypeResolveRef

Interfaces: ConditionallySpeculatable, InferShapedTypeOpInterface, InferTypeOpInterface, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.rsqrt (::mlir::TFL::RsqrtOp)

Reciprocal of square root operator

Computes element-wise reverse square root of input

Traits: AlwaysSpeculatableImplTrait, QuantizableResult, SameOperandsAndResultShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.svdf (::mlir::TFL::SVDFOp)

Single value decomposition filter operator

The SVDF op is a decomposition of a densely connected op into low rank filters. For details: https://research.google.com/pubs/pub43813.html https://arxiv.org/abs/1812.02802

Traits: QuantizableResult, quant::AccumulatorUniformScale<3, 2, 4>

Interfaces: DynamicRangeQuantizedOpInterface, TFL_StatefulOp, TflRuntimeVerifyOpInterface

Attributes:

Operands:

Results:

tfl.scatter_nd (::mlir::TFL::ScatterNdOp)

Scatter_nd operator

Scatter updates into a new tensor according to indices

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.segment_sum (::mlir::TFL::SegmentSumOp)

SegmentSum operator

Computes the sum along segments of a tensor.

Traits: AlwaysSpeculatableImplTrait

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.select (::mlir::TFL::SelectOp)

Select operator

Select values of 'x' if the corresponding value of 'condition' is true or the value of 'y' if false. There are valid condition input sizes:

1. Either the same shape (in which case the select is elementwise), or
2. condition must be Rank 1 and match over the first dimension.

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.select_v2 (::mlir::TFL::SelectV2Op)

SelectV2 operator

Select values of 'x' if the corresponding value of 'condition' is true or the value of 'y' if false. There are valid condition input sizes:

1. Either the same shape (in which case the select is elementwise), or
2. Broadcastable shapes between 'condition', 'x' and 'y'.

Traits: ::mlir::OpTrait::TFLRuntimeOpTrait, AlwaysSpeculatableImplTrait, QuantizableResult, ResultsBroadcastableShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.shape (::mlir::TFL::ShapeOp)

Shape operator

Returns the shape of a tensor.

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.sign (::mlir::TFL::SignOp)

Sign operation

Returns NaN if x is NaN, 0 if x is 0, -1 if x < 0 and 1 if x > 0.

Traits: AlwaysSpeculatableImplTrait, SameOperandsAndResultElementType, SameOperandsAndResultShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.sin (::mlir::TFL::SinOp)

Sine operator

Computes element-wise Sine of input

Traits: AlwaysSpeculatableImplTrait, InferTensorType, TF::SameOperandsAndResultTypeResolveRef

Interfaces: ConditionallySpeculatable, InferShapedTypeOpInterface, InferTypeOpInterface, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.slice (::mlir::TFL::SliceOp)

Return a slice from 'input'.

The output tensor is a tensor with dimensions described by 'size' whose values are extracted from 'input' starting at the offsets in 'begin'.

begin is zero-based; size is one-based. If size[i] is -1, all remaining elements in dimension i are included in the slice. In other words, this is equivalent to setting: size[i] = input.dim_size(i) - begin[i]

Requirements : 0 <= begin[i] <= begin[i] + size[i] <= Di for i in [0, n)

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.softmax (::mlir::TFL::SoftmaxOp)

Softmax operator

Computes element-wise softmax activations with the following formula

exp(input) / tf.reduce_sum(exp(input * beta), dim)

Traits: AlwaysSpeculatableImplTrait, QuantizableResult, SameOperandsAndResultShape

Interfaces: ConditionallySpeculatable, FixedOutputRangeInterface, NoMemoryEffect (MemoryEffectOpInterface), TflArithmeticCountOpInterface, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.space_to_batch_nd (::mlir::TFL::SpaceToBatchNdOp)

SpaceToBatchNd operator

This operation reshapes space dimensions into the "batch" dimension 0

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.space_to_depth (::mlir::TFL::SpaceToDepthOp)

SpaceToDepth operator

Rearranges blocks of spatial data, into depth. More specifically, this op outputs a copy of the input tensor where values from the height and width dimensions are moved to the depth dimension. block_size indicates the input block size.

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.pseudo_sparse_const (::mlir::TFL::SparseConstOp)

Sparse constant pseudo op.

Represents a sparse constant value in TensorFlow Lite dialect. This is not an actual operation and it will be lowered to buffer instead.

Traits: AlwaysSpeculatableImplTrait, FirstAttrDerivedResultType, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Results:

tfl.pseudo_sparse_qconst (::mlir::TFL::SparseQConstOp)

Sparse quantized constant pseudo op

Represents a sparse quantized constant value in TensorFlow Lite dialect. This is not an actual operation and it will be lowered to buffer instead. The quantization parameters are stored as a type attribute in this constant.

Traits: AlwaysSpeculatableImplTrait, FirstAttrDerivedResultType

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Results:

tfl.sparse_to_dense (::mlir::TFL::SparseToDenseOp)

Converts a sparse representation into a dense tensor.

Builds an array dense with shape output_shape such that

# If sparse_indices is scalar
dense[i] = (i == sparse_indices ? sparse_values : default_value)

# If sparse_indices is a vector, then for each i
dense[sparse_indices[i]] = sparse_values[i]

# If sparse_indices is an n by d matrix, then for each i in [0, n)
dense[sparse_indices[i][0], ..., sparse_indices[i][d-1]] = sparse_values[i]

All other values in dense are set to default_value . If sparse_values is a scalar, all sparse indices are set to this single value.

Indices should be sorted in lexicographic order, and indices must not contain any repeats. If validate_indices is true, these properties are checked during execution.

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.split (::mlir::TFL::SplitOp)

Splits a tensor into num_split tensors along one dimension.

Splits the value tensor along split_dim into a number of sub-tensors with same shape as the original one, except for split_dim . Same as tf.Split.

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.split_v (::mlir::TFL::SplitVOp)

Splits a tensor into num_split tensors along one dimension.

Splits the value tensor along split_dim into a number of sub-tensors with same shape as the original one, except for split_dim . The grouping of the resultant sub-tensors is decided by size-splits . Same as tf.SplitV.

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.sqrt (::mlir::TFL::SqrtOp)

Square root operator

Computes element-wise Square root of input

Traits: AlwaysSpeculatableImplTrait, InferTensorType, TF::SameOperandsAndResultTypeResolveRef

Interfaces: ConditionallySpeculatable, InferShapedTypeOpInterface, InferTypeOpInterface, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.square (::mlir::TFL::SquareOp)

Square operator

Computes element-wise Square of input

Traits: AlwaysSpeculatableImplTrait, InferTensorType, TF::SameOperandsAndResultTypeResolveRef

Interfaces: ConditionallySpeculatable, InferShapedTypeOpInterface, InferTypeOpInterface, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.squared_difference (::mlir::TFL::SquaredDifferenceOp)

Squared difference operator

Element-wise squared difference operation.

Traits: ::mlir::OpTrait::TFLRuntimeOpTrait, AlwaysSpeculatableImplTrait, QuantizableResult, ResultsBroadcastableShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.squeeze (::mlir::TFL::SqueezeOp)

Removes dimensions of size 1 from the shape of a tensor.

Given a tensor input , this operation returns a tensor of the same type with all dimensions of size 1 removed. If you don't want to remove all size 1 dimensions, you can remove specific size 1 dimensions by specifying squeeze_dims .

For example:

# 't' is a tensor of shape [1, 2, 1, 3, 1, 1]
shape(squeeze(t)) ==> [2, 3]

Or, to remove specific size 1 dimensions:

# 't' is a tensor of shape [1, 2, 1, 3, 1, 1]
shape(squeeze(t, [2, 4])) ==> [1, 2, 3, 1]

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.strided_slice (::mlir::TFL::StridedSliceOp)

StridedSlice Op

Return a strided slice from input .

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.sub (::mlir::TFL::SubOp)

Subtraction operator

Element-wise subtraction operation.

Traits: ::mlir::OpTrait::TFLRuntimeOpTrait, AlwaysSpeculatableImplTrait, QuantizableResult, ResultsBroadcastableShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflArithmeticCountOpInterface, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.sum (::mlir::TFL::SumOp)

Sum operator

Computes the sum reduction along the specified axes

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.tanh (::mlir::TFL::TanhOp)

Hyperbolic tangent operator

Computes element-wise Hyperbolic tangent of input

Traits: AlwaysSpeculatableImplTrait, QuantizableResult, SameOperandsAndResultShape

Interfaces: ConditionallySpeculatable, FixedOutputRangeInterface, NoMemoryEffect (MemoryEffectOpInterface), TflArithmeticCountOpInterface, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.tile (::mlir::TFL::TileOp)

Tile operator.

Constructs a tensor by tiling a given tensor.

This operation creates a new tensor by replicating input multiples times. The output tensor's i'th dimension has input.dims(i) * multiples[i] elements, and the values of input are replicated multiples[i] times along the 'i'th dimension. For example, tiling [abcd] by [2] produces [abcdabcd].

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.topk_v2 (::mlir::TFL::TopKV2Op)

TopK operator

Returns the top k largest element along each last dimensional slice of input and the indices of values within the last dimension of the input tensor.

Results are always sorted in the descending order.

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.transpose_conv (::mlir::TFL::TransposeConvOp)

Transpose convolution operator

Performs transpose convolution operation on input.

Traits: AlwaysSpeculatableImplTrait, QuantizableResult, quant::AccumulatorUniformScale<3, 1, 2>, quant::AffineOpCoefficient<0, 1>

Interfaces: AffineQuantizedOpInterface, ConditionallySpeculatable, DynamicRangeQuantizedOpInterface, NoMemoryEffect (MemoryEffectOpInterface), TFL_SparseOp, TflArithmeticCountOpInterface, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.transpose (::mlir::TFL::TransposeOp)

Transpose operator

Returns the Transpose of x

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.unidirectional_sequence_lstm (::mlir::TFL::UnidirectionalSequenceLSTMOp)

Unidirectional sequence lstm operator

A recurrent neural network specified by an LSTM cell. This Op supports unrolling the input along the time or batch dimensions, and implements the following operation for each element in the sequence s = 1...sequence_length: outputs[s] = state = activation(LSTMOp(inputs[s]))

where LSTMOp is LSTM TF Lite Op and the “activation” is the function passed as the “fused_activation_function” argument (if not “NONE”).

Traits: QuantizableResult

Interfaces: DynamicRangeQuantizedOpInterface, InferTypeOpInterface, TFL_StatefulOp, TflRuntimeVerifyOpInterface

Attributes:

Operands:

Results:

tfl.unidirectional_sequence_rnn (::mlir::TFL::UnidirectionalSequenceRNNOp)

Unidirectional sequence rnn operator

A recurrent neural network specified by an RNN cell. This Op takes in input in a format {batch_size, seq_len, input_size} or {seq_len, batch_size, input_size} if it's time-majored.

It implements the following operation for each element in the sequence s = 1...sequence_length: outputs[s] = state = activation(RNNOp(inputs[s]))

where RNNOp is RNNOp TF Lite Op and the “activation” is the function passed as the “fused_activation_function” argument (if not “NONE”).

Traits: QuantizableResult

Interfaces: DynamicRangeQuantizedOpInterface, TFL_StatefulOp, TflRuntimeVerifyOpInterface

Attributes:

Operands:

Results:

tfl.unique (::mlir::TFL::UniqueOp)

Unique Op.

This operation returns a tensor output containing all of the unique elements of input sorted in the same order that they occur in input . This operation also returns a tensor idx the same size as x that contains the index of each value of input in the unique output output . In other words:

Traits: AlwaysSpeculatableImplTrait, QuantizableResult

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.unpack (::mlir::TFL::UnpackOp)

Unpacks a tensor along a dimension into multiple tensors

Unpacks a given dimension of a rank- R tensor into num rank- (R-1) tensors.

Unpacks num tensors from value by chipping it along the axis dimension. For example, given a tensor of shape (A, B, C, D) ;

If axis == 0 then the i'th tensor in output is the slice value[i, :, :, :] and each tensor in output will have shape (B, C, D) . (Note that the dimension unpacked along is gone, unlike split ).

If axis == 1 then the i'th tensor in output is the slice value[:, i, :, :] and each tensor in output will have shape (A, C, D) . Etc.

This is the opposite of pack .

Traits: AlwaysSpeculatableImplTrait, QuantizableResult, SameOperandsAndResultElementType

Interfaces: ConditionallySpeculatable, InferTypeOpInterface, NoMemoryEffect (MemoryEffectOpInterface), SameOperandsAndResultsScale, TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Attributes:

Operands:

Results:

tfl.unsorted_segment_max (::mlir::TFL::UnsortedSegmentMaxOp)

UnsortedSegmentMax operator

Computes the maximum value along segments of a tensor such that output[i] = max(data[j....]) where segment_ids[j...] = i if the maximum is empty for a given segment ID i, it outputs the smallest possible value for the specific numeric type, output[i] = numeric_limits::lowest(). Note the values of segment_ids are always validated to be less than num_segments and an error is thrown for out-of-bound indices.

Traits: AlwaysSpeculatableImplTrait

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.unsorted_segment_min (::mlir::TFL::UnsortedSegmentMinOp)

UnsortedSegmentMin operator

Computes the minimum value along segments of a tensor such that output[i] = min(data[j....]) where segment_ids[j...] = i if the minimum is empty for a given segment ID i, it outputs the largest possible value for the specific numeric type, output[i] = numeric_limits::max(). Note the values of segment_ids are always validated to be less than num_segments and an error is thrown for out-of-bound indices.

Traits: AlwaysSpeculatableImplTrait

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.unsorted_segment_prod (::mlir::TFL::UnsortedSegmentProdOp)

UnsortedSegmentProd operator

Computes the product along segments of a tensor.

Traits: AlwaysSpeculatableImplTrait

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.unsorted_segment_sum (::mlir::TFL::UnsortedSegmentSumOp)

UnsortedSegmentSum operator

From a tensor segmentation, computes the output resulting from summing together elements mapped to the same segment_id. Ie output[i] is equal to the tensor sum of all elements from the input tensor mapped to segment_id i . If no tensors are mapped to a particular included segment_id, the output at that indice will be a zero tensor with the appropriate shape. Note the values of segment_ids are always validated to be less than num_segments and an error is thrown for out-of-bound indices

Traits: AlwaysSpeculatableImplTrait

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.var_handle (::mlir::TFL::VarHandleOp)

Returns a handle to a variable resource from its name.

Returns a handle for a variable resource from its name. container: the container this variable is placed in. shared_name: the name by which this variable is referred to.

Interfaces: TflRuntimeVerifyOpInterface

Attributes:

Results:

tfl.where (::mlir::TFL::WhereOp)

Returns locations of nonzero / true values in a tensor.

This operation returns the coordinates of true elements in condition . The coordinates are returned in a 2-D tensor where the first dimension (rows) represents the number of true elements, and the second dimension (columns) represents the coordinates of the true elements. Keep in mind, the shape of the output tensor can vary depending on how many true values there are in condition . Indices are output in row-major order.

Traits: AlwaysSpeculatableImplTrait

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

tfl.while (::mlir::TFL::WhileOp)

While loop

output = input; while (cond(output)) { output = body(output) }

While loop where all values are passes through arguments with implicit capture.

input: A list of input tensors whose types are T. output: A list of output tensors whose types are T. cond: A region that takes 'input' and returns a boolean scalar tensor. body: A region that takes a list of tensors and returns another list of tensors. Both lists have the same types.

Traits: SingleBlockImplicitTerminator

Interfaces: LoopLikeOpInterface, TflRuntimeVerifyOpInterface

Attributes:

Operands:

Results:

tfl.yield (::mlir::TFL::YieldOp)

Yield operation

The "yield" operation represents a return operation within the conditional and body of structured control flow (eg, while), and a terminator for ControlNodeOp. The operation takes a variable number of operands and produces no results. The operand number and types must match the signature of the region that contains the operation.

Traits: AlwaysSpeculatableImplTrait, QuantizableResult, Terminator

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

tfl.zeros_like (::mlir::TFL::ZerosLikeOp)

ZerosLike operator

Returns a tensor of zeros with the same shape and type as the input tensor.

Traits: AlwaysSpeculatableImplTrait, SameOperandsAndResultShape

Interfaces: ConditionallySpeculatable, NoMemoryEffect (MemoryEffectOpInterface), TflRuntimeVerifyOpInterface

Effects: MemoryEffects::Effect{}

Operands:

Results:

Attribute definition

DimensionMetadataAttr

Dimension metadata.

Syntax:

#tfl.dimension_metadata<
  ::mlir::TFL::DimensionTypeAttr,   # format
  int32_t,   # dense_size
  ::llvm::ArrayRef<int32_t>,   # segments
  ::llvm::ArrayRef<int32_t>   # indices
>

Parameters:

SparsityParameterAttr

Sparsity parameter.

Syntax:

#tfl.sparsity_parameter<
  ::llvm::ArrayRef<int32_t>,   # traversal_order
  ::llvm::ArrayRef<int32_t>,   # block_map
  ::llvm::ArrayRef<DimensionMetadataAttr>   # dim_metadata
>

Parameters:

ConstBytesAttr

A string attribute representation of compiled bytes

Syntax Examples:

#tfl<const_bytes : "0xDEADBEEF">

Parameters:

DimensionTypeAttr

dimension_type

Syntax:

#tfl.dimension_type_attr<
  ::mlir::TFL::DimensionType   # value
>

Parameters:

LSTMKernelTypeAttr

lstm_kernel_type

Syntax:

#tfl.lstm_kernel_type_attr<
  ::mlir::TFL::LSTMKernelType   # value
>

Parameters:

MirrorPaddingTypeAttr

mirror_pad_enum

Syntax:

#tfl.mirror_pad_attr<
  ::mlir::TFL::MirrorPaddingType   # value
>

Parameters: