CPU specialization of TensorOps for mathematical operations. More...

Inheritance diagram for Mila::Dnn::Compute::Cpu::MathOps:

Static Public Member Functions
template<TensorDataType TDataType, typename TMemoryResource> requires isValidTensor<TDataType, TMemoryResource>
static void	add (const Tensor< TDataType, TMemoryResource > &a, const Tensor< TDataType, TMemoryResource > &b, Tensor< TDataType, TMemoryResource > &result, IExecutionContext *exec_context=nullptr)
	Element-wise addition of two tensors (CPU implementation).
template<TensorDataType TDataType, typename TMemoryResource> requires isValidTensor<TDataType, TMemoryResource>
static void	divide (const Tensor< TDataType, TMemoryResource > &a, const Tensor< TDataType, TMemoryResource > &b, Tensor< TDataType, TMemoryResource > &result, IExecutionContext *exec_context=nullptr)
	Element-wise division of two tensors (CPU implementation).
template<TensorDataType TDataType, typename TMemoryResource> requires isValidTensor<TDataType, TMemoryResource>
static void	multiply (const Tensor< TDataType, TMemoryResource > &a, const Tensor< TDataType, TMemoryResource > &b, Tensor< TDataType, TMemoryResource > &result, IExecutionContext *exec_context=nullptr)
	Element-wise multiplication of two tensors (CPU implementation).
template<TensorDataType TDataType, typename TMemoryResource> requires isValidTensor<TDataType, TMemoryResource>
static void	subtract (const Tensor< TDataType, TMemoryResource > &a, const Tensor< TDataType, TMemoryResource > &b, Tensor< TDataType, TMemoryResource > &result, IExecutionContext *exec_context=nullptr)
	Element-wise subtraction of two tensors (CPU implementation).
template<TensorDataType TDataType, typename TMemoryResource> requires isValidTensor<TDataType, TMemoryResource>
static float	sum (const Tensor< TDataType, TMemoryResource > &tensor, IExecutionContext *exec_context=nullptr)
	Computes sum of all tensor elements (CPU implementation).

Static Private Member Functions
template<TensorDataType TDataType, typename TMemoryResource, typename TBinaryOp> requires isValidTensor<TDataType, TMemoryResource>
static void	performElementwiseOperation (const Tensor< TDataType, TMemoryResource > &a, const Tensor< TDataType, TMemoryResource > &b, Tensor< TDataType, TMemoryResource > &result, TBinaryOp op)
	Performs element-wise operation using the provided binary function.
template<TensorDataType TDataType, typename TMemoryResource>
static void	validateShapeCompatibility (const Tensor< TDataType, TMemoryResource > &a, const Tensor< TDataType, TMemoryResource > &b, const Tensor< TDataType, TMemoryResource > &result, const std::string &operation_name)
	Validates that three tensors have compatible shapes for element-wise operations.

Detailed Description

CPU specialization of TensorOps for mathematical operations.

Implements element-wise operations for CPU tensors using standard library algorithms with optional parallel execution for large tensors.

Key features:

Synchronous execution (no stream management needed)
Parallel execution for large tensors (>10000 elements)
Accepts ExecutionContext for API consistency (unused on CPU)
Automatic type conversion via TensorHostTypeMap
Zero-copy direct memory access

Member Function Documentation

◆ add()

template<TensorDataType TDataType, typename TMemoryResource>
requires isValidTensor<TDataType, TMemoryResource>

void Mila::Dnn::Compute::Cpu::MathOps::add	(	const Tensor< TDataType, TMemoryResource > &	a,
		const Tensor< TDataType, TMemoryResource > &	b,
		Tensor< TDataType, TMemoryResource > &	result,
		IExecutionContext *	exec_context = nullptr )

inlinestatic

Element-wise addition of two tensors (CPU implementation).

Performs element-wise addition a[i] + b[i] for all elements and stores the result in a pre-allocated result tensor. Both input tensors must have identical shapes matching the result tensor shape.

Template Parameters

TDataType	Abstract tensor data type
TMemoryResource	Memory resource (must be CPU-accessible)

Parameters

a	First operand tensor
b	Second operand tensor
result	Pre-allocated result tensor (must have matching shape)
exec_context	Optional execution context (unused for CPU, accepted for API consistency)

Exceptions

std::invalid_argument If tensor shapes don't match or tensors are empty

Note: ExecutionContext parameter ignored but present for uniform API across devices; Uses parallel execution for tensors with >10000 elements

Here is the call graph for this function:

◆ divide()

template<TensorDataType TDataType, typename TMemoryResource>
requires isValidTensor<TDataType, TMemoryResource>

void Mila::Dnn::Compute::Cpu::MathOps::divide	(	const Tensor< TDataType, TMemoryResource > &	a,
		const Tensor< TDataType, TMemoryResource > &	b,
		Tensor< TDataType, TMemoryResource > &	result,
		IExecutionContext *	exec_context = nullptr )

inlinestatic

Element-wise division of two tensors (CPU implementation).

Performs element-wise division a[i] / b[i] for all elements and stores the result in a pre-allocated result tensor.

For floating-point types, follows IEEE 754 standards:

Division by zero produces infinity or NaN For integer types:
Division by zero throws std::runtime_error

Template Parameters

TDataType	Abstract tensor data type
TMemoryResource	Memory resource (must be CPU-accessible)

Parameters

a	First operand tensor (dividend)
b	Second operand tensor (divisor)
result	Pre-allocated result tensor (must have matching shape)
exec_context	Optional execution context (unused for CPU, accepted for API consistency)

Exceptions

std::invalid_argument	If tensor shapes don't match or tensors are empty
std::runtime_error	If division by zero in integer division

Note: ExecutionContext parameter ignored but present for uniform API across devices; Uses parallel execution for tensors with >10000 elements

Here is the call graph for this function:

◆ multiply()

template<TensorDataType TDataType, typename TMemoryResource>
requires isValidTensor<TDataType, TMemoryResource>

void Mila::Dnn::Compute::Cpu::MathOps::multiply	(	const Tensor< TDataType, TMemoryResource > &	a,
		const Tensor< TDataType, TMemoryResource > &	b,
		Tensor< TDataType, TMemoryResource > &	result,
		IExecutionContext *	exec_context = nullptr )

inlinestatic

Element-wise multiplication of two tensors (CPU implementation).

Performs element-wise multiplication a[i] * b[i] for all elements and stores the result in a pre-allocated result tensor.

Template Parameters

TDataType	Abstract tensor data type
TMemoryResource	Memory resource (must be CPU-accessible)

Parameters

a	First operand tensor
b	Second operand tensor
result	Pre-allocated result tensor (must have matching shape)
exec_context	Optional execution context (unused for CPU, accepted for API consistency)

Exceptions

std::invalid_argument If tensor shapes don't match or tensors are empty

Note: ExecutionContext parameter ignored but present for uniform API across devices; Uses parallel execution for tensors with >10000 elements

Here is the call graph for this function:

◆ performElementwiseOperation()

template<TensorDataType TDataType, typename TMemoryResource, typename TBinaryOp>
requires isValidTensor<TDataType, TMemoryResource>

void Mila::Dnn::Compute::Cpu::MathOps::performElementwiseOperation	(	const Tensor< TDataType, TMemoryResource > &	a,
		const Tensor< TDataType, TMemoryResource > &	b,
		Tensor< TDataType, TMemoryResource > &	result,
		TBinaryOp	op )

inlinestaticprivate

Performs element-wise operation using the provided binary function.

Applies a binary operation to corresponding elements of two input tensors and stores the result in the output tensor. Uses parallel execution for improved performance on large tensors (>10000 elements).

Template Parameters

TDataType	Abstract tensor data type
TMemoryResource	Memory resource type
TBinaryOp	Binary operation function type

Parameters

a	First input tensor
b	Second input tensor
result	Output tensor (must be pre-allocated with correct shape)
op	Binary operation to apply (e.g., std::plus, std::minus)

Here is the call graph for this function:

Here is the caller graph for this function:

◆ subtract()

template<TensorDataType TDataType, typename TMemoryResource>
requires isValidTensor<TDataType, TMemoryResource>

void Mila::Dnn::Compute::Cpu::MathOps::subtract	(	const Tensor< TDataType, TMemoryResource > &	a,
		const Tensor< TDataType, TMemoryResource > &	b,
		Tensor< TDataType, TMemoryResource > &	result,
		IExecutionContext *	exec_context = nullptr )

inlinestatic

Element-wise subtraction of two tensors (CPU implementation).

Performs element-wise subtraction a[i] - b[i] for all elements and stores the result in a pre-allocated result tensor.

Template Parameters

TDataType	Abstract tensor data type
TMemoryResource	Memory resource (must be CPU-accessible)

Parameters

a	First operand tensor (minuend)
b	Second operand tensor (subtrahend)
result	Pre-allocated result tensor (must have matching shape)
exec_context	Optional execution context (unused for CPU, accepted for API consistency)

Exceptions

std::invalid_argument If tensor shapes don't match or tensors are empty

Note: ExecutionContext parameter ignored but present for uniform API across devices; Uses parallel execution for tensors with >10000 elements

Here is the call graph for this function:

◆ sum()

template<TensorDataType TDataType, typename TMemoryResource>
requires isValidTensor<TDataType, TMemoryResource>

float Mila::Dnn::Compute::Cpu::MathOps::sum	(	const Tensor< TDataType, TMemoryResource > &	tensor,
		IExecutionContext *	exec_context = nullptr )

inlinestatic

Computes sum of all tensor elements (CPU implementation).

Reduces tensor to a single scalar value representing the sum of all elements. Uses parallel reduction for large tensors.

Template Parameters

TDataType	Abstract tensor data type
TMemoryResource	Memory resource (must be CPU-accessible)

Parameters

tensor	Input tensor
exec_context	Optional execution context (unused for CPU, accepted for API consistency)

Returns: Sum of all elements as float

Note: ExecutionContext parameter ignored but present for uniform API across devices; Uses parallel reduction for tensors with >10000 elements; Always returns after computation (synchronous operation)

Here is the call graph for this function:

◆ validateShapeCompatibility()

template<TensorDataType TDataType, typename TMemoryResource>

void Mila::Dnn::Compute::Cpu::MathOps::validateShapeCompatibility	(	const Tensor< TDataType, TMemoryResource > &	a,
		const Tensor< TDataType, TMemoryResource > &	b,
		const Tensor< TDataType, TMemoryResource > &	result,
		const std::string &	operation_name )

inlinestaticprivate

Validates that three tensors have compatible shapes for element-wise operations.

Template Parameters

TDataType	Abstract tensor data type
TMemoryResource	Memory resource type

Parameters

a	First input tensor
b	Second input tensor
result	Result tensor
operation_name	Name of the operation for error reporting

Exceptions

std::invalid_argument If shapes don't match or tensors are empty

Here is the call graph for this function:

Here is the caller graph for this function:

The documentation for this struct was generated from the following file:

/__w/Mila/Mila/Mila/Src/Dnn/Compute/Devices/Cpu/Tensors/Operations/CpuTensorOps.Math.ixx

Static Public Member Functions

Static Private Member Functions

Detailed Description

Member Function Documentation

◆ add()

◆ divide()

◆ multiply()

◆ performElementwiseOperation()

◆ subtract()

◆ sum()

◆ validateShapeCompatibility()