TensorOps.Math.ixx File Reference

Device-dispatching math helpers for tensor arithmetic operations. More...

#include <concepts>
#include <memory>
import Compute.DeviceType;
import Compute.ExecutionContext;
import Dnn.TensorOps.Base;
import Dnn.TensorDataTypeMap;
import Dnn.TensorDataTypeTraits;
import Dnn.TensorDataType;
import Dnn.Tensor;

Namespaces
namespace	Mila
	Mila main API namespace.
namespace	Mila::Dnn

Functions
template<TensorDataType TDataType, typename TMemoryResource> requires isValidTensor<TDataType, TMemoryResource>
void	Mila::Dnn::add (const Tensor< TDataType, TMemoryResource > &a, const Tensor< TDataType, TMemoryResource > &b, Tensor< TDataType, TMemoryResource > &result, IExecutionContext *exec_context=nullptr)
	Element-wise addition with optional ExecutionContext (device-dispatched).
template<TensorDataType TDataType, typename TMemoryResource> requires isValidTensor<TDataType, TMemoryResource>
void	Mila::Dnn::divide (const Tensor< TDataType, TMemoryResource > &a, const Tensor< TDataType, TMemoryResource > &b, Tensor< TDataType, TMemoryResource > &result, IExecutionContext *exec_context=nullptr)
	Element-wise division with optional ExecutionContext (device-dispatched).
template<TensorDataType TDataType, typename TMemoryResource> requires isValidTensor<TDataType, TMemoryResource>
void	Mila::Dnn::multiply (const Tensor< TDataType, TMemoryResource > &a, const Tensor< TDataType, TMemoryResource > &b, Tensor< TDataType, TMemoryResource > &result, IExecutionContext *exec_context=nullptr)
	Element-wise multiplication with optional ExecutionContext (device-dispatched).
template<TensorDataType TDataType, typename TMemoryResource> requires isValidTensor<TDataType, TMemoryResource>
Tensor< TDataType, TMemoryResource >	Mila::Dnn::operator* (const Tensor< TDataType, TMemoryResource > &a, const Tensor< TDataType, TMemoryResource > &b)
	Element-wise multiplication operator (always synchronous).
template<TensorDataType TDataType, typename TMemoryResource> requires isValidTensor<TDataType, TMemoryResource>
Tensor< TDataType, TMemoryResource >	Mila::Dnn::operator+ (const Tensor< TDataType, TMemoryResource > &a, const Tensor< TDataType, TMemoryResource > &b)
	Element-wise addition operator (always synchronous).
template<TensorDataType TDataType, typename TMemoryResource> requires isValidTensor<TDataType, TMemoryResource>
Tensor< TDataType, TMemoryResource >	Mila::Dnn::operator- (const Tensor< TDataType, TMemoryResource > &a, const Tensor< TDataType, TMemoryResource > &b)
	Element-wise subtraction operator (always synchronous).
template<TensorDataType TDataType, typename TMemoryResource> requires isValidTensor<TDataType, TMemoryResource>
Tensor< TDataType, TMemoryResource >	Mila::Dnn::operator/ (const Tensor< TDataType, TMemoryResource > &a, const Tensor< TDataType, TMemoryResource > &b)
	Element-wise division operator (always synchronous).
template<TensorDataType TDataType, typename TMemoryResource> requires isValidTensor<TDataType, TMemoryResource>
void	Mila::Dnn::subtract (const Tensor< TDataType, TMemoryResource > &a, const Tensor< TDataType, TMemoryResource > &b, Tensor< TDataType, TMemoryResource > &result, IExecutionContext *exec_context=nullptr)
	Element-wise subtraction with optional ExecutionContext (device-dispatched).
template<TensorDataType TDataType, typename TMemoryResource> requires isValidTensor<TDataType, TMemoryResource>
float	Mila::Dnn::sum (const Tensor< TDataType, TMemoryResource > &tensor, IExecutionContext *exec_context=nullptr)
	Sum reduction with optional ExecutionContext (device-dispatched).

Detailed Description

Device-dispatching math helpers for tensor arithmetic operations.

This partition provides the high-level, device-agnostic entry points for tensor math operations (e.g., element-wise addition). Each helper forwards to the device-specific TensorOps<ComputeDeviceTag>::... implementation (see CPU and CUDA specializations).

The templates are constrained with isValidTensor<TDataType, TMemoryResource> to ensure the tensor configuration is valid (memory resource compatibility, type traits available, and device accessibility).

ExecutionContext handling:

• Optional ExecutionContext parameter for stream control (borrowed, not owned)
• When provided, operations use the context's stream (caller controls sync)
• When null, operations use default stream and synchronize before returning
• Raw pointer semantics ensure zero overhead

Usage:

• Call add(a, b, result) for element-wise addition of two tensors with the same abstract data type and memory resource. The call is automatically dispatched to the appropriate device implementation.
• Optionally provide ExecutionContext for explicit stream control: add(a, b, result, ctx.get())

Preconditions:

• All operands must satisfy isValidTensor and have matching shapes.
• Result tensor must be pre-allocated with matching shape.
• Device-specific implementations validate shapes and perform operations efficiently.
• ExecutionContext (if provided) must outlive the function call.