TensorOps.Zero.ixx File Reference

Device-dispatched fast zero operation for tensor buffers. More...

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

Namespaces
namespace	Mila
	Mila main API namespace.

Functions
template<TensorDataType TDataType, typename TMemoryResource> requires isValidTensor<TDataType, TMemoryResource>
void	Mila::Dnn::zero (Tensor< TDataType, TMemoryResource > &tensor, IExecutionContext *exec_context=nullptr)
	Zero a tensor using the fastest backend implementation.

Detailed Description

Device-dispatched fast zero operation for tensor buffers.

Provides the high-level, device-agnostic entry point zero(...) that forwards to the backend TensorOps<device>::zero(...) implementation.

Function Documentation

zero()

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

void Mila::Dnn::zero ( Tensor< TDataType, TMemoryResource > & tensor, IExecutionContext * exec_context = nullptr )

export

Zero a tensor using the fastest backend implementation.

Forwards to the device-specific TensorOps<device>::zero implementation.

Template Parameters

TDataType	Abstract tensor data type.
TMemoryResource	Memory resource type backing the tensor.

Parameters

tensor	Destination tensor to be zeroed.
exec_context	Optional execution context for stream control (borrowed).

Note

If exec_context is provided the backend should schedule the zero on the context's stream and avoid synchronizing. If exec_context is null the backend may synchronize before returning to provide synchronous semantics.