Mila 0.13.48
Deep Neural Network Library
Loading...
Searching...
No Matches
Mila::Dnn::TensorOps< TDevice > Struct Template Referenceexport

Device-dispatched TensorOps interface template. More...

Detailed Description

template<Compute::DeviceType TDevice>
struct Mila::Dnn::TensorOps< TDevice >

Device-dispatched TensorOps interface template.

Specialize TensorOps<TDevice> for each supported Compute::DeviceType to provide backend implementations of tensor operations (elementwise, reductions, copy, fill, etc.).

Requirements for specializations:

  • Provide the operations used by the framework (static or instance methods), matching the signatures expected by TensorOps callers.
  • Use the device's memory resource and execution context types to access device-specific APIs and streams.
  • Respect host/device accessibility guarantees: CPU specializations must operate on host-accessible memory, CUDA specializations on device memory.

Usage example:

template<>
struct TensorOps<Compute::DeviceType::Cpu>
{
static void copy(const ITensor& src, ITensor& dst);
// ...
};
Mila::Dnn::ITensor
Abstract interface providing essential tensor information and data access.
Definition ITensor.ixx:40
Mila::Dnn::Compute
Definition Device.ixx:15
Mila::Dnn::copy
void copy(const Tensor< TSrcDataType, TSrcMemoryResource > &src, Tensor< TDstDataType, TDstMemoryResource > &dst, IExecutionContext *exec_context=nullptr)
Copies tensor data from source to destination tensor with optional ExecutionContext.
Definition TensorOps.Transfer.ixx:88
Mila::Dnn::TensorOps
Device-dispatched TensorOps interface template.
Definition TensorOps-Base.ixx:44
Template Parameters
TDeviceCompute device type to specialize for (DeviceType::Cpu, DeviceType::Cuda, ...)
The documentation for this struct was generated from the following file: