Interface for an computations acceleration backend. More...

#include <accelerator.hpp>

Inheritance diagram for spla::Accelerator:

Public Member Functions
virtual	~Accelerator ()=default

virtual Status	init ()=0

virtual Status	set_platform (int index)=0

virtual Status	set_device (int index)=0

virtual Status	set_queues_count (int count)=0

virtual const std::string &	get_name ()=0

virtual const std::string &	get_description ()=0

virtual const std::string &	get_suffix ()=0

Detailed Description

Interface for an computations acceleration backend.

Accelerator is an optional library computations backend, which may provided customized and efficient implementations of some operations over matrices and vectors.

Accelerator can implement additional and custom storage schemas on top of the default schemas in matrices and vectors and optional store any data along with default in order to speed-up computations.

Typical accelerator implementation is a GPUs utilization by usage of OpenCL or CUDA API. In this case additional device resident data stored with host data and kernels dispatched in order to perform computations.