HMLP: High-performance Machine Learning Primitives
Public Member Functions | Public Attributes | List of all members
hmlp::Device Class Reference

This class describes devices or accelerators that require a master thread to control. A device can accept tasks from multiple workers. All received tasks are expected to be executed independently in a time-sharing fashion. Whether these tasks are executed in parallel, sequential or with some built-in context switching scheme does not matter. More...

#include <device.hpp>

Inheritance diagram for hmlp::Device:
hmlp::gpu::Nvidia

Public Member Functions

 Device ()
 Device implementation.
 
virtual class CacheLine * getline (size_t size)
 
virtual void prefetchd2h (void *ptr_h, void *ptr_d, size_t size, int stream_id)
 
virtual void prefetchh2d (void *ptr_d, void *ptr_h, size_t size, int stream_id)
 
virtual void waitexecute ()
 
virtual void wait (int stream_id)
 
virtual void * malloc (size_t size)
 
virtual void malloc (void *ptr_d, size_t size)
 
virtual size_t get_memory_left ()
 
virtual void free (void *ptr_d, size_t size)
 

Public Attributes

DeviceType devicetype
 
std::string name
 
class Cache cache
 

Detailed Description

This class describes devices or accelerators that require a master thread to control. A device can accept tasks from multiple workers. All received tasks are expected to be executed independently in a time-sharing fashion. Whether these tasks are executed in parallel, sequential or with some built-in context switching scheme does not matter.

The documentation for this class was generated from the following files: