ipc.hpp File Reference

wrappers for CUDA's facilities for sharing on-device memory addresses and CUDA events between host processes (Inter- Process Communication) More...

#include "context.hpp"
#include "types.hpp"
#include "error.hpp"
#include <string>

Include dependency graph for ipc.hpp:

This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Classes
class	cuda::memory::ipc::imported_ptr_t
	A smart-pointer-like class for memory obtained via inter-process communication. More...

Namespaces
	cuda
	Definitions and functionality wrapping CUDA APIs.
	cuda::memory
	Representation, allocation and manipulation of CUDA-related memory, of different.
	cuda::event
	CUDA timing functionality, via events and their related code (not including the event wrapper type event_t itself)
	cuda::event::ipc
	Definitions and functionality related to CUDA events (not including the event wrapper type event_t itself)

Typedefs
using	cuda::memory::ipc::ptr_handle_t = CUipcMemHandle
	The concrete value passed between processes, used to tell the CUDA Runtime API which memory area is desired.
using	cuda::event::ipc::handle_t = CUipcEventHandle
	The concrete value passed between processes, used to tell the CUDA Runtime API which event is desired. More...

Functions
imported_ptr_t	cuda::memory::ipc::wrap (void *ptr, bool owning) noexcept
	Construct an instance of our wrapper class for IPC-imported memory from a raw pointer to the mapping.
ptr_handle_t	cuda::memory::ipc::export_ (void *device_ptr)
	Obtain a handle for a region of on-device memory which can be transmitted for use in another operating system process. More...
imported_ptr_t	cuda::memory::ipc::import (const ptr_handle_t &ptr_handle)
	Import memory from another process, given the appropriate handle.
handle_t	cuda::event::ipc::export_ (const event_t &event)
	Enable use of an event which this process created by other processes. More...
event_t	cuda::event::ipc::import (const device_t &device, const handle_t &event_ipc_handle)
	Obtain a proper CUDA event, corresponding to an event created by another process, using a handle communicated via operating-system inter-process communications. More...
event_t	cuda::event::ipc::import (const context_t &context, const handle_t &event_ipc_handle)

Detailed Description

wrappers for CUDA's facilities for sharing on-device memory addresses and CUDA events between host processes (Inter- Process Communication)

CUDA addresses into device memory are not valid across different host processes - somewhat, but not entirely, similarly to the case of host memory addresses. Still, there is no reason why different processes should not be able to interact with the same on-device memory region. The same is also true for other entities, such as streams and events.

CUDA provides several functions to enable different processes to share at least memory addresses and events, which are wrapped here. In addition to the free-standing functions, the class cuda::memory::ipc::imported_ptr_t is defined, usable by receiving processes as an 'adapter' to incoming handles which may be passed as-is to code requiring a proper pointer.

Function Documentation

export_()

ptr_handle_t cuda::memory::ipc::export_ ( void *  device_ptr )

inline

Obtain a handle for a region of on-device memory which can be transmitted for use in another operating system process.

Note

The name contains an underscore so as not to clash with the C++ reserved word export

Parameters

device_ptr

beginning of the region of memory to be shared with other processes

Returns

a handle which another process can call detail_::import() on to obtain a device pointer it can use