cuda::device Namespace Reference

Definitions and functionality related to CUDA devices (not including the device wrapper type device_t itself) More...

Classes
struct	compute_architecture_t
	A numeric designator of an architectural generation of CUDA devices. More...
struct	compute_capability_t
	A numeric designator of the computational capabilities of a CUDA device. More...
struct	pci_location_t
	Location "coordinates" for a CUDA device on a PCIe bus. More...
struct	properties_t
	A structure holding a collection various properties of a device. More...
struct	stream_priority_range_t
	A range of priorities supported by a CUDA device; ranges from the higher numeric value to the lower. More...

Typedefs
using	id_t = int
	Numeric ID of a CUDA device used by the CUDA Runtime API.
using	attribute_t = cudaDeviceAttr
	CUDA devices have both "attributes" and "properties". More...
using	attribute_value_t = int
	All CUDA device attributes (cuda::device::attribute_t) have a value of this type.
using	pair_attribute_t = cudaDeviceP2PAttr
	While Individual CUDA devices have individual "attributes" (attribute_t), there are also attributes characterizing pairs; this type is used for identifying/indexing them, aliasing cudaDeviceP2PAttr.

Enumerations
enum	: device::id_t { default_device_id = 0 }

Functions
device_t	get (id_t device_id) noexcept
	Returns a proxy for the CUDA device with a given id. More...
device_t	get (pci_location_t pci_id)
	Obtain a proxy to a device using its PCI bus location. More...
device_t	get (const ::std::string &pci_id_str)
	Obtain a proxy to a device using a string with its PCI bus location. More...
constexpr compute_capability_t	make_compute_capability (unsigned combined) noexcept
	A named constructor idiom for {compute_capability_t}. More...
constexpr compute_capability_t	make_compute_capability (unsigned major, unsigned minor) noexcept
	A named constructor idiom for {compute_capability_t}.
device::id_t	count ()
	Get the number of CUDA devices usable on the system (with the current CUDA library and kernel driver) More...

Detailed Description

Definitions and functionality related to CUDA devices (not including the device wrapper type device_t itself)

Typedef Documentation

attribute_t

using cuda::device::attribute_t = typedef cudaDeviceAttr

CUDA devices have both "attributes" and "properties".

This is the type for attribute identifiers/indices, aliasing cudaDeviceAttr.

Enumeration Type Documentation

anonymous enum

anonymous enum : device::id_t

Enumerator
default_device_id	If the CUDA runtime has not been set to a specific device, this is the ID of the device it defaults to.

Function Documentation

count()

device::id_t cuda::device::count ( )

inline

Get the number of CUDA devices usable on the system (with the current CUDA library and kernel driver)

Note

This should be returning an unsigned value; unfortunately, device::id_t is signed in CUDA for some reason and we maintain compatibility (although this might change in the future). So... the returned type is the same as in cudaGetDeviceCount, a signed integer.

Returns

the number of CUDA devices on this system

Exceptions

cuda::error

if the device count could not be obtained

get() [1/3]

device_t cuda::device::get ( id_t  device_id )

inlinenoexcept

Returns a proxy for the CUDA device with a given id.

Parameters

device_id

the ID for which to obtain the device proxy

Note

direct constructor access is blocked so that you don't get the idea you're actually creating devices

get() [2/3]

device_t cuda::device::get ( pci_location_t  pci_id )

inline

Obtain a proxy to a device using its PCI bus location.

Parameters

pci_id

The domain-bus-device triplet locating the GPU on the PCI bus

Returns

a device_t proxy object for the device at the specified location

get() [3/3]

device_t cuda::device::get ( const ::std::string &  pci_id_str )

inline

Obtain a proxy to a device using a string with its PCI bus location.

Parameters

pci_id_str

A string listing of the GPU's location on the PCI bus

Returns

a device_t proxy object for the device at the specified location

Note

I'm not very happy about the assumption that get-device-by-string means get-device-by-pci-location; that's not such a great assumption to make IMHO. But - it's convenient for now and there's no immediate risk from some other obvious source of CUDA-device-identifying strings.

make_compute_capability()

constexpr compute_capability_t cuda::device::make_compute_capability ( unsigned  combined )

noexcept

A named constructor idiom for {compute_capability_t}.

Parameters

combined

A combination of the major and minor version number, e.g. 91 for major 9, minor 1