uni20::cuda::DeviceCapabilities Struct Reference

Immutable hardware properties cached for one CUDA device ordinal. More...

#include <device.hpp>

Public Attributes
int	async_engine_count = 0
	Number of asynchronous copy engines.
bool	can_map_host_memory = false
	Whether the device can map page-locked host memory.
int	compute_capability_major = 0
	CUDA compute-capability major version.
int	compute_capability_minor = 0
	CUDA compute-capability minor version.
bool	concurrent_kernels = false
	Whether the device supports concurrent kernel execution.
bool	concurrent_managed_access = false
	Whether CPU and device may concurrently access managed memory.
bool	cooperative_launch = false
	Whether cooperative kernel launch is supported.
bool	ecc_enabled = false
	Whether ECC is enabled on the device.
bool	managed_memory = false
	Whether the device supports managed-memory allocation.
int	max_threads_per_block = 0
	Maximum threads in one thread block.
int	max_threads_per_multiprocessor = 0
	Maximum resident threads per streaming multiprocessor.
bool	memory_pools_supported = false
	Whether stream-ordered CUDA memory pools are supported.
int	multiprocessor_count = 0
	Number of streaming multiprocessors.
std::string	name {}
	CUDA runtime device name.
bool	pageable_memory_access = false
	Whether the device can directly access pageable host memory.
int	pci_bus = 0
	PCI bus identifier.
int	pci_device = 0
	PCI device identifier.
int	pci_domain = 0
	PCI domain identifier.
std::size_t	shared_memory_per_block = 0
	Maximum shared memory per thread block in bytes.
std::size_t	shared_memory_per_multiprocessor = 0
	Maximum shared memory per streaming multiprocessor in bytes.
bool	stream_priorities_supported = false
	Whether CUDA stream priorities are supported.
std::size_t	total_global_memory = 0
	Total device global memory in bytes.
bool	unified_addressing = false
	Whether the device uses unified virtual addressing.
std::array< std::uint8_t, 16 >	uuid {}
	Stable 16-byte CUDA device UUID.
int	warp_size = 0
	Hardware warp size in threads.

Detailed Description

Immutable hardware properties cached for one CUDA device ordinal.

Device::capabilities() returns the process-wide cached instance. Provider-library support and mutable execution resources are not device hardware capabilities and are tracked separately.

Member Data Documentation

async_engine_count

int uni20::cuda::DeviceCapabilities::async_engine_count = 0

Number of asynchronous copy engines.

can_map_host_memory

bool uni20::cuda::DeviceCapabilities::can_map_host_memory = false

Whether the device can map page-locked host memory.

compute_capability_major

int uni20::cuda::DeviceCapabilities::compute_capability_major = 0

CUDA compute-capability major version.

compute_capability_minor

int uni20::cuda::DeviceCapabilities::compute_capability_minor = 0

CUDA compute-capability minor version.

concurrent_kernels

bool uni20::cuda::DeviceCapabilities::concurrent_kernels = false

Whether the device supports concurrent kernel execution.

concurrent_managed_access

bool uni20::cuda::DeviceCapabilities::concurrent_managed_access = false

Whether CPU and device may concurrently access managed memory.

cooperative_launch

bool uni20::cuda::DeviceCapabilities::cooperative_launch = false

Whether cooperative kernel launch is supported.

ecc_enabled

bool uni20::cuda::DeviceCapabilities::ecc_enabled = false

Whether ECC is enabled on the device.

managed_memory

bool uni20::cuda::DeviceCapabilities::managed_memory = false

Whether the device supports managed-memory allocation.

max_threads_per_block

int uni20::cuda::DeviceCapabilities::max_threads_per_block = 0

Maximum threads in one thread block.

max_threads_per_multiprocessor

int uni20::cuda::DeviceCapabilities::max_threads_per_multiprocessor = 0

Maximum resident threads per streaming multiprocessor.

memory_pools_supported

bool uni20::cuda::DeviceCapabilities::memory_pools_supported = false

Whether stream-ordered CUDA memory pools are supported.

multiprocessor_count

int uni20::cuda::DeviceCapabilities::multiprocessor_count = 0

Number of streaming multiprocessors.

name

std::string uni20::cuda::DeviceCapabilities::name {}

CUDA runtime device name.

pageable_memory_access

bool uni20::cuda::DeviceCapabilities::pageable_memory_access = false

Whether the device can directly access pageable host memory.

pci_bus

int uni20::cuda::DeviceCapabilities::pci_bus = 0

PCI bus identifier.

pci_device

int uni20::cuda::DeviceCapabilities::pci_device = 0

PCI device identifier.

pci_domain

int uni20::cuda::DeviceCapabilities::pci_domain = 0

PCI domain identifier.

shared_memory_per_block

std::size_t uni20::cuda::DeviceCapabilities::shared_memory_per_block = 0

Maximum shared memory per thread block in bytes.

shared_memory_per_multiprocessor

std::size_t uni20::cuda::DeviceCapabilities::shared_memory_per_multiprocessor = 0

Maximum shared memory per streaming multiprocessor in bytes.

stream_priorities_supported

bool uni20::cuda::DeviceCapabilities::stream_priorities_supported = false

Whether CUDA stream priorities are supported.

total_global_memory

std::size_t uni20::cuda::DeviceCapabilities::total_global_memory = 0

Total device global memory in bytes.

unified_addressing

bool uni20::cuda::DeviceCapabilities::unified_addressing = false

Whether the device uses unified virtual addressing.

uuid

std::array<std::uint8_t, 16> uni20::cuda::DeviceCapabilities::uuid {}

Stable 16-byte CUDA device UUID.

warp_size

int uni20::cuda::DeviceCapabilities::warp_size = 0

Hardware warp size in threads.

---

The documentation for this struct was generated from the following file:

• device.hpp