Mila::Dnn::Compute::CudaDeviceProps Class Reference

Wrapper for CUDA device properties with cached values. More...

Public Member Functions
	CudaDeviceProps (int device_id)
	Constructs device properties for specified CUDA device.
std::pair< int, int >	getComputeCapability () const
	Gets compute capability as major/minor version pair.
std::string	getName () const
	Gets the device name.

Public Attributes
int	clockRate { 0 }
int	l2CacheSize { 0 }
int	major { 0 }
int	maxGridSize [3] { 0, 0, 0 }
int	maxThreadsDim [3] { 0, 0, 0 }
int	maxThreadsPerBlock { 0 }
int	memoryBusWidth { 0 }
int	memoryClockRate { 0 }
int	minor { 0 }
int	multiProcessorCount { 0 }
std::string	name
int	pciBusID { 0 }
int	pciDeviceID { 0 }
int	pciDomainID { 0 }
size_t	sharedMemPerBlock { 0 }
size_t	totalGlobalMem { 0 }
int	warpSize { 0 }

Private Attributes
int	device_id_ { 0 }

Detailed Description

Wrapper for CUDA device properties with cached values.

Queries and caches CUDA device properties on construction. The underlying cudaDeviceProp struct is used only during construction as a local variable and is not retained as a member to keep this class safe to import from any module that exports a class template.

CUDA 13.0+ Compatibility:

• clockRate and memoryClockRate fields are deprecated (always 0).
• Uses cudaDeviceGetAttribute for clock rates instead.

Constructor & Destructor Documentation

CudaDeviceProps()

Mila::Dnn::Compute::CudaDeviceProps::CudaDeviceProps ( int device_id )

inlineexplicit

Constructs device properties for specified CUDA device.

Queries CUDA runtime for device properties and caches all needed values. For CUDA 13.0+, clock rates are queried via device attributes.

Parameters

device_id

CUDA device ID (0-based).

Exceptions

std::runtime_error

If device properties query fails.

Here is the call graph for this function:

Member Function Documentation

getComputeCapability()

std::pair< int, int > Mila::Dnn::Compute::CudaDeviceProps::getComputeCapability ( ) const

inline

Gets compute capability as major/minor version pair.

Returns

std::pair<int, int> e.g. {8, 9} for SM 8.9.

getName()

std::string Mila::Dnn::Compute::CudaDeviceProps::getName ( ) const

inline

Gets the device name.

Returns

std::string Device name.

Member Data Documentation

clockRate

int Mila::Dnn::Compute::CudaDeviceProps::clockRate { 0 }

device_id_

int Mila::Dnn::Compute::CudaDeviceProps::device_id_ { 0 }

private

l2CacheSize

int Mila::Dnn::Compute::CudaDeviceProps::l2CacheSize { 0 }

major

int Mila::Dnn::Compute::CudaDeviceProps::major { 0 }

maxGridSize

int Mila::Dnn::Compute::CudaDeviceProps::maxGridSize[3] { 0, 0, 0 }

maxThreadsDim

int Mila::Dnn::Compute::CudaDeviceProps::maxThreadsDim[3] { 0, 0, 0 }

maxThreadsPerBlock

int Mila::Dnn::Compute::CudaDeviceProps::maxThreadsPerBlock { 0 }

memoryBusWidth

int Mila::Dnn::Compute::CudaDeviceProps::memoryBusWidth { 0 }

memoryClockRate

int Mila::Dnn::Compute::CudaDeviceProps::memoryClockRate { 0 }

minor

int Mila::Dnn::Compute::CudaDeviceProps::minor { 0 }

multiProcessorCount

int Mila::Dnn::Compute::CudaDeviceProps::multiProcessorCount { 0 }

name

std::string Mila::Dnn::Compute::CudaDeviceProps::name

pciBusID

int Mila::Dnn::Compute::CudaDeviceProps::pciBusID { 0 }

pciDeviceID

int Mila::Dnn::Compute::CudaDeviceProps::pciDeviceID { 0 }

pciDomainID

int Mila::Dnn::Compute::CudaDeviceProps::pciDomainID { 0 }

sharedMemPerBlock

size_t Mila::Dnn::Compute::CudaDeviceProps::sharedMemPerBlock { 0 }

totalGlobalMem

size_t Mila::Dnn::Compute::CudaDeviceProps::totalGlobalMem { 0 }

warpSize

int Mila::Dnn::Compute::CudaDeviceProps::warpSize { 0 }

---

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

• /__w/Mila/Mila/Mila/Src/Dnn/Compute/Devices/Cuda/CudaDeviceProps.ixx