reprospect.tools.device_properties module
Lightweight wrapper around CUDA driver and runtime libraries.
Provides a minimal interface for interacting with CUDA-capable devices and querying their properties.
Note
This module has some functional overlap with cuda-bindings. The API and the implementation may
evolve to better complement or integrate with the NVIDIA CUDA Python bindings.
- class reprospect.tools.device_properties.Cuda(flags: int = 0)View on GitHub
Bases:
object- __init__(flags: int = 0) NoneView on GitHub
- classmethod check_driver_api_call(*, func: str) AnyView on GitHub
Wrap CUDA driver API call func to raise an exception if the call is not successful.
- classmethod check_driver_status(*, status: CudaDriverError, info: Any) NoneView on GitHub
Check that status is successful, raise otherwise.
- classmethod check_runtime_api_call(*, func: str) AnyView on GitHub
Wrap CUDA runtime API call func to raise an exception if the call is not successful.
- classmethod check_runtime_status(*, status: CudaRuntimeError, info: Any) NoneView on GitHub
Check that status is successful, raise otherwise.
- property device_count: intView on GitHub
Wrap
cudaGetDeviceCount.
- get_device_attribute(*, value_type: type, attribute: CUdevice_attribute, device: int) AnyView on GitHub
Retrieve an attribute of device.
- get_device_compute_capability(*, device: int) ComputeCapabilityView on GitHub
Get compute capability of device.
- get_device_name(*, device: int, length: int = 150) strView on GitHub
Get name of device.
- get_device_total_memory(*, device: int) intView on GitHub
Get device total memory.
- classmethod load() NoneView on GitHub
Load CUDA library.
- class reprospect.tools.device_properties.CudaDriverError(value: int)View on GitHub
Bases:
objectCUDA driver error code.
- property success: boolView on GitHub