Once the Driver API is initialised and an execution context is created This context can be accessed with the moduleį.Primary. For example, to create a new executionĬontext on CUDA device 0: > ctx > destroy ctxĮach device also has a unique context which is used by the Runtime API. Operations, such as memory allocation and kernel execution, take place A context is associated with a particular device, and all Once you have chosen a device to use, the next step is to create a CUDAĬontext. Seeį for additional operations to query the This package also includes the executable 'nvidia-device-query', which whenĮxecuted displays the key properties of all available devices. The number of available CUDA-capable devices is given viaĭeviceProperties Given a device handle, we can query the properties of that device using
To a compute device at a given ordinal using the device operation. Is assigned a unique identifier (beginning at zero). Next, we must select a GPU that we will execute operations on. This is becauseĬUDA maintains CPU-local state, so operations should always be run fromīefore any operation can be performed, the Driver API must be Should be launched with the option -fno-ghci-sandbox. The steps canīe copied into a file, or run directly in ghci, in which case ghci The following is a short tutorial on using the Driver API.
Furthermore, since itĭoes not require compiling and linking the program with nvcc, theĭriver API provides better inter-language compatibility. Although more difficult to use initially, the DriverĪPI provides more control over how CUDA is used. With operations such as initialisation, context management, and loading Using the Driver API, the programmer must deal explicitly Is a lower-level interface to CUDA devices than that provided by the This module defines an interface to the CUDA driver API.
0 kommentar(er)
