Struct coaster::frameworks::cuda::Cuda

pub struct Cuda { /* private fields */ }

Provides the Cuda Framework.

Implementations

impl Cuda

pub fn initialise_cublas(&mut self) -> Result<(), Error>

Create a handle to CUBLAS and assign it to CUDA Object

Creating a handle when the CUDA object is created initially will cause CUDA_ERROR_LAUNCH_FAILED when an attempt is made to use the pointer. This can also affect global initialisation of the pointer, and so the initialise must run after the CUDA Driver is fully initialised, or (theoretically) a call is done to CUDA Free or DeviceSynchronise.

pub fn initialise_cudnn(&mut self) -> Result<(), Error>

Create a handle to CUDNN and assign it to CUDA Object

pub fn cudnn(&self) -> &Cudnn

Return a reference to the CUDNN Handle

pub fn cublas(&self) -> &Context

Return a reference to the CUBLAS Handle

Trait Implementations

impl Clone for Cuda

fn clone(&self) -> Cuda

Returns a copy of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for Cuda

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl IFramework for Cuda

fn new_device(&self, hardwares: &[Device]) -> Result<Self::D, Error>

Creates a new Cuda context for computation.

Cuda’s context differs from OpenCL’s context. Multi device support works different in Cuda. This function currently suppports only one device, but should be a wrapper for multi device support.

type H = Device

The Hardware representation for this Framework.

type D = Context

The Device representation for this Framework.

type B = Module

The Binary representation for this Framework.

fn ID() -> &'static str

Defines the Framework by a Name.

fn new() -> Cuda

Initializes a new Framework.

fn load_hardwares() -> Result<Vec<Device>, Error>

Initializes all the available hardwares.

fn hardwares(&self) -> &[Device]

Returns the cached and available hardwares.

fn binary(&self) -> &Self::B

Returns the initialized binary.