Object representing a kernel. More...

#include <CudaKernel.hpp>

Public Member Functions
	operator const void * () const
	Implicitly convert to void* so this works in CUDA API functions. More...

cudaFuncAttributes	getAttributes () const
	Query function attributes. More...

void	setAttribute (cudaFuncAttribute attr, int value) const

void	setCacheConfig (cudaFuncCache config)
	Set the cache mode. More...

void	setSharedMemConfig (cudaSharedMemConfig config)
	Set the shared memory mode. More...

int	getMaxActiveBlocksPerSM (int blockSize, size_t dynamicSMemSize)

int	getMaxActiveBlocksPerSMIgnoreCaching (int blockSize, size_t dynamicSmemSize)

Static Public Member Functions
static CudaKernel &	get (const void *func)
	Get the global object representing the given kernel. More...

template<typename... Args>
static CudaKernel &	get (void func(Args...))

Detailed Description

Object representing a kernel.

A single, global Kernel object exists for every GPU kernel in the program. This object stores the per-kernel state that can be manipulated via the API, such as cache mode, shared memory bank size, and so on.

This object's constructors are private. Use CudaKernel::get() for initialisation.

When doing kernel launches there are several options to choose from.

See also: Stream::launchKernel

Member Function Documentation

◆ get()

static CudaKernel & sp::CudaKernel::get ( const void * func )

static

Get the global object representing the given kernel.

◆ getAttributes()

cudaFuncAttributes sp::CudaKernel::getAttributes ( ) const

Query function attributes.

See also: cudaFuncGetAttributes()

◆ getMaxActiveBlocksPerSM()

int sp::CudaKernel::getMaxActiveBlocksPerSM	(	int	blockSize,
		size_t	dynamicSMemSize
	)

See also: cudaOccupancyMaxActiveBlocksPerMultiprocessor()

◆ getMaxActiveBlocksPerSMIgnoreCaching()

int sp::CudaKernel::getMaxActiveBlocksPerSMIgnoreCaching	(	int	blockSize,
		size_t	dynamicSmemSize
	)

See also: cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlags()

◆ operator const void *()

sp::CudaKernel::operator const void * ( ) const

Implicitly convert to void* so this works in CUDA API functions.

◆ setCacheConfig()

void sp::CudaKernel::setCacheConfig ( cudaFuncCache config )

Set the cache mode.

See also: cudaFuncSetCacheConfig()

◆ setSharedMemConfig()

void sp::CudaKernel::setSharedMemConfig ( cudaSharedMemConfig config )

Set the shared memory mode.

See also: cudaFuncSetSharedMemConfig

Public Member Functions

Static Public Member Functions