Core macros and utilities. More...

Data Structures
class	exception
	Exception class for CUDA errors. More...

Functions
void	check_error (const char *file, const int line)
	Terminate if cudaGetLastError() returns error. More...

void	async_check_error (const char *file, const int line)
	Terminate if cudaGetLastError() returns error. More...

void	safe_call (cudaError_t err, const char *file, const int line)
	Wrapper to check error code for any cuda call that returns one. Terminates if error is returned. More...

void	handler (std::exception &e, const char *file, const int line)
	Simple exception handler. More...

template<typename T >
T *	device_malloc (int n) throw (exception)
	Allocate memory on device. More...

template<typename T >
T *	mapped_malloc (int n) throw (exception)
	Allocate mapped memory on host. More...

template<typename T >
T *	host_malloc (int n) throw ( exception )
	Allocate host memory. More...

template<typename T >
void	device_free (T *ptr) throw ( exception )
	Free device memory. More...

template<typename T >
void	mapped_free (T *ptr) throw ( exception )
	Free mapped memory. More...

template<typename T >
void	host_free (T *ptr)
	Free host memory (included just for consistency). More...

template<typename T >
void	copy_device_to_host (T dest, T src, int n) throw ( exception )
	Copy from device to host. More...

template<typename T >
void	copy_host_to_device (T dest, T src, int n) throw ( exception )
	Copy from host to device. More...

template<typename T >
void	copy_device_to_device (T dest, T src, int n) throw ( exception )
	Copy from device to device. More...

template<typename T >
void	copy_host_to_host (T dest, T src, int n) throw ( exception )
	Copy from host to host. More...

void	device_synchronize () throw ( exception )
	Wrapper for cudaDeviceSynchronize() More...

void	get_memory_info () throw ( exception )
	Wrapper for cudaMemGetInfo() More...

template<typename T1 , typename T2 >
__global__ void	fill (T1 *z, T2 a, int n)
	Fill device array z with constant a. More...

template<typename T1 , typename T2 >
__global__ void	seq (T1 *z, T2 first, int n)
	Fill device array z with sequence. More...

template<typename T1 , typename T2 >
__global__ void	seq (T1 *z, T2 first, T2 inc, int n)
	Fill device array z with sequence. More...

template<typename T1 , typename T2 >
__global__ void	rep (T1 z, int n, T2 x, int nx, int ncopies)
	Fill device array z with values from x, repeated ncopies times. More...

template<typename T1 , typename T2 >
__global__ void	tile (T1 z, int n, T2 x, int nx, int ncopies)
	Fill device array z with values from x, tiled ncopies times. More...

__global__ void	aX_plus_bY (float z, float a, float x, float b, float *y, int n)
	Elementwise operation, z = ax + by. More...

__global__ void	max_X_Y (float z, float x, float *y, int n)
	Elementwise operation, z = max( x, y ) More...

__global__ void	min_X_Y (float z, float x, float *y, int n)
	Elementwise operation, z = min( x, y) More...

__global__ void	log_X (float z, float x, int n)
	Elementwise operation, z = log( x ) More...

__global__ void	exp_X (float z, float x, int n)
	Elementwise operation, z = exp( x ) More...

Detailed Description

Core macros and utilities.

Function Documentation

void mycuda::async_check_error	(	const char *	file,
		const int	line
	)

Terminate if cudaGetLastError() returns error.

Notes

Calls cudaDeviceSynchronize() first.
Call using macro, e.g., CUDA_ASYNC_CHECK_ERROR().

Definition at line 66 of file mycuda.h.

__global__ void mycuda::aX_plus_bY	(	float *	z,
		float	a,
		float *	x,
		float	b,
		float *	y,
		int	n
	)

Elementwise operation, z = a*x + b*y.

Definition at line 411 of file mycuda.h.

void mycuda::check_error	(	const char *	file,
		const int	line
	)

Terminate if cudaGetLastError() returns error.

Notes

Call using macro, e.g., CUDA_CHECK_ERROR().

Definition at line 47 of file mycuda.h.

template<typename T >

void mycuda::copy_device_to_device	(	T *	dest,
		T *	src,
		int	n
	)
throw	(	exception
	)

Copy from device to device.

Example

copy_device_to_device( dest, src, n );

Definition at line 276 of file mycuda.h.

template<typename T >

void mycuda::copy_device_to_host	(	T *	dest,
		T *	src,
		int	n
	)
throw	(	exception
	)

Copy from device to host.

Example

copy_device_to_host( dest, src, n );

Definition at line 245 of file mycuda.h.

template<typename T >

void mycuda::copy_host_to_device	(	T *	dest,
		T *	src,
		int	n
	)
throw	(	exception
	)

Copy from host to device.

Example

copy_host_to_device( dest, src, n );

Definition at line 261 of file mycuda.h.

template<typename T >

void mycuda::copy_host_to_host	(	T *	dest,
		T *	src,
		int	n
	)
throw	(	exception
	)

Copy from host to host.

Notes

Use for mapped memory copies.

Example

copy_host_to_host( dest, src, n );

Definition at line 294 of file mycuda.h.

template<typename T >

void mycuda::device_free	(	T *	ptr)
throw	(	exception
	)

Free device memory.

Definition at line 209 of file mycuda.h.

template<typename T >

T* mycuda::device_malloc	(	int	n)
throw	(	exception
	)

Allocate memory on device.

Example

float *x = device_malloc<float>( 1024 );

Definition at line 155 of file mycuda.h.

void mycuda::device_synchronize	(	)
throw	(	exception
	)

Wrapper for cudaDeviceSynchronize()

Definition at line 304 of file mycuda.h.

__global__ void mycuda::exp_X	(	float *	z,
		float *	x,
		int	n
	)

Elementwise operation, z = exp( x )

Definition at line 471 of file mycuda.h.

template<typename T1 , typename T2 >

__global__ void mycuda::fill	(	T1 *	z,
		T2	a,
		int	n
	)

Fill device array z with constant a.

Definition at line 328 of file mycuda.h.

void mycuda::get_memory_info	(	)
throw	(	exception
	)

Wrapper for cudaMemGetInfo()

Definition at line 314 of file mycuda.h.

void mycuda::handler	(	std::exception &	e,
		const char *	file,
		const int	line
	)

Simple exception handler.

Example

void bar() { try { foo(); } catch( std::exception& e ) { mycuda::handler( e ); } }

Definition at line 139 of file mycuda.h.

template<typename T >

void mycuda::host_free ( T * ptr)

Free host memory (included just for consistency).

Definition at line 231 of file mycuda.h.

template<typename T >

T* mycuda::host_malloc	(	int	n)
throw	(	exception
	)

Allocate host memory.

Notes

Included for consistency. Better:

float * x = new float[1024]; delete x;

Definition at line 197 of file mycuda.h.

__global__ void mycuda::log_X	(	float *	z,
		float *	x,
		int	n
	)

Elementwise operation, z = log( x )

Definition at line 458 of file mycuda.h.

template<typename T >

void mycuda::mapped_free	(	T *	ptr)
throw	(	exception
	)

Free mapped memory.

Definition at line 220 of file mycuda.h.

template<typename T >

T* mycuda::mapped_malloc	(	int	n)
throw	(	exception
	)

Allocate mapped memory on host.

Example:

float *x = mapped_malloc<float>( 1024 );

Notes

call cudaDeviceSynchronize() before accessing data written on device from host.

Definition at line 177 of file mycuda.h.

__global__ void mycuda::max_X_Y	(	float *	z,
		float *	x,
		float *	y,
		int	n
	)

Elementwise operation, z = max( x, y )

Definition at line 424 of file mycuda.h.

__global__ void mycuda::min_X_Y	(	float *	z,
		float *	x,
		float *	y,
		int	n
	)

Elementwise operation, z = min( x, y)

Definition at line 441 of file mycuda.h.

template<typename T1 , typename T2 >

__global__ void mycuda::rep	(	T1 *	z,
		int	n,
		T2 *	x,
		int	nx,
		int	ncopies
	)

Fill device array z with values from x, repeated ncopies times.

E.g, if x={1,2,3} and ncopies=2, then z={1,1,2,2,3,3}

Definition at line 381 of file mycuda.h.

void mycuda::safe_call	(	cudaError_t	err,
		const char *	file,
		const int	line
	)

Wrapper to check error code for any cuda call that returns one. Terminates if error is returned.

Notes

Call using macro, e.g., CUDA_SAFE_CALL( cudaMalloc( &ptr, n*sizeof(float) )).

Definition at line 84 of file mycuda.h.

template<typename T1 , typename T2 >

__global__ void mycuda::seq	(	T1 *	z,
		T2	first,
		int	n
	)

Fill device array z with sequence.

E.g., z = {first, first+1, ..., first+n-1 }

Definition at line 346 of file mycuda.h.

template<typename T1 , typename T2 >

__global__ void mycuda::seq	(	T1 *	z,
		T2	first,
		T2	inc,
		int	n
	)

Fill device array z with sequence.

E.g., z = {first, first+inc, ..., first+(n-1)*inc }

Definition at line 364 of file mycuda.h.

template<typename T1 , typename T2 >

__global__ void mycuda::tile	(	T1 *	z,
		int	n,
		T2 *	x,
		int	nx,
		int	ncopies
	)

Fill device array z with values from x, tiled ncopies times.

E.g, if x={1,2,3} and ncopies=2, then z={1,2,3,1,2,3}

Definition at line 399 of file mycuda.h.

Data Structures

Functions

Detailed Description

Function Documentation

Notes

Notes

Example

Example

Example

Notes

Example

Example

Example

Notes

Example:

Notes

Notes