Platforms

Platforms and platform-specific syntaxes. More...

Modules
	C++ Syntax
	CUDA Syntax
	OpenACC Syntax
	NVIDIA GPU

Classes
class	tinker::ThrustCache
	Device memory cache (allocator) for the Thrust Library. More...

Macros
#define	TINKER_CUDART   1
#define	TINKER_GPULANG_OPENACC   0
#define	TINKER_GPULANG_CUDA   0

Enumerations
enum class	tinker::Platform { tinker::Platform::UNKNOWN = 0x000 , tinker::Platform::ACC = 0x001 , tinker::Platform::CUDA = 0x002 }

Functions
std::string	tinker::cxxCompilerName ()
	Returns the name of the C++ compiler. More...
std::string	tinker::accCompilerName ()
	Returns the name of the OpenACC compiler. More...
std::string	tinker::cudaCompilerName ()
	Returns the name of the CUDA compiler. More...
void	tinker::cudalibData (RcOp)
	Sets up the CUDA variables including but not limited to CUDA streams, CUDA library handles, CUDA memory buffers, and integer units for the OpenACC queues. More...
std::string	tinker::exec (const std::string &cmd)
	Executes a command in shell and returns the output in a string. More...
void	tinker::throwExceptionMissingFunction (const char *functionName, const char *file, int lineNum)
void	tinker::gpuData (RcOp)
	Sets up the GPU card. More...
int	tinker::gpuGridSize (int nthreadsPerBlock)
int	tinker::gpuMaxNParallel (int idev)
void	tinker::platformData (RcOp)

Variables
void *	tinker::pinned_buf
	Preallocated pinned host memory. More...
void *	tinker::dptr_buf
	Preallocated device memory. More...
int	tinker::ndevice
	Total number of visible devices. More...
int	tinker::idevice
	Device ID number in use. More...
Platform	tinker::pltfm_config

Detailed Description

Platforms and platform-specific syntaxes.

---

Class Documentation

tinker::ThrustCache

class tinker::ThrustCache

Device memory cache (allocator) for the Thrust Library.

The strategy of this allocator is as follows: it only expands the size of the allocated memory, and does not free the memory on calling deallocate(value_type*, size_t), until clear() is called.

Public Types
using	value_type = char
Public Member Functions
	ThrustCache ()
value_type *	allocate (ptrdiff_t)
	Expands and returns the allocated memory upon request. More...
void	deallocate (value_type *, size_t)
	Does not do anything. More...
void	clear ()
	Deallocates the memory. More...
Static Public Member Functions
static ThrustCache &	instance ()
	Returns the reference to the singleton object. More...
static void	allocate ()
	Allocates the memory managed by the singleton object. More...
static void	deallocate ()
	Deallocates the memory managed by the singleton object. More...

Member Typedef Documentation

value_type

using tinker::ThrustCache::value_type = char

Constructor & Destructor Documentation

ThrustCache()

tinker::ThrustCache::ThrustCache

(

)

Member Function Documentation

allocate() [1/2]

static void tinker::ThrustCache::allocate ( )

static

Allocates the memory managed by the singleton object.

allocate() [2/2]

value_type * tinker::ThrustCache::allocate

(

ptrdiff_t

)

Expands and returns the allocated memory upon request.

clear()

void tinker::ThrustCache::clear

(

)

Deallocates the memory.

deallocate() [1/2]

static void tinker::ThrustCache::deallocate ( )

static

Deallocates the memory managed by the singleton object.

deallocate() [2/2]

void tinker::ThrustCache::deallocate	(	value_type *	,
		size_t
	)

Does not do anything.

instance()

static ThrustCache & tinker::ThrustCache::instance ( )

static

Returns the reference to the singleton object.

Macro Definition Documentation

TINKER_CUDART

#define TINKER_CUDART   1

Macro for the CUDA runtime-enabled GPU code. Defined to 0 when the code is compiled on the CPU platform.

TINKER_GPULANG_CUDA

#define TINKER_GPULANG_CUDA   0

Macro for the CUDA GPU code. Defined to 0 when

• the code is compiled on the CPU platform;
• OpenACC code is in use on the GPU platform.

TINKER_GPULANG_OPENACC

#define TINKER_GPULANG_OPENACC   0

Macro for the OpenACC GPU code. Defined to 0 when

• the code is compiled on the CPU platform;
• only the CUDA code is in use on the GPU platform.

Enumeration Type Documentation

Platform

enum class tinker::Platform

strong

Flags for the major platforms.

Enumerator
UNKNOWN	Flag for the unknown platform.
ACC	Flag for the OpenACC platform.
CUDA	Flag for the CUDA platform.

Function Documentation

accCompilerName()

std::string tinker::accCompilerName

(

)

Returns the name of the OpenACC compiler.

cudaCompilerName()

std::string tinker::cudaCompilerName

(

)

Returns the name of the CUDA compiler.

cudalibData()

void tinker::cudalibData

(

RcOp

)

Sets up the CUDA variables including but not limited to CUDA streams, CUDA library handles, CUDA memory buffers, and integer units for the OpenACC queues.

cxxCompilerName()

std::string tinker::cxxCompilerName

(

)

Returns the name of the C++ compiler.

exec()

std::string tinker::exec

(

const std::string &

cmd

)

Executes a command in shell and returns the output in a string.

gpuData()

void tinker::gpuData

(

RcOp

)

Sets up the GPU card.

gpuGridSize()

int tinker::gpuGridSize

(

int

nthreadsPerBlock

)

Parameters

nthreadsPerBlock

Dimension of the thread blocks.

Returns

Recommended dimension of the thread grids.

gpuMaxNParallel()

int tinker::gpuMaxNParallel

(

int

idev

)

Parameters

idev	Device number.

Returns

Maximum number of parallel threads on the device.

platformData()

void tinker::platformData

(

RcOp

)

throwExceptionMissingFunction()

void tinker::throwExceptionMissingFunction	(	const char *	functionName,
		const char *	file,
		int	lineNum
	)

Variable Documentation

dptr_buf

void* tinker::dptr_buf

extern

Preallocated device memory.

idevice

int tinker::idevice

extern

Device ID number in use.

ndevice

int tinker::ndevice

extern

Total number of visible devices.

pinned_buf

void* tinker::pinned_buf

extern

Preallocated pinned host memory.

pltfm_config

Platform tinker::pltfm_config

extern

Platform in use.