Tissue Forge GPU acceleration on CUDA-supporting devices. More...

Classes
struct	Angle

struct	AngleConfig
	CUDA runtime control interface for Tissue Forge angles. More...

struct	Bond

struct	BondConfig
	CUDA runtime control interface for Tissue Forge bonds. More...

struct	BoundaryCondition

struct	BoundaryConditions

struct	CUDAContext
	A convenience wrap of the CUDA context for JIT-compiled Tissue Forge programs. More...

struct	CUDADevice
	A simple interface with a CUDA device. More...

struct	CUDAFunction
	A CUDA kernel from a JIT-compiled Tissue Forge program. More...

struct	CUDARTProgram
	A JIT-compiled CUDA Tissue Forge program. More...

struct	CUDARTSource
	Convenience class for loading source from file and storing, here intended for CUDA. More...

struct	DPDPotentialData

struct	EngineConfig
	CUDA runtime control interface for Tissue Forge engine. More...

struct	Flux

struct	Fluxes

struct	FluxTypeIdPair

struct	Particle

struct	Potential

struct	PotentialData

struct	queue_cuda

struct	SimulatorConfig
	CUDA runtime control interface for Simulator. More...

struct	task_cuda

Enumerations
enum	ErrorCode : int { TFCUDAERR_ok = 0 , TFCUDAERR_setdevice , TFCUDAERR_setblocks , TFCUDAERR_setthreads , TFCUDAERR_ondevice , TFCUDAERR_notondevice , TFCUDAERR_cleardevices , TFCUDAERR_refresh , TFCUDAERR_send , TFCUDAERR_pull , TFCUDAERR_LAST }

Functions
int	Angle_setThreads (const unsigned int &nr_threads)

int	Angle_setBlocks (const unsigned int &nr_blocks)

int	Angle_getDevice ()

int	Angle_toDevice (engine *e)

int	Angle_fromDevice (engine *e)

int	Angle_refresh (engine *e)

int	Angle_refreshAngle (engine e, AngleHandle a)

int	Angle_refreshAngles (engine e, AngleHandle *angles, int nr_angles)

int	engine_cuda_add_angle (AngleHandle *ah)

int	engine_cuda_finalize_angle (int aind)

int	engine_cuda_finalize_angles (engine e, int ainds, int nr_angles)

int	engine_cuda_finalize_angles_all (engine *e)

int	engine_angle_eval_cuda (struct TissueForge::Angle angles, int N, struct engine e, float *epot_out)

int	Bond_setThreads (const unsigned int &nr_threads)

int	Bond_setBlocks (const unsigned int &nr_blocks)

int	Bond_getDevice ()

int	Bond_setDevice (engine *e, const int &deviceId)

int	Bond_toDevice (engine *e)

int	Bond_fromDevice (engine *e)

int	Bond_refresh (engine *e)

int	Bond_refreshBond (engine e, BondHandle b)

int	Bond_refreshBonds (engine e, BondHandle *bonds, int nr_bonds)

int	engine_cuda_add_bond (TissueForge::Bond *b)

int	engine_cuda_add_bonds (TissueForge::Bond *bonds, int nr_bonds)

int	engine_cuda_finalize_bond (int bind)

int	engine_cuda_finalize_bonds (engine e, int binds, int nr_bonds)

int	engine_cuda_finalize_bonds_all (engine *e)

int	engine_bond_eval_cuda (struct TissueForge::Bond bonds, int N, struct engine e, float *epot_out)

__device__ void	Flux_getFluxes (int fxind_cuda, Fluxes fluxes_cuda)

__device__ void	Flux_getNrFluxes (unsigned int *nr_fluxes)

__device__ void	Flux_getNrStates (unsigned int *nr_states)

TissueForge::Potential	toCUDADevice (const TissueForge::Potential &p)
	Loads a potential onto a CUDA device.

__host__ __device__ void	cudaFree (TissueForge::Potential *p)

__device__ void	potential_eval_r_cuda (struct TissueForge::Potential p, FPTYPE r, FPTYPE e, FPTYPE *f)
	Evaluates the given potential at the given point (interpolated).

__device__ void	potential_eval_cuda (struct TissueForge::Potential p, float r2, float e, float *f)
	Evaluates the given potential at the given point (interpolated).

CUresult	cuda_errorchk (CUresult retCode, const char *file, int line)

nvrtcResult	nvrtc_errorchk (nvrtcResult retCode, const char *file, int line)

cudaError_t	cudart_errorchk (cudaError_t retCode, const char *file, int line)

void	init ()
	Initialize CUDA.

void	setGLDevice (const int &deviceId)

std::string	getDeviceName (const int &deviceId)
	Get the name of a device.

size_t	getDeviceTotalMem (const int &deviceId)
	Get the total memory of device.

int	getDeviceAttribute (const int &deviceId, const int &attrib)
	Get the attribute value of a device.

int	getNumDevices ()
	Get number of available compute-capable devices.

std::string	getDevicePCIBusId (const int &deviceId)
	Get the PCI bus id of a device.

int	getCurrentDevice ()
	Get the device id of the current context of the calling CPU thread.

int	maxThreadsPerBlock (const int &deviceId)
	Maximum number of threads per block.

int	maxBlockDimX (const int &deviceId)
	Maximum x-dimension of a block.

int	maxBlockDimY (const int &deviceId)
	Maximum y-dimension of a block.

int	maxBlockDimZ (const int &deviceId)
	Maximum z-dimension of a block.

int	maxGridDimX (const int &deviceId)
	Maximum x-dimension of a grid.

int	maxGridDimY (const int &deviceId)
	Maximum y-dimension of a grid.

int	maxGridDimZ (const int &deviceId)
	Maximum z-dimension of a grid.

int	maxSharedMemPerBlock (const int &deviceId)
	Maximum amount of shared memory available to a thread block in bytes.

int	maxTotalMemConst (const int &deviceId)
	Memory available on device for constant variables in a CUDA C kernel in bytes.

int	warpSize (const int &deviceId)
	Warp size in threads.

int	maxRegsPerBlock (const int &deviceId)
	Maximum number of 32-bit registers available to a thread block.

int	clockRate (const int &deviceId)
	The typical clock frequency in kilohertz.

bool	gpuOverlap (const int &deviceId)
	Test if the device can concurrently copy memory between host and device while executing a kernel.

int	numMultiprocessors (const int &deviceId)
	Number of multiprocessors on the device.

bool	kernelExecTimeout (const int &deviceId)
	Test if there is a run time limit for kernels executed on the device.

bool	computeModeDefault (const int &deviceId)
	Test if device is not restricted and can have multiple CUDA contexts present at a single time.

bool	computeModeProhibited (const int &deviceId)
	Test if device is prohibited from creating new CUDA contexts.

bool	computeModeExclusive (const int &deviceId)
	Test if device can have only one context used by a single process at a time.

int	PCIDeviceId (const int &deviceId)
	PCI device (also known as slot) identifier of the device.

int	PCIDomainId (const int &deviceId)
	PCI domain identifier of the device.

int	clockRateMem (const int &deviceId)
	Peak memory clock frequency in kilohertz.

int	globalMemBusWidth (const int &deviceId)
	Global memory bus width in bits.

int	L2CacheSize (const int &deviceId)
	Size of L2 cache in bytes. 0 if the device doesn't have L2 cache.

int	maxThreadsPerMultiprocessor (const int &deviceId)
	Maximum resident threads per multiprocessor.

int	computeCapabilityMajor (const int &deviceId)
	Major compute capability version number.

int	computeCapabilityMinor (const int &deviceId)
	Minor compute capability version number.

bool	L1CacheSupportGlobal (const int &deviceId)
	Test if device supports caching globals in L1 cache.

bool	L1CacheSupportLocal (const int &deviceId)
	Test if device supports caching locals in L1 cache.

int	maxSharedMemPerMultiprocessor (const int &deviceId)
	Maximum amount of shared memory available to a multiprocessor in bytes.

int	maxRegsPerMultiprocessor (const int &deviceId)
	Maximum number of 32-bit registers available to a multiprocessor.

bool	managedMem (const int &deviceId)
	Test if device supports allocating managed memory on this system.

bool	multiGPUBoard (const int &deviceId)
	Test if device is on a multi-GPU board.

int	multiGPUBoardGroupId (const int &deviceId)
	Unique identifier for a group of devices associated with the same board.

void	test (const int &numBlocks, const int &numThreads, const int &numEls, const int &deviceId=0)
	Tests JIT-compiled program execution and deployment.

std::string	tfIncludePath ()
	Returns the current path to the installed Tissue Forge include directory.

HRESULT	setTfIncludePath (const std::string &_path)
	Set the current path to the installed Tissue Forge include directory.

std::string	tfPrivateIncludePath ()
	Returns the path to the installed Tissue Forge private include directory.

std::string	tfResourcePath ()
	Returns the current path to the installed Tissue Forge resource directory.

HRESULT	setTfResourcePath (const std::string &_path)
	Set the current path to the installed Tissue Forge resource directory.

std::string	CUDAPath ()
	Returns the path to the installed Tissue Forge CUDA resources directory.

std::string	CUDAIncludePath ()
	Returns the current path to the installed CUDA include directory.

HRESULT	setCUDAIncludePath (const std::string &_path)
	Set the current path to the installed CUDA include directory.

std::string	CUDAResourcePath (const std::string &relativePath)
	Returns an absolute path to a subdirectory of the install Tissue Forge CUDA resources directory.

std::string	CUDAPTXObjectRelPath ()
	Returns the relative path to the installed Tissue Forge CUDA PTX object directory.

std::vector< std::string >	CUDAArchs ()
	Returns the supported CUDA architectures of the installation.

Detailed Description

Tissue Forge GPU acceleration on CUDA-supporting devices.

Function Documentation

◆ clockRate()

int TissueForge::cuda::clockRate ( const int & deviceId )

The typical clock frequency in kilohertz.

Returns: int

◆ clockRateMem()

int TissueForge::cuda::clockRateMem ( const int & deviceId )

Peak memory clock frequency in kilohertz.

Returns: int

◆ computeCapabilityMajor()

int TissueForge::cuda::computeCapabilityMajor ( const int & deviceId )

Major compute capability version number.

Returns: int

◆ computeCapabilityMinor()

int TissueForge::cuda::computeCapabilityMinor ( const int & deviceId )

Minor compute capability version number.

Returns: int

◆ computeModeDefault()

bool TissueForge::cuda::computeModeDefault ( const int & deviceId )

Test if device is not restricted and can have multiple CUDA contexts present at a single time.

Returns: true; false

◆ computeModeExclusive()

bool TissueForge::cuda::computeModeExclusive ( const int & deviceId )

Test if device can have only one context used by a single process at a time.

Returns: true; false

◆ computeModeProhibited()

bool TissueForge::cuda::computeModeProhibited ( const int & deviceId )

Test if device is prohibited from creating new CUDA contexts.

Returns: true; false

◆ CUDAArchs()

std::vector< std::string > TissueForge::cuda::CUDAArchs ( )

Returns the supported CUDA architectures of the installation.

Returns: std::vector<std::string>

◆ CUDAIncludePath()

std::string TissueForge::cuda::CUDAIncludePath ( )

Returns the current path to the installed CUDA include directory.

Returns: std::string

◆ CUDAPath()

std::string TissueForge::cuda::CUDAPath ( )

Returns the path to the installed Tissue Forge CUDA resources directory.

Returns: std::string

◆ CUDAPTXObjectRelPath()

std::string TissueForge::cuda::CUDAPTXObjectRelPath ( )

Returns the relative path to the installed Tissue Forge CUDA PTX object directory.

The path is relative to the Tissue Forge CUDA resources directory, and depends on the build type of the installation.

Returns: std::string

◆ CUDAResourcePath()

std::string TissueForge::cuda::CUDAResourcePath ( const std::string & relativePath )

Returns an absolute path to a subdirectory of the install Tissue Forge CUDA resources directory.

Parameters

relativePath

Returns: std::string

◆ getCurrentDevice()

int TissueForge::cuda::getCurrentDevice ( )

Get the device id of the current context of the calling CPU thread.

Returns: int

◆ getDeviceAttribute()

int TissueForge::cuda::getDeviceAttribute	(	const int &	deviceId,
		const int &	attrib )

Get the attribute value of a device.

Parameters

deviceId
attrib

Returns: int

◆ getDeviceName()

std::string TissueForge::cuda::getDeviceName ( const int & deviceId )

Get the name of a device.

Parameters

deviceId

Returns: std::string

◆ getDevicePCIBusId()

std::string TissueForge::cuda::getDevicePCIBusId ( const int & deviceId )

Get the PCI bus id of a device.

Parameters

deviceId

Returns: std::string

◆ getDeviceTotalMem()

size_t TissueForge::cuda::getDeviceTotalMem ( const int & deviceId )

Get the total memory of device.

Parameters

deviceId

Returns: size_t

◆ getNumDevices()

int TissueForge::cuda::getNumDevices ( )

Get number of available compute-capable devices.

Returns: int

◆ globalMemBusWidth()

int TissueForge::cuda::globalMemBusWidth ( const int & deviceId )

Global memory bus width in bits.

Returns: int

◆ gpuOverlap()

bool TissueForge::cuda::gpuOverlap ( const int & deviceId )

Test if the device can concurrently copy memory between host and device while executing a kernel.

Returns: true; false

◆ kernelExecTimeout()

bool TissueForge::cuda::kernelExecTimeout ( const int & deviceId )

Test if there is a run time limit for kernels executed on the device.

Returns: true; false

◆ L1CacheSupportGlobal()

bool TissueForge::cuda::L1CacheSupportGlobal ( const int & deviceId )

Test if device supports caching globals in L1 cache.

Returns: true; false

◆ L1CacheSupportLocal()

bool TissueForge::cuda::L1CacheSupportLocal ( const int & deviceId )

Test if device supports caching locals in L1 cache.

Returns: true; false

◆ L2CacheSize()

int TissueForge::cuda::L2CacheSize ( const int & deviceId )

Size of L2 cache in bytes. 0 if the device doesn't have L2 cache.

Returns: int

◆ managedMem()

bool TissueForge::cuda::managedMem ( const int & deviceId )

Test if device supports allocating managed memory on this system.

Returns: true; false

◆ maxBlockDimX()

int TissueForge::cuda::maxBlockDimX ( const int & deviceId )

Maximum x-dimension of a block.

Returns: int

◆ maxBlockDimY()

int TissueForge::cuda::maxBlockDimY ( const int & deviceId )

Maximum y-dimension of a block.

Returns: int

◆ maxBlockDimZ()

int TissueForge::cuda::maxBlockDimZ ( const int & deviceId )

Maximum z-dimension of a block.

Returns: int

◆ maxGridDimX()

int TissueForge::cuda::maxGridDimX ( const int & deviceId )

Maximum x-dimension of a grid.

Returns: int

◆ maxGridDimY()

int TissueForge::cuda::maxGridDimY ( const int & deviceId )

Maximum y-dimension of a grid.

Returns: int

◆ maxGridDimZ()

int TissueForge::cuda::maxGridDimZ ( const int & deviceId )

Maximum z-dimension of a grid.

Returns: int

◆ maxRegsPerBlock()

int TissueForge::cuda::maxRegsPerBlock ( const int & deviceId )

Maximum number of 32-bit registers available to a thread block.

Returns: int

◆ maxRegsPerMultiprocessor()

int TissueForge::cuda::maxRegsPerMultiprocessor ( const int & deviceId )

Maximum number of 32-bit registers available to a multiprocessor.

Returns: int

◆ maxSharedMemPerBlock()

int TissueForge::cuda::maxSharedMemPerBlock ( const int & deviceId )

Maximum amount of shared memory available to a thread block in bytes.

Returns: int

◆ maxSharedMemPerMultiprocessor()

int TissueForge::cuda::maxSharedMemPerMultiprocessor ( const int & deviceId )

Maximum amount of shared memory available to a multiprocessor in bytes.

Returns: int

◆ maxThreadsPerBlock()

int TissueForge::cuda::maxThreadsPerBlock ( const int & deviceId )

Maximum number of threads per block.

Returns: int

◆ maxThreadsPerMultiprocessor()

int TissueForge::cuda::maxThreadsPerMultiprocessor ( const int & deviceId )

Maximum resident threads per multiprocessor.

Returns: int

◆ maxTotalMemConst()

int TissueForge::cuda::maxTotalMemConst ( const int & deviceId )

Memory available on device for constant variables in a CUDA C kernel in bytes.

Returns: int

◆ multiGPUBoard()

bool TissueForge::cuda::multiGPUBoard ( const int & deviceId )

Test if device is on a multi-GPU board.

Returns: true; false

◆ multiGPUBoardGroupId()

int TissueForge::cuda::multiGPUBoardGroupId ( const int & deviceId )

Unique identifier for a group of devices associated with the same board.

Returns: int

◆ numMultiprocessors()

int TissueForge::cuda::numMultiprocessors ( const int & deviceId )

Number of multiprocessors on the device.

Returns: int

◆ PCIDeviceId()

int TissueForge::cuda::PCIDeviceId ( const int & deviceId )

PCI device (also known as slot) identifier of the device.

Returns: int

◆ PCIDomainId()

int TissueForge::cuda::PCIDomainId ( const int & deviceId )

PCI domain identifier of the device.

Returns: int

◆ potential_eval_cuda()

__device__ void TissueForge::cuda::potential_eval_cuda	(	struct TissueForge::Potential *	p,
		float	r2,
		float *	e,
		float *	f )

Evaluates the given potential at the given point (interpolated).

Parameters

p	The #potential to be evaluated.
r2	The radius at which it is to be evaluated, squared.
e	Pointer to a floating-point value in which to store the interaction energy.
f	Pointer to a floating-point value in which to store the magnitude of the interaction force divided by r.

Note that for efficiency reasons, this function does not check if any of the parameters are NULL or if sqrt(r2) is within the interval of the #potential p.

◆ potential_eval_r_cuda()

__device__ void TissueForge::cuda::potential_eval_r_cuda	(	struct TissueForge::Potential *	p,
		FPTYPE	r,
		FPTYPE *	e,
		FPTYPE *	f )

Evaluates the given potential at the given point (interpolated).

Parameters

p	The #potential to be evaluated.
r	The radius at which it is to be evaluated.
e	Pointer to a floating-point value in which to store the interaction energy.
f	Pointer to a floating-point value in which to store the magnitude of the interaction force.

Note that for efficiency reasons, this function does not check if any of the parameters are NULL or if sqrt(r2) is within the interval of the #potential p.

◆ setCUDAIncludePath()

HRESULT TissueForge::cuda::setCUDAIncludePath ( const std::string & _path )

Set the current path to the installed CUDA include directory.

Parameters

_path absolute path

Returns: HRESULT

◆ setTfIncludePath()

HRESULT TissueForge::cuda::setTfIncludePath ( const std::string & _path )

Set the current path to the installed Tissue Forge include directory.

Parameters

_path absolute path

Returns: HRESULT

◆ setTfResourcePath()

HRESULT TissueForge::cuda::setTfResourcePath ( const std::string & _path )

Set the current path to the installed Tissue Forge resource directory.

Parameters

_path absolute path

Returns: HRESULT

◆ test()

void TissueForge::cuda::test	(	const int &	numBlocks,
		const int &	numThreads,
		const int &	numEls,
		const int &	deviceId = 0 )

Tests JIT-compiled program execution and deployment.

Enable logger at debug level to see step-by-step report.

Parameters

numBlocks	number of blocks
numThreads	number of threads
numEls	number of elements in calculations
deviceId	ID of CUDA device

◆ tfIncludePath()

std::string TissueForge::cuda::tfIncludePath ( )

Returns the current path to the installed Tissue Forge include directory.

Returns: std::string

◆ tfPrivateIncludePath()

std::string TissueForge::cuda::tfPrivateIncludePath ( )

Returns the path to the installed Tissue Forge private include directory.

Returns: std::string

◆ tfResourcePath()

std::string TissueForge::cuda::tfResourcePath ( )

Returns the current path to the installed Tissue Forge resource directory.

Returns: std::string

◆ toCUDADevice()

TissueForge::Potential TissueForge::cuda::toCUDADevice ( const TissueForge::Potential & p )

Loads a potential onto a CUDA device.

Parameters

p	The potential

Returns: The loaded potential, or NULL if failed

◆ warpSize()

int TissueForge::cuda::warpSize ( const int & deviceId )

Warp size in threads.

Returns: int

Classes

Enumerations

Functions

Detailed Description

Function Documentation

◆ clockRate()

◆ clockRateMem()

◆ computeCapabilityMajor()

◆ computeCapabilityMinor()

◆ computeModeDefault()

◆ computeModeExclusive()

◆ computeModeProhibited()

◆ CUDAArchs()

◆ CUDAIncludePath()

◆ CUDAPath()

◆ CUDAPTXObjectRelPath()

◆ CUDAResourcePath()

◆ getCurrentDevice()

◆ getDeviceAttribute()

◆ getDeviceName()

◆ getDevicePCIBusId()

◆ getDeviceTotalMem()

◆ getNumDevices()

◆ globalMemBusWidth()

◆ gpuOverlap()

◆ kernelExecTimeout()

◆ L1CacheSupportGlobal()

◆ L1CacheSupportLocal()

◆ L2CacheSize()

◆ managedMem()

◆ maxBlockDimX()

◆ maxBlockDimY()

◆ maxBlockDimZ()

◆ maxGridDimX()

◆ maxGridDimY()

◆ maxGridDimZ()

◆ maxRegsPerBlock()

◆ maxRegsPerMultiprocessor()

◆ maxSharedMemPerBlock()

◆ maxSharedMemPerMultiprocessor()

◆ maxThreadsPerBlock()

◆ maxThreadsPerMultiprocessor()

◆ maxTotalMemConst()

◆ multiGPUBoard()

◆ multiGPUBoardGroupId()

◆ numMultiprocessors()

◆ PCIDeviceId()

◆ PCIDomainId()

◆ potential_eval_cuda()

◆ potential_eval_r_cuda()

◆ setCUDAIncludePath()

◆ setTfIncludePath()

◆ setTfResourcePath()

◆ test()

◆ tfIncludePath()

◆ tfPrivateIncludePath()

◆ tfResourcePath()

◆ toCUDADevice()

◆ warpSize()