SoCuda Class Reference
[CUDA device]

Cuda module class More...

#include <Inventor/cuda/SoCuda.h>

List of all members.

Classes
struct	SoCudaModule
	Defines a CUDA module.
Public Types
enum	SoCudaError {   SO_CUDA_SUCCESS,   SO_CUDA_MODULE_LOAD_ERROR,   SO_CUDA_MODULE_FUNCTION_ERROR,   SO_CUDA_INVALID_PARAMETER,   SO_CUDA_INVALID_CONTEXT }
Static Public Member Functions
static void	init ()
static void	finish ()
static bool	isInitialized ()
static SbBool	isAvailable ()
static SoCudaError	loadBinaryModule (const SbString &filename, const SbString &moduleName)
static SoCudaError	loadBinaryModule (const void *moduleData, size_t dataSize, const SbString &moduleName)
static SoCudaHandle *	findFunction (const SbString &moduleName, const SbString &functionName)
static SoCudaError	findFunction (SoCudaHandle *handle, const SbString &moduleName, const SbString &functionName)

---

Detailed Description

Cuda module class

The Open Inventor CUDA module provides a set of useful functions for CUDA development including memory management, module handling and device management functions.

NOTE: The classes in this module do not exist in Open Inventor 10.0 and later.

• Memory management is provided by the SoCudaBufferObject class.
• Module handling is included in the SoCuda class.
• Device management functions are provided by the classes SoCudaContext and SoCudaDevice.

This module also provides built-in CUDA functions to quickly integrate in your application the CUDA functions useful to improve your computing performences.

The integrated modules are organized in different topics:

• Conversion
• Volumes
• ....

USAGE

After the initialization of OpenInventor the CUDA module must be initialized using the init function. Before exiting the application the finish function must be called, in order to free all the memory allocated by the module.

NOTES

After the initialization of the CUDA module there is no default CudaContext, so don't forget to create a CUDA context before performing any CUDA operation. See the class SoCudaContext for more information.

EXAMPLE

This example shows how to call a specific CUDA function.

  // OpenInventor Initialization
  ...

  SoCuda::init();
  SoCudaDevice* pMainDevice = SoCudaDevice::findFirstAvailableDevice();
  if ( pMainDevice ) {
    SoCudaContext* pContext = new SoCudaContext( pMainDevice );
    if ( pContext ) {
      pContext->bind();
      SoCudaHandle* pHandle = SoCuda::findFunction( "ModuleName", "FunctionName" );
      if ( pHandle ) {
        SoCudaApi::launch( pHandle, 10, 1 );
        SoCudaApi::freeHandle( pHandle );
      }
      pContext->unbind();
    }
  }
  SoCuda::finish();

SEE ALSO

SoCudaApi, SoCudaContext, SoCudaDevice, SoCudaBufferObject, SoCudaArithmetic, SoCudaConversion, SoCudaConvolution, SoCudaSeismic

See related examples:

HelloCUDA, SimpleCUDAKernel, ComputeSlice, CudaMarchingCubes, LDMMandelBulb

---

Member Enumeration Documentation

enum SoCuda::SoCudaError

This enum defines the possible errors returned by the Cuda API.

Enumerator:

SO_CUDA_SUCCESS	No error, the operation is a success.
SO_CUDA_MODULE_LOAD_ERROR	Error: The specified module cannot be loaded.
SO_CUDA_MODULE_FUNCTION_ERROR	Error: The specified function cannot be loaded.
SO_CUDA_INVALID_PARAMETER	Error: One of the specified paramaters is invalid.
SO_CUDA_INVALID_CONTEXT	This error indicates that there is no valid CUDA context to perform the requested operation. See the SoCudaContext class to create a CUDA context.

---

Member Function Documentation

static SoCudaError SoCuda::findFunction	(	SoCudaHandle *	handle,
		const SbString &	moduleName,
		const SbString &	functionName
	)			[static]

This function resolves a specific function in a specified module.

Parameters:

	handle	The handle which is updated with the function pointer.
	moduleName	The name of the module which is supposed to contain the function.
	functionName	The name of the function to resolve.

Returns:

An error code.

static SoCudaHandle* SoCuda::findFunction	(	const SbString &	moduleName,
		const SbString &	functionName
	)			[static]

This function resolves a specific function in a specified module.

Parameters:

	moduleName	The name of the module which is supposed to contain the function.
	functionName	The name of the function to resolve.

Returns:

A CUDA handle on the function, or NULL if the function was not found!

static void SoCuda::finish

(

)

[static]

Unload the CUDA module and free the data.

static void SoCuda::init

(

)

[static]

Initialize the CUDA module.

static SbBool SoCuda::isAvailable

(

)

[static]

This function indicates if CUDA is available on this platform.

Returns:

True if at least one CUDA enabled device is available in the system.

static bool SoCuda::isInitialized

(

)

[static]

Returns TRUE if module is currently initialized.

static SoCudaError SoCuda::loadBinaryModule	(	const void *	moduleData,
		size_t	dataSize,
		const SbString &	moduleName
	)			[static]

This function loads a cubin from the specified buffer and register it in the list of modules.

Parameters:

	moduleData	The cubin buffer.
	dataSize	The size in bytes of the cubin buffer.
	moduleName	The name of the module linked to the file.

Returns:

An error code is returned on error, 0 means no error.

static SoCudaError SoCuda::loadBinaryModule	(	const SbString &	filename,
		const SbString &	moduleName
	)			[static]

This function loads a cubin file and register it in the list of modules.

Parameters:

	filename	The cubin filename.
	moduleName	The name of the module linked to the file.

Returns:

An error code is returned on error, 0 means no error.

---

The documentation for this class was generated from the following file:

• Inventor/cuda/SoCuda.h