CudaHelpers.h File Reference

All CUDA helper functions and structs are declared here. More...

#include <nvneural/CudaTypes.h>
#include <nvneural/CoreHelpers.h>
#include <cassert>
#include <cuda_runtime_api.h>

Go to the source code of this file.

Macros
#define	CheckCudaDriverErrors(status_)
	CheckCudaDriverErrors is an error-handling macro. More...
#define	CheckCudartErrors(status_)
	CheckCudartErrors is an error-handling macro. More...

Functions
NeuralResult	nvneural::CheckCudaDriverErrors_Impl (CUresult status, const char file[], int line)
	Implementation function for CheckCudaDriverErrors. More...
NeuralResult	nvneural::CheckCudartErrors_Impl (cudaError_t status, const char file[], int line)
	Implementation function for CheckCudartErrors. More...

Variables
const size_t	nvneural::STANDARD_CUDA_BLOCK_SIZE = 32
	A commonly used Cuda Thread Block size. More...

Detailed Description

All CUDA helper functions and structs are declared here.

Macro Definition Documentation

CheckCudaDriverErrors

#define CheckCudaDriverErrors

(

status_

)

Value:

    do                                                                                                \
    {                                                                                                 \
        const auto localResult = nvneural::CheckCudaDriverErrors_Impl((status_), __FILE__, __LINE__); \
        if (failed(localResult))                                                                      \
        {                                                                                             \
            return localResult;                                                                       \
        }                                                                                             \
    } while (false)

CheckCudaDriverErrors is an error-handling macro.

It returns NeuralResult::Failure from the caller if the provided CUresult expression evaluates to anything but CUDA_SUCCESS.

Use this for CUDA driver API calls.

Example: CheckCudaDriverErrors(cuCtxSynchronize());

CheckCudartErrors

#define CheckCudartErrors

(

status_

)

Value:

    do                                                                                            \
    {                                                                                             \
        const auto localResult = nvneural::CheckCudartErrors_Impl((status_), __FILE__, __LINE__); \
        if (failed(localResult))                                                                  \
        {                                                                                         \
            return localResult;                                                                   \
        }                                                                                         \
    } while (false)

CheckCudartErrors is an error-handling macro.

It returns NeuralResult::Failure from the caller if the provided cudaError_t expression evaluates to anything but cudaSuccess.

Use this for CUDA runtime API calls.

Example: CheckCudartErrors(cudaDeviceSynchronize());

Function Documentation

CheckCudaDriverErrors_Impl()

NeuralResult nvneural::CheckCudaDriverErrors_Impl ( CUresult  status, const char  file[], int  line  )

inline

Implementation function for CheckCudaDriverErrors.

Parameters

status	CUDA driver result to check
file	Filename holding the check
line	Line number of the check

CheckCudartErrors_Impl()

NeuralResult nvneural::CheckCudartErrors_Impl ( cudaError_t  status, const char  file[], int  line  )

inline

Implementation function for CheckCudartErrors.

Parameters

status	CUDA runtime result to check
file	Filename holding the check
line	Line number of the check

Variable Documentation

STANDARD_CUDA_BLOCK_SIZE

const size_t nvneural::STANDARD_CUDA_BLOCK_SIZE = 32

A commonly used Cuda Thread Block size.

This block size is in no way mandatory. Users are encouraged to use whatever launch scheme makes the most sense for their kernel and target hardware.