a1.cpp File Reference

#include <iostream>
#include <cuda_runtime.h>
#include <device_launch_parameters.h>
#include <cuda_fp16.h>

Classes
class	Matrix< T >
	Base MAT class for GPU memory with reference counting. More...

Macros
#define	uchar   unsigned char
#define	MAT_8   1
#define	MAT_8U   3
#define	MAT_16   2
#define	MAT_16I   5
#define	MAT_32   4
#define	MAT_64   8
#define	BIG_LIMIT   1000000
#define	CUDA_CHECK(call)
	检查CUDA函数的错误
#define	CUFFT_CHECK(call)
	检查cuFFT函数的错误

Functions
template<typename T >
__global__ void	matrixAddKernel (T *a, T *b, T *result, size_t rows, size_t cols)
	Add Kernel.
template<typename T >
__global__ void	matrixSubtractKernel (T *a, T *b, T *result, size_t rows, size_t cols)
	Sub Kernel.
template<typename T >
__global__ void	matrixMulKernel (T *a, T *b, T *result, size_t aRows, size_t aCols, size_t bCols)
	mul Kernel be faster! https://github.com/njuhope/cuda_sgemm/blob/master/gemm.cu
template<typename T >
__global__ void	LUDecomposition (T *A, int n)
	LU Kernel be faster! https://github.com/njuhope/cuda_sgemm/blob/master/gemm.cu.

Macro Definition Documentation

BIG_LIMIT

#define BIG_LIMIT   1000000

CUDA_CHECK

#define CUDA_CHECK

(

call

)

Value:

do { \
    cudaError_t status = call; \
    if (status != cudaSuccess) { \
        fprintf(stderr, "CUDA error in file '%s' in line %i: %s\n", \
                __FILE__, __LINE__, cudaGetErrorString(status)); \
        exit(EXIT_FAILURE); \
    } \
} while(0)

检查CUDA函数的错误

Parameters

call

CUFFT_CHECK

#define CUFFT_CHECK

(

call

)

Value:

do { \
    cufftResult_t status = call; \
    if (status != CUFFT_SUCCESS) { \
        fprintf(stderr, "cuFFT error in file '%s' in line %i\n", \
                __FILE__, __LINE__); \
        exit(EXIT_FAILURE); \
    } \
} while(0)

检查cuFFT函数的错误

Parameters

call

MAT_16

#define MAT_16   2

MAT_16I

#define MAT_16I   5

MAT_32

#define MAT_32   4

MAT_64

#define MAT_64   8

MAT_8

#define MAT_8   1

MAT_8U

#define MAT_8U   3

uchar

#define uchar   unsigned char

Function Documentation

LUDecomposition()

template<typename T >

__global__ void LUDecomposition	(	T *	A,
		int	n )

LU Kernel be faster! https://github.com/njuhope/cuda_sgemm/blob/master/gemm.cu.

Template Parameters

T

Parameters

array1
n

Returns

global

matrixAddKernel()

template<typename T >

__global__ void matrixAddKernel	(	T *	a,
		T *	b,
		T *	result,
		size_t	rows,
		size_t	cols )

Add Kernel.

Template Parameters

T

Parameters

array1
array2
result_array
rows
cols

Returns

global

matrixMulKernel()

template<typename T >

__global__ void matrixMulKernel	(	T *	a,
		T *	b,
		T *	result,
		size_t	aRows,
		size_t	aCols,
		size_t	bCols )

mul Kernel be faster! https://github.com/njuhope/cuda_sgemm/blob/master/gemm.cu

Template Parameters

T

Parameters

array1
array2
result_array
rows
cols

Returns

global

matrixSubtractKernel()

template<typename T >

__global__ void matrixSubtractKernel	(	T *	a,
		T *	b,
		T *	result,
		size_t	rows,
		size_t	cols )

Sub Kernel.

Template Parameters

T

Parameters

array1
array2
result_array
rows
cols

Returns

global