cuda-流

Data.h

#pragma once
#include "cuda_runtime.h"
#include "device_launch_parameters.h"
class Data
{
public:
	Data();
	~Data();

	float * h_a;
	float * h_b;
	float * h_c;
	float * h_d;

	float * d_a;
	float * d_b;
	float * d_c;
	float * d_d;

	int size = 10;	/** 变量长度 */

	//分配资源
	void allocateResource();

	//初始化相关参数
	void init();

	//内存拷贝
	void memcpyHost2Device();

	//内存拷贝
	void memcpyDevice2Host();

	//释放资源
	void releaseResource();
};

data.cpp

#include "Data.h"
#include <iostream>
using namespace std;


Data::Data()
{
}


Data::~Data()
{
}

void Data::allocateResource()
{
	h_a = (float*)malloc(sizeof(float)*size);
	h_b = (float*)malloc(sizeof(float)*size);
	h_c = (float*)malloc(sizeof(float)*size);
	h_d = (float*)malloc(sizeof(float)*size);

	cudaMalloc(&d_a, sizeof(float)*size);
	cudaMalloc(&d_b, sizeof(float)*size);
	cudaMalloc(&d_c, sizeof(float)*size);
	cudaMalloc(&d_d, sizeof(float)*size);
}

void Data::init()
{
	for (int i = 0; i < size; i++)
	{
		h_a[i] = i;		h_b[i] = i;		h_c[i] = i;		h_d[i] = i;
	}
}

void Data::memcpyHost2Device()
{
	cudaMemcpy(d_a, h_a, sizeof(float)*size, cudaMemcpyHostToDevice);
	cudaMemcpy(d_b, h_b, sizeof(float)*size, cudaMemcpyHostToDevice);
	cudaMemcpy(d_c, h_c, sizeof(float)*size, cudaMemcpyHostToDevice);
	cudaMemcpy(d_d, h_d, sizeof(float)*size, cudaMemcpyHostToDevice);
}

void Data::memcpyDevice2Host()
{
	cudaMemcpy(h_a,d_a, sizeof(float)*size, cudaMemcpyDeviceToHost);
	cudaMemcpy(h_b,d_b, sizeof(float)*size, cudaMemcpyDeviceToHost);
	cudaMemcpy(h_c,d_c, sizeof(float)*size, cudaMemcpyDeviceToHost);
	cudaMemcpy(h_d,d_d, sizeof(float)*size, cudaMemcpyDeviceToHost);
}


void Data::releaseResource()
{
	cudaFree(d_a);
	cudaFree(d_b);
	cudaFree(d_c);
	cudaFree(d_d);
	free(h_a);
	free(h_b);
	free(h_c);
	free(h_d);
}

kernel.cu

#include "cuda_runtime.h"
#include "device_launch_parameters.h"

#include <stdio.h>
#include <iostream>

#include "Data.h"


__global__ void multiply(float c[],float d[])
{
	printf("执行Step1\n");
}

__global__ void add(float a[],float b[])
{
	printf("执行Step2\n");
}

void main()
{
	//创建流
	cudaStream_t stream1;
	cudaStreamCreate(&stream1);

	//初始化设备
	Data data;
	data.allocateResource();
	data.init();
	
	add << <1, data.size,0,stream1>> > (data.d_a, data.d_b);				//控制先执行
	multiply << <1, data.size, 0, stream1 >> > (data.d_c, data.d_d);		//控制后执行
	
	cudaStreamSynchronize(stream1);

	data.releaseResource();

	cudaStreamDestroy(stream1);

	cudaDeviceReset();
	
	system("pause");
    return;
}

输出结果，add会在multiply之前执行。

参考《GPU高性能编程CUDA实战》

补充：Data类可以作为模板使用