cuda c源代码-1

记录下学习cuda过程中的源代码，附有注释以便回顾学习。

#include 
using namespace std;

__global__ void add(int a, int b, int *c) {
	*c = a + b;
}

int test1(void) { // 设备指针和生成设备内存
	int c;
	int *dev_c;
	cudaMalloc((void**)&dev_c, sizeof(int));
	add << <1, 1 >> > (1, 3, dev_c);
	//printf("%d\n", *dev_c);
	cudaMemcpy(&c, dev_c, sizeof(int), cudaMemcpyDeviceToHost);
	printf("%d\n", c);
	cout << c << endl;
	cudaFree(dev_c);
	//test();
	return 0;
}

int test2() { // 测试在主机代码使用设备指针
	int *dev;
	cudaMalloc((void**)&dev, sizeof(int));
	*dev = 3; // 不能这样，虽然编译不会出现问题，但是运行过程中会中途退出
	cout << "hello" << endl;
	//cout << dev << endl;
	return 0;
}

int test3() { // 获取设备信息
	int count;
	cudaGetDeviceCount(&count); // GPU数量
	cout << "有" << count << "个GPU设备" << endl;
	cudaDeviceProp prop;
	for (int i = 0; i < count; ++i) {
		cudaGetDeviceProperties(&prop, i);
		cout << "  ---  Genenal Information of Device  --  " << endl;
		cout << "name: " << prop.name << endl;
		cout << "Compute capability: " << prop.major << "." << prop.minor << endl;
		cout << "Clock Rate: " << prop.clockRate << endl;

		cout << "\n  __  Memory Information for device __  " << endl;
		cout << "Total global memery: " << prop.totalGlobalMem << endl;
		cout << "Total constant memery: " << prop.totalConstMem << endl;// 常量内存总量，什么鬼
	}
	return 0;
}

void test4() { // 根据条件选择设备，版本号大于等于1.3
	cudaDeviceProp prop;
	int dev;
	cudaGetDevice(&dev);
	cout << "ID of current device is " << dev << endl;

	memset(&prop, 0, sizeof(cudaDeviceProp));
	prop.major = 1;
	prop.minor = 3;
	cudaChooseDevice(&dev, &prop);
	cout << "the id of choose device is " << dev << endl;
	cudaSetDevice(dev);
}

const int N = 10;

__global__ void add(int *a, int *b, int *c) {
	int tid = blockIdx.x;
	if (tid < N) {
		c[tid] = a[tid] + b[tid];
	}
}

void test5() { // cuda进行十位的矢量相加
	int a[N], b[N], c[N];
	for (int i = 0; i < 10; ++i) { // 初始化矢量
		a[i] = -i - 3;
		b[i] = i * i;
	}
	int *dev_a, *dev_b, *dev_c; // 在GPU上分配内存
	cudaMalloc((void**)&dev_a, N*sizeof(int));
	cudaMalloc((void**)&dev_b, N * sizeof(int));
	cudaMalloc((void**)&dev_c, N * sizeof(int));
	cudaMemcpy(dev_a, a, sizeof(a), cudaMemcpyHostToDevice);
	cudaMemcpy(dev_b, b, sizeof(b), cudaMemcpyHostToDevice);
	for (int i = 0; i < 100000; ++i)
		add << > > (dev_a, dev_b, dev_c); // 左线程块的数量，右一个线程块的线程数量
	cudaMemcpy(c, dev_c, sizeof(c), cudaMemcpyDeviceToHost);

	for (int i = 0; i < N; ++i) {
		cout << a[i] << " + " << b[i] << " = " << c[i] << endl;
	}

	cudaFree(dev_a);
	cudaFree(dev_b);
	cudaFree(dev_c);
}

__global__ void add_2(int *a, int *b, int *c) {
	int tid = threadIdx.x;
	if (tid < N) {
		c[tid] = a[tid] + b[tid];
	}
}

void test6() { // cuda进行十位的矢量相加 // 使用并行线程而非并行线程块实现
	int a[N], b[N], c[N];
	for (int i = 0; i < 10; ++i) { // 初始化矢量
		a[i] = -i - 3;
		b[i] = i * i;
	}
	int *dev_a, *dev_b, *dev_c; // 在GPU上分配内存
	cudaMalloc((void**)&dev_a, N * sizeof(int));
	cudaMalloc((void**)&dev_b, N * sizeof(int));
	cudaMalloc((void**)&dev_c, N * sizeof(int));
	cudaMemcpy(dev_a, a, sizeof(a), cudaMemcpyHostToDevice);
	cudaMemcpy(dev_b, b, sizeof(b), cudaMemcpyHostToDevice);
	for (int i = 0; i < 100000; ++i)
		add_2 << <1, N >> > (dev_a, dev_b, dev_c); // 左线程块的数量，右一个线程块的线程数量
	cudaMemcpy(c, dev_c, sizeof(c), cudaMemcpyDeviceToHost);

	for (int i = 0; i < N; ++i) {
		cout << a[i] << " + " << b[i] << " = " << c[i] << endl;
	}

	cudaFree(dev_a);
	cudaFree(dev_b);
	cudaFree(dev_c);
}

__global__ void add_3() {
	int bx = blockIdx.x, tx = threadIdx.x;
	int bd = blockDim.x, gd = gridDim.x; // 测试有多少个
	printf("bx = %d  tx = %d  bd = %d  gd = %d\n\n", bx, tx, bd, gd);
	//cout << "hello, world" << endl; // 核函数里面似乎不能使用cout？可以使用printf？
	//cout << "bx = " << bx << "  tx = " << tx << endl;
	//cout << "bd = " << bd << "  gd = " << gridDim.x << endl << endl;
}

void test7() {
	add_3 << <2, 5 >> > ();
}

int main() {
	test7();
}