关于随机数生成的速度测试

随机数生成有两种方法，一种是主机上生成拷贝到设备上，另一种是直接主机上调用，在设备上生成：下面就是关于这两种方法的速度测试：

int main(){
    clock_t start,finish;
    int *d_data;
    int m = 1000;
    CUDA_CALL(cudaMalloc((void **)&d_data, m*sizeof(int)));
    int rand1[1000];
    start = clock();
    
    memset(rand1,0,m*sizeof(int));
    for(int i=0;i<m;i++)
        rand1[i] = RAND::Rand(100);
    
    CUDA_CALL(cudaMemcpy(d_data,rand1,m*sizeof(int),cudaMemcpyHostToDevice));
    finish=clock();
    printf("%d--%d --- %dn",start,finish,finish-start);

    clock_t start1,finish1;
    start1=clock();
    size_t n = 1000;
    size_t i;
    curandGenerator_t gen;
    float *devData;
    CUDA_CALL(cudaMalloc((void **)&devData, n*sizeof(float)));
    CURAND_CALL(curandCreateGenerator(&gen, CURAND_RNG_PSEUDO_DEFAULT));
    CURAND_CALL(curandSetPseudoRandomGeneratorSeed(gen, 1234));
    CURAND_CALL(curandGenerateUniform(gen, devData, n));

    CURAND_CALL(curandDestroyGenerator(gen));
   
    finish1=clock();
    printf("%d --- %d",finish1,finish1-start1);
     CUDA_CALL(cudaFree(devData));
    getchar();
    return 0;
}

速度测试结果：