学习记录 --CUDA基本语法 --CUDA10.2

设备信息

void getDeviceInfo(){    
    int count;
    cudaDeviceProp prop;
    cudaGetDeviceCount(&count);

    for (int i = 0; i < count; i++) {
        cudaGetDeviceProperties(&prop, i);
        printf("显卡名称: %s \n", prop.name);
        printf("显卡版本号: %d.%d \n", prop.major,prop.minor);
        printf("全局总内存: %lld GB\n", prop.totalGlobalMem / (1000*1000*1000));
        printf("线程块最大共享内存: %lld KB \n", prop.sharedMemPerBlock / 1000 );
        printf("线程块(Block)最大线程数量: %d \n", prop.maxThreadsPerBlock );
        printf("多维线程块中每一维最大线程数量: %d %d %d \n", 
            prop.maxThreadsDim[0], prop.maxThreadsDim[1], prop.maxThreadsDim[2] );
        printf("一个Grid中每一维最大Block数量: %d %d %d \n",
            prop.maxGridSize[0], prop.maxGridSize[1], prop.maxGridSize[2]);
        printf("最大并行处理器数量数量: %d \n",
            prop.multiProcessorCount);
    }
}

申请空间

int *data_arr_on_dev;
cudaError_t cudaStatus ;

cudaStatus = cudaMalloc((void**)&data_arr_on_dev, arr_size * sizeof(int));
if (cudaStatus != cudaSuccess) {
    fprintf(stderr, "cudaMalloc failed!");
    goto err;
}

内存复制

主机向设备

cudaStatus = cudaMemcpy(data_arr_on_dev, data_arr_on_host, 
    arr_size * sizeof(arr_type),
    cudaMemcpyHostToDevice);
if (cudaStatus != cudaSuccess) {
    fprintf(stderr, "cudaMemcpy failed!");
    goto err;
}

设备向主机

cudaStatus = cudaMemcpy(data_arr_on_host, data_arr_on_dev, 
    arr_size * sizeof(arr_type),
    cudaMemcpyDeviceToHost);
if (cudaStatus != cudaSuccess) {
    fprintf(stderr, "cudaMemcpy failed!");
    goto err;
}

 核函数调用

kernel << < blocksPerGrid, threadsPerBlock >> > (param_a, param_b);

核函数定义

__global__ void kernel( int param_a, int param_b){

    __shared__ int shared_int; // 共享内存 仅可在block内共享

    threadIdx.x    // block内线程编号
    blockDim.x     // block大小

    blockIdx.x     // grid内block编号
    gridDim.x      // grid大小

    __syncthreads();    // 线程同步 同步block内的所有线程

}