2015年11月12日 星期四

CUDA: Query Device Information With The Runtime API

since: 2015/11/12
update: 2015/11/12

reference:
1. Professional CUDA C Programming
2. I touchs: Get Started On NVIDIA Jetson TK1
3. I touchs: How to re-flash your Jetson TK1 Development Kit 
4. I touchs: Using the Jetson TK1 as a remote development environment for CUDA

5. CUDA Runtime API :: CUDA Toolkit Documentation

A. 在 Nsight Eclipse Edition 撰寫程式如下:
     // 檔名: checkDeviceInfor.cu
#include <cuda_runtime.h>
#include <stdio.h>

/*
 * Display a variety of information on the first CUDA device in this system,
 * including driver version, runtime version, compute capability, bytes of
 * global memory, etc.
 */


void CHECK(cudaError_t call)
{
    const cudaError_t error = call;
    if (error != cudaSuccess)
    {
        fprintf(stderr, "Error: %s:%d, ", __FILE__, __LINE__);
        fprintf(stderr, "code: %d, reason: %s\n", error,
                cudaGetErrorString(error));
    }
}

int main(int argc, char **argv)
{
    printf("\n%s Starting...\n\n", argv[0]);

    int deviceCount = 0;
    cudaGetDeviceCount(&deviceCount);

    if (deviceCount == 0)
    {
        printf("There are no available device(s) that support CUDA\n");
    }
    else
    {
        printf("Detected %d CUDA Capable device(s)\n", deviceCount);
    }

    int dev = 0, driverVersion = 0, runtimeVersion = 0;
    CHECK(cudaSetDevice(dev));
    cudaDeviceProp deviceProp;
    CHECK(cudaGetDeviceProperties(&deviceProp, dev));
    printf("Device %d: \"%s\"\n\n", dev, deviceProp.name);

    cudaDriverGetVersion(&driverVersion);
    cudaRuntimeGetVersion(&runtimeVersion);

    printf("  CUDA Driver Version / Runtime Version          %d.%d / %d.%d\n",
           driverVersion / 1000, (driverVersion % 100) / 10,
           runtimeVersion / 1000, (runtimeVersion % 100) / 10);

    printf("  CUDA Capability Major/Minor version number:    %d.%d\n\n",
           deviceProp.major, deviceProp.minor);

    printf("  Total amount of global memory:                 %.2f MBytes (%llu "
           "bytes)\n", (float)deviceProp.totalGlobalMem / pow(1024.0, 3),
           (unsigned long long)deviceProp.totalGlobalMem);

    printf("  GPU Clock rate:                                %.0f MHz (%0.2f "
           "GHz)\n", deviceProp.clockRate * 1e-3f,
           deviceProp.clockRate * 1e-6f);

    printf("  Memory Clock rate:                             %.0f Mhz\n",
           deviceProp.memoryClockRate * 1e-3f);

    printf("  Memory Bus Width:                              %d-bit\n",
           deviceProp.memoryBusWidth);

    if (deviceProp.l2CacheSize)
    {
        printf("  L2 Cache Size:                                 %d bytes\n\n",
               deviceProp.l2CacheSize);
    }

    printf("  Max Texture Dimension Size (x,y,z)             1D=(%d), "
           "2D=(%d,%d), 3D=(%d,%d,%d)\n", deviceProp.maxTexture1D,
           deviceProp.maxTexture2D[0], deviceProp.maxTexture2D[1],
           deviceProp.maxTexture3D[0], deviceProp.maxTexture3D[1],
           deviceProp.maxTexture3D[2]);

    printf("  Max Layered Texture Size (dim) x layers        1D=(%d) x %d, "
           "2D=(%d,%d) x %d\n\n", deviceProp.maxTexture1DLayered[0],
           deviceProp.maxTexture1DLayered[1], deviceProp.maxTexture2DLayered[0],
           deviceProp.maxTexture2DLayered[1],
           deviceProp.maxTexture2DLayered[2]);

    printf("  Total amount of constant memory:               %lu bytes\n",
           deviceProp.totalConstMem);

    printf("  Total amount of shared memory per block:       %lu bytes\n",
           deviceProp.sharedMemPerBlock);

    printf("  Total number of registers available per block: %d\n\n",
           deviceProp.regsPerBlock);

    printf("  Warp size:                                     %d\n\n",
           deviceProp.warpSize);

    printf("  Maximum number of threads per multiprocessor:  %d\n",
           deviceProp.maxThreadsPerMultiProcessor);

    printf("  Maximum number of threads per block:           %d\n",
           deviceProp.maxThreadsPerBlock);

    printf("  Maximum sizes of each dimension of a block:    %d x %d x %d\n",
           deviceProp.maxThreadsDim[0],
           deviceProp.maxThreadsDim[1],
           deviceProp.maxThreadsDim[2]);

    printf("  Maximum sizes of each dimension of a grid:     %d x %d x %d\n",
           deviceProp.maxGridSize[0],
           deviceProp.maxGridSize[1],
           deviceProp.maxGridSize[2]);

    printf("  Maximum memory pitch:                          %lu bytes\n\n",
           deviceProp.memPitch);

    exit(EXIT_SUCCESS);
}

-----------------------------------------------------------------------------------------------

B. 執行結果:

-----------------------------------------------------------------------------------------------

C. 備註:

     還可以使用 command-line 的工具: nvidia-smi (System Management Interface),
     不過 Jetson TK1 似乎不支援. 

沒有留言:

張貼留言

注意:只有此網誌的成員可以留言。