Skip to content

Instantly share code, notes, and snippets.

@delijati

delijati/hw.cpp

Created November 22, 2022 12:21
Show Gist options

  • Select an option

    Learn more about clone URLs
  • Save delijati/200b1b8edfee4227da560b83e0d1c563 to your computer and use it in GitHub Desktop.

Select an option

Learn more about clone URLs
Save delijati/200b1b8edfee4227da560b83e0d1c563 to your computer and use it in GitHub Desktop.
Download ZIP

Revisions

  1. delijati created this gist Nov 22, 2022.
    112 changes: 112 additions & 0 deletions hw.cpp
    View file
    Original file line number Diff line number Diff line change
    @@ -0,0 +1,112 @@
    #include <iostream>
    #include <cuda_runtime.h>
    #include <cudnn.h>


    /**
    * Minimal example to apply sigmoid activation on a tensor
    * using cuDNN.
    **/
    int main(int argc, char** argv)
    {
    int numGPUs;
    int driverVersion = 0, runtimeVersion = 0;
    cudaGetDeviceCount(&numGPUs);
    std::cout << "Found " << numGPUs << " GPUs." << std::endl;
    cudaSetDevice(0); // use GPU0
    int device;
    struct cudaDeviceProp devProp;
    cudaGetDevice(&device);
    cudaGetDeviceProperties(&devProp, device);
    cudaDriverGetVersion(&driverVersion);
    cudaRuntimeGetVersion(&runtimeVersion);

    std::cout << "Device: " << devProp.name << std::endl;
    std::cout << "Driver Version: " << driverVersion<<"\n";
    std::cout << "Runtime Version: " << runtimeVersion<<"\n";
    std::cout << "Compute capability:" << devProp.major << "." << devProp.minor << std::endl;
    std::cout << "Total amount of global memory: "<<(unsigned long long)devProp.totalGlobalMem<<" bytes\n";
    std::cout << "Total amount of constant memory: "<<devProp.totalConstMem<<"bytes\n";
    std::cout << "Total amount of shared memory per block: "<<devProp.sharedMemPerBlock<<" bytes\n";
    std::cout << "Total number of registers available per block: "<<devProp.regsPerBlock<<"\n";
    std::cout << "Warp size: "<<devProp.warpSize<<"\n";

    cudnnHandle_t handle_;
    cudnnCreate(&handle_);
    std::cout << "Created cuDNN handle" << std::endl;

    // create the tensor descriptor
    cudnnDataType_t dtype = CUDNN_DATA_FLOAT;
    cudnnTensorFormat_t format = CUDNN_TENSOR_NCHW;
    int n = 1, c = 1, h = 1, w = 10;
    int NUM_ELEMENTS = n*c*h*w;
    cudnnTensorDescriptor_t x_desc;
    cudnnCreateTensorDescriptor(&x_desc);
    cudnnSetTensor4dDescriptor(x_desc, format, dtype, n, c, h, w);

    // create the tensor
    float *x;
    cudaMallocManaged(&x, NUM_ELEMENTS * sizeof(float));
    for(int i=0;i<NUM_ELEMENTS;i++) x[i] = i * 1.00f;
    std::cout << "Original array: ";
    for(int i=0;i<NUM_ELEMENTS;i++) std::cout << x[i] << " ";

    // create activation function descriptor
    float alpha[1] = {1};
    float beta[1] = {0.0};
    cudnnActivationDescriptor_t sigmoid_activation;
    cudnnActivationMode_t mode = CUDNN_ACTIVATION_SIGMOID;
    cudnnNanPropagation_t prop = CUDNN_NOT_PROPAGATE_NAN;
    cudnnCreateActivationDescriptor(&sigmoid_activation);
    cudnnSetActivationDescriptor(sigmoid_activation, mode, prop, 0.0f);

    cudnnActivationForward(
    handle_,
    sigmoid_activation,
    alpha,
    x_desc,
    x,
    beta,
    x_desc,
    x
    );

    cudnnDestroy(handle_);
    std::cout << std::endl << "Destroyed cuDNN handle." << std::endl;
    std::cout << "New array: ";
    for(int i=0;i<NUM_ELEMENTS;i++) std::cout << x[i] << " ";
    std::cout << std::endl;
    cudaFree(x);
    return 0;
    }
    /*
    Info:
    $ lsb_release -a
    LSB Version: n/a
    Distributor ID: ManjaroLinux
    Description: Manjaro Linux
    Release: 22.0.0
    Codename: Sikaris
    $ uname -a
    Linux papagayo 5.15.78-1-MANJARO #1 SMP PREEMPT Thu Nov 10 20:50:09 UTC 2022 x86_64 GNU/Linux
    $ nvidia-smi -L
    GPU 0: NVIDIA GeForce MX450
    Build:
    $ g++ -I/opt/cuda/include -I/opt/cuda/targets/ppc64le-linux/include -o hw.o -c hw.cpp
    $ nvcc -ccbin g++ -m64 -gencode arch=compute_80,code=sm_80 -o hw hw.o -I/opt/cuda/include -I/opt/cuda/targets/ppc64le-linux/include -L/opt/cuda/lib64 -L/opt/cuda/targets/ppc64le-linux/lib -lcublasLt -lcudart -lcublas -lcudnn -lstdc++ -lm
    $ ./hw
    Found 1 GPUs.
    Device: NVIDIA GeForce MX450
    Driver Version: 11080
    Runtime Version: 11080
    Compute capability:7.5
    Total amount of global memory: 1969815552 bytes
    Total amount of constant memory: 65536bytes
    Total amount of shared memory per block: 49152 bytes
    Total number of registers available per block: 65536
    Warp size: 32
    Created cuDNN handle
    Original array: 0 1 2 3 4 5 6 7 8 9
    Destroyed cuDNN handle.
    New array: 0.5 0.731059 0.880797 0.952574 0.982014 0.993307 0.997527 0.999089 0.999665 0.999877
    */