driver_api_stereo_multi.cpp 5.16 KB
Newer Older
wester committed
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222
/* This sample demonstrates working on one piece of data using two GPUs.
   It splits input into two parts and processes them separately on different
   GPUs. */

// Disable some warnings which are caused with CUDA headers
#if defined(_MSC_VER)
#pragma warning(disable: 4201 4408 4100)
#endif

#include <iostream>
#include "opencv2/cvconfig.h"
#include "opencv2/core/core.hpp"
#include "opencv2/highgui/highgui.hpp"
#include "opencv2/cudastereo.hpp"

#ifdef HAVE_TBB
#  include "tbb/tbb_stddef.h"
#  if TBB_VERSION_MAJOR*100 + TBB_VERSION_MINOR >= 202
#    include "tbb/tbb.h"
#    include "tbb/task.h"
#    undef min
#    undef max
#  else
#    undef HAVE_TBB
#  endif
#endif

#if !defined(HAVE_CUDA) || !defined(HAVE_TBB) || defined(__arm__)

int main()
{
#if !defined(HAVE_CUDA)
    std::cout << "CUDA support is required (CMake key 'WITH_CUDA' must be true).\n";
#endif

#if !defined(HAVE_TBB)
    std::cout << "TBB support is required (CMake key 'WITH_TBB' must be true).\n";
#endif

#if defined(__arm__)
    std::cout << "Unsupported for ARM CUDA library." << std::endl;
#endif

    return 0;
}

#else

#include <cuda.h>
#include <cuda_runtime.h>

using namespace std;
using namespace cv;
using namespace cv::cuda;

struct Worker { void operator()(int device_id) const; };
void destroyContexts();

#define safeCall(expr) safeCall_(expr, #expr, __FILE__, __LINE__)
inline void safeCall_(int code, const char* expr, const char* file, int line)
{
    if (code != CUDA_SUCCESS)
    {
        std::cout << "CUDA driver API error: code " << code << ", expr " << expr
            << ", file " << file << ", line " << line << endl;
        destroyContexts();
        exit(-1);
    }
}

// Each GPU is associated with its own context
CUcontext contexts[2];

void inline contextOn(int id)
{
    safeCall(cuCtxPushCurrent(contexts[id]));
}

void inline contextOff()
{
    CUcontext prev_context;
    safeCall(cuCtxPopCurrent(&prev_context));
}

// GPUs data
GpuMat d_left[2];
GpuMat d_right[2];
Ptr<cuda::StereoBM> bm[2];
GpuMat d_result[2];

static void printHelp()
{
    std::cout << "Usage: driver_api_stereo_multi_gpu --left <left_image> --right <right_image>\n";
}

int main(int argc, char** argv)
{
    if (argc < 5)
    {
        printHelp();
        return -1;
    }

    int num_devices = getCudaEnabledDeviceCount();
    if (num_devices < 2)
    {
        std::cout << "Two or more GPUs are required\n";
        return -1;
    }

    for (int i = 0; i < num_devices; ++i)
    {
        cv::cuda::printShortCudaDeviceInfo(i);

        DeviceInfo dev_info(i);
        if (!dev_info.isCompatible())
        {
            std::cout << "GPU module isn't built for GPU #" << i << " ("
                 << dev_info.name() << ", CC " << dev_info.majorVersion()
                 << dev_info.minorVersion() << "\n";
            return -1;
        }
    }

    // Load input data
    Mat left, right;
    for (int i = 1; i < argc; ++i)
    {
        if (string(argv[i]) == "--left")
        {
            left = imread(argv[++i], cv::IMREAD_GRAYSCALE);
            CV_Assert(!left.empty());
        }
        else if (string(argv[i]) == "--right")
        {
            right = imread(argv[++i], cv::IMREAD_GRAYSCALE);
            CV_Assert(!right.empty());
        }
        else if (string(argv[i]) == "--help")
        {
            printHelp();
            return -1;
        }
    }


    // Init CUDA Driver API
    safeCall(cuInit(0));

    // Create context for GPU #0
    CUdevice device;
    safeCall(cuDeviceGet(&device, 0));
    safeCall(cuCtxCreate(&contexts[0], 0, device));
    contextOff();

    // Create context for GPU #1
    safeCall(cuDeviceGet(&device, 1));
    safeCall(cuCtxCreate(&contexts[1], 0, device));
    contextOff();

    // Split source images for processing on GPU #0
    contextOn(0);
    d_left[0].upload(left.rowRange(0, left.rows / 2));
    d_right[0].upload(right.rowRange(0, right.rows / 2));
    bm[0] = cuda::createStereoBM();
    contextOff();

    // Split source images for processing on the GPU #1
    contextOn(1);
    d_left[1].upload(left.rowRange(left.rows / 2, left.rows));
    d_right[1].upload(right.rowRange(right.rows / 2, right.rows));
    bm[1] = cuda::createStereoBM();
    contextOff();

    // Execute calculation in two threads using two GPUs
    int devices[] = {0, 1};
    tbb::parallel_do(devices, devices + 2, Worker());

    // Release the first GPU resources
    contextOn(0);
    imshow("GPU #0 result", Mat(d_result[0]));
    d_left[0].release();
    d_right[0].release();
    d_result[0].release();
    bm[0].release();
    contextOff();

    // Release the second GPU resources
    contextOn(1);
    imshow("GPU #1 result", Mat(d_result[1]));
    d_left[1].release();
    d_right[1].release();
    d_result[1].release();
    bm[1].release();
    contextOff();

    waitKey();
    destroyContexts();
    return 0;
}


void Worker::operator()(int device_id) const
{
    contextOn(device_id);

    bm[device_id]->compute(d_left[device_id], d_right[device_id], d_result[device_id]);

    std::cout << "GPU #" << device_id << " (" << DeviceInfo().name()
        << "): finished\n";

    contextOff();
}


void destroyContexts()
{
    safeCall(cuCtxDestroy(contexts[0]));
    safeCall(cuCtxDestroy(contexts[1]));
}

#endif