mirror of
https://github.com/opencv/opencv.git
synced 2025-06-07 09:25:45 +08:00
Merge pull request #11245 from alalek:cuda_samples_drop_low_level_api
This commit is contained in:
Vadim Pisarevsky
commit
569011a121
@ -53,9 +53,6 @@ endif()
|
||||
foreach(sample_filename ${all_samples})
|
||||
ocv_define_sample(tgt ${sample_filename} gpu)
|
||||
ocv_target_link_libraries(${tgt} ${OPENCV_LINKER_LIBS} ${OPENCV_CUDA_SAMPLES_REQUIRED_DEPS})
|
||||
if(HAVE_CUDA AND NOT ANDROID)
|
||||
ocv_target_link_libraries(${tgt} ${CUDA_CUDA_LIBRARY})
|
||||
endif()
|
||||
if(HAVE_opencv_xfeatures2d)
|
||||
ocv_target_link_libraries(${tgt} opencv_xfeatures2d)
|
||||
endif()
|
||||
|
||||
@ -1,156 +0,0 @@
|
||||
/* This sample demonstrates the way you can perform independent tasks
|
||||
on the 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.hpp"
|
||||
#include "opencv2/cudaarithm.hpp"
|
||||
|
||||
#ifdef HAVE_TBB
|
||||
# include "tbb/tbb.h"
|
||||
# include "tbb/task.h"
|
||||
# undef min
|
||||
# undef max
|
||||
#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];
|
||||
|
||||
int main()
|
||||
{
|
||||
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 << "CUDA module isn't built for GPU #" << i << " ("
|
||||
<< dev_info.name() << ", CC " << dev_info.majorVersion()
|
||||
<< dev_info.minorVersion() << "\n";
|
||||
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));
|
||||
|
||||
CUcontext prev_context;
|
||||
safeCall(cuCtxPopCurrent(&prev_context));
|
||||
|
||||
// Create context for GPU #1
|
||||
safeCall(cuDeviceGet(&device, 1));
|
||||
safeCall(cuCtxCreate(&contexts[1], 0, device));
|
||||
|
||||
safeCall(cuCtxPopCurrent(&prev_context));
|
||||
|
||||
// Execute calculation in two threads using two GPUs
|
||||
int devices[] = {0, 1};
|
||||
tbb::parallel_do(devices, devices + 2, Worker());
|
||||
|
||||
destroyContexts();
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
void Worker::operator()(int device_id) const
|
||||
{
|
||||
// Set the proper context
|
||||
safeCall(cuCtxPushCurrent(contexts[device_id]));
|
||||
|
||||
Mat src(1000, 1000, CV_32F);
|
||||
Mat dst;
|
||||
|
||||
RNG rng(0);
|
||||
rng.fill(src, RNG::UNIFORM, 0, 1);
|
||||
|
||||
// CPU works
|
||||
cv::transpose(src, dst);
|
||||
|
||||
// GPU works
|
||||
GpuMat d_src(src);
|
||||
GpuMat d_dst;
|
||||
cuda::transpose(d_src, d_dst);
|
||||
|
||||
// Check results
|
||||
bool passed = cv::norm(dst - Mat(d_dst), NORM_INF) < 1e-3;
|
||||
std::cout << "GPU #" << device_id << " (" << DeviceInfo().name() << "): "
|
||||
<< (passed ? "passed" : "FAILED") << endl;
|
||||
|
||||
// Deallocate data here, otherwise deallocation will be performed
|
||||
// after context is extracted from the stack
|
||||
d_src.release();
|
||||
d_dst.release();
|
||||
|
||||
CUcontext prev_context;
|
||||
safeCall(cuCtxPopCurrent(&prev_context));
|
||||
}
|
||||
|
||||
|
||||
void destroyContexts()
|
||||
{
|
||||
safeCall(cuCtxDestroy(contexts[0]));
|
||||
safeCall(cuCtxDestroy(contexts[1]));
|
||||
}
|
||||
|
||||
#endif
|
||||
@ -1,217 +0,0 @@
|
||||
/* 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.hpp"
|
||||
#include "opencv2/highgui.hpp"
|
||||
#include "opencv2/cudastereo.hpp"
|
||||
|
||||
#ifdef HAVE_TBB
|
||||
# include "tbb/tbb.h"
|
||||
# include "tbb/task.h"
|
||||
# undef min
|
||||
# undef max
|
||||
#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 --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
|
||||
Loading…
Reference in New Issue
Block a user