|
|
|
@ -7,42 +7,24 @@ |
|
|
|
|
#endif |
|
|
|
|
|
|
|
|
|
#include <iostream> |
|
|
|
|
#include "cvconfig.h" |
|
|
|
|
#include "opencv2/core/core.hpp" |
|
|
|
|
#include "opencv2/gpu/gpu.hpp" |
|
|
|
|
|
|
|
|
|
#if !defined(HAVE_CUDA) || !defined(HAVE_TBB) || defined(__arm__) |
|
|
|
|
|
|
|
|
|
#if 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> |
|
|
|
|
#include "opencv2/core/internal.hpp" // For TBB wrappers |
|
|
|
|
|
|
|
|
|
using namespace std; |
|
|
|
|
using namespace cv; |
|
|
|
|
using namespace cv::gpu; |
|
|
|
|
|
|
|
|
|
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) |
|
|
|
|
{ |
|
|
|
@ -50,65 +32,28 @@ inline void safeCall_(int code, const char* expr, const char* file, int line) |
|
|
|
|
{ |
|
|
|
|
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() |
|
|
|
|
struct Worker: public ParallelLoopBody |
|
|
|
|
{ |
|
|
|
|
int num_devices = getCudaEnabledDeviceCount(); |
|
|
|
|
if (num_devices < 2) |
|
|
|
|
Worker(int num_devices) |
|
|
|
|
{ |
|
|
|
|
std::cout << "Two or more GPUs are required\n"; |
|
|
|
|
return -1; |
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
for (int i = 0; i < num_devices; ++i) |
|
|
|
|
{ |
|
|
|
|
cv::gpu::printShortCudaDeviceInfo(i); |
|
|
|
|
|
|
|
|
|
DeviceInfo dev_info(i); |
|
|
|
|
if (!dev_info.isCompatible()) |
|
|
|
|
count = num_devices; |
|
|
|
|
contexts = new CUcontext[num_devices]; |
|
|
|
|
for (int device_id = 0; device_id < num_devices; device_id++) |
|
|
|
|
{ |
|
|
|
|
std::cout << "GPU module isn't built for GPU #" << i << " (" |
|
|
|
|
<< dev_info.name() << ", CC " << dev_info.majorVersion() |
|
|
|
|
<< dev_info.minorVersion() << "\n"; |
|
|
|
|
return -1; |
|
|
|
|
CUdevice device; |
|
|
|
|
safeCall(cuDeviceGet(&device, device_id)); |
|
|
|
|
safeCall(cuCtxCreate(&contexts[device_id], 0, device)); |
|
|
|
|
} |
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
// 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}; |
|
|
|
|
parallel_do(devices, devices + 2, Worker()); |
|
|
|
|
|
|
|
|
|
destroyContexts(); |
|
|
|
|
return 0; |
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
void Worker::operator()(int device_id) const |
|
|
|
|
{ |
|
|
|
|
virtual void operator() (const Range& range) const |
|
|
|
|
{ |
|
|
|
|
for (int device_id = range.start; device_id != range.end; ++device_id) |
|
|
|
|
{ |
|
|
|
|
// Set the proper context
|
|
|
|
|
safeCall(cuCtxPushCurrent(contexts[device_id])); |
|
|
|
|
|
|
|
|
@ -138,13 +83,56 @@ void Worker::operator()(int device_id) const |
|
|
|
|
|
|
|
|
|
CUcontext prev_context; |
|
|
|
|
safeCall(cuCtxPopCurrent(&prev_context)); |
|
|
|
|
} |
|
|
|
|
} |
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
~Worker() |
|
|
|
|
{ |
|
|
|
|
if ((contexts != NULL) && count != 0) |
|
|
|
|
{ |
|
|
|
|
for (int device_id = 0; device_id < count; device_id++) |
|
|
|
|
{ |
|
|
|
|
safeCall(cuCtxDestroy(contexts[device_id])); |
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
delete[] contexts; |
|
|
|
|
} |
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
CUcontext* contexts; |
|
|
|
|
int count; |
|
|
|
|
}; |
|
|
|
|
|
|
|
|
|
void destroyContexts() |
|
|
|
|
int main() |
|
|
|
|
{ |
|
|
|
|
safeCall(cuCtxDestroy(contexts[0])); |
|
|
|
|
safeCall(cuCtxDestroy(contexts[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::gpu::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; |
|
|
|
|
} |
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
// Init CUDA Driver API
|
|
|
|
|
safeCall(cuInit(0)); |
|
|
|
|
|
|
|
|
|
// Execute calculation
|
|
|
|
|
parallel_for_(cv::Range(0, num_devices), Worker(num_devices)); |
|
|
|
|
|
|
|
|
|
return 0; |
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
#endif |
|
|
|
|