Skip to content

Lazy loading nvcuda.dll #2215

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Aug 14, 2019
Merged

Lazy loading nvcuda.dll #2215

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
160 changes: 120 additions & 40 deletions modules/cudaoptflow/src/nvidiaOpticalFlow.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,11 +29,14 @@ CV_Error(cv::Error::HeaderIsNull, "Nvidia Optical Flow headers not found. Make s

//macro for dll loading
#if defined(_WIN64)
#define MODULENAME TEXT("nvofapi64.dll")
#define OF_MODULENAME TEXT("nvofapi64.dll")
#define CUDA_MODULENAME TEXT("nvcuda.dll")
#elif defined(_WIN32)
#define MODULENAME TEXT("nvofapi.dll")
#define OF_MODULENAME TEXT("nvofapi.dll")
#define CUDA_MODULENAME TEXT("nvcuda.dll")
#else
#define MODULENAME "libnvidia-opticalflow.so.1"
#define OF_MODULENAME "libnvidia-opticalflow.so.1"
#define CUDA_MODULENAME "libcuda.so"
#endif

#define NVOF_API_CALL(nvOFAPI) \
Expand Down Expand Up @@ -112,6 +115,114 @@ using namespace cv::cuda;

namespace
{
class LoadNvidiaModules
{
private:
typedef int(*PFNCudaCuCtxGetCurrent)(CUcontext*);
typedef NV_OF_STATUS(NVOFAPI *PFNNvOFAPICreateInstanceCuda)
(uint32_t apiVer, NV_OF_CUDA_API_FUNCTION_LIST* cudaOf);

PFNCudaCuCtxGetCurrent m_cudaDriverAPIGetCurrentCtx;
PFNNvOFAPICreateInstanceCuda m_NvOFAPICreateInstanceCuda;
HMODULE m_hOFModule;
HMODULE m_hCudaModule;
bool m_isFailed;

LoadNvidiaModules() :
m_cudaDriverAPIGetCurrentCtx(NULL),
m_NvOFAPICreateInstanceCuda(NULL),
m_isFailed(false)
{
//Loading Cuda Library
#if defined(_WIN32) || defined(_WIN64)
HMODULE hCudaModule = LoadLibrary(CUDA_MODULENAME);
#else
void *hCudaModule = dlopen(CUDA_MODULENAME, RTLD_LAZY);
#endif

if (hCudaModule == NULL)
{
m_isFailed = true;
CV_Error(Error::StsBadFunc, "Cannot find Cuda library.");
}
m_hCudaModule = hCudaModule;

#if defined(_WIN32)
m_cudaDriverAPIGetCurrentCtx = (PFNCudaCuCtxGetCurrent)GetProcAddress(m_hCudaModule, "cuCtxGetCurrent");
#else
m_cudaDriverAPIGetCurrentCtx = (PFNCudaCuCtxGetCurrent)dlsym(m_hCudaModule, "cuCtxGetCurrent");
#endif
if (!m_cudaDriverAPIGetCurrentCtx)
{
m_isFailed = true;
CV_Error(Error::StsBadFunc,
"Cannot find Cuda Driver API : cuCtxGetCurrent() entry in Cuda library");
}

//Loading Optical Flow Library
#if defined(_WIN32) || defined(_WIN64)
HMODULE hOFModule = LoadLibrary(OF_MODULENAME);
#else
void *hOFModule = dlopen(OF_MODULENAME, RTLD_LAZY);
#endif

if (hOFModule == NULL)
{
m_isFailed = true;
CV_Error(Error::StsBadFunc, "Cannot find NvOF library.");
}
m_hOFModule = hOFModule;

#if defined(_WIN32)
m_NvOFAPICreateInstanceCuda = (PFNNvOFAPICreateInstanceCuda)GetProcAddress(m_hOFModule, "NvOFAPICreateInstanceCuda");
#else
m_NvOFAPICreateInstanceCuda = (PFNNvOFAPICreateInstanceCuda)dlsym(m_hOFModule, "NvOFAPICreateInstanceCuda");
#endif
if (!m_NvOFAPICreateInstanceCuda)
{
m_isFailed = true;
CV_Error(Error::StsBadFunc,
"Cannot find NvOFAPICreateInstanceCuda() entry in NVOF library");
}
};

~LoadNvidiaModules()
{
if (NULL != m_hCudaModule)
{
#if defined(_WIN32) || defined(_WIN64)
FreeLibrary(m_hCudaModule);
#else
dlclose(m_hCudaModule);
#endif
}
if (NULL != m_hOFModule)
{
#if defined(_WIN32) || defined(_WIN64)
FreeLibrary(m_hOFModule);
#else
dlclose(m_hOFModule);
#endif
}
m_hCudaModule = NULL;
m_hOFModule = NULL;
m_cudaDriverAPIGetCurrentCtx = NULL;
m_NvOFAPICreateInstanceCuda = NULL;
}

public:
static LoadNvidiaModules& Init()
{
static LoadNvidiaModules LoadLibraryObj;
if (LoadLibraryObj.m_isFailed)
CV_Error(Error::StsError, "Can't initialize LoadNvidiaModules Class Object");
return LoadLibraryObj;
}

PFNCudaCuCtxGetCurrent GetCudaLibraryFunctionPtr() { return m_cudaDriverAPIGetCurrentCtx; }
PFNNvOFAPICreateInstanceCuda GetOFLibraryFunctionPtr() { return m_NvOFAPICreateInstanceCuda; }
};

class NvidiaOpticalFlowImpl : public cv::cuda::NvidiaOpticalFlow_1_0
{
private:
Expand Down Expand Up @@ -169,7 +280,6 @@ class NvidiaOpticalFlowImpl : public cv::cuda::NvidiaOpticalFlow_1_0
NvOFHandle GetHandle() { return m_hOF; }

protected:
HMODULE m_hModule; //module handle to load nvof dll
std::mutex m_lock;

public:
Expand Down Expand Up @@ -198,6 +308,8 @@ NvidiaOpticalFlowImpl::NvidiaOpticalFlowImpl(
m_cuContext(nullptr), m_format(NV_OF_BUFFER_FORMAT_GRAYSCALE8),
m_gridSize(NV_OF_OUTPUT_VECTOR_GRID_SIZE_4)
{
LoadNvidiaModules& LoadNvidiaModulesObj = LoadNvidiaModules::Init();

int nGpu = 0;

cuSafeCall(cudaGetDeviceCount(&nGpu));
Expand All @@ -208,7 +320,8 @@ NvidiaOpticalFlowImpl::NvidiaOpticalFlowImpl(

cuSafeCall(cudaSetDevice(m_gpuId));
cuSafeCall(cudaFree(m_cuContext));
cuSafeCall(cuCtxGetCurrent(&m_cuContext));

cuSafeCall(LoadNvidiaModulesObj.GetCudaLibraryFunctionPtr()(&m_cuContext));

if (m_gridSize != NV_OF_OUTPUT_VECTOR_GRID_SIZE_4)
{
Expand Down Expand Up @@ -253,38 +366,9 @@ NvidiaOpticalFlowImpl::NvidiaOpticalFlowImpl(
m_costBufElementSize = sizeof(uint32_t);
}

#if defined(_WIN32) || defined(_WIN64)
HMODULE hModule = LoadLibrary(MODULENAME);
#else
void *hModule = dlopen(MODULENAME, RTLD_LAZY);
#endif

if (hModule == NULL)
{
CV_Error(Error::StsBadFunc,
"Cannot find NvOF library.");
}
m_hModule = hModule;

typedef NV_OF_STATUS(NVOFAPI *PFNNvOFAPICreateInstanceCuda)
(uint32_t apiVer, NV_OF_CUDA_API_FUNCTION_LIST* cudaOf);

#if defined(_WIN32)
PFNNvOFAPICreateInstanceCuda NvOFAPICreateInstanceCuda
= (PFNNvOFAPICreateInstanceCuda)GetProcAddress(m_hModule, "NvOFAPICreateInstanceCuda");
#else
PFNNvOFAPICreateInstanceCuda NvOFAPICreateInstanceCuda
= (PFNNvOFAPICreateInstanceCuda)dlsym(m_hModule, "NvOFAPICreateInstanceCuda");
#endif
if (!NvOFAPICreateInstanceCuda)
{
CV_Error(Error::StsBadFunc,
"Cannot find NvOFAPICreateInstanceCuda() entry in NVOF library");
}

m_ofAPI.reset(new NV_OF_CUDA_API_FUNCTION_LIST());

NVOF_API_CALL(NvOFAPICreateInstanceCuda(NV_OF_API_VERSION, m_ofAPI.get()));
NVOF_API_CALL(LoadNvidiaModulesObj.GetOFLibraryFunctionPtr()(NV_OF_API_VERSION, m_ofAPI.get()));
NVOF_API_CALL(GetAPI()->nvCreateOpticalFlowCuda(m_cuContext, &m_hOF));

memset(&m_initParams, 0, sizeof(m_initParams));
Expand Down Expand Up @@ -416,9 +500,7 @@ void NvidiaOpticalFlowImpl::calc(InputArray _frame0, InputArray _frame1, InputOu
}
}

cuSafeCall(cuCtxPushCurrent(m_cuContext));
inputStream.waitForCompletion();
cuSafeCall(cuCtxPopCurrent(&m_cuContext));

//Execute Call
NV_OF_EXECUTE_INPUT_PARAMS exeInParams;
Expand All @@ -436,9 +518,7 @@ void NvidiaOpticalFlowImpl::calc(InputArray _frame0, InputArray _frame1, InputOu
m_hCostBuffer : nullptr;;
NVOF_API_CALL(GetAPI()->nvOFExecute(GetHandle(), &exeInParams, &exeOutParams));

cuSafeCall(cuCtxPushCurrent(m_cuContext));
outputStream.waitForCompletion();
cuSafeCall(cuCtxPopCurrent(&m_cuContext));

if (_flow.isMat())
flowXYGpuMat.download(_flow);
Expand All @@ -460,7 +540,7 @@ void NvidiaOpticalFlowImpl::calc(InputArray _frame0, InputArray _frame1, InputOu
else
CV_Error(Error::StsBadArg, "Incorrect cost buffer passed. Pass Mat or GpuMat");
}
cuSafeCall(cuCtxSynchronize());
cuSafeCall(cudaDeviceSynchronize());
}

void NvidiaOpticalFlowImpl::collectGarbage()
Expand Down