mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2024-11-29 05:29:54 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Merge pull request #19222 from YashasSamaga:cuda4dnn-fix-build-diagnostics

Browse Source
This commit is contained in:
Alexander Alekhin 2021-03-10 17:40:36 +00:00
commit fbb38cc245
2 changed files with 19 additions and 31 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

36
modules/dnn/src/cuda4dnn/init.hpp
View File

@ -17,28 +17,18 @@ namespace cv { namespace dnn { namespace cuda4dnn {
void checkVersions() void checkVersions()
{ {
int cudart_version = 0; // https://docs.nvidia.com/deeplearning/cudnn/developer-guide/index.html#programming-model
CUDA4DNN_CHECK_CUDA(cudaRuntimeGetVersion(&cudart_version)); // cuDNN API Compatibility
if (cudart_version != CUDART_VERSION) // Beginning in cuDNN 7, the binary compatibility of a patch and minor releases is maintained as follows:
// Any patch release x.y.z is forward or backward-compatible with applications built against another cuDNN patch release x.y.w (meaning, of the same major and minor version number, but having w!=z).
// cuDNN minor releases beginning with cuDNN 7 are binary backward-compatible with applications built against the same or earlier patch release (meaning, an application built against cuDNN 7.x is binary compatible with cuDNN library 7.y, where y>=x).
// Applications compiled with a cuDNN version 7.y are not guaranteed to work with 7.x release when y > x.
auto cudnn_bversion = cudnnGetVersion();
auto cudnn_major_bversion = cudnn_bversion / 1000, cudnn_minor_bversion = cudnn_bversion % 1000 / 100;
if (cudnn_major_bversion != CUDNN_MAJOR || cudnn_minor_bversion < CUDNN_MINOR)
{ {
std::ostringstream oss; std::ostringstream oss;
oss << "CUDART reports version " << cudart_version << " which does not match with the version " << CUDART_VERSION << " with which OpenCV was built"; oss << "cuDNN reports version " << cudnn_major_bversion << "." << cudnn_minor_bversion << " which is not compatible with the version " << CUDNN_MAJOR << "." << CUDNN_MINOR << " with which OpenCV was built";
CV_LOG_WARNING(NULL, oss.str().c_str());
}
auto cudnn_version = cudnnGetVersion();
if (cudnn_version != CUDNN_VERSION)
{
std::ostringstream oss;
oss << "cuDNN reports version " << cudnn_version << " which does not match with the version " << CUDNN_VERSION << " with which OpenCV was built";
CV_LOG_WARNING(NULL, oss.str().c_str());
}
auto cudnn_cudart_version = cudnnGetCudartVersion();
if (cudart_version != cudnn_cudart_version)
{
std::ostringstream oss;
oss << "CUDART version " << cudnn_cudart_version << " reported by cuDNN " << cudnn_version << " does not match with the version reported by CUDART " << cudart_version;
CV_LOG_WARNING(NULL, oss.str().c_str()); CV_LOG_WARNING(NULL, oss.str().c_str());
} }
} }
@ -57,9 +47,6 @@ namespace cv { namespace dnn { namespace cuda4dnn {
bool isDeviceCompatible() bool isDeviceCompatible()
{ {
if (getDeviceCount() <= 0)
return false;
int device_id = getDevice(); int device_id = getDevice();
if (device_id < 0) if (device_id < 0)
return false; return false;
@ -80,9 +67,6 @@ namespace cv { namespace dnn { namespace cuda4dnn {
bool doesDeviceSupportFP16() bool doesDeviceSupportFP16()
{ {
if (getDeviceCount() <= 0)
return false;
int device_id = getDevice(); int device_id = getDevice();
if (device_id < 0) if (device_id < 0)
return false; return false;

14
modules/dnn/src/dnn.cpp
View File

@ -239,11 +239,10 @@ private:
#endif #endif
#ifdef HAVE_CUDA #ifdef HAVE_CUDA
if (haveCUDA() && cuda4dnn::isDeviceCompatible()) if (haveCUDA())
{ {
backends.push_back(std::make_pair(DNN_BACKEND_CUDA, DNN_TARGET_CUDA)); backends.push_back(std::make_pair(DNN_BACKEND_CUDA, DNN_TARGET_CUDA));
if (cuda4dnn::doesDeviceSupportFP16()) backends.push_back(std::make_pair(DNN_BACKEND_CUDA, DNN_TARGET_CUDA_FP16));
backends.push_back(std::make_pair(DNN_BACKEND_CUDA, DNN_TARGET_CUDA_FP16));
} }
#endif #endif
} }
@ -2363,6 +2362,9 @@ struct Net::Impl : public detail::NetImplBase
CV_Assert(preferableBackend == DNN_BACKEND_CUDA); CV_Assert(preferableBackend == DNN_BACKEND_CUDA);
#ifdef HAVE_CUDA #ifdef HAVE_CUDA
if (!cudaInfo) /* we need to check only once */
cuda4dnn::checkVersions();
if (cuda4dnn::getDeviceCount() <= 0) if (cuda4dnn::getDeviceCount() <= 0)
CV_Error(Error::StsError, "No CUDA capable device found."); CV_Error(Error::StsError, "No CUDA capable device found.");
@ -2373,7 +2375,10 @@ struct Net::Impl : public detail::NetImplBase
CV_Error(Error::GpuNotSupported, "OpenCV was not built to work with the selected device. Please check CUDA_ARCH_PTX or CUDA_ARCH_BIN in your build configuration."); CV_Error(Error::GpuNotSupported, "OpenCV was not built to work with the selected device. Please check CUDA_ARCH_PTX or CUDA_ARCH_BIN in your build configuration.");
if (preferableTarget == DNN_TARGET_CUDA_FP16 && !cuda4dnn::doesDeviceSupportFP16()) if (preferableTarget == DNN_TARGET_CUDA_FP16 && !cuda4dnn::doesDeviceSupportFP16())
CV_Error(Error::StsError, "The selected CUDA device does not support FP16 operations."); {
CV_LOG_WARNING(NULL, "The selected CUDA device does not support FP16 target; switching to FP32 target.");
preferableTarget = DNN_TARGET_CUDA;
}
if (!cudaInfo) if (!cudaInfo)
{ {
@ -2384,7 +2389,6 @@ struct Net::Impl : public detail::NetImplBase
auto d2h_stream = cuda4dnn::csl::Stream(true); // stream for background D2H data transfers auto d2h_stream = cuda4dnn::csl::Stream(true); // stream for background D2H data transfers
cudaInfo = std::unique_ptr<CudaInfo_t>(new CudaInfo_t(std::move(context), std::move(d2h_stream))); cudaInfo = std::unique_ptr<CudaInfo_t>(new CudaInfo_t(std::move(context), std::move(d2h_stream)));
cuda4dnn::checkVersions();
} }
cudaInfo->workspace = cuda4dnn::csl::Workspace(); // release workspace memory if any cudaInfo->workspace = cuda4dnn::csl::Workspace(); // release workspace memory if any