tksharpless's profile - activity

cuda brox optical flow exchanging X and Y?? Trying out cuda::BroxOpticalFlow in OCV 3.4 with MSVC 12 (2013). Computing

OK, it seems that it fastFree() really is in opencv_cored.lib -- am I right?. The problem was badly obscured by a typically opaque & stupid MSVC configuration issue. The program was being compiled as 32 bit (by the MSVC wizard's default -- gnash --) and the libraries of course are x64. It took a bit of fooling to finally get the linker to admit that was a problem. Then it happily loaded the "missing" entries (and of course showed me a who le bunch more undefined & conflicted symbols, which were not so hard to sort out).

DUMPBIN shows that my static opencv_core300d.lib has many references to "?fastFree@cv@@YAXPEAX@Z" but does not define it. That at least shows this is not a name-mangling issue.

I guess it is pointless to ask this. I have seen lots of similar questions but NO coherent answers. It seems that nobody who knows how the OpenCV libraries are organized is willing to tell.

Building with MSVC2013. Trying to link with static libraries of OCV 3.0, that was built and installed with MSVC2013.

My program (at peresent) does nothing more than declare a Mat. Get link error

unresolved external symbol "void __cdecl cv::fastFree(void *)" (?fastFree@cv@@YAXPAX@Z) referenced in function "public: __thiscall cv::Mat::~Mat(void)" (??1Mat@cv@@QAE@XZ)

Have confirmed that linker searched the OCV core library. So either core does not export fastFree(), or does so under a different mangled name than the one the compiler generated.

Is fastFree in some other library? And if so, why, for the love of heaven?

RESOLVED by updating display driver. Did not want to do that because my machine has occasional display driver crashes, more often with newer drivers. But opencv-ocl evidently needs OpenCL support beyond what the old driver provided. So i shall just have to live with more crashes for the duration of this project -- or move it to a different machine.

My platform is a Clevo P170HM with nVidia GTX460M running Windows 7. I build a lot of open s/w with MinGW64, most recently OpenCV. CUDA is only feasible with MSVC, so I asked for OpenCL and got a build with no errors. But it won't run a simple hello world: at the first ocl call (doesn't seem to matter which) I get "invalid parameter passed to C function", apparently from nVidia's nvopencl.dll, which gdb reports as having been loaded and then unloaded during the execution of that line. Moreover test_ocl and perf_ocl report nothing but failures, mostly "C++ exception with description "C:\Users\Public\3DSoftware\opencv-2.4.7\opencv\s ources\modules\ocl\src\cl_context.cpp:551: error: (-215) currentContext != NULL in function getContext" thrown in the test body"

I did have both Intel's and AMD's OpenCL SDKs installed (wrongly, as I now suppose) when the build was configured. I have since uninstalled those and re-installed the nVidia driver (which includes their OpenCL support). Then re-run Cmake from scratch, and in modules/ocl in the build tree, make clean; make. That remade a lot of opencv, maybe all of it. But the test pgms still fail the same way, and my hello world still gets the same error.

Any suggestions?

First time build OpenCV 2.4.7 with MinGW64 on Win7. I have a working MinGW64 build of Qt 4.8 installed, Cmake found it, system PATH finds its dlls. BUT opencv test pgms crash at launch with "procedure entry point _Z17qt_message_output9QtMsgTypePKc not found in QtCore4.dll". Even though dependency walker reports that entry point is present in the QtCore4.dll that would be loaded.

Can anybody explain this, or provide a workaround?