dnn(conv1d): invalid memory access (2022-12-27)

[alalek]

Valrind report: http://pullrequest.opencv.org/buildbot/builders/master_valgrind-lin64-debug/builds/100008

Sporadic test crashes:

• http://pullrequest.opencv.org/buildbot/builders/precommit_windows64/builds/100536
• https://github.com/opencv/opencv/actions/runs/3782660057/jobs/6430574854

relates #22905

[ RUN      ] Conv3D.conv3d/1, where GetParam() = (GFLOPS=0.000, K=[3 x 3 x 3], IN={1, 2, 19, 19, 19}, OCN=2, G=2, S=[2 x 2 x 2], P=(1, 1) x (1, 1) x (1, 1), BIAS, OCV/CPU)
==18788== Invalid read of size 4
==18788==    at 0x50246F5: packData8 (fast_convolution.cpp:302)
==18788==    by 0x50246F5: operator() (fast_convolution.cpp:750)
==18788==    by 0x50246F5: std::_Function_handler<void (cv::Range const&), cv::dnn::runFastConv(cv::_InputArray const&, cv::_OutputArray const&, cv::Ptr<cv::dnn::FastConv> const&, int, cv::Ptr<cv::dnn::dnn4_v20221220::ActivationLayer> const&, std::vector<float, std::allocator<float> > const&, bool)::{lambda(cv::Range const&)#1}>::_M_invoke(std::_Any_data const&, cv::Range const&) (std_function.h:316)
==18788==    by 0x501E6F6: operator() (std_function.h:706)
==18788==    by 0x501E6F6: cv::ParallelLoopBodyLambdaWrapper::operator()(cv::Range const&) const (utility.hpp:604)
==18788==    by 0x662F91A: cv::(anonymous namespace)::ParallelLoopBodyWrapper::operator()(cv::Range const&) const (parallel.cpp:352)
==18788==    by 0x6646062: execute (parallel_impl.cpp:332)
==18788==    by 0x6646062: cv::ThreadPool::run(cv::Range const&, cv::ParallelLoopBody const&, double) (parallel_impl.cpp:647)
==18788==    by 0x66463EB: cv::parallel_for_pthreads(cv::Range const&, cv::ParallelLoopBody const&, double) (parallel_impl.cpp:750)
==18788==    by 0x662FF07: parallel_for_impl (parallel.cpp:609)
==18788==    by 0x662FF07: cv::parallel_for_(cv::Range const&, cv::ParallelLoopBody const&, double) (parallel.cpp:520)
==18788==    by 0x5022CF9: parallel_for_ (utility.hpp:612)
==18788==    by 0x5022CF9: cv::dnn::runFastConv(cv::_InputArray const&, cv::_OutputArray const&, cv::Ptr<cv::dnn::FastConv> const&, int, cv::Ptr<cv::dnn::dnn4_v20221220::ActivationLayer> const&, std::vector<float, std::allocator<float> > const&, bool) (fast_convolution.cpp:531)
==18788==    by 0x4FDC95F: cv::dnn::ConvolutionLayerImpl::forward(cv::_InputArray const&, cv::_OutputArray const&, cv::_OutputArray const&) (convolution_layer.cpp:1391)
==18788==    by 0x50F6FE1: cv::dnn::dnn4_v20221220::Net::Impl::forwardLayer(cv::dnn::dnn4_v20221220::detail::LayerData&) (net_impl.cpp:727)
==18788==    by 0x50ED137: cv::dnn::dnn4_v20221220::Net::Impl::forwardToLayer(cv::dnn::dnn4_v20221220::detail::LayerData&, bool) (net_impl.cpp:881)
==18788==    by 0x51071AD: cv::dnn::dnn4_v20221220::Net::Impl::forward(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&) (net_impl.cpp:906)
==18788==    by 0x50E8D6B: cv::dnn::dnn4_v20221220::Net::forward(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&) (net.cpp:80)
==18788==  Address 0x1ba761ac is 28 bytes after a block of size 80 in arena "client"
==18788== 

[alalek]

@zihaomu @vpisarev Please take a look on this.

There is gdb debugger available with valdind: https://valgrind.org/docs/manual/manual-core-adv.html

[zihaomu]

I'm working on it, try to fix it today.