MNN icon indicating copy to clipboard operation
MNN copied to clipboard

在国产芯片兆芯RX7000上, 运行时crash

Open tianzunmengyan opened this issue 1 month ago • 3 comments

求助如何解决, 详细信息如下

代码版本: 最新3.3.0

构建命令: cmake .. -DCMAKE_BUILD_TYPE=Release -DMNN_BUILD_SHARED_LIBS=ON -DMNN_SEP_BUILD=OFF -DMNN_USE_SYSTEM_LIB=ON -DMNN_BUILD_OPENCV=ON -DMNN_OPENCL=ON -DMNN_BUILD_DEMO=ON -DMNN_BUILD_BENCHMARK=ON

测试命令: ./benchmark.out ../benchmark/models/ 1 1 3

crash堆栈: (gdb) bt full #0 0x00007fffedfd40a4 in ?? () from /lib/x86_64-linux-gnu/cx4oclcompiler.so No symbol table info available. #1 0x00007fffedfa69c9 in ?? () from /lib/x86_64-linux-gnu/cx4oclcompiler.so No symbol table info available. #2 0x00007fffee4b157e in ?? () from /lib/x86_64-linux-gnu/cx4oclcompiler.so No symbol table info available. #3 0x00007fffee4b5794 in ?? () from /lib/x86_64-linux-gnu/cx4oclcompiler.so No symbol table info available. #4 0x00007fffee4c0543 in ?? () from /lib/x86_64-linux-gnu/cx4oclcompiler.so No symbol table info available. #5 0x00007fffee4cd065 in ?? () from /lib/x86_64-linux-gnu/cx4oclcompiler.so No symbol table info available. #6 0x00007fffee4ce29e in ?? () from /lib/x86_64-linux-gnu/cx4oclcompiler.so No symbol table info available. #7 0x00007fffee4d2726 in ?? () from /lib/x86_64-linux-gnu/cx4oclcompiler.so No symbol table info available. #8 0x00007fffee4d7fed in ?? () from /lib/x86_64-linux-gnu/cx4oclcompiler.so No symbol table info available. #9 0x00007fffee4d83ef in ?? () from /lib/x86_64-linux-gnu/cx4oclcompiler.so No symbol table info available. #10 0x00007fffedd53cee in ?? () from /lib/x86_64-linux-gnu/cx4oclcompiler.so No symbol table info available. #11 0x00007fffedd53f82 in ?? () from /lib/x86_64-linux-gnu/cx4oclcompiler.so No symbol table info available. #12 0x00007fffedd54178 in ?? () from /lib/x86_64-linux-gnu/cx4oclcompiler.so No symbol table info available. #13 0x00007fffedb35327 in ?? () from /lib/x86_64-linux-gnu/cx4oclcompiler.so No symbol table info available. #14 0x00007fffedb375c2 in __CLC_build_program_with_source_string () from /lib/x86_64-linux-gnu/cx4oclcompiler.so No symbol table info available. #15 0x00007ffff714df28 in ?? () from /lib/x86_64-linux-gnu/cx4_driocl.so No symbol table info available. #16 0x00007ffff7ce7397 in MNN::OpenCLRuntime::buildProgram(std::__cxx11::basic_string<char, std::char_traits, std::allocator > const&, cl::Program*) () from /home/maxhub/software/mnn/build/libMNN.so No symbol table info available. #17 0x00007ffff7ce9fd6 in MNN::OpenCLRuntime::buildKernelWithCache(std::__cxx11::basic_string<char, std::char_traits, std::allocator > const&, std::__cxx11::basic_string<char, std::char_traits, std::allocator > const&, std::set<std::__cxx11::basic_string<char, std::char_traits, std::allocator >, std::less<std::__cxx11::basic_string<char, std::char_traits, std::allocator > >, std::allocator<std::__cxx11::basic_string<char, std::char_traits, std::allocator > > > const&, int, MNN::Tensor const*, MNN::Tensor const*, bool) () from /home/maxhub/software/mnn/build/libMNN.so No symbol table info available. #18 0x00007ffff7cd343e in MNN::OpenCL::copyBufferToImage(MNN::OpenCLRuntime*, cl::Buffer const&, cl::Image const&, int, int, int) () from /home/maxhub/software/mnn/build/libMNN.so No symbol table info available. #19 0x00007ffff7d65ae0 in MNN::OpenCL::ConvCommonExecution::ConvCommonExecution(MNN::Op const*, MNN::Backend*, bool) () from /home/maxhub/software/mnn/build/libMNN.so No symbol table info available. #20 0x00007ffff7d66886 in MNN::OpenCL::ConvExecution::ConvExecution(std::vector<MNN::Tensor*, std::allocatorMNN::Tensor* > const&, std::vector<MNN::Tensor*, std::allocatorMNN::Tensor* > const&, MNN::Op const*, MNN::Backend*, bool) () from /home/maxhub/software/mnn/build/libMNN.so No symbol table info available. #21 0x00007ffff7d68d6c in MNN::OpenCL::ConvolutionCreator::onCreate(std::vector<MNN::Tensor*, std::allocatorMNN::Tensor* > const&, std::vector<MNN::Tensor*, std::allocatorMNN::Tensor* > const&, MNN::Op const*, MNN::Backend*) const () from /home/maxhub/software/mnn/build/libMNN.so No symbol table info available. #22 0x00007ffff7cc264b in MNN::OpenCL::OpenCLBackend::onCreate(std::vector<MNN::Tensor*, std::allocatorMNN::Tensor* > const&, std::vector<MNN::Tensor*, std::allocatorMNN::Tensor* > const&, MNN::Op const*) () from /home/maxhub/software/mnn/build/libMNN.so No symbol table info available. #23 0x00007ffff7a88acd in MNN::OpCommonUtils::createExecutionWithExternal(MNN::Backend*, std::vector<MNN::Tensor*, std::allocatorMNN::Tensor* > const&, std::vector<MNN::Tensor*, std::allocatorMNN::Tensor* > const&, MNN::Op const*, MNN::FileLoader*, std::shared_ptrMNN::BufferStorage&) () from /home/maxhub/software/mnn/build/libMNN.so No symbol table info available. #24 0x00007ffff7a96e2b in MNN::Pipeline::allocMemory(bool, bool) () from /home/maxhub/software/mnn/build/libMNN.so No symbol table info available. #25 0x00007ffff7a9f8d7 in MNN::Session::resize() () from /home/maxhub/software/mnn/build/libMNN.so No symbol table info available. #26 0x00007ffff7a6aef2 in MNN::Interpreter::createMultiPathSession(std::vector<MNN::ScheduleConfig, std::allocatorMNN::ScheduleConfig > const&, std::pair<std::map<MNNForwardType, std::shared_ptrMNN::Runtime, std::less<MNNForwardType>, std::allocator<std::pair<MNNForwardType const, std::shared_ptrMNN::Runtime > > >, std::shared_ptrMNN::Runtime > const&) () from /home/maxhub/software/mnn/build/libMNN.so No symbol table info available. #27 0x00007ffff7a6b007 in MNN::Interpreter::createMultiPathSession(std::vector<MNN::ScheduleConfig, std::allocatorMNN::ScheduleConfig > const&) () from /home/maxhub/software/mnn/build/libMNN.so No symbol table info available. #28 0x00007ffff7a6bb14 in MNN::Interpreter::createSession(MNN::ScheduleConfig const&) () from /home/maxhub/software/mnn/build/libMNN.so No symbol table info available. #29 0x0000000000404f0a in doBench(Model&, int, int, int, bool, int, int, float, int, bool, bool) [clone .constprop.104] () No symbol table info available. #30 0x000000000040392e in main ()

tianzunmengyan avatar Nov 19 '25 13:11 tianzunmengyan

可以用CPU后端测试一下

wangzhaode avatar Nov 20 '25 01:11 wangzhaode

Compatibility is very poor.

GreenShadows avatar Nov 20 '25 19:11 GreenShadows

可以用CPU后端测试一下

用cpu后端试过, 可以正常运行完毕benchmark所有测试

tianzunmengyan avatar Nov 21 '25 02:11 tianzunmengyan