search

ceccocats / tkDNN

Deep neural network library and toolkit to do high performace inference on NVIDIA jetson platforms
GNU General Public License v2.0
718 stars 209 forks source link

ninja ERROR #252

Closed B1ackg0d closed 3 years ago

B1ackg0d commented 3 years ago

Hello，everyone！When I used ninja -j4，I meet this problems.Would you do me a favor? ninja -j4 [1/49] Building CXX object CMakeFiles\tkDNN.dir\src\CenterTrack.cpp.obj FAILED: CMakeFiles/tkDNN.dir/src/CenterTrack.cpp.obj C:\PROGRA~2\MICROS~2\2019\COMMUN~1\VC\Tools\MSVC\1429~1.301\bin\Hostx64\x64\cl.exe /nologo /TP -DTKDNN_PATH=\"D:/tkDNN-master1/tkDNN-master\" -DtkDNN_EXPORTS -I..\include\tkDNN -I"E:\CUDA\CUDA Development\include" -IC:\opt\x64-windows\include\eigen3 -I..\include -IC:\opt\x64-windows\include /O2 /FS /EHsc -DOPENCV /O2 /Ob2 /DNDEBUG -MD /showIncludes /FoCMakeFiles\tkDNN.dir\src\CenterTrack.cpp.obj /FdCMakeFiles\tkDNN.dir\ /FS -c ..\src\CenterTrack.cpp ..\src\CenterTrack.cpp(311): warning C4838: 从“int”转换到“float”需要收缩转换 ..\src\CenterTrack.cpp(311): warning C4838: 从“int”转换到“float”需要收缩转换 ..\src\CenterTrack.cpp(472): error C2131: 表达式的计算结果不是常数 ..\src\CenterTrack.cpp(472): note: 因读取超过生命周期的变量而失败 ..\src\CenterTrack.cpp(472): note: 请参见“this”的用法 ..\src\CenterTrack.cpp(473): error C2131: 表达式的计算结果不是常数 ..\src\CenterTrack.cpp(472): note: 因读取超过生命周期的变量而失败 ..\src\CenterTrack.cpp(472): note: 请参见“this”的用法 ..\src\CenterTrack.cpp(475): error C3863: 不可指定数组类型“bool [sl]” ..\src\CenterTrack.cpp(476): error C2131: 表达式的计算结果不是常数 ..\src\CenterTrack.cpp(476): note: 因读取超过生命周期的变量而失败 ..\src\CenterTrack.cpp(476): note: 请参见“this”的用法 ..\src\CenterTrack.cpp(478): error C3863: 不可指定数组类型“bool ['函数']” ..\src\CenterTrack.cpp(481): error C3863: 不可指定数组类型“bool ['函数']” ..\src\CenterTrack.cpp(483): error C3863: 不可指定数组类型“bool [sl]” ninja: build stopped: subcommand failed.

perseusdg commented 3 years ago

Hello , there are some minor issues with the current master branch on windows,i have fixed some (centertrack is still somewhat slow on windows compared to linux and segmentation doesnt work properly) of them here https://github.com/perseusdg/tkDNN (master branch) ,running seg_demo results in a cuda illegal memory access on windows ,i will create a pull request for it once its sorted

B1ackg0d commented 3 years ago

Hi,perseusdg！Does it mean that this problem can't be solved yet? This looks like a syntax problem during compilation. In fact, I have modified a lot of source code to allow ninja to progress. Thank you for your reply！

perseusdg commented 3 years ago

If you are talking about running the segmentation network(the network has some caveats on linux too,as mentioned in the docs it runs with cuda 10.2 only) then yeah this problem cant be solved yet ,if you are talking about the performance of certain networks (like centertrack) when compared to linux,i think the issue boils down to wddm mode ,maybe someone with a card than can run in tcc mode can achieve a similar performance to running it on linux

B1ackg0d commented 3 years ago

Sorry, I still don't understand what you mean. In fact, I am just a beginner. I just want to successfully compile tkdnn, try to run the demo. But using the command ninja, it reported a lot of compilation errors (many syntax) in cpp. My msvc is 16.1, will this affect it? But I see that others can run successfully.

perseusdg commented 3 years ago

In short use the master branch on this fork https://github.com/perseusdg/tkDNN (it will run every network except the segmentation(seg_demo) ones ) to use it on windows and I am not sure about msvc 16.1 ,I tested it on 16.7,16.9 and 16.10 .If you want to stick to this repo and use it on windows ,check out the repo at this point https://github.com/ceccocats/tkDNN/tree/c306b368608893e92925bf143e7cf14f19525aeb (this works with msvc 16.7 and 16.9)

B1ackg0d commented 3 years ago

Sorry I made a mistake, my msvc is 16.11. image Is this the msvc version?Or something like this？ image

perseusdg commented 3 years ago

Yeah this version should work

B1ackg0d commented 3 years ago

Thanks for your help.The branch you posted can be used.

grandprixgp commented 3 years ago

@B1ackg0d funny running into you here :)

@perseusdg Just to clarify, I assume the newly released ShelfNet support does not function on Windows? I was just in the process of implementing this (Windows) and stumbled upon this issue when having build issues.

Do you have any idea what the timeline is looking like for proper ShelfNet on tkDNN on Windows support?

perseusdg commented 3 years ago

At the moment shelfnet doesn't work on windows and about the timeline,I am not really sure about it.