ArrayFire test builds are failing
CNugteren opened this issue · comments
Somewhere between 4 months ago and one month ago something changed on the ArrayFire CI that caused all builds to be failing. Example device history: https://ci.arrayfire.org:8010/#/builders/clblast-linux-nvidia-t4 and example failed job: https://ci.arrayfire.org:8010/#/builders/39/builds/64
The error message is:
ninja: error: '/opt/intel/mkl/lib/intel64/libmkl_rt.so', needed by 'clblast_test_diagnostics', missing and no known rule to make it
This is caused by the way CMake is invoked:
cmake -G Ninja -DCMAKE_INSTALL_PREFIX:STRING=package -DCMAKE_BUILD_TYPE:STRING=Release -DBUILD_SHARED_LIBS:BOOL=ON -DCLIENTS:BOOL=OFF -DSAMPLES:BOOL=ON -DTESTS:BOOL=ON -DCBLAS_INCLUDE_DIRS:PATH=/opt/intel/mkl/include -DCBLAS_LIBRARIES:FILEPATH=/opt/intel/mkl/lib/intel64/libmkl_rt.so ..
Perhaps MKL is no longer installed, or it is in a different location?
Could someone from ArrayFire perhaps have a look? @umar456 or @9prady9? Many thanks 🙏
Sorry about that. We made a couple of dependency updates a few months ago and I think that is what is causing the issue. I can take a look later today.
Great, I see that the Linux builds now pass, e.g. https://ci.arrayfire.org:8010/#/builders/12/builds/144 👍
However the Windows builds still fail (e.g. https://ci.arrayfire.org:8010/#/builders/48/builds/26/steps/4/logs/stdio), but that seems due to changes I recently made:
C:\Users\BuildBot\buildbot\worker\clblast-windows-amd-r9\build\test\correctness\testblas.cpp(25): error C2766: explicit specialization; 'kTransposes' has already been defined
C:\Users\BuildBot\buildbot\worker\clblast-windows-amd-r9\build\test/correctness/testblas.hpp(161): note: see previous definition of 'public: static std::vector<clblast::Transpose,std::allocator<clblast::Transpose> > const clblast::TestBlas<unsigned short,unsigned short>::kTransposes'
I'll try to resolve those.
-
On one machine Windows builds now compile fine, but then all tests segfault. E.g.:
https://ci.arrayfire.org:8010/#/builders/48/builds/29/steps/5/logs/stdio
Any idea what might be going on there? -
And on another Windows machine CMake doesn't even run, it seems there is some issue with the installed compiler:
https://ci.arrayfire.org:8010/#/builders/49/builds/43
I was able to solve the segfaults in the r9 machine(its actually a Vega GPU but I need to rename the job). I still have to figure out the issue with the other build. I will look at it again soon.
Fixed the last windows builder. I am not sure why CMake was selecting the GNU version of the Intel LLVM compiler. I forced the selection of the Microsoft compiler for that build