![]() Either the CMake config needs to be amended, or the directory structure in the wheel needs to be amended to reflect what the CMake file is looking for.Įverything works just fine on Linux and MacOs (which have sane linking models and directory structures), so this is a uniquely Windows problem. The wheel contains the ".lib" files directly in the "lib" folder and ".dll" files in the "bin" folder of the wheel, but the CMake config file tries to find these in " Library/redist/intel64/vc14". The CMake config file distributed with this package references the wrong directory structure compared to the contents of the wheel. Please, add to the question post more details (code) about the way how the library 'A' finds the library 'B' and links with it. ![]() then your situation probably differs from the common one. Unfortunately, there is another problem with the "tbb-devel" Python package. 11 1 Setting CMAKEPREFIXPATH is normally a correct way for hint CMake about custom location of a library. Second, to try and circumvent the problem above on Windows, I tried to use TBB as a threading backend for MKL. So my CMake build system is a little lost. Ideally, one would distribute this as part of a "intel-openmp-devel" package, that is a dependency of "mkl-devel" My project has used GitHub Actions for CI for a long time on Linux and MacOS, and as I am trying to add Windows to our CI tests, it seems that the pre-built Windows installations from don't contain llvm-config.exe, nor any LLVMConfig.cmake as far as I can tell. ![]() The file libiomp5md.lib (link file, not the statically linked version) needs to be present to link the application, and CMake fails to find MKL when this is not present. Unfortunately, I've run into a couple of problems:įirst, on Windows the OpenMP link stub file is missing in the mkl-devel wheel, meaning that one cannot build and link a program that uses intel threading using this distribution. I've been putting together a new build pipeline for a library I've been working on, and I was hoping to use the python distributions of mkl to greatly simplify the build system - and avoid installing the whole oneAPI base package in the CI system. ![]()
0 Comments
Leave a Reply. |
Details
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |