I’m trying to add a production module for inferencing. The problem is that all of our build machines are running in docker (without docker windows) and so in order to produce windows builds we use x86_64-w64-mingw32-g++ which wont link against MSVC. So, I’ve been wondering if there is a way to build LibTorch w/CPU only using mingw. My trial has yielded some problems with the build so far (i.e. std::mutex not being defined) which I can try and figure out a solution for, but I wanted to know before I sink a lot of time into this, if people think that this approach is unlikely to yield results without an enormous amount of effort.
2 Likes
My advice is to give up, minGW conflicts with Libtorch, because Libtorch is compiled by MSVC, unless you can find a version of Libtorch compiled by minGW, which I hear exists