Ok, this is all very exciting. Added -fexceptions
to the make/local file, but not really sure if that was what sorted it out. But if it works, it works! ¯_(ツ)_/¯
Overall CPU-load is around 25% (have 6 physical cores, 12 hyperthreaded). Time to start reading the manual on threading and try bigger models! Thanks for your help @ahartikainen.
I think the current Windows instructions are fundamentally flawed and will not work until threading bug in winpthread
/mingw
is fixed. (see my earlier message)
I think we should remove the current instructions from wiki?
Any ideas?
Sounds sensible to me. If Ubuntu is the best current solution, that would save end-users like me some time and frustrastion.
I did try with clang (mingw-w64 headers and msvc) and with mingw-w64 headers I can build CmdStan, but it fail at compiling thread_local parts.
msvc headers fail with make somehow.
It seems that “my custom python env” + pystan + clang-cl(msvc headers) is the only way to use thread_local on Windows. So I guess thread_local is not supported.
MPI will probably fail too, but Open-CL could work.
I’m thinking that Docker + easy interface would be the best option on Windows (something that could be doable in future with httpstan)
I’m thinking that Docker + easy interface would be the best option on Windows (something that could be doable in future with httpstan)
If any Windows people finds this thread in the future: Save yourself the aggrevation, and use Stan with Docker. I needed a Win Pro license, but apart from the getting started is very easy.
https://www.rocker-project.org/use/shared_volumes/
https://hub.docker.com/r/jrnold/rstan/
2 Likes