Hi, I have a number of questions about the internal architecture of Theano itself, simply out of a wish to understand what I'm using.
If I understand correctly, when I use the "cpu" backend, Theano generates a bunch of C++ code and compiles and runs it. My first question is: how is it possible that Theano uses g++ on Windows, when I know that Python is built with Visual Studio and therefore C++ code built with other compilers does not link correctly to Python? When I use the "gpu" backend, Theano uses the same approach, but incorporates nvcc into the mix as well. My second question: when I build libgpuarray and switch to the "cuda" backend, how does it work? What is the role of libgpuarray? Does it contain all the GPU code so that no new code has to be generated at runtime? Why can't libgpuarray be distributed as a prebuilt binary? Is there a technical reason why I can't just "pip install pygpu"? Márton -- --- You received this message because you are subscribed to the Google Groups "theano-users" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
