> > Very cool, David! > Thanks. :-)
Does it work with windows? CUDA is great but I'm > stuck at the moment with computers running windows > rather than unix and (unfortunately) CUDA doesn't > speak gcc for windows platforms. > Ostensibly, yes, but there are caveats and I'm not sure about the details: my CUDA box does not have Windows on it. The caveats arise because nvcc assumes you're using cl if you're on a Windows box unless you specify the certain command-line options. These options can be passed to perl_nvcc and will be properly sent along to nvcc, so it should work, but it'll take a bit of research to figure out exactly what those flags should be. Part of my push to get PDL fully ported to windows > is to allow for direct CUDA support. > Yes. With the work I have here, PDL::CUDA may come sooner than you had expected. :-) I'm working on releasing perl_nvcc to CPAN some time in the next week or so. After I get perl_nvcc out to CPAN, the next step will be Inline::CUDA (to wrap up all of those flags in an automatic way). All of that will probably take the rest of the month to get worked out, at which point I'll move the discussion of the CUDA bindings over to the PDL list, where it would make a lot more sense. David
