IMHO, memcpy and to_device should act as closely to C memcpy as possible - just blindly copying bytes. Anything else has the potential to be very unintuitive and frustrating to C programmers, which seems like a bad thing to me seeing as CUDA is C(++).
If you want something to be smart about arrays, isn't that what GPUArray is for? Just my 2 cents. Ian Cullinan ________________________________________ From: [email protected] [[email protected]] On Behalf Of Andreas Klöckner [[email protected]] Sent: Monday, 2 March 2009 8:21 AM To: Nicholas Tung Cc: [email protected] Subject: Re: [PyCuda] pycuda memcpy_htod 2) I'm torn about the flag check. I don't want people falling into the same trap as you. On the other hand, your change outlaws legitimate code (and forces it into a sketchy-sounding "_unchecked" code path). Also, having this check just for C-contiguity and not F is kind of asymmetric. If there are strong opinion one way or the other, now is the time to say something. _______________________________________________ PyCuda mailing list [email protected] http://tiker.net/mailman/listinfo/pycuda_tiker.net
