On Wed, 31 Aug 2011 17:33:07 -0400, Anthony LaTorre <tlator...@gmail.com> wrote: Non-text part: multipart/mixed Non-text part: multipart/alternative > I'm trying to understand how and when device memory gets freed. For example, > look at this piece of code:
Looks fine to me--after all, PyCUDA relies on the Python garbage collector to tell it when some chunk of memory is no longer needed. The only slightly suboptimal thing is that there seems to be a reference loop in GPUArrays, i.e. array -> device allocation -> array (?), so that deallocation really has to wait until the GC finds it. Also note that PyCUDA will run the GC if it encounters an out-of-memory situation on the device. If that's not good enough, you can also call a_gpu.gpudata.free() to manually free the memory. HTH, Andreas
pgpJKzwg2njOT.pgp
Description: PGP signature
_______________________________________________ PyCUDA mailing list PyCUDA@tiker.net http://lists.tiker.net/listinfo/pycuda
