Hi all,

I'm writing to let you know that the initial slicing support in PyCUDA
and PyOpenCL has had a slightly unintended performance consequence due
to this numpy bug:

https://github.com/numpy/numpy/issues/3375

I've written about this in the release notes here:

http://documen.tician.de/pyopencl/misc.html#version-2013-1
http://documen.tician.de/pycuda/misc.html#version-2013-1

The upshot is that arithmetic with numpy scalars and Py{CUDA,OpenCL}
arrays could be *very* slow under some circumstances. I've submitted a
patch to numpy to fix this (see the linked bug log). In the meantime,
check the two links above for possible workarounds.

Andreas

Attachment: pgpH9VyEaoIHa.pgp
Description: PGP signature

_______________________________________________
PyCUDA mailing list
[email protected]
http://lists.tiker.net/listinfo/pycuda

Reply via email to