Hi Ying Wai,
On Freitag 11 Dezember 2009, Ying Wai (Daniel) Fan wrote:
I am using your PyCUDA on some image processing
applications. I have
modified some PyCUDA files for additional functionality. You may find
these small changes useful to be included in next release of PyCUDA.
I have modified functions in elementwise.py to for additional codes to
run before and after the GPU threaded for-loop. In dtype_to_ctype of
tools.py, I have added support for complex numbers. I have attached the
Regarding your changes to elementwise.py, a similar patch for "loop_prep" has
already made it in, I've added a post_loop parameter to suit your needs.
I haven't merged your changes to tools.py, because I don't understand the
reasoning behind most of them:
- from struct import calcsize
- return calcsize('l') * 8
+ return tuple.__itemsize__ * 8
Why is this better?
+ elif dtype == numpy.complex64:
+ return "float2"
This is incorrect.
+ elif tp in ["bool"]: return "?"
"?" is not a documented type character in Python's "struct"
Please give me credit for these changes when
Done, see revlog.
I have also written some codes for doing Fast Fourier
complex number arithmetics with PyCUDA. If you think these can be
included in PyCUDA, I can send you these files too.
Please do, I think this could be very useful.
Further, the preferred way of submitting changesets is by sending a diff
against a recent git tree, rather than entire modified files from the last
Finally, please subscribe to the PyCUDA mailing list and send future
modifications there instead of to my personal address.
Thanks for your contribution.