Hey Andreas,
Sorry for writing in german the first time, I'm switching to english
from now on. By the way Andreas Kloeckner sounds german, doesn't it ?
Back to the problem, thanks for testing and attaching the little sample.
The mistake I made in my program was pretty simple but ugly (as I said
Pyhton is not my strength ^^)!
Instead of:
self.kernel.XYZ(self.commandQueue, (data.size, ), (self.WORKSIZE, ),
*(kernelargs), global_offset=(data.baseNum[i], ))
I did:
self.kernel.XYZ(self.commandQueue, (data.size, ), (self.WORKSIZE, ),
*(kernelargs), global_offset=data.baseNum[i])
The problem here was, that the host app didn't give an error message
(which pyOpenCL would have shown for the above construct I guess), but
simply crashed :-/ with a fault in kernelbase.dll (im on Win7 x64). Well
now it works like a charm!
Thanks for your help,
Philip
> I hope you don't mind me replying in English.
>
> I just tried running an example with an offset:
>
> http://tiker.net/tmp/demo-offset.py
>
> My results were as follows:
>
> - AMD CPU: fine
> - Intel CPU: fine
> - AMD GPU (Beaver Creek, APU): NaN
> - Nvidia GPU (260): NaN
>
> I think PyOpenCL's support is fine. Can you try and get a backtrace on
> your crash with gdb?
>
> gdb --args python my-script.py arg1 arg2
>
> r[Enter]
> (wait for crash)
> bt[Enter]
> (copy and paste output of that)
>
> Andreas
_______________________________________________
PyOpenCL mailing list
[email protected]
http://lists.tiker.net/listinfo/pyopencl
