Simon Ekeberg <[email protected]> writes:

> Hi again, thanks for quick response :)
>
> I've managed to get pyopencl working on a linux machine at work. I'm using
> nvidias implementation from their latest cuda sdk.
> All pyopencl tests runs fine now.

Great. I would also be interested in the output of the tests running on
Apple, if you can produce those. Run

PYOPENCL_TEST="apple:DEVICE_NAME_SUBSTRING" py.test --tb=native

for GPU and CPU.

> Boxtrees demo script has some problems still:
>
> pyopencl.RuntimeError: clBuildProgram failed: invalid binary -
> Build on <pyopencl.Device 'Quadro 4000' on 'NVIDIA CUDA' at 0x26a2490>:
> ptxas error   : Entry function 'scan_scan_intervals' uses too much shared
> data (0xc0b4 bytes, 0xc000 max)
>
> I'm not shure if this is related to that I didn't install the drivers
> coming with the sdk. The opencl lib that we have installed seems to be
> 1.0.0 while the headers I used from the sdk is 1.1 (I think). I will try to
> get the latest driver installed by our admin.
>
> Please note that this is just for fun, so don't prioritize:)

Boxtree is very new. I've brought it up using the Intel and AMD CPU
implementations on Linux and am still working on actual GPU support. So
this is valuable feedback. I've also got my own CI server running tests,
that's showing that there are still some issues--it's pre-release code
after all! But it'll get fixed up in due course.

Btw I've cc'd the pyopencl list, hope you don't mind.

Andreas


_______________________________________________
PyOpenCL mailing list
[email protected]
http://lists.tiker.net/listinfo/pyopencl

Reply via email to