Dnia 2012-08-21, wto o godzinie 11:49 -0400, Andreas Kloeckner pisze:
[ cut ]
> This seems to indicate that 'out of device memory' likely shouldn't be
> reported as a subclass of MemoryError. pytest seems to treat that as
> unrecoverable, and I have to admit that it has a point there.
> 
> I've removed MemoryError from the base classes of PyOpenCL's and
> PyCUDA's MemoryError classes.
> 
> If you catch MemoryErrors and mean to catch device memory errors, this
> will break your code. (But I don't think that's a likely scenario, so I
> went ahead and made the change.)
> 

OK, this change seems sensible.
I have rebuild latest PyOpenCL and:
1. On 1GB GTX 460 test_array.py passes on both Python2 and Python3
2. On Loveland test_array.py gives 5 failures on Python2 and 6 failures
on Python3.
3. On ION (GeForce 9400M) test_array.py gives 5 errors on
Most of those are
E   pyopencl.RuntimeError: clEnqueueNDRangeKernel failed: out of
resources
although I have also
test_array.py:1109: AssertionError

While those failures do not show any problems with PyOpenCL, but
situation when tests pass or fail depending on hardware configuration
seems strange to me. Should I look into those tests and try to fix them
not to fail but maybe to finish earlier?

Best regards.

-- 
Tomasz Rybak  GPG/PGP key ID: 2AD5 9860
Fingerprint A481 824E 7DD3 9C0E C40A  488E C654 FB33 2AD5 9860
http://member.acm.org/~tomaszrybak

Attachment: signature.asc
Description: This is a digitally signed message part

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

Reply via email to