Dnia 2013-04-23, wto o godzinie 00:27 -0400, Andreas Kloeckner pisze: > Hi Tomasz, > > Tomasz Rybak <[email protected]> writes: > > Sorry if there are two copies of this message. > > I have sent it to the list but received no confirmation > > (nor any error) and checked that archive does not show > > any message from January. > > I can see that there is already new version (2013.1) in docs, > > marked "in development". I would like for it not to be released > > before fixing problems with parallel prefix scan. > > > > Problems with scan are only visible on APU Loveland. > > (snip) > > I think I may have some good news here. I was encountering these same > failures on a "Devastator" (who makes up these names?!) APU (on a > Trinity chip). > > After upgrading the OpenCL ICD (and only that) to 13.3 beta 3 from here, > the issue seems to have disappeared. > > http://support.amd.com/us/kbarticles/Pages/AMDCatalyst13-3LINBetaDriver.aspx > > To be precise, I've just ripped these files out of their driver package, > uninstalled the Debian-packaged 13.1 ICD, pointed an .icd file in > /etc/OpenCL/vendors at libamdocl64.so, and all appeared to be well. > > libamdocl64.so > libamdsc64.so > libSlotMaximizerAg.so > libSlotMaximizerBe.so > > It'd be great if you could try this and confirm. That'd let me be fairly > confident about rolling a release soon. > > Andreas
Thanks for the tip. I've downloaded latest git revision, built package on 13.1 and got 5 failures from test_algorithms.py on 13.1 OpenCL 1.2 AMD-APP (1084.4). Then I've put 4 files from 13.3 into /usr/lib/x86_64-linux-gnu/ and run tests again; got 7 failures :-| After reboot and removing __pycache__ from test/ directory I got 2 failures on 13.3 OpenCL 1.2 AMD-APP (1124.2): test_algorithm.py:827: AssertionError test_key_value_sorter[ctx_factory=<context factory for <pyopencl.Device 'Loveland' on 'AMD Accelerated Parallel Processing' at 0x3569ca0>>] > assert sorted(mydict[i]) == sorted(lists[start:end]) E assert [4300, 7248, ...3, 10861, ...] == [1359271250154...25015456, ...] E At index 0 diff: 4300 != 135927125015456 test_algorithm.py:770: AssertionError test_sort[ctx_factory=<context factory for <pyopencl.Device 'Loveland' on 'AMD Accelerated Parallel Processing' at 0x3569ca0>>] It's much better, but not ideal. Interesting thing is that sorting seems to be a problems. But my nose says that it is some problem with optimization in the driver or OpenCL implementation - new drivers fixed some of the problems and I did not seen any problems on NVIDIA implementation or AMD CPU. I've just got email that Simon Richter created Debian package for Intel Beignet OpenCL implementation so I intend to test PyOpenCL with it. I'll try to play more with AMD OpenCL drivers on Friday or during the weekend; now my internet connection is unstable and I do not want to swap graphics drivers in my system without ability to download stable versions of packages to restore it. BTW - I tested PyOpenCL on NVIDIA hardware and just noticed you have fixed problem with crash on Image tests (patch from 2013-01-22, adding check for context version).Thanks. 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
signature.asc
Description: This is a digitally signed message part
_______________________________________________ PyOpenCL mailing list [email protected] http://lists.tiker.net/listinfo/pyopencl
