Hello Ricardo, I committed the patch with 'cl_amd_fp64' support to git, please check if it works for you. As for the errors in test_errors.py, I managed to find a machine with Radeon 46-something, so I will try to investigate this.
Best regards, Bogdan 2011/7/1 Ricardo Amézquita <[email protected]>: > Dear Bogdan > > As far as I know the results I'm obtaining are correct despite the warnings. > I'll try to make some more detailed tests and let you know. > > About the pyfft tests, the performance test runs, the func test runs but the > error test gives an error message and does not run. > > There is something else I forgot to tell you, in the kernel.mako file I had > to make a change to be able to use double numbers, I had to change the line > 11 from > > #pragma OPENCL EXTENSION cl_khr_fp64 : enable > > to > > #pragma OPENCL EXTENSION cl_amd_fp64: enable > > because this boards do not accept the first pragma (i guess the double > support is not complete) > > I had to do a similar change in pyopencl, so the kernels for sum could > compile > > Attached I'm sending the result for the tests. If you need anything else > please let me know, > > About Andreas question: > The max work group size reported is > > a.get_info(cl.device_info.MAX_WORK_GROUP_SIZE) > Out[8]: 256 > > but I think, I saw somewhere (I'm not completely sure and I cant find the > source right now), that when barriers where used, the group size was limited > to 64. > > > > One more thing, I would like to thank you, and also to thank Andreas for all > the work you have put into this pieces of software. > > Regards > > -- > Ricardo Amézquita O > > Combustión Ingenieros Ltda > Cll 25B # 85B - 76, Bogotá / Colombia > Tel: 57-1-2635232 > Fax: 57-1-4296516 > http://www.cihologramas.com > > _______________________________________________ PyOpenCL mailing list [email protected] http://lists.tiker.net/listinfo/pyopencl
