- firedrake - imperial.ac.uk

Re: [firedrake] Chemora: A PDE Solving Framework for Modern HPC Architectures
by Fabio Luporini 12 Oct '14

12 Oct '14

I've just skimmed through the paper, it seems they share lots of high-level concepts with Firadrake/FEniCS, although it's weird they don't systematically compare to these approaches (OK, fine for Firedrake that is a new system, but FEniCS...) As for the code optimizations: the loop fission thing is interesting, although it's exactly the work the Carlo and OP2 people did in OP2. And they don't say the most important aspects: *when* they have to apply that and what "semi-automatically" means. As for the "explicit" vectorization, I'm really surprised a compiler can't do what they claim, perhaps that's why they don't even mention how they compiled the code, as far as I saw (btw, they also contradict themselves, since they also say "Since Kranc generates out- put in a high-level language (e.g. C++), it does not have to deal with very low-level architecture details such as register allocation or instruction selection." It would be interesting to know more. -- Fabio 2014-10-11 20:01 GMT+01:00 McRae, Andrew <a.mcrae12(a)imperial.ac.uk>: > "...The two most important optimisations were loop fission to not > overflow the instruction cache, and manual vectorisation..." > > Sounds like they have their own Fabio! > > On 11 October 2014 11:22, Rathgeber, Florian < > florian.rathgeber(a)imperial.ac.uk> wrote: > >> In Chemora, PDEs are expressed either in a high-level LaTeX-like >> language or in Mathematica. It's based on Cactus: >> >> http://arxiv.org/abs/1410.1764 >> >> Florian >> >> >

1 0

Re: [firedrake] Chemora: A PDE Solving Framework for Modern HPC Architectures
by Andrew McRae 12 Oct '14

12 Oct '14

"...The two most important optimisations were loop fission to not overflow the instruction cache, and manual vectorisation..." Sounds like they have their own Fabio! On 11 October 2014 11:22, Rathgeber, Florian < florian.rathgeber(a)imperial.ac.uk> wrote: > In Chemora, PDEs are expressed either in a high-level LaTeX-like > language or in Mathematica. It's based on Cactus: > > http://arxiv.org/abs/1410.1764 > > Florian > >

2 1

firedrake on ARCHER: PETSc issue
by Eike Mueller 11 Oct '14

11 Oct '14

Dear firedraker, after installing PETSc and petsc4py in my own $WORK, I can run sequentially, but if I run on more than one core it hangs when importing firdrake. I traced this down to the call of PETSc._initialise(args, comm) in the method init() in petsc4py/build/lib.linux-x86_64-2.7/petsc4py/__init__.py, which just does not return. It does pick up my PETSc installation correctly (printed out path, arch in ImportPETSc method in lib/__init__.py). I build the PETSc branch mlange/plex-distributed-overlap (same as in $FDRAKE_DIR) with the same configure options use there, and then I build petsc4py with make. Any ideas? Thanks, Eike -- Dr Eike Hermann Mueller Research Associate (PostDoc) Department of Mathematical Sciences University of Bath Bath BA2 7AY, United Kingdom +44 1225 38 5803 e.mueller(a)bath.ac.uk http://people.bath.ac.uk/em459/

3 4

Re: [firedrake] Firedrake paper results
by David Ham 11 Oct '14

11 Oct '14

Hi Florian, I'm in trying to make some progress on my bits of this. I think the results look really good. One issue currently is that you talk everywhere about strong scaling to n processors. I don't think this makes any sense. I think the relevant measure is that we strong scale down to m dofs per core. We could always strong scale to more processors by starting with a bigger problem, but the number of dofs per core is the actual limit. Cheers, David On 11 October 2014 11:27, Rathgeber, Florian < florian.rathgeber(a)imperial.ac.uk> wrote: > I'd still be interested in feedback on this! > > On 04/10/14 09:12, Florian Rathgeber wrote: > > Dear all, > > > > I finished a first draft of the results section for the Firedrake paper. > > Any feedback gratefully received! > > > > If you don't have access to the repository you can get a PDF from > > > > https://wwwhomes.doc.ic.ac.uk/~fr710/paper.pdf > > > > Cheers, > > Florian > > -- Dr David Ham Departments of Mathematics and Computing Imperial College London http://www.imperial.ac.uk/people/david.ham

1 0

Chemora: A PDE Solving Framework for Modern HPC Architectures
by Florian Rathgeber 11 Oct '14

11 Oct '14

In Chemora, PDEs are expressed either in a high-level LaTeX-like language or in Mathematica. It's based on Cactus: http://arxiv.org/abs/1410.1764 Florian

1 0

Sympy
by Colin Cotter 08 Oct '14

08 Oct '14

Dear all, I'm looking for a way to convert Sympy expressions into UFL expressions, where the symbols are Python functions. Any tips? Cheers Cjc

4 5

Re: [firedrake] Installing Firedrake
by Florian Rathgeber 07 Oct '14

07 Oct '14

Please send questions like this to the mailing list, which I have cc'ed. This backtrace by itself is unfortunately useless. Which machine is that? Which architecture? Which MPI? Which compiler did you use to compile PETSc, petsc4py, the extension modules? Which versions of PETSc, petsc4py, Firedrake, PyOP2 are you using? What are you running? Have you successfully run the PETSc tests on that machine? Florian P.S.: Lawrence's Edinburgh email address is no longer active, you should update your address book On 07/10/14 15:49, Bercea, Gheorghe-teodor wrote: > Hi Lawrence, Florian, > > I am having trouble installing firedrake on the LLNL computers, I followed all the steps online but when i want to import dmplex then I get a segfault. GDB gives the following backtrace: > #0 __strncpy_ssse3 () at ../sysdeps/x86_64/multiarch/strcpy.S:94 > #1 0x00002aaaad6b1080 in MPID_VIA_Init (argc=<value optimized out>, argv=0x2aaaac3dcd28, size=0x2aaaad8e7230, rank=0x2aaaad8e7234) > at /usr/local/tools/mvapich-gnu-1.2/src/mpid/ch_smp/viainit.c:137 > #2 0x00002aaaad68de4c in MPID_Init (argc=0x2aaaac3dcd24, argv=0x2aaaac3dcd28, config=<value optimized out>, error_code=0x7fffffff558c) at mpid_init.c:66 > #3 0x00002aaaad67b1cc in MPIR_Init (argc=0x2aaaac3dcd24, argv=0x2aaaac3dcd28) at initutil.c:355 > #4 0x00002aaaac4e8e85 in PetscInitialize () from /g/g90/bercea1/.local/lib/python2.7/site-packages/petsc/lib/libpetsc.so.3.05 > #5 0x00002aaaac10f638 in __pyx_f_8petsc4py_5PETSc_initialize (__pyx_v_args=<value optimized out>, __pyx_v_comm=0x77e800) at src/petsc4py.PETSc.c:216152 > #6 0x00002aaaac1108b9 in __pyx_pf_8petsc4py_5PETSc__initialize (__pyx_self=<value optimized out>, __pyx_args=<value optimized out>, __pyx_kwds=<value optimized out>) > at src/petsc4py.PETSc.c:216630 > #7 __pyx_pw_8petsc4py_5PETSc_1_initialize (__pyx_self=<value optimized out>, __pyx_args=<value optimized out>, __pyx_kwds=<value optimized out>) at src/petsc4py.PETSc.c:19970 > #8 0x00000000004a6c93 in call_function (f=<value optimized out>, throwflag=<value optimized out>) at Python/ceval.c:4012 > #9 PyEval_EvalFrameEx (f=<value optimized out>, throwflag=<value optimized out>) at Python/ceval.c:2665 > #10 0x00000000004a8767 in PyEval_EvalCodeEx (co=0x8cd8b0, globals=<value optimized out>, locals=<value optimized out>, args=<value optimized out>, argcount=1, kws=0x93c2b8, kwcount=0, > defs=0x8ac748, defcount=3, closure=0x0) at Python/ceval.c:3252 > #11 0x00000000004a695e in fast_function (f=<value optimized out>, throwflag=<value optimized out>) at Python/ceval.c:4108 > #12 call_function (f=<value optimized out>, throwflag=<value optimized out>) at Python/ceval.c:4033 > … > > > I can't work out why this is happening, have you got any ideas? > > Cheers, > > --Doru

1 0

Re: [firedrake] dmplex.so
by Florian Rathgeber 07 Oct '14

07 Oct '14

Issues like this should be raised on GitHub or sent to the list, which I have cc'ed. As explained in reply to your previous email, the problem is that you have your PYTHONPATH set wrong. firedrake is the *package* and therefore any modules within the firedrake directory are scoped by the package and there is no name clash if you set your PYTHONPATH correctly. I don't understand why you're suggesting to move dmplex.so. What would that solve? Where would you want to move it to? It's part of the firedrake package and imported by several other modules. Florian On 07/10/14 15:11, Bercea, Gheorghe-teodor wrote: > Hi Florian, > > I think it would be better if dmplex.so wasn't in /firedrake/firedrake because as soon as you add that path to the python path any gzip operation will import the io.py in that folder which is of course the wrong one in that case. Either that or change the name of io.py file. I think the latter would be better because that way we avoid the name clash altogether. > Let me know what you think. > > --Doru

1 0

Re: [firedrake] gzip
by Florian Rathgeber 07 Oct '14

07 Oct '14

Questions like this should go to the mailing list, which I have cc'ed. My crystal ball tells me that you have /usr/gapps/exmatex/firedrake/firedrake/firedrake on your PYTHONPATH, whereas you should have /usr/gapps/exmatex/firedrake/firedrake. Therefore you have a name clash with the io module from the standard library and the one from firedrake is picked up instead. Those paths are rather awkward, which system is this? Florian On 07/10/14 13:18, Bercea, Gheorghe-teodor wrote: > Hi Florian, > > Thanks for fixing the previous bug, now I have another. I'm trying to install Petsc via pip with --user enabled and I'm getting an error when importing gzip: > > Traceback (most recent call last): > File "/g/g90/bercea1/.local/bin/pip", line 5, in <module> > from pkg_resources import load_entry_point > File "build/bdist.linux-x86_64/egg/pkg_resources.py", line 22, in <module> > File "/usr/apps/python/lib/python2.7/zipfile.py", line 6, in <module> > import io > File "/usr/gapps/exmatex/firedrake/firedrake/firedrake/io.py", line 8, in <module> > from pyop2.logger import warning, RED > File "/usr/gapps/exmatex/firedrake/PyOP2/pyop2/__init__.py", line 9, in <module> > from op2 import * > File "/usr/gapps/exmatex/firedrake/PyOP2/pyop2/op2.py", line 39, in <module> > import base > File "/usr/gapps/exmatex/firedrake/PyOP2/pyop2/base.py", line 46, in <module> > from caching import Cached, ObjectCached > File "/usr/gapps/exmatex/firedrake/PyOP2/pyop2/caching.py", line 37, in <module> > import gzip > File "/usr/apps/python/lib/python2.7/gzip.py", line 36, in <module> > class GzipFile(io.BufferedIOBase): > > It might be the case that it's picking up some io.py file from my python path. > This is a related issue: > http://stackoverflow.com/questions/6315440/how-to-solve-attributeerror-when… > > Do you have any idea if this is the case? > > Cheers, > > --Doru

1 0

Re: [firedrake] Pytest
by Florian Rathgeber 06 Oct '14

06 Oct '14

The reason for this is a change in the way conftest files are discovered in pytest 2.6.3 [1]. Will push a fix. Florian [1]: http://pytest.org/latest/changelog.html#id1 On 06/10/14 18:43, Bercea, Gheorghe-teodor wrote: > Hi Florian, > > I'm trying to run the PyOP2 unit test using 'make test'. The first three tests pass but starting from the 4th test every test which refers to the 'backend' fixture fails: > fixture 'backend' not found > available fixtures: pytestconfig, x, iter2ind1, iter2ind2, tmpdir, iterset, g, capfd, indset2, xl, recwarn, x2, mat, monkeypatch, iter2ind22, capsys, y, indset > use 'py.test --fixtures [testpath]' for help on them. > > I was wondering what I'm missing because it seems that 'backend' is in conftest.py but it doesn't seem to get picked up. I'm using pytest 2.6.4. > > Let me know if you have any idea about what could be going wrong. > > Many thanks, > > --Doru

1 0