March 2011 - petsc-dev - mailman3.cels.anl.gov

Re: [petsc-dev] [petsc-maint #68731] Installation problem with petsc4py
by Jed Brown 31 Mar '11

31 Mar '11

As a related matter, could we add a field PETSC_PYTHON to petscvariables so that Makefile users can run a consistent version?

1 0

Re: [petsc-dev] MatNullSpaceDestroy
by Lisandro Dalcin 31 Mar '11

31 Mar '11

On 24 March 2011 13:43, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > >> o now MatNullSpace instances cannot be freed with >> PetscObjectDestroy()... This change broke bad petsc4py Python+PETSc >> refcounting management. BTW, PetscObjectDestroy() does not take a >> pointer... And IMHO PetscHeaderDestroy should also take a pointer... > > Thank you for the explanation. petsc4py relies on PetscObjectDestroy() which won't work when some destroys take & and some don't. I didn't know you needed PetscObjectDestroy() to always work. > > The correction solution is to switch to using the & in all destroys. It sounds like this has to happen sooner rather than later. > In the mean time, I've pushed a quick, one-line workaround for this, plus a (non public) View routine. -- Lisandro Dalcin --------------- CIMEC (INTEC/CONICET-UNL) Predio CONICET-Santa Fe Colectora RN 168 Km 472, Paraje El Pozo 3000 Santa Fe, Argentina Tel: +54-342-4511594 (ext 1011) Tel/Fax: +54-342-4511169

1 0

Re: [petsc-dev] [petsc-maint #68731] Installation problem with petsc4py
by Barry Smith 31 Mar '11

31 Mar '11

On Mar 30, 2011, at 2:49 PM, David Maxwell wrote: > Barry, Andy: > > I suppose you can avoid both these gotchas by involving the end user in the choice rather than making it arbitrarily. For example, in reconfigure: > > * Don't hardwire the python executable, but > > * Check that the python executable running the script is the original one (compare sys.executable with a stored value). > > If there is a mismatch, print a meaningful error message telling the user to either run the script explicitly with the original python, or to set a command line flag indicating the current python is to be used (and is then a part of the reconfiguration). Good idea. Satish, unless you can find a problem with this suggestion could you please make this change. > > Speaking of meaningful error messages, the message from petsc4py.Configure.configureLibrary: > > realpath of /usr/lib/libpython.dylib (/System/Library/Frameworks/Python.framework/Versions/2.6/Python) does not point to Python library path (/Library/Frameworks/Python.framework/Versions/6.1/Python) for current Python; > Are you not using the Apple python? Fixed. Barry > > is misleading since the current python library path is not obliged to be /usr/lib/libpython's realpath, but can also be one of a handful of hardcoded macports pythons as well. > > -David > > >> >> We've had endless discussions about this. If we don't hardwire it and someone runs the reconfigure... after a path change then they get something different and get very confused. If we do hardware it then it cannot easily use a different python the way a user might think it would. We've had people screwed on both sides of this (we didn't used to hardware it). >> >> If you have a solution distinct from those two flawed schemes we'd love to try it. >> >> Barry >> >>> >>> Cheers, >>> >>> Andy >>> >>> >>> On Mar 29, 2011, at 11:09 PM, Matthew Knepley wrote: >>> >>>> On Tue, Mar 29, 2011 at 9:18 PM, Andy Aschwanden <andy.aschwanden(a)arsc.edu> wrote: >>>> Dear PETSc developers >>>> >>>> I'm trying to install petsc-3.1-p4 with --download-petsc4py on my OSX 10.6, and get the following error: >>>> >>>> TESTING: configureLibrary from PETSc.packages.petsc4py(config/PETSc/packages/petsc4py.py:43) ******************************************************************************* >>>> UNABLE to CONFIGURE with GIVEN OPTIONS (see configure.log for details): >>>> ------------------------------------------------------------------------------- >>>> realpath of /usr/lib/libpython.dylib (/System/Library/Frameworks/Python.framework/Versions/2.6/Python) does not point to Python library path (/Library/Frameworks/Python.framework/Versions/6.1/Python) for current Python; >>>> Are you not using the Apple python? >>>> >>>> Detailed configure.log is attached. >>>> >>>> The answer to the above question is: I'm *NOT* using the Apple python, instead I have python26 installed via MacPorts. There are, however, various python installations on my system, including an EPD 6.1 in /Library/Frameworks/Python.framework/Versions/6.1/Python. >>>> >>>> I will look at the log, but this install is no longer the easiest one. If you just install >>>> PETSc, and then use >>>> >>>> pip install petsc4py >>>> >>>> when you have PETSC_DIR and PETSC_ARCH defined, it should work. >>>> >>>> Thanks, >>>> >>>> Matt >>>> >>>> I did >>>> $ python_select python26 >>>> to select the MacPorts python distribution I want to use with petsc4py, and indeed >>>> >>>> $ which python >>>> /opt/local/bin/python >>>> >>>> points to the right place. I don't know why petsc picks up the wrong python. Is there a way to tell petsc which python version it should use for petsc4py. >>>> >>>> >>>> Thanks, >>>> >>>> Andy >>>> >>>> CC: David Maxwell >>>> >>>> >>>> Dr. Andy Aschwanden >>>> ARSC Postdoctoral Fellow >>>> Geophysical Institute >>>> 903 Koyukuk Drive >>>> Fairbanks AK 99775 >>>> USA >>>> >>>> email andy.aschwanden(a)arsc.edu >>>> phone +1 (907) 474-7199 >>>> fax +1 (907) 474-7290 >>>> >>>> >>>> >>>> >>>> >>>> >>>> -- >>>> What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead. >>>> -- Norbert Wiener >>> >>> Dr. Andy Aschwanden >>> ARSC Postdoctoral Fellow >>> Geophysical Institute >>> 903 Koyukuk Drive >>> Fairbanks AK 99775 >>> USA >>> >>> email andy.aschwanden(a)arsc.edu >>> phone +1 (907) 474-7199 >>> fax +1 (907) 474-7290 >>> >>> >>> >>> >> > >

1 0

Re: [petsc-dev] PETSc Sorting with duplicate entries
by Barry Smith 31 Mar '11

31 Mar '11

On Mar 31, 2011, at 11:34 AM, Todd Munson wrote: > >> PetscSortIntWithPermutation() implements bubble sort for n<8 and >> insertion sort otherwise, >> which does not give ordered index for duplicate entries. If you can >> find the algorithm used by Matlab, please >> let us know. We can implement it. Otherwise, I think you may write a >> subroutine to sort index() of >> duplicate entries, which takes O(n) operations, a low order operation >> comparing to sort(). > > Why can't the sort routine simply use two comparisons? In particular, > you use the value as the first comparison and in case of ties use the > index to break ties? This means that there are no identical values > assuming all indices are unique. This results in a slower sort than one that ignores the index values. Usually one wants sorts to be as fast as possible so we don't include this. If you want the double sort you should provide that as "yet another routine" that would be called by people who care about this additional feature. Barry > > Cheers, Todd. >

1 0

Re: [petsc-dev] PETSc Sorting with duplicate entries
by Jed Brown 31 Mar '11

31 Mar '11

On Thu, Mar 31, 2011 at 19:29, Hong Zhang <hzhang(a)mcs.anl.gov> wrote: > PetscSortIntWithPermutation() implements bubble sort for n<8 and > insertion sort otherwise, > insertion sort for n<8 and quicksort otherwise

1 0

Re: [petsc-dev] PETSc Sorting with duplicate entries
by Todd Munson 31 Mar '11

31 Mar '11

> PetscSortIntWithPermutation() implements bubble sort for n<8 and > insertion sort otherwise, > which does not give ordered index for duplicate entries. If you can > find the algorithm used by Matlab, please > let us know. We can implement it. Otherwise, I think you may write a > subroutine to sort index() of > duplicate entries, which takes O(n) operations, a low order operation > comparing to sort(). Why can't the sort routine simply use two comparisons? In particular, you use the value as the first comparison and in case of ties use the index to break ties? This means that there are no identical values assuming all indices are unique. Cheers, Todd.

1 0

Re: [petsc-dev] PETSc Sorting with duplicate entries
by Hong Zhang 31 Mar '11

31 Mar '11

Gautam: > I'm was sorting an array which that has duplicate entries: (x = [39 9 19 39 29]). I noticed that for duplicate entries after sorting, the index returned are not in ascending. The example below will explain this further. > > Here is the code snippet: > > ============================================ > do ii=1,5 > index(ii) = ii > endo > > index = index -1 > call PetscSortIntWithPermutation( 5, x, index, ierr) > index = index + 1 > > do ii=1,5 > write(*,*), x(ii), index(ii), x(index(ii)) > enddo > ============================================ > > > Output: > 39 2 9 > 9 3 19 > 19 5 29 > 39 4 39 > 29 1 39 > PetscSortIntWithPermutation() implements bubble sort for n<8 and insertion sort otherwise, which does not give ordered index for duplicate entries. If you can find the algorithm used by Matlab, please let us know. We can implement it. Otherwise, I think you may write a subroutine to sort index() of duplicate entries, which takes O(n) operations, a low order operation comparing to sort(). Hong > Output (Expected by me, similar to MATLAB output): Note the difference in the index(4) and index(5) values > 39 2 9 > 9 3 19 > 19 5 29 > 39 1 39 > 29 4 39 > > > I tired using PetscSortIntWithArray and go similar result. I was wondering if there is a sorting option that can get me the output which I'm expecting. The only way around I came up for this was to look at the index(:) entries that correspond to duplicate values and sort them again. > > > -Gautam. >

1 0

PETSc Sorting with duplicate entries
by Bisht, Gautam 31 Mar '11

31 Mar '11

Hi, I'm was sorting an array which that has duplicate entries: (x = [39 9 19 39 29]). I noticed that for duplicate entries after sorting, the index returned are not in ascending. The example below will explain this further. Here is the code snippet: ============================================ do ii=1,5 index(ii) = ii endo index = index -1 call PetscSortIntWithPermutation( 5, x, index, ierr) index = index + 1 do ii=1,5 write(*,*), x(ii), index(ii), x(index(ii)) enddo ============================================ Output: 39 2 9 9 3 19 19 5 29 39 4 39 29 1 39 Output (Expected by me, similar to MATLAB output): Note the difference in the index(4) and index(5) values 39 2 9 9 3 19 19 5 29 39 1 39 29 4 39 I tired using PetscSortIntWithArray and go similar result. I was wondering if there is a sorting option that can get me the output which I'm expecting. The only way around I came up for this was to look at the index(:) entries that correspond to duplicate values and sort them again. -Gautam.

1 0

Re: [petsc-dev] PetscBinaryRead for complex double dense matrices
by Barry Smith 31 Mar '11

31 Mar '11

Thank you for letting us know. This is now fixed in petsc-dev Barry On Mar 31, 2011, at 9:20 AM, Alexander Grayver wrote: > Hallo, > > May be this topic has already been discussed, I've not done search, but anyway it seems that PetscBinaryRead from 3.1-p7 doesn't support reading complex double dense matrices. > > I had to add this code: > > if (nz == -1) > if strcmp(comp,'complex') > s = read(fd,2*m*n,'double'); > A = reshape(complex(s(1:2:2*m*n),s(2:2:2*m*n)),n,m).'; > else > s = read(fd,m*n,'double'); > A = reshape(s,n,m)'; > end > else > > to be able to read them. > > Regards, > Alexander

1 0

PetscBinaryRead for complex double dense matrices
by Alexander Grayver 31 Mar '11

31 Mar '11

Hallo, May be this topic has already been discussed, I've not done search, but anyway it seems that PetscBinaryRead from 3.1-p7 doesn't support reading complex double dense matrices. I had to add this code: if (nz == -1) if strcmp(comp,'complex') s = read(fd,2*m*n,'double'); A = reshape(complex(s(1:2:2*m*n),s(2:2:2*m*n)),n,m).'; else s = read(fd,m*n,'double'); A = reshape(s,n,m)'; end else to be able to read them. Regards, Alexander

1 0