April 2012 - petsc-dev - mailman3.cels.anl.gov

Re: [petsc-dev] Recommended Petsc Build Procedure
by Dave Nystrom 22 Apr '12

22 Apr '12

Matthew Knepley writes: > On Sun, Apr 22, 2012 at 10:42 AM, Dave Nystrom <Dave.Nystrom(a)tachyonlogic.com> wrote: > > > At the end of configure.log, there are two possible ways to build petsc-dev > > that are specified. Which is the recommended way to build - using make or > > using python? I have been using make. > > > > Also, one is labeled as legacy and one is labeled as experimental. That > > gives the impression of having a choice between an old, archaic method or a > > new, experimental approach. Should one just be labeled as production? > > You do not have CMake, and thus did not get the "production" printout. % which cmake ~/bin/cmake % cmake --version cmake version 2.8.6 I also used the "--download-cmake=yes" for petsc configure and I see that petsc did build cmake. > The make is indeed legacy, but just as fast as CMake (Aron), as I believe > the Python is as well. The Python is experimental, but I have been using > it for a year and it works fine. > > Matt

1 0

Re: [petsc-dev] Recommended Petsc Build Procedure
by Aron Ahmadia 22 Apr '12

22 Apr '12

You can do parallel builds with CMake... A On Sun, Apr 22, 2012 at 5:57 PM, Dave Nystrom <Dave.Nystrom(a)tachyonlogic.com > wrote: > I have cmake on my system but configure.log does not document that > approach. > What is your build command to build petsc with cmake? > > Thanks, > > Dave > > Aron Ahmadia writes: > > I use the CMake build because it's the fastest, though it requires you > to > > have CMake installed on your system. > > > > A > > > > On Sun, Apr 22, 2012 at 5:42 PM, Dave Nystrom < > Dave.Nystrom(a)tachyonlogic.com > > > wrote: > > > > > At the end of configure.log, there are two possible ways to build > petsc-dev > > > that are specified. Which is the recommended way to build - using > make or > > > using python? I have been using make. > > > > > > Also, one is labeled as legacy and one is labeled as experimental. > That > > > gives the impression of having a choice between an old, archaic > method or a > > > new, experimental approach. Should one just be labeled as production? > > > > > > Thanks, > > > > > > Dave > > > >

1 0

Re: [petsc-dev] Recommended Petsc Build Procedure
by Matthew Knepley 22 Apr '12

22 Apr '12

On Sun, Apr 22, 2012 at 10:57 AM, Dave Nystrom < Dave.Nystrom(a)tachyonlogic.com> wrote: > I have cmake on my system but configure.log does not document that > approach. > We did not find it. This is a configure question, and should be mailed to petsc-maint, not the dev list. > What is your build command to build petsc with cmake? > make Matt > Thanks, > > Dave > > Aron Ahmadia writes: > > I use the CMake build because it's the fastest, though it requires you > to > > have CMake installed on your system. > > > > A > > > > On Sun, Apr 22, 2012 at 5:42 PM, Dave Nystrom < > Dave.Nystrom(a)tachyonlogic.com > > > wrote: > > > > > At the end of configure.log, there are two possible ways to build > petsc-dev > > > that are specified. Which is the recommended way to build - using > make or > > > using python? I have been using make. > > > > > > Also, one is labeled as legacy and one is labeled as experimental. > That > > > gives the impression of having a choice between an old, archaic > method or a > > > new, experimental approach. Should one just be labeled as production? > > > > > > Thanks, > > > > > > Dave > > > > -- 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

1 0

Re: [petsc-dev] Recommended Petsc Build Procedure
by Dave Nystrom 22 Apr '12

22 Apr '12

I have cmake on my system but configure.log does not document that approach. What is your build command to build petsc with cmake? Thanks, Dave Aron Ahmadia writes: > I use the CMake build because it's the fastest, though it requires you to > have CMake installed on your system. > > A > > On Sun, Apr 22, 2012 at 5:42 PM, Dave Nystrom <Dave.Nystrom(a)tachyonlogic.com > > wrote: > > > At the end of configure.log, there are two possible ways to build petsc-dev > > that are specified. Which is the recommended way to build - using make or > > using python? I have been using make. > > > > Also, one is labeled as legacy and one is labeled as experimental. That > > gives the impression of having a choice between an old, archaic method or a > > new, experimental approach. Should one just be labeled as production? > > > > Thanks, > > > > Dave > >

1 0

Re: [petsc-dev] Recommended Petsc Build Procedure
by Matthew Knepley 22 Apr '12

22 Apr '12

On Sun, Apr 22, 2012 at 10:42 AM, Dave Nystrom < Dave.Nystrom(a)tachyonlogic.com> wrote: > At the end of configure.log, there are two possible ways to build petsc-dev > that are specified. Which is the recommended way to build - using make or > using python? I have been using make. > > Also, one is labeled as legacy and one is labeled as experimental. That > gives the impression of having a choice between an old, archaic method or a > new, experimental approach. Should one just be labeled as production? > You do not have CMake, and thus did not get the "production" printout. The make is indeed legacy, but just as fast as CMake (Aron), as I believe the Python is as well. The Python is experimental, but I have been using it for a year and it works fine. Matt > Thanks, > > Dave > -- 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

1 0

Re: [petsc-dev] Recommended Petsc Build Procedure
by Aron Ahmadia 22 Apr '12

22 Apr '12

I use the CMake build because it's the fastest, though it requires you to have CMake installed on your system. A On Sun, Apr 22, 2012 at 5:42 PM, Dave Nystrom <Dave.Nystrom(a)tachyonlogic.com > wrote: > At the end of configure.log, there are two possible ways to build petsc-dev > that are specified. Which is the recommended way to build - using make or > using python? I have been using make. > > Also, one is labeled as legacy and one is labeled as experimental. That > gives the impression of having a choice between an old, archaic method or a > new, experimental approach. Should one just be labeled as production? > > Thanks, > > Dave >

1 0

Recommended Petsc Build Procedure
by Dave Nystrom 22 Apr '12

22 Apr '12

At the end of configure.log, there are two possible ways to build petsc-dev that are specified. Which is the recommended way to build - using make or using python? I have been using make. Also, one is labeled as legacy and one is labeled as experimental. That gives the impression of having a choice between an old, archaic method or a new, experimental approach. Should one just be labeled as production? Thanks, Dave

1 0

Re: [petsc-dev] MatAXPY with DIFFERENT_NONZERO_PATTERN
by Alexander Grayver 22 Apr '12

22 Apr '12

Matt, Here are matrices (double complex, 3 MB each): http://dl.dropbox.com/u/60982984/X.dat http://dl.dropbox.com/u/60982984/Y.dat Code: static char help[] = ""; int main(int argc,char **args) { Mat A,B; PetscErrorCode ierr; PetscViewer viewer; PetscScalar cone; PetscInitialize(&argc,&args,(char *)0,help); ierr = PetscViewerBinaryOpen(PETSC_COMM_WORLD,"Y.dat",FILE_MODE_READ,&viewer);CHKERRQ(ierr); ierr = MatCreate(PETSC_COMM_WORLD,&A);CHKERRQ(ierr); ierr = MatSetFromOptions(A);CHKERRQ(ierr); ierr = MatLoad(A,viewer);CHKERRQ(ierr); ierr = PetscViewerDestroy(&viewer);CHKERRQ(ierr); ierr = PetscViewerBinaryOpen(PETSC_COMM_WORLD,"X.dat",FILE_MODE_READ,&viewer);CHKERRQ(ierr); ierr = MatCreate(PETSC_COMM_WORLD,&B);CHKERRQ(ierr); ierr = MatSetFromOptions(B);CHKERRQ(ierr); ierr = MatLoad(B,viewer);CHKERRQ(ierr); ierr = PetscViewerDestroy(&viewer);CHKERRQ(ierr); cone = 1.0; ierr = MatAYPX(A,cone,B,DIFFERENT_NONZERO_PATTERN); ierr = PetscFinalize(); return 0; } It works fine if you run it with: mpirun -n 10 solveTest -mat_type aij and crashes if n > 10 Thanks. On 22.04.2012 14:51, Matthew Knepley wrote: > On Sun, Apr 22, 2012 at 8:41 AM, Alexander Grayver > <agrayver(a)gfz-potsdam.de <mailto:[email protected]>> wrote: > > Sorry, I don't quite get. Is this something I can avoid by doing > preallocation properly or petsc issue that will be fixed? > > > This looks like a PETSc bug to me, but it would be nice to get input > that triggers it. > > Thanks, > > Matt > > On 20.04.2012 14:23, Jed Brown wrote: >> On Fri, Apr 20, 2012 at 05:18, Matthew Knepley <knepley(a)gmail.com >> <mailto:[email protected]>> wrote: >> >> Did someone intend to fix this, and then >> stop? MatAXPY_BasicWithPreallocation() is the same >> as MatAXPY_Basic(). >> >> >> Look how it's called. I don't know what's going on here. >> >> ierr = >> MatAXPYGetPreallocation_SeqAIJ(yy->A,xx->A,nnz_d);CHKERRQ(ierr); >> ierr = >> MatAXPYGetPreallocation_SeqAIJ(yy->B,xx->B,nnz_o);CHKERRQ(ierr); >> ierr = >> MatMPIAIJSetPreallocation(B,0,nnz_d,0,nnz_o);CHKERRQ(ierr); >> ierr = MatAXPY_BasicWithPreallocation(B,Y,a,X,str);CHKERRQ(ierr); >> > > > -- > Regards, > Alexander > > > > > -- > 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 -- Regards, Alexander

1 0

Re: [petsc-dev] MatAXPY with DIFFERENT_NONZERO_PATTERN
by Matthew Knepley 22 Apr '12

22 Apr '12

On Sun, Apr 22, 2012 at 8:41 AM, Alexander Grayver <agrayver(a)gfz-potsdam.de>wrote: > ** > Sorry, I don't quite get. Is this something I can avoid by doing > preallocation properly or petsc issue that will be fixed? > This looks like a PETSc bug to me, but it would be nice to get input that triggers it. Thanks, Matt > On 20.04.2012 14:23, Jed Brown wrote: > > On Fri, Apr 20, 2012 at 05:18, Matthew Knepley <knepley(a)gmail.com> wrote: > >> Did someone intend to fix this, and then >> stop? MatAXPY_BasicWithPreallocation() is the same as MatAXPY_Basic(). >> > > Look how it's called. I don't know what's going on here. > > ierr = > MatAXPYGetPreallocation_SeqAIJ(yy->A,xx->A,nnz_d);CHKERRQ(ierr); > ierr = MatAXPYGetPreallocation_SeqAIJ(yy->B,xx->B,nnz_o);CHKERRQ(ierr); > ierr = MatMPIAIJSetPreallocation(B,0,nnz_d,0,nnz_o);CHKERRQ(ierr); > ierr = MatAXPY_BasicWithPreallocation(B,Y,a,X,str);CHKERRQ(ierr); > > > > -- > Regards, > Alexander > > -- 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

1 0

Re: [petsc-dev] MatAXPY with DIFFERENT_NONZERO_PATTERN
by Alexander Grayver 22 Apr '12

22 Apr '12

Sorry, I don't quite get. Is this something I can avoid by doing preallocation properly or petsc issue that will be fixed? On 20.04.2012 14:23, Jed Brown wrote: > On Fri, Apr 20, 2012 at 05:18, Matthew Knepley <knepley(a)gmail.com > <mailto:[email protected]>> wrote: > > Did someone intend to fix this, and then > stop? MatAXPY_BasicWithPreallocation() is the same as MatAXPY_Basic(). > > > Look how it's called. I don't know what's going on here. > > ierr = > MatAXPYGetPreallocation_SeqAIJ(yy->A,xx->A,nnz_d);CHKERRQ(ierr); > ierr = > MatAXPYGetPreallocation_SeqAIJ(yy->B,xx->B,nnz_o);CHKERRQ(ierr); > ierr = MatMPIAIJSetPreallocation(B,0,nnz_d,0,nnz_o);CHKERRQ(ierr); > ierr = MatAXPY_BasicWithPreallocation(B,Y,a,X,str);CHKERRQ(ierr); > -- Regards, Alexander

1 0