[petsc-users] combine single and double precision

2016-02-09 Thread albert gutiérrez
Hello everyone, At times it can be useful to combine different data types in the same problem. For example, there are parts of an algorithm which may be memory bounded (requiring float) and operations where more precision is needed (requiring double or float_128). Can single and double precision

Re: [petsc-users] PCFIELDSPLIT question

2016-02-09 Thread Hom Nath Gharti
Thanks a lot Matt! Were you referring to http://www.mcs.anl.gov/petsc/petsc-current/src/snes/examples/tutorials/ex62.c.html ? I do not see any statements related to PCFieldSplit there. Am I missing something here? Thanks, Hom Nath On Tue, Feb 9, 2016 at 10:19 AM, Matthew Knepley

Re: [petsc-users] combine single and double precision

2016-02-09 Thread Jed Brown
albert gutiérrez writes: > Hello everyone, > > At times it can be useful to combine different data types in the same > problem. For example, there are parts of an algorithm which may be memory > bounded (requiring float) and operations where more precision is needed >

Re: [petsc-users] PCFIELDSPLIT question

2016-02-09 Thread Hom Nath Gharti
Thank you so much Barry! For my small test case, with -pc_fieldsplit_block_size 4, the program runs, although the answer was not correct. At least now I get something to look upon. I am using PCFieldSplitSetIS to set the fields. Do I still need to use -pc_fieldsplit_block_size? In my case each

Re: [petsc-users] Different results with different distribution of rows

2016-02-09 Thread Matthew Knepley
On Tue, Feb 9, 2016 at 7:06 AM, Florian Lindner wrote: > Hello, > > I use PETSc with 4 MPI processes and I experience different results > when using different distribution of rows amoung ranks. The code looks > like that: > The default PC is BJacobi/ILU. This depends on the

Re: [petsc-users] Different results with different distribution of rows

2016-02-09 Thread Dave May
If you don't specify preconditioner via -pc_type XXX, the default being used is BJacobi-ILU. This preconditioner will yield different results on different numbers of MPI-processes, and will yield different results for a fixed number of different MPI-processes, but with a different matrix

Re: [petsc-users] PCFIELDSPLIT question

2016-02-09 Thread Matthew Knepley
On Tue, Feb 9, 2016 at 9:10 AM, Hom Nath Gharti wrote: > Thank you so much Barry! > > For my small test case, with -pc_fieldsplit_block_size 4, the program > runs, although the answer was not correct. At least now I get > something to look upon. I am using PCFieldSplitSetIS

[petsc-users] Different results with different distribution of rows

2016-02-09 Thread Florian Lindner
Hello, I use PETSc with 4 MPI processes and I experience different results when using different distribution of rows amoung ranks. The code looks like that: KSPSetOperators(_solver, _matrixC.matrix, _matrixC.matrix); // _solverRtol = 1e-9 KSPSetTolerances(_solver, _solverRtol, PETSC_DEFAULT,

Re: [petsc-users] Different results with different distribution of rows

2016-02-09 Thread Florian Lindner
Addition. The KSP Solver shows very different convergence: WRONG: [0] KSPConvergedDefault(): Linear solver has converged. Residual norm 6.832362172732e+06 is less than relative tolerance 1.e-09 times initial right hand side norm 6.934533099989e+15 at iteration 8447 RIGHT: [0]

Re: [petsc-users] PCFIELDSPLIT question

2016-02-09 Thread Matthew Knepley
On Tue, Feb 9, 2016 at 9:31 AM, Hom Nath Gharti wrote: > Thanks a lot Matt! > > Were you referring to > > http://www.mcs.anl.gov/petsc/petsc-current/src/snes/examples/tutorials/ex62.c.html > ? > > I do not see any statements related to PCFieldSplit there. Am I > missing

Re: [petsc-users] PCFIELDSPLIT question

2016-02-09 Thread Hom Nath Gharti
Sounds interesting! Thanks a lot Matt! I will have a look. Hom Nath On Tue, Feb 9, 2016 at 10:36 AM, Matthew Knepley wrote: > On Tue, Feb 9, 2016 at 9:31 AM, Hom Nath Gharti wrote: >> >> Thanks a lot Matt! >> >> Were you referring to >> >>