On Mon, Oct 27, 2014 at 3:02 PM, Filippo Leonardi <
filippo.leona...@sam.math.ethz.ch> wrote:
> On Monday 27 October 2014 14:31:45 Mark Adams wrote:
> >
> > I'm not sure what you are asking. starting i=0 or i=1, and with and
> > without mg. So four possi
>
> Yeah, that should work; an alternative is to defer the estimate until
> the next solve.
>
Sounds better. As I said one of my users has a zero RHS on the first solve.
and run in debug mode if possible.
> But the problem seems to still be there, also tried:
> -ksp_chebyshev_estimate_eigenvalues_random
> but does not change.
>
> Just tell me what test you need and I'll try to do them.
>
> Best,
> Filippo
>
> On Wednesday 29 October 2
; On Tuesday 04 November 2014 12.21:29 Mark Adams wrote:
> > On Tue, Nov 4, 2014 at 2:55 AM, Filippo Leonardi <
> >
> > filippo.leona...@sam.math.ethz.ch> wrote:
> > > Sorry for the late update.
> > >
> > > I tested the code with the branch:
It does not look like you are in fact using the branch with this fix. The
branch code looks like:
if (n) PetscValidScalarPointer(r,3);
*and*
if (n) PetscValidScalarPointer(r,4);
So if you now put the check for argument 4 inside of your if statement then
you will have reimplemented my branch,
I have a code that repartitions the grid using PETSc. I would like to know
the simplest instructions that I can give my users (Pat Worley in this
case) to verify that ParMetis is actually called. I had difficulty with
this my self in that -info did not seem to say ParMetis was being used. I
did
rdering, or 2 for parallel analysis and ictnl(29) ordering (None)
> -mat_mumps_icntl_29 <0>: ICNTL(29): parallel ordering 1 = ptscotch 2 =
> parmetis (None)
>
> Hong
>
> On Mon, Nov 17, 2014 at 9:12 AM, Mark Adams wrote:
>
>> I have a code that repartitions the grid using PETSc
Ah yes, -mat_partitioning_view,
Thanks,
Mark
On Mon, Nov 17, 2014 at 11:42 AM, Barry Smith wrote:
>
> MatPartitioningView() ?
>
>
> > On Nov 17, 2014, at 9:12 AM, Mark Adams wrote:
> >
> > I have a code that repartitions the grid using PETSc. I would like to
&g
Richardson damping factor does not seem to do anything. I set it to zero
and the solve was fine.
Am I missing something?
Thanks,
Mark
Down solver (pre-smoother) on level 3 ---
KSP Object:(mg_levels_3_) 1024 MPI processes
type: richardson
R
Oh, SOR probably grabs the iteration ...
Never mind.
On Fri, Dec 19, 2014 at 10:20 AM, Mark Adams wrote:
>
> Richardson damping factor does not seem to do anything. I set it to zero
> and the solve was fine.
>
> Am I missing something?
>
> Thanks,
> Mark
>
>
On Sat, Dec 27, 2014 at 5:00 AM, Alp Kalpalp wrote:
> Hi,
>
> I implemented a newmark time stepping algorithm without using TS
> structure. I am following ex59 about PCBDDC.
>
>
use TS
> ComputeMatrix
> ComputeKSPBDDC
>
> for // a time loop
> {
> Compute RHS
> KSPSolve
> GatherResults
> MoveToN
On Thu, Jan 1, 2015 at 12:20 AM, TAY wee-beng wrote:
> Hi,
>
> I used to run my CFD code with 96 procs, with a grid size of 231 x 461 x
> 368.
>
> I used MPI and partition my grid in the z direction. Hence with 96 procs
> (8 nodes, each 12 procs), each procs has a size of 231 x 461 x 3 or 231 x
>
On Sun, Dec 28, 2014 at 12:24 PM, Alp Kalpalp wrote:
> Thanks for the answers,
>
> Please forgive me, I forgot to say that my stiffness matrix is not
> changing during time steps. I could not remember directly but just after a
> google search..I just hit this
>
> http://web.stanford.edu/group/frg
I am advancing a two equation system with TS that has an additional
constraint equation. I build a 3x3 composite matrix a_ts%FJacobean that
has my F(U). I then do:
call
MatDuplicate(a_ts%FJacobean,MAT_DO_NOT_COPY_VALUES,a_ts%FJacobean2,ierr)
call
TSSetIJacobian(a_ts%ts,a_ts%FJacobean2,a_ts%F
Sorry, I was getting "composite" from DMCompositeGetAccessArray, I am
talking about MatNest. I can get the matrices with MatNestSetSubMats
>
> > ! copy in linear operator
> > call MatCopy(a_ts%FJacobean,Jpre,ierr);CHKERRQ(ierr)
> > ! shift 1 & 2 by 'shift'
> > call MatShift(mat00,shif
Not sure what is going on. It stopped coarsening for some reason. Does
the code set the number of levels at 2? Could you try remove the
-pc_gamg_coarse_eq_limit
200 ... maybe it is gett junk. Also add -pc_gamg_verbose 2. That might
give me a clue (probably not).
Mark
On Wed, Feb 18, 2015 at
On Thu, Feb 19, 2015 at 12:45 PM, Hong wrote:
> Fabian,
> Too much time was spent on the matrix operations during setup phase, which
> has plenty room for optimization.
>
Yea, a little slow but not crazy. With 2M equation on one proc you may be
having cache problems.
> Can you provide us a
>
>
> >
> > The RAP is often more than 50% of PCSetUp, so this might not save much.
>
>Hee,hee in some of my recent runs of GAMG the RAP was less than 25% of
> the time. Thus skipping the other portions could really pay off.*
>
>
This percentage is very problem dependant. 3D elasticity is hig
It sounds like AMG is used on the whole system. I've done this for a
different saddle point system but it required enough algorithm extension
that I wrote a paper on it. These methods are not in GAMG.
I would at least not use smoothed aggregation, use plane aggregation
(-pc_gamg_agg_nsmooths 0).
FYI, stüben used classical AMG for elasticity but he has articulated his
code for elasticity more than Hypre as I understand it. Hypre can work OK
for elasticity in my experience. Its worth a try.
Mark
On Thu, Mar 5, 2015 at 5:27 PM, David Knezevic
wrote:
> OK, got it, thanks!
>
> David
>
>
>
ried fare poorly. I guess it's because the
> highly contrasting stiffnesses give a large condition number. Is there a
> standard way to get the preconditioner compensate for this?
>
>
> Thanks,
> David
>
>
>
> On Sat, Mar 7, 2015 at 5:57 PM, Mark Adams wrote:
&g
.
On Sun, Mar 8, 2015 at 11:38 AM, David Knezevic
wrote:
> On Sun, Mar 8, 2015 at 10:31 AM, Mark Adams wrote:
>
>>
>>
>>
>>> Another situation I've been meaning to ask about: If the elasticity
>>> model includes some "near rigid" regions
I declare
PCCompositeType::pcctype
But PCFieldSplitGetType does not seem to like it?
Mark
[1]PETSC ERROR: - Error Message
--
[1]PETSC ERROR: Invalid pointer
[1]PETSC ERROR: Invalid Pointer to Int: Parameter # 2
[1
) ));
}
On Fri, Apr 17, 2015 at 4:12 PM, Barry Smith wrote:
>
>Likely need to run in the debugger to see what generates the error.
> Perhaps a missing ierr?
>
> Barry
>
>
> > On Apr 17, 2015, at 9:10 AM, Mark Adams wrote:
> >
> > I declare
> &g
I am getting this error on Titan with Intel programming env.
I run this from an interactive batch shell to avoid the batch business.
This workflow seemed to work for pgi & gnu programming envs.
Any ideas?
Thanks,
Mark
configure.log
Description: Binary data
wrote:
> On Fri, 8 May 2015, Mark Adams wrote:
>
> > I am getting this error on Titan with Intel programming env.
> >
> > I run this from an interactive batch shell to avoid the batch business.
> >
> > This workflow seemed to work for pgi & gnu programmi
I added the LD path stuff and that seems to have fixed this problem ... but
there are still problems.
On Fri, May 8, 2015 at 1:23 PM, David Trebotich wrote:
> This has all come up because I cannot build Chombo-crunch using gnu
> prgenv---cannot find mkl anymore. So olcf consultants advised Inte
ldn't be the problem]
>
Yea, I tried this workflow with PGI and it worked. I'm giving up on Intel
and see if we can make GNU work for Treb.
Thanks,
>
> Satish
>
> On Fri, 8 May 2015, Mark Adams wrote:
>
> > I added the LD path stuff and that seems to have fixed thi
I think I did a super clean build... I will try again.
On Mon, May 11, 2015 at 5:04 PM, Matthew Knepley wrote:
> On Mon, May 11, 2015 at 6:55 PM, Mark Adams wrote:
>
>> More Titan fun. I have built for PGI on Titan but tests are failing to
>> build: -lhwloc?
>>
>&
I did a fresh build again. No luck. Is this how I turn it off?
'--download-hwloc=0',
On Mon, May 11, 2015 at 5:14 PM, Matthew Knepley wrote:
> On Mon, May 11, 2015 at 7:06 PM, Mark Adams wrote:
>
>> I think I did a super clean build... I will try again.
>&g
>
> Notes: ./configure ran fine and detected -lhwloc in some standard system
> install location under normal circumstances it couldn't just disappear for
> a different example.
>
>
I configured in an interactive shell, so on a compute node. I tried to
'make ex56' on a login node, as usual. So I
I have setup the equations attached using TSSetIJacobian and
TSSetIFunction.
This is intended to be a fully implicit solvers with some auxiliary
variables.
We want to verify the code with the current forward Euler solver. Are
there command line that would give me a solver that solves the two
au
of the time step, whereas the forward Euler (or what the existing
code does) is computed with the state at the beginning of the time step.
This may not be the best way to verify this code ... Thanks again.
On Wed, May 27, 2015 at 9:41 PM, Jed Brown wrote:
> Mark Adams writes:
>
> >
:43 PM, Matthew Knepley wrote:
> >
> > Does bfort have to be told that PCCompositeType is an enum?
> >
> >Matt "bfort's biggest fan"
> >
> > On Fri, Apr 17, 2015 at 4:37 PM, Mark Adams wrote:
> > It looks like fieldsplitf.c has an extr
> >
> > > > FYI, this does not seem to be fixed.
> > > >
> > > > On Sat, Apr 18, 2015 at 5:49 AM, Barry Smith
> wrote:
> > > >
> > > > Looks like a bunch of these things are missing from
> > > lib/petsc/conf/bfort-
Well I used a fresh petsc clone and that worked.
On Fri, May 29, 2015 at 1:05 PM, Mark Adams wrote:
> Humm, I cleaned everything, and even cut and pasted your configure line,
> same thing. I am now trying with a fresh petsc clone.
>
> I add these modules:
>
> modul
Eduardo, as Matt said you problem is ill conditioned but and you might find
that if you add more elements and make the mesh less anisotropic that your
solve faster, and you get a better solution obviously.
I'm not sure what options there are for better discretization but you can
probably do a lot
On Tue, Jun 2, 2015 at 6:00 PM, Jed Brown wrote:
> Eduardo writes:
>
> > I am solving a FEM solid mechanics linear elasticity model, for now the
> > only problem is the mesh that has needle-shaped and very flat elements.
>
> Why does it have such elements? Is the material highly anisotropic
> (
>
>
> Mark, how do you turn on heavy-edge matching?
>
>
-mat_coarsen_type hem
Don't try this yet. Try the parameters as Matt suggested.
> Thanks,
>
> Matt
>
>
>> Thanks,
>> Justin
>>
>
>
>
> --
> What most experimenters take for granted before they begin their
> experiments is infin
On Thu, Jun 4, 2015 at 12:29 PM, Matthew Knepley wrote:
> On Thu, Jun 4, 2015 at 10:31 AM, Justin Chang wrote:
>
>> Yeah I saw his recommendation and am trying it out. But I am not sure
>> what most of those parameters mean. For instance:
>>
>> 1) What does -pc_gamg_agg_nsmooths refer to?
>>
>
>
greatest amount of
> workload and number of calls. My only hesitation is how much traffic
> MatMatMults may take (assuming I go off of the same assumptions as in that
> paper).
>
> 6) And/or, are there any other functions that I missed that might be
> important to calculate as well?
>
>
>>
> The overwhleming cost of AMG is the Galerkin triple-product RAP.
>
>
That is overstating it a bit. It can be if you have a hard 3D operator and
coarsening slowly is best.
Rule of thumb is you spend 50% time is the solver and 50% in the setup,
which is often mostly RAP (in 3D, 2D is much
-
>
>--
> *From:* petsc-users-boun...@mcs.anl.gov [petsc-users-boun...@mcs.anl.gov]
> on behalf of Justin Chang [jychan...@gmail.com]
> *Sent:* Saturday, June 06, 2015 5:29 AM
> *To:* Mark Adams
> *Cc:* petsc-user
are multiplied by kappa.
>>
>>
>>
>> --
>> Matthew Young
>> Graduate Student
>> Boston University Dept. of Astronomy
>> --
>>
>>--
>> *From:* Matthew Knepley [knep...@gmail.com]
>&
t; Matthew Young
> Graduate Student
> Boston University Dept. of Astronomy
> --
>
>--
> *From:* Mark Adams [mfad...@lbl.gov]
> *Sent:* Wednesday, June 10, 2015 10:33 AM
> *To:* Young, Matthew, Adam
> *Subject:* Re: GAMG
>
> The manual has material
gt; --
> Matthew Young
> Graduate Student
> Boston University Dept. of Astronomy
> ------
>
>
>
> From: Jed Brown [j...@j
GAMG does not know about this explicitly and apparently it is not getting
picked up my PCMG correctly (probably my fault).
I don't understand what this is supposed to do. I would not worry about
it, but should GAMG care about this?
Mark
On Thu, Jun 11, 2015 at 7:48 AM, David Scott wrote:
> He
Sorry for the delay. The problem is that the eigen estimates are bad for
the Chebyshev smoother. Unfortunately this fails catastrophically.
First, I do not understand why we do not get an error message here:
13:57 PICell> ~/Codes/petsc/src/ksp/ksp/examples/tutorials$ mpirun -n 8
./ex54 -ne 1023
end
> anymore.
>
> Is there something else to change apart from the option
> -mg_levels_esteig_ksp_type cg ?
>
> Thanks,
> Benoit
>
>
>
> On Wed 08 Jul 2015 03:31:32 PM CEST, Mark Adams wrote:
>
>> Sorry for the delay. The problem is that the eigen estimates are ba
Do we have any data with respect to the number of PETSc users? Better yet
number of GAMG users?
Mark
simple solver like change:
-pressure_pc_type gamg
to
-pressure_pc_type jacobi
And add:
*-*pressure_*ksp_monitor_true_residual*
Mark
-- Forwarded message --
From: Greg Miller
Date: Mon, Jul 13, 2015 at 2:08 PM
Subject: same petsc problem
To: Mark Adams
Cc: David Trebotich
Hi
then it is an
overflow problem.
-- Forwarded message --
From: Greg Miller
Date: Mon, Jul 13, 2015 at 3:54 PM
Subject: Re: Fwd: same petsc problem
To: Mark Adams
Cc: "t...@lbl.gov"
thanks. here it is.
I'm not getting the nan problem with this code - not sure
Note, you want to tell the matrix that it has these blocks with
MatSetBlockSize, and do not remove a single DOF (like for BCs). That is,
xyzxyxyz is not allowed. The block size is critical for GAMG and would
think HYPRE uses it also because it does OK on elasticity.
On Wed, Jul 15, 2015 at 2:52 P
Where is this list of (known) apps?
Thanks,
On Tue, Jul 14, 2015 at 6:09 PM, Jed Brown wrote:
> Mark Adams writes:
>
> > Do we have any data with respect to the number of PETSc users? Better
> yet
> > number of GAMG users?
>
> Mailing list statistics, downstream sof
u should be able to set that block size to
3. If this get submatrix happens inside of FieldSplit then I think there
is a solution for this.
>
>
> Sander
>
> On Thu, Jul 16, 2015 at 12:26 AM, Mark Adams wrote:
>
>> Note, you want to tell the matrix that it has these block
understood,
On Wed, Jul 15, 2015 at 9:25 PM, Paul T. Bauman wrote:
>
>
> On Wed, Jul 15, 2015 at 9:23 PM, Paul T. Bauman
> wrote:
>
>> (Shame on me, but is there a relevant GAMG publication?)
>>
>
> Arg, this came off bad, sorry. I meant, is there a particular publication
> for which GAMG shoul
>
>
> Just count the number of email threads that have GAMG in the text and
> give that number to David Brown :-)
>
>
138 since last January. I have seriously considered this!
his into the web page when it is accepted.
>
> On Wed, Jul 15, 2015 at 8:32 PM, Barry Smith wrote:
>
>>
>> > On Jul 15, 2015, at 6:53 PM, Mark Adams wrote:
>> >
>> > Where is this list of (known) apps?
>>
>> http://www.mcs.anl.gov/petsc/publi
>
>
>
> Yeah, gamg tries to put all the rows of the matrix on one process and
> leave the other processes empty,
Actually, GAMG does not do that anymore. But there are two parameters that
govern this and the defaults result in one active process on the coarse
grid: -pc_gamg_process_eq_limit
On Thu, Jul 16, 2015 at 8:18 PM, Michele Rosso wrote:
> Barry,
>
> thank you very much for the detailed answer. I tried what you suggested
> and it works.
> So far I tried on a small system but the final goal is to use it for very
> large runs. How does PCGAMG compares to PCMG as far as perf
I have a 4x4 block matrix, in one big AIJ matrix. I use (global) ISs to
address individual fields. I have a mass matrix for one field. I would
like to add a scaled mass matrix to one field at a time.
MatGetLocalSubMatrix looks like what I want but it takes local indices. My
ISs are global. It
This was fixed on June 18 (Stephen Kramer):
> Can you please test this and see if works, without your fix?
>
> 'git fetch; git checkout mark/gamg-crs',
>
Great! That fixes everything for me: it uses KSP+PREONLY for mg_coarse by
default again and I can also override it with -mg_coarse_ksp_type
On Thu, Jul 16, 2015 at 5:53 PM, Michele Rosso wrote:
> Hi,
>
> I am performing a series of solves inside a loop. The matrix for each
> solve changes but not enough to justify a rebuilt of the PC at each solve.
> Therefore I am using KSPSetReusePreconditioner to avoid rebuilding unless
> necess
Coming in late to this thread but you are doing frequency domain NV.
Start by getting your time domain (definite, no omega shift) solves
working. This can be a challenge for NV. There are techiques for this but
we do not have them. Start with plane aggregation (-pc_gamg_nsmooths 0),
this should
take some time for me, but I’ll be back…
>
>
>
> What does NV abbreviate?
>
Naviers ... typing too fast
>
>
> Mahir
>
>
>
>
>
> *From:* Mark Adams [mailto:mfad...@lbl.gov]
> *Sent:* den 26 juli 2015 15:41
> *To:* Jed Brown
> *Cc:* Ülker-Kaustell,
Hopper at NERSC does configure with '--with-mpiexec=aprun' (but Edison
does). I do not see a mpiexec.aprun anymore in PETSc. Is there an easy
fix for this?
Mark
never mind, make MPIEXEC=aprun works.
On Wed, Jul 29, 2015 at 8:59 AM, Mark Adams wrote:
> Hopper at NERSC does configure with '--with-mpiexec=aprun' (but Edison
> does). I do not see a mpiexec.aprun anymore in PETSc. Is there an easy
> fix for this?
> Mark
>
>
i Mark and Barry,>>>> >>>> I am sorry for
> > my late reply: it was a busy week!>>>> I run a test case for a larger
> > problem with as many levels (i.e. 5) of MG I could and GAMG as PC at the
> > coarse level. I attached the output of info
On Tue, Aug 11, 2015 at 3:25 PM, Matthew Knepley wrote:
> On Tue, Aug 11, 2015 at 2:10 PM, Mani Chandra wrote:
>
>> Hi,
>>
>> Is there an example where Petsc's SNES has been used with Chombo, and
>> perhaps with an automatic Jacobian assembly? I'd like to know if Petsc can
>> pick out the number
On Wed, Aug 12, 2015 at 2:29 PM, Mani Chandra wrote:
> Hi,
>
>
>> Chombo (me) creates an MPIAIJ matrix. So automatic Jacobian assembly
>> should work.
>>
>> I have put a SNES in a Chombo code, but did not use automatic Jacobian
>> assembly.
>>
>
> Do you have an example?
>
If you want to make a
R solve, to a AIJ matrix.
I don't think you want to look at DMs.
Mark
> just not sure it would work with Chombo. But I'll give it a shot.
>
> Cheers,
> Mani
>
> On Wed, Aug 12, 2015 at 5:56 PM, Mark Adams wrote:
>
>>
>>
>> On
Has this been fixed in latter releases? Would removing -options_left fix
this?
[0]PETSC ERROR: No support for this operation for this object type!
[0]PETSC ERROR: Only supports ASCII viewer!
[0]PETSC ERROR:
[0]PETSC ERROR: P
the way we
build or something. So our user is happy now. Lets just hope this problem
goes back to where it came from.
On Fri, Oct 25, 2013 at 8:31 PM, Satish Balay wrote:
> I can't reproduce this.
>
> Is this reproduceable with the PETSc examples with your build?
>
> Satish
This sounds like a hypre issues. Euclid/Hypre is not use much AFAIK so it
might have some problems that have not been addressed.
On Tue, Nov 19, 2013 at 12:11 PM, Manav Bhatia wrote:
> Hi,
>
> I am using ilu(k) preconditioner from Euclid/Hypre on my Mac OS 10.9
> box with 12 processors and
'-pc_gamg_verbose 2' will print out stuff where you can see the number of
active processors. We have not been super motivated to use subcoms because
there are no norms taken on coarse grids, except in the eigen estimator for
cheby, unless you use krylov in the solver, which is not usually useful.
>
>
>
> [0]PETSC ERROR: Petsc has generated inconsistent data!
> [0]PETSC ERROR: Unhandled case, must have at least two fields, not 1
This can happen if you give fieldsplit options to a non-fieldsplit
matrix/PC. So so calling XXXSetFromOptions with a KSP/PC and an AIJ Mat
(ie, non-nested Mat) an
ile=0x2ea3a08
"src/ksp/pc/impls/fieldsplit/fieldsplit.c", n=77, p=PETSC_ERROR_INITIAL,
mess=0x2ea4090 "Unhandled case, must have at least two fields, not %d") at
/global/u2/m/madams/petsc_private/src/sys/error/err.c:353
On Sat, Nov 30, 2013 at 6:33 PM, Jed Brown wrote:
> Mark Adams wr
>
> > else ! 1 field 00 solver, no prefix
> > call PetscOptionsClearValue('-ksp_monitor',ierr)
> > call PetscOptionsClearValue('-ksp_converged_reason',ierr)
>
> Yuck. The options table is for the (run-time) user. Manually futzing
> with options, especially clearing options, is a dirty
>
>
>
> if (i == 0) {
> for (f = 0; f < numFields; ++f) {
> ierr = PCFieldSplitSetIS(pc, fieldNames[f],
> fields[f]);CHKERRQ(ierr);
> ierr = PetscFree(fieldNames[f]);CHKERRQ(ierr);
> ierr = ISDestroy(&fields[f]);CHKERRQ(ierr);
> }
>
> numFields sh
>
>
> So this is a different KSP; not the one you want to use fieldsplit on?
>
Yes,
>
> Please always use different prefixes for different solvers.
>
We use prefixes. I left them out of the example. So things like
'-s2_pc_type fieldsplit' cause an error. The 's2_' solver is a one field
(Lapl
>
>
> So when you say to use fieldsplit for a solve with only one field, you
> get an error saying that fieldsplit requires at least two fields?
>
>
Yes
> 1. Don't do that.
>
>
My point was that this user got this message and so I suggesting a possible
problem: giving fieldsplit parameters to a s
We have a lot of 5-point stencil operators on ~50x100 grids to solve.
These are not symmetric and we have been using LU. We want to move this
onto GPUs (Titan). What resources are there to do this?
Mark
On Thu, Dec 12, 2013 at 4:38 PM, Ed D'Azevedo wrote:
> Hi Karli,
>
> Yes, each MPI process is responsible for solving a system of nonlinear
> equations on a number of grid cells.
>
Just to elaborate, and Ed can correct me, each MPI process has a few 100 to
a few 1000 (spacial) cells. We solve a
This thread has lost the main developer Ed ... cc'ed along with the PI.
Ed and CS, I will forward a few messages on this thread.
On Thu, Dec 12, 2013 at 6:29 PM, Dominic Meiser wrote:
> Hi Karli,
>
>
> On 12/12/2013 02:50 PM, Karl Rupp wrote:
>
>
> Hmm, this does not sound like something I wo
Try running prometheus with -out_verbose 2
and gamg with -pc_gamg_verbose 2
and send me the output.
Mark
On Thu, Jan 30, 2014 at 4:40 AM, Thomas Gross wrote:
> Please find enclosed the output for GAMG using -mg_levels_ksp_max_it 1:
>
> Prometheus:
> -ksp_type cg -pc_type prometheus -log_summar
And is this just a simple cube problem in elasticity as it looks like from
your input file name?
On Thu, Jan 30, 2014 at 4:40 AM, Thomas Gross wrote:
> Please find enclosed the output for GAMG using -mg_levels_ksp_max_it 1:
>
> Prometheus:
> -ksp_type cg -pc_type prometheus -log_summary -ksp_mo
>
>
>
> We're seeing a bit lower performance in MatMult with GAMG, perhaps
> because we are not using block formats specialized for elasticity.
>
>
The block info seems to be there.
> Mark, what else is different?
* Prometheus seems to be coarsening slower and taking nearly twice the
iterations
Also try running with -pc_gamg_repartition true
>
>
>
> I meant that GAMG isn't using BAIJ, but Prometheus is. That affects
> kernel efficiency.
>
>
Ah yes, that must be it.
2: Module 'PrgEnv-pgi' is currently not
> loaded
> ModuleCmd_Switch.c(172):ERROR:152: Module 'PrgEnv-pgi' is currently not
> loaded
> ModuleCmd_Switch.c(172):ERROR:152: Module 'PrgEnv-pgi' is currently not
> loaded
> ModuleCmd_Switch.c(172):ERROR:152:
I fixed this by not downloading MPI. I used to not need to specify MPI at
all. But I seemed to when I use 'maint'.
On Sun, Mar 23, 2014 at 8:46 AM, Mark Adams wrote:
> Yes, I was told:
>
> Comments:
> 2014-03-21 14:44:49 - Harvey Wasserman (Additional comments)
> 1.
I get this error in the middle of a GMRES iteration. Any ideas?
0 SNES Function norm 6.067786990073e+17
Residual norms for fsa_ solve.
0 KSP Residual norm 6.067786990073e+17
1 KSP Residual norm 6.037192129591e+17
2 KSP Residual norm 5.852390318109e+17
3 KSP Residua
I have this code:
#if PETSC_VERSION_GT(3,4,0) || !PETSC_VERSION_RELEASE
ierr = KSPSetOperators(m_ksp,m_mat,m_mat);CHKERRQ(ierr);
#else
ierr =
KSPSetOperators(m_ksp,m_mat,m_mat,SAME_NONZERO_PATTERN);CHKERRQ(ierr);
#endif
And it fails with a v3.4 at TACC (eg /opt/apps/intel13/mvapich2_1
>
>
> Are you sure it is not 3.4.x?
>
No. But I do not see a 3.4.1.
Perhaps I should change
#if PETSC_VERSION_GT(3,4,0) || !PETSC_VERSION_RELEASE
to
#if PETSC_VERSION_GE(3,5,0) || !PETSC_VERSION_RELEASE
This interface change will not go into a release until 3.5 right?
Sorry for getting to this late. I think you have figured it out basically
but there are a few things:
1) You must set the block size of A (bs=2) for the null spaces to work and
for aggregation MG to work properly. SA-AMG really does not make sense
unless you work at the vertex level, for which we
Stephan, I have pushed a pull request to fix this but for now you can just
use -mg_levels_ksp_type chebyshev -mg_levels_pc_type jacobi. This used to
be the default be we move to SOR recently.
Mark
On Sat, Mar 29, 2014 at 5:52 PM, Mark Adams wrote:
> Sorry for getting to this late. I th
hy can't singletons be elided completely? Then you would have a zero
> row of P.
> View this pull
> request<https://bitbucket.org/petsc/petsc/pull-request/161/gamg-shift-coarse-grids-to-avoid-zero#comment-1559358>or
> add a comment by replying to this email.
> In reply t
If the solver is degrading as the coefficients change, and I would assume
get more nasty, you can try deleting the solver at each time step. This
will be about 2x more expensive, because it does the setup each solve, but
it might fix your problem.
You also might try:
-pc_type hypre
-pc_hypre_typ
solution and report back.
> Again, thank you.
>
> Michele
>
>
> On 05/22/2014 09:34 AM, Mark Adams wrote:
>
> If the solver is degrading as the coefficients change, and I would assume
> get more nasty, you can try deleting the solver at each time step. This
> wi
>
>
> Mark, if PCReset (via KSPReset) does not produce the same behavior as
> destroying the KSP and recreating it, it is a bug. I think this is the
> case, but if it's not, it needs to be fixed.
>
Yes it should work. I've just never used it.
1 - 100 of 1032 matches
Mail list logo