[Users] Problem with CarpetRegrid2/AMR

Tue Sep 13 11:00:08 CDT 2011

Ian,

I'd not found the problem yet. Hopefully this will be the hint we need
to get this fixed.

Thanks again,
Hal

On Tue, 2011-09-13 at 15:31 +0100, Ian Hawke wrote:
> Erik, Hal,
> 
> Did you have any luck tracking down this error? I've just come back to 
> this and am seeing the same error message; it appears to arise when two 
> grids on a refined level merge, as in:
> 
>     [3][0][0]   exterior: [-0.010000,-0.020000,-0.020000] : 
> [0.000000,0.020000,0.020000] : [0.001250,0.001250,0.001250]
>     [3][0][1]   exterior: [0.010000,-0.020000,-0.020000] : 
> [0.040000,0.020000,0.020000] : [0.001250,0.001250,0.001250]
> 
> becomes
> 
>   [3][0][0]   exterior: [-0.010000,-0.020000,-0.020000] : 
> [0.040000,0.020000,0.020000] : [0.001250,0.001250,0.001250]
> 
> It seems that the old grids are destroyed before the data is 
> copied/populated in Recompose (either that or the old grid structure is 
> not referred to in the data transfer).
> 
> Ian
> 
> On 07/09/11 01:03, Erik Schnetter wrote:
> > Hal
> >
> > This is where numbers are assigned to components. The communication
> > schedule decides which component needs to send data to which other
> > component (which may be located on another process or not); this
> > schedule is created for each refinement level independently, and may
> > (if there is an error) refer to component numbers that don't exist.
> > This schedule is set up in dh.cc.
> >
> > Can you send me the example you are currently running (your source
> > code and parameter file)? I will try to give it a try.
> >
> > -erik
> >
> > On Tue, Sep 6, 2011 at 7:44 PM, Hal Finkel<hfinkel at anl.gov>  wrote:
> >> On Thu, 2011-09-01 at 15:16 -0400, Erik Schnetter wrote:
> >>> On Thu, Sep 1, 2011 at 3:05 PM, Hal Finkel<hfinkel at anl.gov>  wrote:
> >>>> On Thu, 2011-09-01 at 14:25 -0400, Erik Schnetter wrote:
> >>>>> On Thu, Sep 1, 2011 at 11:51 AM, Hal Finkel<hfinkel at anl.gov>  wrote:
> >>>>>> On Thu, 2011-09-01 at 11:37 -0400, Erik Schnetter wrote:
> >>>>>>> On Thu, Sep 1, 2011 at 10:53 AM, Hal Finkel<hfinkel at anl.gov>  wrote:
> >>>>>>>> On Tue, 2011-08-30 at 21:06 -0400, Erik Schnetter wrote:
> >>>>>>>>> On Tue, Aug 30, 2011 at 5:28 PM, Hal Finkel<hfinkel at anl.gov>  wrote:
> >>>>>>>>>> Could I also decrease the block size? I currently have
> >>>>>>>>>> CarpetRegrid2::adaptive_block_size = 4, could it be smaller than that?
> >>>>>>>>>> Is there a restriction based on the number of ghost points?
> >>>>>>>>> Yes, you can reduce the block size. I assume that both the regridding
> >>>>>>>>> operation and the time evolution will become slower if you do that,
> >>>>>>>>> because more blocks will have to be handled.
> >>>>>>>> Regardless of what I do, once we get past the first coarse time step,
> >>>>>>>> the program seems to "hang" at "INFO (Carpet): [ml=0][rl=0][m=0][tl=0]
> >>>>>>>> Regridding map 0...".
> >>>>>>>>
> >>>>>>>> Overall, it is in dh::regrid(do_init=true). It spends most of its time
> >>>>>>>> in bboxset<int, 3>::normalize() and, specifically, mostly in the loop:
> >>>>>>>> for (typename bset::iterator nsi = nbs.begin(); nsi != nbs.end(); ++
> >>>>>>>> nsi). The normalize() function does exit, however, so it is not hanging
> >>>>>>>> in that function.
> >>>>>>>>
> >>>>>>>> The core problem seems to be that it takes a long time to execute:
> >>>>>>>> boxes  = boxes .shift(-dir) - boxes;
> >>>>>>>> in dh::regrid(do_init=true). Probably because boxes has 129064 elements.
> >>>>>>>> The coarse grid is now only 30^3 and I've left the regrid box size at 4.
> >>>>>>>> I'd think, then, that the coarse grid should have a maximum of 30^3/4^3
> >>>>>>>> ~ 420 refinement regions.
> >>>>>>>>
> >>>>>>>> What is the best way to figure out what is going on?
> >>>>>>> Hal
> >>>>>>>
> >>>>>>> Yes, this function is very slow. I did not expect it to be
> >>>>>>> prohibitively slow. Are you compiling with optimisation enabled?
> >>>>>> I've tried with optimizations enabled (and without for debugging).
> >>>>>>
> >>>>>>> The bboxset represents the set of refined regions, and it is
> >>>>>>> internally represented as a list of bboxes (regions). Carpet performs
> >>>>>>> set operations on these (intersection, union, complement, etc.) to
> >>>>>>> determine the communication schedule, i.e. which ghost zones of which
> >>>>>>> bbox need to be filled from which other bbox. Unfortunately, the
> >>>>>>> algorithm used for this is O(n^2) in the number of refined regions,
> >>>>>>> and set operations when implemented via lists themselves are O(n^2) in
> >>>>>>> the set size, leading to a rather unfortunate overall complexity. The
> >>>>>>> only cure is to reduce the number of bboxes (make them larger) and to
> >>>>>>> regrid fewer times.
> >>>>>> This is what I suspected, but nevertheless, is there something wrong?
> >>>>>> How many boxes do you expect that I should have? The reason that it does
> >>>>>> not finish, even with optimizations, is that there are 129K boxes in the
> >>>>>> loop (that's at least 16 billion box normalizations?).
> >>>>>>
> >>>>>> The coarse grid is only 30^3, and the regrid box size is 4, so at
> >>>>>> maximum, there should be ~400 level one boxes. Even if some of those
> >>>>>> have level 2 boxes, I don't understand how there could be 129K boxes.
> >>>>> The refinement structure itself should have one bbox per refined 4^3
> >>>>> box, and both CarpetRegrid2 and CarpetLib would try to combine these
> >>>>> into fewer boxes where possible, i.e. where one can form rectangles or
> >>>>> larger cubes. I would thus expect no more than (30/4)^2 = 64 bboxes on
> >>>>> level one.
> >>>> That makes sense. I think that there is a bug somewhere which is causing
> >>>> the box set to be much too big. Furthermore, it does not happen on every
> >>>> run, only sometimes. When it does not happen, I hit another bug after a
> >>>> few coarse timesteps:
> >>>>
> >>>> I get a range-check exception from std::vector in a call to:
> >>>> gh::get_local_component (rl=1, c=8)
> >>>> the problem is that this returns:
> >>>> local_components_.AT(rl).AT(c);
> >>>> and local_components_[1].size() is 8
> >>>> The call to get_local_component is coming from ggf::transfer_from_all
> >>>> at:
> >>>> int const lc2 = h.get_local_component(rl2,c2);
> >>>> where c2 is from psend.component.
> >>>> So it looks like there is an off-by-one error somewhere.
> >>> Very strange. This code should be quite solid by now. psend is set in
> >>> the file dh.cc in thorn Carpet/CarpetLib; there is one (large) routine
> >>> that calculates the communication schedule. Some of the indexing
> >>> errors there in the past included confusing the number of components
> >>> on different refinement levels, which led to indexing errors such as
> >>> the one you describe.
> >> The bad component numbers are not coming from:
> >> preg.component = tmpncomps.AT(m)++;
> >> in Carpet/src/Recompose.cc
> >>
> >> Where else are the component numbers assigned?
> 
> _______________________________________________
> Users mailing list
> Users at einsteintoolkit.org
> http://lists.einsteintoolkit.org/mailman/listinfo/users

-- 
Hal Finkel
Postdoctoral Appointee
Leadership Computing Facility
Argonne National Laboratory
1-630-252-0023
hfinkel at anl.gov