Re: [DuMux] Step post-processing in MPI mode

[Dmitry Pavlov]

Timo,

Thank you very much, that worked like a charm.

On another note, did you have a chance to look into MR 2914 and my 
message from October 28?


Best regards,

Dmitry


On 18.11.2021 17:19, Timo Koch wrote:

Hi Dmitry,

you want to only iterate over the internal cells to not sum up things 
twice on the process boundary, to this end use


for (const auto& element : elements(this->gridGeometry().gridView(), 
|Dune::Partitions::interior|))


After the loop you actually have to sum up over the process if you 
want the summed result, e.g. try


const auto& comm = this->gridGeometry().gridView().comm();
for (int i = 0; i < FluidSystem::numComponents; ++i)
fluxes[i] = comm.sum(fluxes[i]);

Best wishes
Timo


On 18. Nov 2021, at 13:40, Dmitry Pavlov > wrote:


Hello,

I have the following code to compute and add up fluxes through 
boundary cells of certain type after each step:



  Scalar fluxes[FluidSystem::numComponents] = {0};

  for (const auto& element : 
elements(this->gridGeometry().gridView()))

  {
  auto fvGeometry = localView(this->gridGeometry());
  fvGeometry.bindElement(element);

  auto elemVolVars = 
localView(gridVariables.curGridVolVars());

  elemVolVars.bind(element, fvGeometry, curSol);

  auto elemFluxVarsCache = 
localView(gridVariables.gridFluxVarsCache());


  for (auto&& scvf : scvfs(fvGeometry))
  {
  if (scvf.boundary())
  {
  const auto boundaryMarkerId = 
gridDataPtr_->getBoundaryDomainMarker(scvf.boundaryFlag());


  if (boundaryMarkerId == 888)
  {
  auto neumannFluxes = neumann(element, 
fvGeometry, elemVolVars, elemFluxVarsCache, scvf);
  neumannFluxes *= scvf.area() * 
elemVolVars[scvf.insideScvIdx()].extrusionFactor();


  for (int i = 0; i < 
FluidSystem::numComponents; i++)

  fluxes[i] += neumannFluxes[i];
  }
  }
  }
  }


It works in single-process mode but not in multi-process. I think 
this has to do with the grid being divided between processes? How do 
I adapt this post-processing code to this scheme?



Best regards,

Dmitry


___
DuMux mailing list
DuMux@listserv.uni-stuttgart.de 
https://listserv.uni-stuttgart.de/mailman/listinfo/dumux


___
DuMux mailing list
DuMux@listserv.uni-stuttgart.de
https://listserv.uni-stuttgart.de/mailman/listinfo/dumux

Re: [DuMux] Step post-processing in MPI mode

[Timo Koch]

Hi Dmitry,

you want to only iterate over the internal cells to not sum up things twice on 
the process boundary, to this end use

for (const auto& element : elements(this->gridGeometry().gridView(), 
Dune::Partitions::interior))

After the loop you actually have to sum up over the process if you want the 
summed result, e.g. try

const auto& comm = this->gridGeometry().gridView().comm();
for (int i = 0; i < FluidSystem::numComponents; ++i)
fluxes[i] = comm.sum(fluxes[i]);

Best wishes
Timo


> On 18. Nov 2021, at 13:40, Dmitry Pavlov  wrote:
> 
> Hello,
> 
> I have the following code to compute and add up fluxes through boundary cells 
> of certain type after each step:
> 
> 
>   Scalar fluxes[FluidSystem::numComponents] = {0};
> 
>   for (const auto& element : 
> elements(this->gridGeometry().gridView()))
>   {
>   auto fvGeometry = localView(this->gridGeometry());
>   fvGeometry.bindElement(element);
> 
>   auto elemVolVars = localView(gridVariables.curGridVolVars());
>   elemVolVars.bind(element, fvGeometry, curSol);
> 
>   auto elemFluxVarsCache = 
> localView(gridVariables.gridFluxVarsCache());
> 
>   for (auto&& scvf : scvfs(fvGeometry))
>   {
>   if (scvf.boundary())
>   {
>   const auto boundaryMarkerId = 
> gridDataPtr_->getBoundaryDomainMarker(scvf.boundaryFlag());
> 
>   if (boundaryMarkerId == 888)
>   {
>   auto neumannFluxes = neumann(element, fvGeometry, 
> elemVolVars, elemFluxVarsCache, scvf);
>   neumannFluxes *= scvf.area() * 
> elemVolVars[scvf.insideScvIdx()].extrusionFactor();
> 
>   for (int i = 0; i < FluidSystem::numComponents; i++)
>   fluxes[i] += neumannFluxes[i];
>   }
>   }
>   }
>   }
> 
> 
> It works in single-process mode but not in multi-process. I think this has to 
> do with the grid being divided between processes? How do I adapt this 
> post-processing code to this scheme?
> 
> 
> Best regards,
> 
> Dmitry
> 
> 
> ___
> DuMux mailing list
> DuMux@listserv.uni-stuttgart.de
> https://listserv.uni-stuttgart.de/mailman/listinfo/dumux

___
DuMux mailing list
DuMux@listserv.uni-stuttgart.de
https://listserv.uni-stuttgart.de/mailman/listinfo/dumux

[DuMux] Step post-processing in MPI mode

[Dmitry Pavlov]

Hello,

I have the following code to compute and add up fluxes through boundary 
cells of certain type after each step:



  Scalar fluxes[FluidSystem::numComponents] = {0};

  for (const auto& element : 
elements(this->gridGeometry().gridView()))

  {
  auto fvGeometry = localView(this->gridGeometry());
  fvGeometry.bindElement(element);

  auto elemVolVars = localView(gridVariables.curGridVolVars());
  elemVolVars.bind(element, fvGeometry, curSol);

  auto elemFluxVarsCache = 
localView(gridVariables.gridFluxVarsCache());


  for (auto&& scvf : scvfs(fvGeometry))
  {
  if (scvf.boundary())
  {
  const auto boundaryMarkerId = 
gridDataPtr_->getBoundaryDomainMarker(scvf.boundaryFlag());


  if (boundaryMarkerId == 888)
  {
  auto neumannFluxes = neumann(element, 
fvGeometry, elemVolVars, elemFluxVarsCache, scvf);
  neumannFluxes *= scvf.area() * 
elemVolVars[scvf.insideScvIdx()].extrusionFactor();


  for (int i = 0; i < 
FluidSystem::numComponents; i++)

  fluxes[i] += neumannFluxes[i];
  }
  }
  }
  }


It works in single-process mode but not in multi-process. I think this 
has to do with the grid being divided between processes? How do I adapt 
this post-processing code to this scheme?



Best regards,

Dmitry


___
DuMux mailing list
DuMux@listserv.uni-stuttgart.de
https://listserv.uni-stuttgart.de/mailman/listinfo/dumux