Add global reduction for "magic heating" #31
Conversation
|
When manually running the regression tests on HPCC, they all pass. Waiting on an MPI build for the automated CI runner. |
|
@pgrete I can reproduce the weird values using the |
src/hydro/hydro.cpp
Outdated
| // global reduction of a vector | ||
| const int numHist = pin->GetOrAddInteger("precipitator", "numHist", 8); | ||
| if (parthenon::Globals::my_rank == 0) { | ||
| std::cout << "Using numHist = " << numHist << std::endl; | ||
| } | ||
| parthenon::AllReduce<PinnedArray1D<Real>> profile_reduce; | ||
| profile_reduce.val = PinnedArray1D<Real>("Reduce me", numHist); | ||
| pkg->AddParam("profile_reduce", profile_reduce, true); | ||
|
|
||
| PinnedArray1D<Real> profile_reduce_zbins("Bin centers", numHist); | ||
| const Real x3min = pin->GetReal("parthenon/mesh", "x3min"); | ||
| const Real x3max = pin->GetReal("parthenon/mesh", "x3max"); | ||
| const Real dz_hist = (x3max - x3min) / numHist; | ||
| for (int i = 0; i < numHist; ++i) { | ||
| profile_reduce_zbins(i) = dz_hist * (Real(i) + 0.5) + x3min; | ||
| } | ||
| pkg->AddParam("profile_reduce_zbins", profile_reduce_zbins, false); | ||
|
|
There was a problem hiding this comment.
Given that this is (currently) still a "single problem generator" related feature, I suggest to move it to that pgen (this also applies to the code below in the hydro_driver where I suggest to use a problem related callback function for the reduction).
If you don't object I'm happy to move stuff around.
There was a problem hiding this comment.
I don't object, but I don't think I fully understand -- do you mean create a new callback function interface for this?
| const cooling::TabularCooling &tabular_cooling = | ||
| pkg->Param<cooling::TabularCooling>("tabular_cooling"); |
There was a problem hiding this comment.
If this function will only work for tabular cooling, I'd move the logic upwards (i.e., where you already check for none) and make sure it's only working in that case (and otherwise return/fail).
I'm thinking of future us' here who might implemented other cooling methods and assume that this works silently.
Co-authored-by: Philipp Grete <pgrete@hs.uni-hamburg.de>
Co-authored-by: Philipp Grete <pgrete@hs.uni-hamburg.de>
|
Have you tried updating the tasks as discussed on Matrix? What's the input file you used for the render above? Is that the cloud one? |
Ah yes, it's the cloud input file. |
|
Ok, so the output issue exists also on the main branch, at least for the cloud problem. |
|
With the changes you suggested on the Matrix chat, the outputs for test problems run with this branch look fine in yt. |
|
The MPI tests appear to be failing due to a memory issue. Might be fixed by parthenon-hpc-lab/parthenon#841. |
|
I've found a cleaner way to implement this that doesn't require changing the hydro driver, so I'm closing this PR for now. |
This adds a global reduction in the hydro driver to compute a 1D vertical profile of the cooling rate.
Due to the way reductions are implemented in Parthenon, this requires adding tasks in hydro driver. However, this is somewhat of a specialized application, so I'm not sure how best to handle this special case. Feedback is welcome!