Skip to content

Structure of the code (aka where is everything?)

K Clough edited this page Nov 15, 2021 · 23 revisions

Yes, we know, GRChombo is a big code. At first the number of files will seem overwhelming, but with time you will start to learn where to find things and the structure will make (some) sense.

On this page we provide some hints on how to find your way around the code, but in the end you just have to dive in and learn as you go.

A useful pdf guide on this topic (with nicer pictures) from the latest GRChombo training day can be found in Useful resources. One should also look at the guides on C++ classes, inheritance and templating, which are used extensively in the code - some basic knowledge of these concepts is assumed below.

Hierarchy of GRChombo

The code is designed to have 3 main levels in its hierarchy, as follows:

  1. Specific Example related files, e.g. for BinaryBH - specific actions relevant to the BinaryBH example - key classes include BHAMR, BinaryBHLevel, SimulationParameters. Also important are the namespaces UserVariables and DiagnosticVariables in the BinaryBH examples folder and BinaryBH (the initial data).

    The functions that are specified at this level include things like setting initial data, calculating example-specific diagnostics, reading in example-specific parameters, specifying the tagging criterion.

This inherits most of the functionality from:

  1. GRChombo - specific physics actions common to most 
 GR problems - key classes include GRAMR, GRAMRLevel, and SimulationParametersBase
. See also the CCZ4UserVariables namespace (and most of the contents of the GRChombo Source folders).

    The functions that are specified at this level include things like performing the RHS calculation for CCZ4 and matter variables, calculating constraints, calculating the finite derivatives, checking for Nans, setting up and reading in the CCZ4 parameters.

This in turn inherits most of the functionality from:

  1. Chombo - overall program flow relevant to any hyperbolic initial value problem with AMR - key classes: AMR, AMRLevel, ChomboParameters
. (See also most of the contents of the src/AMRTimeDependent folder.)

    The functions that are specified at this level include things like setting up the initial AMR hierarchy and performing the AMR regridding and performing the Runge-Kutta update.

Where to find the files

Logically, all of the files related to level 1:BinaryBH should be in the specific Example folder Examples/BinaryBH. However, there are a few exceptions:

  • The initial data class BinaryBH is considered sufficiently general (ie, it will be used in many examples without modification) to be included in the Source code of GRChombo rather than in the Example folder itself, so it is in Source/InitialConditions/BlackHoles. For matter classes it is more likely that you will want to put the initial condition code in the Example folder itself, as it is more likely to be problem specific (as in InitialScalarData ).

  • Similarly many functions related to evolving the black holes, and tracking their punctures, are generally useful enough to be in Source. See in particular the Source/BlackHoles folder.

  • We often include tagging criteria in the Source code as they are reusable in many examples and provide a useful library of examples. See the Source/TaggingCriteria folder.

Logically, all of the files related to level 2:GRChombo should be in the GRChombo/Source folder. However, there are exceptions:

  • Some of the Example specific code is in here too if it is sufficiently general in use, as discussed above.
  • Some functions that logically belong in Chombo are here too, as discussed below.

Logically, all of the files related to level 3:Chombo should be in the Chombo/src folder. However, there are exceptions:

  • The SetupFunctions and ChomboParameters classes are included in Source/GRChomboCore as we have made some changes to the Chombo setup to suit our applications.

  • The AMRInterpolator (see Extraction and integration) is a functionality that logically belongs to the Chombo code, but it didn't exist, so we wrote it and it lives in GRChombo/Source/AMRInterpolator.

  • The BoundaryConditions code is a functionality that logically belongs to the Chombo code, but it didn't exist in the form we wanted, so we wrote our own and it lives in GRChombo/Source/GRChomboCore.

Note that the main Chombo file is AMR.cpp, which controls the overall program flow. You don't really need to understand this file - you can just trust it will use the functions you give it in your Example Level file at the right moment, but if you want to really understand what is going on it is worth taking a quick look. If your question is "when and why is my code doing this step?" the answer probably lies in here somewhere.

Hooks and virtual functions

Each part of the Chombo/GRChombo/BinaryBH Hierarchy has some awareness of the part above and below it. This is provided by, for example, AMRLevel providing a virtual function that does nothing, but gets overwritten in GRAMRLevel. Functions may also provide hooks for certain actions, again defined via overwriting virtual functions, at certain points.

Example:

Go to AMR.cpp and look for the calls to the postTimeStep function on each level, i.e. m_amrlevels[ilev]->postTimeStep(). Notice how they happen in a loop over all levels ilev of the hierarchy. This is used to tell each Level to do something after its final update at each timestep (ie, after synchronisation).

Now go and find the the AMRLevel::posttimestep() function in Chombo's AMRLevel.H. It does nothing.

Now look at GRAMRLevel and find its postTimeStep function. This overwrites the virtual function in AMRLevel, so now the function does some coarse to fine averaging, and enforces the boundary conditions. It also contains a call to another virtual function specificPostTimeStep(), which is again just a blank virtual function - find it in GRAMRLevel.hpp.

Now look at BinaryBHLevel. You should find the function BinaryBHLevel::specificPostTimeStep() here, which overwrites the virtual one in GRAMRLevel and does some calculations and outputs. Now you know where it will be called in the whole AMR process, and in what order in the levels.

This kind of exciting detective work is often required for finding the connections between functions. The command grep is your friend here.

A note on AMR (GRAMR/BHAMR) versus AMRLevel (GRAMRLevel/BinaryBHLevel)

This is a key point which most users fail to grasp initially, and even experienced users mess up. It is always worth some extra thinking time, and probably also some outputting to check what is going on matches what you meant to do. (Don't ever feel ashamed to add a line pout() << "I am here doing X on level " << m_level << endl; to the code.)

AMR controls the program flow for the entire hierarchy - it knows that, for example, 6 levels of refinement exist, with the coarsest level having a certain value of dx, etc.

AMRLevel is then a class that exists for each of these six levels. So there are 6 copies of this that get called in turn, in an order that is determined by the AMR class. Any instructions in an AMRLevel class will happen on each level in turn, and won't affect the other levels unless you explicitly ask them to talk to each other. Usually they only know and can access data on the levels above and below them, but they can appeal to the AMR class for wider control (this is required to use the AMRInterpolator, for example).

So, for example, if you write in the postTimeStep() function a command to write out "hello world", in the pout files, you will get this output on every level of the hierarchy, after each of its timesteps. This will be a lot of output.

If instead you want something to only happen on the coarsest level, you will need to bracket it by an if statement that requires `if(m_level == 0){}'. An example of this is something like writing out a diagnostic. You don't want this to happen 6 times.

If you want something to happen on every level but only on every coarsest timestep, you need to have an if statement that checks the timestep is a multiple of the coarsest one, i.e. if(at_level_timestep_multiple(0)) as in the BinaryBHLevel. An example of this is something like calculating the values of a diagnostic variable across the whole grid (ie, on every level) for output only on the coarsest timestep. It would of course not crash the code to call this on every level timestep, but it would be a waste of computing time since it would only be output once, whereas on the 6th level there will be 2^6 timesteps (and therefore 2^6 calculating of the diagnostic) in between the output, which only relies on the final calculation of the values.

Clone this wiki locally