Highlights of the new preCICE release v2.4

Just out: Release v2.4.0 · precice/precice · GitHub

As usual, we fixed quite a few bugs. We mention an important fix below. This alone should already be incentive enough to upgrade :grin: … but, of course, we also have a few new handy features:

New export variants

The export functionality now supports many more data formats:

The csv data, you can for example import to any Python script with pandas: :eyes:

def loadParallelCSV(name):
  import glob, pandas
  return pandas.concat([pandas.read_csv(name, sep=";") for name in glob.glob(f"{name}_*.csv")], ignore_index=True)

def loadParallelCSVSeries(name)
  import re, glob, pandas
  l = [(re.search("dt(\d+)_", s).group(1), s) for s in glob.glob(f"{name}.dt*_*.csv")]
  retrun pandas.concat([pandas.read_csv(file, sep=";").assign(dt=dt) for dt, file in l], ignore_index=True)

pointData       = loadParallelCSV("A-ExporterTwo.dt1")
pointDataSeries = loadParallelCSVSeries("A-ExporterTwo")

Introducing precice-tools

We renamed binprecice to precice-tools (the old name is still usable till preCICE v3.0) and added some additional functionality:

Building

We added a few small details that should make your and our lives easier:

  • New CMake options PRECICE_RELEASE_WITH_DEBUG_LOG, PRECICE_RELEASE_WITH_ASSERTIONS, and PRECICE_RELEASE_WITH_TRACE_LOG to switch on Debug logging, assertions, and Trace logging, respectively when in Release mode (all CMake options). This is useful for debugging preCICE with larger test cases, especially when dealing with connectivity :rocket:.
  • A pkg-config file for using preCICE directly from the build directory. (Add pkg-config file for binary directory by fsimonis · Pull Request #1238 · precice/precice · GitHub) – very handy for adapter developers who rely on pkg-config and develop versions of preCICE. Simply add the build directory to the PKG_CONFIG_PATH environment variable after setting up LD_LIBRARY_PATH.

We also added tooling shortcuts to CMake, which simplify contributing:

  • make doxygen: Generate doxygen documentation
  • make format: Format the codebase (requires local install of the correct version of clang-format)
  • make soucesIndex: Update the CMake source index
  • make changelog: Add changelog entry using the GitHub PR number

Dependencies

With the new Ubuntu LTS release 22.04, we drop the support of the oldest LTS, namely Ubuntu 18.04 LTS, making Ubuntu 20.04 LTS the new baseline. This means, preCICE now requires Boost version 1.71.0 and CMake version 3.16.3 (Lift baseline from Ubuntu 18.04 LTS to 20.04 LTS by fsimonis · Pull Request #1259 · precice/precice · GitHub). We are now actively investigating a possible move to C++17.

Nearest-neighbor gradient mapping (still experimental)

We added a new mapping method nearest-neighbor-gradient (Added API Methods for writing the gradient data needed for Nearest Neighbor Gradient Mapping by ariguiba · Pull Request #1169 · precice/precice · GitHub), which was implemented, tested, and documented as part of the bachelor’s thesis of @ariguiba. Users can now give gradients of data to preCICE and get a second-order mapping method this way.

writeScalarGradientData(int dataID, int valueIndex, const double* gradientValues)

Compared to the nearest-projection mapping, no connectivity is required (but gradient data). First tests with ASTE showed very promising results. See all new API methods and a full code example.

Time interpolation (still experimental)

Time interpolation is a topic we are working on since longer, see for example the paper on quasi-Newton waveform iteration from last year:

Rüth, B., Uekermann, B., Mehl, M., Birken, P., Monge, A. and Bungartz, H.J., 2021. Quasi‐Newton waveform iteration for partitioned surface‐coupled multiphysics applications. International Journal for Numerical Methods in Engineering, 122(19), pp.5236-5257. https://doi.org/10.1002/nme.6443

We now merged and released a first larger chunk of functionality. We now have linear interpolation in time, but still restricted to parallel-implicit coupling.

Switch it on:

<solver-interface experimental="true" ... >
...
    <participant name="FluidSolver">
        <use-mesh name="FluidMesh" provide="yes"/>
        <write-data name="Forces" mesh="MyMesh"/>
        <read-data name="Displacements" mesh="FluidMesh" waveform-order="1"/>
    </participant>
...
</solver-interface>

… and then read data whereever (in time) you want:

void readBlockVectorData(int dataID, int size, const int* valueIndices, 
                         double relativeReadTime, double* values) const;

Everything is still experimental – use with care. To find out more and stay up to date, have a look at the full documentation

Julia bindings

Probably worth mentioning, we recently also added bindings for Julia: GitHub - precice/PreCICE.jl: Julia language bindings for preCICE :smile: . We also added them to the Julia General Package Registry so that they should be easy to install. To test the bindings, you could then run the example solverdummy code.

Bug in socket communication

We finally discovered the root course of a rare and difficult to reproduce bug in the sockets communication back-end. A race condition in the send queue led to crashes in send-intensive situations. This was observed in cases that explicitly enable the gather-scatter algorithm in communications using: <m2n:sockets enforce-gather-scatter="true" ... />. Furthermore, this could be observable in uni-directional coupling, especially if the sending participant finishes time-windows faster than the receiving participant.

Even more

We definitely forgot something in this list, but anyway, enough reasons to upgrade right now! :smile:

7 Likes

Well, I am happy to read about new changes that are being introduced in the latest version.

2 Likes