Gary Macindoe

[git-lfs](https://git-lfs.github.com/) may be of interest here. It allows you to host large files somewhere and version them within a git repository.

I have added code (in [81b1441ad085](http://pauli.chem.ucl.ac.uk/trac/hemelb/changeset/3420/)) to enable instrumentation after 5 cells have been added and stop it after a further 1000 iterations have passed. In order not to break...

From analysing the callgrind output I believe the most optimisation can be had from refactoring `hemelb::redblood::DivideConquerCells::pair_range::nextDist()` in `CellCell.cc`. This has 13.82% of the call time with just over 1% in...

Profile run for 1000 iterations with 3 cells: https://gist.github.com/garymacindoe/f92df705a9504a3b6f5f

The source of the calls to `free` are from `std::vector`s declared in `Interpolation.h` on lines 130, 132 and 134. The `InterpolationIterator` that they are part of is used in a...

As expected, replacing the runtime stencil stuff with compile-time constants removed all the calls to `free` and resulted in a 10% speed increase (benchmark now starts profiling at ~1350s instead...

I have changed the stencil type to be a template parameter of `CellController` with the default specified in `CMakeLists.txt`. This allows hemelb to be built to use a particular stencil...

The `stencil` XML attribute is completely ignored as the stencil is now set by CMake. I will remove the code that reads the stencil from XML.

Python has an [H5MD implementation](https://github.com/pdebuyl/pyh5md) that seems to have a sane design (there is a [C implementation](https://github.com/pdebuyl/ch5md) by the same author which doesn't). It wraps the pyh5 library which in...

I'm personally leaning towards using NetCDF, possibly with the new C++ API. All formats seem to use HDF5 underneath or at least something very close to it. These use a...