GitLab

When a morton index isn't mapped on any mpi process it is now mapped on
the last one

Uniform Cube, Ellipsoid and Plummer can be generated (default is Uniform
Cube).

Add define to use file or not in testBlockedMpiChebyshev.cpp testBlockedMpiUniform.cpp and testFmmAlgorithmThreadProc.cpp

Add a fix to put -1 instead of 0 when there is no communication on many node.

Add correct number of thread.

Fix job bug in semple_mpi.sh. It was moving the entire directory because STARPU_FXT_PREFIX was empty

Add information in the wrap for each different plot (add bsize)

Script to generate only gantt chart.
Add speed plot (not speedup).

uniform kernel.

Now it take the idx of the block instead of the first Morton index.
To avoid a possible overflow with tree higher than 8.

Replace handle type. It was Variable type, and i switch it to vector
type to do a partition on it.

Add RANDOM_PARTICULE in testBlockedUniform.

Before you'd have to split the total amount of particles between the
number of node ...

Now there is one script per algorithm and kernel.
The number of node and particles are handle before.
Plus, job scripts load envirronement from a script on plafrim.

Add gantt and communications volumes : R scripting and python parsing.
Add start_profiling and stop_profiling in python script.
Add calls to gen_comm and gen_gantt in generate_graph.R.

export tarball. Add check for time in python script. Small change in the
timer in the tests (not that much).