Add a synchronization point at the end of engine_dump_restarts() to make sure everything has been written before proceeding and risking a crash

Use a different function of the ParMETIS API to avoiding caching of the first AC_CHECK_LIB call

See merge request !1004

Sphenix pressure floor

See merge request !1002

Fixes for checks on optimized AXV2 architectures

See merge request !997

Align the memory allocation in the cbrt() unit test to prevent FPEs in the peel and remainder loops.

Nifty clutser solution

See merge request !915

Fix an undefined behaviour problem

Closes #646

See merge request !1000

Fix an undefined behaviour problem in engine_split_particles() by using the more expensive (but correct) functions to relink the gpart to other particles.

Backport a fix from the COLIBRE repo to solve a segfault in the cleaning routines when ending a run that restarted from check-point files.

Time-step limiter and time-step synchronization over MPI

See merge request !999

Take care with structure alignments, we should allocate cells with posix_memalign and when passing aligned structs, like runner, do this by pointer not value, not sure how aligned memory will work on the stack

Was crashing on COSMA6, so AVX not AVX2 this time

Note part of this squashes the report 'note: The ABI for passing parameters with 64-byte alignment has changed in GCC 4.6' when calling test_boundary_conditions()

Avoid overflow problems with the number of waits/unlocks of tasks. Now that we go beyong 64bits for tasks it makes no sense to worry about memory footprint for these anymore.

Logger: small fix

See merge request !998

Add -march=core-avx2 for AMD CPUs compiled using ICC

See merge request !993

From http://www.prace-ri.eu/IMG/pdf/Best-Practice-Guide-AMD.pdf, note we add these to CFLAGS as the icc_flags is a choice vector

intel/2018

Logger: add doc

See merge request !971

Sphenix SPH

See merge request !994

Speeds up self-gravity x2

Random: add seed parameter

See merge request !988