Cleaned out all but 3 omp statements. Might also have corrected a race condition in the space construction.


Former-commit-id: e0b9e6872f5388f09b94489c65869acc28e3eb7c

Autotools update

Fix problem configuring against HDF5 without CFLAGS and LDFLAGS set.

Fix problem building without MPI support.

Improve the installation documentation.

See merge request !27

Former-commit-id: ac0293bd397169e4e42eaad4e3163bd1624c6490

Former-commit-id: 21620a3f08de0f2e1cfd759802a2b367bbe83f6b

Former-commit-id: 109682c3f976c1ddafa3b4aa540ad07c082479d7

These are needed when the environment CFLAGS do not point at the HDF5 include directory. Use CPPFLAGS as CFLAGS is not set


Former-commit-id: cd1da96aeecf3e8abce21946d0b1168ae6159e53

Default gridfactors

Previous fix didn't compile when GCC is treating warnings as errors.
Warnings are not nice anyway so squash these and remove the temporary
variables (slight loss of clarity).

See merge request !26


Former-commit-id: da4a93ae77291f4143ebb9bdc40ff92743046a28

GCC doesn't like unused variables...


Former-commit-id: 7444d6297d58323b46843d39a0a4de7a3f35cd4a

Default gridfactors

Adds code to generate a default grid.

Constantly forgetting this myself, so consider if this is appropriate for the mainline.

See merge request !25


Former-commit-id: dc5e5946f421347abceee8e9a2e0d29d81b14e73

Result is OK, but may not be optimal


Former-commit-id: 657158a64354b32a2dd6fa95605cd4db164e1a4e

Previously --with-metis=<location> only picked up the library, not the headers, we now require the installation root, not the directory with the library


Former-commit-id: cad63fbc2562b553aba188801450de8d20d68fbd

Autotools update

Check that we have at least OpenMPI version 1.8.6 and warn if we do not.

See merge request !24


Former-commit-id: 5b04516684c7a6d4c89b88c4a453f198cff693b9

Former-commit-id: c8b7f31c1076d38184bdfcb19a77e7284cb53b4e

Autotools update

Generate the version.h file when it is not present, regardless of whether there is a git command
available or not. This can happen when a download of the gitlab is taken using the zip/tar facilities.

See merge request !23


Former-commit-id: 5251c17793262deed3b9dd5d746fdf090a69bf98

This can happen if a zip/tar of the source code is download directly from gitlab


Former-commit-id: 87b39ab3cf4d53a625511937b8bdda5c9b3a9cf1

mpi_fixes

Fixed broken scaling in task/resource graph in engine_repartition, was causing load imbalances across MPI nodes (see #34).

See merge request !22


Former-commit-id: 07836a868ad9b27b874c0a813029021562ea7521

Former-commit-id: b0776d4e3f2c4a6bbb463bb384b1c63b2c47fb3e

Mpi fixes

See merge request !21


Former-commit-id: 3fd9731a809d549a681492fee5e620b278320c1e

Former-commit-id: 5779ae9399763e919f6888d60921a7f95a11f67b

Former-commit-id: 6b8cc8d01d17b1455f7729ec9338a2ece94e168a

Big cosmo volume

Following issue #30, I have added a new script to generate arbitrarily large initial conditions.

The scripts downloads an initial tile and then replicates it along each axis multiple times and/or downsamples the initial particle load. The script takes two arguments:

python makeIC.py factor copies

where factor is the downsampling factor [0, 1] and copies is an integer giving the number of copies along each axis.

See merge request !20


Former-commit-id: 21a0e5d16d4324566e3331f92f9c04ec00b38a67

Added a new test case: A big cosmological volume. The script downloads an initial tile and replicates it multiple times. The script can also downsample the particles. The initial tile has 51M particles and is taken from the EAGLE simulation.


Former-commit-id: e52abd720300078c0bcead0e411bfdc1a903d218

Mpi fixes

Fix particle array overrun when exchanging strays.

See merge request !19


Former-commit-id: 19a5dd7186517b069c527cdd4ca67fdaea324073

don't just assume that we have enough space to accommodate for strays, re-allocate the parts array if needed.


Former-commit-id: 3dc666ba37ab16f0fab536586e1d50498d0bc73e

Autotools update

Add option to force serial IO when using a parallel HDF5.

See merge request !18


Former-commit-id: 842eeb1c2e4fafd0526640c4ed09acf47245071a

Mpi fixes

This should fix issue #28.

Tested it with the SodShock example, was able to reproduce the bug and catch the exact location using the address sanitizer. Have not yet tested it in a "real" parallel environment, e.g. a cosma run, but only on my laptop.

Could you test this with the CosmoVolume and verify that it indeed no longer crashes?

See merge request !17


Former-commit-id: 79fbeba71a8f2e5e6e3986f09302bc304880d947

fixed the segfaults on reshuffling, which were due to out-of-bounds particles in engine_redistribute. also made the scaling of the vertex and edge weights to prevent integer overflows more sane.


Former-commit-id: a365ccebe62978ca1f3f8d79f94d313bc0c73cfe

Former-commit-id: bc3ef3b278dc751e8d0ae3ab821dd499b5e156ad

Former-commit-id: 0cffbe34cf5f2c782442aebdeb4a007944ca851b

Clean up of the Doxygen documentation

Modified the Doxyfile to properly document the inline functions at the core of the algorithm.
Also corrected some typos in the documentation of functions that were mentioned in #15.

The doxygen documentation can now be built cleanly.

The changelog is massive due to the update of the Doxyfile to 1.8.x from 1.6.x. All the changes to the source files are in commented sections.

See merge request !15


Former-commit-id: 8a6a08f229f323265667e914c45b4b032ed9563f

Parallel HDF5 autoconf test

Add a HDF5 autoconf check

See merge request !16


Former-commit-id: 2a5e8f452c4bcc9a24535901a64a16999b466e56

Allow the code to do i/o in serial while MPI is activated

Quite a big change. Main goal is to solve issue #29 and hence offer a workaround #22.

The old non-MPI code is now in single_io.[ch] and is called when WITH_MPI is not defined.

The old MPI code with parallel-HDF5 is called when WITH_MPI and HAVE_PARALLEL_HDF5 are both defined.

The new MPI code is in serial_io.[ch] and is celled when WITH_MPI is defined but HAVE_PARALLEL_HDF5 is not.

In this serial mode, the code forces the different MPI ranks to write to the same file one after the other, appending their particles at the end of the pre-existing set of particles.

This merge request also corrects issue #27 and a bug in the attributes of the arrays in HDF5 that got the wrong units.

SWIFT now runs with Platform-MPI !!

See merge request !14


Former-commit-id: 8c6aaed9c5fbc30e205a7246838b7013a7be4bdc

Former-commit-id: 0a19e8d9e30d6159fd575e18aa64e63deb208292

Former-commit-id: 5b078c03983e09461e12729551919f2cc8faa404

Doing this to get the examples to work correctly.


Former-commit-id: def96923e54cdb82902539fd83bab6f86b9de69a

When building with MPI but serial HDF5


Former-commit-id: c14dc046667847891e309cfaf8260f3dff242cdb

Former-commit-id: b506d1bb917124e96c060792920ebfe0856d83f6

Modified the doxyfile such that the GCC __attribute__(blabla) statements get ignored and do no mess up the documentation.


Former-commit-id: a4c5d9832722d716a54901a10e6cf096b8b45923

Updated the Doxyfile and corrected the doxygen documentation tags that raised warnings. Build is now cleaner.


Former-commit-id: 5ca3a6d2612545e910dc0a74f3df5ed043c05ee9

Former-commit-id: eead9a984fdab0f5311ee9a9e4d89704ef042c16

The code can now write output in serial with MPI. Each rank writes its particle after the other in the same file.


Former-commit-id: 835922aa8ad8a8f961a238002c820d67f368a3bb