Erebus

This release mainly fixed the unbound numerical accurancy bugs by setting a maximum number of teams and threads. For more information see here

Some bugs where also fixed in the command line arguments logics and we simplified the almost_equal implementation

Etna

-Add support for min, max` reduction (enable in Tier 2). Thanks to @naromero77!
-Fix code generation for math function inside a SIMD region (enable in Tier 2)
-Fix some bash issue regarding Argument parsing

Piton de la Fournaise

Version 1.0!
920 test in tiers 1, 1928 tests in tiers 2 and a few hundred thousand in tiers 3.
Added support for multi-device, and solved a few bugs.

Not expecting major change now, maybe some code refactoring.

Maybe in the futur:

• nowait clause for host-threaded,
• get_max_teams, get_max_thread for better floating point presion in non-balenced case.
• More Fortran Math Function

For sure in the futur:

• Follow the evolution of the specs

Last Before Stable (hopefully)

Globally happy with the test generation, the tiers list, and the printing summary and the quality of internal testing.

Things to potentially do before version 1.0:

• Not really happy with the internal naming convention but I don't know if I can do better. How to name 'omp loop', 'omp for do loop', and do, for.
• Missing 2008 Fortran Math functions (https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gfortran/Intrinsic-Procedures.html#Intrinsic-Procedures).
• Missing nested parallelisms (https://software.intel.com/content/www/us/en/develop/articles/exploit-nested-parallelism-with-openmp-tasking-model.html)

PR welcome!

Introduction of Tiers

• A lot of rewrite on the source code who generate and parse the tests. It should be more maintainable now.
• Add Tiers lists. See readme.
• If nobody complains, or after addressing complain, we will release v1.0

First Version

Legacy version.