Releases: fbpic/fbpic
0.26.0
0.25.0
What's Changed
- Update format and dependencies of CI test by @RemiLehe in #646
- Fix CI by @ax3l in #583
- Add the option to fix the random seed by @RemiLehe in #599
- Avoid warning and use new interface in
test_boosted_particle_output.py
by @RemiLehe in #654 - Fix pyflakes error by @RemiLehe in #658
- Remove obsolete "python setup.py test" command by @RemiLehe in #657
- Correct positions in antenna emission by @RemiLehe in #656
- Fix version of numexpr to avoid error message in CI by @RemiLehe in #661
- Doc: Perlmutter CUDA 11.7 Update by @ax3l in #663
- Support for lasy files by @RemiLehe in #645
- Check the lasy version when reading lasy files by @RemiLehe in #665
- add info on mkl issue with apple silicon by @PrometheusPi in #662
New Contributors
- @ax3l made their first contribution in #583
- @PrometheusPi made their first contribution in #662
Full Changelog: 0.24.0...0.25.0
0.24.0
0.23.0
0.21.0
This release expands support for the PICMI interface, in particular for boosted-frame simulations. In addition, this release added the new optional argument boost_positions_in_dens_func
to the function add_new_species
.
This allows the pass the density function directly in the lab frame. (See 463)
0.20.4
0.20.3
This is a bug-fix release. It improves FBPIC's compatibility with recent versions of some of the software dependencies (namely numba
and mkl
).
-
The latest version of
numba
(numba 0.53.
) raised numerous warnings when running previous versions of FBPIC on GPU. These warnings do not appear anymore with this new release. (See #538) -
FBPIC can use the MKL distribution from
conda
, but was unable to use the MKL distribution frompip
. This is now fixed. (See #537)
0.20.2
0.20.1
0.20.0
This release introduces several improvements and bug fixes.
Better handling of multiple GPUs on clusters:
- The attribution of individual GPUs to the different MPI ranks is more robust, and avoids oversubscribing certain GPUs while leaving others idle (see #523). In addition, GPU oversubscription is now automatically detected, and a corresponding error message is printed. (see #524)
- Out-of-memory error are handled more properly, and will abort multi-GPU simulations whenever they are encountered by one of the GPUs. (see #521)
- The hostname is now printed even for single-rank simulation. (see #495)
- Kernel launch parameters have been fine-tuned for A100 architecture. (see #525)
Bug fixes:
- The external fields used to be applied to all species, instead of the species specified by the user. This is now fixed. (see #498)
- The installation instructions were updated. (see #515)
New features:
- Added a
Mirror
class to block the propagation of the laser e.g. in multi-stage simuations. (see #507)