News
New members:
- Panagiotis (NTUA) who will work remotely on GPU acceleration
- Together with a new technical student to arrive
New users (Simon):
- Diamond-II light source in UK
- Trying BLonD for beam-loading + high-harm. cavity + almost full machine
- Usecase for partial histograms and multi-turn wakes
Performance tests object (Alex)
- New object to collect runtime results over x runs
- Comparison of functions/implementations
- Fitting profiles: gather measured profiles for all machines as example, e.g. curve_fit() and minimize()
Global parameters (Markus)
- Consistent handling of parameters for distribution and fitting, see BLonD_common/tree/fitting_overhaul
- e.g. bunch length -> different definitions, consistent with getters and setters
- later: think what else could be put (interpolation etc.)
- function to check kwargs?
- Distribution object (line distribution): different calling signatures implemented (Gaussian and binomial so far)
- just an object
- in: time_array -> out: data_array
- in: data_array -> out: fit performed
- 2D distribution to be implemented (based on action and Hamiltonian)
- Alex: propagate structure to all other distributions
- Kostis: kwargs -> check whether some parameters are not consumed
- Nested functions can use a subset but check on the highest-level fucntion
Profiling & BLonD-mpi (Kostis)
- MPI version has more and more users
- Started to look at integration to master branch
- No errors for sequential running => should be easy to implement
- git-prompt: small repo for git usage in cluster
- More informative output on console
- E.g. to see a branch, and whether files are commited
- Profiling: pyprof module
- timing -> runtime of regions
- papiprof -> HW details
- mpiprof -> mpi code profiling
- perfdeluxe -> architectural events reported
- Support for python3.5 dropped
- now testing python3.8 -> incompatibilities...
- C++ code coverage: can be reported, but needs to be set up
New implementation of drift equation (Alex)
- PSB test with alpha_0, alpha_1, alpha_2
- Benchmark: non-linear terms improved with new implementation
- Unittests: comparison_drift.py (different implementations) and test_drift.py (py vs cpp)
- Using assert_allclose to evaluate relative and absolute tolerances
- Tracker options: simple/legacy/new available
- Pull request launched
There are minutes attached to this event.
Show them.