Developer Meeting: 2016-12-07

[ax3l]

date: 2016-12-07, 1:10pm-2:30pm style: scrum (3'-5' each) protocol: Richard Pausch (@PrometheusPi)

Attendees

• HZDR: Axel Huebl (organizer), René Widera, Alexander Matthes, Marco Garten, Heiko Burau, Alexander Debus, Richard Pausch, Fabian Koller
• GSI/HI Jena (remote via WebRTC): Bifeng Lei, Stefan Tietze

Introduction

all attendees, each 1min

Progress in Topics

Axel Huebl (@ax3l)

• additions to meeting topics not yet sent in? -> none
• release PIConGPU 0.2.1 is out -> beta status still expect changes in input files in the upcoming versions for input API, read CHANGELOG.md
• we will try to generate our change log automatically next time by parsing the merged pull-requests: all developers should take time to use meaningful titles and descriptions in pull requests
  • physics oriented: e.g., "what does a fixed PMacc bug affect for a PIConGPU user?"
  • title: precise and not too general: e.g., "ParticlePusher: Fix C++ Warning in Boris", "PMacc reduce: fix summation (not used before)"
  • description: detailed as always, add if a bug affects the user-land results (vice versa: also add if it does not and why)
• openPMD (adios and hdf5) in PIConGPU adds a large collection of post-processsing tools, see openPMD project
• mallocMC -> no in-house developer anymore
  • needs continuation
  • extend to multi-core CPUs

Alexander Matthes (@theZiz)

• Resource management: we will replace our transaction system which overlaps computation & communication
• Problem: currently programmed by hand, hard to extend, plugins do not use event system, ...
• Solution: All "algorithmic parts" have to declare what resources they will access
• Userland: just write public typedefs / constexpr which resources a functor will access

Marco Garten (@n01r)

• will implement Thomas-Fermi ionization model for collisional LTE ionization
• based on Thomas Fermi atom model
• requires ion density and temperature -> thus two fields (FieldTmp: one scalar field)
• model already implemented but still missing both fields at the same time
• useful for over-dense targets

Axel Huebl (@ax3l)

• collisional ionization model outside the LTE (local thermal equilibrium) conditions
• based on SCFLY
  • needs energy and density grid
  • solve time dependent rate equation
• combine 0D SCIFLY code with 3D plasma simulation (PIConGPU)
• Marco Garten (@n01r) will later add radiation transport

Heiko Burau (@Heikman)

• Strong fields effect electron dynamics (QED)
• Master's thesis: https://doi.org/10.5281/zenodo.192116
• non-linear Compton scattering
  • can replace Liénard-Wiechert radiation plugin (because it is faster)
  • if a_0>10 spectra are fine
  • if a_0<10 one gets a smeared out spectra because there is no interference
  • only works in synchrotron approximation
• Bremsstrahlung (pull request #1504)
  • still not merged to dev
  • might provide useful synthetic diagnostic for electron dynamics (in over-dense targets)

Alexander Debus (@BeyondEspresso)

• extend laser implementation
  • experiments need laser that deviate from simple Gauss laser
  • include Gauss Laguerre laser (rotational symmetric) (pull request #1580)
  • this extension should be continued for a longer time till we have a flexible laser infrastructure
• possible issues with boundary conditions

Richard Pausch (@PrometheusPi)

• form factor extended (pull requests #1641 #1671)
• Projects: clean up radiation plugin (as an example of projects)

Marco Garten (@n01r)

• ParaTAXIS (developed by Alexander Grund (@Flamefire))
• Marco took it over for his PhD
• right now only offline tool
• use 3D electron density and scatter X-ray photons using a Monte-Carlo method
  • photon/x-ray laser implementation
  • actual interaction: model photon scattering
  • accumulate photons (in-phase) to emulate a detector
• fixes orientation in laser read-in from SIMEX
• goal: include in PIConGPU to model pump probe experiments

René Widera (@psychocoderHPC)

• alpaka -> not only run on GPUs anymore (PICon... everything!)
• initial author: Benjamin Worpitz (@BenjaminW3, Master's thesis: https://doi.org/10.5281/zenodo.49768)
• René and Erik Zenker (@erikzenker) -> use alpaka with PIConGPU
• papers: DOI:10.1109/IPDPSW.2016.50, DOI:10.1007/978-3-319-46079-6_21
• currently picongpu-alpaka is a prototyping repo with core PIC algorithm
• now: will port piece wise to PIConGPU
  • will require change in build process
  • not all plugins will be available on CPU from start (but in the end)
• cupla: wrapper on top of alpaka to ease the use of alpaka

Axel Huebl (@ax3l)

• "how do we work on GitHub" (only effects developers)
• due to breaking changes in the upcoming weeks please contact René or Axel to know when best to start your development (to avoid conflicts with updates)

Bifeng Lei

• Question: Is it possible to simulate external injection with PIConGPU?
• Axel: Main problem is to get the field self-consistently - there are methods to do this for cold bunches in PIConGPU (see Diploma thesis of Axel, https://doi.org/10.5281/zenodo.15924) but they do not work out of the box/mainline right now
• Axel: an other way is to accelerate from a resting bunch: we can show you how to set this up in detail

René Widera (@psychocoderHPC)

• parallel iterators (how to move over particles)
• currently particles are grouped in super cells and it is not easily possible to get just particles in a specific cell
• needed for particle-particle interaction
• student working on that (Sebastian Hahn)
• re-factoring of that work is going on

Axel Huebl (@ax3l)

• particle merging algorithms needed for ionization and photons

Fabian Koller (@C0nsultant)

• develops C++ (& Python) API that speaks openPMD
• to replace/extend libSplash (e.g., 2.0?)
• make it easier to read & write openPMD compatible output with C++
• HDF5/ADIOS/JSON/netCDF/... agnostic
• also improve usability of libSplash
• API design: writing soon ready, reading will take a little longer

Axel Huebl (@ax3l)

• in order to ask questions: open an issue on GitHub - we will notify the appropriate developer (alternative: write to the mailing list)
• external people can visit us to get started as users or developers as Stefan Tietze (@stetie) did
• next meeting in 1 month: 2017-01-11 13:00 (CET) - email will be send a week in advance

post-meeting reminder:

René Widera (@psychocoderHPC) & Axel Huebl (@ax3l)

• deprecated coding styles, replace with C++11
  • BOOST_PP
  • #ifdef in PIConGPU code
  • put const on the right, read left to right: const int -> int const, const int * const -> int const * const
  • above will be enforced in reviews
  • official style (not enforced until tools available): https://github.com/ComputationalRadiationPhysics/contributing
  • rewrite traits with float numbers to constexpr
    • @BeyondEspresso: when will constexpr float std::cos() work? C++14?
    • @n01r: can we replace PMACC_CONST_VECTOR? constexpr int[3] = {1, 2, 3};?

[n01r]

I added a Wiki entry for an ongoing list of developer meeting minutes that refers to this issue.

[ax3l]

ok, but we added a flag for it anyway and I am not a big fan of workflows manually updating things at various places. I edited your wiki entry to link to all meeting minutes.

[n01r]

Oh, well, a far better solution if it is already automated. ^^;