Project

General

Profile

Mar 25 2019

Apologies

Mark on holiday

Absent

Completed since last meeting (Mar 18 2019)

  • more nbnxm cleanup

Upcoming targets

  • 2019.2 April 15 - see https://redmine.gromacs.org/versions/86
  • BioExcel KPIs and risks due ASAP for D6.2
  • BioExcel D1.1 software development plan draft due March 31
  • BioExcel D3.1 – Use Case Work Plans March 31, our input in excel-sheets to be restructured into document template, Christian

Decisions needed

  • Decide between bComputeGlobals, doComputeGlobals and computeGlobals and update dev guide
    - computeGlobals should be the function, doComputeGlobals the boolean
  • How to split off BioExcel specific discussion from GROMACS meetings?
    - rather project summary than very specific administration details

Progress since last meeting

  • ORNL open to the idea of giving us access, resource for CI possibly available in future, but no suggestion from them of manpower just now - Mark to continue
  • patches preparing cryo-EM refinement module - Christian
  • patches using mdspan infrastructure - Christian
  • sketch for abstraction layer for external data input, put on hold for now - Joe, Mark, Christian, Erik
  • Planning for FEP meeting - Mark, Berk, Michael Shirts
  • Pure-GPU MD loop draft up - Artem
  • Modular integrator feature branch available - Mark
  • Python API patches up (and feature branch available) - Joe, Mark
  • New cuda redmine issues from Alan need feedback - Szilard, Artem, Mark
  • Constraints tests + GPU implementation - Artem, Mark
  • CUDA buffer ops patches - Szilard, Berk, Mark
  • do_force improvements
    • PBC setup in force loop improvements - Szilard
  • BioExcel-2 WP1
    • finalize D1.1 content for GROMACS * create tasks in ClickUp - Mark, Paul, Berk, Christian * iterate on KPIs proposal - Mark, Berk, Paul, Erik, Christian * iterate on UEDIN QM/MM CP2K benchmarks and reports
  • BioExcel-1 AWH benchmarking writeup - Berk
  • Topology data structure cleanup - Paul, depends on more cleanup
  • High level GPU manager - Szilard, Artem, Mark
    • High level object for long-lived GPU objects (hack struct with void pointers) - Szilard
  • GPU bonded kernel fusion proposed #2818 - Szilard
  • broken virial in SHAKE #2879 - Mark
  • migrated gmx dump to use gmx::Options - Mark
  • proposed gmx::compat::optional - Mark
  • migrating check and mdrun to use gmx::Options (for gmxapi) - Mark

New items

  • Documentation built in docker infrastructure as test for docker infrastructure for continuous integration and testing, Joe
  • GPU-only medium term goals
    - what features does it support?
    - what is the performance behavior, where does it add most to performance
    - what is to be done to get into release

Stale items

  • Draft single-rank single-GPU MD schedules - Szilard, Berk, Artem
  • native CUDA in cmake problematic #2831 - no progress, Mark has ideas but not yet time
  • how to find libstdc++ for icc and clang #2842 - Mark, ready but needs follow up (Cray CC is broken already)
  • naming conventions for GROMACS module and file names #2839 - Mark to promulgate decision
  • using Redmine for project organization - Mark
  • Email gmx-developers about development feature flags - Mark
  • Adopt clang-format #2107 - Berk