Skip to content

Latest commit

 

History

History
69 lines (53 loc) · 2.41 KB

logbook.org

File metadata and controls

69 lines (53 loc) · 2.41 KB

Setup of basic MD features

<2015-11-26 Thu>

Velocity verlet and RATTLE for the dimer are working properly and most of the MD code is moved to modules.

A round of profiling

<2015-11-27 Fri>

Using test/setup_single_dimer.f90 as a basis, identify most time-consuming parts of the code:

  • simple_mpcd_step
  • maximum_displacement
  • compact_p_to_h

The MPCD step requires line profiling and/or a parallel RNG to speed things up.

The displacement routine has been accelerated by resetting particles in the PBC box only upon resorting, so that the call to rel_pos can be removed.

No idea yet for compact_p_to_h.

Documentation

<2016-06-14 Tue>

The organization for the documentation is:

  • A general documentation is generated with sphinx on the basis of the doc/*.rst files.
  • A doxygen-generated API, reading from the Fortran source files in src/, programs/ and test/.
  • An explicit relation the documentation and the API is provided in doxygen by linking to the relevant documentation page and in the documentation by a sphinx directive that locates doxygen names.
  • The website for RMPCDMD is based on the sphinx documentation.

Build for the mac

<2016-06-15 Wed>

Building on OS X is possible, although the lack of built-in package manager makes it a bit longer to set up.

The procedure, whose first version was contributed by Sébastien Michelin, is the following:

sudo port install gcc-5 cmake git
git clone --recursive https://github.com/pdebuyl-lab/RMPCDMD
cd RMPCDMD
CC=gcc-mp-5 FC=gfortran-mp-5 ./scripts/download_and_build_hdf5.sh
mkdir build
cd build
HDF5_ROOT=../_hdf5-1.8.17 cmake .. -DCMAKE_C_COMPILER=gcc-mp-5 -DCMAKE_Fortran_COMPILER=gfortran-mp-5
make

On the way, the lack of md5sum and wget on the mac was circumvented by using openssl md5 -r and curl instead. CC and FC must be specified, else the system-provided compiler takes priority for both the HDF5 build and for the cmake build.

Scaling and profiling with Allinea map

<2016-10-10 Mon>

A number of scaling tests have been performed with Allinea map on the VSC Tier-1 system Breniac. Many scaling issues were identified as OpenMP errors in the code leading to probable race conditions. Using also Intel Inspector to reveal potential races, many missing private attributes were set.