Releases: Parallel-in-Time/PFASST
Releases · Parallel-in-Time/PFASST
MPI Bugfix Release
v0.5.0 -- The MPI Bugfix Release (2015/06/01)
DOI: 10.6084/m9.figshare.1431794
$ git diff --stat v0.4.0..v0.5.0
> 45 files changed, 1109 insertions(+), 758 deletions(-)
Notable Features
- random deadlocks with various setups due to open and unused MPI request handlers
(#210, #212, #216) - PFASST with one processor was not MLSDC
(#193, #196) - MPI is now the default when compiling
(#197) - expected quadrature error now calculated and logged
(#206) - runtime parameters can now also be read from config file instead of command line
(#199) - colourful logging can now be disabled via runtime flag
(#202) - file name for logs can be specified via runtime parameters
(#203)
Details
VectorEncapsulation
andMPIVectorEncapsulation
merged into one
(#176, #200)- version now set by Python script not requireing recompilation of everything when not changed
(#195) - various enhancements and fixes of build system
(#201, #211) - little code cleanup
(#194, #204, #205, #208) - some documentation
(#191, #207)
Contributors
- Matthew Emmett, Lawrence Berkeley National Laboratory (memmett)
- Torbjörn Klatt, Jülich Supercomputing Centre (torbjoernk)
- Daniel Ruprecht, Institute of Computational Science, University of Lugano (danielru)
- Robert Speck, Jülich Supercomputing Centre (pancetta)
- Selman Terzi, Jülich Supercomputing Centre (selmanTerzi)
MPI-Bugfix Release
The MPI communication of PFASST was buggy, incomplete and wrong.
The Big Improvement
DOI: 10.6084/m9.figshare.1381721
$ git diff --stat v0.3.0..v0.4.0
> 131 files changed, 15182 insertions(+), 6940 deletions(-)
Notable Features
- Iteration controll for PFASST via relative and absolute residual tolerances.
(#139, #141, #142) - Adding an implicit sweeper with support for LU decomposition of quadrature matrix.
(#150) - General improvements to the logging and configuration functionality.
(#144, #145, #146, #148, #163, #168, #169, #177) - Successful tests on supercomputers (i.e., JUQUEEN)
(#168, #176) - Improved code organization.
(#156, #157, #158, #170, #181) - Rewamped documentation.
(#149, #160, [#179][], #183) - Examples: Boris supports multiple particles and multi-level coarsening
(#152, #172, #180)
Details
- logging framework [Easylogging][] has been updated from v9.75 to v9.80.
(#163) - root finding algorithm for polynomials uses fixed number of iterations or residual tolerance.
(#162) - reworked how
pfasst::init(...)
is used.
(#148) - minimum CMake version bumped to 2.8.6
(#153) git describe
-based version number aspfasst::VERSION
(#185)- script to generate and compile test coverage report was rewritten in Python 3.x
(#154)
Contributors
- Matthew Emmett, Lawrence Berkeley National Laboratory (memmett)
- Torbjörn Klatt, Jülich Supercomputing Centre (torbjoernk)
- Daniel Ruprecht, Institute of Computational Science, University of Lugano (danielru)
- Robert Speck, Jülich Supercomputing Centre (pancetta)
- Selman Terzi, Jülich Supercomputing Centre (selmanTerzi)
The Big Cleanup Release (v0.3.0)
Notable Features
- Complete rewrite of quadrature functions.
- Addition of iteration control mechanism for SDC and MLSDC.
(extention for PFASST is scheduled) - Introduced framework for passing command line parameters to programs
using PFASST++. - Versatile and colourful logging framework based on
Easylogging++ (https://github.com/easylogging/easyloggingpp). - New Example: Boris-SDC.
Details
- Using Eigen3 (http://eigen.tuxfamily.org/) matrix data types and
functionality instead of Boost's uBLAS
(http://www.boost.org/doc/libs/1_57_0/libs/numeric/ublas/doc/index.html). - Introduced callback interception points in sweepers:
post_step
,
post_sweep
andpost_predict
. - In sweepers:
state(0)
is not assumed to be same asstart_state
. - Examples are now in their own namespace
pfasst::examples::
- Bunch of CMake-related fixes and extentions.
- Example of using PFASST++ with a Makefile project (see
advection-diffusion example). - Support for HashDist (https://github.com/hashdist/hashdist) to
provide dependencies. - A few corrections and extentions to the documentation.
Contributors
- Matthew Emmett, Lawrence Berkeley National Laboratory
- Torbjörn Klatt, Jülich Supercomputing Centre
- Daniel Ruprecht, Institute of Computational Science, University of
Lugano - Robert Speck, Jülich Supercomputing Centre
The MPI-PFASST Release (v0.2.0)
Features:
- PFASST implementation using MPI
- Scalar example
First Release
Features:
- basic implementations of SDC and MLSDC