[Users] Meeting Minutes

[Roland Haas]

Present: Vassili, Peter, Frank, Philipp, Yosef, Roland, Tanja, Ian

Tickets:
* remove ADM from ET: no objections in principle (already in
CactusArchive), need to check which test suites rely on it. Ian will run
testsuite without ADM compiled in and check which tests are not run. ADM
can take a long time to compile on some compilers (intel?, using DEBUG
options)

* new set of GRHydro updates, please review, will be applied on Thursday
unless objections are raised

GRHydro:
* have large number of implementations of con2prim, with overlap and
code and functionality, some of which fail for cases they claim to work
* Philipp is importing a new C based con2prim written by a grad student
that handles all cases, is based in a published paper and is faster than
the current implementation. It is written and C with GPUs in mind. Will
keep the old implementations around and select between the two based on
a parameter on the scheduler level. Decide come time before next release
if new version is stable enough to make default or not. If default,
deprecate old code and remove within one year.
* code is being rewritten to simplify (auto)vectorization by compiler,
currently Prim2Con and PPM are finished/being rewritten. New code is in
C since more control over what is going on. Will decide before next
release whether to replace current code or offer option. Prim2Con most
likely can be replaced since the logic is simple, PPM might not offer
the very same set of features as the current code.
* there was a suggestion to move the tests into a thorn of their own,
this suggestion found not support
* vectorization changes: Caltech (Christian Ott mostly) is working on
re-writing GRHydro to take advantage of the compiler's ability to
vectorize code. Depending on routine and CPU this can result in large
speedups (combined with other optimizations).
* some code in GRHydro is (nearly) orphaned: Philip and Tanja will
(re)adopt the vector potential code, entropy evolution is not yet adopted

Workshop
* local organizers to update wiki page, organize breakfast and coffee,
provide maps etc, test room setup with remote participants
* topics:
** vectorization
** AMR grid setup
** AMR scaling
** IO scaling
** cleanup of GRHydro
** automated performance tests to verify that changes are actually
beneficial
** MHD development
* previous suggestions:
** Refluxing
** A-field MHD
* David Rideout offered to give a presentation (private email to Erik,
hope I did not get this wrong)
* want to have a relatively large list of possible topics then select
which ones to actually work on

Misc:
* Dennis Castleberry (LSU) is working on changing the testsuite driver
script to run multiple tests in parallel if the MPI processes,
OMP_NUM_THREADS and machine core number settings allow. Desire to make
sure that this interacts smoothly with simfactory's testsuite option.
Ian is most familiar with this code in simfactory.
* there is a desire to have fewer (ideally) one mailing lists for all
commit messages or to be able to forward Cactus commits to the ET list.
Problematic to allow users to Reply to these messages and have the
answer forwarded to the original list. Shelved for now.
* Alternative/stopgap measure: list the "other" mailing lists eg for
Cactus/Kranc/Carpet/Lllama/Outflow/Other-Externally-Hosted-Thorns on ET
Support page
* please provide short fully functional parfiles similar to the one in
the gallery http://einsteintoolkit.org/about/gallery/ (Roland has in the
past promised a NsNs parfile, now that Llama and CTGamma are public,
this is much easier :-)).

Yours,
Roland











-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 262 bytes
Desc: OpenPGP digital signature
Url : http://lists.einsteintoolkit.org/pipermail/users/attachments/20130701/c0258671/attachment.bin