Meeting minutes:
Present were: Ian, Tanja, Roland, Frank, Christian, Peter, Gabrielle, Eloisa, Bruno, Josh
* GRHydro: ** a development branch to add finite-temperature evolution was created in order to keep trunk stable for the release ** temp_and_ent_evolution_method will be renamed to temperature_evolution_method
* Testsuites: ** we suspect that failures of MacLachlan testsuites might be related to OpenMP settings and numerical accuracy ** for now we will remove the Hamiltonian constraint from the testsuite ** in the future we want per variable tolerances ** Peter and Eloisa will look into the failing AHFinderDirect testsuites ** Peter will look into the failing ADM testsuite ** Roland will look into why some GRHydro testsuites fail
*upcoming ET release: ** name: Chandrasekhar ** Ian will create git tags for the release (Kranc, Carpet, MacLachlan?) ** Frank will create svn tags for the release ** will have to look into build issues with libjpeg (might already be resolved by now)
*future ET release: ** simfactory should be able to run the testsuites ** run testsuites using OpenMP ** per variable tolerances ** use EOS_Omni in GRHydro, remove one+ EOS interface ** create a tarball with the complete checkout for easier download by new users
Yours, Roland
Hi,
It's Thursday and we agreed on Monday to try to get the testsuites in shape until yesterday, so here is the current status:
The following testsuites which failed on Monday now pass:
test_waveell test_binary_1 gw3d_McLachlan_ord4_*
This means that one some machines (e.g. the philip cluster at LSU or the numrel workstations at LSU with the intel compiler, but possibly also more machines once the tests are rerun) we have no failing testsuites for 1-process runs.
However, there are still some issues:
- most of the AHFinderDirect testsuites fail on 2 processes - some of the GRHydro testsuites fail on some machines, possibly especially on machines not using the intel compiler. This is especially bad because that affects Ranger and Kraken, the two largest machines on the TeraGrid. - test_ADM_2 fails on two processors
All of the above already have volunteers working on them, but we don't yet have a volunteer for the following:
- the DistortedBHIVP fails on half of the machines, also possibly connected to the compiler used
Please reply to this message if you can try to help with this.
Did anybody see build problems? I am especially asking because of known (hopefully past) problems with Formaline and/or libjpeg.
Frank Loeffler
Hello all,
However, there are still some issues:
- most of the AHFinderDirect testsuites fail on 2 processes
- some of the GRHydro testsuites fail on some machines, possibly especially on machines not using the intel compiler. This is especially bad because that affects Ranger and Kraken, the two largest machines on the TeraGrid.
Ranger seems to use the Intel compiler (10.1). When I tested on Ranger the error persists even for -O0 and with "-fp-model precise". On the other hand the test passes on my workstation using gcc (4.1.2).
Yours, Roland
On Thu, Nov 04, 2010 at 10:33:06AM -0400, Roland Haas wrote:
Ranger seems to use the Intel compiler (10.1). When I tested on Ranger the error persists even for -O0 and with "-fp-model precise". On the other hand the test passes on my workstation using gcc (4.1.2).
When you say 'the test', which one do you mean? Did you by any chance also look into the similar testsuites from HydroInitExcision? Can you commit the test results from your workstation?
thanks, Frank
Hello all,
When you say 'the test', which one do you mean?
Ah, sorry about that. I was speaking of GRHydro_test_tov_ppm_ML both using 1 and 2 MPI processes and no OpenMP.
Did you by any chance also look into the similar testsuites from HydroInitExcision?
I did not look into them any further beyond looking at halfsphere_pugh_ppm and noticing that the errors are in scon[0] and of relative order 2e-11.
Can you commit the test results from your workstation?
Just did (results from yesterday so they don't include the wavebinary fixes yet).
Yours, Roland
On 4 Nov 2010, at 15:20, Frank Loeffler wrote:
However, there are still some issues:
- most of the AHFinderDirect testsuites fail on 2 processes
- some of the GRHydro testsuites fail on some machines, possibly
especially on machines not using the intel compiler. This is especially bad because that affects Ranger and Kraken, the two largest machines on the TeraGrid.
- test_ADM_2 fails on two processors
All of the above already have volunteers working on them,
Just as an observation, the AHFinderDirect and test_ADM_2 tests also failed for the last release.
Damiana is now in as good a situation as it was for the last release, and I am rerunning the Kraken tests now.
users@lists.einsteintoolkit.org
-
Frank Loeffler -
Ian Hinder -
Roland Haas