I just updated the Einstein Toolkit test results on http://einsteintoolkit.org/release-info/parse_testsuite_results.php for many systems.
For the systems that do not yet have current results, please either update them, or remove them from the list if the systems are outdated, or give me access so that I can run the tests there as well.
There seem to be two main problems that need addressing: (1) On Hopper (and Gordon and Trestles), many tests running on 2 MPI processes fail. This is likely an MPI problem, or a configuration problem related to how we use MPI. (2) On Blue Waters, a more tests (8) than on most other systems are failing. We should investigate.
This is a community effort. If you are a maintainer of the Einstein Toolkit, you are expected to contribute to the release process.
-erik
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Hello Erik, all
I just updated the Einstein Toolkit test results on http://einsteintoolkit.org/release-info/parse_testsuite_results.php
for many systems. Very nice. Thank you very much for running these.
For the systems that do not yet have current results, please either update them, or remove them from the list if the systems are outdated, or give me access so that I can run the tests there as well.
I would like to suggest to remove kraken from the list since it is going to be (http://www.nics.tennessee.edu/event/kraken-decommission) decommissioned very soon.
The table shows both zwicky and zwicky-intel11.1. Looking at the option list used it seems to me that for zwicky already the intel 11 compiler was used. So zwicky-intel11.1 should be removed.
There seem to be two main problems that need addressing: (1) On Hopper (and Gordon and Trestles), many tests running on 2 MPI processes fail. This is likely an MPI problem, or a configuration problem related to how we use MPI. (2) On Blue Waters, a more tests (8) than on most other systems are failing. We should investigate.
I can have a look at the bluewaters failures (one of them is a hydro failure so they default to me anyhow).
Yours, Roland
- -- My email is as private as my paper mail. I therefore support encrypting and signing email messages. Get my PGP key from http://keys.gnupg.net.
Hello all,
I can have a look at the bluewaters failures (one of them is a hydro failure so they default to me anyhow).
The hydro failure (balsara4_2d) goes away when I compile with -O3 instead of -Ofast so it seems that this is another case where hydro is very susceptible to roundoff level errors being amplified. Please note that balsara4_1d (the 1d version of this test) already has larger error threshold so I think I will raise the 2d thresholds to the same value. This test has very high Lorentz factors in the data so can very hard for the code. With this I still get test failures for Schwarzschild_EF, regression_test (in SphericalHarmonicRecon) and twopunctures_carpet.
Yours, Roland
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Hello all,
For the systems that do not yet have current results, please either update them, or remove them from the list if the systems are outdated, or give me access so that I can run the tests there as well.
I also realize that we unfortunately did not include the IllinoisGRMHD code in the list of compiled thorns (which would have been nice since then it would be tested on all machines, though Zach has already tested it with PGI compilers which I take to be the most exotic ones we compile with).
Yours, Roland
- -- My email is as private as my paper mail. I therefore support encrypting and signing email messages. Get my PGP key from http://keys.gnupg.net.
On 23 Apr 2014, at 16:48, Roland Haas roland.haas@physics.gatech.edu wrote:
Signed PGP part Hello all,
For the systems that do not yet have current results, please either update them, or remove them from the list if the systems are outdated, or give me access so that I can run the tests there as well.
I also realize that we unfortunately did not include the IllinoisGRMHD code in the list of compiled thorns (which would have been nice since then it would be tested on all machines, though Zach has already tested it with PGI compilers which I take to be the most exotic ones we compile with).
We did this deliberately as it is not yet a part of the toolkit. It has not undergone a review, nor had agreement that it should be added. As such, efforts related to it should happen after the release. We put it in the thornlist commented-out at the author's request, as it would make it more convenient for people to download it.
Hello all,
We did this deliberately as it is not yet a part of the toolkit. It has not undergone a review, nor had agreement that it should be added. As such, efforts related to it should happen after the release. We put it in the thornlist commented-out at the author's request, as it would make it more convenient for people to download it.
I see. I did not want to suggest including IllinoisGRMHD in the list of testsuites or claim it is tested in the release. It just seemed to me that this would be a good opportunity to check if the that code actually compiles on all machines (since we are compiling on all of them anyway and eg the author may not even have access to all of them). Before we aim to include it in the next release (hopefully). No big deal though.
Yours, Roland
On Apr 23, 2014, at 18:44 , Roland Haas roland.haas@physics.gatech.edu wrote:
Hello all,
We did this deliberately as it is not yet a part of the toolkit. It has not undergone a review, nor had agreement that it should be added. As such, efforts related to it should happen after the release. We put it in the thornlist commented-out at the author's request, as it would make it more convenient for people to download it.
I see. I did not want to suggest including IllinoisGRMHD in the list of testsuites or claim it is tested in the release. It just seemed to me that this would be a good opportunity to check if the that code actually compiles on all machines (since we are compiling on all of them anyway and eg the author may not even have access to all of them). Before we aim to include it in the next release (hopefully). No big deal though.
Starting the tests (on all machines) takes two minutes (combined). What takes long is to figure out why things don't work on particular machines. Adding a few additional thorns will be easy; please remind me after the release.
-erik
On Wed, Apr 23, 2014 at 02:33:04PM +0200, Erik Schnetter wrote:
I just updated the Einstein Toolkit test results on http://einsteintoolkit.org/release-info/parse_testsuite_results.php for many systems.
Thanks a lot for providing these! This is a lot of help for all of us.
Frank
users@lists.einsteintoolkit.org
-
Erik Schnetter -
Erik Schnetter -
Frank Loeffler -
Ian Hinder -
Roland Haas