Trac January 2014

trac@lists.einsteintoolkit.org

1 participants
145 discussions

[Einstein Toolkit] #1326: running loopcontrol on strange number of threads fails
by Einstein Toolkit 22 Mar '15

22 Mar '15

#1326: running loopcontrol on strange number of threads fails -----------------------------------+---------------------------------------- Reporter: rhaas | Owner: Type: defect | Status: new Priority: minor | Milestone: Component: EinsteinToolkit thorn | Version: Keywords: LoopControl | -----------------------------------+---------------------------------------- my machine has 8 cores (according to /proc/cpuinfo). Running eg the trigger test with 3 threads fails inside of loopcontrol. To reproduce: {{{ export OMP_NUM_THREADS=3 mpirun -n 2 exe/cactus_bns_all arrangements/AEIThorns/Trigger/test/trigger.par }}} -- Ticket URL: <https://trac.einsteintoolkit.org/ticket/1326> Einstein Toolkit <http://einsteintoolkit.org> The Einstein Toolkit

1 1

[Einstein Toolkit] #1361: disable hyperthreading in loopcontrol by default
by Einstein Toolkit 22 Mar '15

22 Mar '15

#1361: disable hyperthreading in loopcontrol by default -----------------------------------+---------------------------------------- Reporter: rhaas | Owner: Type: defect | Status: new Priority: major | Milestone: Component: EinsteinToolkit thorn | Version: Keywords: LoopControl | -----------------------------------+---------------------------------------- when running with both openmp and sufficiently many threads that hyperthreading threads are used, many tests using LoopControl (Cartoon, RotatingSymmety180, RotatingSymmetry90) fail. This can be tracked down to disabling hyperthreading support in LoopControl (ie. turning of hyperthreading makes things work). In particular on bethe with smt and 8 physical cores: The Cartoon/test_cartoon_2.par test shows differences from the recorded results when run with 16 threads (but not with 8 threads). If I then go ahead and disable OMP in all ML source files but ML_BSSN_enforce *and* comment out the #include "loopcontrol.h", then the difference goes away. Adding back #include "loopcontrol.h" brings back the error. Some further experimenting with LoopControl's options shows that indeed the use_smt_threads option is what causes problems. If I turn it off things work fine even with a vanilla source tree. Otherwise relative differences are on the order 1e-7 and absolute 1e-11 (in momx_z_[2][2].xg). Without smt the results are identical to the stored values. The issue only occurs in combination of OpenMP, vectorization and hyperthreading. The issue is independent of the compiler (both intel 13 and gcc 4.4 show the same behaviour), and vectorization (sse2) and many threads (up to 4 times the number of physical cores) works fine on non-smt machines. I propose to disable LoopControl::use_smt_threads by default. Note that we cannot completely remove it since apparently for Vesta (a Blue Gene/Q) smt is required to get and multi-threading at all. -- Ticket URL: <https://trac.einsteintoolkit.org/ticket/1361> Einstein Toolkit <http://einsteintoolkit.org> The Einstein Toolkit

1 4

[Einstein Toolkit] #1204: carpet bug
by Einstein Toolkit 22 Mar '15

22 Mar '15

#1204: carpet bug -------------------------------------+-------------------------------------- Reporter: abdik@… | Owner: eschnett Type: defect | Status: new Priority: major | Milestone: Component: Carpet | Version: Keywords: | -------------------------------------+-------------------------------------- The latest version of carpet seems to contain a bug that affects AMR+multipatch runs. My stderr and stdout and par file are attached. Found by Roland and Ernazar. -- Ticket URL: <https://trac.einsteintoolkit.org/ticket/1204> Einstein Toolkit <http://einsteintoolkit.org> The Einstein Toolkit

1 3

[Einstein Toolkit] #997: problem in appending output after recovery
by Einstein Toolkit 22 Mar '15

22 Mar '15

#997: problem in appending output after recovery ----------------------------------------+----------------------------------- Reporter: corvino.giovanni@… | Owner: eschnett Type: defect | Status: new Priority: major | Milestone: Component: Carpet | Version: Keywords: | ----------------------------------------+----------------------------------- I have a problem in appending output files from Carpet. I used to produce 3d HDF5 output of grid variables and write the output in the same directory also after recovery from checkpoint. The new output was automatically appended to the existing one. Now I am producing h5 output also on 2D slices but in this case the output is overwritten so I lost the data for all but the last recovery. -- Ticket URL: <https://trac.einsteintoolkit.org/ticket/997> Einstein Toolkit <http://einsteintoolkit.org> The Einstein Toolkit

1 3

[Einstein Toolkit] #1518: Parameter parser and CCTK_ParameterSet interpret leading zeros in numbers differently
by Einstein Toolkit 28 Jan '15

28 Jan '15

#1518: Parameter parser and CCTK_ParameterSet interpret leading zeros in numbers differently --------------------+------------------------------------------------------- Reporter: rhaas | Owner: Type: defect | Status: new Priority: minor | Milestone: Component: Cactus | Version: development version Keywords: | --------------------+------------------------------------------------------- the parameter parser allows things like: {{{ thorn::param1 = 011 thorn::param2 = 012.34 }}} in parameter files. For floating point values this is a bit unexpected but otherwise mostly harmless. For integers the situation is a bit more complex since in C a leading zero is used to indicate a octal number. And (worse) while the parameter file parser converts the string "011" to the number 11 the Cactus call CCTK_ParameterSet will convert it to 9. The difference is ultimately the difference between calling atof (Parser) and strtol (CCTK_ParameterSet). To avoid confusion it would likely be good to change CCTK_ParameterSet to behave the way the Parameter parser does. This is a change in behaviour compared to the pre-Piraha parser, however I suspect the number of users that actually used octal (or hexedecimal) notation in their parameter files is small. The change is to change {{{inval = strtol (value, &endptr, 0);}}} to {{{inval = strtol (value, &endptr, 10);}}} in line 2209 of src/main/Parameters.c and similar in line 2270. -- Ticket URL: <https://trac.einsteintoolkit.org/ticket/1518> Einstein Toolkit <http://einsteintoolkit.org> The Einstein Toolkit

1 27

[Einstein Toolkit] #428: Forbid configuration names ending in -reconfig
by Einstein Toolkit 12 Jan '15

12 Jan '15

#428: Forbid configuration names ending in -reconfig ----------------------+----------------------------------------------------- Reporter: eschnett | Owner: Type: defect | Status: new Priority: minor | Milestone: Component: Cactus | Version: Keywords: | ----------------------+----------------------------------------------------- I accidentally created a Cactus configuration with a name like "sim- reconfig". This confuses Cactus, because the command "make sim-reconfig" can then mean either to build the "sim-reconfig" configuration, or to reconfigure the "sim" configuration. Cactus should catch and forbid these cases. This is probably most cleanly handled by using a script instead of a Makefile to interpret the user commands. -- Ticket URL: <https://trac.einsteintoolkit.org/ticket/428> Einstein Toolkit <http://einsteintoolkit.org> The Einstein Toolkit

1 12

[Einstein Toolkit] #1039: Web-based documentation should be automatically updated
by Einstein Toolkit 09 Dec '14

09 Dec '14

#1039: Web-based documentation should be automatically updated -------------------------------------+-------------------------------------- Reporter: hinder | Owner: Type: enhancement | Status: new Priority: minor | Milestone: Component: EinsteinToolkit website | Version: Keywords: Documentation | -------------------------------------+-------------------------------------- We have documentation for Cactus and the ET on their respective websites. This should be updated automatically when the source is changed, so it is never out of date. There should be documentation for the current release as well as the current development version. There is one technical problem standing in the way of this. The system should be automated, and it needs to have commit rights to a specific directory in the SVN repository hosting the files. SVN accounts for CCT machines tend to be the same as CCT login accounts, so we don't want to store the username and password. I think the best solution is to create an SVN account specifically for the documentation build system which is not tied to a CCT login account. This account would be given commit access to just the directories necessary. The "checkout, build doc, commit" script should be kept in version control somewhere, and a machine should be chosen on which to run it. It could be run regularly, or just in response to commits. -- Ticket URL: <https://trac.einsteintoolkit.org/ticket/1039> Einstein Toolkit <http://einsteintoolkit.org> The Einstein Toolkit

1 2

[Einstein Toolkit] #1487: Documentation build should be tested as part of the automated build and test
by Einstein Toolkit 08 Dec '14

08 Dec '14

#1487: Documentation build should be tested as part of the automated build and test --------------------+------------------------------------------------------- Reporter: hinder | Owner: hinder Type: defect | Status: new Priority: minor | Milestone: Component: Other | Version: development version Keywords: | --------------------+------------------------------------------------------- The build of the ET documentation should be tested as part of the automated build and test. -- Ticket URL: <https://trac.einsteintoolkit.org/ticket/1487> Einstein Toolkit <http://einsteintoolkit.org> The Einstein Toolkit

1 3

[Einstein Toolkit] #791: Output timer tree as XML
by Einstein Toolkit 08 Dec '14

08 Dec '14

#791: Output timer tree as XML -------------------------+-------------------------------------------------- Reporter: hinder | Owner: eschnett Type: enhancement | Status: new Priority: minor | Milestone: Component: Carpet | Version: Keywords: | -------------------------+-------------------------------------------------- The attached patch to Carpet adds a parameter (off by default) which outputs the timer tree from each process to an XML file in the output directory at the end of the run. The output looks like this: {{{ <timer name = "main"> 7.7702 <timer name = "CallFunction"> 0.000241 <timer name = "thorns"> 0.000233 <timer name = "CaKernel_FreeDevMem"> 0.000221 <timer name = "PostCall"> 2e-06 </timer> <timer name = "PreCall"> 1e-06 </timer> </timer> </timer> </timer> <timer name = "CarpetStartup"> 0.00982 <timer name = "AllocateGridHierarchy"> 5e-06 </timer> }}} An alternative schema would be to have the name and timer value as subelements; i.e. {{{ <timer> <name>main</name> <value>7.7702</value> <children> <timer> <name>CallFunction</name> <value<0.000241</name> ... </timer> }}} The <children> tag might not be necessary, but might make it easier to parse. Having one file per process is not ideal; we would like to add reductions across processes to give min, max, average, standard deviation etc, also for the standard output display. OK to commit as a work-in-progress? We can change the schema later if it turns out to be easier to deal with. -- Ticket URL: <https://trac.einsteintoolkit.org/ticket/791> Einstein Toolkit <http://einsteintoolkit.org> The Einstein Toolkit

1 15

[Einstein Toolkit] #727: Test output changes every time some tests are run
by Einstein Toolkit 08 Dec '14

08 Dec '14

#727: Test output changes every time some tests are run -----------------------------------------------------------+---------------- Reporter: hinder | Owner: Type: defect | Status: new Priority: major | Milestone: Component: EinsteinToolkit thorn | Version: Keywords: testsuites QuasiLocalMeasures IDAxiOddBrillBH | -----------------------------------------------------------+---------------- Each time a new run of the test suites is performed, the test output of IDAxiOddBrillBH and QuasiLocalMeasures changes a little. These changes are below the tolerances set in the test.ccl files, so the tests pass, but there should be no change at all when the tests are run on the same machine. An example of the difference from one test run to the next is shown at http://git.barrywardell.net/EinsteinToolkitTestResults.git/blobdiff/95fe088…. Another example, from QuasiLocalMeasures, is at http://git.barrywardell.net/EinsteinToolkitTestResults.git/blobdiff/95fe088… /qlm-ks-shifted/admbase::metric.average.asc. The only thorns to exhibit this behaviour are IDAxiOddBrillBH and QuasiLocalMeasures. All other test data remains unchanged from one test run to the next. Note that the two test runs will be performed with different executables, since a new configuration is created for each run. I think I remember testing that simply re-running the same executable resulted in the same test output. Is it possible that there are a small number of points which access uninitialised memory which changes on each run for these thorns? -- Ticket URL: <https://trac.einsteintoolkit.org/ticket/727> Einstein Toolkit <http://einsteintoolkit.org> The Einstein Toolkit

1 6

← Newer
1
...
4
5
6
7
8
9
10
...
15
Older →

Jump to page:

2026

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

Trac January 2014