Skip to content
Snippets Groups Projects

Repository graph

You can move around the graph by using the arrow keys.
Select Git revision
  • FMM
  • MPI_qsched
  • aidan
  • autotools-update
  • cleanup
  • fortran_bindings
  • fortran_with_timers
  • gpu_friendly_bh
  • master default protected
  • merge_from_svn
  • ompss_fixes
  • openMP_locks
  • paper_fixes
  • paper_simulated_memcpy
  • paper_updates
  • resource_reuse
  • task_timers
  • test_bh_2
  • unbiased_gravity
19 results
Created with Raphaël 2.2.018Apr1617May1194327Apr2625211914137Mar1Apr31Mar2918Dec109873126Nov25191711109543230Oct29282726242321201615141312876530Jun171095426May2512328Apr27232119131Mar261886543225Feb1211106528Jan2221201816141373119Dec1412111027Nov2625232018171311220Oct17127643230Sep2928272321141398743231Aug286331Jul292522189876Jun6May17Apr161514139730Mar28201924Feb21192Dec13Nov12105131Oct302928272510Jun69AprStarted copying over the genetic partitioning attempt from swift to see what we can do here in a standalone testbedMPI_qschedMPI_qschedFixed test_matadd and the errors it found to do with send/recv tasks becoming visible to user code (rather than being solely contained in the library. It finds severe issues with load balancing (I believe) slash movement of data when the initial layout would be load balanced. Need a python script to generate task plots from this.Added a trivial matrix addition example to test the ability to cope with data sources from multiple nodes and to test if it scales ok for completely trivial problems.Fixing mess I maderesource_reuseresource_reuseReadding stuff i accidentally removedUpdated the task script to match peter's new version for SWIFTtask_timerstask_timersMerge branch 'openMP_locks' of gitlab.cosma.dur.ac.uk:swift/quicksched into openMP_locksopenMP_locksopenMP_locksfixed the exampleFixed some issues that seemed to appear between C and FORTRAN that i don't think should be requiredfixed some typosUpdated test_n2started on a naive n squared particle algorithm using tasksImplemented a version for #16, tested and works correctly as far as I've foundAttempt to fix up the Makefile.am to make building more reliablefortran_bindingsfortran_bindingsMerge branch 'master' into fortran_with_timers to add the rand bugfix into this branch to be used todayfortran_with_ti…fortran_with_timersChanged the use of rand to rand_r to hopefully improve performance and to be thread-safemastermasterAdded fortran ability to use the task timersFixed merge issuesWorking scriptSWIT task plot script available for #15. I don't think this works as I want but need to spend some itme rewriting it from scratchUpdate to #15, now TASKTIMERS is output and configure is fixedSome infrastructure for #15. Not tested yet, needs integrating with FORTRAN bindings once they are added. May be done before this is added to master so I can use themMissed the module file for including when using make installAdded various qsched_..._none values to fortran bindingsFixed a bug with pthread IDs and made the scheduler threadsafe using our gnu99 not really safe method #13getting thread ID is now just a simple call that doesn't require the user to pass any values. Its slightly complex and relies on pthread_getspecific returning NULL when not called from a pthread - however this is what the pthread specification says so I think its fineAdded a thread ID function as discussed in Issue #13, currently requires the scheduler. Also added tests in the example directory to show it works with both OpenMP and pthreadsMakeplots script addedpaper_updatespaper_updatesMissed someNew first version of the plots, old version still available as .old filesfixed copyright stuffAdded the missing qsched_free call to the fortran testWorking version of FORTRAN QuickSched bindings, however the fortran examples do not seem to be built by automakeVersion 0 of the FORTRAN bindings for FORTRANFixed the same bug that occurred twice.Fixed a bug that would occur if we had local and non-local tasks with joint children & no parents that lock on the same resource (this doesn't happen in our tests, so doesn't affect our results).I think I actually fixed the bug now.aidanaidanHopefully fixed a bug that has never occured, but could occur if hierarchical resources were small enough. Didn't break anythingMerge branch 'paper_fixes' into 'master' Merge branch 'MPI_qsched' of gitlab.cosma.dur.ac.uk:swift/quicksched into MPI_qsched