Home - Waterfall Grid T-Grid Console Builders Recent Builds Buildslaves Changesources Users - JSON API - About

Console View

Legend:   Passed Failed Warnings Failed Again Running Exception Offline No data

58346 harman
This should fix the HTML links for sus output.

58345 dav
Fix bad var name from sed
58344 jholmen
* Added bulletproofing to avoid building GPU-specific code for non-GPU builds.
58343 harman
update outputProblemSpec()

58342 harman
Added machinery to restart a timestep.  Currently, if the number of particles deleted exceeds
a user defined value then restart the timestep with a lower delT.  The machinery is easy to
extend to other conditions.

<TSR_max_particles_deleted>  X  </TSR_max_particles_deleted>

58341 bpeterson
Undoing the multiple streams approach for charox.  Multiple streams are going to be harder to add due to that second task layer within Arches. 
58340 ahumphrey
Actually remove -fopenmp flag.
Generate new configure script with Dav's added bulletproofing.
58339 bpeterson
Getting char oxidation to also support multiple streams

58338 bpeterson
This gets the stream partitioning tuning parameter fully work, and also got the parallel_for transitioned over to using multiple streams per loop.  Tested and verified with Poisson1.cc

58337 bisaac
This commit adds all of Minmins work on the nles model. No impact on regression tests.
58336 jas
Revert commit 58331 which broke GPU configure.
58335 jholmen
* Added a "-cuda_streams_per_task" command line option.

This option allows the number of CUDA streams used per task to be specified at run time.
58334 bpeterson
Getting rid of some old multi-GPU per node code which caused ridiculously task graph compile times on certain problems.

58333 harman
Rephrase exception message per Dav's suggestion.

58332 harman
  - tell the user how to use xmllint to find problems with ups file.

  • Linux-Optimize-GPU: configure failed -  stdio
58331 jas
Use -fopenmp flag only when Kokkos has been enabled.
58330 bpeterson
This gets Arches recompiling for GPUs, and now works for non-OpenMP Kokkos builds

58329 bpeterson
This commit targets two major items.  1) There was a conflict between the Unified Scheduler + GPU (uses Uintah managed Pthreads), Kokkos OpenMP (the default host execution space), and the Kokkos Random Number library tool.  The Kokkos Random library wanted to use Kokkos OpenMP for a backend, but that conflicts when more than 1 pthread is used, and it crashed.  To get Kokkos Cuda working, a non-openmp/cuda only Kokkos must be configured.  This also has a side effect which insists that the Unified Scheduler should not be a targetted Kokkos Cuda scheduler, we must further develop John's Kokkos scheduler to handle CUDA work.  2) compare_uda and a handful of additional side tools can now compile with Kokkos mixed in.  I can't get all side tools to compile yet, graph_view is having problems.  3) Note that this commit doesn't allow for GPU char oxidation to compile yet, it seems Derek made some changes which I need to clean up
58328 bpeterson
Made a typo.  Hopefully third time is the charm

58327 bpeterson
Part 2

58326 bpeterson
Latest changes to try and support two different types of locks

58325 derekhar
Added new initialize interface. An example of this API:


When kokkos is enabled these variables are KokkosView3, and var4_View is a view of KokkosView3s. 
When kokkos is not enable these variables are CCVar NCvar SFCXVar SFCYVar or SFCZVar or std::vectors of these variables.

The kokkos version uses managed kokko::views which apparently have high overhead compared to their unmanaged counterparts , and may need to be revisited.

58324 harman
Fixed bug when the values to be updated contained white spaces.

58323 jbhooper
-- Ported diffusion framework into SerialMPM
-- Added Arruda-Boyce 8 Chain Rubber Elastomer constitutive model
-- Misc. formatting cleanup (particularly in MPM/Materials/Diffusion)
58322 bpeterson
This fixes a race condition problem with the Unified Scheduler using the Kokkos Random number library.  Specifically a mutex lock was needed to ensure the Random library was loaded only once, but previously it was trying to use omp locks instead of std::mutex locks.  Unified Scheduler needs std::mutex.  This fix changes MasterLock.h so it knows which threaded environment it is in, and can lock accordingly
58321 harman
Added missing comma.

  • Linux-Optimize-Test: Running MPMICE opt tests failed -  stdio
58320 harman
Added 2D Rate stick simulation to nightly regression tester.
  • Linux-Optimize-Test: Running MPMICE opt tests failed -  stdio
58319 harman
Removed input files that used features that no longer exit.  Material addition.

58318 harman
consistent file names
58317 harman
more house cleaning
58316 harman
house clean and organization.
58315 bpeterson
Cleaned up the temporary UnifiedScheduler GPU computes hack.  Again, it's temporary. 
Also, fixed and RMCRT bug.

58314 harman

Dust off script and take a pass through it.

58313 derekhar
update RMCRT to use new GPU_cellObjects header
58312 derekhar
Fix template bug.  Update headers to use new GPU_cellObjects header. Add conversion to GPUVector.
58311 derekhar
Added new cell object, renamed stencil7 to be more generic since it now holds Vector and Stencil7.
58310 derekhar
removed compiler warning signed / unsigned int comparison
58309 derekhar
adopt invalid material type for gpu DW API
58308 derekhar
Corrected new Parallel_for name
58307 derekhar
Corrected issue when compiling with cuda.  (missing Kokkos_inline)