Skip to content
Snippets Groups Projects
Select Git revision
  • 13-mpi-poisson-solvers
  • 14-development-documentation protected
  • 14-development-documentation-2
  • guix-package
  • joss
  • marthe
  • master default protected
  • rocm_singularity_support
  • rocm_support
  • wipGuixRecipe
  • 2024.11
11 results
You can move around the graph by using the arrow keys.
Created with Raphaël 2.2.014Nov1313Oct1Jul22May2120230Apr29282524232217109732126Mar25242113121176528Feb27262418765331Jan30242322212014109612Dec54329Nov28272019138628Oct25141110743230Sep252320181613121196530Aug281431Jul30262524191716119543219Jun176430May292827242322211716151310730Apr2927252419181716121110543229Mar28272522151413121187654127Feb262522212019181312876131Jan3029252322211817Dec14131276530Nov26212014626Oct252423128Sep2120191475331Aug30281025Jul1312111065429Jun22drag/lift, tempAjout patch pour passer en simple precision. Corrections de tests pour la simple presision.Update drag and lifts.Clean cmake config. Fix (at least ...) libparmes find bugFix build env pbfix old cmake behaviorfix old cmake behaviorFix(?) cmake weird behavior for fftw and libs. Remove scitools picklerientemp drag and lift. Untest, unstableMulti-GPU communications optimisation on 1k advection-remeshingCleanupDynamic multi-gpu communication size on remeshing for to left comm.Dynamic multi-gpu communication size on remeshing for to right comm.Multi-GPU profiling outputTemp drag. Some tests fail (test_advection_nullVelocity, test_advection_randomVelocity) following merge with JM branch. To be fixedAdd penal_vort + tests, momentum drag and lift (not fully tested), tests ok, Noca to be reviewedfix penalization bug. All tests okImprove multi-gpu advection profiling. Add overlaped communications in diffusionFix profiling of communication kernelsCleanup multi-GPU new communicationsOk for 1k version. before overlapping.Fix pipelined transfers for velocity. All cases ok (1k, 2k, MS ans SingleScale). Need comm overlap for 1k.Fix velocity pipelined transferPipeline transfer for velocityExtend use of pinned memory to optimize standard Host-Device transfersFix profiler for oprators namesRelease the pinned pagesFix pinned memory usagepipelined transfer with pinned memory (host) for scalarOptimize multi-GPU remeshing for communications ovelarp by computationsfix testsfix testsNew subsets + update penalization op. All tests okMulti-GPU remeshing: overlaping communications by computationsazodhazoduhFix multi-gpu, remove assertion on 1D topology for scalesAdd a max dt for adaptative timestepFix last bugs in subsets. All tests ok except python advection and penalization (--> obstacles obsolete)Fix last bugs in subsets. All tests ok except python advection and penalization (--> obstacles obsolete)