Skip to content
Snippets Groups Projects
Select Git revision
  • 66-absolute-access-is-probably-not-copied-correctly-after-_eval_subs
  • const_fix
  • fhennig/hip-target
  • fhennig/v2.0-deprecations
  • fma
  • gpu_bufferfield_fix
  • gpu_liveness_opts
  • holzer-master-patch-46757
  • hyteg
  • improved_comm
  • master default protected
  • rangersbach/cuda_blocksizes
  • rangersbach/reductions
  • target_dh_refactoring
  • v2.0-dev
  • vectorization_sqrt_fix
  • zikeliml/124-rework-tutorials
  • zikeliml/Task-96-dotExporterForAST
  • release/1.3.7 protected
  • release/1.3.6 protected
  • release/2.0.dev0 protected
  • release/1.3.5 protected
  • release/1.3.4 protected
  • release/1.3.3 protected
  • release/1.3.2 protected
  • release/1.3.1 protected
  • release/1.3 protected
  • release/1.2 protected
  • release/1.1.1 protected
  • release/1.1 protected
  • release/1.0.1 protected
  • release/1.0 protected
  • release/0.4.4 protected
  • last/Kerncraft
  • last/OpenCL
  • last/LLVM
  • release/0.4.3 protected
  • release/0.4.2 protected
38 results
You can move around the graph by using the arrow keys.
Created with Raphaël 2.2.012Mar11107654327Feb2625242120191817141312111065431Jan302928272423222120171615141318Dec131018Nov13127628Oct21141110128Sep824Aug191227Jul24231917161598543111Jun105131May282218161329Apr262322218543228Mar27252019151413121110876528Feb27252019183231Jan30282625191817161514121110918Dec124128Nov1417Oct29Sep282019181513729Aug2824231825Jul20171312530Jun292824227431May3011Apr109331Mar282724221711823Feb22Dec10Nov974325Oct241029Sep282728Jul2123May20update pyproject.toml to distribute all headers in subfolders of `include`Format gpu_indexing.pyIntroduce assume_warp_aligned_block_size & use_block_size_fitting optionsAdapt 3d default gpu block size to reduce register pressureMerge branch 'v2.0-dev' into rangersbach/cuda_blocksizesMerge branch 'fhennig/fix-2d-gpu-indexing' into 'v2.0-dev'v2.0-devv2.0-devFix Linear3D Indexing if Iteration Space Rank < 3Use default block size if block size fitting failsMerge remote-tracking branch 'origin/rangersbach/cuda_blocksizes' into rangersbach/cuda_blocksizesDistribute default block size cubicly over dimsAdd more edge cases to block size fitting testTerminate block size fitting if block size is large enough and round to next multiple of warp sizeFix ceil_to_multipleMerge branch 'fhennig/include-cleanup' into 'v2.0-dev'Some cleanup to the pystencils runtime headersSmall adaptations to gpu_kernels.md user manualAdapt docstring for DynamicBlockSizeLaunchConfigurationRename determine_block_size to fit_block_sizeMerge remote-tracking branch 'origin/rangersbach/cuda_blocksizes' into rangersbach/cuda_blocksizesAdd test checking generated GPU launch configurationsSlightly adapt block size fitting algorithmAdapt formatting of gpu_indexing.pyRemove max_block_sizes config option and add getters for gpu targets in GpuIndexing. Encapsulate in HardwareProperties dataclassMinor fix for parameters of automatic launch configApply 2 suggestion(s) to 2 file(s)Refactor condition for failure in check_sizes_and_return and emit reason in error messageOmit lambda for warp size in GpuIndexingMove ceil_to_multiple to utils moduleMerge branch 'rangersbach/structural' into 'v2.0-dev'Introduction of structural ast nodesFix lintAdd small description of block config determination in docs for gpu codegenAdapt formattingFix lintFix lintFix getter for thread exec condition for dense/sparse iteration spaces in cuda.pyRemove unnecessary conditionSlightly adapt algorithm for determining block sizes with known iteration spaceAdd support for "single" type in parse_type_name functionzikeliml/Task-9…zikeliml/Task-96-dotExporterForASTV0.3 of dot printer, now supports colored precision output, integrates the IRPrinter and uses a context manager and dynamic lifetyle halndling of subgraphs.