Skip to content
GitLab
Projects Groups Topics Snippets
  • /
  • Help
    • Help
    • Support
    • Community forum
    • Submit feedback
    • Contribute to GitLab
  • Sign in
  • pystencils pystencils
  • Project information
    • Project information
    • Activity
    • Labels
    • Members
  • Repository
    • Repository
    • Files
    • Commits
    • Branches
    • Tags
    • Contributor statistics
    • Graph
    • Compare revisions
  • Issues 17
    • Issues 17
    • List
    • Boards
    • Service Desk
    • Milestones
  • Merge requests 7
    • Merge requests 7
  • CI/CD
    • CI/CD
    • Pipelines
    • Jobs
    • Schedules
  • Deployments
    • Deployments
    • Environments
    • Releases
  • Monitor
    • Monitor
    • Incidents
  • Analytics
    • Analytics
    • Value stream
    • CI/CD
    • Repository
  • Wiki
    • Wiki
  • Snippets
    • Snippets
  • Activity
  • Graph
  • Create a new issue
  • Jobs
  • Commits
  • Issue Boards
Collapse sidebar
  • pycodegen
  • pystencilspystencils
  • Merge requests
  • !106

WIP: Cuda autotune

  • Review changes

  • Download
  • Patches
  • Plain diff
Closed Stephan Seitz requested to merge seitz/pystencils:cuda-autotune into master Dec 12, 2019
  • Overview 2
  • Commits 3
  • Pipelines 11
  • Changes 6

This PR introduces two one changes:

  • rotate (32,1,1) depending on field strides to fastest dimension. So (1,1,32) for c-layout and (32,1,1) for fortran layout. So pystencils will be fast also for c-layout (this will always be performed)
  • auto-tune the block dimensions to whatevers is fastest for a specific kernel on localhost. On first kernel call different layouts are tried and the kernel will be called henceforth with the fastest configuration (disk_cached). This could be intersting for OpenCL where we don't know which launch config is the fastest (on OpenCL the runtime can alternatively give a hint on that).

One drawback: the test calls are only correct if input and output fields do not overlap (so no in-place kernels).

Edited Oct 07, 2020 by Stephan Seitz
Assignee
Assign to
Reviewers
Request review from
Time tracking
Source branch: cuda-autotune