Skip to content
GitLab
Projects Groups Snippets
  • /
  • Help
    • Help
    • Support
    • Community forum
    • Submit feedback
    • Contribute to GitLab
  • Sign in
  • Chameleon Chameleon
  • Project information
    • Project information
    • Activity
    • Labels
    • Members
  • Repository
    • Repository
    • Files
    • Commits
    • Branches
    • Tags
    • Contributors
    • Graph
    • Compare
  • Issues 18
    • Issues 18
    • List
    • Boards
    • Service Desk
    • Milestones
  • Merge requests 11
    • Merge requests 11
  • Deployments
    • Deployments
    • Releases
  • Packages and registries
    • Packages and registries
    • Package Registry
    • Container Registry
    • Infrastructure Registry
  • Monitor
    • Monitor
    • Incidents
  • Analytics
    • Analytics
    • Value stream
    • Repository
  • Activity
  • Graph
  • Create a new issue
  • Commits
  • Issue Boards
Collapse sidebar
  • solverstack
  • ChameleonChameleon
  • Issues
  • #74
Closed
Open
Issue created Jan 17, 2019 by Ghost User@ghost

Slow runs of ZLANGE for Max and Frobenius norms

Hello!

I think I found a bug in workspace allocation of chameleon_pzlange_generic function, lines 438-440 and 450-452. Idea behind these descriptors is to store result of LANGE operation on each tile. So, it should be MT-by-NT matrix for Max norm (maximum per each block) and 2*MT-by-NT matrix for Frobenius norm (scale and scaled sum of squares per each block). However, what I see in the code is allocation of MT-by-N and 2*MT-by-N matrices. Fix for this is obvious, so I am not putting any pull request.

My test code was running for 12 seconds to compute norm of 10k-by-10k matrix before fix and 0.5 seconds after this fix.

Edit: chameleon_pzlansy_generic is also corrupted by this bug.

Edited Jan 17, 2019 by Ghost User
To upload designs, you'll need to enable LFS and have an admin enable hashed storage. More information
Assignee
Assign to
Time tracking