Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

batched - dense: Testing and fixing Serial QR #2342

Open
wants to merge 4 commits into
base: develop
Choose a base branch
from

Conversation

lucbv
Copy link
Contributor

@lucbv lucbv commented Sep 19, 2024

The serial QR algorithms does not have unit-tests and is failing for non square matrices. See issue #2328.
This first commit fixes the issue with rectangular matrices and adds a basic test for that use case.
Next will work on adding a test that exercises the interfaces on multiple matrices of different sizes within a parallel_for. Finally equivalent tests will be added for the square case as well.

Tasks:

  • analytical test on rectangular matrix (SerialQR, ApplyQ, FormQ, Q*Q^t=I)
    • SerialQR
    • ApplyQ
    • FormQ
  • multiple variable size rectangular matrices (same API as above)
    • SerialQR
    • ApplyQ
    • FormQ
  • analytical test on square matrix
    • SerialQR
    • ApplyQ
    • FormQ
  • multiple variable size square matrices
    • SerialQR
    • ApplyQ
    • FormQ

Copy link
Contributor

@cwpearson cwpearson left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

edit - duplicated comment

Copy link
Contributor

@cwpearson cwpearson Oct 10, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can any of these structs be moved to Impl while you're at it? e.g. SerialQR_FormQ_Internal. Or would you prefer to do that as a separate deprecating PR?

}
#endif

// #if defined(KOKKOSKERNELS_INST_COMPLEX_FLOAT)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Leave an explanation for why this is disabled.

@lucbv lucbv self-assigned this Oct 15, 2024
@lucbv lucbv force-pushed the batched_QR_fixes branch 2 times, most recently from 18fc349 to b24686b Compare November 6, 2024 20:45
lucbv added 2 commits December 4, 2024 13:39
The serial QR algorithms does not have unit-tests and is failing
for non square matrices. See issue kokkos#2328.
This first commit fixes the issue with rectangular matrices and
adds a basic test for that use case. Next will work on adding a
test that exercises the interfaces on multiple matrices of different
sizes within a parallel_for. Finally equivalent tests will be added
for the square case as well.
Fixing unused variable error
It looks like the Left NoTranspose ApplyQ is not doing the correct
thing. Will have a look at that next.

Signed-off-by: Luc <[email protected]>
Signed-off-by: Luc Berger-Vergiat <[email protected]>
We did not pass the stride of the work array to internal routines
and we are not enforcing a contiguous memory allocation either so
when using subviews to pass the work array, we run into troubles.

Signed-off-by: Luc Berger-Vergiat <[email protected]>
@lucbv lucbv removed the AT: WIP label Dec 4, 2024
After fixing the issue with the workspace utilization now
looking at issues when the problem and batch sizes are increased.

Signed-off-by: Luc Berger-Vergiat <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants