Skip to content

WIP: Use dense LAPACK routines via block-tridiagonal view for banded * and qr #159

New issue

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

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

Already on GitHub? # to your account

Open
wants to merge 3 commits into
base: master
Choose a base branch
from

Conversation

dlfivefifty
Copy link
Member

This should make QR and banded * banded operations much faster by leveraging fast dense linear algebra for blocks

@codecov
Copy link

codecov bot commented Feb 23, 2020

Codecov Report

Merging #159 into master will decrease coverage by 1.16%.
The diff coverage is 0.00%.

Impacted file tree graph

@@            Coverage Diff             @@
##           master     #159      +/-   ##
==========================================
- Coverage   72.77%   71.60%   -1.17%     
==========================================
  Files          21       22       +1     
  Lines        2516     2557      +41     
==========================================
  Hits         1831     1831              
- Misses        685      726      +41     
Impacted Files Coverage Δ
src/generic/diagblockview.jl 0.00% <0.00%> (ø)

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update da0c126...15addda. Read the comment docs.

@MasonProtter
Copy link
Member

What's needed for this to merge? Just more tests?

@dlfivefifty
Copy link
Member Author

It hasn't been done yet, it's just a bunch of experiments at the moment.

I got disheartened when I realized LAPACK's tpqrt2! is slower than just copying to a matrix and doing a dense QR.

For banded * banded, the issue is that there is apparently no BLAS triangular * triangular.

# for free to join this conversation on GitHub. Already have an account? # to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants