CQRRP (blocked CQRRPT) #58
Conversation
…tau vector on its main diagonal; Q in its current form is the same as output from geqrf.
…t was present before the output format switch.
…ctor from Cholesky QR
… plotting are in MATLAB
…to lapack functions in HQRRP, unless a workspace query is required. That is, we can get rid of the larfb wrapper if we want (but it will be tideous).
|
Our build recipes were outdated. |
|
Looks like we are using a 2020 version of MKL. |
|
@rileyjmurray I changed the way we install MKL; it now downloads and installs the most recent version. |
Update
|
Should have commented this yesterday, but I've updated MacOS-related stuff and everything works now! |
|
@rileyjmurray Since we realized that PAQR is not of interest to reproduce in RandLAPACK, i think this PR is ready to be reviewed. |
There was a problem hiding this comment.
I left a handful of comments. Please respond to each, @TeachRaccooon!
This PR adds CQRRP - an extension of CQRRPT's idea in a form of an iterative algorithm for QRCP that uses randomization. The output format is identical to that of GEQP3. Most of the algorithm's computations are taking place in the original matrix A's space, as described in detail in the algorithm's comments. We have also updated rl_hqrrp.cc, allowing to switch between GEQRF+LARFT (original strategy for getting Q, R factors) and CholQR+ORHR_col. We also add Kahan matrix generator and overhaus some utility routines. Updated yaml workflow file to be in synch with the most recent LAPACK++/BLAS++ build recipes and to pull the most recent MKL version. --------- Co-authored-by: TeachRaccooon <[email protected]>
This PR adds CQRRP - an extension of CQRRPT's idea in a form of an iterative algorithm for QRCP that uses randomization. The output format is identical to that of GEQP3.
Most of the algorithm's computations are taking place in the original matrix A's space, as described in detail in the algorithm's comments.
We have also updated rl_hqrrp.cc, allowing to switch between GEQRF+LARFT (original strategy for getting Q, R factors) and CholQR+ORHR_col.
We also add Kahan matrix generator and overhaus some utility routines.
Updated yaml workflow file to be in synch with the most recent LAPACK++/BLAS++ build recipes and to pull the most recent MKL version.