Add needs_square_A trait
#400
Conversation
Codecov Report
@@ Coverage Diff @@
## main #400 +/- ##
==========================================
- Coverage 68.95% 68.37% -0.58%
==========================================
Files 26 26
Lines 1907 1929 +22
==========================================
+ Hits 1315 1319 +4
- Misses 592 610 +18
... and 2 files with indirect coverage changes 📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more |
avik-pal
force-pushed
the
ap/needs_square_A
branch
from
d814042
to
0cf30e1
Compare
avik-pal
force-pushed
the
ap/needs_square_A
branch
from
0cf30e1
to
988e3f7
Compare
src/LinearSolve.jl
Outdated
| function for custom algorithms! | ||
| """ | ||
| needs_square_A(::Nothing) = false # Linear Solve automatically will use a correct alg! | ||
| function needs_square_A(alg::SciMLLinearSolveAlgorithm) |
There was a problem hiding this comment.
Just remove the fallback please.
There was a problem hiding this comment.
Or rather, the fallback is true. That's a safe fallback, methods can opt-in to supporting non-square, but anything that supports non-square supports square so that's a safe assumption.
| DefaultAlgorithmChoice.QRFactorization | ||
| if is_underdetermined(A) | ||
| # Underdetermined | ||
| DefaultAlgorithmChoice.QRFactorizationPivoted |
There was a problem hiding this comment.
QRFactorization is already pivoted?
There was a problem hiding this comment.
oh I see. @YingboMa what do you think of just always defaulting to pivoted QR?
There was a problem hiding this comment.
Yeah, pivoted QR handles degenerate systems too.
There was a problem hiding this comment.
But it's quite a bit slower though.
avik-pal
force-pushed
the
ap/needs_square_A
branch
from
23297f9
to
45adbd0
Compare
Makes the following changes:
needs_square_Atrait for solvers