Support for Diagonal
#50
Conversation
Codecov Report❌ Patch coverage is
... and 3 files with indirect coverage changes 🚀 New features to boost your workflow:
|
lkdvos
force-pushed
the
diagonal
branch
2 times, most recently
from
6e119a5 to
0a8df6b
Compare
| end | ||
|
|
||
| p .+= (0:(n - 1)) .* n | ||
| U[p] .= Ref(one(eltype(U))) |
There was a problem hiding this comment.
I don't think this Ref is necessary due to the broadcasting behaviour of scalars, but I assume it also has no cost. Other than this, I have no comments and think this looks great.
There was a problem hiding this comment.
Semi related, why not use @. broadcast syntax to minimize the number of array iterations?
There was a problem hiding this comment.
I think I was thinking when implementing this that since the Diagonal implementation really requires very little from the scalars, it might be that someone at some point uses this for some exotic type that actually has some unwanted broadcasting behavior, so better to be explicit about it. Realistically though, this totally doesn't matter 🙃
Here I've added specialized implementations for working with
Diagonals, through a dedicated backend.