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.

Sign up for GitHub

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

Only preallocated one vector for preconditioners #187

Merged
merged 2 commits into from
Mar 24, 2020
Merged

Only preallocated one vector for preconditioners #187

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
ba10ab7
Only preallocated one vector for preconditioners
amontoison Mar 23, 2020
e1c1b84
Add a comment about how preconditioners are preallocated
amontoison Mar 24, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 3 additions & 1 deletion src/variants.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,7 +4,9 @@ using LinearAlgebra
function wrap_preconditioners(kwargs)
if (haskey(kwargs, :M) && typeof(kwargs[:M]) <: AbstractMatrix) || (haskey(kwargs, :N) && typeof(kwargs[:N]) <: AbstractMatrix)
k = keys(kwargs)
v = Tuple(typeof(arg) <: AbstractMatrix ? PreallocatedLinearOperator(arg) : arg for arg in values(kwargs))
# Matrix-vector products with Mᵀ and Nᵀ are not required, we can safely use one vector for products with M / Mᵀ and N / Nᵀ
Copy link
Member

Choose a reason for hiding this comment

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

Please explain that that's why you set symmetric=true somewhere. Otherwise, I'm sure we'll forget and puzzle over this in 6 months time.

# One vector for products with M / Mᵀ and N / Nᵀ is used when the option symmetric is set to true with a PreallocatedLinearOperator
v = Tuple(typeof(arg) <: AbstractMatrix ? PreallocatedLinearOperator(arg, symmetric=true) : arg for arg in values(kwargs))
Copy link
Member

Choose a reason for hiding this comment

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

With GMRES & co., preconditioners could be non-symmetric.

Copy link
Member Author

Choose a reason for hiding this comment

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

Yes, but we only need M⁻¹ * v products. I did the same thing with A for CGS, DIOM and DQGMRES even if A is not symmetric.

Copy link
Member

Choose a reason for hiding this comment

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

I see. It's worth adding a comment around your last change to explain that.

Copy link
Member Author

Choose a reason for hiding this comment

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

With the option symmetric=true, we preallocate matrix-vector products with A and A' with the same vector. If we don't use / need products with A' it's the good thing to do even if A is not symmetric.

kwargs = Iterators.Pairs(NamedTuple{k, typeof(v)}(v), k)
end
return kwargs
Expand Down