Concretization cleanup #183
Conversation
| stencil_pivot = div(stencil_length,2) | ||
| end | ||
| SparseArrays.SparseMatrixCSC(A::DerivativeOperator{T}, N::Int=A.len) where T = | ||
| copyto!(spzeros(T, N, N+2), A, N) |
There was a problem hiding this comment.
This would be really slow?
There was a problem hiding this comment.
Why would it? I overloaded Base.copyto! at the top of the file, the code of which is exactly the same as was in all three versions for different target matrices.
I might add we've used this pattern to great effect in ContinuumArrays.jl et al.
There was a problem hiding this comment.
Adding to a sparse matrix using indexing is really slow: the indices should all be declared up front.
There was a problem hiding this comment.
oh I see, ehh, it's not any worse than before, but it's still bad. If no one wants to actually do the (I,J,V), then it's best to probably just build the banded matrix and then convert.
There was a problem hiding this comment.
Sure :) I was anyway interested in the banded versions, but could not keep my hands off refactoring this.
There was a problem hiding this comment.
Yeah, 99% of the time people should just call sparse and get the banded version.
Before starting to look at #182, I refactored the concretization, since so much was in common between the different array types.
Impacts #173.