add faster sum/any/all/iszero

[SobhanMP]

closes #237
i'll write tests tomorrow

would it make sense that the functions for transpose/adjoint be added to LinearAlgebra.jl?

can nonzeros be larger than nnz? there are code where it would be valid and code that it would not be.

[dkarrasch]

But LinearAlgebra doesn't know about SparseXYZ. I think these functions should live here. I remember we have made an effort to implement efficient reduction functions. Don't they apply here? Maybe there is some dispatch going in the wrong direction? Looking at sparsevector.jl lines 1425ff, there is sparsity-aware code.

[dkarrasch]

Wait. This doesn't apply to SparseVectors, but to sparse matrices? But then again, we have code in sparsematrix.jl lines 2132ff, and seemingly a specialization for +.

[dkarrasch]

I believe this is all actually handled well, see my comment in the original issue. What is missing is a count specialization for Bool sparse arrays.

[SobhanMP]

But LinearAlgebra doesn't know about SparseXYZ.

sure but what ever the parent is, shouldn't Transpose just forward it to the parent? like always as these are not position aware functions

[dkarrasch]

I see what you mean. Like

iszero(x::AdjOrTrans) = iszero(parent(x))

right?

[SobhanMP]

yeah

[SobhanMP]

closed in favor #244