I realized BLAS.gbmv works with both orderings using the âTâ option, so its probably worth supporting both formats.
Anyone know if there is a BandedMatrix*BandedMatrix routine in BLAS/LAPack? I have it implemented in Julia, but it would probably be a lot faster using an inbuilt version. > On 29 Mar 2016, at 2:25 AM, Tim Holy <[email protected]> wrote: > > On Sunday, March 27, 2016 06:43:19 PM Sheehan Olver wrote: >> I've split out the BandedMatrix data structure in BandedMatrices.jl: >> >> https://github.com/ApproxFun/BandedMatrices.jl >> >> Right now the entries are in the "wrong" order for LAPack: the rows are >> stored (instead of the columns) . I'm debating whether to support both >> orderings, or replace the current implementation with the "correct" column >> version. > > You internal debate is also going on elsewhere: > https://github.com/JuliaLang/julia/issues/4774#issuecomment-149349751 > https://github.com/JuliaLang/julialang.github.com/blob/teh/arrays/blog/_posts/2016-03-27-arrays-iteration.md#row-major-matrices-permuteddimensionarrays-and-taking-transposes-seriously > > Best, > --Tim > >> >> On Thursday, March 24, 2016 at 9:10:12 AM UTC+11, Tony Kelman wrote: >>> +1 for a specific standalone BandedMatrices package. Once it's up and >>> running and registered, could even move the Lapack bindings for gb* >>> functions out to the package (and pb* and sb* if you also have symmetric), >>> since we don't have a type for them in base. I think it'll be important to >>> work out how to get base to support additional sparse and structured >>> matrix >>> formats well, without needing to have the types themselves in base. >>> >>> On Wednesday, March 23, 2016 at 2:43:16 PM UTC-7, Sheehan Olver wrote: >>>> It would be great to have a broader SparseFormats.jl package. >>>> >>>> Maybe in the time being I'll start it as BandedMatrices.jl (since that's >>>> already implemented) which can be renamed later. >>>> >>>> Sheehan >>>> >>>> Sent from my iPhone >>>> >>>> On 24 Mar 2016, at 03:07, Christoph Ortner <[email protected]> wrote: >>>> >>>> Here is a short list that in the long-run would be useful: >>>> * compressed row >>>> * triplet >>>> * block-sparse matrices >>>> (* linked-list variants of CC and CR??) >>>> >>>> C >>>> >>>> On Wednesday, 23 March 2016 13:53:30 UTC, Tim Holy wrote: >>>>> Base has UniformScaling, Diagonal, Bidiagonal, Tridiagonal, >>>>> SymTridiagonal, >>>>> UpperTriangular, LowerTriangular, in addition to SparseVector and >>>>> SparseMatrixCSC. But not general banded. >>>>> >>>>> --Tim >>>>> >>>>> On Wednesday, March 23, 2016 04:19:22 AM Christoph Ortner wrote: >>>>>> I'd be interested as well; more generally: does Julia still only have >>>>> >>>>> CCS >>>>> >>>>>> implemented? For me it would sound productive to start a general >>>>>> SparseFormats.jl (or similar) package and collect various other >>>>> >>>>> formats, >>>>> >>>>>> which could eventually be incorporated into Base or a >>>>> >>>>> standard-library. >>>>> >>>>>> Christoph >
