That's right. It is there as an additional knob when you need it. For large 
matrices where 32-bit indices suffice, this leads to useful space savings 
and should also have lower indexing overhead inside sparse solvers.

-viral

On Thursday, July 10, 2014 7:32:26 AM UTC-7, Mauro wrote:
>
> It's the type of the index variables.  So, potentially useful to save 
> some space by using Uint32 or Uint16 for them.  Of course that reduces 
> the maximal size of the matrix. 
>
> There is an "example" in test/sparse.jl: 
> S1290 = SparseMatrixCSC(3, 3, Uint8[1,1,1,1], Uint8[], Int64[]) 
>
> I think you have to construct it with SparseMatrixCSC, as `sparse` just 
> uses Int. 
>
> On Thu, 2014-07-10 at 13:52, Simon Byrne <[email protected]> wrote: 
> > Out of curiosity, what is the purpose of the Ti parameter in 
> > SparseMatrixCSC type? 
> > 
> > I managed to track down this old commit: 
> > 
> https://github.com/JuliaLang/julia/commit/a0961e5df69da9f736ac8cb54a1186433fa6c1ba
>  
> > which suggests it was introduced in order to conserve space for small 
> > matrices, but now it seems to always be the default Int type. 
> > 
> > -simon 
>
>

Reply via email to