The devectorize package https://github.com/lindahua/Devectorize.jl can deal
with this:
f(x) = @devec x[:]=2.*x
'f' won't allocate any new memory.
On Wednesday, February 19, 2014 12:11:02 PM UTC-8, Tim Holy wrote:
>
> Unfortunately, I'm pretty sure you can't do what you're hoping for.
> Consider:
>
> julia> A = rand(2,2)
> 2x2 Array{Float64,2}:
> 0.63259 0.109017
> 0.667425 0.112597
>
> julia> pointer(A)
> Ptr{Float64} @0x000000000446f6a0
>
> julia> A *= 2
> 2x2 Array{Float64,2}:
> 1.26518 0.218033
> 1.33485 0.225195
>
> julia> pointer(A)
> Ptr{Float64} @0x0000000005696650
>
> I think A *= 2 gets translated by the parser into A = A*2. There are some
> old
> posts that can do a better job than I explaining why this is important.
>
> So I think your only bet is to do something like this:
>
> for i = 1:length(A)
> A[i] *= 2
> end
>
> which does work in-place.
>
> --Tim
>
> On Wednesday, February 19, 2014 11:26:48 AM Madeleine Udell wrote:
> > Currently in-place arithmetic operations turn shared arrays into normal
> > ones: eg
> >
> > b = Base.shmem_randn(10)
> > b *= 2
> >
> > results in b becoming a (not-shared) array. I'd like to overload these
> > in-place arithmetic operations so that instead of writing
> >
> > typeof(b)==SharedArray? b.s *= 2 : b *= 2
> >
> > I can just write
> >
> > b *= 2
> >
> > and know that the shared status of my array will be preserved. How might
> > one overload these operations?
> >
> > (Note: I don't want c = b + b to automatically create a shared array c,
> > since that allocates new shared memory.)
>
