Júlio Hoffimann (Feb 08 2024 at 21:20): Júlio Hoffimann (Feb 08 2024 at 21:20):The Transducers.jl tcollect function is super useful, but it seems to always return a Vector regardless of the input type.
Is there a way to preserve the input type with Transducers.jl or Base.Threads?
julia> using Transducers
julia> using CircularArrays
julia> tcollect((1,2,3))
3-element Vector{Int64}:
1
2
3
julia> tcollect([1,2,3])
3-element Vector{Int64}:
1
2
3
julia> tcollect(CircularVector([1,2,3]))
3-element Vector{Int64}:
1
2
3
The underlying issue is that we have a couple of map(fun, iter) that we would like to parallelize. But replacing map with tcollect doesn't work as expected.
Mason Protter (Feb 08 2024 at 22:26):Yeah, this is something Transducers.jl doesn't really handle well. I think the right function here is tcopy since that doesn't try to follow the behaviour of collect
Mason Protter (Feb 08 2024 at 22:27):but trying your examples with tcopy, none of them work still
Mason Protter (Feb 08 2024 at 22:27):which I guess we should classify as bugs that need to be fixed.
Mason Protter (Feb 08 2024 at 22:27):The circular arrays one could potentially be a package extension
Mason Protter (Feb 08 2024 at 22:28):but this is in general a hard problem
Mason Protter (Feb 08 2024 at 22:29):I would suggest though that if your problem is so time consuming that it requires parallelism, then maybe it's not so bad to have to do a convert after, i.e.
julia> Tuple(tcollect((1,2,3)))
(1, 2, 3)
julia> CircularVector(tcollect(CircularVector([1,2,3])))
3-element CircularVector(::Vector{Int64}):
1
2
3
but I get that's also annoying
Júlio Hoffimann (Feb 08 2024 at 22:33):Yes, we ended up following this route.
Júlio Hoffimann (Feb 08 2024 at 22:35):It would be super nice if these things worked. Also, is it too big of a dream to imagine a future where we can just replace map by a pmap and choose the form of parallelism? Threads vs GPU threads vs processes...
Júlio Hoffimann (Feb 08 2024 at 22:35):I wish we had a general pmap that performed all sorts of parallelism with keyword options
Mason Protter (Feb 08 2024 at 22:39):Also, is it too big of a dream to imagine a future where we can just replace map by a pmap and choose the form of parallelism? Threads vs GPU threads vs processes...
That's what Transducers.jl / Folds.jl already does. They provide sequential, threaded, and distributed backends
Mason Protter (Feb 08 2024 at 22:39):Taka worked on a GPU backend but it's broken now
Mason Protter (Feb 08 2024 at 22:39):would be great if we could revive it
Júlio Hoffimann (Feb 08 2024 at 22:40):Perhaps it is an issue of documentation then. I never saw the GPU case for instance
Mason Protter (Feb 08 2024 at 22:41):https://github.com/JuliaFolds/FoldsCUDA.jl
Mason Protter (Feb 08 2024 at 22:41):It was always experimental though
Mason Protter (Feb 08 2024 at 22:42):but yes, the idea is that Transducers can give us a very general way of doing parallelism that can be re-implemented for many different backends
Last updated: Dec 28 2024 at 04:38 UTC