longemen3000 / cachedfunctions.jl Goto Github PK

This issue is used to trigger TagBot; feel free to unsubscribe.

If you haven't already, you should update your TagBot.yml to include issue comment triggers.
Please see this post on Discourse for instructions and more details.

If you'd like for me to do this for you, comment TagBot fix on this issue.
I'll open a PR within a few hours, please be patient!

using CachedFunctions
function f!(y,x)
    y[1] = exp(sum(x))
    y[2] = x[1]+x[2] - x[3]
    y
end

x1 = rand(3)
x2 = rand(3)
x3 = rand(Float32,3)
x4 = rand(Float32,3)

f = CachedFunction(f!,3,2)

Returns

ERROR: MethodError: no method matching prototype(::Int64)

I'm using Version 1.5.4 (2021-03-11)

Thanks a lot for the package, I think this will really help me reduce allocation-based slowdown in my application.

One incompatibility I have is that my functions are for dynamical systems and are of the form f!(y,x,t), i.e. I pass in the time t::Number in addition to the state vector x::Vector, and y=similar(x).

This seems like the sort of thing Julia should be able to generalize to cover really easily, but I can't see how to force this into the mold of a single input state vector without allocating push!(x,t) which defeats the purpose of caching.

as mencioned by tkf in discourse, there is a need to make this structure Thread-safe. there are some approaches to this:

• lock on call: everytime that a call is done, lock, wait for the result and unlock. but this has the obvious disadvantage that in a multithreaded setting, the execution would be linear, defeating the purpose
• duplicate caches: each thread can only access one cache. that maybe be acomplished by using the key(eltype(x),Val(threadId)). What i don't know is if the get call to the thread function is safe itself.

To test this, a good example is also necessary