Decrease allocations done

[Datseris]

A standard evolve call has at least 100,000 allocations even though everything is done through SVectors.

I do not know how or why this happens and also how to solve it.

Just do

using DynamicalBilliards, BenchmarkTools
bt = billiard_sinai()
p = randominside(bt)
evolve!(p, bt, 1.0) # compile
@btime evolve!(p, bt, 1000.0)

to see what I mean.

[dapias]

What kind of billiard bt is?

[Datseris]

It doesn't really matter. The allocations come from evolve! so I am quite sure they happen in any combination of billiard tables.

And so it should be, since all the methods are identical no matter the obstacles. What changes is only sings and numerical values...

[Datseris]

I guess the answer is simply "SVectors result in allocations" ? I cannot know for sure though.

What I know for sure is that a lot of SVectors are created all the time during evolve!

[dapias]

Check out this case:

using DynamicalBilliards, BenchmarkTools

space = 3.0
r = 1.0
polygon_sides = 6
bth = billiard_polygon(polygon_sides, space/sqrt(3); setting = "periodic")
d = Disk([0.,0.], r)
push!(bth, d)
t = 1000.0
p = randominside(bth)
@btime evolve!(p, bth, t)

I get

478.572 μs (7459 allocations: 137.20 KiB)

after having compiled once.

PS: Sometimes in Sydney, the connection to github is down. I wanted to share this piece of code two days ago...

[dpsanders]

Note that when using the benchmarking tools, any global variables should be preceded by a $, as e.g.
@btime evolve!($p, $bth, $t).
However, I don't think this will make much difference here.

[dapias]

Oh ok. Thank you @dpsanders

[Datseris]

This is apparently due to the type instabilities that result from using Vector{Obstacle{T}} as the billiard table.

I am hoping the combo of #30 and #51 will resolve it at least partly.

[Datseris]

Closed in #72