Strange Memory Allocations with saveat

[MicheleCeresoli]

Hello, I'm observing very strange and large memory allocations while solving a standard ODE and evaluating it at given times. However, after the first benchmark, if (and only if!) I recompile the integrated function these allocations disappear.

The function I'm integrating is:

@inbounds function dyn!(du::Vector{Float64}, u::Vector{Float64}, 
                                 p, t::Float64)
    du[1] =- u[1]
    du[2] = -u[2]
    du[3] = -u[3]
   
    nothing 
end

while test is the function that evaluates the solution at the given times:

import DifferentialEquations: Vern9, ODEProblem, solve
function test(x::Vector{Float64}, tms::Vector{Float64})

    user_prob = ODEProblem(dyn!, x, [0., 86400])
    user_sol = solve(user_prob, Vern9(), abstol=1e-10, reltol=1e-9, saveat=tms, save_everystep=false).u

    return user_sol 
end

Calling it with:

x =  [3., 5., 10];
tms = collect(LinRange(0., 86400., 86400));

using BenchmarkTools
@benchmark test($x, $tms)

yields:

BenchmarkTools.Trial: 30 samples with 1 evaluation.
 Range (min … max):  129.553 ms … 535.830 ms  ┊ GC (min … max): 11.32% … 74.42%
 Time  (median):     151.338 ms               ┊ GC (median):    14.82%
 Time  (mean ± σ):   167.691 ms ±  71.469 ms  ┊ GC (mean ± σ):  21.46% ± 11.46%

    ██                                                           
  ▆▇██▆▁▆▁▆▄▄▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▄ ▁
  130 ms           Histogram: frequency by time          536 ms <

 Memory estimate: 304.61 MiB, allocs estimate: 432505.

If i recompile dyn! the new benchmark is:

BenchmarkTools.Trial: 81 samples with 1 evaluation.
 Range (min … max):  49.294 ms … 106.972 ms  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     59.441 ms               ┊ GC (median):    0.00%
 Time  (mean ± σ):   62.334 ms ±  12.016 ms  ┊ GC (mean ± σ):  2.41% ± 4.19%

  █ ▁    ▁▁    ▃                                                
  █▆█▇▄▇▇██▇▆▇▇█▁▇▆▆▆▆▄▆▆▆▄▄▁▁▄▁▁▆▆▁▁▁▄▁▁▁▁▁▁▁▆▁▁▄▁▄▄▁▁▁▁▁▁▁▁▄ ▁
  49.3 ms         Histogram: frequency by time         98.8 ms <

 Memory estimate: 17.22 MiB, allocs estimate: 173311.

Any idea what's happening? I see the Garbage Collector is taking quite a lot of time

[ChrisRackauckas]

I've seen this in other spots, like SciML/OrdinaryDiffEq.jl#1502 . It seems like it's something to do with type inference or inlining being dependent on the method table, and so redefining functions can change what inlines and what does not. But it seems like a core Julia thing and I'm not sure what we could specifically do to avoid it (see some of the links to JuliaLang/julia issues, i.e. JuliaLang/julia#32552 and JuliaLang/julia#35800).

I'm closing this as a case of a Base Julia issue, and thus it's not actionable in this library, but I'll ask some of the folks working on inference what's the current status of fixing this.