# Confusing memory allocations when using the integrator of DifferentialEquations.jl
**URL:** https://discourse.julialang.org/t/confusing-memory-allocations-when-using-the-integrator-of-differentialequations-jl/74436
**Category:** Performance
**Tags:** question, diffeq, memory-allocation
**Created:** [January 11, 2022, 9:31pm UTC](https://discourse.julialang.org/t/confusing-memory-allocations-when-using-the-integrator-of-differentialequations-jl/74436 "2022-01-11T21:31:53Z")
**Posts on this page:** 11
**Page:** 1
### Author:  [@duan](https://discourse.julialang.org/u/duan)
#### Post date: [January 11, 2022, 9:31pm UTC](https://discourse.julialang.org/t/confusing-memory-allocations-when-using-the-integrator-of-differentialequations-jl/74436/1 "2022-01-11T21:31:53Z")
Here is a prototype of my problem:
```julia
using OrdinaryDiffEq
n = 2^18;
u0 = rand(ComplexF64, n);
function f(du, u, p, t)
for i in 1:length(u)
du[i] = -1.0im * u[i]
end
return
end
mutable struct A
u0::Vector{ComplexF64}
integrator
function A(u0)
a = new()
a.u0 = u0
prob = ODEProblem(f, u0, (0, 1), nothing)
a.integrator = init(prob, Tsit5(), save_on=false, save_start=false, save_end=false, calck=false, alias_u0=true, abstol=1e-8, reltol=1e-8);
return a
end
end
function g(u1, u2)
x = 0.0im
for i in 1:length(u1)
x += u1[i] * conj(u2[i])
end
return x
end
function g2(a::A)
u1 = a.u0
u2 = a.integrator.u
x = 0.0im
for i in 1:length(u1)
x += u1[i] * conj(u2[i])
end
return x
end
a = A(u0);
step!(a.integrator, 1, true)
```
Basically, `g()` and `g2()` calculate the dot product of the initial and final states. (Yes, I could have used `dot()`, but this is a prototype.) Naively, I expect they behave similarly. It turns out that `g2()` allocates extra memory in each call. Can someone please give me some insight? Thanks!
```nohighlight
julia> @time g(a.u0, a.integrator.u)
0.000627 seconds (1 allocation: 32 bytes)
174863.6176038208 + 0.0im
julia> @time g2(a)
0.042788 seconds (1.57 M allocations: 43.992 MiB, 36.44% gc time)
174863.6176038208 + 0.0im
```
---
### Author:  [@Oscar\_Smith](https://discourse.julialang.org/u/Oscar_Smith)
#### Post date: [January 11, 2022, 9:42pm UTC](https://discourse.julialang.org/t/confusing-memory-allocations-when-using-the-integrator-of-differentialequations-jl/74436/2 "2022-01-11T21:42:57Z")
Your struct isn’t type stable. Specifically, the compiler can’t figure out what the type of `a.integrator` is.
---
### Author:  [@ChrisRackauckas](https://discourse.julialang.org/u/ChrisRackauckas)
#### Post date: [January 11, 2022, 9:50pm UTC](https://discourse.julialang.org/t/confusing-memory-allocations-when-using-the-integrator-of-differentialequations-jl/74436/3 "2022-01-11T21:50:30Z")
```julia
mutable struct A{I}
u0::Vector{ComplexF64}
integrator::I
```
---
### Author:  [@duan](https://discourse.julialang.org/u/duan)
#### Post date: [January 11, 2022, 11:13pm UTC](https://discourse.julialang.org/t/confusing-memory-allocations-when-using-the-integrator-of-differentialequations-jl/74436/4 "2022-01-11T23:13:31Z")
Indeed. Since I know the type of a.integrator.u, the problem is fixed by redefining `g2()`.
```julia
function g2(a::A)
u1 = a.u0
u2 = a.integrator.u::Vector{ComplexF64}
x = 0.0im
for i in 1:length(u1)
x += u1[i] * conj(u2[i])
end
return x
end
```
---
### Author:  [@duan](https://discourse.julialang.org/u/duan)
#### Post date: [January 11, 2022, 11:16pm UTC](https://discourse.julialang.org/t/confusing-memory-allocations-when-using-the-integrator-of-differentialequations-jl/74436/5 "2022-01-11T23:16:10Z")
It would work except that the concrete type of an ODE integrator is very complicated 😓
---
### Author:  [@ChrisRackauckas](https://discourse.julialang.org/u/ChrisRackauckas)
#### Post date: [January 11, 2022, 11:18pm UTC](https://discourse.julialang.org/t/confusing-memory-allocations-when-using-the-integrator-of-differentialequations-jl/74436/6 "2022-01-11T23:18:48Z")
That’s why you just parameterize it as `I`.
---
### Author:  [@duan](https://discourse.julialang.org/u/duan)
#### Post date: [January 11, 2022, 11:22pm UTC](https://discourse.julialang.org/t/confusing-memory-allocations-when-using-the-integrator-of-differentialequations-jl/74436/7 "2022-01-11T23:22:07Z")
In my problem I don’t know the concrete type of the integrator until during the construction of the structure …
---
### Author:  [@ChrisRackauckas](https://discourse.julialang.org/u/ChrisRackauckas)
#### Post date: [January 11, 2022, 11:29pm UTC](https://discourse.julialang.org/t/confusing-memory-allocations-when-using-the-integrator-of-differentialequations-jl/74436/8 "2022-01-11T23:29:20Z")
You don’t need to. Use parametric typing.
```julia
struct A{T}
x::T
end
A(2.0) # A{Float64}(2.0)
A(2) # A{Int}(2)
```
---
### Author:  [@duan](https://discourse.julialang.org/u/duan)
#### Post date: [January 11, 2022, 11:35pm UTC](https://discourse.julialang.org/t/confusing-memory-allocations-when-using-the-integrator-of-differentialequations-jl/74436/9 "2022-01-11T23:35:19Z")
In the code example I posted, the type of `a.integrator` is known inside the constructor. It is given by `init()`depending on the parameters I chose.
---
### Author:  [@ChrisRackauckas](https://discourse.julialang.org/u/ChrisRackauckas)
#### Post date: [January 11, 2022, 11:52pm UTC](https://discourse.julialang.org/t/confusing-memory-allocations-when-using-the-integrator-of-differentialequations-jl/74436/10 "2022-01-11T23:52:58Z")
```julia
mutable struct A{I}
u0::Vector{ComplexF64}
integrator::I
function A(u0)
u0 = u0
prob = ODEProblem(f, u0, (0, 1), nothing)
integrator = init(prob, Tsit5(), save_on=false, save_start=false, save_end=false, calck=false, alias_u0=true, abstol=1e-8, reltol=1e-8);
return new(u0,integrator)
end
end
```
---
### Author:  [@duan](https://discourse.julialang.org/u/duan)
#### Post date: [January 12, 2022, 12:28am UTC](https://discourse.julialang.org/t/confusing-memory-allocations-when-using-the-integrator-of-differentialequations-jl/74436/11 "2022-01-12T00:28:46Z")
Thanks for the suggestions @ChrisRackauckas. I misremembered. I actually passed `a` as the last parameter to `ODEProblem()` (replacing `nothing`) in the real code. Because Julia doesn’t support mutual reference, I defined `a.integrator` not in the constructor but in a separate function.