# Help making function faster

**URL:** <https://discourse.julialang.org/t/help-making-function-faster/3793>\
**Category:** General Usage\
**Created:** [May 18, 2017, 9:18pm UTC](https://discourse.julialang.org/t/help-making-function-faster/3793 "2017-05-18T21:18:30Z")\
**Posts on this page:** 4\
**Page:** 1

<div class="post-metadata">

**Author:** ![favba](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/favba/32/2735_2.png) [@favba](https://discourse.julialang.org/u/favba)\
**Post date:** [May 18, 2017, 9:18pm UTC](https://discourse.julialang.org/t/help-making-function-faster/3793/1 "2017-05-18T21:18:30Z")

</div>

In order to save memory in my computations I created a module with array wrappers (named TransformableFields) and in-place rfft! and irfft!.  
The module can be found here [InPlaceRealFFT.jl](https://gist.github.com/favba/c43987780b67f5bed089e4060bde8b87). There is no dependencies, so It can be just copied and one is able to use it.  
I am using fft’s to compute derivates (among other things), but the inplace version is given much slower results. Here is a sample code

```julia
using InPlaceRealFFT

function rfftfreq(n::I,s::R) where {I<:Integer,R<:Real}
  d = 2π*s/n
  [(n/2 - i)/(d*n) for i = n/2:-1:0]
end

function dx(field::A,len::Float64,n=1::Int) where {T<:Union{Float32,Float64},A<:AbstractArray{T,3}}
  nx,ny,nz = size(field)
  fieldhat = rfft(field,1)
  kim = (rfftfreq(nx,len) .* im) .^ n

  Threads.@threads for l = 1:nz
    for j = 1:ny
      for i = 1:length(kim)
        @inbounds fieldhat[i,j,l] = fieldhat[i,j,l]*kim[i]
      end
    end
  end
  return irfft(fieldhat,nx,1)
end

function dx!(field::TransformableField{T,3},len::T,n=1::Int) where T<:Union{Float64,Float32}

  nx,ny,nz = size(field.r)

  rfft!(field,1)

  kim = (rfftfreq(nx,len) .* im) .^ n
  fieldhat = field.c
  nx,ny,nz = size(field)
  Threads.@threads for l = 1:nz
    for j = 1:ny
      for i = 1:nx
        @inbounds fieldhat[i,j,l] = fieldhat[i,j,l]*kim[i]
      end
    end
  end
  return irfft!(field,1)
end

lx=1.0;
a = randt(1024,512,512); # randomly generated TransformableField
b = copy(a.r); # b Simple Array
@time c = dx(b,lx);
@time dx!(a,lx);

```

If I run this script I have the following output:

```julia
21.337080 seconds (1.60 M allocations: 4.088 GiB, 2.14% gc time)
93.240138 seconds (777.40 M allocations: 27.121 GiB, 3.63% gc time)

```

It seems to me that `dx!` is spending most of the time in the for loop. I can’t figure out why that is allocating so much.

Probably both functions can be optimised, but I expected that they would give similar results the way they are written now.

This was run using Julia 0.6rc1.

---

<div class="post-metadata">

**Author:** ![ksmcreynolds](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/ksmcreynolds/32/860_2.png) [@ksmcreynolds](https://discourse.julialang.org/u/ksmcreynolds)\
**Post date:** [May 19, 2017, 2:18pm UTC](https://discourse.julialang.org/t/help-making-function-faster/3793/2 "2017-05-19T14:18:29Z")

</div>

The issue appears to be due to the behavior of threads in 0.6. I’ve had a similar problem (see [Multithreading performance regressions in 0.6?](https://discourse.julialang.org/t/multithreading-performance-regressions-in-0-6/3748)). You can solve this by wrapping the threaded loop in its own function and calling that. Doing that I get `dx!` to be about twice as fast as `dx`.

Also you should run each function once before timing it or you will include the compilation time that only occurs on the first run.

---

<div class="post-metadata">

**Author:** ![kristoffer.carlsson](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/kristoffer.carlsson/32/22_2.png) [@kristoffer.carlsson](https://discourse.julialang.org/u/kristoffer.carlsson)\
**Post date:** [May 19, 2017, 2:32pm UTC](https://discourse.julialang.org/t/help-making-function-faster/3793/4 "2017-05-19T14:32:29Z")

</div>

The code that is profiled is inside a function?

---

<div class="post-metadata">

**Author:** ![favba](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/favba/32/2735_2.png) [@favba](https://discourse.julialang.org/u/favba)\
**Post date:** [May 19, 2017, 7:29pm UTC](https://discourse.julialang.org/t/help-making-function-faster/3793/6 "2017-05-19T19:29:40Z")

</div>

Thank you, that was it. I also get `dx!` to be about twice as fast as `dx` now.