# FFTW plan applied to array with wrong memory alignment

**URL:** https://discourse.julialang.org/t/fftw-plan-applied-to-array-with-wrong-memory-alignment/101672
**Category:** Signal and Image Processing
**Tags:** question
**Created:** [July 16, 2023, 9:43am UTC](https://discourse.julialang.org/t/fftw-plan-applied-to-array-with-wrong-memory-alignment/101672 "2023-07-16T09:43:30Z")
**Posts on this page:** 1
**Page:** 1

<div class="post-metadata">

### Author: ![ymtoo](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/ymtoo/32/18859_2.png) [@ymtoo](https://discourse.julialang.org/u/ymtoo)
#### Post date: [July 16, 2023, 9:43am UTC](https://discourse.julialang.org/t/fftw-plan-applied-to-array-with-wrong-memory-alignment/101672/1 "2023-07-16T09:43:30Z")

</div>

MWE:

```julia
using ChainRulesCore: @ignore_derivatives
using DSP
using DSP.Periodograms: forward_plan
using Zygote

function mystft(x::AbstractVector{T}, n, noverlap) where {T}
    numsamples = length(x)
    numframes = numsamples >= n ? div((numsamples - n), n - noverlap) + 1 : 0
    inputtype = fftintype(T)
    outtype = fftouttype(T)
    oneton = Base.OneTo(n)
    tmp = similar(x, outtype, T <: Real ? (n >> 1)+1 : n)
    plan = @ignore_derivatives forward_plan(zeros(inputtype, n), tmp)
    map(1:numframes) do i
        offset = (i-1) * (n - noverlap)
        x1 = inputtype.(x[offset .+ oneton])
        plan * x1
    end |> stack
end

xF64 = randn(Float64, 10000)
xF32 = Float32.(xF64)

n = 128
noverlap = 64
stft_xF32 = stft(xF32, n, noverlap)
stft_xF64 = stft(xF64, n, noverlap)
mystft_xF32 = mystft(xF32, n, noverlap)
mystft_xF64 = mystft(xF64, n, noverlap)

stft_xF32 == mystft_xF32 # true
stft_xF64 == mystft_xF64 # true
mystft_xF64 ≈ mystft_xF32 # true

gradient(x -> sum(abs, mystft(x, n, noverlap)), xF64) # this works
gradient(x -> sum(abs, mystft(x, n, noverlap)), xF32) 
# ERROR: ArgumentError: FFTW plan applied to array with wrong memory alignment

```

The gradient of `mystft` with respect to Vector{Float64} works but it throws `ERROR: ArgumentError: FFTW plan applied to array with wrong memory alignment` when Vector{Float32} is the input argument. Is this a bug?

Julia and package versions:

```julia
julia> versioninfo()
Julia Version 1.9.2
Commit e4ee485e909 (2023-07-05 09:39 UTC)
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 8 × Intel(R) Core(TM) i7-10510U CPU @ 1.80GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-14.0.6 (ORCJIT, skylake)
  Threads: 4 on 8 virtual cores
Environment:
  LD_LIBRARY_PATH = /opt/ros/noetic/lib::/usr/local/cuda-11.7/lib64
  JULIA_EDITOR = code
  JULIA_NUM_THREADS = 4

(test_stft) pkg> st
Status `~/Projects/tmp/test_stft/Project.toml`
  [d360d2e6] ChainRulesCore v1.16.0
  [717857b8] DSP v0.7.8
  [e88e6eb3] Zygote v0.6.62

```