# FFTW MKL sub dimensional FFT create plan failed

**URL:** <https://discourse.julialang.org/t/fftw-mkl-sub-dimensional-fft-create-plan-failed/81086>\
**Category:** Internals & Design\
**Tags:** fftw, mkl\
**Created:** [May 15, 2022, 8:53am UTC](https://discourse.julialang.org/t/fftw-mkl-sub-dimensional-fft-create-plan-failed/81086 "2022-05-15T08:53:43Z")\
**Posts on this page:** 4\
**Page:** 1

<div class="post-metadata">

**Author:** ![Lightup1](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/lightup1/32/35410_2.png) [@Lightup1](https://discourse.julialang.org/u/Lightup1)\
**Post date:** [May 15, 2022, 8:53am UTC](https://discourse.julialang.org/t/fftw-mkl-sub-dimensional-fft-create-plan-failed/81086/1 "2022-05-15T08:53:43Z")

</div>

MKL provider failed for plan sub dimension FFT.

```julia
test=Array{ComplexF64}(undef, [[4,4];[4,4]]...);
p1=plan_fft(test,[3,4])
result1=fft(reshape([1:4^4;],4,4,4,4))
result2=p1*reshape([1:4^4;],4,4,4,4)
result1==result2 #false means things are right

```

But for a MKL provider

```julia
julia> plan_fft!(test,[3,4])
ERROR: FFTW could not create plan
Stacktrace:
 [1] error(s::String)
   @ Base .\error.jl:33
 [2] macro expansion
   @ C:\Users\Baiyi Yu\.julia\packages\FFTW\SDUwi\src\fft.jl:576 [inlined]
 [3] (FFTW.cFFTWPlan{ComplexF64, -1, true, 4})(X::Array{ComplexF64, 4}, Y::Array{ComplexF64, 4}, region::Vector{Int64}, flags::UInt32, timelimit::Float64)
   @ FFTW C:\Users\Baiyi Yu\.julia\packages\FFTW\SDUwi\src\FFTW.jl:49
 [4] #plan_fft!#7
   @ C:\Users\Baiyi Yu\.julia\packages\FFTW\SDUwi\src\fft.jl:695 [inlined]
 [5] plan_fft!(X::Array{ComplexF64, 4}, region::Vector{Int64})
   @ FFTW C:\Users\Baiyi Yu\.julia\packages\FFTW\SDUwi\src\fft.jl:695
 [6] top-level scope
   @ REPL[45]:1

```

Does MKL not support sub-dimensional FFT?  
This kind of FFT is quite ordinary for operator of tensor

---

<div class="post-metadata">

**Author:** ![pengtu](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/pengtu/32/19411_2.png) [@pengtu](https://discourse.julialang.org/u/pengtu)\
**Post date:** [May 17, 2022, 9:49pm UTC](https://discourse.julialang.org/t/fftw-mkl-sub-dimensional-fft-create-plan-failed/81086/2 "2022-05-17T21:49:34Z")

</div>

The transform along an interior dimension (dimension 3 of 4 here) is not supported in MKL because it can’t be mapped to a constant distance between starting points of the FFTs.

MKL has a single parameter for this (DFTI\_INPUT\_DISTANCE, DFTI\_OUTPUT\_DISTANCE), implying a 1D array of starting points equally spaced at that distance. FFTW can take a multi-dimensional array of starting points through parameters called “howmany\_rank” and “howmany\_dims” in their [guru interface](https://www.fftw.org/fftw3_doc/Guru-vector-and-transform-sizes.html).

The limitation in MKL is documented in the FFTW3 interface [docs](https://www.intel.com/content/www/us/en/develop/documentation/onemkl-developer-reference-c/top/appendix-d-fftw-interface-to-onemkl/fftw3-interface-to-onemkl/using-fftw3-wrappers.html) (“the only supported values for parameter howmany\_rank in guru and guru64 plan creation functions are 0 and 1”).

---

<div class="post-metadata">

**Author:** ![N5N3](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/n5n3/32/17663_2.png) [@N5N3](https://discourse.julialang.org/u/N5N3)\
**Post date:** [May 18, 2022, 3:33am UTC](https://discourse.julialang.org/t/fftw-mkl-sub-dimensional-fft-create-plan-failed/81086/3 "2022-05-18T03:33:11Z")

</div>

You could use some lazy reshape “trick” to make MKL work on this example.  
(As dim1 and dim2 could be merged.)  
Ideally we could let `FFTW.jl` do this for us, but `plan_fft(test,[2,4])` is still “broken”.  
We need to loop dim3 mannually in this case, but it’s hard to make it efficient on all size.

Edit: If you can’t use FFTW backend for some reason, and you don’t need `rfft`.  
You can try this patched version: [GitHub - N5N3/FFTW.jl at SelfUse](https://github.com/N5N3/FFTW.jl/tree/SelfUse).

---

<div class="post-metadata">

**Author:** ![Lightup1](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/lightup1/32/35410_2.png) [@Lightup1](https://discourse.julialang.org/u/Lightup1)\
**Post date:** [June 25, 2022, 7:23am UTC](https://discourse.julialang.org/t/fftw-mkl-sub-dimensional-fft-create-plan-failed/81086/4 "2022-06-25T07:23:31Z")

</div>

Thank you all for the detailed explanations!

The reason I raise the question is that in my personal computer I find that `MKL` makes `fft` faster and support allocation free multithread.  
Now I move to `cuFFT` for large 3D fft in a cluster and it works well except that `CUDA.jl` now has no high level wrapper for multi-gpu fft support. `PencilFFT` now supports `CuArray` but it is complicated to implement if the cluster do not have cuda-aware `MPI` installed.

Hope that someday `CUDA.jl` and `MPI.jl` can just implement cuda-aware `MPI` with artifacts.