Elliott Mahler

One thing worth looking into with all of this: When comparing benchmarks to rustFFT, is the difference in performance between power2 sizes down to algirthms, or is it down to...

I've done some exploration into split radix with AVX, and I've got promising results. According to the benchmarks, my hardcoded sizes are actually faster than FFTW, Larger split radix sizes...

I implemented a few different hardcoded sizes using both mixed radix and split radix. For example I wrote both a split radix size 16, and a mixed radix 4x4. And...

Worth pointing out: The "prime factor algorithm" in Fourier isn't mathematically accurate: https://en.wikipedia.org/wiki/Prime-factor_FFT_algorithm The prime factor algorithm is the name for a specific algorithm where no twiddle factors are necessary,...

Not directly. But you can compute a Nd FFT by computing FFTs for the rows, transposing, doing FFTs for the columns, transposing again, etc. check out the “transpose” crate for...

I’ll have to look into it but my understanding is that aside from incremental optimizations, you aren’t going to get any crazy improvements by using a specialized algorithm On Sun,...

One approach that could take less time is: For each dimension, copy strided elements into a contiguous scratch buffer, then compute the FFT, then copy them back into their original...

> Which brings up another question - going from Box to Box is easy because you just move the offset. But what if you try to go from Box to...

I just ran my head into this, and it's left me pretty disappointed. I'm developing the [rust_dct](https://github.com/ejmahler/rust_dct) crate. Each of the Discrete Cosine transforms type 1 through for has its...

> Alternately… do you actually need to be using trait objects in the first place? Just from a glance at your use case, my guess is that having my_algorithm* use...