I'd like to suggest the Renoise team to provide "sub-sample" loop points and I'd like to ask you if this makes sense or if there's a flaw in my thinking.

I've got no idea whether Kontakt etc use floating-point loop points, do they? The issue is, that with Renoise/trackers you often set loops just to have small wavecycles, so it's a bit different use there.

When i'd like to let a sampler loop somewhere in audio data (single cycles) where only frequencies occur that are divisors or the sample frequency, then I certainly would not need to specify floating-point loop points, right?

So with less optimal (recorded, drawn) wave data, isn't it better to be able to set loop points (by trying and listening) that are floating point?

And the implementation in the sampler software should be a no-brainer, since the sampler would just need to add a value when it wraps to the beginning of the loop (cycle), correct?

Secondly, can todays samplers (Kontakt? Alchemy?) do resynthesis based on loop points? I mean, that the first part of a voice would be resampled without IFFT and after that IFFT would go on to hold the voice based on loop points. Is this hard to implement?

Oh, and: Are cross-fade loops worse or better than IFFT? As far as I could see this type of loops crossfades volume of two signals around a loop point. That seems to be simmilar to windowing wave data for use in FFT, but it probably introduces more sound artefacts? (since it only operates on time domain)

Any help would be appreciated, thanks a lot!

A problem with sub-sample loop points is the exact waveform between the samples depends on the type of interpolation used for playback, so while it's good for single-cycle loops (better than rounding to the nearst sample and being out of tune) you can't rely on the loop sounding the same in different samplers or in different interpolation/upsampling modes of the same sampler.

true, but once you have selected a particular interpolation method, i'd say, it makes a lot of sense to set the loop points to the zero-crossings of the interpolant. an option around this problem could be to tell the sampler: the loop point is between sample n and n+1 and let the sampler automatically choose the correct fractional value between n and n+1, taking the selected interpolator into consideration.

edit: such a strategy would be only appropriate for longer loops, containing many cycles of the pitch period. for single-cycle loops, the distance between the loop-start and loop-end needs to be rigidly fixed, otherwise detuning would occur (edit2: ...which, on the other hand, could also be compensated by the scanning speed).

AFAIK AKAI sampler allow loop fine tuning.

Still what people usually do nowadays is to have some xfade printed offline for smooth loops