Allpass interpolated delay (/analog toolkit) re-optimization

[Tepeix]

Hi,
here's a re-optimization of the allpass interpolation delay.

Like the "Analog toolkit Interpolated Delay", the "Allpass interpolated delay Trog" or the interpolated delay prim..
(Don't know if another version exist)

(It preserve more the high that a normal linear interpolation)

I try to check the cpu.
But cpu measurement are not so much reliable.. So i'm not sure if what i see is true.
But it must be logicaly faster, not using the 80 bit register.

[Tepeix]

And here's another version that is PolySafe for use inside a synth.

It works !) But it was not easy.
I was misleading by the fact that the delay need another envelope.
(This is more audible when using longer delay)

I used the envelope after the delay and the osc..
This way the delay record an osc with no envelope which would click..

This one need 2 part. The index must be calculated externally in blue to avoid any reset on new voice.
If others delays have the same memory size they could be feed with the same index.

[Tepeix]

Edit.. Sorry Finally there's some unsolved problem with the poly safe version which is not poly safe..
Seams to work but not really.. It's like some note are not processed..
I hope one day i find a fix...

[Tepeix]

So finally it work in blue.
But in white/poly this optimization do something wrong.
I'm not sure what..
My current theory is that when it write constantly in a array in white/poly, it must write the four sse channel one after another.
When it try to write the four sse in one instruction, the new voice will not initialize a new part of the array because it see that this part is writing.

Here's an example with the Trog delay working and mine doing bad thing with multiple note.

[Tepeix]

Oups... I'm very very sorry...

I have to test and debug more my code before to post them !!

So my big error was to not test if the four channel react well.
Testing them with the same input and delay..

But it was reading only the first channel SSE so the input was mono !(

This could also have confuse me for the poly thing..

Here's a four channel working version :

[Tepeix]

Strange.. Now i read that allpass interpolation could produce some click.
But i never hear some..

Maybe i don't understand right this article.. Or maybe it is for very large abrupt change ?

What did you think about it ?

https://quod.lib.umich.edu/cgi/p/pod/do ... format=pdf