Something to keep Spogg busy during self-isolation

[deraudrl]

Try this: https://www.dropbox.com/sh/4ct0soq9v0j3 ... zZnLa?dl=0
and let me know if it works...haven't used Dropbox in about a decade.

[tulamide]

I couldn't get CubicPadSynth to run either. Scanning reports a certain error that assumes the plugin doesn't stick to the definitions. That's with Cakewalk on Windows 10, so clearly it has not to do with Windows versions. AVX was introduced with SandyBridge as the successor of SSE 4, and one could argue that you Spogg, have a processor of the first AVX generation, which might have a glitchy first AVX implementation. But I have a 3rd gen i5, which is Ivy Bridge, and therefore the argument doesnt count.

I fear that the developer actually uses AVX2 or even AVX-512, which would only include pretty recent CPUs.

EDIT: I recommend to read the source code, it's fantastic! He uses descriptive names which makes it easy to follow even if you (like me) have no further knowledge of C++, but did program before. I found that he explicitly checks for all 3 AVX revisions and does a fallback down to AVX (1st gen), if CPU reports so. So that can't be the issue either.
The code is full of programming highlights. For example, I use a simplified algorithm for cubic interpolation in my Spline class, but he even simplified it another step more. I might make good use of it for the spline editing of Trog's envelope generator!
https://github.com/ryukau/VSTPlugins/tree/master/CubicPadSynth

--------------------------------------

Here's a little more search results of Rock Hardburn's Padpal:
A video playing the presets
A discussion on KVR with the dev involved
A forum that seems to host the synth, but you need to register first

--------------------------------------

On the page I listed about the algorithm description, there is a link to ready-made C algo code (ready in terms of "can be used in your synth right out of the box"). It would just need someone with knowledge of C to convert it to DSP/ASM.

[Spogg]

@deraudrl
Yessss! I got it and it works, so I was able to dive in deep and listen to what I wanted. Thank you so much!

@tulamide
Thank you so much for testing in Windows 10 and I really appreciate the links and comments.

@trog
Many thanks for your explanation. The maths and principles of FFT and IFFT are way beyond me I must admit. But my ears and faith in your knowledge convince me you are right (as expected). I could not provoke any wrap-around sounds in the PadPal synth, even though someone on that forum did mention it.

--------------------

I still find it remarkable that so little CPU is needed for PadPal. And the speed at which the table is updated is amazing. Just a small click when you release the bandwidth knob and it’s ready.

Now I’m thinking about making something using brute force (lots of oscillators needed) which could be modulated in real time. Who knows where that will lead but I think tulamide’s 2 weeks expectations are a little optimistic!

Thanks everyone

Spogg

[martinvicanek]

Here is a proof of concept that I put together. It is really barebone, but I find the sound very promising. What do you think?

[trogluddite]

I found that he explicitly checks for all 3 AVX revisions and does a fallback down to AVX (1st gen), if CPU reports so. So that can't be the issue either.

This may still not solve the problem. The only CPU capability conditionals seem to be compiler directives - they are optimising the code for the system upon which the C++ source code is compiled, not the system upon which the final object code (i.e. the plugin .dll) will be executed. I may have missed something when skimming the code, but I didn''t notice anything which would preclude compatibility problems if the final plugin .dll is loaded on a different system to the one which compiled it.

I still find it remarkable that so little CPU is needed for PadPal. And the speed at which the table is updated is amazing

Low CPU load for note-playback doesn't surprise me, as it's literally just a sample-player once the wave rendering has been done. The iFFT rendering speed is very impressive though, and I think it would be a challenge to match that kind of speed with the native features of FS (though maybe the extended assembler of the Alphas would help).

It would just need someone with knowledge of C to convert it to DSP/ASM.

1st rule of porting: never assume that a port will be simple unless you are well versed in both source and target languages!

Actually, we should already have a solution for the hardest part - the iFFT - unless MyCo's ASM version has any size limitations (possible, but I don't remember there being any). However, even the data creation code is nothing like as trivial as those friendly C++ high-level abstractions make it look, and DSP code is out of the question, as it can't do the necessary variable-length loops. It's certainly all possible in ASM, but it's not just a quick brain-teaser for an idle hour or two - per-sample number-crunching is relatively easy, but structured code is a different matter entirely!

[martinvicanek]

Actually, we should already have a solution for the hardest part - the iFFT - unless MyCo's ASM version has any size limitations (possible, but I don't remember there being any).

If I remember correctly MyCo's code makes use of 64k long sine and cosine tables so that could be a limitation. However, these tables may be generated for any size on the fly using an iterative formula. So yes, I believe we do have all the ingredients.

[tektoog]

Here is a proof of concept that I put together. It is really barebone, but I find the sound very promising. What do you think?

Sounds phat... and promising !
Very thick rich sound screaming for filtering and modelling

[tulamide]

Here is a proof of concept that I put together. It is really barebone, but I find the sound very promising. What do you think?

Very promising is an understatement!
This comes right out of the box? I would use it in my production as it is, with no fx, filtering or anything else. What an awesome sound. And this is just the beginnings? Wow!

I had a hard time finding out that I need to touch the detune knob first to generate a table to be able to hear audio. After that I was enjoying to hear how the bandwidth for the partials work. From a single frequency to a full ensemble with just a detune knob, that's fantastic! (And so different from unison)

Please guys, keep at it. This is pure gold from a musician's perspective!

[martinvicanek]

Exploring possibilities...

[adamszabo]

Very very nice! Now it only needs is some BLEP antialiasing