So I'm a humble freelance web developer and as techie-geek-nerd as it gets. Recently I toy from time to time with VSTs using C++ and JUCE. Very different beasts than my work's tools and languages. Yesterday I had time to waste so my goal was:

Develop a dead simple but fast convolution plugin with JUCE

1. Convolution plugin. It's easy given the examples included in JUCE-> SUCCESS-> in 15 min

2. Fast. As many of you know the key is a fast FFT library. The JUCE included one is nice but slow (KissFFT). For some unknown reason I happen to have downloaded and installed Intel's IPP library which is the fastest on PC.

2.1 Fast convolution plugin with Intel's IPP. JUCE has integration for this library. Needs a bit of tweaking but not much-> SUCCESS in 30 min

At this point I could just benchmark and enjoy the two of them but... that's where the procrastination kicked in...

- PROCASTRINATION STRIKES: All well and good but Intel is evil and IPP library is commercial. Wouldn't be nice to use a free one? -> totally pointless for my goal, let alone I barely use legit soft at all

In the next four to six hours...
- Installed the fastest free FFT library, FFTW. Still relatively easy -> SUCCESS in 40 min

- PROCASTRINATION STRIKES AGAIN: Realized there're much newer versions of FFTW but not compiled (ready to use) for Windows->even I could tell I was asking for trouble lol
- Tried to compile them with Visual Studio although FFTW is designed to be compiled in Linux -> FAIL: but complete success in getting in trouble , about 50 min
- Remembered I have a Linux server virtual machine installed for my work. No GUI at all, no C++ compilers installed-> FAIL: just too much trouble in 45 min
- Realized there're other FFT libs almost as fast as FFTW for plugin tasks (audio FFT, 1D real signal) -> I might just had set myself on fire as well
- Downloaded and tried to compile three of them: mufft, pFFT and fftReal-> SURPRSING SUCCESS with mufft and fftReal, about 1 hour.
But...
- PROCASTRINATION STRIKES YET AGAIN: Remembered that you can do CPU tasks with GPUs. No reason at all for my goal (bad for real-time, bad for plugins cross-platform) but heck, sounds cool -> plain and straight death wish
- Tried to compile NVidia's CUDA cuFFT library -> FAIL: half done but dropped, fucking hell, too much pain , 1 hour
- Tried to compile OpenCL (GPU standard to help CPU tasks) FFT library -> FAIL: borderline S&M stuff , 45 min
- Tried to use Microsoft windows-only alternatives: DirectCompute, C++ AMP, Directx12 Compute Shaders...-> if you're gonna die, die with your boots on

At this point I didn't even remember what my fucking goal was and I can't tell how much time I wasted, ended up asleep in bed with the jeans and belt on. I still have slight marks in my legs and waist..

Yeah, you can join me laughing at myself

Edited: those who know me will agree with my best answer selection lol
PS: I think @Haliax and @demberto will love this

 

Last edited: Nov 13, 2020

I didn't know you were that nerdy...

 

Sometimes even I don't know how much

It's true this is one of my greatest messes with geek procrastination. I've had other nerdy-free ones quite impressive as well.

 

Oh, I forgot. One of the things that got me worked up was reading those @Olymoon 's posts about fancy IR/convolution plugins.
So you could say it's all Olymoon's fault...
... ok, may be that's stretching it a bit too much

 

Can you supply a link for this please? I always thought that convolution of a 1-dimensional signal is super easy.
You just treat every sample point as dirac impulse.

 

You're right. Technically the convolution is very easy, quite like you describe it (it applies to more dimensions too). But extremely slow as soon the length of what would be the IR increases. In DAW terms it has no latency (as long as your computer can handle it).

The FFT-based convolution is the fastest method to compute a convolution. Not necessary if you don't need speed. It introduces latency. Much more complicated.

I don't know how much technical you want to get, but here's one link about it:
https://www.dsprelated.com/freebooks/sasp/FFT_versus_Direct_Convolution.html

As a side note: cool avatar!

 

Exactly what I was looking for. Thanks

 

hey @Xupito your controller post and this one inspired me to learn more about JUCE and audio programming , can you recommend a resource to start with ?

 

Sure thing. It's not easy but for learning JUCE:

1. Their own nice tutorials:
https://juce.com/learn/tutorials

2. This is the best YT channel for the task hands down, lots of video-tutorials:
https://www.youtube.com/c/TheAudioProgrammer/playlists


For learning audio plugin programming this is a gem. I'd start with this. A plugin that lets you program what it does with a language similar to C++ but a bit easier:
https://www.bluecataudio.com/Products/Product_PlugNScript/

 

Last edited: Nov 10, 2020

I feel your pain, so does stack overflow.

https://stackoverflow.com/questions/39675436/how-to-get-fftw-working-on-windows-for-dummies

Loved the read too

 

too good hahahaha. Take care tooo lol

Yeah, I followed a similar guide for the .lib creation and Juce also has almost-direct integration for FFTW
If only I had stopped there...

 

Last edited: Nov 11, 2020

UPDATE 2020.11.12 TWO DAYS LATER...

This is good
A while ago I thought I may switch to "work-mode" (aka how would I do this if my salary was at stake?) and try again. The thing went like this:

1. Search properly before throwing myself in the mud. SUCCESS: 15 min
2. Type this on the command line:
git clone https://github.com/microsoft/vcpkg
bootstrap-vcpkg.bat
vcpkg integrate install
vcpkg install fftw3[core:avx2:openmp:sse2:threads] --triplet x64-windows-static

SUCCESS: 10 min while having a coffee and joi... cigarette
2.1. Last FFTW3 version installed with source code patched for Visual Studio
2.2 Automatically compiled(ready to go) all versions I possibly need including multi-core, sse2, avx512
2.3. Static version too. The static are the ones that you can embed in the main program, so you can have only one xupito_is_dumb_convolution_plugin.vst3

3. Compile the plugin: plain and straight trivial shit

I know what you're thinking. Yeah, that too

 

Last edited: Nov 11, 2020

You are becoming agile

 

I consider myself more of a desire oriented... wait a min, that doesn't sound right...

 

I never tried making plugins, because I was stuck on the VST Lister @Xupito knows about. I even posted it here somewhere, guess its probably no use. C++ is first class SHITE, but I still like it for some reason. I spent like 3 months to understand this interface system of VST3 and finally got how plugin and DAW interact with each other

 

C++ has come a long way since I first used it about 30 years ago. Look on the bright side, at least you don't have to code VSTs in Java

 

I feel your pain bro ;) Especially since you went the low-level hard way.

I'm sure someone has tried
But half of what @demberto says is about the concrete implementation of the VST3 standard. Is based on those over bloated OO paradigms that made many C++ programs run even slower than Java under certain conditions at the early 2000s. I'm amazed that once compiled still run as fast as the VST2 ones. Oh, but like x3 times slower compiling time.

Anyway, right now C++ could be a complex yet cool language if they stick to a subset that doesn't require, for instance, to learn twelve ways of initializing.

 

You might wanna see my REAPER & JSFX for Audio Prototyping & Development doc (rationale & alternatives).
And https://www.cockos.com/wdl/ which may make DJBFFT easier to use.

 

Gonna check that out, thanks ;)
I already have in mind to use ReaJS or BCA Plug'n script for my midi plugin.
I've to check the latest "available" SynthEdit too, just in case.

Already checked long ago. So cool library but there's barely examples and no documentation.

 

Last edited: Nov 12, 2020