Cytomic 'The Glue' Compressor

[andy-cytomic]

sorry about being the forum userbase equivalent of a clingy girlfriend.

also that other stuff you posted about how the programming is going is fascinating, i've only ever read about this sort of stuff being used in rendering engines and simulations and stuff.

although i mean, this technically IS a simulation. i guess it just doesn't operate in the same domains as most other simulations I can think of.

LOL! No worries When I'm quiet it's usually because I'm working hard.

The memory throughput of audio is the biggest challenge here for GPUs since gamers don't really care if a scene takes a couple of seconds to load all the textures, but for audio a couple of seconds of silence is a bit crap All the cards with higher memory throughput (ie speed to and from the GPU) are typically more expensive and more niche scientific computing type affairs, so I won't really be able to support those types of devices.

But the good news is that this PDK method means I can use AVX256 for the core number crunching, which is a feature all current CPUs have. Right now I can only use AVX256 for the computation of non-linear functions like exp(x), not for the matrix solver itself, but the PDK method can use it for both, and keeps a cap on the peak iteration count at the same time. Very excited times for me right now, it's been around 6 months of R&D and it's almost production ready

[sleepcircle]

Right now I can only use AVX256 for the computation of non-linear functions like exp(x), not for the matrix solver itself, but the PDK method can use it for both, and keeps a cap on the peak iteration count at the same time.

Yes! This is sort of what I meant, I knew there were certain things like exponents and bleedin' square roots that you could just hand over tidily to an existing function, which is what I assume most plugins mean when they say they are "SSE Optimized for Intel Whatever", and in some cases synthesizers multi-thread things by handing off different voices to different processes, but I couldn't think of any particular audio plugin which is doing what you're talking about! This is quite exciting and I'd like to congratulate you on figuring it all out. Honestly I'm trying to think intelligently about this but I cannot imagine how you pulled it off.

(I am not a very good programmer, it must be said.)

[andy-cytomic]

Yes! This is sort of what I meant, I knew there were certain things like exponents and bleedin' square roots that you could just hand over tidily to an existing function, which is what I assume most plugins mean when they say they are "SSE Optimized for Intel Whatever", and in some cases synthesizers multi-thread things by handing off different voices to different processes, but I couldn't think of any particular audio plugin which is doing what you're talking about! This is quite exciting and I'd like to congratulate you on figuring it all out. Honestly I'm trying to think intelligently about this but I cannot imagine how you pulled it off.

(I am not a very good programmer, it must be said.)

Normally "SSE optimised" refers to the fact you can do stereo processing for the same cost as mono using SSE2 with pretty much zero change to any code other than declaring the type you're using to be "__m128" instead of "double". Optimisations can also be made for FIR type operations where there are no implicit equations or feedback loops complicating things, which rules out most analog modelling stuff since not only are there loads of feedback loops but the feedback loops have to be solved without inserting additional delays. Optimisations beyond this are typically only applicable to polyphonic synth / sampler type situations where each voice of the synth can be computed in parallel, since that is an entirely parallel operation (unless you're doing a piano simulation with feedback between strings etc). In The Drop and The Scream I calculate parallel non-linearities, which definitely helps, but the PDK method is another level above this in terms of parallelisation (yes, that is actually a word! https://en.wiktionary.org/wiki/parallelization )

[andy-cytomic]

I'm still working away flat out, and everything is still looking great for the new PDK method. The updated expression class is doing brilliant optimisations, it's pulled out every common sub-expression that it can and pre-computes as many terms as possible to make the innermost loop lean. Looking at the generated code there is nothing I could do manually to make it any faster, it is a thing of beauty if if you're into raw number crunching

[sleepcircle]

That is REALLY cool to read, about the expression class—well done.

[Vertion]

Woot!

viewtopic.php?f=1&t=512525

[ElevatedBeing]

I'm really curious to hear how HD mode will alter The Glue's overall behavior. Really looking forward to running audio through this HD mode.
Andy, if The Glue HD is part of The Glue 2... then what exactly is The Glue 2 ?

The Glue 2 will be the current version of The Glue with a few things added, but enough so I think it's worth a full version number update. You will be able to load current presets and get the same sounds, but internally I'll switch over to using a fixed cpu DK-method solver, which should help reduce cpu usage for your current projects - this is called MD mode. I'll also add an HD mode, which will have increased cpu load with lots more detail and non-linearities in the model, and a dual mono to stereo variable link control.

Will there be a small upgrade fee for version 1 users? Or do we have to pay full price?

[andy-cytomic]

Will there be a small upgrade fee for version 1 users? Or do we have to pay full price?

The Glue v2 will be a free upgrade for all existing customers of The Glue v1

An existing customer is a anyone that has an order and matching license for that order in their Cytomic accounts.

[meloco_go]

Will there be a small upgrade fee for version 1 users? Or do we have to pay full price?

The Glue v2 will be a free upgrade for all existing customers of The Glue v1

An existing customer is a anyone that has an order and matching license for that order in their Cytomic accounts.

Can I buy it one more time, but you release it sooner?

[sleepcircle]

The Glue v2 will be a free upgrade for all existing customers of The Glue v1

that is extraordinarily generous of you. i almost feel bad getting it for free. maybe you could make a 'donation' option for upgrades or something.

[Delta Sign]

Yeah, that's indeed very generous! Thanks a lot

[andy-cytomic]

The Glue v2 will be a free upgrade for all existing customers of The Glue v1

that is extraordinarily generous of you. i almost feel bad getting it for free. maybe you could make a 'donation' option for upgrades or something.

That's an interesting idea, but there is something that would help more. We want to support our customers with ongoing updates, to say thanks for supporting us with your purchase.

If you like the update once you've used it, and think that it makes The Glue even better value, then please let your friends know about it. Increasing sales and increasing the user base means we can develop more products more quickly and at a higher level, and customers talking about our Cytomic products and spreading the word, and getting genuinely excited by them is the best way to do this! Talking to friends directly, posts to forums and social media, uploading production videos, all that sort of stuff is worth the most to us

[DJMaytag]

As far as V1 goes, I’m trying to sort out a phasing issue I’m having in Reason 10.2, on a bass drum and a parallel bass drum, upon rendering.

Is there any potential for a bug that the real-time va render delay compensation isn’t working properly in Reason, be it on your end or Reason’s (ie Reason doesn’t know how to handle DC differently in render mode)?

EDIT: a mixdown definitely sounds correct now that I’ve set all my instances of The Glue to the same real-time vs render settings (2x).

[andy-cytomic]

As far as V1 goes, I’m trying to sort out a phasing issue I’m having in Reason 10.2, on a bass drum and a parallel bass drum, upon rendering.

Is there any potential for a bug that the real-time va render delay compensation isn’t working properly in Reason, be it on your end or Reason’s (ie Reason doesn’t know how to handle DC differently in render mode)?

EDIT: a mixdown definitely sounds correct now that I’ve set all my instances of The Glue to the same real-time vs render settings (2x).

I've never tested Reason's plugin delay compensation because I don't have a copy of Reason. When oversampling The Glue correctly reports to all hosts with sample accuracy how much latency is introduced by the oversampling process. You should be able to have each instance with its own oversampling settings just fine. Please contact Propellerhead software and get them to have a look at things, I'm happy to provide them with NFRs of Cytomic software to help them test with, and if they can't reproduce the issue easily, and they provide me a copy of Reason to test with I can have a look at it.

[DJMaytag]

I got a reply of “it’s up to the VST to report the delay” and basically “sorry, not our problem.”

I did follow up and ask how Reason handles real-time (which works fine) versus render (where the problem appears). Clearly there is a problem when rendering, so either Reason doesn’t “tell” VST’s that rendering is happening, or The Glue isn’t somehow reporting the increased latency in render mode.

Whatever the case may be, the recommendation might be to use the same real-time and render oversampling setting to avoid this problem in Reason (I’m on 10.2, but I assume 9.5 and above have the same issue).