What's the general status of Reaktor?
I notice that none of the "OS version X compatibility" and "M1 silicon compatibility" pages even mention Reaktor, even though they mention a bunch of other Komplete synths.
No mention of Reaktor: https://support.native-instruments.com/hc/en-us/articles/360014683497-Apple-Silicon-M1-Compatibility-News
Why is this? Is Reaktor splitting off from Native Instruments? Is it a legacy that will never be supported in later OS-es and hardware? Is it somehow its own special snowflake that has a totally different set of support articles I haven't been able to find?
Where would I go to find out whether-and-when on Reaktor support for M1 silicon?
Comments
-
Reaktor does not support M1 natively yet. Pretty sure they will be singing it from the rooftops when it does. Currently I expect that's all the Reaktor devs are working on it as top priority, but it's one of those "it's done when it's done kind of things". No chance that they will be giving a projected date before release.
Problem is that Reaktor isn't just an 'app' it's an optimising compiler, it generates optimised machine code. So with 'apple silicon' it has to be rewritten to generate completely different optimised machine code for a totally different processor architecture. That is no trivial task.
Most applications will just need to be recompiled with a different target cpu, then debugged, tweaked and tested. If you did that for Reaktor, it would run on M1, but the core compiler would still generate code for intel x86 architecture, so none of the ensembles would work.
Maybe something good will come out of it though. Apple silicon is ARM based. So is iOS and Android... hmm, this would be a great opportunity to update the code base so a future version for phones and tablets might be possible... here's hoping :)
It's also a great opportunity for developers to become more familiar with the subtleties of the code base - particularly those that have joined the team since it was originally developed. With a bit of luck future updates might be more significant, or even more frequent on all platforms as a result?
5 -
The good news are tools for translating code are in place and sooner or later things should come in place too
That's my point though - tools for translating code would allow them to easily make their existing code for Reaktor run on Apple Silicon, but it would still generates x86 binaries. Which would be useless.
They need some completely new code for the core compiler to generated ARM binaries. There is no point in translating their old code, so translation tools are useless
I'm sure they are well on the way to getting native M1 compatibility, but it is unrealistic to expect it to happen as quickly as for other apps where translation tools can be applied to the whole process.
0 -
Remember, every time you initialise a Reaktor ensemble that includes core (that's most of them), Reaktor runs a multi part compilation process turning all those boxes and wires into binary machine code.
A tool that can turn x86 binaries into ARM binaries is only useful here if it is free to package into your product without licencing restrictions, and can convert any large binary optimally in milliseconds every time without fail and without human tweaking or testing... do they have that?
It would have to be open source as well otherwise Apple would then control the future of your flagship product...
0 -
It would be good to hear some official word from someone at NI about the future of Reaktor. Theyβve made public posts with roadmaps for Kontakt and Maschine, but in the Reaktor front itβs crickets.
2 -
I understand that Reaktor generates code -- it's the right way to make fine-grained configurability work at high performance!
I also understand why Rosetta isn't a good answer for Reaktor -- it needs a much lower-level implementation.
Technically, one could have hoped that they might be using an abstraction layer like LLVM for their code generation back-end, which could simplify the transition, but given that Reaktor has a very long history ("it is very experienced") then they probably started it before LLVM was a real thing, and thus have their own back-end, and ARM SIMD works just differently enough from x64 SIMD to make an engineer sad.
That's actually not the main question I'm asking. The question I'm asking is why the documentation on M1 transition and OS version transitions don't even mention Reaktor. There's a bunch of tools that are listed as "not on M1 yet" on those pages, and a few (mainly Kontakt,) that are. But Reaktor is in neither list. Which seems like a ... very large omission?
I had good luck with @Jeremy_NI promising a documentation update in another thread, so here's hoping lightning can strike twice if I call his name :-D
1 -
@Jon Watte Reaktor 6 isn't listed on the KB article because it's not officially supported. If you check the document, we're only listing what is compatible either via Rosetta 2 or natively. It's not legacy or splitting from the brand but I imagine we can't work on everything at once, especially when considering there is other compatibility topics other than Silicon.
Unfortunately, we don't really have a timeline available for Reaktor 6 at the moment. We can see products like Maschine, Komplete Kontrol, Massive X or even Guitar Rig 6 Pro are in the making which means Reaktor is not a "coming very soon" sort of thing.
1 -
Technically, one could have hoped that they might be using an abstraction layer like LLVM for their code generation back-end, which could simplify the transition
An extra set of abstraction layer's for a VM would slow down the compilation process. Code gets compiled at every initialisation, so when you load a preset, it recompiles, when you make an edit, when you reset the audio engine..
I would imagine slowing that down significantly is not an option?
I expect that there are parts of the process that will translate easily - building dependency graphs etc., more abstract parts of the compilation process. The main problem would be the code generation itself with optimisations, on a completely different architecture.
Also expect that the Dev team is small - just big enough to handle maintenance, and to have small to medium updates every few years. Expecting the same team to quickly rewrite the most complex and challenging parts of the code base seem's unrealistic. Particularly when targeting a technology they have no experience of.
Just seems to me that we should all just be patient. And those who can't handle this sort of difficult transition should maybe look towards Apple for answers and for slinging mud.
I'm sure Apple could have approached the transition in a way that made things easier for developers like NI. But they didn't so go hassle them ;)
1 -
An extra set of abstraction layer's for a VM would slow down the compilation process.
At the risk of getting a bit too technical for the audience, there's nothing extra there. You need a compiler back-end that can JIT compile. LLVM is one such back-end. It doesn't add any "extra" layering compared to anything you write yourself (assuming you configure/use it appropriately.) There may be other reasons not to use that particular one (including the particular trade-offs chosen in that library,) but "an extra layer" doesn't seem like a reason to me, because no matter what, you need a library to poke the data dependency and control graph at, and get code out of.
Anyway: Right now, a reasonably human reading available communication from Native Instruments will believe that Reaktor isn't even on the roadmap, and is abandonware.
2 -
Anyway: Right now, a reasonably human reading available communication from Native Instruments will believe that Reaktor isn't even on the roadmap, and is abandonware.
I hope not! it has been very quiet for many months though
I expect it's just that the devs are very busy with native M1 support, and now with more of an NI presence on the forums from Matt, Jeremy et al., devs don't need to check here at all... ? π€
0 -
but "an extra layer" doesn't seem like a reason to me, because no matter what, you need a library to poke the data dependency and control graph at, and get code out of.
A virtual machine involves an extra layer, because you have a new 'language' of virtual bytecode instructions.
So with a virtual machine, you
- do all the dependency graphs and whatnot
- then you compile that into bytecode instructions... that's your target machine code (virtual of course)
- then you use a (hopefully near optimal pre-existing, developed by someone else) stage to convert the bytecode into machine code for whatever platform you are using.
without a VM you
- do all the dependency graphs and whatnot
- then compile that directly into machine code instructions
There is no way around that. The whole point of VM is that its an extra layer of abstraction to improve portability. The cost is that there is an extra layer of processing, in many cases, its a price worth paying because in many contexts, a little extra time to deal with JIT lag, and slightly slower binaries is not an issue. And you might be targetting many platforms, so the idea of only having to maintain one target, the VM is extremely attractive.
Unfortunately, one area where that would be problematic is audio DSP for various reasons (particularly control of timing, and missed opportunities for exploiting platform specific optimisations. JIT and GC would be problems not benefits), and on top of that Reaktor also has the requirement for an extremely fast compilation process - so adding another translation layer would be... sub optimal.
Then there's the fact that Reaktor has never targeted more than two cpu architectures, and only one since 2006(ish) when Apple went intel, in that context, it wouldn't make any sense to use a VM
1 -
LLVM is not mainly a virtual machine, although LLVM IR data can be interpreted if you really want to. LLVM is mainly a retargetable compiler code generation back end/library, used by clang, Rustc, and a number of other compilers, FWIW. Yes, LLVM has an IR, but so does any other code generation back-end, no matter whether you wrote it yourself or not. LLVM is not a JIT, it's a code generator back-end.
Native code, built by Xcode, for ARM M1 Macs, use clang, which uses LLVM for codegen. Your DAW hosts and native plugins are most likely all compiled through LLVM, especially if they're M1 native. (There may also be a smattering of assembly involved, but native intrinsics generally do the job well enough.) It's pretty good infrastructure, as far as such things go. I think the fact that LLVM was not a production quality library when Reaktor was first built, has much more to do with it!
2 -
That's pretty cool. I haven't read up on anything like that for many years... just assumed LLVM was relating to some sort of VM. Pretty stupid name really, calling it LLVM, then insisting that it's not an acronym π
...Turns out it was originally Low Level Virtual Machine... it can be a JIT but doesn't have to be...
...doing some more searching, I can't find any examples of LLVM being used with dataflow paradigm languages, it seems to be more in line with control flow stuff, C++ alikes, etc.
Are there any examples? Can the LLVM optimisation system 'understand' Dataflow code the way it 'understands' control flow based paradigms?
0 -
I replied, just in case you two missed it π
1 -
It is pretty hard task to port Reaktor to AS. It may mean to write/rewrite substantial part of Reaktor. But it also depends what code Reaktor creates. It may be C/C++ that is compiled to target maschine code, or it may create x86 source code, that is translated to x86 maschine code by internal compiler. Or....
Because Reaktor used to support two different CPUs in past, I guess (hope) Reaktor generates high level language source code, that is further translated to target maschine code by general purpose compiler... If it is like that, NI needs to get compiler to AS and to optimize/recompile libraries for AS. And lots and lots of testing, optimizing and so on.
Even this simplest case is a lot of work. And it may work differently making porting to AS harder and harder....
0 -
So, the SIMD parts of Reaktor (basically, Primary modules) are not really a big problem to port. There are libraries like xsimd or SIMDe that make it pretty easy to have SIMD intrinsics ported to whichever architecture.
As colB said, Core compiler is the main problem, and it depends on one person only. Yes, it needs a pretty substantial rewrite I'm pretty sure. No, LLVM is probably not the solution.
1
Categories
- All Categories
- 18 Welcome
- 1.7K Hangout
- 67 NI News
- 894 Tech Talks
- 4.6K Native Access
- 17.8K Komplete
- 2.2K Komplete General
- 4.8K Komplete Kontrol
- 6.3K Kontakt
- 1.1K Reaktor
- 407 Battery 4
- 917 Guitar Rig & FX
- 465 Massive X & Synths
- 1.5K Other Software & Hardware
- 6.4K Maschine
- 8.2K Traktor
- 8.2K Traktor Software & Hardware
- Check out everything you can do
- Create an account
- See member benefits
- Answer questions
- Ask the community
- See product news
- Connect with creators