no_std support

[mert-kurttutan]

I know it might be too early.
What is current status for no_std support?
It will potentially be more important for cpu-based workloads.

[ZuseZ4]

I think it should just work, Enzyme works on another level and won't include any std functionality if you don't use it.
At least a while ago I also knew another user who verified that, so please let me know if you encounter issues with it.

[jedbrown]

I think it'll be an issue if allocation is needed. We should eventually expose a pre-allocated tape as can be done using C-Enzyme.

[ZuseZ4]

https://doc.rust-lang.org/alloc/
Afaik alloc is available through core even when no_std is used?

[wsmoses]

Unless you registered a custom allocator, Enzyme will use malloc and free for things

[ZuseZ4]

Yes, but luckily Rust stopped shipping it's own allocator a long time ago, so they should match behaviour.
I just wanted to clarify that I think people won't be surprised if allocations happen in no-std code, let it be Enzyme or normal code.
I'd be open to accept a PR to have custom-alloc / tape support, but I have more important bugs to fix first (like autodiff sometimes not running), so I don't think I'll have time for this in the near future myself.

[mert-kurttutan]

I can have a look at this, no promise though.
Any link to a developer guide would be great

[ZuseZ4]

Unfortunately there is no developer guide for Enzyme core, and I don't have the knowledge to write one.
The Rust PRs are quite well documented (at least the part which got reviewed during upstreaming) and I'd write more docs once I understand how to register derivatives, but I think just experimenting won't be super efficient for you, unfortunately.
I hope to figure out at some point how to solve this issue: #190
From there on it should hopefully be easier to also support a custom allocator, assuming that is what you are interested in.