Replacing a Rust Enum with a 64-Bit Word Made My Interpreter 17% Faster

Posted by metrofun 3 days ago

Counter60Comment25OpenOriginal

Comments

Comment by fpoling 1 hour ago

The article title is misleading. It is not that Rust compiler was not able to optimize some low-level operations. Rather the author came up with encoding schema that fit most things the interpreter dealt with into 64 bit. This replaced the previous schema that used 128 bit for everything but that can be directly mapped into Rust enums. The catch was that it was necessary to allocate some things on the heap and use pointer indirection but that was used for rare values so on average the new schema provided nice win.

One cannot expect a compiler to come up with such encoding.

Comment by pbiggar 21 minutes ago

When I think of how a "compiler" could make these optimizations, I think the right place is an optimizing LLM (so, just a regular coding agent that you prompted to find optimizations like this one), making the changes in source at the request of the developer. That provides the dev with adequate input on whether they would like to opt-in to an unsafe optimization like this one. The compiler can continue to do deterministically-safe optimizations.

Comment by trickypr 11 minutes ago

That seems like a horrible idea:

1. Do you really want the rust compiler to run at the speed of an llm?

2. Compiler optimisations are already extremely unpredictable with deterministic compilers[1], I hate to think how unpredictable your compiler would be.

3. What if someone else wants to build the software, do they have to decide on optimisations now? What if the optimisation depends on your features not available on old generations of CPU? (There is a reason we don’t compile with -march=native)

4. Compilers already have “unsafe” optimisations, but people rarely enable them (-ffast-math)

[1]: https://faultlore.com/blah/oops-that-was-important/

Comment by pbiggar 5 minutes ago

You misunderstand me. I'm saying that the developers can make these optimizations with LLMs, at the source level, and thus they don't need to be added to compilers.

Like just open Claude Code and ask it to find optimizations. That's the right place for this kind of optimization.

Comment by win311fwg 1 hour ago

What is misleading about the title? A custom encoding scheme is exactly what it suggests. Maybe it has been edited since your comment was posted?

Comment by dymk 36 minutes ago

It wasn’t replacing one rust enum, it was replacing what are effectively multiple enums

Comment by dzaima 22 minutes ago

How so? It's replacing multiple enum variants, but just one enum, "enum Value".

(also; if anything, the title is implying the exact opposite of "Rust compiler was able to optimize ...", "Replacing a Rust [...] with [...]" is clearly moving away from Rust-magic to something else)

Comment by lowbloodsugar 4 hours ago

Take a look at triomphe's ArcUnion and extrapolate from there. Basically make a crate for just your 64bit union type, do it unsafe there, test with miri, and now you have a safe 64bit type you can use with match. You're happy digging around assembly so this is well within your wheelhouse. The only challenge will be if you do use miri to verify then you need to use the 'provenance-preserving' pointer adjusting functions. Worth the learning experience in my opinion. I did one for my system and it was super fun and had the performance impact you describe.

Comment by krick 2 hours ago

That's very unpleasant to hear. It's sad to be reminded that Rust compiler is not magic and cannot just... do these things somehow. Sure, all abstractions do have some cost, but, man, 17% performance gain by virtue of replacing enum with this monstrosity? That's very annoying.

Comment by maplant 1 hour ago

It can't do these things because it's not wanted. Say you have the following:

  enum Value {
      Float(f64),
      Ptr(*const T),
  }
Do you want the compiler to disallow certain bit patterns in the Float variant simply so that it can implement NanBoxing?

Comment by vlovich123 1 hour ago

Probably with an annotation around a NanBoxable(f64) type that tells it to do that.

That being said, the optimization is complex that may be insufficient:

> For my boxing scheme, I picked a bias value such that the lowest two bits end up being 10. That 1 in bit index 1 indicates that doubles can't be directly compared for equality. Amazingly, we only lose two bits of exponent, and we keep the full precision of the mantissa, meaning we lose no significant digits in the flonum representation.

This suggests the optimization needs more information about specifically how you want to box the float. There probably is some primitives worth considering standardizing to make this kind of optimization possible so that the tunable parameters are passed as const generic values.

Comment by speedstyle 30 minutes ago

A 64-bit sum type can't magically combine an i64, f64, and several raw pointers, each of which carry a full 64 bits themselves. You have to change the semantics of the code. Some semantics could be expressed more easily with compiler improvements, allowing eg `Aligned<T>` like `NonNull<T>`, or `FiniteF64` like `NonZeroU64`, or even `#[range(0..1<<60)] u64`, but you still couldn't overlap two `Aligned`s in one enum, because only one can be stored unchanged, the others need masking off before usage. Even if the enum semantics allowed this, I'm not sure the compiler should do this kind of compute/memory tradeoff automagically. Which doesn't mean you can't write nice abstractions over it, there's a few tagged ptr crates which aim to do it for you

Comment by gigatexal 4 hours ago

But isn’t the enum far more readable and maintainable than having to do bit operations on things?

Comment by compiler-guy 3 hours ago

The reason the enum is so nice is that it works as a terrific language-supplied abstraction that covers up those bit manipulations. It's very nice to get those abstractions for free like you do in Rust, but it can't be optimal for every specialized use case.

This new code also supplies similar abstractions. That actual specific code is much harder to reason about, but most users--and even the next person who works on the interpreter--simply won't care, or even know what is going on underneath the hood. The abstractions provided by the author do that work and apparently do it cleanly.

For most use-cases, that extra hand-written code isn't worth it. But in specific cases it can be, and the author has actually measured the value and determined that it is.

Comment by steveklabnik 3 hours ago

> As you can see, it has many convenience methods to make it easy to work with, compensating for the loss of the Rust enum.

Also, Rust does try to do some of these optimizations itself. These aren't exposed in the stable language to let you do some more advanced things, but it wouldn't be impossible for you to get the best of both worlds by letting you communicate this stuff more directly to the compiler. Right now those things are more like "this value is where you should put the tag" than the more advanced stuff here, though. Would be cool to see someday!

Comment by lowbloodsugar 3 hours ago

quibble: unsafe is stable. you can't do this in safe rust, but you can do it in unsafe rust. just isolate all the unsafe code in a single type, ideally a tiny crate.

Comment by tialaramex 2 hours ago

Steve wasn't talking about unsafe. He was talking about being able to mint your own non-enum types with user defined niches.

Rust provides for example NonZeroU8 which is an 8-bit unsigned integer that's never zero, leaving it with 255 possible values and a convenient niche. You cannot make one of these yourself directly, because the mechanism used by Rust itself is a deliberately perma-unstable compiler-only proc macro which says "Hey compiler, I promise I only ever use bit patterns 0x01 through 0xFF inclusive".

Today you can either - hide a NonZero type inside your type and use that to get the niche, or, use an enum itself which automatically knows ever pattern it didn't use is a niche. In the future a hypothetical "Pattern Types" feature would let you make such types yourself as easily as Rust does

Personally I would like to make a Balanced set of types, like BalanacedI8 (the 8-bit integers except the most negative, so -127 to +127 inclusive) because I think lots of people have a use for types like i8 or i32 but don't need their unbalanaced most-negative value and could re-purpose it this way. And you can make such types... indeed I have... but it's only really practical in unstable Rust.

Comment by fpoling 1 hour ago

Arbitrary subranges of int types were available in Ada for over 40 years via static enforcement via Spark and compilers were able to optimize them nicely.

For a system language I wish Rust would support such things rather than coming with NonZero hacks.

Comment by steveklabnik 1 hour ago

This is called "pattern types" in Rust land, as your parent mentioned, and is exactly the kind of work being talked about.

NonZero isn't a hack: it's an example of a common pattern. If pattern types were available today, you'd still want NonZero, as an example of a pretty standard pattern.

The idea is, as always: prove out the specific version, then generalize.

Comment by tom_ 4 hours ago

The computer's the one running the code, and it'll be running it a lot (or so its author hopes), so it's probably worth bearing its limitations in mind in the interests of making its life easier (so to speak) rather than prioritising the people who will modify the interpreter - a far less common occurrence.

The bit operations involved are pretty simple and won't take you long to figure out even if you've never done them before.

Comment by diath 3 hours ago

Highly optimized code in hot paths is rarely readable.

Comment by dzaima 2 hours ago

Unless the highly-optimized parts are wrapped by an interface that looks similar to the non-optimized version.

In the case of a tagged object in Rust, depending on how well the compiler can wrangle through it, you might even be able to add a `.unpack()` method that returns a pretty enum from a packed value, that you can pattern-match on or whatever, and let the compiler remove all the code of unpacking unused cases.

(using that directly for the addition example would end up less efficient of course, but still most likely beneficial. It's after this when there's a potential true readability vs performance tradeoff)

Comment by locknitpicker 3 hours ago

> Highly optimized code in hot paths is rarely readable.

...unless it's supported by the language as a first class feature. See for example C++ and RVO.

Comment by diath 3 hours ago

Not necessarily, in C++ you'd still drop from smart pointers to raw pointers, from virtual dispatch to switches/computed gotos, from std::function to function pointers and so on. These abstractions all come at a cost.

Comment by mwkaufma 3 hours ago

If you add "fits in a register" to your list of correctness requirements, then it's no-go even if the source has less cognitive overhead.

Comment by nwhitehead 20 minutes ago

[dead]