A tail-call interpreter in (nightly) Rust
20 hours ago
- #Tail-Call Optimization
- #Performance
- #Rust
- Using Rust's nightly become keyword to implement a tail-call interpreter for Uxn CPU emulation.
- Tail-call interpreter outperforms previous Rust and ARM64 assembly implementations on ARM64.
- Performance on x86-64 is worse than assembly backend, especially in Fibonacci microbenchmark.
- WebAssembly performance with tail calls is significantly worse than VM interpreter.
- Implementation involves storing VM state in function arguments and ending each function with a call to the next.
- Macro used to reduce boilerplate in defining tail-call functions.
- Codegen analysis shows good optimization on ARM64 but suboptimal on x86-64 with register spills.
- Tail-call interpreter is merged and released, with defaults based on architecture.