Beating the compiler
Summary
Matt Keeter explores beating the compiler by implementing a hand-written AArch64 interpreter for the Uxn CPU, using an indirect-threaded dispatch and a register-heavy design. The post includes Rust and assembly sketches, analyzes how moving core data into registers and end-of-opcode threading impacts speed, and reports roughly 30% faster performance over the baseline Rust interpreter. It also covers device I/O, C shims, and experiments that ablate dispatch strategies to show that centralized dispatch hurts performance due to branch prediction.