ttfx’s assembly version reportedly reaches an 8x mean speedup on an Intel 135U
A user says moving ttfx from Rust to assembly brought much larger gains after hours of AI-assisted Rust optimization had begun yielding only modest improvements.
TLDR
A user reports an 8x mean speedup for ttfx’s assembly version on an Intel 135U, describing much larger gains than continued Rust optimization. Earlier on September 26, they credited Opus with further improvements and claimed speeds up to 450x faster than the original Python implementation. They also say it supports SSE2, AVX2 and AVX-512, falling back to the slower Rust implementation otherwise.
