A compiler that solves Anthropic's VLIW optimization challenge

Anthropic's take-home interview challenge asks candidates to manually optimize a VLIW SIMD assembly program together with an LLM.

Instead, this is a general optimizing compiler for that virtual machine.

It performs only general-purpose compiler optimizations. There are no benchmark-specific or algorithm-specific optimizations, yet it substantially reduces cycle count.

The repository contains the compiler, optimization pipeline, and documentation explaining the design.