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.
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.