Large Language Models for Compiler Optimization

Source: Hacker News

Article note: It sounds like they're using the LLM trained on assembly not to optimize the assembly, but to select and order passes in LLVM to optimize the assembly. ...And it still regularly produces code with different semantics than the input. And optimization was measured by number of instructions, not runtime on a target. Nifty, but one of those hype>>reality things, like everything around LLMs.
This entry was posted in News. Bookmark the permalink.

Leave a Reply

Your email address will not be published. Required fields are marked *