Finding bugs in LLVM IR optimisers: programmatically generate small (loop free, no control flow) functions, run your favourite middle end passes, and verify the output is equivalent!
https://blog.regehr.org/archives/1510
(Uses Alive to verify and 4000 CPUs of compute.)
Related Posts
Bril is a cute intermediate language for teaching (think simplified LLVM IR): https://www.cs.cornell.edu/~asampson/blog/bril.html
The idea of providing a standard JSON format to help students write basic passes is really elegant.
Go has an elegant approach to defining example functions, which are shown in docs as `main()` with the output: https://go.dev/blog/examples
Whilst LLMs don't always give an accurate answer, the UI is really compelling. I keep finding users whose favourite way of doing research is an LLM.