Re: What’s more useful, Guile to C or Guile to MLIR?

On Sat, 01 Nov 2025, Srayan Jana <[email protected]> wrote:
> Hello again!
> I'm seriously considering spending some time for the rest of the year working on some sort of Ahead of Time compiler for Guile that reuses parts of Hoot. However, I have a question that is going to decide what I'm going to spend most of my time on.
> Should I try to aim for Guile to C or Guile to MLIR?
> Personally, I'm much more interested in MLIR (especially since MLIR
> has an option to emit c anyways)

My 2 cents.  To IR, simply because we can apply more optimizations at
that level if needed or we can transform the code to other
representation like GPU kernel instead of C.

As for MLIR, I never heard of it, but I would avoid targeting a complex
IR base on LLVM.

Have you considered something like QBE [0] ?

Thanks,
Olivier

[0] https://c9x.me/compile/doc/llvm.html

-- 
Olivier Dion
oldiob.ca