Is there any ongoing work to provide a LLVM backend for OCaml? By googling around I have found some previous work: http://ocaml.org/meetings/ocaml/2012/abstracts/oud2012-paper7.pdf (github: ) https://github.com/whitequark/ocaml-llvm-ng From what I can gather, the primary challenges are calling …

GHC now uses LLVM, and there were a bunch of additions to the LLVM IR to make that possible. The LLVM people seem to be very willing to make IR alterations to make it easier for modern languages to adopt LLVM as a back end. It certainly has a bunch of advantages for any language front end using the…

It is worth noting that LLVM is not really a good back-end for languages with garbage collection or really functional languages in general. It doesn’t provide support for doing a proper tracing GC, it’s slowly getting better but the last I checked it still wasn’t good enough. GHC gets around this b…

[image] lpw25: It is worth noting that LLVM is not really a good back-end for languages with garbage collection or really functional languages in general. It doesn’t provide support for doing a proper tracing GC, it’s slowly getting better but the last I checked it still wasn’t good enough. I…

Hmm. I’ve had a look at the most recent work on GC support, and it might now be expressive enough. As I said, it has been slowly getting better. It still looks like it’s going to unnecessarily get in the way of a lot of LLVM’s optimisations. Given previous attempts to use LLVM for GC I would enter t…

Well, one of the nice things about LLVM is it is purely a back end. That is to say, it’s a pretty low level representation. And one of the nice things about the OCaml compiler is that it’s retargetable. So one can experiment with this without terribly much commitment, and even maintain both native a…

Why is it worse than in code you generate for yourself? Just the lack of direct access to the lower level post-register allocated representation for open coding barriers? Yeah that’s the problem, you need to pass the register information to the runtime. Although to be fair it was a long time sinc…

[image] lpw25: Yeah that’s the problem, you need to pass the register information to the runtime. Although to be fair it was a long time since I’d last looked at this, and I had somewhat confused two separate issues in my mind. For a long time it was the case that the only way to use a GC with L…

[image] perry: BTW, the LLVM link time optimization stuff might provide some interesting opportunities for things like handling parametric polymorphism with polyinstantiation followed by optimization of the type specific instantiations to take advantage of type specific optimization opportunitie…

[image] fyquah95: Unlike C/C++ compilation, when compiling a .ml file, the compiler can be given visibility over the IR (in the form of .cmx files) of modules that it depend on – hence various cross-module optimisations can be done at this stage of the compilation. I didn’t know that. This im…

LLVM Backend for OCaml

Ecosystem

lpw25 November 17, 2017, 10:09pm 5

The problem is that you cannot have GC roots in registers across points where the GC might run. This is not really acceptable for a proper GC. The additional spilling will severely harm performance.

Topic		Replies	Views
OCaml LLVM backend? Ecosystem	19	6095	June 2, 2025
Ocamlopt vs Duplo/LLVM code generation Ecosystem compiler , compilation , code-generation	11	1801	September 16, 2020
OCaml compiler development newsletter, issue 4: October 2021 Community compiler-newsletter	9	2840	November 25, 2021
Advantages of OCaml over Rust Community	76	33379	September 4, 2019
A proposal for a resource-management model for OCaml Community	29	7610	January 1, 2021

LLVM Backend for OCaml

Related topics