Re: [Caml-list] ocamlc compiles hello world, ocamlopt not

[Gabriel Scherer <gabriel.scherer@gmail.com>]

Benedikt, let me jump on the occasion to provide more pointers you
might be interested in:

- Raphaël Amiard, a student at Paris 6 university, recently worked on
translating OCaml byte into LLVM:
https://github.com/raph-amiard/CamllVM ; I have not seen a definite
comparison with Colin's and Benoit's work, but they appear to hit the
same kind of issues (performance degradation of exceptions, handling
GC roots that are passed in registers)

- VeriML is a research language for writing tactics and proof
assistants; developed at Yale during Antonis Stampoulis recently
finished thesis ( http://www.cs.yale.edu/homes/stampoulis/ ); they
have a HOL-like mode of interaction, where the user-interface to the
prover is directly a toplevel (forked from OCaml's toplevel), and they
use OCamlJIT2 to provide good performances within this model.

On Wed, Jan 16, 2013 at 11:32 PM, Benedikt Meurer
<benedikt.meurer@googlemail.com> wrote:
>
> On Jan 16, 2013, at 10:48 , Basile Starynkevitch <basile@starynkevitch.net> wrote:
>
>> On Wed, Jan 16, 2013 at 10:01:06AM +0100, Gerd Stolpmann wrote:
>>> Am 16.01.2013 09:56:55 schrieb(en) Francois Berenger:
>>>> On 01/16/2013 05:02 PM, r0ller@freemail.hu wrote:
>>>>> Hi All,
>>>>>
>>>>> By the way, is there any doc available about porting ocaml
>>>>> (google didn't find any) or any hint on how to move forward? I'd
>>>>> really like to have the native compiler on minix too.
>>>>
>>>> I think in the past it was possible to generate C code from OCaml
>>>> code.
>>>
>>> There is no real C backend (and has never been). The only thing you
>>> can do is to embed bytecode in a C-friendly way (what -output-obj
>>> does). But it is still bytecode.
>>
>> Benoit Vaugon & Michel Mauny worked recently on OcamlCC
>> (a translator from Ocaml bytecode to C)
>> http://oud.ocaml.org/2012/abstracts/oud2012-final10.pdf
>> https://github.com/ocaml-bytes/ocamlcc
>
> Very interesting work. I somehow missed that so far. I did something similar prior to starting with OCamlJIT2, but that turned out to be very slow with the compilation of the C code. I'd be interested to hear about results from generating LLVM code instead of C. Actually I'm still thinking that a tracing JIT using LLVM to optimize/generate code could be a viable option for the bytecode runtime (this is one of the results we gathered while working on Colin's thesis, the LLVM backend for the native code compiler).
>
> greets,
> Benedikt
> --
> Caml-list mailing list.  Subscription management and archives:
> https://sympa.inria.fr/sympa/arc/caml-list
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs