Re: [Caml-list] Linear Scan Register Allocator for ocamlopt/ocamlnat

[Pierre-Alexandre Voye <ontologiae@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 1701 bytes --]

I have a stupid question : I wonder if it would not be a bad idea that Ocaml
output C code and let gcc do its work, so compile code with good
performances in a lot of architecture ? Gcc is able to do autovectorization
(SSE, MMX, Larabee in the futur, etc...), very specific processor
optimization, etc...
But maybe it's a stupid idea ?

2011/8/25 Gerd Stolpmann <info@gerd-stolpmann.de>

> Am Donnerstag, den 25.08.2011, 11:34 +0200 schrieb Benedikt Meurer:
> > On Aug 25, 2011, at 10:02 , Benedikt Meurer wrote:
> >
> > >>> -
> http://ps.informatik.uni-siegen.de/~meurer/tmp/compiletime_timings.pdfcontains a comparison of the ocamlopt invocations.
> > >>> - http://ps.informatik.uni-siegen.de/~meurer/tmp/runtime_timings.pdfcontains comparison of the generated code.
> > >>>
> > >>> As can be seen from the results, amd64 is more sensitive to register
> allocator changes than i386.
> > >>
> > >> Well, this particular i386 CPU model is a strange guy - Northwoods
> have
> > >> this extremely long pipeline, which is very sensitive to unforeseen
> > >> jumps. It would be more interesting to see this test on a modern CPU
> in
> > >> i386 mode. My guess is that it behaves then more like amd64.
> > >
> > > Modern CPUs most probably don't run ocaml in 32bit mode, but more
> likely in long mode. That's why we choose to run the i386 on "real 32bit
> hardware", where the ocaml i386 port is actually used.
>
> Right. However, if you look at new 32-bit-only CPUs like older Atoms
> these base on modern cores where only some of the circuits were omitted
> that are needed for 64-bit execution.--



> [cut]
>
---------------------
https://twitter.com/#!/ontologiae/
http://linuxfr.org/users/montaigne

[-- Attachment #2: Type: text/html, Size: 2655 bytes --]