[MLton-devel] icfp

Suresh Jagannathan MLton@mlton.org
Mon, 24 Feb 2003 17:56:23 -0500


I think it's doable; we should at least try given the makeup
of the committee, and that whatever we write would fit naturally
into the opus backend.   The question that I'm not clear on is
exactly what information the profiler could give us to motivate 
our claims.  The two most obvious ones I see are (a) the 
allocation costs for manipulating simple types (e.g., space
allocated due to polyvariance in the defunctorizer, duplication
in the monomorphiser, and defunctionalization for closure
conversion) is "small" relative to a program's total allocation;
and (b) the time spent in creating these types is "small" relative
to the program's total running time.  This, along with overall
performance results, may be enough, but I'd like to think there's other
interesting data we can manipulate the profiler to reveal that
might strengthen things further.  I'm not quite sure what though.
At any rate, we've got over a month so it still seems worth a shot.
I'm writing opus related stuff now anyway, so it should be possible
to plug sections together quickly if we need to.  The profiler itself
is novel enough to warrant a significant section, and its implementation
ties pretty well with this general claim that simple (and simply-typed)
ILs are a good idea.



On Monday 24 February 2003 05:38 pm, Henry Cejtin wrote:
> What we did long ago (maybe the closure conversion paper) was static and
> dynamic counts of the number of dispatches taked and their various sizes.
> Do you really think ICFP is do-able?  I told Andrew that it was too early.
>
>
> -------------------------------------------------------
> This sf.net email is sponsored by:ThinkGeek
> Welcome to geek heaven.
> http://thinkgeek.com/sf
> _______________________________________________
> MLton-devel mailing list
> MLton-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/mlton-devel


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
MLton-devel mailing list
MLton-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/mlton-devel