Re: [Caml-list] Closing the performance gap to C

[Yotam Barnoy <yotambarnoy@gmail.com>]

> The issue could very well be addressed outside flambda, at the cmmgen level.
> See https://caml.inria.fr/mantis/view.php?id=5894 .   Making flambda aware
> of boxing can help its inlining decisions, but the notion of having
> specialized unboxing calling convention is not really tied to flambda.
>
> Pushing #5894 forward could have a direct impact on numerical code.  I think
> it would really be useful to work on that, even if flambda also does
> something along these lines, at some point.
>
> -- Alain

I think it's not worth the effort. You need to examine all the code
dealing with a parameter (ie. its flow) to see if any generic function
is called on that parameter. This is work best left to Flambda. Even
if a generic function is called on said parameter, you can still
express a cost-based approximation to determine whether unboxing is
worthwhile. This is all large-scale optimization rather than peephole
optimization, and fits better under Flambda's umbrella IMO.

-Yotam