[m-rev.] for review: fix bug in dep_par_conj

[Zoltan Somogyi]

On 14-Oct-2008, Peter Wang <novalazy at gmail.com> wrote:
> > +            ParConjunctStatus = par_conjunct_is_proc_body
> > +            % We don't need to rename anything, XXX.
> > +        )
> 
> What's the XXX for?

To remind me to cover this in the paper.

I followed your other suggestions.

BTW, due to the bug, I could not make the oct 10 deadline for DAMP,
but I am trying for the nov 14 deadline for PLDI. However, I am considering
changing the algorithm, first in the compiler and then in the paper to be
more aggressive in delaying waits: if some arms of e.g. a switch do not
consume a variable, then the compiler can make the arms that do consume it
wait for distinct renamed versions of it, and let any code after the switch
wait for it again. Actually, the compiler should be able to choose, preferably
on the basis of coverage profiling information, between that approach and the
current one.

Conversely, we probably want to back off the aggression in specialization.
If the called procedure needs an input immediately or (more likely) produces
an output only at the very end, then creating a specialized version with the
wait or signal pushed into it increases program size without reducing runtime.
I will do this change first, after committing this diff.

Zoltan.
--------------------------------------------------------------------------
mercury-reviews mailing list
Post messages to:       mercury-reviews at csse.unimelb.edu.au
Administrative Queries: owner-mercury-reviews at csse.unimelb.edu.au
Subscriptions:          mercury-reviews-request at csse.unimelb.edu.au
--------------------------------------------------------------------------