Re: Right of Optimize Eager Evaluation Away

[Nick Roberts <nickroberts@callnetuk.com>]

Laurent Guerby wrote:
> 
> Robert A Duff <bobduff@world.std.com> writes:
> > Laurent Guerby <guerby@acm.org> writes:
> > > The question is: is a smart Ada 95 compiler allowed to generate code
> > > that looks like this:
> > >
> > >    if C then
> > >       X := Super_Expensive_Function_Call_1;
> > >    else
> > >       X := Super_Expensive_Function_Call_2;
> > >    end if;
> > >
> > > that is to say, be lazy about its argument, and so might
> > > save execution time.
> > No.  The function arguments may be evaluated in either order, so in case
> > of exception, the other one might not happen.  But without exceptions,
> > both arguments will be evaluated.
> 
> Is the potential raising of an exception the only language barrier
> here (since as you mention the compiler is free to swap calls
> defeating a class of side effects)? This seams easy to fix (via a
> pragma), if one is interested in keeping the functional aspect of the
> code instead of resorting to statements of course.

I didn't receive Laurent's original post (poor UK feed), so - forgive me
- I must guess at what he was originally suggesting, that a call such as

   P1(X,Y);

where P1 was declared by

   procedure P1 (A: Boolean; B: T) is
   begin
      if A then
         P2;
      else
         P3(B);
      end if;
   end;

would be optimised to evaluate X first, and omit the evaluation of Y if
X turned out to be True.

The RM95 says that a conforming implementation is allowed to execute an
Ada program in any way which produces the required external effects
[RM95 1.1.3(8-15)]. (Which is plain common sense, anyway, really.)

This means that a compiler can perform any optimisation that doesn't
change the external effects of a program. So the suggested optimisation
is perfectly acceptable, provided the compiler can _guarantee_ that it
will not change the external effects of the program.

I get the impression that, in practice, there are not many Ada compilers
which 'strongly' optimise (maybe there aren't any). However, I think a
strongly optimising compiler (by today's standards) would be very
likely, in effect, to produce the suggested optimisation, in the case of
the subprogram P1 being expanded inline, or local and independent, and
provided the compiler could be sure that the evaluation of Y could not
have raised an exception or caused any external effect.

-- 
Nick Roberts
http://www.adapower.com/lab/adaos
Always call for the professionals. (If they don't help, call for me ;-)