Re: Fortran: The Road Ahead.

[Paul Koning <pkoning at xedia dot com>]

>>>>> "Toon" == Toon Moene <toon@moene.indiv.nluug.nl> writes:

 Toon> Several weeks ago I promised to write up the optimisation
 Toon> opportunities I see as being applicable to g77 generated RTL as
 Toon> soon as the first release of egcs hit the archives.

 Toon> That write-up is now in front of you.

Nice writeup.  I'd like to add to it that some of this is not just
valuable to Fortran code.  Recently I tried out some simple DSP code
on egcs to see what it would do with it.  That exposed a number of the
same limitations Toon mentioned.  In C it was fairly easy to work
around some of these by fiddling with pointers instead of using
indexing, but for clarity of code it certainly would have helped to
have some of the things Toon mentioned.

By the way, perhaps unrelated but observed at the same time: the
compiler seemed to produce stack references for locals even when it
didn't need to, i.e., the register allocation was missing
opportunities to use registers instead of stack memory temporaries.
And, even more curiously, at some point I managed to eliminate all
stack references in the body of my function but even so ended up with
a stack locals frame and a store into a stack temporary that was never
referenced again.

	paul