🚀 go-pugleaf

gary@fairlight.oz (Gary Evesson) writes:
>This is an idea that I thought about in the early days of the Mac. I think it's
>a good idea, but convincing someone to do it is another question. Writing
>optimisers is a art in itself - not an easy thing to do. It doesn't seem that
>any of the manufacturers of compilers for these machines are interested
>(I would welcome a contradiction from anyone) as far as I can tell. I guess the
>trick is to apply enough market pressure for them to do something about the
>quality of their code, which I agree is atrocious.

        Leon Frenkel's Benchmark Modula-2 compiler for the Amiga is fairly
well optimised.  I've had a few long discussions with him as to how to write
optimised source, and it is suprising the degree to which the coder can be
sloppy in his structuring and still get the same results.  Still it is
essential that certain coding guidlines be followed when possible (such as
creating optimized mathmatical expressions).

        I've been trying to convince Leon to implement some kind of programmer
controlled register allocation, and his counter argument has been that it is
non-standard, and in most cases the compiler makes better overall register
allocations than programmers would.  A strong supporting argument on his sid
is that if the coder should change previously optimized code, the compiler
will automatically re-optimize whereas the programmer often would not.
        None-the-less, I think at some point he will include programmer
register allocations via compiler directives, which I suggested (as it avoids
most transportability problems) and I think he will do, eventually...


                                                        Thanks,


                                                            Wade.


UUCP: {cbosgd, hplabs!hp-sdd, sdcsvax, nosc}!crash!pnet01!haitex
ARPA: crash!pnet01!haitex@nosc.mil
INET: haitex@pnet01.CTS.COM