lindsay@k.gp.cs.cmu.edu (Donald Lindsay) (01/01/70)
Anyone writing an optimizing compiler would love to have a timing table, which tells him exactly how many clocks any given thing takes. Complex machines have complex tables, because clock counts depend on the prior machine state. Plus, complex instructions can have interactions between the addressing modes. Still, if that were all, (and if you ignore caches) the tables could be useful. The real killer is that fact that any successful machine will be re-implemented, differently. I'm sure that code tuned for the 68010 can be retuned for the 68020. You don't want to know how many 1750A designs there are out there. For vaxen, compiler writers have given up. The idea now is that "smallest is probably fastest". On RISC machines, you would hope for a timing table that was of real use. Is there any data yet about multiple implementations ? -- Don lindsay@k.gp.cs.cmu.edu CMU Computer Science