glen herrmannsfeldt wrote:
Michael J. Mahon wrote: (snip)The FREQUENCY statement that provided the probabilities of conditional branches for this simulation was relatively short-lived, partly because we now know that programmers tend to be *terrible* estimators of branch probabilities!Yes, but it was optional so I don't see why it needed to be removed. Now that branch prediction is more important, maybe it should be added back again.
Branch frequency information is much more accurately obtained as a result of profile feedback. Programmer estimates are notoriously inaccurate. Since profiles reflect only the test workload, there is motivation to collect profiles dynamically, and to optimize frequently executed code discovered in this way. On systems with fewer memory constraints, this has been demonstrated to produce significant speed improvements with relatively low overhead. I'm not aware of any experiments along these lines on the Apple II. -michael NadaPong: Network game demo for Apple II computers! Home page: http://members.aol.com/MJMahon/ "The wastebasket is our most important design tool--and it's seriously underused."