At 4:47 PM 07/07/94 +0100, Graham Toal wrote:
PS I dunno what superoptimisizer Perry is talking about but I've never heard of a real one that works. You have to feed in a complete machine description at register transfer level and i don't know if those exist for real machines; also the problem is almost certainly exponential time for a *guaranteed* solution as Perry claims is possible.
The only tool I have ever seen that created real results was a tool that caused more headaches than solutions. (Inside, proprietary tool, can't go into details) It only worked on its native platform and one could feed it up to about 4K of code to analyse. In one test I fed it approx 60 lines of code. It built a database of 1.2 gigabytes, crunched for about 20 days, and bombed. This repeated for quite a while. Someone eventually got it to work, and it really did provide a pretty optimal solution. I have to wonder if people could have done it in less than 4 months... -j -- "Blah Blah Blah" ___________________________________________________________________ Jamie Lawrence <jamiel@sybase.com>