Re: [hatari-devel] 68030 cycle timings for 64bit div/mul

[ Thread Index | Date Index | More Archives ]

Yes that's roughly what I was thinking of - issuing a limited set of quasi-random samples for each 2^n range, to get a sensible bound and decent distribution, and figure out a curve or pattern to the costs.


I started work on the 32-bit divides but that's paused ATM.  The problem there
of course is that it's impossible to perform every possible divide and check
the results, so the code tries to do quasi-random, quasi-boundary tests.  I
have some preliminary results, but I need to fire up the TT again to continue.
If anyone is interested in me continuing (or in getting my code & pursuing it
themselves), let me know.

Roger Burrows

Mail converted by MHonArc 2.6.19+