I added this table at the end of 1996 (1996-12-28) to my PL/I version of the program (the PC version is written in Pascal with an unhealthy amount of code converted into assembler), and the algorithm was quite smart, or so I thought. Working in the Netherlands at the time, at a site where I could use Strobe, I soon realized that my "quite smart" algorithm was not very smart at all, despite all the optimizations had had come up with, but couldn't think of anything better.
So in 1998 I asked around in the comp.lang.pascal.borland Usenet group, and soon after I got two better versions. I don't have the Strobe report any more, but occasionally I compile the old AD 1998 program with the PL/I "COUNT" option and the results are pretty shocking, when compared to the current implementation:
Code: Select all
NAT_SCAN 125,241 0.7952% NAT_SCAN 1,815,960,038 98.8388%
Moral of the story, first check your algorithm, and only then start thinking about optimisation!