Article 52199 of comp.sys.dec: Jeroen T. Vermeulen wrote: > > In article <33831687.190C@world.std.com> David Chase writes: > > > Since then, I've also looked at the return address prediction claim, > > (a stack implemented in a ring of registers; hard to think of > > another way to do it if you intend to do this, and DEC has the > > patent on it) and I strongly suspect that Intel does something > > very similar to that, based on their code generation advice to > > "always pair calls and returns". > > This reminds me... When the Pentium came out there was mention of a mysterious > "Appendix H" describing Pentium scheduling rules and other optimizations, which > apparently was so hard to come by that I recall one Intel employee posting that > his copy had been stolen from his desk. > > Might this secrecy have had anything to do with covering up potential patent > infringements? AFAICS it could hardly have served to protect Intel's own > inventions, since Intel could simply have patented those for themselves. > Releasing a public Pentium optimization guide, even giving it away for free, > would then have benefited Intel at least as much as anybody else! > > Example: IIRC Terje Mathisen discovered and posted some puzzling irregularities > in Pentium branch performance. Intel might have increased their chips' > effective performance by publicizing expected irregularities in advance, no? The Pentium Branch Target Buffer is actually even more interesting than what I posted about in that long-gone article. Agner Fog and Jitendra Karki took my initial guesses and did a real thorough reverse engineering job on the BTB hardware. I believe you can find the article they wrote on Agner's home page: (http://announce.com/agner/) most probably in the the Pentium Optimization guide (http://announce.com/agner/assem/pentopt.zip) Terje PS. I should perhaps reiterate that I have no ties whatsoever with Intel, besides the fact that some of my friends work there. The software workaround I developed (under Cleve Moler's direction) for the FDIV bug was done for free, in my spare time. -- - Using self-discipline, see http://www.eiffel.com/discipline "almost all programming can be viewed as an exercise in caching"