Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Not allowed yet.

Ex-Intel here. It takes years to gestate new instructions. First specs are in controlled documents available to Intel employees only. Eventually, when things are nailed down, preliminary specs are available under an NDA where a VP approves/signs the Intel side of the NDA. Tool vendors will get those specs. Finally, when the chip is announced the previously NDA documents become public. I was a CPU designer at multiple companies, and everyone follows a similar process.



What about the infamous i386 LOADALL instruction? Certainly there is at least one that is kept secret.

http://www.rcollins.org/articles/loadall/tspec_a3_doc.html


You're just talking about the design phase of a new ISA extension. I think the parent was under the impression that there were instructions in released processors that were undocumented.


And that impression would be correct.

Furthermore, if you are a big enough customer you can also negotiate for special isa extensions yourself (can't provide citation but heard from people at Intel), then of course you'd be the only one to have the documentation to take advantage of them.


There is at least one Intel processor with a public erratum to an instruction (or other mechanism?) that is not public. It's a timed MWAIT, I think.


That was exactly my impression. I don't have a problem with new instructions in new processors being undocumented at first, but purposely undocumented instructions are troubling and would push me away from Intel for new designs.


So, you are the decision maker for whether or not Intel CPU's get designed into your product? If so, I'm sure several Intel salespeople have you on speed dial.

The 286 had purposely undocumented instructions, of this sort: "Ooops this is b0rk3d. It will always be b0rk3d. Let's pretend it didn't happen." So for generations there were holes in the op code map that people tip-toed around. Especially since Intel (meaning the internal grey-beard collective) also forgot exactly what those opcodes were and what they were supposed to do. You care, why, exactly?

It's not like the NSA slips extra opcodes into executables that you compile with your own compiler in order to spy on you. They have much easier ways to spy on you.

Also, it's not like it is that hard to throw unused opcodes at the decoder and see which ones give you the illegal instruction exception, and which ones do something else. You now have a homework assignment. Have fun, let us know what you find.


There are.


The story of FXSAVE/SSE (I wonder if you were there for it):

http://web.archive.org/web/20000817193452/http://www.tbcnet....

http://web.archive.org/web/20000817084210/http://www.tbcnet....

Obviously, XSAVE/AVX was handled better.


> previously NDA documents become public

I'd say some are eventually made public. I went to the IDF in San Francisco expecting this sort of information on Skylake. What a waste of a day.

Not that I'm complaining. Intel's information is excellent. It just arrives when it arrives and it is what it is at that point in time. There are many ways involving effort which glean more information: Agner Fog, articles, patents, .... BTW, Intel folks are helpful on their dev board.


XED author here...

@dbcurtis: Yes, that's what I meant. Thanks for clarifying.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: