Re: SUMMARY: pentium optimized debian discussion
Jason Gunthorpe <jgg@gpu.srv.ualberta.ca> writes:
> The size gain in pentium optimized binaries is not from nops but from
> alignment constrains. Structure members are aligned on paragraphs and
> some function elements are aligned on pargraphs too.
Yes, that makes more sense. I know a lot about supserscalar
architecture design in general (I've read Mike Johnson's book, etc.)
but not much about Intel architecture in particular. Most
superscalars have an advanced enough prefetch so that instruction
misalignment is not really a problem, but data misalignment still is.
> I actually don't belive there will be a measuable slowdown on 486's. (386s
Neither do I. I don't even have a 486 or 386 to test it on. Would
somebody with access to one produce some hard data? I'm happy to
recompile some software on my Pentium to see what sort of CPU increase
and size increase we get. What are some reasonable benchmarks?
Recompiling certain CPU-intensive programs is a laudable goal, but
perhaps the complexity of a second architecture, more archive bloat,
the necessary changes to dpkg, etc. aren't necessary.
Guy
--
To UNSUBSCRIBE, email to debian-devel-request@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Reply to: