Re: [gentoo-dev] x32 fun pants

[Joshua Kinard <kumba@gentoo.org>]

[-- Attachment #1: Type: text/plain, Size: 1958 bytes --]

On 09/16/2011 09:36, Donnie Berkholz wrote:

> 
> For anyone interested how the performance compares to amd64 in more 
> comprehensive tests, check out the slides from the Linux Plumbers 
> Conference (particularly 14-21):
> 
> http://linuxplumbersconf.org/2011/ocw/proposals/531
> 
> In summary, on those benchmarks it looks like a small global win (maybe 
> 5%) on integer calculations with a few huge wins of ≥25%, but a net loss 
> around 5% pretty much globally for floating-point calculations.
> 
> Most people probably do a lot more integer calculations unless they're 
> science geeks like me, plus it should have lower memory use, so my 
> understanding is that it probably makes sense to switch to x32 no matter 
> what you're using now (x86 or amd64).
> 
> Mike, would you agree?


Again, extremely similar to MIPS cases from a few years ago.  While even n32
is fairly stable in Linux (last few years, at least), the idea was always
that in an ideal multilib scenario, you'd use pure 32bits (MIPS o32) in very
limited cases (programs that just didn't work in either n32 or n64), n32 for
a majority of the system, and pure 64bit (n64) for specific applications,
like databases, crypto, or science applications.

That's supposed to provide the balance so that float-intensive apps can use
pure 64bit w/o penalty, but things that simply don't need 64bits' full power
can make use of n32/x32.

And yeah, lower memory use because the size of codewords is smaller in
memory overall.

Anyone wanting to compare x32 and n32 can see the old n32 ABI guide here:
ftp://ftp.linux-mips.org/pub/linux/mips/doc/ABI2/MIPS-N32-ABI-Handbook.pdf

-- 
Joshua Kinard
Gentoo/MIPS
kumba@gentoo.org
4096R/D25D95E3 2011-03-28

"The past tempts us, the present confuses us, the future frightens us.  And
our lives slip away, moment by moment, lost in that vast, terrible in-between."

--Emperor Turhan, Centauri Republic


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 834 bytes --]