Re: [tlug] CentOS to Gentoo-Pros/Cons(if any)?

[Josh Glover <jmglov@example.com>]

On 14/12/05, Stephen J. Turnbull <stephen@example.com> wrote:

> But my main point is that for a system like XEmacs (substitute Python,
> Perl, Java, Ruby, VM-based-interpreter-du-jour) smaller _is_ faster.

Interesting

> But I can also say that whenever people do optimization tests on XEmacs, -Os is the only
> thing that gives measurable speed up (~2%) over -O2 or -O3.[1]

*Very* interesting! I will have to try this with some of my code to
see if I get similar results.

> I guess if you're the bad old AT&T you can afford custom-built CPUs
> with quadruple-size on-chip cache, but really, is that any way to run
> a stock exchange?

Certainly not, but that is The Way Things Are Done in telecom. ;)

However, the principle of trading size for speed (or vice-versa) is a
sound engineering practise, and every coder should know how to do it.

> Footnotes:
> [1]  You will also find places (in src/buffer.c) where code
> duplication and C-level loop unrolling is rampant.  This also gives
> measured speed up, but it's carefully targeted, deliberate
> optimization.

"[C]arefully targeted, deliberate optimization" is the only kind that
is worth a damn, as both you and Donald Knuth have noted.

Speaking of which, as I was walking through the warehouse a couple of
days ago, I saw that we had at least one copy of "The Art of Computer
Programming" boxed set[1] in stock! I may have drooled a little.

-Josh

[1] http://www.amazon.co.jp/exec/obidos/ASIN/0201485419/qid=1134525553/sr=8-6/ref=sr_8_xs_ap_i6_xgl14/249-2441462-6881135