Gentoo Archives: gentoo-performance

From: Bart Alewijnse <scarfboy@×××××.com>
To: gentoo-performance@l.g.o
Subject: Re: [gentoo-performance] inline considered harmful
Date: Thu, 22 Jul 2004 16:59:43
Message-Id: b71082d8040722095938f69b08@mail.gmail.com
In Reply to: Re: [gentoo-performance] inline considered harmful by Jerry McBride
1 > I've been there, done that and bought the t-shirt... Well... maybe I just
2 > looked at the t-shirt....
3 >
4 > Anyway, I tried Os, O2 and O3 on a number of different packages on a gentoo
5 > box running on an AMD XP2500 clocked to 2.5ghz. My test results were at best
6 > ambigeous and didn't reflect what the main stream purporters of "BIG
7 > PERFORMANCE GAINS" with magical compiler settings. At first I thought I was
8 > onto something, but then, like yourself figured out that it was mostly "smoke
9 > and mirrors". Compiler options are best used to optimized specific
10 > applications. Over all use, across an entire linux installation... no one
11 > option is better than the other. That's been my observation, shoot me if you
12 > wish. :')
13
14 Heck no. You're right, of course, and if you want best performance
15 there's a metric arseoads of things to consider, not least of all the
16 processor, motherboard, memory model, speed, latency settings, size of
17 cache lines, cache algorithms -- and perhaps more importantly the
18 number of running threads. All this is rarely worth the bother,
19 especially the uninformed bother (read: just abuot everyone's
20 fiddling, including mine). I don't prentend to know particularly much,
21 it's just that these days there's so little difference I figured I
22 might as well try to save your cache from unnecessary use - with 80
23 processes being something of a minimum, ideally in cache, as that what
24 it's for - and if the code runs no slower, at worst (and probably
25 quite commonly) there's no gain. So I use it, for both my systems.
26
27 I'm just wondering whether Os does better on average on modern cpu's
28 all else being the same, because of the shifts of importance of moving
29 data, and with bigger caches avoiding loading code from memory all the
30 time. Considering the amount of threads on a system these days.
31 There's a small parallel with swapping here. Uh, yeah, that, hrm, slow
32 medium, 'main memory'. Okay, maybe not so much:)
33
34 So really, I'm just looking for the largest affector and spamming this
35 list with my fiddling, heh.
36
37 > I just run
38 >
39 > CHOST="i686-pc-linux-gnu"
40 > CFLAGS="-O2 -march=athlon-xp -fomit-frame-pointer -pipe"
41 > CXXFLAGS="${CFLAGS}"
42 >
43 > in /etc/make.conf and forget about the optimization madness...
44 Same here. Except I'm having fun with -Os right now. Oh, and -ftracer
45 because a friend sugegsted that in the same line of reasoning, simpler
46 code is better. But yeah.
47
48 > Besides, most of the ebuilds specify their own optimization settings
49 > anyways...
50
51 I know. Often because if they don't things break, or it makes
52 supporting harder. Joy:)
53
54 --Bart Alewijnse
55
56 --
57 gentoo-performance@g.o mailing list

Replies

Subject Author
Re: [gentoo-performance] inline considered harmful Jerry McBride <mcbrides9@×××××××.net>