1 |
You could just enable -O2 and the following: -frename-registers and -fweb. -O3 is simply -O2 plus those 3 options. |
2 |
|
3 |
On Wed, 21 Jul 2004 20:50:01 -0400 |
4 |
Adam Petaccia <adam@×××××××××.com> wrote: |
5 |
|
6 |
> With gcc, is there a way to enable all -O3 options but function |
7 |
> inlines? Would -fno-inline work or something like that? |
8 |
> |
9 |
> Mario Domenech Goulart wrote: |
10 |
> |
11 |
> >Hello, |
12 |
> > |
13 |
> >There's an interesting discussion in the OpenBSD mailing |
14 |
> >list about the use of inline. |
15 |
> > |
16 |
> >Here's the beginning of the thread about this topic: |
17 |
> > |
18 |
> >,----[ http://www.sigmasoft.com/cgi-bin/wilma_hiliter/openbsd-tech/200407/msg00175.html ] |
19 |
> >| inline considered harmful. |
20 |
> >| |
21 |
> >| * To: tech@×××××××.org |
22 |
> >| * Subject: inline considered harmful. |
23 |
> >| * From: Artur Grabowski <art@××××××××.org> |
24 |
> >| * Date: 21 Jul 2004 03:54:46 +0200 |
25 |
> >| * User-agent: Gnus/5.09 (Gnus v5.9.0) Emacs/21.2 |
26 |
> >| |
27 |
> >| Today we did a bunch of removal of inline functions in the kernel. |
28 |
> >| It all started to make floppies fit, but now it's a quest. |
29 |
> >| |
30 |
> >| If you think that I'm crazy doing this because it might hurt your |
31 |
> >| precious performance, go back to your vax and leave the performance |
32 |
> >| tuning to people who have a cache. |
33 |
> >| |
34 |
> >| Every single inline we removed today (and there are more in the |
35 |
> >| pipeline and even more waiting to be fixed) shrunk the code and MADE |
36 |
> >| IT FASTER. Yes, modern cpus have something called "cache". The cache |
37 |
> >| prefers the code to be smaller, rather than free from function calls. |
38 |
> >| Yes, some cpus have expensive function call overhead. Don't use them. |
39 |
> >| i386 has quite expensive function calls, on the other hand it doesn't |
40 |
> >| have any relevant amount of registers either. So a function call |
41 |
> >| instead of the same function inlined can potentially make the job |
42 |
> >| easier for the register allocator in the compiler which could eat the |
43 |
> >| overhead. At the same time the instruction cache can run the same code |
44 |
> >| in the same place, instead of loading it from main memory 4711 times. |
45 |
> >| And guess what? The stack on i386 is in the cache too, so the function |
46 |
> >| call overhead isn't that bad anyway. |
47 |
> >| |
48 |
> >| I'm tired of seeing code where everything is made inline just because |
49 |
> >| someone acted on a meme that hasn't been true for over a decade. Bloat, |
50 |
> >| bloat and more bloat. Since people can't use inline correctly (it does |
51 |
> >| have valid and correct uses), from now on inline in the OpenBSD kernel |
52 |
> >| is considered to be a bug until proven otherwise. So. Next time I see |
53 |
> >| code that adds to the bloat with inlines, I expect performance figures |
54 |
> >| and kernel size comparisons that show that the inline actually |
55 |
> >| contributes anything. Otherwise the code does not go in. |
56 |
> >| |
57 |
> >| There's still a lot of work to be done in the kernel (yes, macros can |
58 |
> >| be evil too, just see nfs), so send diffs. And there's a whole |
59 |
> >| unexplored field in userland too. |
60 |
> >| |
61 |
> >| //art |
62 |
> >`---- |
63 |
> > |
64 |
> >Mario |
65 |
> > |
66 |
> > |
67 |
> >-- |
68 |
> >gentoo-performance@g.o mailing list |
69 |
> > |
70 |
> > |
71 |
> > |
72 |
> > |
73 |
> |
74 |
> -- |
75 |
> gentoo-performance@g.o mailing list |
76 |
> |