1 |
On Thu, 12 Feb 2009 06:20:28 -0800, Mark Knecht wrote: |
2 |
|
3 |
> > You have more aggressive flags on the failing machine, have you tried |
4 |
> > using the more conservative (less ricey) flags? |
5 |
|
6 |
> I have not and intend to do that today. The flags have been set like |
7 |
> this since 2003 so I'm surprised that it might be something like this |
8 |
> but it's clearly possible. |
9 |
> |
10 |
> I'm unqualified to say what part of the differences is making this set |
11 |
> more aggressive. Line by line what's your opinion? |
12 |
> |
13 |
> -O3 vs -O2 ? |
14 |
|
15 |
Not much difference. |
16 |
|
17 |
> -march=athlon-xp vs -march=i686 |
18 |
|
19 |
The former is better, providing you're using an AthlonXP. |
20 |
|
21 |
> -funroll-loops |
22 |
|
23 |
You only have to look at http://funroll-loops.org/ to know this is a bad |
24 |
option... |
25 |
|
26 |
> not using -fomit-frame-pointer |
27 |
|
28 |
That's pretty safe, the frame-pointer isn't much use anyway if you're |
29 |
compiling without debug symbols, which most build do by default. |
30 |
|
31 |
> Strange taht after 6-years this would come up but stranger thing happen. |
32 |
|
33 |
Indeed. |
34 |
|
35 |
|
36 |
-- |
37 |
Neil Bothwick |
38 |
|
39 |
Obscenity is the crutch of inarticulate motherfuckers. |