1 |
On Fri, 18 Jul 2008 15:34:53 -0400 |
2 |
Richard Freeman <rich0@g.o> wrote: |
3 |
> Ciaran McCreesh wrote: |
4 |
> > The more interesting question, then, is whether users run any |
5 |
> > non-trivial cpu-bound programs. We know the applied science types |
6 |
> > do, but they tend to be the ones who're doing clever things with icc |
7 |
> > anyway. What about normal users? |
8 |
> |
9 |
> I'm sure they do on some occasion if they encode compressed |
10 |
> audio/video, or when compressing data with zip/etc. That is probably |
11 |
> the biggest application of cpu-bound software. |
12 |
|
13 |
How much of that is memory bound? Of the things that aren't, how many |
14 |
aren't written in assembly anyway? Of those things, what proportion of |
15 |
the runtime is spent in those areas? |
16 |
|
17 |
If you double the speed of something that takes up 2% of the overall |
18 |
execution time, you can't measure the improvement. |
19 |
|
20 |
Or looking at it the other way -- is there any reason to believe that |
21 |
using icc (which can end up being a substantial pain in the arse, given |
22 |
the way it tries to use gcc's c++ headers but doesn't support some of |
23 |
the extensions or quirks that g++ does) will provide a genuine gain |
24 |
for people who aren't already doing clever profile-directed trickery |
25 |
anyway? |
26 |
|
27 |
> I'd probably benefit from using -O3 on the aforementioned |
28 |
> CPU-intensive apps. |
29 |
|
30 |
The problem with -O3 is that function inlining can lead to a |
31 |
substantial cache hit. Unless you're using profile-directed |
32 |
optimisations, which Gentoo doesn't support, it's extremely hit and |
33 |
miss as to whether O3 helps or hurts. |
34 |
|
35 |
-- |
36 |
Ciaran McCreesh |