| 1 |
On 5/16/06, James Ausmus <james.ausmus@×××××.com> wrote: |
| 2 |
> > Does the `O' (uppercase oh) have an `s' component? |
| 3 |
> > |
| 4 |
> Yes - the -O setting is the level of code optimization that gcc does - |
| 5 |
> -Os is very similar to -O2, but also does code size optimization, |
| 6 |
> which may, under some circumstances, result in slower performance, but |
| 7 |
> will also, in other circumstances, result if faster performance. |
| 8 |
|
| 9 |
BTW, I measured the performance differences for the things I care |
| 10 |
about (compression, media encoding, and dm-crypt encryption), and |
| 11 |
ended up choosing -Os for my Core Duo system. As James says, some |
| 12 |
things run faster, other things run slower. And it isn't across the |
| 13 |
board...bzip2 -1 can give completely different results than bzip2 -9. |
| 14 |
The differences generally are within +/- 10% of the -O2 performance. |
| 15 |
|
| 16 |
The major advantage to -Os (aside from smaller executables), is |
| 17 |
compile time. Especially C++ programs take noticably less time to |
| 18 |
compile with -Os than with -O2 or -O3. |
| 19 |
|
| 20 |
-Richard |
| 21 |
|
| 22 |
-- |
| 23 |
gentoo-user@g.o mailing list |
Archives