| 1 |
On Mon, 09 Apr 2007 05:48:07 +0400, maxim wexler <blissfix@×××××.com> |
| 2 |
wrote: |
| 3 |
|
| 4 |
> Hi group, |
| 5 |
> |
| 6 |
> I note two schools of thought on the best CFLAGS for |
| 7 |
> the Pentium III processor. |
| 8 |
> |
| 9 |
> One suggests using -O3 -pipe, the other, -O2 without |
| 10 |
> the pipe. |
| 11 |
> |
| 12 |
> How much difference does this make? Is the extra level |
| 13 |
> of optimization with pipe the equivalent of the lower |
| 14 |
> level without? |
| 15 |
> |
| 16 |
|
| 17 |
Man gcc. -pipe allows to use pipes (memory) instead of temporary files |
| 18 |
(disk). -O controls the resulting binary sometimes making it fast, large, |
| 19 |
and unreliable. I have seen ebuilds that enforce -O2 even when -O3 is |
| 20 |
specified. This can create a faulty impression that -O3 always works fine. |
| 21 |
|
| 22 |
-O and -pipe interact through the size of temporary files. Theoretically, |
| 23 |
-pipes may slow down compiles if there is not enough memory. Another |
| 24 |
factor that should be taken into account is the number of threads for |
| 25 |
make. For a PIII machine the generally recommended number is 2, but when |
| 26 |
the RAM is limited, say 384 megs, 1 thread with pipes works better. |
| 27 |
|
| 28 |
Ideally you should measure compile times for large projects like Open |
| 29 |
Office and find out what works best on your system. I would start with 1 |
| 30 |
thread -O2 and -pipe. |
| 31 |
|
| 32 |
-- |
| 33 |
Andrei Gerasimenko |
| 34 |
-- |
| 35 |
gentoo-user@g.o mailing list |
Archives