| 1 |
David Nielsen said: |
| 2 |
> The -mcpu flag is depricated in GCC 3.3 - and should not be used |
| 3 |
anymore. So in all fairness it would be best not to record it's usage, |
| 4 |
it might mislead some users. |
| 5 |
|
| 6 |
The only depreciation I could find concerning mcpu is in the mips port, |
| 7 |
where it might not me appropriate. For x86 architectures it is available |
| 8 |
though and it also serves a needed option as it determines instruction |
| 9 |
scheduling. The march option does not, the march option determines which |
| 10 |
instructions are used. As scheduling for an architecture lower than the |
| 11 |
instructions is pointless implying mcpu makes sense. It is perfectly legal |
| 12 |
though to specify |
| 13 |
CFLAGS="-march=pentium -mcpu=pentium4" This would create code that runs on |
| 14 |
anything bigger than a pentium, but is optimally scheduled for the |
| 15 |
pentium4 processor. Scheduling has become increasingly important with the |
| 16 |
newer (pentium and further) processors as they have the ability to run |
| 17 |
some (but not all) instructions paralel to eachother. A p4 though provides |
| 18 |
more pipelines than a p1, so instructions need to be scheduled differently |
| 19 |
to optimally use all pipelines. |
| 20 |
|
| 21 |
Paul |
| 22 |
|
| 23 |
-- |
| 24 |
Paul de Vrieze |
| 25 |
Researcher |
| 26 |
Mail: pauldv@××××××.nl |
| 27 |
Homepage: http://www.devrieze.net |
| 28 |
|
| 29 |
|
| 30 |
-- |
| 31 |
Paul de Vrieze |
| 32 |
Researcher |
| 33 |
Mail: pauldv@××××××.nl |
| 34 |
Homepage: http://www.devrieze.net |
| 35 |
|
| 36 |
|
| 37 |
|
| 38 |
|
| 39 |
-- |
| 40 |
gentoo-dev@g.o mailing list |
Archives