| 1 |
On Dienstag 03 Februar 2009, Grant Edwards wrote: |
| 2 |
> Whenever I see a write-up of Gentoo, it's describe as a system |
| 3 |
> similar to BSD "ports" where you build packages from source. |
| 4 |
> The main benefit claimed for this approach is that you get |
| 5 |
> better performance because all executables are optimized for |
| 6 |
> exactly the right instruction set. |
| 7 |
> |
| 8 |
> Where did that bit of apocrypha come from, and why is it |
| 9 |
> parroted by so many people? |
| 10 |
|
| 11 |
because it was true in the beginning, when most distris were still built for |
| 12 |
386 and the difference of an optimized built was that you could watch dvd |
| 13 |
movies without hangs and frame loss. |
| 14 |
|
| 15 |
It is still true to a certain degree today - code compiled for 386 runs much |
| 16 |
slower than code compiled for core2 - on a core2. But on todays overpowered |
| 17 |
cpus you don't see it as prominent as back on k6-2 400 or p3 650 .... |
| 18 |
|
| 19 |
|
| 20 |
> AFAICT, the "performance" benefit due to compiler optimization |
| 21 |
> is practically nil in real-world usage. |
| 22 |
|
| 23 |
nope, they are there. But compiler optimiziations are a very delicate thing. |
| 24 |
You can't just throw funroll-all-loops into make.conf and think that was it. |
| 25 |
And for a general set, march is the most important one. It does do a lot of |
| 26 |
good - the rest is just minor at best. |
Archives