| 1 |
On Sat, Feb 25, 2012 at 10:30:21PM +1100, Adam Carter wrote |
| 2 |
|
| 3 |
> Playing video is one of few situations in which optimisation makes a |
| 4 |
> lot of difference though, thanks to the mmx/sse stuff, which is post |
| 5 |
> i686. So a video benchmark will should show that up, but the boost may |
| 6 |
> be lost in a more general suite of weighted benchmarks. |
| 7 |
|
| 8 |
Today's web is heavily multimedia. There are some text-oriented |
| 9 |
websites where w3m or links is sufficient, but they're few and far |
| 10 |
between. And as I mentioned earlier, the appropriate comparison is |
| 11 |
between a fully optimised system versus a fully default system, dual |
| 12 |
booting on the same machine to eliminate hardware biases. Linux |
| 13 |
works with libraries. Firefox calls a lotlibraries, which call other |
| 14 |
libraries and the kernel, too. That's why it has be the entire system |
| 15 |
being optimized. |
| 16 |
|
| 17 |
> Also, different versions of gcc optimise differently. Usually the |
| 18 |
> optimisation gets better but there a quite a few cases where |
| 19 |
> performance regresses. |
| 20 |
|
| 21 |
??? Can you name any cases where "-march=native" is worse than than |
| 22 |
the default i686 build? And please do an apples-to-apples comparison. |
| 23 |
I.e. pgo bin versus pgo build from source. |
| 24 |
|
| 25 |
-- |
| 26 |
Walter Dnes <waltdnes@××××××××.org> |
Archives