| 1 |
Many thanks for your hints! |
| 2 |
|
| 3 |
On 05/14/2017 07:13:05 AM, Walter Dnes wrote: |
| 4 |
> Why aren't you running "-march=native"? You're probably missing |
| 5 |
> quite |
| 6 |
> a few features of your cpu. BTW, "-march=native" implies |
| 7 |
> "-mtune=native" |
| 8 |
> and "-msse3" (if the cpu supports it). Out of sheer curiousity, what |
| 9 |
> does gcc report as your native cpu when you run... |
| 10 |
> |
| 11 |
> gcc -c -Q -march=native --help=target | grep march= |
| 12 |
|
| 13 |
-march=amdfam10 |
| 14 |
|
| 15 |
> |
| 16 |
> > and quite a e few |
| 17 |
> > CFLAGS="-O3 -mtune=native -pipe -msse -msse2 -msse3 -msse4a -m3dnow" |
| 18 |
> > (for my aged AMD64 Phenom II machine) |
| 19 |
> > |
| 20 |
> > in /etc/portage/env/... |
| 21 |
> |
| 22 |
> "-O3"... yikes! Not really recommended. If you want speedup, use |
| 23 |
> "-march=native" instead. |
| 24 |
|
| 25 |
I didn't have any problems with that. I only use these setting non |
| 26 |
compute intensive packages which |
| 27 |
are not system critical. |
| 28 |
> |
| 29 |
> In make.conf, I have... |
| 30 |
> |
| 31 |
> CFLAGS="-O2 -march=native -mfpmath=sse -fomit-frame-pointer -pipe |
| 32 |
> -fno-unwind-tables -fno-asynchronous-unwind-tables" |
| 33 |
> |
| 34 |
> The 2 "unwind_tables" flags reduce code bloat. See |
| 35 |
> http://comments.gmane.org/gmane.linux.busybox/36695 for details. The |
| 36 |
> guys at busybox do what they can to reduce code bloat. Knocking 15% |
| 37 |
> off |
| 38 |
> the size of an executable or library helps. This results in faster |
| 39 |
> loading at startup (less to load), and less stuff in ram cuts down on |
| 40 |
> swapping. |
| 41 |
|
| 42 |
Many thanks, again |
| 43 |
Helmut |
Archives