1 |
Billy Holmes wrote: |
2 |
|
3 |
> Richard Fish wrote: |
4 |
> |
5 |
>> FYI, -mcpu is deprecated, and a synonym for -mtune, which is implied |
6 |
>> by -march. So you can just take it out and get the same effect. |
7 |
> |
8 |
> |
9 |
> I think some ebuilds filter out march, but don't filter out |
10 |
> mcpu/mtune, so you can still get some processor specific optimizations |
11 |
> out of it. |
12 |
|
13 |
|
14 |
A quick grep of /usr/portage shows that many builds will use |
15 |
replace-flags to replace one -march setting with another or with |
16 |
-mtune/-mcpu. |
17 |
|
18 |
There are a few that can filter -march altogether: |
19 |
|
20 |
1. net-firewall/ipsec-tools: filters all "-march=c3" |
21 |
2. media-libs/libvorbis: filters "-march=pentium?". |
22 |
3. dev-lang/squeak: filters all -march, -mtune, -mcpu, and many many |
23 |
other flags. |
24 |
4. dev-libs/libffi: filters -march=k8, -march=athlon64, and march=opteron. |
25 |
5. app-editors/emacs: filters all -march (only in 21.4-r1) |
26 |
6. app-pda/iripdb: filters march=k8, -march=athlon64, and -march=opteron |
27 |
7. media-tv/mythtv: filters all -march, -mtune, and -mcpu. |
28 |
|
29 |
So for #1, 2, 4, 5, and 6, you are correct that mtune/mcpu could have |
30 |
some benefit. However, IMO these are actually bugs...the ebuild should |
31 |
be using replace-flags to set the appropriate -mtune option if it cannot |
32 |
support -march. |
33 |
|
34 |
For #3 & 7, you get no benefit of any march/mtune/mcpu settings. |
35 |
|
36 |
Cheers, |
37 |
-Richard |
38 |
|
39 |
-- |
40 |
gentoo-user@g.o mailing list |