| 1 |
On Tue, Dec 9, 2008 at 1:23 PM, Sami Näätänen <sn.ml@××××××××××××.fi> wrote: |
| 2 |
> So hi from a amd64 newbie. Not so newbie with Gentoo though. :) |
| 3 |
> |
| 4 |
> My system is an Intel quad core core2 with a 2.4 GHz clock speed coupled with |
| 5 |
> a 4GB of memory. No overclocking etc. Want this to be stable. :) |
| 6 |
> |
| 7 |
> I'm just curious what people use as their stable CFLAGS in amd64 Gentoo? |
| 8 |
> (Sorry if this has been up lately, but I just switched to 64bit env so...) |
| 9 |
> |
| 10 |
> |
| 11 |
> Here is mine and some explanation of why (And I use ~arch system with gcc 4.3) |
| 12 |
> |
| 13 |
> The flags are in order they are used in my CFLAGS and CXXFLAGS. |
| 14 |
> |
| 15 |
> Gives stable base |
| 16 |
> -O2 |
| 17 |
> |
| 18 |
> Want to optimize for my system, but don't want "native" |
| 19 |
> -march=core2 |
| 20 |
> |
| 21 |
> If some ebuilds filter march this will still cache optimize etc for my system |
| 22 |
> -mtune=core2 |
| 23 |
> |
| 24 |
> Faster floating point math and better chance of vectorization |
| 25 |
> -mfpmath=sse |
| 26 |
> |
| 27 |
> These because of the march might get filtered |
| 28 |
> -mmmx -msse -msse2 -msse3 -mssse3 |
| 29 |
> |
| 30 |
> For loop vectorization |
| 31 |
> -ftree-vectorize |
| 32 |
> |
| 33 |
> Just to get some Idea how much vectorized loops there will be. |
| 34 |
> By the way I surprised the amount of "LOOP VECTORIZED" notes in the compile |
| 35 |
> output. And only have seen couple of two versions |
| 36 |
> -ftree-vectorizer-verbose=1 |
| 37 |
> |
| 38 |
> Of course I don't want temp files :) |
| 39 |
> -pipe |
| 40 |
> |
| 41 |
> |
| 42 |
> I don't use any loop unrolling etc, because it would only add the code size. |
| 43 |
> I'm not so brave that I would dare to use -Os. |
| 44 |
> |
| 45 |
> So what's your experiences and reasoning behind what you do? |
| 46 |
> Any benchmarks or so? |
| 47 |
> |
| 48 |
> |
| 49 |
> PS. If you see same post without this added postscript. Just ignore it, it's |
| 50 |
> the same post, but I forgot to change my default identity for this ML. |
| 51 |
> |
| 52 |
|
| 53 |
Dear Sami, |
| 54 |
|
| 55 |
I have a Q9300 and used this: |
| 56 |
|
| 57 |
CFLAGS="-march=nocona -O2 -pipe" |
| 58 |
CXXFLAGS="${CFLAGS}" |
| 59 |
USE="mmx sse sse2 <snip>" |
| 60 |
|
| 61 |
Stable gcc version is 4.2.x. I switched to 4.3.2 by adding |
| 62 |
|
| 63 |
>=sys-devel/gcc-4.3.2 |
| 64 |
>=sys-libs/glibc-2.7-r2 |
| 65 |
|
| 66 |
to /etc/portage/package.keywords. With 4.3.2 I use: |
| 67 |
|
| 68 |
CFLAGS="-march=native -O2 -pipe" |
| 69 |
|
| 70 |
With only a small effort, you get most of the benefits. So fine-tuning |
| 71 |
to the edge will give you issues to solve with only a very small |
| 72 |
percentage of performance increase in return. |
| 73 |
|
| 74 |
My 2 cents.. |
| 75 |
|
| 76 |
Martin |
Archives