| 1 |
On Mon, Sep 20, 2010 at 7:48 PM, Duncan <1i5t5.duncan@×××.net> wrote: |
| 2 |
> Frank Peters posted on Mon, 20 Sep 2010 21:49:33 -0400 as excerpted: |
| 3 |
> |
| 4 |
>> After doing my periodical emerge update, I notice that gcc-4.5.1 is now |
| 5 |
>> unmasked and presumably stable. I also notice that a couple of new |
| 6 |
>> options are possible with this new release, namely Graphite and FLTO |
| 7 |
>> optimizations. (Actually, Graphite has been available for a while but I |
| 8 |
>> haven't been following gcc development too closely and only now do I |
| 9 |
>> become aware of it.) |
| 10 |
>> |
| 11 |
>> Does anyone have any comments on the use of Graphite and/or FLTO? Are |
| 12 |
>> they capable of significant improvements on a amd64 system? |
| 13 |
>> |
| 14 |
>> Supposedly FLTO will greatly increase file size and thus could be |
| 15 |
>> selectively enabled for only a few packages. Would building gcc itself |
| 16 |
>> with FLTO be advisable? |
| 17 |
>> |
| 18 |
>> There are reports that many packages fail to build with FLTO: |
| 19 |
>> http://blog.flameeyes.eu/2010/04/25/tinderbox-getting-friendly-with- |
| 20 |
> gcc-4-5 |
| 21 |
>> |
| 22 |
>> Since I have encountered little detailed discussion on this topic, I am |
| 23 |
>> posting my question to this list. |
| 24 |
> |
| 25 |
> Two excerpts from $equery u gcc |
| 26 |
> |
| 27 |
> - - graphite : Add support for the framework for loop optimizations |
| 28 |
> based on a polyhedral intermediate representation |
| 29 |
> |
| 30 |
> - - lto : Add support for link-time optimizations (unsupported, |
| 31 |
> use at your own risk). |
| 32 |
> |
| 33 |
> So at minimum, you'll need to enable the appropriate USE flags for those |
| 34 |
> features. (As you can see, I haven't.) And, link-time-optimization isn't |
| 35 |
> Gentoo supported at all -- probably partly as a result of flameeyes' |
| 36 |
> testing. |
| 37 |
> |
| 38 |
> Personally, I've been waiting for some general level documentation to come |
| 39 |
> across my radar before I get involved with either of these. I've been |
| 40 |
> running 4.5 for some time now, without issue (except that the new version |
| 41 |
> makes use of more instructions for certain later Intel chips, and I had a |
| 42 |
> problem as a result in my 32-bit chroot, trying to build for my atom based |
| 43 |
> netbook, until I dialed back the allowed hardware instructions to what was |
| 44 |
> available on my older amd opteron based build machine, gcc 4.4 apparently |
| 45 |
> didn't use the instructions at all, even if they were marked as |
| 46 |
> available), but that's partly because I've been a bit conservative in |
| 47 |
> enabling the newly available features such as graphite and link-time- |
| 48 |
> optimization. |
| 49 |
|
| 50 |
I gave 4.5.0 a go a few months back & had issues with mysql (5.1 & |
| 51 |
5.5) no longer compiling. This might be resolved by now tho. |
| 52 |
|
| 53 |
Wil |
Archives