| 1 |
On Tue, May 29, 2012 at 11:09 PM, Alan McKinnon <alan.mckinnon@×××××.com> wrote: |
| 2 |
> On Tue, 29 May 2012 22:56:07 +0200 |
| 3 |
> Rafa Griman <rafagriman@×××××.com> wrote: |
| 4 |
> |
| 5 |
>> Hi Vaeth :) |
| 6 |
>> |
| 7 |
>> On Tue, May 29, 2012 at 7:02 PM, Vaeth |
| 8 |
>> <vaeth@××××××××××××××××××××××××.de> wrote: |
| 9 |
>> > On Tue, 29 May 2012, Rafa Griman wrote: |
| 10 |
>> >> |
| 11 |
>> >> gawk: cmd. line:3: error: Unmatched [ or [^: /[^[:space:]]/ |
| 12 |
>> > |
| 13 |
>> > |
| 14 |
>> > Your gawk is broken. This happens if you emerged gawk with |
| 15 |
>> > current gcc and aggressive FLAGS like -DNDEBUG or -flto. |
| 16 |
>> > Not sure whether it is a bug of gawk or gcc. |
| 17 |
>> |
| 18 |
>> |
| 19 |
>> So it seems to be the CFLAGS :( As I said in my previous e-mail, I was |
| 20 |
>> "experimenting" a bit with them ... I'll be more conservative ;) |
| 21 |
> |
| 22 |
> Well, on the one hand the flags don;t seem to be too bad overall, but |
| 23 |
> on the other hand there's the golden rule of Gentoo: |
| 24 |
> |
| 25 |
> Don't stuff around with CFLAGS |
| 26 |
> |
| 27 |
> Why not? Well, there's the ricer phenomenon where changed CFLAGS |
| 28 |
> worsen performance and stability but the user's own bias convinces |
| 29 |
> him/her that it's actually vastly improved. |
| 30 |
|
| 31 |
|
| 32 |
I know that rule and tried not to be too agressive but seems I was O:) |
| 33 |
Since it's not my min machine, I was going to experiment (nothing |
| 34 |
"scientific", just curiosity) trying different options on different |
| 35 |
partitions and see how they work out. |
| 36 |
|
| 37 |
|
| 38 |
> Some tweaks are perfectly OK, like the multimedia and cpu-specific |
| 39 |
> stuff (sse, mmx and all their related cousins). These are normally safe |
| 40 |
> and can give huge gains with video playback. Likewise for the GPU tweaks |
| 41 |
> once those hit mainline usage. |
| 42 |
|
| 43 |
|
| 44 |
That was mainly the reason: video playback and some other apps |
| 45 |
(scientific). But nothing serious. |
| 46 |
|
| 47 |
|
| 48 |
> One last tweak, if the code is built on the machine that will run it |
| 49 |
> (the usual case, replace -march and -mtune with just "-march=native". |
| 50 |
> It has exactly the same effect but is easier to read, is |
| 51 |
> self-documenting and allows the compiler to attempt it's best (the |
| 52 |
> compiler usually does know much better what to do than you do) |
| 53 |
|
| 54 |
I totally agree with you: the compiler knos better ;) |
| 55 |
|
| 56 |
TIA |
| 57 |
|
| 58 |
Rafa |
Archives