| 1 |
Duncan wrote: |
| 2 |
> |
| 3 |
> |
| 4 |
> Well, you say you want stable, but then say you use ~arch, so I see |
| 5 |
> you're not too stick in the mud. =:^) |
| 6 |
> |
| 7 |
> Here's mine, for a dual Opteron 290: |
| 8 |
> |
| 9 |
> CFLAGS="-march=opteron-sse3 -pipe -O2 -frename-registers -fweb -fmerge- |
| 10 |
> all-constants -fgcse-sm -fgcse-las -fgcse-after-reload -ftree-vectorize - |
| 11 |
> fdirectives-only -freorder-blocks-and-partition -combine" |
| 12 |
> |
| 13 |
> CXXFLAGS="-march=opteron-sse3 -pipe -O2 -frename-registers -fweb -fmerge- |
| 14 |
> all-constants -fgcse-sm -fgcse-las -fgcse-after-reload -ftree-vectorize - |
| 15 |
> fdirectives-only" |
| 16 |
> |
| 17 |
> You can look them up in the gcc manpage, or look back a year or so when I |
| 18 |
> explained most of them, altho that was a couple gcc versions ago and they |
| 19 |
> weren't quite the same. |
| 20 |
> |
| 21 |
> |
| 22 |
<SNIP> |
| 23 |
|
| 24 |
|
| 25 |
Been there, done praactically that, but it didn't make one quark of |
| 26 |
difference overall, except throwing gcc in a coma now and then, |
| 27 |
lenghtening compile problems and causing odd ( but rare ) bugs. |
| 28 |
|
| 29 |
I tried to time several C programs of mine and found that plain -O1 |
| 30 |
worked substantially better than plain -O2. |
| 31 |
|
| 32 |
After that, I said sod all and used plain vanilla CLFAGS on new gcc and |
| 33 |
with right march. Works fine, with same speed, faster compiles and much |
| 34 |
less headaches on average. |
| 35 |
|
| 36 |
In my experience, exotic CFLAGS can make a difference, but this varies |
| 37 |
wildldy from program part to program part, so unless one knows exactly |
| 38 |
what he is doing, he might be better of trusting compiler to use sane |
| 39 |
path with -O2. Besides that, portage doesn't have an option to compile |
| 40 |
just some part of the code with another, non_default CFLAGS... |
| 41 |
|
| 42 |
|
| 43 |
|
| 44 |
> But my basic strategy is this: Because memory is so much slower than |
| 45 |
> cache on a modern processor, in general it should pay to optimize for |
| 46 |
> size even if it costs a few CPU cycles once in awhile. |
| 47 |
True, but he is asking for P4, which was notorious for having long |
| 48 |
pipelina and a neadache after cache miss, so for him -O2 or even -03 |
| 49 |
might be better in _some_ cases. |
| 50 |
But even so, IMVHO it is simply not worth the time and effort to fiddle |
| 51 |
with this, I'd use golden default with right march here also and be |
| 52 |
done with it. |
Archives