Igor Schein on Thu, 12 Dec 2002 12:08:30 -0500

[Date Prev] [Date Next] [Thread Prev] [Thread Next] [Date Index] [Thread Index]

Re: Preparing bugfix release 2.1.5

On Thu, Dec 12, 2002 at 08:56:06AM -0800, Roland Dreier wrote:
> >>>>> "Henri" == Henri Cohen <Henri.Cohen@math.u-bordeaux.fr> writes:
>     Henri> Dear Bill, Please change -malign... into -falign... (at
>     Henri> least on Linux), otherwise gcc >= 3 complains.
> This is actually a little more difficult to resolve.  gcc 2.95 does
> not accept -falign...
> I really don't know whether there is a set of flags that works with
> both gcc 2.95 and 3.
>  - R.

Well, you can always if/else.  The point is, do -malign/-falign flags
really improve performance?  About 4 years ago I ran some comparison
testing with different flags.  I think it has to be done again with
gcc 3.x to determine the set of optimizations resulting in the best
performance.  For a example, an attractive flag -funroll-all-loops.
As of 4 years ago it wasn't effecting the performance, but maybe that
changed in gcc-3.x?  Also, default optimization for gcc is i386.  So
people with Athlons and especially P4's should be able to specify
-march=athlon-xp or -march=pentium4, and not just by editing dft file
manually, but with some configure switch.  etc., the list can go on.