Re: Safe parms for i386 architecture

["Tim Prince" <tprince at computer dot org>]

At the risk of pouring fuel on a religious fire which I would be better
off not starting, I'll offer the following opinions:

The compiler flags which were not used in the testsuites may not have
had adequate testing.

 -O3 (including inline-functions) has been tested nearly as thoroughly
as -O2, OTOH it does not bring nearly as much benefit on 386 as on other
architectures supported by gcc.

-ffast-math has been satisfactory in my experience with 2.95.2.  It was
unsatisfactory on 2.96 for months, showing that not everyone was testing
it.

Someone else would be more of an authority, but I believe the problem
which led to frequent specification of -fno-strength-reduce was solved
before 2.95.2.  I never saw any need for it.

Tim Prince
----- Original Message -----
From: "Jean Francois Martinez" <jfm2@club-internet.fr>
To: <gcc@gcc.gnu.org>
Sent: Thursday, July 13, 2000 11:24 AM
Subject: Safe parms for i386 architecture


>
> Is there an URL documenting what compiler flags are safe in the 386
> architecture for gcc 2.95.2?
>
> Doc does not go into particulars about architecture.
>
> I am particularly interested into fast-math, inline-functions and if
it
> is still necessary to use no-strengthreduce since I have seen these
> flags used in Linux distributions.
>
>                                                     JF Martinez
>
>
>
>