Re: Regression mzero2.c -- alpha

[craig at jcb-sc dot com]

>On Thu, Feb 18, 1999 at 01:50:47PM -0700, Donn Terry wrote:
>> Anyway... the question: what's the policy w.r.t. this: ignore it,
>> set it as an XFAIL or find a way to have the option passed to the
>> compiler only for the Alpha.  (And probably a different option(?)
>> for hppa.)
>
>The $prefix/share/dejagnu/baseboard/unix.exp file on my Alphas contains
>
> set_board_info ieee_multilib_flags "-mieee";
>
>which makes the compiler DTRT for the test case.

To expand a bit on Richard's answer: the gcc compilers currently
follow the path set by Digital's own Alpha-based compilers, in that
they default to generating code assuming limited use of the IEEE
floating-point ranges and features.

I would just refer to the gcc documentation for the Alpha port's
-m flags, but I just noticed this little gem -- Richard or someone,
can you suggest a fix to this recursive definition?

    `su'
          Like `su', but the instructions are marked to be safe for
          software completion (see Alpha architecture manual for
          details).

As of egcs 1.2, it (supposedly) becomes more practical to actually
*use* `-mieee' with g77, because libg2c (g77's version of libf2c)
is built using the "multilib" facility, which should mean that
linking with `-mieee' actually selects a version of libg2c that,
itself, was compiled with `-mieee'.  This avoids crashes doing I/O
on "interesting" IEEE values (which probably would be better avoided
by libf2c code changing somehow).

I have yet to personally investigate how well the above works, as
my Alpha is awaiting a fresh install.  (BTW, now that Dell has
come, fixed my Pentium II's network card for "free" -- turns out
I had two-year in-home support -- installing the Alpha becomes
actually useful, as I can now talk to it.)

        tq vm, (burley)