[MPlayer-dev-eng] ppc runtime-cpu-detection fails with gcc 3.3

[Alan Curry]

Guillaume POIRIER writes the following:
>
>The idea is good in principle, but: What about Apple's GCC? IIRC, OSX
>10.3 has gcc-3.3, and people with that version of OSX are mostly stuck
>with this version unless they compile another version themselves or
>upgrade to 10.4.
>
>It would be nice to check if Apple's GCC-3.3 suffers from the same bug
>or not, and possibly only blacklist FSF's GCC (which mostly means that
>the OS is not darwin) or both if they both suck.

Here's a simple test; interested parties report back with results and
configure may be improved.

  echo 'void foo(void) { setjmp(); }' > gccbug.c
  gcc -maltivec -mabi=altivec -S -o - gccbug.c | grep stv

With a good gcc, you'll get no output. With the 3.3 bug, you'll get something
like this:

        stvx 20,1,0
        stvx 21,1,0
        stvx 22,1,0
        stvx 23,1,0
        stvx 24,1,0
        stvx 25,1,0
        stvx 26,1,0
        stvx 27,1,0
        stvx 28,1,0
        stvx 29,1,0
        stvx 30,1,0
        stvx 31,1,0

I already know 3.3.5 is bad. I'm almost certain that all 3.3.x's are bad, and
everything >=3.4.0 is good. At least some versions <=3.2.x are good,
according to the bugzilla reports on gcc.gnu.org, but I don't know exactly
which ones. The Apple fork is a complete unknown...