New ACATS failures at -O2 (again)

[Laurent GUERBY]

Between 20040415 and 17, four new ACATS tests started failing at -O2 on
x86-linux: c41103b cc51d01 cc51d02 cc70003. There were no gcc/ada
changes.

There are now 13 ACATS tests failing at -O2, all instances of
silent wrong code generation: 7 of those are related to
Long_Long_Float handling, see ada/14988, not really
worrysome, but the 6 recent new failures (regressions vs 3.4) 
are more likely to be used code.

I strongly believe ACATS should be run at -O2 on HEAD.

Laurent