I have some c++ code that produces incorrect results when compiled with -O2 or -O3 (but not -O). It is floating-point math. Have cases like this been reported before? This is egcs-1.0.2 i686-pc-linux-gnu glibc-2.0.6.