Different behaviour when compiling with or without the optimizer

[Martin v. Loewis]

> I have then trimmed down the program as much as I can with the same
> problem still there and attaches that to this email.

Thanks for your bug report. While I could reproduce it with gcc
2.95.1, it seems to work with 2.95.2. I believe this was a well-known
bug in the 386 back-end, which was somehow mishandling the floating
point registers.

Regards,
Martin