gcc 4.1.1 has a regression with -fdelayed-branch on hppa (tested on my b2000) and probably mips too. Since it's enabled by default on -O1+ it's kinda bad mojo, and it's known to break at least perl unpack() in some scenarios -> http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=374396
*** Bug 144885 has been marked as a duplicate of this bug. ***
well lemme know when it hits the 4.1 branch ;P
this will be in gcc-4.1.1-r2