------- Comment #6 from ebotcazou at gcc dot gnu dot org 2007-02-16 13:12 ------- > Any hints on what it might be and why it's reproducible only on SPARC and GCC > 4.x ?
If the trigger happens to be -fstrict-aliasing, it's very likely a violation of the type-based aliasing rules of the C/C++ languages. They are usually exposed by the scheduler, which is more aggressive on SPARC than on x86 for example. It's another (and more convoluted) story for -fwrapv. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30819
