From: Bruno Haible <bruno@cl...> - 2004-03-24 21:37:54
Will Newton wrote:
> I get a single test failing on Linux/hppa:
> Form: (ISQRT 77371252437321868671713407)
> CORRECT: 8796093021183
> CLISP : 8796093022207
> Has anyone got any debugging tips?
- Try to remove the C compiler optimizations.
- Try the newest GCC.
- Compile with debugging info and step through it in gdb.
Last time I saw this, was on HPPA as well, under HP-UX, and the bug
showed up in UDS_divide_. I.e. the division routine was probably miscompiled.
I wasn't able to point out precisely the wrong raw instruction, though.
(HPPA is such an awful RISC!)