This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Need help tracking down and fixing PR44364 (e500 FP register save bug)
- From: "Moffett, Kyle D" <Kyle dot D dot Moffett at boeing dot com>
- To: "gcc at gcc dot gnu dot org" <gcc at gcc dot gnu dot org>
- Cc: Sebastian Andrzej Siewior <sebastian at breakpoint dot cc>, "Schmidt, Nicholas H" <Nicholas dot H dot Schmidt at boeing dot com>, "markmarkmark dot workman at gmail dot com" <markmarkmark dot workman at gmail dot com>
- Date: Sat, 5 Jun 2010 10:01:13 -0500
- Subject: Need help tracking down and fixing PR44364 (e500 FP register save bug)
Hello,
Several of us are trying to put together a "PowerPCSPE" Debian port (powerpc-linux-gnuspe, based around the e500v2), and we're experiencing a pretty bad FP register-save bug (PR44364 [0]). Specifically, it seems like GCC will occasionally only save half of the 64-bit double register on the stack, allowing the other half to be corrupted by the called function.
Unfortunately neither of us is experienced enough with GCC's internals to be able to do much more than find a decent test-case; I'm hoping that maybe I can get a little more visibility on the bug and maybe find the right person to help get it fixed. Heck... there's 3 of us here who will chip in for a case of your favorite beverage if it's a quick fix. :-D
Cheers,
Kyle Moffett
PS: Please keep the CC list
[0] http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44364