This is the mail archive of the
gcc-bugs@gcc.gnu.org
mailing list for the GCC project.
[Bug bootstrap/30790] [4.3 regression] Bootstrap comparison failure i386:
- From: "rridge at csclub dot uwaterloo dot ca" <gcc-bugzilla at gcc dot gnu dot org>
- To: gcc-bugs at gcc dot gnu dot org
- Date: 17 Feb 2007 05:32:18 -0000
- Subject: [Bug bootstrap/30790] [4.3 regression] Bootstrap comparison failure i386:
- References: <bug-30790-231@http.gcc.gnu.org/bugzilla/>
- Reply-to: gcc-bugzilla at gcc dot gnu dot org
------- Comment #1 from rridge at csclub dot uwaterloo dot ca 2007-02-17 05:32 -------
This also happens on MinGW, and I've been able to reproduce on i386-linux.
It seems to be due to the changes made in r121821. In this change
a number of functions made to use set_unique_reg_note(), but this
function doesn't set REG_EQUIV notes on instructions with multiple set
operations. While compiling stage2 this causes set_even_probabilities()
to be miscompiled because it has a divide instruction, a multple set
insn on i386. Because this function is miscompiled, different cfg
probabilities are computed in stage3 than in stage2 and so the object
files don't compare as equal.
Reverting local-alloc.c to the version before r121821 allowed me to
bootstrap on i386-linux successfully.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30790