This is the mail archive of the
gcc-bugs@gcc.gnu.org
mailing list for the GCC project.
[Bug target/17990] [3.4/4.0 Regression] unaligned xmm movaps on the stack with -O2 -msse because of the frame pointer
- From: "rth at gcc dot gnu dot org" <gcc-bugzilla at gcc dot gnu dot org>
- To: gcc-bugs at gcc dot gnu dot org
- Date: 13 Dec 2004 18:20:09 -0000
- Subject: [Bug target/17990] [3.4/4.0 Regression] unaligned xmm movaps on the stack with -O2 -msse because of the frame pointer
- References: <20041014032810.17990.shadow@serverart.org>
- Reply-to: gcc-bugzilla at gcc dot gnu dot org
------- Additional Comments From rth at gcc dot gnu dot org 2004-12-13 18:20 -------
I have been completely unable to reproduce an unaligned stack, with either 3.4
or 4.0. Can those gcc folk that claim to have reproduced it in the past still
do so? If so, what are you trying?
I can say that the proximate cause of the existance of the SSE instruction in
the first place is a test in negsf2 using TARGET_SSE that ought to be using
TARGET_SSE_MATH. I can say that the reason reload fails to rematerialize the
vector constant (and spills it on the stack) is due to the movss split firing
before reload and removing the obviousness of the vector constant.
For 3.4 I can make two minor tweaks that will prevent the manifestation of this
bug, but leave -mfpmath=sse less than correct. For 4.0 I will attempt to fix
that as well.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17990