https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84616
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Keywords| |needs-reduction, wrong-code
Target| |x86_64-*-*, i?86-*-*
Status|WAITING |NEW
CC| |rguenth at gcc dot gnu.org
Component|c++ |target
Version|unknown |7.3.1
--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
Hmm, I get (with local Eigen library) at -O0:
1 -0 0 -0
-0 1 -0 0
0 -0 1 -0
-0 0 -0 1
so it looks like Eigen has issues with signed zeros? I can reproduce your
bogus output already with -O -fno-signed-zeros with my eigen version.
Your preprocessed source also produces those signed zero issues seen above
but requires -funsafe-math-optimizations to reproduce the issue.
So, "confirmed", but given the complexity of Eigen I wouldn't rule out
issues with the library itself.
Interestingly with -m32 -Ofast -mno-sse I see correct results (apart from
the signed zeros).
Needs to be tracked down to sth simpler than Eigen...
