------- Comment #3 from amacleod at redhat dot com 2007-04-21 22:23 -------
I think this is actually fallout from the removal of V_MUSTDEFs, which predated
mem-ssa going in. Although it may have been checked into mainline at the same
time as mem-ssa now that I think about it. The work was originally separate.
My guess would be that FRE was not updated at that time to deal properly with
the absence of the V_MUSTDEF. IIRC, MUSTDEFS were considered to NOT have any
virtual operand USES, so this might now require a tuning of the
ZERO_SSA_OPERANDS conditional to check the RHS if there is one and only one
virtual def on the stmt to see whether its a relevant stmt or not.
Since this is a missed optimization instead of a segfault or wrong code, it was
probably easy to not notice it.
--
amacleod at redhat dot com changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |aldyh at gcc dot gnu dot
| |org, amacleod at redhat dot
| |com
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31651
