[Bug tree-optimization/81556] [7/8 Regression] Wrong code at -O2

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81556

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 41832
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=41832&action=edit
gcc8-pr81556.patch

Untested fix.  The current code whether a SSA_NAME can be reused or not is
based on whether all the rhs2 ops are the same from the toplevel stmt (and
whether there is powi/negate).  That should work fine as long as the number of
ops is the same after optimization, but in this case there is a removal of
reduncant op, at which point that doesn't work anymore.