[Bug tree-optimization/87609] [7/8/9 Regression] miscompilation with restrict and loop

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87609

--- Comment #8 from Richard Biener <rguenth at gcc dot gnu.org> ---
Created attachment 45788
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=45788&action=edit
second part

This is the second part (was already posted for review in October), I've
removed
the extra checking (which fires...) and added an assert.

First one is separately in testing plus both ones together.

Note that with all the copying overflowing of clique just became more likely
so I'll eventually abstract away the ++fun->last_clique and overflow to
zero (no dependence info).  But of course 32000 cliques should be enough
forever!