[Bug tree-optimization/81611] [8 Regression] gcc un-learned loop / post-increment optimization

[aoliva at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81611

--- Comment #8 from Alexandre Oliva <aoliva at gcc dot gnu.org> ---
Created attachment 42970
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=42970&action=edit
alternative (?) complementary candidate patch

This addresses the concern of post-increment in non-loops.  It solves the
problem along with the initial partial patch, but I'm still undecided as to the
second.  I already know it is broken (there's a thinko in the case that handles
all uses in a single stmt), and once the postinc auto_inc addressing modes are
handled by this third patch, I'm not sure yet whether the SSA conflict
reduction the second one affords are worth the trouble any more.  I'll try to
actually measure the effects, once I verify that it (or rather a fixed version
thereof) works.