[Bug tree-optimization/65930] Reduction with sign-change not handled

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65930

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #42730|0                           |1
        is obsolete|                            |

--- Comment #17 from Richard Biener <rguenth at gcc dot gnu.org> ---
Created attachment 46875
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=46875&action=edit
prototype

So here's a patch that can vectorize the testcase.

Preparatory cleanup would be moving almost all validity checks from
force_simple_reduction to vectorizable_reduction plus storing the
core reduction operation (to be used in the epilogue then) from the
SCC detection somewhere, not relying on the stmt_info code of the
stmt feeding the PHI backedge in vectorizable_reduction (in fact on
the forceslp branch that stmt gets vectorized as if being an internal_def,
the epilogue is created because its def is "live").