[Bug tree-optimization/61299] [4.9/4.10 Regression] Performance regression for the SIMD rotate operation with GCC vector extensions

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61299

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2014-05-24
                 CC|                            |glisse at gcc dot gnu.org,
                   |                            |jakub at gcc dot gnu.org
   Target Milestone|---                         |4.9.1
            Summary|[4.9 Regression]            |[4.9/4.10 Regression]
                   |Performance regression for  |Performance regression for
                   |the SIMD rotate operation   |the SIMD rotate operation
                   |with GCC vector extensions  |with GCC vector extensions
     Ever confirmed|0                           |1

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Started with r198772.  We now recognize it as vector rotate by fold-const
(though, at gimple level r198769 still doesn't handle vector types, maybe it
should), but if there is no corresponding vector compare, guess
tree-vect-generic.c should transform it back to shifts.