[Bug rtl-optimization/83565] RTL combine pass breaks shift result (at least on ia64)

[wilson at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83565

Jim Wilson <wilson at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2017-12-29
     Ever confirmed|0                           |1

--- Comment #23 from Jim Wilson <wilson at gcc dot gnu.org> ---
There should be no correctness issue with turning off WORD_REGISTER_OPERATIONS.
 The only issue should be performance.  If we lose 5+% performance, then we
should try to get the broken nonzero_bits1 patch reverted.  If performance
stays about the same, then there is probably no point in fighting this and we
should just turn off WORD_REGISTER_OPERATIONS.  None of the 3 most important
targets set it, so maybe it doesn't work correctly anymore, except maybe on
targets like sparc that have shift instructions that read subwords but write
the entire word.