[Bug target/87079] nios2 optimization for size - case of regression relatively to 5.3.0

[sandra at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87079

sandra at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|                            |2018-11-02
                 CC|                            |sandra at gcc dot gnu.org
           Assignee|unassigned at gcc dot gnu.org      |sandra at gcc dot gnu.org
     Ever confirmed|0                           |1

--- Comment #1 from sandra at gcc dot gnu.org ---
This looks like a problem with the nios2 instruction costs.  Things are going
wrong in expand_widening_mult (in expmed.c); compiling with -O2 takes the
expand_binop path at the end to use the provided umulsidi3 expander, but with
-Os it thinks the expand_mult_const path is better.  I'll try to poke at this
some more and come up with a patch.