[Bug rtl-optimization/59535] [4.9 regression] -Os code size regressions for Thumb1/Thumb2 with LRA

[vmakarov at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59535

Vladimir Makarov <vmakarov at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |vmakarov at gcc dot gnu.org

--- Comment #9 from Vladimir Makarov <vmakarov at gcc dot gnu.org> ---
(In reply to Richard Earnshaw from comment #5)
> Number of register-register move operations in the testcase
> lra:    208
> no-lra: 105

That is huge degradation.  And I guess it is not only because of 2-ops insns. 
It works adequate on x86/x86-64.  Although I know some issues with 2-ops
commutative insns (I am trying to fix it too).

I think major problem is in wrong alternative choices as thumb lo/hi reg usage
is complicated.

I take this bug very seriously.  If I cannot fix it till end of Jan (sorry, it
is a vacation time), probably we should switch to reload pass for thumb.