[Bug target/64479] [4.8 Regression][SH] wrong optimization delayed-branch

[olegendo at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64479

Oleg Endo <olegendo at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|sh3                         |sh*-*-*
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2015-01-06
                 CC|                            |kkojima at gcc dot gnu.org,
                   |                            |olegendo at gcc dot gnu.org
            Summary|wrong optimization          |[4.8 Regression][SH] wrong
                   |delayed-branch  for SH      |optimization delayed-branch
     Ever confirmed|0                           |1

--- Comment #1 from Oleg Endo <olegendo at gcc dot gnu.org> ---
Thanks for pinpointing and reporting this.

trunk (GCC 5): OK
4.9 branch: OK
4.8 branch: NG

This is a problem of the flaky cbranch optimization in 4.8 in the *cbranch_t
splitter code. 

For some reason, modified_between_p seems to ignore the fact that the T_REG is
clobbered in call insns.  Thus it thinks that it's OK to extend the T_REG
lifetime across a call insn.