[Bug target/53272] wrong condition-codes for strict-low-part destination and small-integer source

["hp at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53272

Hans-Peter Nilsson <hp at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-05-08
         AssignedTo|unassigned at gcc dot       |hp at gcc dot gnu.org
                   |gnu.org                     |
     Ever Confirmed|0                           |1

--- Comment #1 from Hans-Peter Nilsson <hp at gcc dot gnu.org> 2012-05-08 00:11:03 UTC ---
Created attachment 27337
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=27337
testcase to be compiled with -O2

The effect of the bug is to eliminate the NULL-pointer check for rtc->ops->f
and always call it, which is not a good idea when it is NULL.  Test-case
derived from a Linux driver.

Unfortunately, presence of a main function makes the bug hide by slightly
changing the generated code, so I have to put the framework in a separate file
to make a runnable test-case. This is a secondary bug, for which I will clone a
PR from this PR: with the marker-asms and attributes noclone, noinline, there
should be no need for that.  It seems static branch predictions change.