[Bug middle-end/53049] expand/TER unappropriate moving unspec volatile

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

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53049

--- Comment #4 from Georg-Johann Lay <gjl at gcc dot gnu.org> 2012-04-19 19:31:12 UTC ---

In tree-ssa-ter.c:is_replaceable_p() there is:

  /* Leave any stmt with volatile operands alone as well.  */
  if (gimple_has_volatile_ops (stmt))
    return false;


and in tree-ssa-ter.c:find_replaceable_in_bb():

      /* Mark expression as replaceable unless stmt is volatile, or the
     def variable has the same root variable as something in the
     substitution list, or the def and use span a call such that
     we'll expand lifetimes across a call.  */
      if (gimple_has_volatile_ops (stmt) || same_root_var

So it leaves volatiles alone -- at least that's what the comments indicate and
why there are tests for has_volatile at all.

If you drag A over B, you can just as well say that B is dragged over A.
This means that no volatile must be crossed as a statement is moved around.
Otherwise, a programm will never be able to state "There is something going on
that is important, but cannot be expressed".

C has no notion of execution time at all, but everyone knows that execution
time in /any/ program is crucial...

Is there a way in the backend to interfere with that malicous transforms?

A builtin is treated as CALL at this level, and calls are skipped to. So it's
all in a mess...