[Bug tree-optimization/56064] New: Optimize VIEW_CONVERT_EXPR with FIXED_CST

[gjl at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56064

             Bug #: 56064
           Summary: Optimize VIEW_CONVERT_EXPR with FIXED_CST
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: gjl@gcc.gnu.org


Created attachment 29235
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=29235
foo.c.165r.expand (RTL dump)

In the followinf C code


static __inline__ __attribute__((__always_inline__))
_Fract rbits (const __INT16_TYPE__ i)
{
    _Fract f;
    __builtin_memcpy (&f, &i, sizeof (_Fract));
    return f;
}

_Fract func (void)
{
    return rbits (__INT16_C(0x1234));
}


The memcpy could be optimized away.  It is represented as VIEW_CONVERT_EXPR but
FIXED_CST is not folded in fold-const.c.

The attachment shows the memory moves of HQmode which is suboptimal.  For more
details, see http://gcc.gnu.org/ml/gcc/2013-01/msg00187.html