Fortran regression on ARM (complex division)

[Richard Earnshaw]

The patch below fixes a regression on the ARM with software floating point 
that was causing an ICE in the fortran compiler in several tests in the 
testsuite.  The problem is that expand_unop does not synthesize abs if 
there is no instruction to do this explicitly: the solution is to call 
expand_abs instead.  Further, it doesn't make sense to take the absolute 
value of an unsigned quantity.  Err, what's an "unsigned" complex number?

Is this ok for the release branch as well?

Richard.

<date>  Richard Earnshaw <rearnsha@arm.com>

	* optabs.c (expand_cmplxdiv_wide): Use expand_abs to get the absolute
	values.