Implement TARGET_ATOMIC_ASSIGN_EXPAND_FENV for powerpc*-*-linux* soft-float and e500.

Implement TARGET_ATOMIC_ASSIGN_EXPAND_FENV for powerpc*-*-linux* soft-float and e500.

This patch implements support for TARGET_ATOMIC_ASSIGN_EXPAND_FENV for
powerpc*-*-linux* soft-float and e500, provided GCC is configured for
glibc 2.19 or later on the target.

New functions __atomic_feholdexcept, __atomic_feclearexcept and
__atomic_feupdateenv were added (to libc) in that glibc version (for
powerpc soft-float / e500 only) in order to support this part of C11.
For soft-float, libc functions are needed because the floating-point
exception state is in TLS variables in libc that aren't directly
accessible outside of glibc.  For e500, they are also needed because
of the prctl syscalls involved in controlling trapping for exceptions
and informing the kernel when certain exception flags have been
cleared.  The actual implementation in GCC is a straightforward matter
of calling those functions.

Tested with no regressions for cross to powerpc-linux-gnu
(soft-float); the c11-atomic-exec-5.c results go from FAIL to PASS.

	* configure.ac (TARGET_GLIBC_MAJOR, TARGET_GLIBC_MINOR): Define
	macros.
	* configure, config.h.in: Regenerate.
	* config/rs6000/linux.h [TARGET_GLIBC_MAJOR > 2 ||
	(TARGET_GLIBC_MAJOR == 2 && TARGET_GLIBC_MINOR >= 19)]
	(RS6000_GLIBC_ATOMIC_FENV): New macro.
	* config/rs6000/linux64.h [TARGET_GLIBC_MAJOR > 2 ||
	(TARGET_GLIBC_MAJOR == 2 && TARGET_GLIBC_MINOR >= 19)]
	(RS6000_GLIBC_ATOMIC_FENV): New macro.
	* config/rs6000/rs6000.c (atomic_hold_decl, atomic_clear_decl)
	(atomic_update_decl): New static variables.
	(rs6000_atomic_assign_expand_fenv) [RS6000_GLIBC_ATOMIC_FENV]:
	Generate calls to __atomic_feholdexcept, __atomic_feclearexcept
	and __atomic_feupdateenv for soft-float and no-FPRs.

From-SVN: r217040