Log Message:
Update to gmp 4.1.2.  Many changes, but the relevant bug fixes seems to be:

	* mpn/generic/rootrem.c: In Newton loop, pad qp with leading zero.

	* mpn/generic/rootrem.c: Allocate 1.585 (log2(3)) times more space
        for pp temporary to allow for worst case overestimate of root.
	Add some asserts.

	* tests/mpz/t-root.c: Generalize and speed up.

        * mpfr/pow.c: Fixed bug (infinite loop) for exact powers.

	* mpfr/sub.c: Fixed wrong inexact flag for a - b where a and b are of
        different signs and EXP(a) < EXP(b).

	* printf/printffuns.c (gmp_fprintf_reps): Make it actually work
	for padding > 256.

        * gmp-impl.h (USE_LEADING_REGPARM): Disable for PIC code generation.

	* mpn/generic/tdiv_qr.c: Remove a bogus assert.

	* mpn/generic/mode1o.c, mpn/alpha/ev5/mode1o.c: Correction to ASSERTs.
	Reported by Christoph Ludwig.

	* mpf/urandomb.c: Truncate nbits to precision of rop.

	* configure.in (sparc v9 *bsd*): Add NetBSD and OpenBSD sparc64.
	(x86 openbsd*): Extra underscore for _GLOBAL_OFFSET_TABLE_.

	* mpn/generic/rootrem.c: Avoid overflow in xnb calculation.
	* mpz/root.c: Avoid overflow in rootnb calculation.

	* gmpxx.h (__gmp_binary_equal, __gmp_binary_not_equal): Fix broken
	mpq/double functions.

	* mpn/m68k/lshift.asm: Fix typo in !scale_available_p code.
        * mpn/m68k/rshift.asm: Likewise.

	* mpz/remove.c: Make src==0 return 0, not do DIVIDE_BY_ZERO.

	* mpfr/acinclude.m4 (MPFR_CONFIGS): Patch by Vincent for an apparent
	float rounding gremlin on powerpc.

	* mpf/inp_str.c: Fix returned count of chars read, reported by Paul
	Zimmermann.  Also fix a memory leak for invalid input.

	* mpfr/set_q.c: Allow for 1 bit numerator or denominator.