javaop.h (jfloat, jdouble): Make them structures mirroring the bit fields of IEEE float and double respectively.

	* javaop.h (jfloat, jdouble): Make them structures mirroring
	the bit fields of IEEE float and double respectively.
	(JFLOAT_FINITE, JFLOAT_QNAN_MASK, JFLOAT_EXP_BIAS,
	JDOUBLE_FINITE, JDOUBLE_QNAN_MASK, JDOUBLE_EXP_BIAS): New.
	(union Word, union DWord): Delete.
	(WORD_TO_FLOAT, WORDS_TO_DOUBLE): Update to match.

	* gjavah.c (java_float_finite, java_double_finite, F_NAN_MASK,
	D_NAN_MASK): Delete.
	(jni_print_float, jni_print_double): New.  Generate
	hexadecimal floating constants.
	(print_field_info): Use jni_print_float/double.

	* jcf-dump.c: Include math.h.  Use ldexp/frexp to assemble
	finite floating point numbers for output; special case
	non-finite floats.

From-SVN: r64671