2005-08-19 Dave Brolley <brolley@redhat.com>

2005-08-19  Dave Brolley  <brolley@redhat.com>

        * cgen-fpu.cxx: New file.
        * cgen-fpu.h: New file.
        * Makefile.am (libcgencpu_la_SOURCES): Add cgen-fpu.cxx.
        * Makefile.in: Regenerated.
        * Contribute the following enhancements to floating point
        support:

        2005-06-14  Dave Brolley  <brolley@redhat.com>

        * fp.h (do_round,do_normal_round,do_normal_underflow,do_normal_overflow)
:
        Now virtual.
        * fp.cxx (do_normal_overflow): Round up to infinity only if
        the result is positive.
        (operator-): Don't apply 'neg' to a nan.
        (neg): Return src if it is a qnan.

        2005-05-20  Dave Brolley  <brolley@redhat.com>

        * fp.cxx (pack_any): Ensure that nan results have a positive sign.
        (operator-): Simply call operator+ with the rhs negated.
        * fp.h (is_negative): New method of fp.

        2005-05-13  Dave Brolley  <brolley@redhat.com>

        * fp.h (fp): All arithmetic operator methods are now members
        of class fp.
        * fp.cxx (fp): Likewise.

        2005-04-27  Dave Brolley  <brolley@redhat.com>

        * fp.cxx (fp::unpack): Only perform sanity check for non-nans.

        2005-04-12  Dave Brolley  <brolley@redhat.com>

        * fp.h: Move static constants and functions to top of file for
        visibility within the file.
        (get_status): New method of fp.
        (get_class): Likewise.
        (qnan_fraction): Likewise.
        (snan_fraction): Likewise.
        (qnan_exponent): Likewise.
        (snan_exponent): Likewise.
        * fp.cxx (fp_to_int): throw fp::overflow for overflow.
        (pack_any): Use [sq]nan_{exponent,fraction} to construct nans.
        (unpack): Initialize status to 'ok'. Set class and status using
        is_qnan_fraction and is_qnan respectively.

        2005-03-18  Dave Brolley  <brolley@redhat.com>

        * fp.cxx: Move static items from here.
        * fp.h: To here. Make some methods virtual. Make private members
        protected.
        (is_inexact): New method.