Changeset 9883:e89a7cf22da3
---------------------------
ext: Include Bochs's SoftFloat implementation

This changeset includes John Hauser's SoftFloat from Bochs's SVN
revision 11804 (2013-09-05). The main difference between Bochs's
version of SoftFloat moves the statical variables controlling floating
point rounding and exceptions to a separate struct that is passed
explicitly to all functions that need it. This version has been
modified to use live in a separate name space (SoftFloat). Internal
Bochs types and macros have been replaced with standard types as
follows:

s/BX_CPP_INLINE/inline/g
s/BX_CONST64/UINT64_C/g

s/Bit64u/uint64_t/g
s/Bit32u/uint32_t/g
s/Bit16u/uint16_t/g
s/Bit8u/uint8_t/g

s/Bit64s/int64_t/g
s/Bit32s/int32_t/g
s/Bit16s/int16_t/g
s/Bit8s/int8_t/g

Additionally, BX_BIG_ENDIAN has been replaced with a test using the
system's endianness macros.

Since some versions of the standard libraries require the define
__STD_CONSTANT_MACROS to enable the UINTxx_C and INTxx_C macros when
compiling C++ code, this changeset also adds that define to the
preprocessor flags.

Various float conversion tests using 80-bit loads/stores and KVM. Seems to work fine.