4.2.1. IEEE Floating Point

The floating point format is specified within the newlib directory in libc/include/machine/ieeefp.h. Details of how the IEEE 754 format is implemented, and variations from the standard, are specified by defining a number of C macros.

__IEEE_BIG_ENDIAN
Define this macro if the floating point format is big endian.
Caution
One, and only one of __IEEE_BIG_ENDIAN and __IEEE_LITTLE_ENDIAN must be defined.
__IEEE_LITTLE_ENDIAN
Define this macro if the floating point format is little endian.
Caution
One, and only one of __IEEE_LITTLE_ENDIAN and __IEEE_BIG_ENDIAN must be defined.
__IEEE_BYTES_LITTLE_ENDIAN
Define this macro in addition to __IEEE_BIG_ENDIAN, where the words of a multi-word IEEE floating point number are in big endian order, but the bytes within each word are in little endian order.
_DOUBLE_IS_32BITS
Define this if double precision floating point is represented using the 32-bit IEEE representation.
_FLOAT_ARG
Floating point arguments are usually promoted to double when passed as arguments. If this is not the case, then this macro should be defined to the type actually used to pass floating point arguments.
_FLT_LARGEST_EXPONENT_IS_NORMAL
Define this if the floating point format uses the largest exponent for finite numbers rather than NaN and infinities. Such a format cannot represent NaNs or infinities, but it's FLT_MAX is twice the standard IEEE value.
_FLT_NO_DENORMALS
Define this if the floating point format does not support IEEE denormalized numbers. In this case, every floating point number with a zero exponent is treated as a zero representation.

	Caution
	Two of these macros (`_FLT_LARGEST_EXPONENT_IS_NORMAL` and `_FLT_NO_DENORMALS`) specify deviations from IEEE 754. These macros only work with single-precision floating point and may not work correctly if hardware floating point support is used (enabled by configuring with `--enable-newlib-hw-fp`).

For most targets it is sufficient to define just one of __IEEE_BIG_ENDIAN or __IEEE_LITTLE_ENDIAN. The definitions should always be surrounded by a conditional, so they are only used when the target architecture is selected. For example the OpenRISC 1000 is big-endian, so we add the following to the header file.

#if defined(__or32__)
#define __IEEE_BIG_ENDIAN
#endif

	Caution
	One, and only one of `__IEEE_BIG_ENDIAN` and `__IEEE_LITTLE_ENDIAN` must be defined.

	Caution
	One, and only one of `__IEEE_LITTLE_ENDIAN` and `__IEEE_BIG_ENDIAN` must be defined.