Release v4.1.0

[Andrew Dawson]

A new release of rpe has been made, the documentation has been updated. The new features in this release are:

New IEEE 754 rounding mode

You can now set an option RPE_IEEE_ROUNDING to use an unbiased "round to nearest, tie to even" rounding scheme, which is specified by IEEE 754. This differs from the default rounding mode which always rounds to nearest. In the next major release (v5) the default and only rounding mode will become the IEEE version, and this flag is a temporary measure to ease the transition to the new rounding mode. Activating IEEE rounding can in some cases produce rather different numerical results (more realistic) and it is therefore recommended that this option is used universally.

Helper for working with FP literals

A new function rpe_literal has been added to simplify code that uses floating-point literals. Previously it was necessary to assign all reduced precision values to an type(rpe_var) in order to control their precision. The new function can be used to avoid this, for example:

type(rpe_var) :: a, b, c
RPE_DEFAULT_SBITS = 10

a = 8.112233
b = 2.31 * a ...                 ! The first operation is evaluated at single precision because the literal 2.31 is single precision
c = rpe_literal(2.31) * a ...    ! The first operation is evaluated with a 10-bit significand

As always, please report any problems to the Github issue tracker, and contributions of any size are always welcome.