Findings cpp_fp_double_backend

[ckormanyos]

Here are some open tasks for further quality enhancement of the new cpp_double_fp_backend<T> template

• cpp_double_fp_backend optimized single-limb operations #728
• Better infinity catches add/sub. Fixed by 715e0b4.
• Catch (max)() and lowest() in eval_convert_to-floating-point.
• Consider numeric_limits values particularly (max)() and epsilon(). Are there even better values for these?
• Improve (or finish) all constexpr for cpp_double_fp_backend Use the methods of Math-ccmath. #727
• Edge cases of pow(x, a) function, in other words $x^a$, for $x{\in}{\mathbb{R}}$, $a{\notin}{\mathbb{R}}$. See also Potential edge-case weakness in default_ops pow() function #722
• Potential edge-case weakness in default_ops pow() function #722
• When testing, it was discovered cpp_dec_float does not compile with complex_adaptor. Fixed by 1ac05b1
• The relative errors when testing data from the table with cpp_double_fp in test_pow.cpp subjectively seem a bit high. Why?
• Investigate and address the TBD comments in test_various_edges_more.cpp

Notes:

• The new cpp_double_fp_backend type is composed of two limbs. At the present time, add/sub/mul/div always do full 2-by-2 limb operations. This issue investigates odd double-limb, single-limb operations like the double-fp type being multiplied by a single fp-type having the same width as its constituents.