* same behavior for positive numbers
* generalisation for negative numbers, treating them as two complement
* improve algorithm, iterate once over the digits, manually perform two complement
* simplify mp_add_d, mp_sub_d
* functions are safe in case of a==c or b==c
* renamed mp_tc_div_2d to mp_signed_rsh (signed right shift)
mp_word is an internal type and it is problematic if it is exposed
in the public api. See for example #216 - MSVC does not support 128 bit
mp_words. But it is perfectly ok to use those internally in the library,
as long as the library is compiled with GCC.
The return type of mp_get_bit was imprecise (either mp_err or mp_bool),
therefore this function is deprecated in favor of s_mp_get_bit for now.
If we need s_mp_get_bit to be public, we should add it under a different
name. However since mp_set_bit is not available, I don't think there any
downstream users (ab)using mp_int as bitsets.
* MP_USE_ENUMS enables enums
* Wc++-compat catches some implicit conversions if MP_USE_ENUMS is defined
* 100% backwards compatible API/ABI if MP_USE_ENUMS is not defined
* deprecate MP_PRNG_ENABLE_LTM_RNG
* custom mp_rand_source is used always if set, which should be more aligned with user expectations
* use custom source in tune.c
* don't call random number generator once per digit, which is slow
* In the default settings, a cutoff X can be modified at runtime
by adjusting the corresponding X_CUTOFF variable.
* Tunability of the library can be disabled at compile time
by defining the MP_FIXED_CUTOFFS macro.
* There is an additional file tommath_cutoffs.h, which defines
the default cutoffs. These can be adjusted manually or by the
autotuner.
Header files which are located in the same directory that the file from where it is included must be included using `" "`, not `< >`.
Otherwise the compiler (gcc 5) cannot understand `#include <tommath_class.h>` in `/usr/include/tommath/tommath.h`.
