gmp-4.1.2 assertion failure

[Kevin Ryde]

Jason <jasonmoxham at btclick.com> writes:
>
>     umul_ppmm (t3, t4, bsize, e);
>     umul_ppmm (t5, t6, cnt - GMP_NAIL_BITS, e);

Yep, that'll be the idea.

>     t2 = GMP_NUMB_BITS;
>     ...
>     udiv_qrnnd (t8, t1, t5, t6, t2);

No, in the normal case where GMP_NUMB_BITS is a power of 2 it should
be shifts, not udiv_qrnnd.  Could hide a 2-limb by numb_bits division
in a macro to avoid junking up the code with conditionals.

> #define MAX_USEFULL_MP_SIZE_T_MAX	(MP_SIZE_T_MAX/BYTES_PER_MP_LIMB)

That has to look at SIZE_T_MAX too, since size_t is what we pass to
the alloc functions.

I'm not sure what we can say about the relative sizes of mp_size_t and
size_t.  Probably mp_size_t >= size_t normally, but on cray where
mp_size_t is an int maybe that's not so.


Incidentally, when mp_limb_t is a longlong bigger than ulong, the
current simple expression can probably be retained, if we felt like
checking that.