Hello, I am perusing through /mpn/powm.c in order to try and solve the
seg-faults on larger operations, and it uses (in some cases) mpn_sqr(), which
uses mpn_toom3_sqr_itch(n) and mpn_toom4_sqr_itch(n) to allocate some stack
memory.
These are both defined inside gmp-impl.h and there might be a typo or
copy-paste bug but I can't know for sure. The defines are:
/* toom33/toom3: Scratch need is 5an/2 + 10k, k is the recursion depth.
We use 3an + C, so that we can use a smaller constant.
*/
#define mpn_toom33_mul_itch(an, bn) \
(3 * (an) + GMP_NUMB_BITS)
#define mpn_toom3_sqr_itch(an) \
(3 * (an) + GMP_NUMB_BITS)
/* toom33/toom3: Scratch need is 8an/3 + 13k, k is the recursion depth.
We use 3an + C, so that we can use a smaller constant.
*/
#define mpn_toom44_mul_itch(an, bn) \
(3 * (an) + GMP_NUMB_BITS)
#define mpn_toom4_sqr_itch(an) \
(3 * (an) + GMP_NUMB_BITS)
As you can see both are defined as "(3 * (an) + GMP_NUMB_BITS)". I am only
bringing this up as the commenting for 3 is "Scratch need is 5an/2 + 10k, k is
the recursion depth" and the commenting for 4 is "Scratch need is 8an/3 + 13k,
k is the recursion depth".
So my question is was the mpn_toom4_sqr_itch() define supposed to use different
(larger) numbers than the mpn_toom3_sqr_itch() define?
-Brett
_______________________________________________
gmp-bugs mailing list
[email protected]
https://gmplib.org/mailman/listinfo/gmp-bugs
