Re: How to do t_INT bit operations?

[hermann@stamm-wilbrandt.de]

On 2023-07-06 09:14, hermann@stamm-wilbrandt.de wrote:
...
> A very first test with splitting exponentiation into 2 and then
> multiply results on slow
> Raspberry Pi400 shows that runtime does not get worse by doing so:
...

On 7600X target CPU number did not decrease.
Reason was that I measured the sum of both exponentiations, while
they should run in parallel. So I reported all exponentiation times
and really both exponentiations on exponents with half the bits set
is always a bit faster than first time, but not much:

hermann@7600x:~/RSA_numbers_factored/pari$ gp -q split.gp
prime p has
2467 decimal digits length
8193 bits length
5 bits set
66ms
0ms
64ms
hermann@7600x:~/RSA_numbers_factored/pari$ vi split.gp
hermann@7600x:~/RSA_numbers_factored/pari$ gp -q split.gp
prime p has
10000 decimal digits length
33218 bits length
16431 bits set
2395ms
2274ms
2276ms
hermann@7600x:~/RSA_numbers_factored/pari$ vi split.gp
hermann@7600x:~/RSA_numbers_factored/pari$ gp -q split.gp
prime p has
36401 decimal digits length
120921 bits length
60327 bits set
57396ms
53865ms
53959ms
hermann@7600x:~/RSA_numbers_factored/pari$

To be tested, whether splitting into exponents with 8.3% bits set will
provide more speedup.

hermann@7600x:~/RSA_numbers_factored/pari$ diff split.gp.orig split.gp
14c14,16
< p = 2 ^ 2 ^ 13 + 897;
---
> R(n) = { (10^n - 1) \ 9; }
>
> \\ p = 2 ^ 2 ^ 13 + 897;
15a18
> p = 34 * R(36400) - 42000040044444004000024 * 10^2264 * R(36400) \ 
> R(4550) - 1;
32a36
> print(Str(gettime())"ms");
34d37
< s=lift(s1*s2);
35a39
> s=lift(s1*s2);
hermann@7600x:~/RSA_numbers_factored/pari$


Regards,

Hermann.