math.big: faster radix_str()

[tankf33der]

Faster conversion of limbs to radix.

Had to split the function into two. They are very similar, as it should be.
Each solves its specific tasks.
Now a larger number of digits is calculated at once.

Performance for radix 10:
o) -current -> 9 times faster;
o) 0.4.10 -> 57 times faster;

Tested against GMP for all radixes in range 2-36.

[huly-for-github]

Connected to Huly®: V_0.6-23024

[enghitalo]

Can you share the benchmark code?

[tankf33der]

Can you share the benchmark code?

Sure :)
Please check and share result.

http://pb1n.de/?4455b5

[spytheman]

With this slightly modified version of the linked code:

import math
import math.big
import benchmark

fn main() {
	mut x := u32(math.pow(3, 2))
	x = u32(math.pow(4, x))
	mut y := big.integer_from_int(5)
	mut b := benchmark.start()
	for _ in 0 .. 5 {
		_ := y.pow(x)
		b.measure('y.pow(x) loop')
	}
	y = y.pow(x)
	b.measure('y.pow(x)')
	str := y.str()
	b.measure('y.str()')
	// println(y)

	println('5^(4^(3^2)) has ${str.len} digits: ${str[..20]} ... ${str[str.len - 20..]}')
	assert str[..20] == '62060698786608744707'
	assert str[str.len - 20..] == '92256259918212890625'
}

on master:

#0 10:33:35 ^ master ~/v>v -cc tcc run x.v
 SPENT   179.234 ms in y.pow(x) loop
 SPENT   178.036 ms in y.pow(x) loop
 SPENT   166.351 ms in y.pow(x) loop
 SPENT   194.824 ms in y.pow(x) loop
 SPENT   171.393 ms in y.pow(x) loop
 SPENT   165.913 ms in y.pow(x)
 SPENT 42602.998 ms in y.str()
5^(4^(3^2)) has 183231 digits: 62060698786608744707 ... 92256259918212890625
#0 10:34:25 ^ master ~/v>
#0 10:34:36 ^ master ~/v>v -prod -cc clang-18 run x.v
 SPENT    34.503 ms in y.pow(x) loop
 SPENT    32.460 ms in y.pow(x) loop
 SPENT    37.107 ms in y.pow(x) loop
 SPENT    39.480 ms in y.pow(x) loop
 SPENT    36.140 ms in y.pow(x) loop
 SPENT    32.663 ms in y.pow(x)
 SPENT 23459.021 ms in y.str()
5^(4^(3^2)) has 183231 digits: 62060698786608744707 ... 92256259918212890625
#0 10:35:16 ^ master ~/v>

... while on the branch of this PR:

#0 10:35:16 ^ master ~/v>gh pr checkout 24666
remote: Enumerating objects: 6, done.
remote: Counting objects: 100% (6/6), done.
remote: Total 6 (delta 5), reused 6 (delta 5), pack-reused 0 (from 0)
Receiving objects: 100% (6/6), 1.02 KiB | 1.02 MiB/s, done.
Resolving deltas: 100% (5/5), completed with 5 local objects.
From github.com:vlang/v
 * [new ref]               refs/pull/24666/head -> radix_str
Switched to branch 'radix_str'
#0 10:35:47 ^ radix_str ~/v>
#0 10:35:48 ^ radix_str ~/v>v -cc tcc run x.v
 SPENT   170.928 ms in y.pow(x) loop
 SPENT   167.950 ms in y.pow(x) loop
 SPENT   166.826 ms in y.pow(x) loop
 SPENT   185.182 ms in y.pow(x) loop
 SPENT   165.408 ms in y.pow(x) loop
 SPENT   164.608 ms in y.pow(x)
 SPENT  5781.445 ms in y.str()
5^(4^(3^2)) has 183231 digits: 62060698786608744707 ... 92256259918212890625
#0 10:35:57 ^ radix_str ~/v>
#0 10:35:58 ^ radix_str ~/v>v -prod -cc clang-18 run x.v
 SPENT    33.097 ms in y.pow(x) loop
 SPENT    36.810 ms in y.pow(x) loop
 SPENT    31.755 ms in y.pow(x) loop
 SPENT    32.860 ms in y.pow(x) loop
 SPENT    36.864 ms in y.pow(x) loop
 SPENT    32.579 ms in y.pow(x)
 SPENT  3313.222 ms in y.str()
5^(4^(3^2)) has 183231 digits: 62060698786608744707 ... 92256259918212890625
#0 10:36:11 ^ radix_str ~/v>

[tankf33der]

In principle, I agree with these suggestions and will make a new push.

[tankf33der]

New push contains all suggestions + cap for st arrays.

update: tested against gmp too.