perf: tiny improvement in scrypt

[ChALkeR]

I see an ~4% improvement from not using % N inside loop over N (and casting to int32)

N is always a power of two, as checked above

[paulmillr]

make sure to add a brief code comment describing why this works

[paulmillr]

also, did you check safari?

[ChALkeR]

@paulmillr I see similar 4-5% improvement on WebKit (in playwright on mac)
Had to fix benchmark code to check that though
Tested only on scrypt(pass, salt, { N: 2 ** 10, r: 8, p: 1, dkLen: 32 })

[ChALkeR]

Approximate:

Engine	n = 2 ** 10	n = 2 ** 19
Node.js 22	505 → 520 (+2.84%)	0.95 → 0.97 (+3.03%)
Chromium	547 → 571 (+4.16%)	1.10 → 1.10 (+2.67%)
WebKit	447 → 477 (+6.99%)	0.82 → 0.88 (+7.50%)
Firefox	314 → 325 (+3.33%)	0.61 → 0.65 (+5.14%)
JSC	450 → 479 (+6.05%)	0.83 → 0.88 (+6.06%)

Raw data

Node.js, main:

scrypt(n: 2 ** 10, r: 8, p: 1) x 505 ops/sec @ 1978μs/op (1888μs..2ms)
scrypt(n: 2 ** 19, r: 8, p: 1) x 0.95 ops/sec @ 1057ms/op

Node.js, pr:

scrypt(n: 2 ** 10, r: 8, p: 1) x 520 ops/sec @ 1922μs/op (1771μs..11ms)
scrypt(n: 2 ** 19, r: 8, p: 1) x 0.97 ops/sec @ 1025ms/op

Chromium, main:

scrypt(n: 2 ** 10, r: 8, p: 1) x 547 ops/sec @ 1826μs/op (1600μs..8ms)
scrypt(n: 2 ** 19, r: 8, p: 1) x 1.1 ops/sec @ 938ms/op

Chromium, pr:

scrypt(n: 2 ** 10, r: 8, p: 1) x 571 ops/sec @ 1750μs/op (1600μs..8ms)
scrypt(n: 2 ** 19, r: 8, p: 1) x 1.1 ops/sec @ 913ms/op (905ms..922ms)

WebKit, main:

scrypt(n: 2 ** 10, r: 8, p: 1) x 447 ops/sec @ 2ms/op (2ms..3ms)
scrypt(n: 2 ** 19, r: 8, p: 1) x 0.82 ops/sec @ 1216ms/op

WebKit, pr:

scrypt(n: 2 ** 10, r: 8, p: 1) x 477 ops/sec @ 2ms/op (2ms..3ms)
scrypt(n: 2 ** 19, r: 8, p: 1) x 0.88 ops/sec @ 1131ms/op

Firefox, main

scrypt(n: 2 ** 10, r: 8, p: 1) x 314 ops/sec @ 3ms/op (2ms..7ms)
scrypt(n: 2 ** 19, r: 8, p: 1) x 0.61 ops/sec @ 1634ms/op

Firefox, pr:

scrypt(n: 2 ** 10, r: 8, p: 1) x 325 ops/sec @ 3ms/op (2ms..6ms)
scrypt(n: 2 ** 19, r: 8, p: 1) x 0.65 ops/sec @ 1550ms/op

jsc, main:

scrypt(n: 2 ** 10, r: 8, p: 1) x 450 ops/sec @ 2ms/op
scrypt(n: 2 ** 19, r: 8, p: 1) x 0.83 ops/sec @ 1204ms/op

jsc, pr:

scrypt(n: 2 ** 10, r: 8, p: 1) x 479 ops/sec @ 2ms/op
scrypt(n: 2 ** 19, r: 8, p: 1) x 0.88 ops/sec @ 1131ms/op

[paulmillr]

thanks