perf: reduce binary operations for Chi() & Maj()

[kewde]

I noticed some interesting albeit insignificant optimizations in some SHA256 implementations.

https://github.com/bitcoin-core/secp256k1/blob/00774d0723af1974e2a113db4adc479bfc47e20f/src/hash_impl.h#L18C14-L18C55

They remove one binary operation, but perhaps not worth the extra mental load for auditors since it does deviate from FIPS etc.

I've also seen variants where the last operator in Maj is ^ instead, I assume that it has the same performance as |.
https://github.com/digitalbazaar/forge/blob/2bb97afb5058285ef09bcf1d04d6bd6b87cffd58/lib/sha256.js#L297

[paulmillr]

Could you run benchmarks and tell the difference? npm run bench

[kewde]

I haven't seen a statistically notable difference on my machine.

I've changed the benchmark to only run sha256 & do 32B for 10M runs.
After running each one for about 10 times, I've picked the best results for each:

master

# 32B
sha256 x 1,381,215 ops/sec @ 724ns/op

PR

# 32B
sha256 x 1,396,648 ops/sec @ 716ns/op

Slightly better, at best a 1% improvement in the best runs

All master runs

user@Mac noble-hashes % yarn bench             
yarn run v1.19.1
$ node benchmark/noble.js
# 32B
sha256 x 1,315,789 ops/sec @ 760ns/op

# 1MB
sha256 x 243 ops/sec @ 4ms/op

✨  Done in 11.91s.
user@Mac noble-hashes % yarn bench
yarn run v1.19.1
$ node benchmark/noble.js
# 32B
sha256 x 1,371,742 ops/sec @ 729ns/op

# 1MB
sha256 x 230 ops/sec @ 4ms/op

✨  Done in 11.37s.
user@Mac noble-hashes % yarn bench
yarn run v1.19.1
$ node benchmark/noble.js
# 32B
sha256 x 1,381,215 ops/sec @ 724ns/op

# 1MB
sha256 x 235 ops/sec @ 4ms/op ± 5.02% (4ms..12ms)

✨  Done in 11.25s.
user@Mac noble-hashes % yarn bench
yarn run v1.19.1
$ node benchmark/noble.js
# 32B
sha256 x 1,375,515 ops/sec @ 727ns/op

# 1MB
sha256 x 233 ops/sec @ 4ms/op

✨  Done in 11.23s.
user@Mac noble-hashes % yarn bench
yarn run v1.19.1
$ node benchmark/noble.js
# 32B
sha256 x 1,335,113 ops/sec @ 749ns/op

# 1MB
sha256 x 242 ops/sec @ 4ms/op

✨  Done in 11.74s.
user@Mac noble-hashes % yarn bench
yarn run v1.19.1
$ node benchmark/noble.js
# 32B
sha256 x 1,355,013 ops/sec @ 738ns/op

# 1MB
sha256 x 242 ops/sec @ 4ms/op

✨  Done in 11.48s.
user@Mac noble-hashes % yarn bench
yarn run v1.19.1
$ node benchmark/noble.js
# 32B
sha256 x 1,342,281 ops/sec @ 745ns/op

# 1MB
sha256 x 228 ops/sec @ 4ms/op

✨  Done in 11.83s.
user@Mac noble-hashes % yarn bench
yarn run v1.19.1
$ node benchmark/noble.js
# 32B
sha256 x 1,355,013 ops/sec @ 738ns/op

# 1MB
sha256 x 241 ops/sec @ 4ms/op

✨  Done in 11.56s.
user@Mac noble-hashes % yarn bench
yarn run v1.19.1
$ node benchmark/noble.js
# 32B
sha256 x 1,356,852 ops/sec @ 737ns/op

# 1MB
sha256 x 243 ops/sec @ 4ms/op

✨  Done in 11.62s.
user@Mac noble-hashes % yarn bench
yarn run v1.19.1
$ node benchmark/noble.js
# 32B
sha256 x 1,356,852 ops/sec @ 737ns/op

# 1MB
sha256 x 242 ops/sec @ 4ms/op

✨  Done in 11.31s.
user@Mac noble-hashes % yarn bench
yarn run v1.19.1
$ node benchmark/noble.js
# 32B
sha256 x 1,344,086 ops/sec @ 744ns/op

# 1MB
sha256 x 245 ops/sec @ 4ms/op

✨  Done in 11.54s.
user@Mac noble-hashes % yarn bench
yarn run v1.19.1
$ node benchmark/noble.js
# 32B
sha256 x 1,366,120 ops/sec @ 732ns/op

# 1MB
sha256 x 232 ops/sec @ 4ms/op

✨  Done in 11.58s.
user@Mac noble-hashes % yarn bench
yarn run v1.19.1
$ node benchmark/noble.js
# 32B
sha256 x 1,366,120 ops/sec @ 732ns/op

# 1MB
sha256 x 233 ops/sec @ 4ms/op

✨  Done in 11.53s.

All PR runs

user@Mac noble-hashes % yarn bench
yarn run v1.19.1
$ node benchmark/noble.js
# 32B
sha256 x 1,373,626 ops/sec @ 728ns/op

# 1MB
sha256 x 243 ops/sec @ 4ms/op

✨  Done in 11.35s.
user@Mac noble-hashes % yarn bench
yarn run v1.19.1
$ node benchmark/noble.js
# 32B
sha256 x 1,396,648 ops/sec @ 716ns/op

# 1MB
sha256 x 243 ops/sec @ 4ms/op

✨  Done in 11.22s.
user@Mac noble-hashes % yarn bench
yarn run v1.19.1
$ node benchmark/noble.js
# 32B
sha256 x 1,392,757 ops/sec @ 718ns/op

# 1MB
sha256 x 244 ops/sec @ 4ms/op

✨  Done in 11.12s.
user@Mac noble-hashes % yarn bench
yarn run v1.19.1
$ node benchmark/noble.js
# 32B
sha256 x 1,375,515 ops/sec @ 727ns/op

# 1MB
sha256 x 219 ops/sec @ 4ms/op ± 5.35% (4ms..15ms)

✨  Done in 11.69s.
user@Mac noble-hashes % yarn bench
yarn run v1.19.1
$ node benchmark/noble.js
# 32B
sha256 x 1,351,351 ops/sec @ 740ns/op

# 1MB
sha256 x 241 ops/sec @ 4ms/op

[paulmillr]

Thanks for the pull request! We will keep it as-is for now. There are different ways of doing that.