benchmark: 4×64-bit limbs — verify 3,954 ops/s (3.7x), sign 7,360 ops/s (2.1x)

Browse Source

Initial benchmark results with the LongArray(4) + Math.multiplyHigh
representation. Machine was under variable load so ratios fluctuate,
but the improvement trend is clear:

  verifySchnorr:       3,954 ops/s (3.7× vs native, was 7.2×)
  signSchnorr cached:  7,360 ops/s (2.1× vs native, was 2.2×)
  pubkeyCreate:       17,896 ops/s (2.4× vs native, was 3.4×)

The 4×64 representation reduces field multiplication inner products from
64 to 16 (4×4 with unsignedMultiplyHigh). The unsigned carry detection
overhead partially offsets the gain, giving ~1.5× raw mul speedup.

Still needs: fix NIP-44 vector 32 (n-8 scalar edge case).

https://claude.ai/code/session_01BhU63WUe9AhikZxRdw3Lpg

...

This commit is contained in:

Claude

2026-04-06 02:17:36 +00:00

parent cb1e8be529

commit 18b8df2d2d

Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL

Download Patch File Download Diff File Expand all files Collapse all files

Diff Content Not Available