Update BLST - Githubissues

status-im / nim-blscurve

Nim implementation of BLS signature scheme (Boneh-Lynn-Shacham) over Barreto-Lynn-Scott (BLS) curve BLS12-381

Apache License 2.0

26 stars 11 forks source link

Update BLST #86

Closed mratsim closed 3 years ago

mratsim commented 3 years ago

This updates BLST from commit Reference diff:

Before https://github.com/supranational/blst/tree/6e0190c from Jul 13, 2020
After https://github.com/supranational/blst/tree/a8398ed from Sep 17, 2020
https://github.com/supranational/blst/compare/6e0190c...a8398ed

State

[x] Support APi changes from https://github.com/supranational/blst/pull/9
[x] Remove vectorization aliasing workaround from #69, https://github.com/status-im/nim-blscurve/commit/2317e9ceac3378f9540bbbbc0b1fdbaf73808a9b, partly address #84 Upstream: https://github.com/supranational/blst/issues/22, and probably related https://github.com/supranational/blst/commit/f8a77bd97fd9318efab20725f3a4b068a1842e85
[x] Change the workaround from https://github.com/status-im/nim-blscurve/commit/da9ae49dab4cbb95b2cdaa68ecc221ee15c67a9a to pass -D__BLST_PORTABLE__ instead of falling back to Milagro if the CPU doesn't support SSE3
[x] Expose the optimized SHA256 primitives for use in hashTreeRoot

mratsim commented 3 years ago

The speedup with BLST SHA256 is 14x~15x

In particular hashTreeRoot is one of our bottleneck in NBC and the BeaconState is 5MB

arnetheduck commented 3 years ago

is that using some fancy instructions or plain code?

mratsim commented 3 years ago

With SSE3 for vectorized xor, shuffles and vectorized shift right:

https://github.com/supranational/blst/blob/master/build/elf/sha256-x86_64.s#L359-L362

This is portable to ARM since they have equivalent vector instructions.