Closed jsign closed 1 year ago
results on my arm64 machine:
name old time/op new time/op delta
FpInverse/go-ipa-12 2.82µs ± 0% 1.93µs ± 0% ~ (p=1.000 n=1+1)
FpInverse/go-blst-12 2.47µs ± 0% 2.53µs ± 0% ~ (p=1.000 n=1+1)
So the speedup is similar.
This PR shows how the optimized
go-ipa
Inverse(...)
code change impacts benchmarks in this repo.The raw
Inverse(...)
benchmark has the same expected speedup improvement as shown in thego-ipa
benchmark:Regarding our other benchmark to specifically measure the performance of trie key generation:
So, we got around a ~1.09x speedup.
And the quite exciting benchmark improvements for updating a Verkle Trie leaf:
This means that updating a Verkle Trie leaf got ~1.6x speedup. This makes sense as shown in my previous document where the flamegraph shows that the cost was dominated by
Inverse(..)
. We also got a slight improvement in full Pedersen Commitment, but that operation is dominated byMul(...)
.Note: I'll keep this as a draft PR until we have an idea if https://github.com/crate-crypto/go-ipa/pull/26 will get merged.