$ the-wire · showcase
Salsa20 gets a loong64 SIMD path
By RepoJournal · Filed · About Go · Composed from the cited sources · methodology
A single commit to golang/crypto takes Salsa20 on LoongArch from 3.8 microseconds to under 500 nanoseconds per 1KB XOR, the first optimization this desk has seen for the architecture.
salsa20: add loong64 SIMD implementation golang/crypto
The new SIMD implementation measured on Loongson-3C6000 at 2.20GHz cuts XOR1K from 3835.5ns to 493.9ns, an 87.12% reduction (p=0.000, n=10). If you run x/crypto/salsa20 on loong64, this is a drop-in speedup in the published module; on every other GOARCH nothing changes.