Curious if you've thought about whether it's beneficial to use the jdk.incubator.vector library. I noticed your posted benchmark for xxHash (not sure which variant) is an impressive 9.5 GB/s, which is just under the native version's XXH32 speed of 9.7 GB/s. Can we get it to "RAM speeds" that the native version's XXH3 (with SSE2) advertises by using explicit vectorization?
Amazing work here in this library btw!
Curious if you've thought about whether it's beneficial to use the jdk.incubator.vector library. I noticed your posted benchmark for xxHash (not sure which variant) is an impressive 9.5 GB/s, which is just under the native version's XXH32 speed of 9.7 GB/s. Can we get it to "RAM speeds" that the native version's XXH3 (with SSE2) advertises by using explicit vectorization?
Amazing work here in this library btw!