FastTree release 2.2.0 makes extensive use of auto-vectorization since it was proven that gcc and clang do handle its "vector" loops pretty well. See morgannprice/fasttree@a3ea39f and morgannprice/fasttree#1.
It might be worth creating four versions, containing the permutation of:
considering that (1) wasm isn't that fast yet and (2) not a lot of f64 vectorization can be done in 128-bit-length wasm vectors.
FastTree release 2.2.0 makes extensive use of auto-vectorization since it was proven that gcc and clang do handle its "vector" loops pretty well. See morgannprice/fasttree@a3ea39f and morgannprice/fasttree#1.
It might be worth creating four versions, containing the permutation of:
considering that (1) wasm isn't that fast yet and (2) not a lot of f64 vectorization can be done in 128-bit-length wasm vectors.