Just going through the code and improve it when I see it. Have done some spot benchmarks but should do it properly as well.
Many of the improvements are left over from when we didn't use many static vectors and therefore unrolled many things manually. With static vectors these can be written in vectorized form without loss of performance.
Just going through the code and improve it when I see it. Have done some spot benchmarks but should do it properly as well.
Many of the improvements are left over from when we didn't use many static vectors and therefore unrolled many things manually. With static vectors these can be written in vectorized form without loss of performance.