Performance improvements broke PQ about jvector HOT 4 CLOSED

this doesn't match what you saw initially, so either we broke it during review, or this only manifests on my machine

i couldn't figure out how to get to your pre-review code, I think this is b/c you squashed on merge, but clearly it exists somewhere since you can still click on the link in github

$ git co 923d2c849fab3e519277e5b36fe4b43be1eee5d7
fatal: reference is not a tree: 923d2c849fab3e519277e5b36fe4b43be1eee5d7

from jvector.

This regression only occurs when running without SIMD (which made it extra hard to track down because at some point in the bisect intellij reset my run configuration)

from jvector.

I bet it's this

  @Override
  public float dotProduct(float[] a, int aoffset, float[] b, int boffset, int length)
  {
    float sum = 0f;
    for (int i = 0; i < length; i++) {
      sum += a[aoffset + i] * b[aoffset + i];
    }

    return sum;
  }

should be using boffset

meeting time, will check afterwards

from jvector.

Sorry about that. I’m gonna add a test to make sure we catch breaks like this

from jvector.