Painless now supports asymmetric bit dot-product, meaning, treating the individual bytes as a "mask" and summing the subsequent components on the query vector.
Right now, this is just a regular 'ole for loop. I think we can make it WAY faster with Panama SIMD.
The two methods in question are: ESVectorUtil.ipByteBit and ESVectorUtil.ipFloatBit
Description
Painless now supports asymmetric bit dot-product, meaning, treating the individual bytes as a "mask" and summing the subsequent components on the query vector.
Right now, this is just a regular 'ole for loop. I think we can make it WAY faster with Panama SIMD.
The two methods in question are:
ESVectorUtil.ipByteBit
andESVectorUtil.ipFloatBit