Affected Component
Distance Computation
Current Behavior
Currently, though FP32 and FP16 are both optimized by using ARM64 SIMD, INT8 is not done yet.
Desired Improvement
INT8 NEON SIMD optimization, which can achieve 4x performance boost expectedly
Impact
Faster queries as well as build processing
Affected Component
Distance Computation
Current Behavior
Currently, though FP32 and FP16 are both optimized by using ARM64 SIMD, INT8 is not done yet.
Desired Improvement
INT8 NEON SIMD optimization, which can achieve 4x performance boost expectedly
Impact
Faster queries as well as build processing