CU8 — K=30 @ “Threads=256”: clamp vs. clamp-avoidance (baseline vs warpmerge vs warpmerge2 sendM=1)

CU7 baseline vs warpmerge vs warpmerge2

ADD results to CU5 & CU6

CU6 — NCU mechanism snapshot

Added Warpmerge Result Analyzation

Added half2 & Warpmerge

Add CUDA Phase

Adjusted figure captions.

Added Performance_summary.md and REPRO.md

Added Phase 6B&C: IVF

Added Phase 5C:INT8 + Query Batching

Added Phase 4B:INT8 & Phase 6:HNSW Baseline

Adjust tables

Adjust tables

Query batching