Optimize Raw Data Search: BF functions and Intermin Index

[cqy123456]

Purpose:
Knowhere provides Brute Force functions or a JIT intermin index for milvus streaming raw data search.
In order to minimize the additional memory overhead of Milvus raw data search and improve search performance, knowhere will make the following optimizations:

1. bruteforce search with zero copy:

• More Simd support for fp16/bf16(batch_4); -> pr: enhance: BF functions support real fp16/bf16 calculate #980
• remove all (fp16/bf16->fp32) conversion, and use fp16/bf16 directly to do brute force; -> pr: enhance: BF functions support real fp16/bf16 calculate #980

**2. choose SCANN base index as intermin index: **

• scann iterator; -> pr: enhance: scann support iterator #992
• thread safe protection;
• sampling training or faster train;

3. a no memory(NM) version of interim index:

• a NM flat as storage;
• multi data type support;
• distances function with get_vector lamdba;

4. milvus related issue:

future works:

1. More data type support: binary/ sparse;
2. ivf flat cc with nm;
3. lock-free verison of thread safe scann;

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions. Rotten issues close after 30d of inactivity. Reopen the issue with /reopen.

[liliu-z]

@cqy123456 are we still tracking this here?

[liliu-z]

/reopen

[sre-ci-robot]

@liliu-z: Reopened this issue.

In response to this:

/reopen

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.