Hello,

I am not able to use nmslib for py3.9 or 3.10, for a code that works very well in a conda+py3.8 install.

Steps to reproduce:

conda create -n testnmslib39 python=3.9 # adapt for 3.8 or 3.10
conda activate testnmslib39
pip install nmslib numpy ipykernel

And then the code:

import nmslib
import numpy as np

oracle = nmslib.init(method='hnsw', space='cosinesimil', data_type=nmslib.DataType.DENSE_VECTOR)
source = np.random.random((1000, 100)) * 2.0 - 1.0
oracle.addDataPointBatch(source)
oracle.createIndex(print_progress=True)

query = np.random.random((1, 100)) * 2.0 - 1.0
print(oracle.knnQuery(query))

With python 3.8, I get:

(array([919, 942, 534,  43, 368, 528,  22, 114, 302, 879], dtype=int32),
 array([0.64898074, 0.70981157, 0.7411968 , 0.75587684, 0.7559525 ,
        0.75614583, 0.76265025, 0.7647664 , 0.76834595, 0.76985985],
       dtype=float32))

With python 3.9 or 3.10, I get:

(array([171, 171, 171, 171, 171, 171, 171, 171, 171, 171], dtype=int32),
 array([0.6919109, 0.6919109, 0.6919109, 0.6919109, 0.6919109, 0.6919109,
        0.6919109, 0.6919109, 0.6919109, 0.6919109], dtype=float32))

Note that:

• It installed nmslib-2.1.1-cp38-cp38-manylinux2010_x86_64.whl for 3.8, nmslib-2.1.1-cp39-cp39-manylinux2010_x86_64.whl for 3.9, and built from nmslib-2.1.1.tar.gz for 3.10
• With python3.8 and 3.9, I get a warning "Your CPU supports instructions that this binary was not compiled to use: SSE3 SSE4.1 SSE4.2 AVX AVX2", but not with 3.10 (seems coherent with the facts that it had to recompile because no binary is available).
• Recompiling on 3.10 with CFLAGS=-msse2 to get the same warning did not solve the issue. On the contrary, recompiling on 3.8 with -march=native did not yield any issue.
• I have the same issue without conda and python3.10 virtualenv
• The results are the same with knnQueryBatch or with knnQuery, and with any parameters inside createIndex and on QueryTimeParams (efS, M, post...)
• It seems really similar to sw-graph Returns the Same Results Over and Over again #561, but as I did more thorough examination and easiest minimal installation, I thought I'd give it more visibility creating a new issue

My computer&distro infos :

Kernel: 6.8.0-52-generic x86_64 bits: 64 compiler: N/A Desktop: Cinnamon 6.0.4 tk: GTK 3.24.33
    wm: muffin vt: 7 dm: LightDM 1.30.0 Distro: Linux Mint 21.3 Virginia base: Ubuntu 22.04 jammy
CPU:
  Info: 6-core model: Intel Core i7-10750H bits: 64 type: MT MCP smt: enabled arch: Comet Lake
    rev: 2 cache: L1: 384 KiB L2: 1.5 MiB L3: 12 MiB
  Flags: avx avx2 ht lm nx pae sse sse2 sse3 sse4_1 sse4_2 ssse3 vmx