IndexPQFastScan train under WSL2 was slow

[solomonwzs]

Hi,
I'm using Faiss (v 1.12.0) on Windows 11 + WSL2. Faiss was built from source with

cmake -DFAISS_ENABLE_GPU=OFF \
      -DBUILD_TESTING=OFF \
      -DFAISS_ENABLE_PYTHON=ON \
      -DCMAKE_BUILD_TYPE=Release \
      -DFAISS_OPT_LEVEL=avx2 \
      -DCMAKE_INSTALL_PREFIX=/usr \
      ..
make

In C++ I create the index by

auto index = faiss::index_factory(dim, desc, type);
index->train(n, x);

When the description is IVF1024,PQ32 the process uses 100 % CPU and training finishes quickly.
With IVF1024,PQ32x4fs, however, CPU utilisation stays low and training becomes extremely slow.
The same code running on a native Linux machine does not show this behaviour.

How can I fix this?

[mnorris11]

We don't officially support WSL2 but converting to discussion if others have the same issue.