Skip to content

WSL.md

Latest commit

 

History

History
211 lines (140 loc) · 5.11 KB

WSL.md

File metadata and controls

211 lines (140 loc) · 5.11 KB
Raw

Setup for WSL2 (Ubuntu)

This guide gets the CUDA build of SAM3DBody-cpp running under WSL2 on Windows. WSL has two quirks that trip people up:

  1. You need a CUDA 12.x toolkit + cuDNN 9.x that match the ONNX Runtime 1.20.1 GPU build (see DEPENDENCIES.md).
  2. WSL prepends every Windows PATH entry to your Linux environment, so the loader can pick up the Windows GPU driver instead of the WSL one unless you fix PATH / LD_LIBRARY_PATH (Step 6).

Do not install the full Linux NVIDIA driver inside WSL. The GPU driver lives on the Windows side. Inside WSL you install only the CUDA toolkit and cuDNN.

Prerequisites

  • Windows 10/11 with WSL2 and an Ubuntu distro installed.

  • A recent NVIDIA driver installed on Windows (the WSL GPU passthrough comes from it).

  • Verify the GPU is visible from inside WSL before continuing:

    nvidia-smi

    This must list your GPU. If it doesn't, fix the Windows driver / WSL setup first — nothing below will work otherwise.

Step 1 — Enable the lunar universe repo (for gcc-12)

The build needs gcc-12, which isn't in the default repos on newer Ubuntu releases. Add the archived lunar (23.04) universe repo to pull it in.

Open the sources file:

sudo nano /etc/apt/sources.list.d/ubuntu.sources

Paste this at the end:

Types: deb
URIs: http://old-releases.ubuntu.com/ubuntu/
Suites: lunar
Components: universe
Signed-By: /usr/share/keyrings/ubuntu-archive-keyring.gpg

Save, then refresh:

sudo apt update

If your Ubuntu already provides gcc-12 (apt-cache policy gcc-12 shows a candidate), you can skip this step.

Step 2 — Install gcc-12

sudo apt install gcc-12

Step 3 — Install CUDA Toolkit 12.x

Install the WSL-Ubuntu CUDA toolkit (this guide uses 12.2):

https://developer.nvidia.com/cuda-12-2-0-download-archive?target_os=Linux&target_arch=x86_64&Distribution=WSL-Ubuntu&target_version=2.0&target_type=deb_local

Pick Linux → x86_64 → WSL-Ubuntu so you get the WSL package (not the full desktop driver).

Verify:

nvcc --version

Expected (version line is what matters):

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2023 NVIDIA Corporation
Cuda compilation tools, release 12.2, V12.2.91
Build cuda_12.2.r12.2/compiler.32965470_0

Step 4 — Install cuDNN 9.x (for CUDA 12)

ONNX Runtime 1.20.1 requires cuDNN 9 — cuDNN 8 will not work and is the #1 cause of Failed to load library libonnxruntime_providers_cuda.so.

Get the install command for your distro here (choose Linux → x86_64 → Ubuntu, cuDNN for CUDA 12):

https://developer.nvidia.com/cudnn-downloads?target_os=Linux&target_arch=x86_64&Distribution=Ubuntu&target_type=deb_local

After installing, refresh the loader cache and confirm cuDNN 9 is present:

sudo ldconfig
ldconfig -p | grep -i cudnn

You should see libcudnn.so.9 listed.

Step 5 — Set your GPU's CUDA architecture in scripts/build.sh

Set -DCMAKE_CUDA_ARCHITECTURES for your card (89 = Ada: RTX 40-series incl. RTX 4060; 86 = Ampere RTX 30-series; 75 = Turing RTX 20-series):

#!/bin/bash

THISDIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"
cd "$THISDIR"
cd ..

mkdir -p build
cd build

cmake .. \
  -DCMAKE_BUILD_TYPE=Release \
  -DCMAKE_CUDA_ARCHITECTURES=89

make -j$(nproc)

exit 0

Then build:

bash scripts/build.sh

Step 6 — Fix the WSL GPU library path before running

WSL appends all Windows PATH entries to your Linux environment, so the loader can resolve the Windows GPU driver path instead of the WSL one. Export clean paths so Linux sees the WSL GPU libs (/usr/lib/wsl/lib) and your CUDA install:

export PATH=/usr/local/cuda-12.2/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
export LD_LIBRARY_PATH=/usr/lib/wsl/lib:/usr/local/cuda-12.2/lib64

Adjust cuda-12.2 to your installed version. To make this permanent, append the two export lines to ~/.bashrc (then source ~/.bashrc).

Verify CUDA EP can load

Before running the pipeline, confirm the ONNX Runtime CUDA provider has all its dependencies (no output = success):

ldd build/onnxruntime_dl/lib/libonnxruntime_providers_cuda.so | grep -i "not found"

If anything shows up here, revisit Step 4 (cuDNN) and Step 6 (paths). See DEPENDENCIES.md for the full troubleshooting flow, including why a CPU fallback fails with Could not find an implementation for Expand(13).

Now run, e.g.:

scripts/webcam.sh

Blender tip — MakeHuman/MPFB installed via Blender extensions

If you installed Blender manually and the MakeHuman (MPFB) extension through Blender's extension system, edit blender/blender_bvh_plugin.py.

Find:

if __name__ == "__main__":

and inside the loop over add-ons add this elif:

elif "mpfb" in addon.lower():   # MPFB installed as a Blender extension
    haveMPFB = True

Thanks to Vivek Chouhan ( https://www.linkedin.com/in/vivek-chouhan/ ) for providing this information!