Install¶

Files needed to run¶

Voxelyze3¶

Voxelyze3 is an executable that split tasks into multiple groups, and call vx_node_worker to execute them on different GPUs. vx_node_worker works with Voxelyze3 and is the one who actually carry out the calculation. (So don’t delete it.)

VoxCAD¶

VoxCAD can visualize history file recorded via Voxelyze3. VoxCAD should be run locally, and GPU is only optional for VoxCAD.

Install Voxelyze3¶

On DeepGreen¶

DeepGreen is UVM’s GPU cluster. We have already compiled on DeepGreen, so it will be quite easy to use Voxelyze3 on DeepGreen.

Follow a five-minute instruction here: https://github.com/liusida/gpuVoxels-dg-installation

On Desktop/Laptop with GPUs¶

If you have access to root, the best way to get started is using Docker, since Nvidia provide ready to use CUDA docker image files.

First, install Docker CE (Community Edition).

Then, install nvidia-container:

sudo apt-get update
sudo apt-get install -y nvidia-container-runtime nvidia-container-toolkit

Then, start the container:

docker run --gpus all --name=gpuVoxels --volume=/tmp:/gpuVoxels/host -it sidaliu/gpuvoxels:1.0

Note

You can copy Voxelyze3 and vx_node_worker to wherever you want and run it there.

Note

You can change the path /tmp to anywhere you want, and keep your code and data there.

Note

If you are not familiar with Docker, here’s some helpful commands you will use.

# The commands below should be run on host command line
# See what containers are running
docker ps
# See what containers are running/suspending
docker ps --all
# You should see a suspended container called gpuVoxels
# Let's re-enter that container
docker start gpuVoxels
docker attach gpuVoxels

How I made the docker image¶

If you’d like to see what’s inside the image sidaliu/gpuvoxels:1.0, here is the way I build it.

docker run --gpus all --name=gpuVoxels --volume=/tmp:/host -it nvidia/cuda:10.1-devel-ubuntu18.04

The meaning of last command is run a docker container using nvidia image cuda, download it if not exists. The name of container is gpuVoxels, and /tmp of the host is shared in the container as /host.

After a while…

Now, I am inside the docker container.

apt update
apt install -y git cmake libboost-all-dev
cd /
mkdir gpuVoxels
git clone https://github.com/liusida/gpuVoxels.git gpuVoxels_src
cd /gpuVoxels_src/Voxelyze3
mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE=Release -DCUDA_DEBUG=OFF ..
cmake --build .
cp Voxelyze3 /gpuVoxels
cp vx_node_worker /gpuVoxels
exit
# Back to the host

docker commit -a "Sida Liu <sliu1@uvm.edu>" gpuVoxels sidaliu/gpuvoxels:1.0
docker push sidaliu/gpuvoxels:1.0

On Server with GPUs but without root¶

It will be too tricky to install all the dependencies without root, especially CUDA 10.x.

If there’s already CUDA 10.1 or 10.2 on the server, you’ll need to compile and install cmake, g++, boost from source.

Install VoxCAD¶

VoxCAD need OpenGL, Boost, QT5. You will need to build VoxCAD from source. Here are instructions for Ubuntu and Mac.

On Ubuntu¶

The package names in Ubuntu are:

OpenGL (libglfw3-dev, freeglut3-dev, libglm-dev, mesa-utils), Boost (libboost-all-dev), QT5 (qtbase5-dev).

Here I demostrate how to do it in Ubuntu 18.04.

sudo apt install -y git cmake libboost-all-dev qtbase5-dev libglfw3-dev freeglut3-dev libglm-dev mesa-utils
# this will take a while...
git clone https://github.com/liusida/gpuVoxels.git gpuVoxels_src
cd /gpuVoxels_src/VoxCAD
mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
cmake --build .
# done.

On Mac¶

brew install cmake
brew install boost
brew install qt5
brew install glfw3
brew cask install xquartz
brew install freeglut
brew install glm
brew install mesa

# this will take a while...

git clone https://github.com/liusida/gpuVoxels.git gpuVoxels_src
cd gpuVoxels_src/VoxCAD/
mkdir build
cd build/
cmake -DCMAKE_BUILD_TYPE=Release ..

Wait a minute, you need to make a little bit of hacking in the source files:

first is VoxCAD/src/QTUtils/QOpenGL.cpp

Change line 11 from
    #include "GL/glu.h"
To
    #include "OpenGL/glu.h"

second is VoxCAD/CMakeFiles.txt

Change line 70 from
    find_package(glm CONFIG REQUIRED) # glm
to
    # find_package(glm CONFIG REQUIRED) # glm

Change line 44 from
    target_link_libraries(VoxCAD PRIVATE ${OpenGL_LIBRARIES} GL)
to
    target_link_libraries(VoxCAD PRIVATE ${OpenGL_LIBRARIES})

OK, let’s continue.

cmake --build .
# done.

Thank Arlo Cohen at University of Vermont for providing the instruction on Mac.

Running VoxCAD¶

Run VoxCAD using command:

./VoxCAD <filename.history>

Note

Now you can copy the file VoxCAD and Default.vxc to wherever you want and run it there.

Note

If you forget the file Default.vxc, the simulation will stuck at every step.