Build with Bombs 💣

A Minecraft Java mod for procedural house generation by diffusion

This repo contains two components:

• inference_dll This contains a C++ DLL that sets up CUDA and calls TensorRT. It provides a number of exported functions for use by Java.
• mod_neoforge This is the Java mod code that calls the inference.dll functions. It handles getting / setting blocks in Minecraft.

Version requirements:

• neoforge-21.1.77
• Minecraft-1.21.1
• TensorRT-10.5.0.18
• CUDA 12.6

Hardware compatibility

TensorRT 10.5 requires an NVIDIA GPU with compute capability >= 7.5. This means it requires an RTX 2060 or better, a GTX 1660 Ti or better, an MX550 or better, or a Tesla T4 or better. See the support matrix. Check this Wikipedia table to find the compute capability of your GPU: Compute capability, GPU semiconductors and Nvidia GPU board products

Setup Guide

This setup guide includes steps for building the .jar Java mod file as well as building the native executable.

Required packages and programs

• CMake: https://cmake.org/download/
• Java 21 JDK: https://www.oracle.com/java/technologies/downloads/#jdk21-windows
• CUDA 12.6: https://developer.nvidia.com/cuda-12-6-0-download-archive
• TensorRT 10.5: https://developer.nvidia.com/tensorrt/download/10x
• (Optional) Make: https://www.gnu.org/software/make/

You can check your environment configuration by running the following commands:

cmake --version

java --version

nvidia-smi
nvcc --version

dpkg -l | grep nvinfer

make --version

Check tensorrt_linux_install_steps.md for TensorRT installation details.

Build steps (Windows)

1. Run ./gradlew build. After the build succeeds, the mod .jar file will be located at mod_neoforge/build/libs/buildwithbombs-0.2.1.jar.

2. Build the inference library and test executable using CMake:

   cd inference_dll
   mkdir build
   cd build
   cmake ..
   cmake --build . --config Release

   This will produce inference.dll in the inference_dll/build/Release directory.

3. Create a run directory inside mod_neoforge if it does not exist, and copy the generated DLL

4. Download the ONNX model from the release page and place it in the mod_neoforge/run directory. This contains the model parameters and must be located next to inference.dll.

5. Make sure inference.dll is able to find the TensorRT and CUDA dynamic libraries. Either copy all DLLs into the mod_neoforge/run directory, or add the CUDA and TensorRT lib folders to the system path.

6. Test the mod by running:

   ./gradlew runClient
   

Build steps (Linux x86_64)

1. Run ./gradlew build. After the build succeeds, the mod .jar file will be located at mod_neoforge/build/libs/buildwithbombs-0.2.1.jar.

2. Build the inference library and test executable:

   Option 1: Using Make (Recommended)

   cd inference_dll
   
   # Build the shared library
   make lib

   Option 2: Using CMake directly

   cd inference_dll
   mkdir build
   cd build
   cmake ..
   cmake --build . --config Release

3. Copy the newly built library (libinference.so) to the mod's run folder. Create the run folder if it doesn't exist:

   cp libinference.so ../mod_neoforge/run

4. Download the ONNX model from the release page and place it in the mod_neoforge/run directory. This contains the model parameters and must be located next to libinference.so.

5. Make sure libinference.so can find the TensorRT and CUDA shared libraries. Either copy all required .so files into mod_neoforge/run, or add the TensorRT and CUDA library folders to your system path:

   export LD_LIBRARY_PATH=/usr/local/tensorrt-10.5/lib:$LD_LIBRARY_PATH

6. Test the mod by running:

   ./gradlew runClient

Build steps (Linux aarch64)

There are still some issues with the Linux aarch64 build. Before we fix them, you can try the following steps to build the mod.

1. following the steps in tensorrt_linux_install_steps.md to install TensorRT and CUDA.

2. Locate the following CUDA and TensorRT and copy them to your Minecraft run directory (the parent directory of /mods). (usually in /usr/local/tensorrt-10.5/lib)

   • libnvinfer_builder_resource.so.10.5.0
   • libnvinfer.so.10.5.0

   (usually in /usr/local/cuda-12.6/lib64)

   • libnvonnxparser.so
   • libcudart.so.12

3. Place buildwithbombs-0.2.1.jar(build from previous step) in the Minecraft /mods folder

4. Place libinference.so (build from previous step) in the Minecraft run directory.

5. Place ddim_single_update.onnx (from release page) in the Minecraft run directory.

6. Start the game. If it loads, you will be given "Diffusion TNT" items upon entering a world. Placing one of these blocks triggers the diffusion process.

once you move all the files to the run directory, your folder structure should be like this:

.
├── ddim_single_update.onnx
├── libcudart.so.12
├── libinference.so
├── libnvinfer_builder_resource.so.10.5.0
├── libnvinfer.so.10
├── libnvonnxparser.so
├── ···
├── mods
│   └── buildwithbombs-0.2.1.jar
└── versions
    └── 1.21.1-NeoForge

6. Start the game. If it loads, you will be given "Diffusion TNT" items upon entering a world. Placing one of these blocks triggers the diffusion process.

Building and running the standalone test

To test your TensorRT installation without running Minecraft, you can build and run a standalone test executable:

On Linux:

cd inference_dll

# Build the shared library libinference.so
make lib

# Build the test executable inference
make test 

# Run the test
make run

# Clean all build files
make clean

On Windows:

1. Build the shared library (inference.dll)

   # Create a build directory for the shared library
   mkdir build_lib
   cd build_lib
   cmake ..
   cmake --build . --config Release

   This will generate inference.dll in the build_lib directory.

2. Build the test executable

   # Create a build directory for the test executable
   mkdir build_test
   cd build_test
   # This will automatically enable test mode without manual code changes
   cmake .. -DSTANDALONE_TEST=ON
   cmake --build . --config StandaloneTest

   This will generate the inference executable in the build_test directory. When building with STANDALONE_TEST=ON:

   • Test code is automatically enabled
   • Console output is enabled (no log file redirection)
   • No manual code changes are needed

3. Run the inference

   ./inference.exe

the test will build two parallel diffusion threads, and for each thread, it will diffuse 1000 timesteps.

In the end, you will see the output in the console like:

job: 0, step = X, sum = Y
job: 1, step = X, sum = Y

If you can see the step goes to 0, and the sum is not 0, then the installation of TensorRT is successful.

Problem you might encounter

ERROR: Mod and diffusion engine don't match! Init failed.

That means you have different version of the mod and the diffusion engine. you should update your code by git pull and build buildwithbombs-0.2.1.jar again.

Social

buildwithbombs.com

Discord to ask questions

Join this server to try it out (no client-side mod required): mc.buildwithbombs.com 🧨