Vulkan Neural Network (vknn)

This project implements a simple feedforward neural network for the MNIST dataset, with the following variants:

• Naive C++ implementation - strightforward implementation using standard library and some sort of OOP (app/naive)
• Eigen-optimized implementation - linear-algebra-friendly implementation using Eigen library (app/eigen)

All variants load the MNIST dataset from embedded resources (embedded with cmakerc library) and train/test a neural network with specified number of hidden layer.

---

Requirements

• Minimum CMake: 3.25
• Compilers tested:
  • gcc (Ubuntu 10.5.0-1ubuntu1~20.04) 10.5.0
• Internet connection (required for CMake FetchContent)

If you use a different version or compiler and encounter a compilation error, please let me know and I'll fix it as soon as possible.

---

Building

cd vulkan_neural_network
mkdir build
cmake -S . -B build
cmake --build build

There may be compilation warnings from some third-party libraries. You can ignore them.

Configure options

• VKNN_BUILD_EIGEN - build Eigen-based implementation - eigen_nn, ON by default
  • VKNN_USE_AVX2 - use avx2 instructions set for the Eigen-based implementation, OFF by default, requires a CPU with avx2 support
• VKNN_BUILD_NAIVE - build naive implementation - naive_nn, ON by default
• VKNN_USE_DOUBLE - use double precision (double) data type instead of single precision (float), OFF by default

To use an option, provide it during the configuration step, for example:

cmake -S . -B build -DVKNN_USE_DOUBLE=ON -DVKNN_BUILD_EIGEN=OFF

---

Project Structure

• app/naive/ — Naive neural network implementation
• app/eigen/ — Eigen-based neural network implementation
• libs/common/ — Common utilities (e.g., data loading, timers)
• libs/warnings/ — Compiler warning settings
• resources/ — MNIST dataset files
• third_party/ — External dependencies (fmt, cmakerc, eigen)

Running

# naive implementation
./build/app/naive/naive_nn

# eigen implementation
./build/app/eigen/eigen_nn

All implementations will train and test a neural network on the embedded MNIST dataset.

Dataset

The following MNIST files are embedded as resources:

• train-images.idx3-ubyte
• train-labels.idx1-ubyte
• t10k-images.idx3-ubyte
• t10k-labels.idx1-ubyte

License

See LICENSE.

Credits

• cmakerc
• Eigen
• fmt