mlib – Minimal Linear Algebra Library

mlib is a lightweight, modern C++ linear algebra library designed for performance, clarity, and portability. It provides efficient dense vector and matrix operations with architecture-specific backends.

Features

• Dense vector container: mlib::Vector<T>
• Dense matrix container: mlib::Matrix<T> (row-major storage)
• Optimized Level-1 vector operations:
  • axpy, scal, copy, swap, dot
• Element-wise vector operations: add, sub, mul, div, abs, neg
• Norms: L1 (sum of absolute values), L2 (Euclidean norm)
• Basic matrix operations

Supported Backends

• Scalar (portable fallback)
• AVX2 (x86-64 with FMA)
• NEON (ARM64 / AArch64, including Apple Silicon)

Requirements

• C++17 compiler (GCC 9+, Clang 8+, MSVC 2019+)
• CMake 3.20+
• AVX2-capable CPU (for AVX2 backend)
• ARM64 CPU (for NEON backend)

Build Instructions

From project root:

mkdir -p build && cd build

Linux / x86-64 (AVX2)

cmake .. -DCMAKE_BUILD_TYPE=Release \
         -DMLIB_ENABLE_AVX2=ON \
         -DMLIB_ENABLE_NEON=OFF \
         -DMLIB_BUILD_EXAMPLES=ON
cmake --build . --parallel $(nproc)

macOS / Apple Silicon (NEON)

cmake .. -DCMAKE_BUILD_TYPE=Release \
         -DMLIB_ENABLE_AVX2=OFF \
         -DMLIB_ENABLE_NEON=ON \
         -DMLIB_BUILD_EXAMPLES=ON
cmake --build . --parallel $(nproc)

Run examples

./examples/basic_vector_op
./examples/vector_elementwise

Usage

Include the single umbrella header for all kernels:

#include <mlib/kernels.hpp>

Vector example

mlib::Vector<float> x(1000), y(1000);
// fill data...

mlib::kernels::axpy_p32(2.5f, x, y);
float res = mlib::kernels::dot_p32(x, y);
mlib::kernels::add_p32(a, b, result);

Matrix example

mlib::Matrix<float> A(200, 300), B(300, 150), C(200, 150);
// fill matrices...

mlib::kernels::gemm_p32(A, B, C, 1.0f, 0.0f);  // C ← A·B

License

MIT License – free to use, modify, and distribute.