isomorphism

Isomorphism

Isomorphism is a platform-agnostic C++ math wrapper that unifies MLX (Apple Silicon) and SYCL (oneMKL/PC) backends. It acts as a routing layer, translating high-level tensor operations into highly optimized native calls for the active hardware, ensuring seamless performance across disparate architectures.

By using Isomorphism, you write your mathematical logic once using a Domain Specific Language (DSL) and deploy it anywhere without worrying about the underlying silicon.

---

Key Features

• Platform-Agnostic DSL: A unified interface for tensor operations including element-wise arithmetic and logical operations.
• Intelligent Routing: Automatically dispatches math to MLX on Apple devices or SYCL/oneMKL on Intel/NVIDIA/AMD hardware.
• Heavy Linear Algebra: Built-in support for SVD, QR decomposition, determinants, and linear system solvers.
• Seamless Integration: Designed to be a lightweight dependency for any C++ project through a modern CMake interface.

---

Installation

The recommended way to install Isomorphism on macOS is via Homebrew. This ensures all backend dependencies, like MLX, are correctly configured.

# 1. Install MLX for macOS compatibility
brew install mlx

# 2. Tap the custom Isomorphism repository
brew tap c0rmac/homebrew-isomorphism

# 3. Install the library
brew install isomorphism

Integrating with Your Project

Once installed, you can integrate Isomorphism into your project using CMake. The package exports a clean target that handles all include paths and backend-specific linking automatically.

CMake Configuration

Add the following to your CMakeLists.txt:

# 1. Locate the Isomorphism package
find_package(isomorphism REQUIRED)

# 2. Your executable
add_executable(my_app main.cpp)

# 3. Link the Isomorphism target
# This automatically handles include paths and backend dependencies (like MLX/SYCL)
target_link_libraries(my_app PRIVATE isomorphism::isomorphism)

Basic Usage

#include <isomorphism/math.hpp>
#include <isomorphism/tensor.hpp>

namespace iso = isomorphism;

int main() {
    // Operations are automatically routed to the active backend (MLX or SYCL)
    iso::Tensor a = iso::math::full({3, 3}, 1.0f, DType::Float32);
    iso::Tensor b = iso::math::eye(3, DType::Float32);
    
    // Perform a batched matrix multiplication
    iso::Tensor result = iso::math::matmul(a, b);
    
    // Explicitly execute the computation graph
    iso::math::eval(result);
    
    return 0;
}

Project Structure

• include/isomorphism/: Contains the public API and hardware-agnostic DSL headers.
• src/backends/mlx/: Implementation files for the Apple Silicon backend using the MLX framework.
• src/backends/sycl/: Implementation files for the PC/Linux backend using SYCL and oneMKL.

---

License

This project is licensed under the MIT License.