Metal-Accelerated Tensor Operations (fasttensor)

This project demonstrates GPU-accelerated tensor operations using Apple's Metal framework. It performs element-wise addition and multiplication on large tensors, leveraging Metal for efficient computation.

---

Features

• GPU acceleration using the Metal framework.
• Lightweight and efficient tensor operations.
• Automated build process using a Makefile.

---

Directory Structure

├── include/
│   ├── MetalExecutor.h
│   └── Tensor.h
├── src/
│   ├── main.cpp
│   ├── MetalExecutor.cpp
│   └── Tensor.cpp
├── shaders/
│   └── tensor.metal
├── build/
│   ├── tensor_app
│   └── tensor.metallib
├── Makefile
└── README.md

---

Prerequisites

• macOS with Xcode installed.
• Command-line tools for Xcode (xcode-select --install).

---

Build and Run

Run the Makefile to compile the Metal shader and C++ source files: make

Run the Application: ./build/tensor_app

Clean the build directory: make clean

---

Metal Shader Details

The shaders/tensor.metal file defines two GPU compute functions:

• add_tensors: Performs element-wise addition of two tensors.
• multiply_tensors: Performs element-wise multiplication of two tensors.

These functions are called from the C++ application using Metal’s pipeline.

---

Example Output

After running the program, you should see output similar to this:

Metal GPU Tensor Multiply Time: 0.00234 s

---

Acknowledgments

Thanks to Apple’s Metal framework documentation. Inspiration drawn from GPU-accelerated tensor libraries like PyTorch and TensorFlow.