High-Performance Lazy Tensor Library (C++23)

Problem Statement: Create a High-Performance Tensor Math Library (Lazy Evaluation)

• High-Level Goal: Implement a NumPy-like library for C++ where operations like A + B * C do not create temporary matrices but instead build an expression tree evaluated only upon assignment.
• Core Concepts: Expression Templates, Move Semantics (Rule of 5), std::span (C++20), Smart Pointers.
• Architectural Challenges: Managing object lifetimes within expression trees. Optimizing for "Return Value Optimization" (RVO).
• Tooling: BLAS/LAPACK (for backend linkage), Google Benchmark.

Solution

A modern, header-only C++23 tensor library designed to demonstrate zero-overhead abstraction, lazy evaluation, and multi-dimensional memory mapping.

This project bypasses legacy C++ idioms (like the Curiously Recurring Template Pattern / CRTP) in favor of cutting-edge C++23 features to build a highly optimized, compile-time Abstract Syntax Tree (AST) for mathematical operations.

Architectural Highlights

• Zero-Overhead Expression Templates: Uses C++23 "Deducing this" (Explicit Object Parameters - P0847) to build lazy-evaluation trees without virtual function dispatch or CRTP boilerplate.
• Zero-Copy Broadcasting & Transposition: Separates physical 1D memory from logical N-dimensional shapes using programmable strides.
• Multidimensional Subscripting: Implements the C++23 multidimensional subscript operator (P2128) for native tensor[row, col] syntax.
• Eager vs. Lazy Separation: Intelligently separates element-wise operations (evaluated lazily in a single loop) from tensor contractions (evaluated eagerly to optimize CPU cache locality).

Prerequisites

To build and run this project, you must have a compiler that fully supports C++23 "Deducing this".

• Compiler: Clang 18.0 or newer (Strictly required)
• Build System: CMake 3.20 or newer
• OS: Linux (Ubuntu 18.1.3 recommended), macOS, or Windows (via Clang-cl)

Build Instructions

This project uses CMake to orchestrate the build and automatically fetches GoogleTest for unit testing.

# 1. Ensure Clang 18 is the active compiler
export CXX=clang++-18
export CC=clang-18

# 2. Configure the build directory
cmake -B build -S .

# 3. Compile the demonstration and test binaries
cmake --build build -j$(nproc)

# 4. Run the unit tests
cd build
ctest --output-on-failure -V

# 5. Run the demonstration
./tensor_demo

Acknowledgments

The problem statement, architectural blueprints, C++23 metaprogramming patterns, and build system configurations for this project were developed in collaboration with an AI assistant as part of an advanced systems-programming exercise.