update-industrial-matrix-ultimate 2.1.0

Ultimate Industrial Matrix Library - High Performance C++ with Python Bindings

Industrial Matrix

High-performance matrix computing library with SIMD optimization and OpenMP parallelization.

Features

• SIMD Optimization: AVX2 vectorized operations for maximum performance
• OpenMP Parallelization: Multi-threaded matrix multiplication and operations
• NumPy Integration: Seamless conversion between NumPy arrays and Matrix objects
• Multiple Data Types: Support for float32, float64, int32, and int64
• Modern C++17: Clean, efficient implementation with PIMPL pattern

Installation

Prerequisites

sudo apt-get update
sudo apt-get install -y python3-dev python3-pip build-essential

Install from source

pip install pybind11 numpy
python setup.py build_ext --inplace
python setup.py install

Or use pip in development mode:

pip install -e .

Quick Start

import industrial_matrix as im
import numpy as np

# Check system capabilities
print(im.system_info())

# Create matrices
A = im.MatrixF64.random(1000, 1000)
B = im.MatrixF64.ones(1000, 1000)

# Matrix operations
C = A @ B  # Matrix multiplication
D = A + B  # Element-wise addition
E = A.transpose()

# NumPy interoperability
np_array = np.random.rand(100, 100)
matrix = im.MatrixF64.from_numpy(np_array)
result = matrix.to_numpy()

# Benchmark
stats = im.benchmark(size=512, trials=10)
print(f"Performance: {stats['gflops']:.2f} GFLOPS")

Performance

The library uses block-based matrix multiplication with cache optimization and SIMD instructions:

• Single-threaded: ~10-30 GFLOPS (depends on CPU)
• Multi-threaded: Scales linearly with core count
• SIMD: 4-8x speedup from AVX2 vectorization

API Reference

MatrixF64 / MatrixF32

Factory Methods

• zeros(rows, cols) - Create zero matrix
• ones(rows, cols) - Create matrix filled with ones
• identity(n) - Create identity matrix
• random(rows, cols, min, max) - Create random matrix

Properties

• rows() - Number of rows
• cols() - Number of columns
• shape() - Tuple of (rows, cols)
• size() - Total number of elements

Operations

• matrix_multiply(other) or @ - Matrix multiplication
• elementwise_add(other) or + - Element-wise addition
• elementwise_multiply(other) or * - Element-wise multiplication
• transpose() - Matrix transpose
• scalar_multiply(scalar) - Scalar multiplication

Advanced Patterns

Scalar Addition Workaround

The library doesn't have a direct scalar_add method. To add a scalar to all elements:

# Add scalar value to all matrix elements
ones = im.MatrixF64.ones(rows, cols)
scalar_matrix = ones.scalar_multiply(5.0)  # Create matrix of [5.0, 5.0, ...]
result = original_matrix.elementwise_add(scalar_matrix)

# Example: Add 10 to all elements
A = im.MatrixF64.random(100, 100)
offset = im.MatrixF64.ones(100, 100).scalar_multiply(10.0)
B = A.elementwise_add(offset)

NumPy Integration

• from_numpy(array) - Convert NumPy array to Matrix
• to_numpy() - Convert Matrix to NumPy array

Requirements

• Python 3.7+
• NumPy 1.19.0+
• C++17 compatible compiler
• OpenMP support (optional, for parallelization)
• AVX2 CPU support (optional, for SIMD)

License

MIT License

Contributing

Contributions welcome! Please submit pull requests or open issues on GitHub.