offc_delphi - High Performance Matrix Operations Library

offc_delphi is a dynamic library written in C, providing efficient matrix multiplication operations for Delphi applications. It implements various advanced algorithms including sequential, OpenMP parallelized, SIMD vectorized (AVX2/FMA), and Strassen's matrix multiplication. This module is designed for High-Performance Computing (HPC) tasks requiring intensive algebraic computations.

Features

• Matrix Multiplication Algorithms: Implements standard sequential (IKJ), OpenMP parallelized, and Strassen's recursive multiplication.
• SIMD Vectorization: Utilizes AVX and AVX2 intrinsics along with loop unrolling to maximize CPU throughput in vector operations.
• Delphi Wrapper: Includes a Delphi wrapper for seamless integration.
• High-Performance Computing: Enables vectorized/optimized computations for improved performance.
• Cross-Language Compatibility: Distributed as a DLL, accessible from various programming environments.

Requirements & Development Environment

The library is developed, tested, and intended to be used in the following environment:

• Operating System: Windows 11 (64-bit)
• Compiler: GCC 15.1.0
• Toolchain: MinGW-w64 12.0.0 UCRT (release 1) from winlibs.com
• Dependencies: OpenMP for multi-threading.
• Build System: mingw32-make
• Delphi Integration: RAD Studio (Delphi 12.1 Community Edition)

Toolchain Setup and Installation Steps

To compile this library, you must install the specific GCC toolchain that supports OpenMP multi-threading. Follow these steps carefully:

1. Download the Correct Compiler Version

• Navigate to winlibs.com.

• Locate the block titled: "GCC 15.1.0 (with POSIX threads) + MinGW-w64 12.0.0 (UCRT) - release 1".

• Under this block, look for the Win64 row (specifically: Win64 without LLVM/Clang/LLD/LLDB).

• Click on 7-Zip archive (recommended for smaller download size) or Zip archive to download the package.

  Note: It is strictly required to choose the POSIX threads variant because GCC's OpenMP implementation (libgomp) relies natively on the POSIX threading standard (pthread) for stable multi-threaded execution.

2. Extraction and Directory Setup

• Extract the downloaded archive.

• Move the extracted folder (usually named mingw64) to a clean, global system path without spaces (e.g.: C:\winlibs).

3. Add to Windows System Path
To call gcc and mingw32-make from any terminal, add the toolchain's bin directory to your system environment variables:

• Press Win + S, type "env", and select Edit the system environment variables.

• In the System Properties window, click on the Environment Variables... button.

• Under System variables (bottom list), select the variable named Path and click Edit....

• Click New on the right side and type the absolute path to the extracted bin directory:

  C:\winlibs\bin
  

• Click OK to close each of the three setup windows

4. Verification
Open a new Command Prompt (cmd) or PowerShell window and run the following commands to verify the installation:

gcc --version
mingw32-make --version

If configured correctly, both commands will output their respective version information (GCC 15.1.0 and GNU Make), confirming that your development environment is ready.

Project Structure

offc_delphi/
│
├── build/                  # Compiled binaries and intermediate files
├── include/                # Public API headers (C)
│   └── offc_delphi.h
├── interface/              # Delphi wrapper
│   └── OffC.pas
├── src/                    # C source code
│   └── offc_delphi.c
├── LICENSE                 # License information
├── Makefile                # Build script for DLL
└── README.md               # Project documentation

Compilation Instructions

Once your toolchain environment is properly set up, building the library is simple:

1. Open a terminal (Command Prompt or PowerShell).

2. Navigate to the root directory of this repository:

   cd path\to\offc_delphi

3. Run the compiler build tool:

   mingw32-make

This will generate offc_delphi.dll in the build\ directory.

Using the Library in Delphi

1. Linking the DLL (Choose One Method)

To allow your Delphi application to load and communicate with the dynamic library, you must make offc_delphi.dll discoverable by Windows. You can achieve this using one of the following methods:

Method A: Add the Build Directory to the System Path (Recommended for Development)
During active development of the C library, you might need to recompile the DLL frequently. Copying the file manually after every compilation is tedious and error-prone.

1. Add the absolute path of your local offc_delphi\build\ directory directly to your Windows system's PATH environment variable (following the same steps used for the compiler setup).

Method B: Place the DLL in the Executable Folder (Recommended for Distribution)
If you are packaging, deploying, or sharing your compiled Delphi application:

1. Locate your Delphi project's output directory (usually named Win64\Debug or Win64\Release inside your project's folder structure).

2. Copy offc_delphi.dll from offc_delphi\build\ and paste it directly into that folder, alongside your compiled .exe file.

2. Integrating the Wrapper in RAD Studio / Delphi

To use the library functions natively inside your Delphi Pascal source code, follow these steps:

1. Open your Delphi project in RAD Studio.

2. Go to the top menu and select Project > Options....

3. In the left panel, navigate to Building > Delphi Compiler.

4. In the main area, click the Search Path field.

5. Click on the ellipsis ... button and add the absolute path to the interface\ folder inside your repository (e.g.: C:\Projects\offc_delphi\interface).

6. Click OK and then Save.

Example Usage:

uses OffC;

// Multiply matrices A and B into C using sequential IKJ algorithm
mm_ikj_seq(@MatrixA[0], @MatrixB[0], @MatrixC[0], Rows, K, Cols);

Academic Citation

If you use this software in your research, please cite it using the metadata provided in the CITATION.cff file located in the root of this repository.

License

This project is licensed under the MIT License - see the LICENSE file for details.