DSP-SIMD is a high-performance, open-source C++ toolkit engineered for computationally intensive Digital Signal Processing (DSP) workloads. By leveraging hardware-level SIMD (Single Instruction, Multiple Data) vector instructions, this library accelerates core mathematical algorithms, transforming execution timelines from serial latency into parallel throughput.

The library is architected for developers building real-time audio processors, software-defined radio (SDR) pipelines, telecommunication stacks, and high-frequency sensor arrays where raw computational efficiency is a hard requirement.

---

🚀 Key Architectural Features

• Hardware-Accelerated Vectorization: Explicitly utilizes processor vector registers to compute multiple data points simultaneously, unlocking near-linear speedups on modern CPU architectures.
• Modular DSP Primitives: Highly decoupled, reusable components designed for seamless drop-in integration into complex DSP pipelines.
• High-Performance Profiling: Includes a robust integration test suite and benchmarking subsystem located under perf-tests/ and unit-tests/ to guarantee algorithmic precision and measure strict execution cycle optimizations.
• Modern C++ Design: Adheres to low-overhead design patterns, zero-cost abstractions, and strict cache-aligned memory management protocols.

---

🛠 Supported SIMD Topologies

The toolkit is designed to abstract platform-specific compiler intrinsics, providing cross-platform scalability across a variety of hardware vectors:

• x86 / x86_64: Intel & AMD hardware acceleration platforms including SSE4.2, AVX, AVX2, and AVX-512 extensions.

---

📦 Core DSP Library Modules

The codebase is systematically isolated into the following key execution modules:

├── src/            # Core SIMD algorithm primitives and vectorized pipelines
├── perf-tests/     # Low-level performance evaluation benchmarks and cycle tracking
└── unit-tests/     # Strict logical correctness verification suites

1. Vectorized Arithmetic & Operations

Optimized block memory arithmetic routines processing contiguous data arrays in single-cycle instruction bounds:

• Vector-Vector and Vector-Scalar Addition, Subtraction, Multiplication, and Division.
• Complex number multi-lane multiplication and fused-multiply-add (FMA) routines.

---

⚙️ Getting Started

Prerequisites

• A modern C++ compiler supporting C++17 or higher (GCC, Clang, or MSVC).
• Hardware support for SIMD extensions (e.g., AVX2 or ARM NEON enabled via compiler flags like -march=native).

Cloning the Repository

Retrieve the source codebase directly from GitHub:

git clone https://github.com/bgin/DSP-SIMD.git
cd DSP-SIMD

Compilation & Optimization Flags

To harness the true performance of this toolkit, ensure target-specific optimization flags are provided to the compiler during compilation:

# Example compilation command enabling native vectorization instructions
g++ -O3 -march=native -std=c++17 src/main.cpp -I ./src -o dsp_simd_app

---

🧪 Verification & Benchmarking

The toolkit features dedicated sub-systems to verify numerical accuracy against standardized floating-point benchmarks, while simultaneously logging instruction performance.

Running Unit Tests

To confirm cross-platform precision and edge-case correctness:

# Navigate to unit test directory and compile suite
cd unit-tests
# Execute unit testing pipeline

Executing Performance Tests

To run raw cycle benchmarks and compute throughput speedup analysis:

cd perf-tests
# Execute performance binaries to monitor SIMD processing improvements

---

📄 Licensing & Contribution

This toolkit is open-source and distributed under the terms of the GNU General Public License v3 (GPLv3).

• Feel free to file issues, feature requests, or submit optimization pull requests via the official GitHub Repository.
• Created and actively maintained by bgin.