Home

AOCL-DLP Documentation Hub

AOCL-DLP (AMD Optimizing CPU Libraries - Deep Learning Primitives) is a high-performance library providing optimized deep learning primitives for AMD processors. It implements GEMM operations for machine learning applications, supporting multiple data types, fused pre/post-operations, and batch processing -- all tuned to leverage AMD hardware capabilities including AVX2, AVX512, AVX512_VNNI, AVX512_BF16, and AVX512_FP16 instruction sets.

New here? Start with the Quick Start Guide to build, install, and run your first GEMM in 5 minutes.

---

Getting Started

• Quick Start -- Install, build your first program, and run it
• Integration Guide -- CMake packages, manual linking, static vs dynamic, troubleshooting
• Examples & Tutorials -- Annotated code examples for every feature

User Guides

• Library Overview -- Architecture, components, data types, hardware abstraction
• GEMM Guide -- Data type combinations, memory layouts, matrix reordering, choosing the right variant
• Post-Operations Guide -- Fused post-ops (BIAS, activations, SCALE, MATRIX_ADD/MUL) via dlp_metadata_t
• Eltwise Operations Guide -- Standalone element-wise operations (separate from GEMM post-ops)
• Quantization Guide -- Symmetric quantization, mixed-precision workflows, scale/zero-point setup
• API Lifecycle -- End-to-end flow: data prep, post-ops setup, compute, threading

Performance & Configuration

• Performance Guide -- Threading, NUMA, memory layout, architecture-specific tips
• Environment Variables -- Complete reference for DLP_NUM_THREADS, AOCL_ENABLE_INSTRUCTIONS, OpenMP tuning

Testing & Benchmarking

• DLP Testing -- Google Test framework, YAML configs, running and writing tests
• DLP Benchmarking -- Google Benchmark framework, YAML configs, performance analysis

Developer Guides

• JIT Code Generation -- Just-In-Time compilation system, Xbyak assembler, kernel debugging

Reference

• FAQ -- Common questions about threading, linking, data types, and performance
• API Reference (Sphinx) -- Full generated API documentation

Project Links

• README -- Feature summary and data type table
• BUILD.md -- Build configuration and CMake options
• INSTALL.md -- Installation steps
• Contributing -- How to contribute
• License -- BSD 3-Clause