This repository provides detailed documentation and tooling for understanding the AMD Radeon RX 7900 XTX (Navi31/gfx1100) GPU architecture, focusing on its various IP blocks and firmware loading mechanisms. It's aimed at kernel developers, reverse engineers, and advanced users interested in low-level GPU operations on Linux.

How It Works

The project delves into the complex architecture of the RDNA3 GPU, breaking down its components like the Graphics Compute Unit (GC), Micro Engines (ME, MEC, MES), Command Processor (CP), and firmware loading process via the Platform Security Processor (PSP). It utilizes tools like umr (User Mode Register) to inspect hardware registers and firmware files, providing insights into driver interactions and hardware initialization.

Quick Start & Requirements

• Install: Requires a Linux environment with the AMDGPU kernel driver.
• Prerequisites: Access to umr tool (likely part of AMD's internal tooling or a community equivalent), and potentially kernel headers for recompiling modules.
• Setup: Basic setup involves ensuring the AMDGPU driver is loaded and potentially adding user groups (render, video). Debugging requires enabling dynamic debug prints in the kernel.
• Links:
  • AMDGPU Driver Core Documentation: https://docs.kernel.org/gpu/amdgpu/driver-core.html
  • RDNA3 Instruction Set Architecture: https://www.amd.com/content/dam/amd/en/documents/radeon-tech-docs/instruction-set-architectures/rdna3-shader-instruction-set-architecture-feb-2023_0.pdf

Highlighted Details

• Detailed breakdown of RDNA3 IP blocks (GC, PSP, SMU, VCN, SDMA, etc.) with firmware versions.
• Examples for dumping and analyzing GPU registers using umr.
• Guidance on recompiling and loading the amdgpu kernel module with specific parameters.
• Extensive list of firmware types and their corresponding kernel loading messages.

Maintenance & Community

• The repository appears to be a personal documentation effort rather than a community-driven project with active maintenance or a dedicated community channel.

Licensing & Compatibility

• The README does not specify a license. The content appears to be for informational and educational purposes.

Limitations & Caveats

• This repository is primarily documentation and analysis; it does not provide a functional library or framework for direct GPU programming. The use of internal AMD tooling like umr might require specific access or setup.