AMD emits fresh open-source GPU tools for HPC, game devs
Fleshes out GPUOpen strategy
AMD has fleshed out its notion of an openly defined GPU architecture, GPUOpen, with the launch of a bunch of open-source tools on GitHub plus a shiny new website.
The move has been welcomed by the gaming press, but GPUOpen is not all about blasting people in 3D death matches – AMD also has the high performance computing (HPC) community in mind. The “Professional Compute” side of the initiative brings together tools like:
- HIP – described as a portable C++ runtime and kernel language for GPUs, with tools to help migrate CUDA code into C++.
- ContinuumAnalytics' HSA-accelerated Python.
- The open source HC Compiler for C++ code.
- HSAIL debugging, an OpenCL-based system health monitor, and a bunch of different HPC-targeted libraries and applications.
AMD's Greg Stoner, senior director Radeon Open Compute, said that for GPUOpen, “we have rebuilt our core foundation on how you access our GPU resources starting from the driver, runtime, compilers and even the programming languages.”
He also announced the Lightning Compiler Initiative, which will feed into AMD's existing Boltzmann Initiative. Announced in September 2015, Boltzmann includes a Linux driver for cluster-class GPUs, and an HPC-optimised HSA+ (heterogeneous system architecture) runtime.
For the game devs, there's code and documentation to give them more control over the GPU, with GPUOpen software hosted at GitHub to keep that collaboration-love ticking along. Game and graphics tools in GPUOpen include:
- The CodeXL Analyzer CLI – offline compilation and performance analysis for OpenCL kernel, DirectX shaders and OpenGL shaders.
- A DirectX plugin offering debugging tools like API trace, GPU trace, linked trace and an object inspector.
Meanwhile, an assortment of tools are available here. All the source code appears to be MIT-licensed. ®