An LLVM-Based Intel GPU 3D Driver Won't Come Soon
This weekend in Brussels I heard about the state of Intel's initiative for migrating to an LLVM-based shader compiler for their open-source Mesa 3D driver to replace their existing custom compiler embedded within the driver. Unfortunately, the state of this LLVM-based compiler is rather meager.
Last August I wrote about how Intel was aiming for a higher performant driver and as part of that to switch to an LLVM-based shader compiler for a big performance win. After being unconvinced by LLVM and its IR for years, Intel Open-Source Technology Center developers last year expressed plans to switch to using an LLVM compiler rather than their own custom creation they have been maintaining for years.
They wanted to use LLVM since most projects in this area are using the open-source compiler infrastructure in some fashion. They want to take advantage of the LLVM IR, LLVM's optimizations, and other existing code rather than continuing to re-create their own code in this area.
I hadn't heard anything about an Intel LLVM compiler since last year in San Diego when it was originally talked about, so this weekend in Brussels at FOSDEM I asked Ian Romanick, Intel's lead Mesa developer, about this project.
While Intel has lots of open-source graphics driver developers working on their graphics stack, the LLVM shader compiler for their graphics driver isn't coming along. As for why there isn't any experimental Intel Mesa branch with an LLVM compiler or any other news on the topic is that this new compiler is no longer aggressively being pursued.
Also in terms of Beignet or any other open-source OpenCL/GPGPU support for Intel on Linux, there isn't anything to officially report on at this time.
Latest Articles & Reviews
Latest Linux News
Most Viewed News This Week