Jay is a new open-source shader compiler being developed for Intel’s open-source OpenGL and Vulkan Linux drivers. Ultimately this Jay shader compiler should help in delivering better Linux graphics performance with modern Intel hardware.
Last year Alyssa Rosenzweig joined Intel who led work with Asahi Linux on their AGX graphics driver code and was a contractor as well for Valve working on the Linux graphics stack. With Rosenzweig’s contributions to Mesa over the years was a lot of work on shader compilation and NIR, so to not much surprise now, she’s been leading work on a new Intel shader compiler.
Though it is a bit surprising to see Jay being developed as besides the “BRW” shader compiler long used by Intel drivers within Mesa, Intel also has the Intel Graphics Compiler (IGC) used by their Compute Runtime for OpenCL and Level Zero on Linux. IGC is also used on Microsoft Windows not only for compute but also their graphics shader compiler too. I was previously told a few years ago that Intel was looking at using IGC for their Mesa drivers, but now it turns out there is a new solution being developed with this “Jay” compiler.
In today’s Mesa merge request introducing Jay, Alyssa explained:
“This MR adds Jay, a new SSA-based compiler for Intel GPUs. This is an early work-in-progress. It isn’t ready to ship, but we’d like to move development in-tree rather than rebasing the world every week. Please don’t bother testing yet – we know the status and we’re working on it!
Jay’s design is similar to other modern NIR backends, particularly ACO, NAK and AGX. It is fully SSA, deconstructing phis after RA. We use a Colombet register allocator similar to NAK, allowing us to handle Intel’s complex register regioning restrictions in a straightforward way. Spilling logical registers is straightforward with Braun-Hack… but how we apply the standard SSA backend design to Intel hardware is not. You can read the code if you’re curious, but I plan to “spill” the beans at XDC… So it’s fitting for several reasons that it’ll be in Toronto this year :-)”
At the moment the Jay compiler is just targeting Intel Xe2 hardware but will be expanded over time. The compiler can pass conformance testing for OpenGL ES 3.0 and OpenCL 3.0 while Vulkan work is progressing.
The performance of Jay so far is quite compelling compared to their current shader compiler:
“For a teaser though… here’s a nasty CTS test (math_bruteforce sin) compiled on both the old (brw) and new (jay) compilers:
jay: 6768 instructions (361:396 spills:fills) – 7.00 seconds
brw: 12980 instructions (578:1144 spills:fills) – 19.91 seconds
Better code than the current compiler in a fraction of the time… the future looks bright for Mesa compilers on Intel :-)”
A very promising first step for Jay. It will be exciting as Jay matures and becomes viable for experimental testing. Jay is written in C and comes in at just over 14k lines of new code. Those curious can see this merge request for the initial details on the new Intel Jay shader compiler.
