Airlie: "Why Sharing Code With Windows Isn't Always A Win"
Following the news today of Intel sharing ~60% of their GPU driver code-base between Windows and Linux and working to bring the Intel Graphics Compiler (IGC) to Mesa in 2021, not everyone is enthusiastic about those prospects.
David Airlie at Red Hat serves as the DRM co-maintainer for the Linux kernel, contributes significantly to Mesa particularly for Lavapipe/LLVMpipe/Clover, and also is the co-founder of the open-source Radeon Vulkan "RADV" driver as well as being involved in the early days of the open-source AMD driver work, among other open-source/Linux accomplishments over the years. Airlie penned a blog post today in response to Intel's news of the increased Windows/Linux code sharing and IGC likely coming to Mesa.
Airlie feels this isn't a great idea and likely bad for Linux-based distributions. The main argument comes down to the fact that IGC is largely developed behind closed doors at Intel and doesn't currently foster a community/open-source-first workflow. Intel at least right now doesn't really engage with the open-source development community over IGC but largely pushes out weekly-ish code snapshots, similar to what AMD does with AMDVLK and LLPC/XGL, etc.
Intel's current NIR compiler within Mesa is meanwhile developed in the open and through the same community review process as other parts of Mesa. Airlie would personally rather see companies like Intel that are serious about code sharing to instead start out from the open-source project(s) that exist and adapt to their Windows/closed-source rather than the approach currently employed by IGC or AMDVLK where there is less community involvement.
As Intel pushes ahead with IGC for Mesa, we'll see if the development approach around the Intel Graphics Compiler changes at all to make it more friendly and transparent for the open-source community.
For end-users at least and consumers not concerning themselves with the development model, all indications are the IGC'ified Mesa performance should be faster than the current compiler back-end.
Airlie's thoughts on the matter can be found via his personal blog.
David Airlie at Red Hat serves as the DRM co-maintainer for the Linux kernel, contributes significantly to Mesa particularly for Lavapipe/LLVMpipe/Clover, and also is the co-founder of the open-source Radeon Vulkan "RADV" driver as well as being involved in the early days of the open-source AMD driver work, among other open-source/Linux accomplishments over the years. Airlie penned a blog post today in response to Intel's news of the increased Windows/Linux code sharing and IGC likely coming to Mesa.
Airlie feels this isn't a great idea and likely bad for Linux-based distributions. The main argument comes down to the fact that IGC is largely developed behind closed doors at Intel and doesn't currently foster a community/open-source-first workflow. Intel at least right now doesn't really engage with the open-source development community over IGC but largely pushes out weekly-ish code snapshots, similar to what AMD does with AMDVLK and LLPC/XGL, etc.
Intel's current NIR compiler within Mesa is meanwhile developed in the open and through the same community review process as other parts of Mesa. Airlie would personally rather see companies like Intel that are serious about code sharing to instead start out from the open-source project(s) that exist and adapt to their Windows/closed-source rather than the approach currently employed by IGC or AMDVLK where there is less community involvement.
As Intel pushes ahead with IGC for Mesa, we'll see if the development approach around the Intel Graphics Compiler changes at all to make it more friendly and transparent for the open-source community.
For end-users at least and consumers not concerning themselves with the development model, all indications are the IGC'ified Mesa performance should be faster than the current compiler back-end.
Airlie's thoughts on the matter can be found via his personal blog.
60 Comments