Intel's Mesa Drivers Using The IGC Compiler Delayed
Last summer I wrote about Intel prototyping their Mesa drivers to use the IGC compiler, which followed Intel transitioning their Windows driver to use this compiler that was originally written for their open-source Linux compute stack. While they were making good progress last year on having their Mesa drivers use the IGC compiler, the project has been pushed back.
The LLVM-based Intel Graphics Compiler (IGC) has proven successful for their OpenCL / Level Zero Compute Runtime stack on Linux and now under Windows for their graphics driver over the past year. Last summer I was told about their prototyping work with their Vulkan "ANV" and Iris "OpenGL" drivers atop IGC as an alternative to their existing shader compiler within Mesa. Later in 2020 I was then informed Intel's Mesa drivers using IGC were progressing with promising performance results (faster than the current code). They hoped to roll out the IGC-based Intel Mesa drivers for initial public testing in H1'2021 and ideally by the end of 2021 that it would possibly be the Intel shader compiler default within Mesa for their open-source drivers.
After word got out that Intel was interested in using IGC in Mesa, it drew some criticism from various Mesa developers over those prospects.
As soon as H1'21 came to an end and not hearing anything about Intel IGC usage within Mesa since last year nor seeing any Mesa merge requests, I reached out to Intel for an update.
The Mesa-IGC project is still being pursued but they had to prioritize work on other unmentioned projects. Thus IGC within Mesa has been pushed out and likely will not arrive until 2022.
The delay on transitioning their Linux graphics drivers to IGC isn't entirely surprising given we have seen no mentions of it in recent months but their open-source engineers have been very busy on other efforts like preparing for their initial and future discrete graphics products and readying the Alder Lake support. On the DG1 graphics card front, with pending patches that should come together for the 5.15 kernel, there is finally the ability to run a working, accelerated desktop but other work is ongoing, including last week seeing initial DG2 bring-up and Xe_HP SDV support.
Switching out the graphics compiler back-end is no small task and prone to regressions, so understandably isn't something that Intel wants to rush amid all of their other driver happenings. IGC at least will continue moving forward thanks to the compute stack and Windows driver usage until that Mesa transition is ready.
The LLVM-based Intel Graphics Compiler (IGC) has proven successful for their OpenCL / Level Zero Compute Runtime stack on Linux and now under Windows for their graphics driver over the past year. Last summer I was told about their prototyping work with their Vulkan "ANV" and Iris "OpenGL" drivers atop IGC as an alternative to their existing shader compiler within Mesa. Later in 2020 I was then informed Intel's Mesa drivers using IGC were progressing with promising performance results (faster than the current code). They hoped to roll out the IGC-based Intel Mesa drivers for initial public testing in H1'2021 and ideally by the end of 2021 that it would possibly be the Intel shader compiler default within Mesa for their open-source drivers.
After word got out that Intel was interested in using IGC in Mesa, it drew some criticism from various Mesa developers over those prospects.
As soon as H1'21 came to an end and not hearing anything about Intel IGC usage within Mesa since last year nor seeing any Mesa merge requests, I reached out to Intel for an update.
The Mesa-IGC project is still being pursued but they had to prioritize work on other unmentioned projects. Thus IGC within Mesa has been pushed out and likely will not arrive until 2022.
The delay on transitioning their Linux graphics drivers to IGC isn't entirely surprising given we have seen no mentions of it in recent months but their open-source engineers have been very busy on other efforts like preparing for their initial and future discrete graphics products and readying the Alder Lake support. On the DG1 graphics card front, with pending patches that should come together for the 5.15 kernel, there is finally the ability to run a working, accelerated desktop but other work is ongoing, including last week seeing initial DG2 bring-up and Xe_HP SDV support.
Switching out the graphics compiler back-end is no small task and prone to regressions, so understandably isn't something that Intel wants to rush amid all of their other driver happenings. IGC at least will continue moving forward thanks to the compute stack and Windows driver usage until that Mesa transition is ready.
2 Comments