With pending code for Mesa, the LLVMpipe software-based Gallium3D/OpenGL driver could soon become faster on modern systems that support [1]AVX-512 . That's assuming the AI/LLVM-generated code is all in order.

A draft merge request was opened today for adding 512-bit wide vector support for fragment shaders to the [2]LLVMpipe CPU-based driver. Red Hat engineer David Airlie toyed with the idea of AVX-512 for LLVMpipe years ago but hadn't completed it. In the past few days, AI via the GPT Terra and Astra AI models they were able to complete the task.

[3]

David AIrlie noted in [4]a blog post that AI was used for closing the final gaps on the code and finding some bugs in the process. He also remarked that he doesn't have any AVX-512 hardware and thus hasn't been able to actually test the AVX-512 LLVMpipe implementation yet.

Those interested can find this AVX-512 LLVMpipe work via [5]this draft merge request .



[1] https://www.phoronix.com/search/AVX-512

[2] https://www.phoronix.com/search/LLVMpipe

[3] https://www.phoronix.com/image-viewer.php?id=2026&image=avx_512_glxgears_lrg

[4] https://airlied.blogspot.com/2026/10/jetlag-diversions-llvmpipe-512-wide.html

[5] https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/45279