GPU build/test farm

gfx-rs project has a set of GPU workload tests that we want to expand and maintain in order to keep the backends functioning without regressions. Problem is that developers are not able to run the reftest framework on all supported platforms (Vulkan, D3D12, Metal-1, Metal-2, GL, WebGL, etc).

In order to grow steadily, we need a continuous integration support for those tests, which requires a set of machines running Windows, Linux, and OSX with access to graphics hardware. This is certainly well beyond what Bors does today, but I’m hoping to start the discussion here.

1 Like

There is no need to modify Bors. What you need is a CI system that can post Github status notifications. Then simply gate bors on that status notification. For example, Nix uses both Travis and Buildbot, and Bors is gated on both. See this link for an example of how to gate on Buildbot.

What you need isn’t a Bors extension. You just need a Buildbot or Jenkins instance, with a GPU-equipped buildslave.