RadeonSI Lands Improved Scaling For Shader Compiler Threads

Merged this week were some minor changes to AMD's RadeonSI Gallium3D OpenGL open-source driver around the shader selector code. One of the changes in particular though is noteworthy.

As part of "radeonsi: shader selector minor changes", there is radeonsi: scale the number of shader compiler threads. While a small code change, this alteration to RadeonSI scales the number of shader compiler threads rather than spawning all of the threads at start-up.

AMD driver developer Pierre-Eric Pelloux-Prayer summed up the change as, "This speeds up short lived programs (eg: piglit runs duration is reduced by ±25%), avoid wasting resources and still make use of multi-threaded capabilities."

Benefiting short-lived shader programs is great but very exciting is the impact for the Piglit OpenGL regression/unit driver testing framework. Piglit is commonly used by Mesa developers and reducing its run-time by around 25% is very significant. This is very useful for improving the developer experience in being able to finish up Piglit regression test runs quicker for saving developer time as well as for continuous integration (CI) testing in being able to complete the Piglit tests in roughly a quarter less time than has been the case with the existing shader compiler threading behavior.

So all around directly and/or indirectly by way of more efficient development/testing pace this change should be beneficial for users of the RadeonSI Gallium3D driver for OpenGL on GCN 1.0 GPUs and newer.