RadeonSI Lands Another "Very Large" Optimization To Further Boost SPECViewPerf

In recent months we have seen a lot of RadeonSI optimizations focused on SPECViewPerf with AMD seemingly trying to get this open-source OpenGL driver into very capable shape moving forward for workstation GL workloads. Hitting Mesa 22.0-devel today is yet another round of patches for tuning SPECViewPerf.

Already with Mesa 21.3-devel all indications are the RadeonSI performance is at or beyond the performance of AMD's proprietary OpenGL driver available on Windows and via the Radeon Software for Linux "PRO" driver package for workloads like SPECViewPerf... For gaming, RadeonSI has long been superior to the proprietary driver. But the AMD open-source OpenGL driver team still isn't done yet optimizing for SPECViewPerf and achieving every possible advantage.

The newest merge to Mesa 22.0-devel is this set of 11 patches to enable pipelined partial buffer uploads and attempting to sync less for clear_buffer and copy_buffer operations. As part of that this merge is implementing pipelined partial buffer uploads using CPU storage.

The performance improvement from this latest work is said to be "very large" for RadeonSI with SPECViewPerf in particular. The option to use CPU storage uploads for the threaded context is said to be as much as 2x or more the performance of the existing behavior.