anv: set ComputeMode.PixelAsyncComputeThreadLimit = 4
authorFelix DeGrood <felix.j.degrood@intel.com>
Wed, 13 Sep 2023 20:56:59 +0000 (20:56 +0000)
committerMarge Bot <emma+marge@anholt.net>
Tue, 17 Oct 2023 18:09:29 +0000 (18:09 +0000)
commitb561bcd78c300bbd27ece05beb98320233c351be
tree6aa5e5c005faef51a459fde65a1f12718e0d669d
parent8ff4847b644d5485edee15504970ad0afe81f290
anv: set ComputeMode.PixelAsyncComputeThreadLimit = 4

Heuristic-based optimization throttling CCS work (async compute).
Without throttling, background compute work consumes all threads,
deminishing performance gains by running dispatch in parallel with
3D work.

Optimization is heuristics based, meaning a workload might slow
down when using async compute.

Best value: PixelAsyncComputeThreadLimit = 4. On DG2, this
equates to a max CCS thread occupancy of 37.5%.

Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25508>
src/intel/vulkan/genX_init_state.c