[Intel-gfx] [PATCH v2 3/6] drm/i915/guc: use a separate GuC client for each engine

[Chris Wilson]

On Fri, Jul 22, 2016 at 11:23:11AM +0100, Dave Gordon wrote:
> Does this patchset change the results in the parallel-submission nop test?

No.

Using Execlists submission
Maximum execution latency on render, 3.392us, total 7.794us per cycle
All (4 engines): 8,257,536 cycles, average 1.211us per cycle

default: 5,789,696 cycles: 3.455us
render: 5,748,736 cycles: 3.479us
bsd: 13,406,208 cycles: 1.492us
blt: 13,852,672 cycles: 1.444us
vebox: 13,680,640 cycles: 1.462us

Using GuC submission (v6.1)
Maximum execution latency on render, 3.674us, total 9.089us per cycle
All (4 engines): 577,536 cycles, average 17.441us per cycle

default: 5,399,552 cycles: 3.704us
render: 5,341,184 cycles: 3.745us
bsd: 13,060,096 cycles: 1.531us
blt: 8,828,928 cycles: 2.265us
vebox: 12,945,408 cycles: 1.545us


-- 
Chris Wilson, Intel Open Source Technology Centre