Skip to contentSkip to conversation

Performance regression, please revert commit 441ef916ae6569c88b3d6abaf7fea4d69be49d76

DISCUSSION Knut Petersen 2 participants 5 messages Jan 13, 2013

lore
Share & tools
lore
#1 lore raw
In reply to Knut Petersen
Quoted 2 lines · Knut Petersen

There is a massive performance regression on i915GM hardware, opening/closing/redrawing looks as if it is done in slow motion.

gen2/3 benefits from it after applying the bugfix to libdrm. -Chris

Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
#0 lore raw

There is a massive performance regression on i915GM hardware, opening/closing/redrawing looks as if it is done in slow motion.

I bisected the problem to:

commit 441ef916ae6569c88b3d6abaf7fea4d69be49d76

Author: Chris Wilson <[email protected]>
Date:   Thu Jan 10 19:14:21 2013 +0000

intel: Throttle harder

Filling the rings is a very unpleasant user experience, so cap the number of batches we allow to be inflight at any one time.

Interestingly, as also found with SNA, throttling can improve performance by reducing RSS. However, typically throughput is improved (at the expense of latency) by oversubscribing work to the GPU and a 10-20% slowdown is commonplace for cairo-traces. Notably, x11perf is less affected and in particular application level benchmarks show no change.

Note that this exposes another bug in libdrm-intel 2.4.40 on gen2/3.

Please revert 441ef (or limit the new code to hardware that benefits from it)

cu,
Knut

#2 lore raw
In reply to Chris Wilson
Quoted 6 lines · Chris Wilson
Quoted 2 lines · Knut Petersen

There is a massive performance regression on i915GM hardware, opening/closing/redrawing looks as if it is done in slow motion.

gen2/3 benefits from it after applying the bugfix to libdrm. -Chris

Which bugfix?

cu,
Knut

#3 lore raw
Quoted 8 lines · Knut Petersen
Quoted 6 lines · Chris Wilson
Quoted 2 lines · Knut Petersen

There is a massive performance regression on i915GM hardware, opening/closing/redrawing looks as if it is done in slow motion.

gen2/3 benefits from it after applying the bugfix to libdrm. -Chris

Which bugfix?

https://patchwork.kernel.org/patch/1963391/

commit fdda97007b1dbf95beb16a0e3510fd36c89e8c33

Author: Chris Wilson <[email protected]>
Date:   Fri Jan 11 00:55:12 2013 +0000

intel: Remove the fence count contributions when clearing relocs
-Chris

Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
#4 lore raw
Quoted 9 lines · Chris Wilson
Quoted 8 lines · Knut Petersen
Quoted 6 lines · Chris Wilson
Quoted 2 lines · Knut Petersen

There is a massive performance regression on i915GM hardware, opening/closing/redrawing looks as if it is done in slow motion.

gen2/3 benefits from it after applying the bugfix to libdrm. -Chris

Which bugfix?

Quoted 9 lines

https://patchwork.kernel.org/patch/1963391/

commit fdda97007b1dbf95beb16a0e3510fd36c89e8c33

Author: Chris Wilson <[email protected]>
Date:   Fri Jan 11 00:55:12 2013 +0000

intel: Remove the fence count contributions when clearing relocs
-Chris

Thanks, that indeed helps.

cu,
Knut