1439760 - http://makeyourmoneymatter.org/ has performance issues with blob image invalidation

Status: RESOLVED FIXED

(Core :: Graphics: WebRender, enhancement, P1)

Description

[Jeff Muizelaar [:jrmuizel]]

We mostly seem to be getting hurt with allocations.

Comment 1

[Jeff Muizelaar [:jrmuizel]]

https://perfht.ml/2omNrRr

Comment 2

[Jeff Muizelaar [:jrmuizel]]

vs https://perfht.ml/2onhEQE

Comment 3

[Jeff Muizelaar [:jrmuizel]]

So there's a big svg background image. We seem to draw this to a similar surface that's very big. We do this for every tile. This makes much sad.

Also, clipping makes things much worse.

Comment 4

[Jeff Muizelaar [:jrmuizel]]

gfxCallbackDrawable::MakeSurfaceDrawable is the caller to CreateSimilarDrawTarget

Comment 5

[Jeff Muizelaar [:jrmuizel]]

We're repeating because of 

   161 	    if ((IsRepeatingExtendMode(aExtendMode) || aOpacity != 1.0 || aContext->CurrentOp() != CompositionOp::OP_OVER) &&
   162 	        !mSurfaceDrawable) {
-> 163 	        mSurfaceDrawable = MakeSurfaceDrawable(aContext, aSamplingFilter);
   164 	    }

Comment 6

[Jeff Muizelaar [:jrmuizel]]

This has been mostly fixed. The remaining problems are mostly caused by bug 1456558

Comment 7

[Andrew Osmond [:aosmond] (he/him)]

(In reply to Jeff Muizelaar [:jrmuizel] from comment #6)
> This has been mostly fixed. The remaining problems are mostly caused by bug
> 1456558

Seems to be performing well. I can probably improve the trimming of surfaces after the fact as it seems if you start with a large size, and scroll to something smaller, it kicks into factor of 2 mode but won't aggressively discard the old large surfaces (because we don't have a factor of 2 replacement...hmmm). But that is a temporary waste of memory that 60 seconds will resolve when they expire out of the cache.