Bug 1480293 should hopefully stick this time, but my testing of those builds on Windows indicates that the reporting doesn't work when the GPU process is enabled. I haven't been able to debug it, and I'm out of time for the week. But my suspicion is that we're bailing out here: https://hg.mozilla.org/integration/mozilla-inbound/rev/91c7687d381bbd5d5ac662ef5feefe965abfd60e#l7.94 My original patch did the reporting in the parent (non-GPU process), which worked, but billed the memory usage to the wrong process heap. So I think we have two options here: * Keep the reporter in the same process as the RenderThread, and find a way to send the message appropriately when we're in the GPU process. * Move the reporter back to the parent process' gfxPlatform::Init, and bill it to the correct process with [1]. We'll want to make sure that we only do this when the GPU process is enabled, and also that we only run the reporter when WR is enabled, to avoid the crashes that caused the last backout. [1] https://searchfox.org/mozilla-central/source/xpcom/base/nsIMemoryReporter.idl#126

Matt, if you have time on Monday, would you mind taking this? It should be fast with a local windows dev environment (which I don't have), and it would be good to get it landed a day sooner.

I ended up getting an extra moment and taking a crack at option 2: https://treeherder.mozilla.org/#/jobs?repo=try&revision=5f87b0d3e060e1ec4f484417a53d0fafdaf5b829

Try build works for me - With css francine open: GPU (pid 12576) Explicit Allocations 198.32 MB (100.0%) -- explicit ├──132.36 MB (66.74%) -- gfx/webrender │ ├──132.06 MB (66.59%) -- gpu-cache │ │ ├──128.00 MB (64.54%) ── cpu-mirror │ │ └────4.06 MB (02.05%) ── metadata │ └────0.30 MB (00.15%) -- (4 tiny) │ ├──0.23 MB (00.12%) ── primitive-stores │ ├──0.04 MB (00.02%) ── render-tasks │ ├──0.02 MB (00.01%) ── clip-stores │ └──0.01 MB (00.00%) ── hit-testers ├───44.40 MB (22.39%) ── heap-unclassified ├───14.83 MB (07.48%) -- threads │ ├──14.38 MB (07.25%) -- stacks │ │ ├───2.00 MB (01.01%) ── WRRenderBackend#1 (tid=16476) │ │ ├───2.00 MB (01.01%) ── WRSceneBuilder#1 (tid=3116) │ │ ├───2.00 MB (01.01%) ── WRSceneBuilderLP#1 (tid=12316) │ │ ├───2.00 MB (01.01%) ── WRWorker#0 (tid=14244) │ │ ├───2.00 MB (01.01%) ── WRWorker#1 (tid=12124) │ │ ├───2.00 MB (01.01%) ── WRWorker#2 (tid=18796) │ │ ├───2.00 MB (01.01%) ── WRWorker#3 (tid=16224) │ │ └───0.38 MB (00.19%) ++ (11 tiny) │ └───0.45 MB (00.23%) ++ overhead ├────6.44 MB (03.25%) -- heap-overhead │ ├──5.02 MB (02.53%) ── bin-unused │ └──1.43 MB (00.72%) ++ (2 tiny) └────0.29 MB (00.14%) ++ (5 tiny)

(In reply to samlh from comment #4) > Try build works for me - Great! Thanks. I'll upload the patches then.

This is how things looked when jrmuizel reviewed them. MozReview-Commit-ID: Gqp1tcNzsjC

Given that it's the weekend, flagging Matt and Jeff - whoever gets to it first can clear the reviews for the others.

MozReview-Commit-ID: Db9mz5jQKwG

Thanks Markus!

https://hg.mozilla.org/mozilla-central/rev/6368a1c271de https://hg.mozilla.org/mozilla-central/rev/5a423ac3eb23 https://hg.mozilla.org/mozilla-central/rev/6807afe94107