All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=828060

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=088f8356edf10b69726d2e816af181d06f408a5b6af946be08089ce926485a70


Bot(s) for this bug's original alert(s):

chromium-webrtc-rel-mac

There are more improvements in linux that are not picked up by the alert somehow. It might be related to the range of results, and I will look at it.
https://chromeperf.appspot.com/report?sid=f470e1b87c985928342b8310babf8cbf08b8b3c32c7572d6cb1fb1df71302b83

Overall, these are the outcomes:
- This only affects SW decoded output.
- In the platforms where there is GMB support(linux-mac), we cut down display latency by ~40%.
- It will be automatically applicable for future GMB supports, i.e. if GMB is supported in Win. dcastagna@ do you know if there is any timeline?

Native GMBs where we can map them on the render and use them as GL images is not going to happen.
We can use shared memory GMBs on windows though. Did you check if that improves latency? Last time Dale tried to enable them it was not a clear improvement.

Sure, I can test that. As long as it is better/faster than VideoResourceUpdater, it is a win for us. Should I test locally, or just push a CL/revert and observe the bots?

Can you sync with Dale? Since he tried already and decided not to switch the flag at the time.

dalecurtis@ can you PTAL at the discussion?

It's worth trying separately for WMPMS, I don't think the results would be independent of what we saw with <video> given the different pipelines. I'll also see about flipping the experiment for the av analysis lab again with my recent improvements to see if anythings better.

I tested on a Lenovo P50 through the existing browser_test. CreateForSoftwarePlanes() takes ~2.5ms on average and ~4.2ms max. GMBVFP takes ~0.5ms on average and ~1ms max. It looks like a clear win as we are interested in reducing the time to display. Note that CreateForSoftwarePlanes() is terrible on worse systems, i.e. this bot takes 11ms on average and 30 ms max: https://chromeperf.appspot.com/report?sid=6a42eee394c4bf35579313391eaab26f97cad0d890c2bd51f3757054b029da15

I think it is worth going forward with this on WMPMS. I can follow it in a new bug if you agree. Which test are you interested in for <video>? I can take a look at that too, unless you want to wait for av analysis lab.

Deleted: gmb-for-realtime comparison 17.5 KB

gmb-for-realtime comparison 17.5 KB View Download

I spinned off the Win GMB discussion to a new issue: https://bugs.chromium.org/p/chromium/issues/detail?id=829524