Closed Bug 873244 Opened 12 years ago Closed 11 years ago

enyo.js sampler "bouncing balls" demo fps regressed from 50+ fps in 1.0.1 to 5 fps in 1.2

Categories

(Firefox OS Graveyard :: General, defect, P3)

ARM
Gonk (Firefox OS)

Tracking

(blocking-b2g:-, firefox24 affected, b2g18 unaffected, b2g18-v1.0.1 unaffected)

RESOLVED WORKSFORME
1.4 S6 (25apr)
blocking-b2g -
Tracking Status
firefox24 --- affected
b2g18 --- unaffected
b2g18-v1.0.1 --- unaffected

People

(Reporter: bkelly, Unassigned)

Details

(Keywords: perf, regression, Whiteboard: [c= p= s=2014.04.25 u=])

The enyo.js framework has a Sampler app that demonstrates many of its various components. Recently they added support for FFOS as discussed here: https://hacks.mozilla.org/2013/03/firefox-os-support-lands-in-enyojs-2-2/ Unfortunately, one of the canvas demos crashes the b2g executable on a relatively recent master/m-c. (I can adb shell into the phone, but ps does not list b2g any more.) On a fresh v1-train, the app runs fine. To reproduce: 1) Install the enyo sampler app by going here: http://enyo2-sampler.canuckcoding.ca/ 2) Open the app and navigate to Canvas->Bouncing Balls 3) Let it run for around 30 seconds to a minute. My unagi crashed reliably within 15 seconds or so. My Buri got quite janky and then crashed closer to a minute. I tried looking at procrank as it crashed on my Buri: APPLICATION PID Vss Rss Pss Uss cmdline b2g 143 136824K 126824K 83484K 40212K /system/b2g/b2g Enyo 2 Sampler 481 186876K 102872K 59621K 16444K /system/b2g/plugin-container Homescreen 384 13648K 13648K 13516K 13456K /system/b2g/plugin-container ------ ------ ------ 164233K 77432K TOTAL
Can you get a crash report URL?
As far as I can tell there are no crash reports on the device. The screen first freezes and then goes black. It does not automatically restart. I can adb shell in, but b2g-ps does not show any processes.
blocking-b2g: --- → koi?
Severity: normal → critical
Keywords: crash, stackwanted
Whiteboard: [b2g-crash]
(In reply to Scoobidiver from comment #3) > To get crash reports, see > https://wiki.mozilla.org/B2G/QA/ > Tips_And_Tricks#Getting_crashes_off_the_Device At the time I filed this I tried getting a crash report using that method, but none was available. I believe the b2g process is getting OOM killed. Should a crash report be generated in that case?
Flags: needinfo?(bkelly)
That's a good question: Do OOMs cause crash reports to be generated? I'd be kind of surprised, because in normal usage of the phone, background apps get killed by OOM all the time. Anyway for tips on investigate OOMs, see https://wiki.mozilla.org/B2G/Debugging_OOMs.
I just tried to reproduce on my dev Buri running m-c at 134166:12cdc8931e48. It appears that the b2g process no longer crashes, but the app runs extremely slowly. For example, my keon on 1.0.1 gets 50+ frames per second. On this Buri I get around 5 fps and the physics calculations are messed up. It seems possible some of the slowness is due to this version of m-c having hwc disabled. (I am specifically avoiding hwc at the moment due to bug 881970.) I no longer have an unagi to test with. Does anyone else have another device on m-c to test? Also, if this is just a slowness issue and the crash is fixed in recent m-c, should we close this and open a new bug? Or repurpose this bug?
re-nom if needed
blocking-b2g: koi? → -
Flags: needinfo?(mlee)
Summary: enyo.js sampler "bouncing balls" demo crashes b2g on master → enyo.js sampler "bouncing balls" demo fps regressed from 50+ fps in 1.0.1 to 5 fps in 1.2
Whiteboard: [b2g-crash] → [c= p= s= u=]
Flags: needinfo?(mlee)
Severity: critical → major
Priority: -- → P3
Ben - Do you know if this issue still reproduces on current Firefox OS versions?
Flags: needinfo?(bkelly)
This seems to be running at 50fps on my v1.4 open-c now. I suspect we fixed this while dealing with various game bugs.
Status: NEW → RESOLVED
Closed: 11 years ago
Flags: needinfo?(bkelly)
Resolution: --- → WORKSFORME
Whiteboard: [c= p= s= u=] → [c= p= s=2014.04.25 u=]
Target Milestone: --- → 1.4 S6 (25apr)
You need to log in before you can comment on or make changes to this bug.