Roberto Agostino Vitillo (:rvitillo) Reporter
	Description • 11 years ago

Firefox is consuming an excessive amount of CPU resources while idling on Instagram compared to Chrome on Linux.

	Roberto Agostino Vitillo (:rvitillo) Reporter
	Comment 1 • 11 years ago

The issue seems to occur when the same image, declared more than once with different width and heights, is used as a background with a size of 100%.


Hottest stacktrace of FF while displaying the testcase:

skia::BGRAConvolve2D(unsigned char const*, int, bool, skia::ConvolutionFilter1D const&, skia::ConvolutionFilter1D)
skia::BGRAConvolve2D(unsigned char const*, int, bool, skia::ConvolutionFilter1D const&, skia::ConvolutionFilter1D const&, int, unsigned char*, bool)
skia::ImageOperations::ResizeBasic(SkBitmap const&, skia::ImageOperations::ResizeMethod, int, int, SkIRect const&, void*)
skia::ImageOperations::Resize(SkBitmap const&, skia::ImageOperations::ResizeMethod, int, int, SkIRect const&, void*)
skia::ImageOperations::Resize(SkBitmap const&, skia::ImageOperations::ResizeMethod, int, int, void*)
mozilla::gfx::Scale(unsigned char*, int, int, int, unsigned char*, int, int, int, mozilla::gfx::SurfaceFormat)
ScaleRunner::Run()
nsThread::ProcessNextEvent(bool, bool*)
NS_ProcessNextEvent(nsIThread*, bool)
mozilla::ipc::MessagePumpForNonMainThreads::Run(base::MessagePump::Delegate*)
MessageLoop::RunInternal()
MessageLoop::Run()
nsThread::ThreadFunc(void*)
_pt_root
start_thread

	Benoit Girard (:BenWa)
	Comment 2 • 11 years ago

Good find. Can you confirm that this background activity doesn't go away after a delay?

	Benoit Girard (:BenWa)
	Comment 3 • 11 years ago

Actually I guess your graph shows that even after 30 seconds we haven't reverted back to the idle state. This is 2 downscale of a 256x256 so it shouldn't take 30 seconds.

Milan this looks like a regression off background image scaling and looks serious.

Can we get a regression window for this bug? This will confirm any assumptions.

	Timothy Nikkel (:tnikkel)
	Comment 4 • 11 years ago

We only support one downscaled size per image, so I bet we just keep looping, downscaling to 40px, then downscaling to 20px, then downscaling to 40px...

	Milan Sreckovic [:milan] (needinfo for best results)
	Comment 5 • 11 years ago

So, the regression is traced all the way back to HQ image downscaling?

	Benoit Girard (:BenWa)
	Comment 6 • 11 years ago

(In reply to Timothy Nikkel (:tn) from comment #4)
> We only support one downscaled size per image, so I bet we just keep
> looping, downscaling to 40px, then downscaling to 20px, then downscaling to
> 40px...

I buy this theory.

(In reply to Milan Sreckovic [:milan] from comment #5)
> So, the regression is traced all the way back to HQ image downscaling?

Sounds like a guess. A regression window should take ~15 mins on this testcase so we should still run it.

	Timothy Nikkel (:tnikkel)
	Comment 7 • 11 years ago

(In reply to Benoit Girard (:BenWa) from comment #6)
> (In reply to Milan Sreckovic [:milan] from comment #5)
> > So, the regression is traced all the way back to HQ image downscaling?
> 
> Sounds like a guess. A regression window should take ~15 mins on this
> testcase so we should still run it.

Or flip the image.high_quality_downscaling.enabled pref.

	Benoit Girard (:BenWa)
	Comment 8 • 11 years ago

IMO with archived builds regression windows are more efficient then educated guesses, at least up to the Nightly window, but just my opinion ^-^.

	Alice0775 White
	Comment 9 • 11 years ago

Dupe Bug 927377 ?

	Vladan Djeric (:vladan)
	Comment 10 • 11 years ago

n-i for Milan to assign someone after March 17th

	David Baron :dbaron:
	Comment 11 • 11 years ago

I think seth might have some related plans here.

	Seth Fowler [:seth] [:s2h]
	Comment 12 • 11 years ago

Yup, bug 977459, which I just recently started working on, should take care of this problem, if comment 4 is accurate.

	Seth Fowler [:seth] [:s2h]
	Comment 13 • 10 years ago

I'm clearing the needinfo for Milan here. This will be resolved by bug 1060200, which I'm uploading a patch for now. (Bug 1060200 was part of bug 977459 but I've decided to split that bug into smaller bugs.)

	Seth Fowler [:seth] [:s2h]
	Comment 14 • 10 years ago

This was fixed by bug 1060200. I just checked, and the problem no longer appears.

	Joshua Mitchell (Inactive)
	Comment 15 • 10 years ago

Issue is resolved - clearing old keywords - qa-wanted clean-up