716462 - Performance regression with xBGR visual and no RENDER extension

Status: NEW

(Core :: Graphics, defect, P3)

Description

[Ginn Chen]

Firefox/Thunderbird 7 or later renders pretty slow on xBGR display and no RENDER extension.

The slow path is
0.   	0.   	gfxContext::Paint(double)
0.   	0.   	_moz_cairo_paint_with_alpha
0.   	0.   	_cairo_gstate_paint
0.   	0.   	_cairo_surface_paint
0.   	0.   	_cairo_surface_fallback_paint
0.   	0.   	_clip_and_composite_region
0.   	0.   	_cairo_surface_composite
0.   	0.   	_cairo_xlib_surface_composite
0.   	0.   	_cairo_xlib_surface_upload
0.   	0.   	_draw_image_surface
0.670	0.670	_XPutPixel32

and

0.   	0.   	gfxContext::Fill()
0.   	0.   	_moz_cairo_fill_preserve
0.   	0.   	_cairo_gstate_fill
0.   	0.   	_cairo_surface_fill
0.   	0.   	_cairo_surface_fallback_fill
0.   	0.   	_clip_and_composite_trapezoids
0.   	0.   	_clip_and_composite_region
0.   	0.   	_cairo_surface_composite
0.   	0.   	_cairo_xlib_surface_composite
0.   	0.   	_cairo_xlib_surface_upload
0.   	0.   	_draw_image_surface
0.680	0.680	_XPutPixel32

_cairo_xlib_surface_upload is new in cairo 1.10.
It seems it caused the regression.

Comment 1

[Ginn Chen]

Attachment: patch

Do not do upload if RGB masks don't match.
It will avoid falling into slow XPutPixel32 path.

The patch improves gfxContext::Fill() performance ~15 times with xBGR visual and no RENDER extension.
(although it is still ~ 10 times slower than the RENDER path.)

Comment 2

[Joe Drew (not getting mail)]

Comment on attachment 586909 [details] [diff] [review]
patch

Let's also accept a review by Jeff here. He can push it upstream if it works for him.

Comment 3

[u279076]

(In reply to Joe Drew (not getting mail) from comment #2)
> Comment on attachment 586909 [details] [diff] [review]
> patch
> 
> Let's also accept a review by Jeff here. He can push it upstream if it works
> for him.

Looks like this was missed - is it still valid?

Comment 4

[Petr Sumbera]

This is still being used for FF build on Solaris. Ginn, can you please comment whether this is still valid?

Comment 5

[BugBot [:suhaib / :marco/ :calixte]]

The bug assignee didn't login in Bugzilla in the last 7 months, so the assignee is being reset.