552142 - Pixman fast-scale branch performance improvements

Status: RESOLVED DUPLICATE of bug 562087

(Core :: Graphics, defect)

Description

[Doug Turner (:dougt)]

see http://lists.freedesktop.org/archives/pixman/2010-March/000086.html

I think we will want this for fennec as soon as possible.

Comment 1

[Oleg Romashin (:romaxa)]

Attachment: Patch against m-c

Comment 2

[Oleg Romashin (:romaxa)]

Attachment: Add missing file

Comment 3

[Oleg Romashin (:romaxa)]

Attachment: oprofile scale_bilinear_no_patch

Comment 4

[Oleg Romashin (:romaxa)]

Attachment: oprofile scale_nearest_no_patch

Comment 5

[Oleg Romashin (:romaxa)]

Attachment: oprofile scale_bilinear_with_patch

Comment 6

[Oleg Romashin (:romaxa)]

Attachment: oprofile scale_nearest_with_patch

Comment 7

[Oleg Romashin (:romaxa)]

Tested with fennec zoom animation patch
According to oprofile results with this patch we are ~2.5x faster.

Also we should set 
ctx.mozImageSmoothingEnabled = false;
to avoid bilinear scaling and use nearest scaling.

Comment 8

[Oleg Romashin (:romaxa)]

In 16bpp path with fennec animated zooming we are using
fast_composite_scale_nearest_src_0565_0565
function.

And yep, now it looks like animated zooming (before it was just too slow that animation was not visible)

Comment 9

[Jeff Muizelaar [:jrmuizel]]

I'd rather take this from upstream. Any good reason we can't do that?

Comment 10

[Siarhei Siamashka]

> I'd rather take this from upstream.

Me too.

> Any good reason we can't do that?

We can, but it just may take a bit longer when it comes through upstream. Fortunately in this particular case, Alexander Larsson is also from redhat, so I guess this issue got a nice priority boost :)

Comment 11

[Siarhei Siamashka]

Fast nearest scaling optimizations have been released in pixman 0.18.0

By the way, could anybody have a look what's wrong with firefox and why it does not want to use normal repeat for tiling in that forum template?
http://lists.cairographics.org/archives/cairo/2010-April/019650.html

Comment 12

[Jeff Muizelaar [:jrmuizel]]

(In reply to comment #11)
> By the way, could anybody have a look what's wrong with firefox and why it does
> not want to use normal repeat for tiling in that forum template?
> http://lists.cairographics.org/archives/cairo/2010-April/019650.html

My guess is that midori doesn't blend between tiles, where as firefox does. Are the tiles zoomed?

Comment 13

[Siarhei Siamashka]

Yes, the tiles are zoomed. I just remember that firefox also used repeat in this case earlier and was a bit surprised. Is the blending really necessary? How does it affect performance?

Comment 14

[Oleg Romashin (:romaxa)]

I think we should update pximan to latest upstream with zoom fixes.

Comment 15

[Jeff Muizelaar [:jrmuizel]]

(In reply to comment #13)
> Yes, the tiles are zoomed. I just remember that firefox also used repeat in
> this case earlier and was a bit surprised. Is the blending really necessary?

The blended results are what designers likely expect. It's certainly more elegant.

> How does it affect performance?

It can have a very negative impact on performance, and we are consider removing it...

Comment 16

[Jeff Muizelaar [:jrmuizel]]

This should be taken care of by the next pixman update.