130027 - [regression] When moving mouse-cursor CPU pegs to 100%

Status: VERIFIED FIXED

(Core Graveyard :: GFX, defect, P1)

Description

[Markus Hübner]

When moving the mouse-cursor Mozilla is having major problems - it's quite 
hard to navigate at all.
Using build 2002030908 on win-xp, 1.1ghz, 327RAM

Comment 1

[R.K.Aa.]

related to bug 93620?

Comment 2

[Peter Nilsson]

www.landor.com worked fine on 20020306 but after installing 20020310 the
performance was really bad.

Comment 3

[Peter Nilsson]

I've made some simple testing and I think it has something to do with the
background image.

Without the background there are no performance problems.

Comment 4

[Markus Hübner]

Probably http://www.go-mono.com/class-status-System.Xml.html is related to 
this regression too.

Comment 5

[Markus Hübner]

The regression starts with 2002030803 (build 2002030703 is fine).

Comment 6

[Arthur]

Sorry for the spam, was to eagerly clicking on submit at the mid-air warning.

Comment 7

[Arthur]

Bug 130055 looks similar.

Comment 8

[Markus Hübner]

No problems with 0.9.9 build 2002031104

Comment 9

[Christine Hoffman]

Changing QA contact

Comment 10

[Markus Hübner]

http://www.scea.com/ is another really worse example.

Comment 11

[Markus Hübner]

This problem does not shop up on Linux.

Another thing I discovered: on http://www.world-direct.com/step2 the problem 
appears if you move the mouse-cursor over the products images or the links 
below the images. If I remove the CSS file it is working fine.

Comment 12

[Markus Hübner]

Happening on so many sites ... this really needs some investigation & a 
profile :)

Comment 13

[Markus Hübner]

Further investigations showed that the regression happened between 2002030710 
and 2002030803.

Comment 14

[Markus Hübner]

Think this regression could have been caused by  
http://bugzilla.mozilla.org/show_bug.cgi?id=117376

Comment 15

[dcone (gone)]

I have a 1.4 ghz machine, 500 meg memory.  I dont see any slowness at all with
http://www.landor.com/ or http://www.landor.com/.  All that the fix for
http://bugzilla.mozilla.org/show_bug.cgi?id=117376 did was allow more cases for
the PatBlt to be used when drawing a tiled background.  If the PatBlt was not
used.. it would actually.. or should actually be slower.. and it would show up
when you scrolled a page.  I will try this bug out later.. probably on monday on
a slower machine I have at work.. if I do see the slowdown.. then I will just
back out that change (its just one line of code) and see if its the cause..
although I don't see how it could be at the moment.

Comment 16

[karnaze (gone)]

I'm seeing the problem on my win2k home system, marking nsbeta1+. It is slow 
when mousing in landor or dragging windows over it. Reassigning to dcone after 
talking to him and putting back the code before his change in bug 117376. This 
is the code we changed:

if ((mAlphaDepth>8) || ((mAlphaDepth==8)&&tryAgain) ||  
(canRaster==DT_RASPRINTER || (256==mNumPaletteColors) || 
(imageScaledWidth>MAX_BUFFER_WIDTH) || (imageScaledHeight>MAX_BUFFER_HEIGHT)) {

Comment 17

[dcone (gone)]

Attachment: Patch to limit the use of PatBlt

Comment 18

[dcone (gone)]

Attachment: last patch was wrong one.. this is correct.

Comment 19

[dcone (gone)]

Attachment: Updated patch.. some merge conflicts.. this fixes that.

Comment 20

[karnaze (gone)]

With the latest patch, I'm not seeing the problems I reported in comment #16.

Comment 21

[Kevin McCluskey (gone)]

*** Bug 130228 has been marked as a duplicate of this bug. ***

Comment 22

[Marc Attinasi]

Comment on attachment 74765 [details] [diff] [review]
Updated patch.. some merge conflicts.. this fixes that.

Can this:
 if(PR_TRUE != gIsWinNT){
become this:
 if(!gIsWinNT){
pretty please???

Also, the ALL CAPS IS OBNOXIOUS

Finally, I don't think you need the useSlow variable since it is always PR_TRUE
if you hit the spot where you test for it (if not useSlow you do the PatBltTile
call and return). A comment at the if(useSlow) location is good enough (unless
I am missing something due to the context-diff). If it is for readability, that
is OK too.

Please consider these comments, and you get the coveted sr=attinasi

Comment 23

[karnaze (gone)]

Comment on attachment 74765 [details] [diff] [review]
Updated patch.. some merge conflicts.. this fixes that.

r=karnaze. Minor nit - either always have space before and after operators <,
&& etc. or not.

Comment 24

[dcone (gone)]

I fixed the spacing issues.. and the 

if(PR_TRUE != gIsWinNT){
TO
if(gIsWinNT)

I left the useSlow for readability.. next version I may use a fuction for this.

Comment 25

[Scott Collins]

Comment on attachment 74765 [details] [diff] [review]
Updated patch.. some merge conflicts.. this fixes that.

a=scc

Comment 26

[Jacek Chmiel]

Build 20020319 - it seems to be even worse on http://www.devx.com than before
(much slower, less than 1 fps refresh rate). (Windows XP Pro, Celeron 1Ghz,
512MB RAM). Hope to see working fix soon.

Comment 27

[dcone (gone)]

fixed.. 

Comment 28

[Jacek Chmiel]

Build 20020320 - it works well! Thanks. But... could you take a look at
www.ntcompatible.com? Image on the left is not in sync with text when scrolling.
Is this related to this bug or should I post it as another one?

Comment 29

[Henrik Gemal]

I'm seeing the same thing in Mail&News with HTML mails with background images....

Comment 30

[Moied]

verified fixed with build ID 20020404 using WINXP