Bug 5526 – SVGs with filters look grainy when scaled

• Component: SVG
• Status: RESOLVED DUPLICATE

[Eric Seidel (no email)]

SVGs with filters look grainy when scaled

Currently we apply filters according to the abstract coordinate system of the graphic, NOT the device 
coordinate system (i.e. the final pixel coords when drawn to the screen).  This can be *bad* for a couple 
reasons.

1.  When you scale the resulting pixels, the result looks grainy.
2.  You can end up rendering 6 billion pixels and only using 1% of that if that part of the image is 
scaled down.

A couple ways to solve this:
1.  Just figure out what the final CTM for the item/group in question and render only the necessary 
number of pixels.
2.  (Better!) *delay* compositing of the result image.  Instead pass the CIImage back to the parent 
container (the one calling draw()) and let it do the compositing (or apply it's own filters using that 
CIImage as an input).  When you finally have to do the compositing, figure out the absolute CTM, etc.

I guess 2. is really just an optimized version of 1.  In either case, this will need to be solved.

[Eric Seidel (no email)]

Erick said he would take a crack at this.

[Dirk Schulze]

If I understand your comment right, we do this right now. I scaled the feGaussianBlur and feColorMatrix of the W3C test suite and i don't see pixels. But maybe this is a result of antialiasing?

[Dirk Schulze]

Committed r69181: <http://trac.webkit.org/changeset/69181>

[Csaba Osztrogonác]

(In reply to comment #3)
> Committed r69181: <http://trac.webkit.org/changeset/69181>

It broke 2 tests on Qt bot:
http://build.webkit.org/results/Qt%20Linux%20Release/r69181%20(21569)/svg/filters/shadow-on-filter-pretty-diff.html
http://build.webkit.org/results/Qt%20Linux%20Release/r69181%20(21569)/svg/filters/shadow-on-rect-with-filter-pretty-diff.html

[Dirk Schulze]

Committed r69182: <http://trac.webkit.org/changeset/69182>

[Dirk Schulze]

Committed r69187: <http://trac.webkit.org/changeset/69187>

[Dirk Schulze]

This bug was fixed with the patch on bug 31370. Marking this bug as duplication of 31370 because fixes for the patch mention this bug. Sorry for that.

*** This bug has been marked as a duplicate of bug 31370 ***

[WebKit Review Bot]

http://trac.webkit.org/changeset/69187 might have broken GTK Linux 32-bit Debug

[WebKit Review Bot]

http://trac.webkit.org/changeset/69188 might have broken GTK Linux 32-bit Debug