Poor performance on crazy DOM raytracer

• Component: DOM
• Status: RESOLVED FIXED

[David Smith]

WebKit loses to Opera by a very large margin on the linked benchmark.

[David Smith]

Created attachment 16205 [details]
Speeds up the "full render" from >400 seconds to ~45 seconds for me

Thanks to everyone on irc for their suggestions and prodding :) Particularly the suggestion of using ListHashSet from Maciej.

[David Smith]

Created attachment 16206 [details]
Fix my email address, and remove a stray formatting change

Minor tweaks, sorry for the spam.

[Maciej Stachowiak]

This seems like a gratuitous change:

-        for ( ; (r = it.current()); ++it) {
+        for ( ; (r = it.current()); ++it ) {

There's no need to use new here, you can just make the Vector a local variable:

+    Vector<RenderObject*>* deadObjects = new Vector<RenderObject*>();

In fact, it might be worthwhile giving it an inline capacity to handle the common case effeciently. Something like:

Vector<RenderObject*, 16> deadObjects;

r- to address these points.

[David Smith]

Created attachment 16229 [details]
Addresses review comments

>This seems like a gratuitous change:
>
>-        for ( ; (r = it.current()); ++it) {
>+        for ( ; (r = it.current()); ++it ) {

Oops, leftover from a find & replace style fixup. Fixed.

>There's no need to use new here, you can just make the Vector a local variable:
>
>+    Vector<RenderObject*>* deadObjects = new Vector<RenderObject*>();
>
>In fact, it might be worthwhile giving it an inline capacity to handle the
>common case effeciently. Something like:
>
>Vector<RenderObject*, 16> deadObjects;

Fixed.

[Maciej Stachowiak]

Comment on attachment 16229 [details]
Addresses review comments

r=me

[Oliver Hunt]

<rdar://problem/5470512>

[David Smith]

Fixed in r25464