[Webkit-unassigned] [Bug 43903] Speeding up SVG filters with multicore (SMP) support

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=43903





--- Comment #26 from Gabor Loki <loki at webkit.org>  2011-03-09 07:41:32 PST ---
Created an attachment (id=85174)
 --> (https://bugs.webkit.org/attachment.cgi?id=85174&action=review)
Measurement result with Methanol

I have measured the performance progression of this SMP framework with our page loading benchmark (called Methanol: http://gitorious.org/methanol).

Two measurement scenarios were used to validate the framework (an artificial test-suite and a collection of real websites).

Scenario 1:
 - Real websites:
   - Contains the first 25 unique websites from Alexa Top 500.
   - Does not contain any SVG.
 - Result
   - Measured on x86 and on ARM
   - It is not surprising that there is no difference between the reference and the two SMP-based approaches.

Scenario 2:
 - Artificial test-suite:
   - Methanol contains a small test-suite for SVG. Some test cases contain only one SVG filter with simple shapes, others are using combined filters.
 - Result
   - Measured on x86 and ARM
   - Methanol reported 30% - 80% performance progression on turbulence-related tests. Others were not changed. See the attached file.
 - An important though
   - The 'turbSmall' test contains several small elements. The turbulence filter is called on these elements independently. The drawing area is for each elements is 100x100 (which is under the initial SMP-limit for turbulence). You can see the performance numbers were the same for each measurements. While the 'turbBig' test performs better with SMP (in this case the elements' size are 400x400).

If someone has another SVG-related benchmark, I would be happy to measure the SMP framework with it.

-- 
Configure bugmail: https://bugs.webkit.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.