[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 #10 from Gabor Loki <loki at webkit.org>  2011-02-24 00:19:50 PST ---
> If I understant correctly, you are doing research. So getting accurate numbers is probably interesting for your papers as well.

I would be the most happiest people if this topic was a research.

> When we considered parallelization for Qt's CPU rasterization, we looked for the boundaries of input for which SMP is helpful VS is harmful. I think this kind of analysis in both interesting for the WebKit project, and for research. (and note that we ended up not using parallelization for most primitives of Qt because it was more harmful than helpful for general rendering).

I agree with you about the importance of boundaries. That's why I left the s_minimalRectDimension for platforms. All other code depends on the number of requested threads. So, if the platform/OS/architecture is able to decide its boundaries this will be an easy task to implement their needs. Of course lots of benchmarking are needed to achieve this, but for introducing this parallelization approach the current limit is fine. Fine tuning of parallel turbulence filter should be done later with the help of benchmarks.

Anyway I will check Qt's benchmark environment. Is the time measuring method of this environment based on clock_t or wall-time (for example: clock_gettime)?

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