[SoC] QGIS on steroids - week #2

[Martin Dobias]

Hi Wolf

On Tue, Jun 8, 2010 at 12:25 AM, Wolf Bergenheim
<wolf+grass at bergenheim.net> wrote:
> On Tue, Jun 8, 2010 at 01:10, Martin Dobias <wonder.sk at gmail.com> wrote:
>> Hi all,
>>
>> This week I've implemented the mechanism to render map layers in
>> separate threads, so that multiple processors/cores can be used for
>> the rendering. On my dual-core test machine, I was able to get close
>> to the ideal 2x speed up.
>
> Nice! I'll have to try that at work, where I have a dual core. Great
> going so far.

I would welcome any testing! The code is still sharp around the edges,
though once I'll make threading as a configurable option, it should be
ready for testing.

>>
>> For the next week, I plan to improve the rendering and caching code.
>> Then I'd like to start with next target which will take few weeks - to
>> enable the rendering to happen in worker thread and GUI will not be
>> blocked while rendering.
>
> That will make all the difference in the world. I remember the pain of
> having to wait for the ui to become responsive, especially with large
> vector maps with loads of features. Do you have any idea how this will
> affect the labeling you worked on last year? I remember you saying
> that it was slower than expected...

The labeling from the last year will not be affected much. Care must
be taken when registering the features from individual layers to avoid
run conditions, I haven't checked that yet.

Regarding the speed of the labeling, it is slow only under some
conditions, most importantly when labeling polygons with free
placement - the evaluation of label candidates is quite costly and
uses naive implementation in some parts. The core algorithms used for
label placement are fast.

Regards
Martin