Use several threads for parallel map tile rendering on multi-core CPUs

[mapsforge]

From thilo.mu...@gmail.com on January 23, 2011 18:04:07

As of today, (Android) smartphones have single core CPUs. But multi-core CPUs will be standard already in the near future. The LG Optimus 2x with its Nvidia Tegra 2 processor is one example, ready to be shipped in the next month.

To make better use of this additional CPU power, map rendering must be parallelised. The number of rendering threads should equal the number of available cores. Implementing this will probably be easy as the map rendering already works asynchronously and all critical data structures are synchronized. Doubling the number of worker threads will not speed up the rendering of a single map tile but double the output of map tiles per time unit.

Original issue: http://code.google.com/p/mapsforge/issues/detail?id=73

[mapsforge]

From thilo.mu...@gmail.com on September 17, 2011 05:31:41

Labels: -Priority-Low Priority-Medium

[mapsforge]

From thilo.mu...@gmail.com on November 13, 2011 03:39:20

Status: Started

[mapsforge]

From thilo.mu...@gmail.com on December 23, 2011 13:43:23

The latest SVN version of the map library is prepared for this feature. All data data structures like the job queue and the tile caches are separated from each other and should be thread-safe. However, in order to use several parallel rendering threads, we need to synchronize during the labeling as this step must be done in a serialized manner.

[ludwigb]

This is work in progress in dev 29fd89a

[devemux86]

I'm closing this, as the feature is considered implemented.

For problems see #591 or open a new issue.