New Marlin-renderer releases 0.4.4 & 0.5.4

[bourges.laurent]

New Marlin-renderer releases are available:

• Marlin 0.4.4: https://github.com/bourgesl/marlin-renderer/releases/tag/v0.4.4
  
• Marlin 0.5.4: https://github.com/bourgesl/marlin-renderer/releases/tag/v0.5.4
  

Marlin 0.5.4 is using sun.misc.Unsafe to use off-heap storage for edge data: it was intensively tested and is considered stable: ~ 5 to 10% faster on JDK 1.8 (no gain on jdk7).

Marlin 0.5.4 is then recommended for JDK 1.8 only; for JDK 1.7, please use Marlin 0.4.4 !

Changes:

• for large crossing / edge pointer arrays (>=75), use to an optimized merge sort for newly added edges (>=10) + final merge pass using auxiliary storage
• refactoring to move statistics into RendererStats class (to avoid allocating them when doStats or doMonitors are disabled)
• improved complex affine transformer allocations (tx, delta and inverse delta p2dconsumer)
• improved createStrokedShape() by using a custom path2d (preallocated large path2D to avoid useless array resizes + efficient cloning)

Your feedback is very important !
Please tell us if your appreciate marlin and if it fits well your workload, use cases...

Just post your comments on the forum:
https://groups.google.com/forum/#!topic/marlin-renderer/biSbzplNlOc

Enjoy !

Laurent B.

[Laurent Bourgès]

Here are the new benchmark results:
https://github.com/bourgesl/marlin-renderer/wiki/Benchmarks

All benchmark were made using Oracle JDK1.8.0_b05 on my fast machine (core-i5) using MapBench (default & shared profiles).

"Marlin 0.5.4 is 17% to 25% faster than marlin 0.3 and is as fast as Oracle ductus !"

Laurent B.