Decode Sub-region

[boxerab]

Building on the omnibus branch, this branch adds fast sub-tile decode.

If a decode region is specified using the -d parameter, all code blocks that do not contribute
to this region will be ignored, boosting decode speed considerably.

This is useful for decoding a sub region inside of a large single tile image.

Fixes #693 and #676

[boxerab]

Strange, all builds pass (ignore ABI test) except gcc 64 bit release, where a lot of tests fail.
Has anyone experienced this, and if so how to trouble shoot ? The only thing I can think of is that
debug build initializes pixel memory to zero, and release does not.

[mayeut]

@boxerab, could it be related to this warning:
openjpeg/src/lib/openjp2/region_mgr.c:194:1: warning: control reaches end of non-void function [-Wreturn-type]

[boxerab]

@mayeut nice catch - I left some debug code in one of my methods. Thanks.

[boxerab]

Well, just 8 broken tests to fix, and they are all related to a single image. Question: do we have any tests in the suite that test the -d flag : setting a decode region ? Also, could we add a test to test decoding a region from a specific tile? i.e. -d 0,0,64,64 -t 0

[mayeut]

There are tests using the -d option. Check in the file tests/nonregression/test_suite.ctest.inand look for -d. In fact, I think the newly broken tests are those using this option (file p0_04.j2k & p1_04.j2k).

[boxerab]

@mayeut great, thanks!

[mayeut]

@boxerab Wouldn't that be the source of the stalled tests under travis ?

[boxerab]

@mayeut once again, nice catch! I checked in some debug code again.
Thanks.

[boxerab]

I tested decoding of a 1000x1000 region from a large 22000x29000 image, tiled with 1024x1024 tiles.

This branch took 600 ms, while the omnibus branch took around 1100 ms, on my system.