[PATCH] new warp algorithms - mode and average

[etourigny]

I have implemented mode and average warping algorithms, based on those used in overview creation. This code has been tested on a few datasets (see attachments) with all datatypes (Byte, Int, Float) but not all cases have been looked into (e.g. color tables, various projections). Any feedback and testing would be appreciated.

[etourigny]

patch

[etourigny]

test data

[etourigny]

test data (0s and 1s)

[etourigny]

There are still some comments in there, because some transparency/nodata issues have not been fully tested.

[etourigny]

Adding Frank and Even in CC, perhaps one of you could give some feedback? (esp. Frank which is the original author).

I am hesitant on a few issues - please look into the TODO comments if you have the time and inclination.

[etourigny]

optim and cleanup

[warmerdam]

Regarding the TODOs:

• The outer loop check of the unified mask is not appropriate for these algorithms since there are multiple input points per output point, so just remove the commented out code.
• The chunk nodata mask is not meaningful in the warp context and that commented out code can be removed.

The code looks good. I'd encourage you to go ahead and commit it.

[etourigny]

ok Frank thanks! am I right to assume this would make it into 1.10 ? If so, I will update docs also.

[warmerdam]

Please do update the docs too. Yes, it seems it will be in 1.10.

[etourigny]

commited in r25884 with updated docs

[Even Rouault]

The pabSuccess array isn't used currently to determine if a target pixel can be computed. Furthermore there should likely be a pabSuccess and pabSuccess2.

[etourigny]

added autotests in r25891

[etourigny]

added optimization for int16 - uses the same code as for byte data (r25895 and 25896)