[Gdal-dev] ECW Source SDK Available - Linux Binaries Prepared

[Frank Warmerdam]

On Mon, 31 Jan 2005 14:45:24 +0100, Nacho <brodin_ign at gva.es> wrote:
> Hi,
> I have compiled gdal with .ecw support in Linux and I am doing some test with
> it.
> I have detected a performance problem and I'd like some help about it.
> When I read a window (100x100 px) smaller than the total width
> of the image (8400x6200 px) it cost same time than reading a window
> with full width (8400x100 px). With gdalinfo I've seen that it has a
> block size of 8400x1 and I think there are some relationship, but I've no
> idea about how to cope with it.
> I've had some equaly puzzling results with mrsid, and no having solved it
> I finally decided to use GeoDSDK library without GDAL.
> Maybe I'm doing a bad use of the library  ... any help welcome

Nacho, 

There are some issues with how GDAL decides whether to do block based
reading, or custom windowed reads with some of the specialized drivers
like the MrSID, ECW and JPEG2000 drivers.  

Are you using the latest CVS version of the ECW driver?  I have substantially
rewritten it since GDAL 1.2.5.  What calls do you make?   A since
GDALRasterBand::RasterIO() for each band?  One call per scanline needed?
Different access patterns can give quite different behavior. 

I am interested in working on the logic to give good general case behavior of
the driver, so if you can write up a small program that demonstrates your access
case, and also provide the configuration information about your file, I can dig
into what is happening. 

Similar issues arise with MrSID, ECW and JPEG2000 though the internal rules
are not the same, so fixing one won't necessary fix the others. 

Best regards,
-- 
---------------------------------------+--------------------------------------
I set the clouds in motion - turn up   | Frank Warmerdam, warmerdam at pobox.com
light and sound - activate the windows | http://pobox.com/~warmerdam
and watch the world go round - Rush    | Geospatial Programmer for Rent