[gdal-dev] Re: Performance improvment for gdal_retile

[Even Rouault]

Le samedi 18 décembre 2010 14:31:21, christian.mueller at nvoe.at a écrit :
> Some last comment about the test case, the current source checks
> 
> test_gdal_retile_2()
> 
> ds.GetRasterBand(4).Checksum() != 35
> ds.GetRasterBand(4).Checksum() != 35
> 
> If I run this on a Linux Intel 32 bit configuration, it has to be
> 
> ds.GetRasterBand(4).Checksum() != 283
> ds.GetRasterBand(4).Checksum() != 122
> 
> Doing the same on a Linux AMD 64 Bit configuration, the values should be
> 
> ds.GetRasterBand(4).Checksum() != 283
> ds.GetRasterBand(4).Checksum() != 126
> 
> It seems that the cecksum is dependent on the  platform. Is this
> desirable ? Could this cause some headaches later ?

Yes, this is a bit annoying.

I guess this is related to some compiler difference that impacts numerical 
precision with the bilinear resampling. What is your OS and compiler versions 
? Do you compile in debug or optimized mode ? Do you use your script against 
GDAL trunk, or do you use it against GDAL 1.7 ?

35 works for me on Ubuntu 10.04 64bits, and it also works on all Windows 
builds of Tamas (http://vbkto.dyndns.org/sdk/) that test 7 combinations of 
32/64 bit and MSVC 1310, 1400, 1500 and 1600.

> 
> Cheers
> Christian
> 
> 
> 
> 
> ----------------------------------------------------------------
> This message was sent using IMP, the Internet Messaging Program.