[Gdal-dev] Floating point exception on x86_64

[vincent at ecovla.nl]

Just FYI,

I've been running gdal for more than one year now on amd64 linux
(Gentoo) without any (well, without much apparently amd64 related)
problems. Somewhere in the misty outbacks of my memory I do recall to
have had some problems in the beginning (could search the mailing list
for that) but those were solved by Frank then.
I always compile gdal from cvs, though, have never used the gentoo
ebuild. That could be some hint.

My gdal configure script:

#!/bin/bash
#gdal configure script
CFLAGS="-O3 -march=k8 " CXXFLAGS="-O3 -march=k8 " \
./configure \
--with-libz \
--with-png \
--with-libtiff=internal \
--with-geotiff=internal \
--with-jpeg \
--with-netcdf=/usr/lib \
--with-hdf4=/usr/lib \
--with-pg \
--with-grass=/usr/local/grass-6.1.cvs \
--with-odbc \
--with-jasper=no \
--with-geos

(and for the careful reader: yes, netcdf *and* hdf4 do work without any
interference, takes some ebuild tricks though).

Regards,
Vincent.


Pierre GM wrote:

>Folks,
>I've been a quasi daily user of gdal for the last three years on x86. For the 
>last week, I've been trying to use gdal-1.26 on two 64b machines, a Mobile 
>AMD64 and a P4_E64. Both machines run Gentoo. I first used the gentoo ebuild. 
>On both machines, installation works, gdal-config works properly, and I'm 
>able to use the GRASS 'r.in.gdal' with success.
>However, gdalinfo, gdalwarp, ogrinfo ... all crash with a floating point 
>exception (even with as simple as a --help as flag). A dmesg gives me (for 
>example):
>gdalinfo[2657] trap divide error rip:3682a0811d rsp:7fffffffe900 error:0
>gdalwarp[2799] trap divide error rip:3682a0811d rsp:7fffffffe910 error:0
>ogrinfo[2922] trap divide error rip:3682a0811d rsp:7fffffffe910 error:0
>
>Here are my regular CFLAGS and FFLAGS
>CFLAGS="-march=k8 -O2 -pipe -fweb"
>FLAGS="-march=k8 -O2 -pipe -fweb"
>On top of that, I have 'pic' in my USE flags (for non-gentooer, that puts an 
>additional -fPIC to the flags when needed).
>
>Compiling the sources 'by hand' leads to the same (absence of) results. Same 
>thing with 1.2.5
>
>Any idea/comment/suggestion would be more than welcome, a GIS without gdal 
>isn't really worth
>
>Thx in advance
>
>  
>