Re: 2.9.8 Build problems on Opteron (RHEL4)

[John W. Eaton]

On  6-Sep-2006, Fredrik Lingvall wrote:

| >     | F77=gfortran CPPFLAGS="-I$HOME/work/OPTERON/usr/local/include"
| >     | CFLAGS="-O3 -m64 -Wall -funroll-loops -march=opteron -mtune=opteron
| >     | -pthread" CXXFLAGS="-I$HOME/work/OPTERON/usr/local/include -O3 -m64
| >     | -Wall -funroll-loops -march=opteron -mtune=opteron -pthread"
| >     | FFLAGS="-I$HOME/work/OPTERON/usr/local/include -O3 -m64 -Wall
| >     | -funroll-loops -march=opteron -mtune=opteron -pthread"
| >     | LDFLAGS="-L$HOME/work/OPTERON/usr/local/lib"
| >     ./configure  --enable-64
| >     | --prefix=$HOME/work/OPTERON/usr/local
| >     | --with-blas=$HOME/work/OPTERON/usr/local/lib/libBLAS.so
| >     | --with-lapack=$HOME/work/OPTERON/usr/local/lib/libLAPACK.so
| >
| >     Are you sure you want --enable-64?  Did you note the warning
| >     issued by
| >     configure when you use it?
| >
| Yes and yes. I used gfortran from gcc 4.1.1 and Goto BLAS v 1.04 so I
| hope its OK. At
| least I do not get any core dumps or other errors.

Was your copy of Goto BLAS compiled such that it will use with 8-byte
integers for dimension and index values?

Also, I don't see a -fdefault-integer-8 option for gfortran.
To make Octave work properly with --enable-64, you must have 8-byte
integers for any dimension and index values used for numeric arrays.

As the warning said, this is still an EXPERIMENTAL option.  It is a
work in progress.  If the conditions above are not met, then there may
be mysterious failures, including incorrect results that could occur
with no warning.

| Yes the patch made 2.9.8 compile nicely. Seems to run fine too. Thanks!
| I did not get these errors on my 32-bit Gentoo laptop.  Do you know why
| it failed on the 64-bit machine?

Because on all systems, nzmax in the MEX inteface is always an "int"
object, which is typically 4 bytes wide even on 64-bit systems, but on
64-bit systems with --enable-64, octave_idx_type is an 8-byte integer.

jwe