Re: [Intel Compiler] Optimisation flag does not like __va_arg_pack / undefined reference to __builtin_is(finite|nan|inf)

[Jaroslav Hajek]

On Wed, Apr 21, 2010 at 9:32 AM, Vincent Letocart
<address@hidden> wrote:
> On 21 April 2010, at 09:26:22 (Epoch+1271838382 sec), Jaroslav Hajek wrote :
>
>> > I have two other options: a guru tells me some tricky tips to get
>> > it working with gcc 4.3.x (flag, change in config/Makefile), or
>> > I pray a lot again and substitue the "compiler-triplet" (icc,ifort,icpc)
>> > by (gcc,ifort,g++) version 4.3.3 ?
>> >
>> > Vincent.
>> >
>> >
>>
>> If you're OK with g++, why not just use (gcc,gfortran,g++)?
>
> Because we (at the Belgian Institute of Space Aeronmy) try
> to use the Intel Compilers as much as possible, and as I also
> compiled the dependencies (fftw3, arpack, glpk, qhull, qrupdate,
> hdf5, metis, suitesparse-metis) with those compilers, I expect
> the upgrade to work the same way. Of course, if there is no
> solution ... then a "fallback" to the GNU compilers is the
> best way to get it working.
>
> Regards
>
> Vincent.
>
> --
> A woman should have compassion.
>                -- Kirk, "Catspaw", stardate 3018.2
>

In that case, I would suggest you compile gcc 4.1.2 to a special
directory. It should be possible to substitute a different stdc++
version for icpc using the -cxxlib flag. However, you may need to
recompile the dependencies with this setting as well.
Since you'll be compiling an older gcc with newer one, I think you may
disable bootstrapping which will make the build much faster.


-- 
RNDr. Jaroslav Hajek, PhD
computing expert & GNU Octave developer
Aeronautical Research and Test Institute (VZLU)
Prague, Czech Republic
url: www.highegg.matfyz.cz