opensuse support

[walter_s_]

Hi

In the thread on the glibc 2.6 I have read about an extension of the support to FC7 and Ubuntu.
But what about OpenSuse ? I think it is a very popular ditribution and I am surprised to see that it is excluded from the supported ones.

Being the Intel compiler an essential tool for me I am going to switch from OpenSuse 10.2 to Mandriva 2007 but that is a pitty for me. Indeed, having used it for a long time, I was very satisfied with OpenSuse.

Regards

Walter

[TimP]

I suppose it's simply a matter of cost-benefit, as to whether anyone can afford to set up servers and complete the full round of testing. I'm amazed that you would claim a switch from OpenSuSE to Mandriva was done on account of ifort. Neither is likely to generate enough revenue in additional ifort sales to justify the testing. Maybe a reason you haven't heard much about OpenSuSE is that ifort installs the same as on SLES (which has undergone sufficient testing to be listed as supported), with no work-arounds required. Besides, OpenSuSE doesn't need special arrangements with hardware OEMs to make it viable.
I've found ifort (and icpc) working OK even on SuSE installations with deficient g++, both before and after installing g++ and gfortran from source in order to remedy the deficiencies. Currently, on my 32-bit laptop, I have an installation of ifort and icpc 10, made under SuSE 9.3, which works unmodified also under OpenSuSE 10.3. On the SuSE OS where ifort is installed, the rpm hooks into YaST provide full YaST accounting of the installation and its dependencies.

[walter_s_]

Hi tim18

I know that ifort installation works on OpenSuse as well as on SLES/SLED.
The problem is that with the last stable release (10.2) I have experienced some problems with the ugage of idb (see my previous post) and also some crashes (internal error) of icc. This happens despite that OpenSuse is based on the "supported" versions of kernel and glibc. But, as said in some other thread by Steve Lionel, there are many other variables beyond glibc and kernel version and it is better to stick with one of the supported distributions.
Among these there are SLED 10 and Mandriva 2007. Having tested both I have found the ifort idb problems disappears with them.
Unfortunately SLED 10 (sp1) has a too limited package selection and miss even the basic thevelopment packages like tk-devel, tcl-devel. So I have decided to go with Mandriva 2007.

walter

[TimP]

The more likely problem in icc is lack of compatibility with current libstdc++. Early 9.1 versions didn't work with any satisfactory g++ 4.x; later versions were OK with g++ 4.1 but not 4.2. I tried icpc 10.0.026 successfully with the g++ 4.2 active in 32-bit openSUSE 10.3, after seeing your post. These improvements run ahead of announced full support. I still have problems with g++ 4.3 active, including when running on RHEL4. I haven't tried 64-bit g++ 4.2 lately. I haven't seen so much impact of g++ version on ifort, which I thought was the subject of the post. I don't know many people who rely on idb exclusively, particularly not with icc.

[walter_s_]

Hi tim

But what else (instead of idb) would you use to debug a fortran 95 code compiled with ifort. Anyway I had the same problem with gdb (which I consider not as good as idb for fortran 95).

Because you have installed opensuse 10.3 I am curious to know ii you have the same problem I have (on OpeSuse 10.2) with the module variables.

To test it you should compile (ifort -g -o test test.f) the following very short file named test.f:

module mod1
implicit none
integer mod1_I1
end module mod1

program test
use mod1, I1=>mod1_I1
implicit none
I1=1
write(*,*) I1
end program

Then after havin launched idb (idb test) you have to enter the commands

break test.f:10
run
print I1

If all is OK you should get: I1=1
On OpenSuse 10.2 I get: I1=


Thanks
Walter