We have installed the 11.1 professional compiler set along with all the Intel developer tools as of November 2010.
I was wondering if the latest compiler set (12.1) and the other tools released around the same time, could be installed alongside the older Intel suite. One concern is that the new set has a new license file and I am not sure if I installing the new license will interrupt access to the older set.
Do I have to make any special arrangements in order to have the two tool sets coexist ?
If you integrate both compilers into the same Visual Studio, normally it would default to the newer compiler, but you can select the other on the pull-down menu. There's a recent answer about this on the Fortran windows forum. As I see in my own copy, tools>options>intel c++>compilers shows you the selection menu. By the way, we haven't reached 12.1 yet; there is "xe 2011" (12.0) update 1.
We are interested in the Linux environmnet exclusively. I assume that different compiler set versions can co-exist in Linux using the same license server, correct? I have an up-to-date license for the 12.1 set but I would like to have both 11.1 and 12.1 be accessible and selectable by users who have reasons to use on or the other.
Another Question: are the 12.1 tools ready for production use at the moment? We use Intel compiler set to built MPI stacks and then complex software on top of it. It would not be so efficient for us to say keep building the dependent software several times....
Yes, the supported way of choosing between 11.1 and "12.0" linux compilers is by sourceing the environment variable setting script in the corresponding compiler installation. There may be a problem in coexistence of "12.0" (xe 2011) updates 0 and 1, where the last one installed in the default path blocks the earlier one, but that can be overcome by choosing a new installation path. For example, I chose /opt/xeon/ instead of /opt/intel/ for my update 1 installation. This problem doesn't arise between 11.1 and the new compiler. I believe the current compiler is as well tested as 11.1 (better, on the distros which are newly designated as supported). The imf-arch-consistency option should help with numerical consistency between different CPU types. Needless to say, there would be individual points on which one or the other may be better, and most developers would make a full round of QA comparisons between old and new compilers. If you were to take full advantage of the new compiler, you might set a consistency option such as '-fp-model source', then use the #pragma simd on C loops for which vectorization is missed. My personal reservation is against the new nested loop transformations at -O3. You may be able to change path to the new compiler without rebuilding the existing MPI, if you built it with simply icc and icpc (no full path) as the compiler, as the new compiler and libraries are compatible with the earlier object files.