parallel studio XE cluster edition initialization behavior

[Othman_B_Intel]

Dear colleagues,

I recently noticed a behavior I did not expected with Parallel Studio XE Cluster Edition 2017. It happens also for 2016.

To use PSXE 2017, I source /opt/intel/parallel_studio_xe_2017.2.174/psxevars.sh (nothing new, this is the last version actually).

But, I kept the old versions. And when I source /opt/intel/parallel_studio_xe_2017.1.043/psxevars.sh, the compilers points to the ones in /opt/intel/parallel_studio_xe_2017.2.050.

To me, the expected behavior is that when I source /opt/intel/parallel_studio_xe_2017.1.043/psxevars.sh the compilers points to the ones in /opt/intel/parallel_studio_xe_2017.1.043...

About VTune, I expect that this is always the last that is executed, as well as Advisor.

The advantage is that it keeps consistancy among the development tools (same version of the compiler, mpi, mkl...). Moreover, why keeping the script psxevars.sh in specific locations identified by version number if it is to load the last version of the tools?

What is your opinion?

Best regards,

Othman.

[gaston-hillar]

I think the right place to post this question is the Intel C++ Compiler forum. Here is the link: https://software.intel.com/en-us/forums/intel-c-compiler

[SergeyKostrov]

>>...I kept the old versions... I tried to keep old versions / updates of Intel PS XE Cluster Edition for Linux and decided it is better to remove old versions to avoid any problems similar to what you've described.