search

WISE-Developers / Prometheus-EOL-Issues

0 stars 0 forks source link

Intel compiler #19

Open RobBryce opened 1 year ago

RobBryce commented 1 year ago

When we integrated the Intel compiler, we found about a 15% gain in performance. Since that time, Intel has migrated from GCC to Clang, and we've seen generations of new processors.

I suggest we do some critical timing operations to determine (a) if/what the performance gain still is and (b) if there's still value when WISE now exists.

If Intel is dropped, then the installer is simplified. If Intel is kept, then we should migrate to the latest version and update all aspects around its GUID accordingly.

BadgerOnABike commented 1 year ago

What is the effort required to determine these time differences?

As you mentioned, does it matter with the existence of WISE.

This strikes me as something that may not be worth doing for EOL. @nealmcloughlin thoughts?

RobBryce commented 1 year ago

I suggest this for 2 reasons: 1.) removing the Intel compiler from the Windows build makes it more consistent with WISE targets. 2.) Every time we upgrade the Intel or Microsoft compilers, I spend a lot of time dealing with minor variations I can contribute to differences in parsing of C++17 syntax, and fighting with related compilation errors. Then there's the minor regression testing that's needed because we have the odd file that has exhibited errors in the compilers.

The effort to remove Intel for a test build means dealing with differences in compilers. The long-term effort includes modification to the install process.