Experiment with FMV for gridding code

This issue was copied from the ASTRON Jira board: RAP-36

Description

WSClean doesn't use separate functions for SSE and AVX2, like AOFlagger does. Hence, optimization of WSClean needs to be done a bit different. Here we can probably use __attribute__((target_clones(...))), which instructs the compiler to generate different versions of the same function for all the features mentioned in ... (see for example this document for more information) .

It is very important to have a good way to measure performance. Andre Offringa will provide information on how this can best be done for WSClean.

Acceptance criteria:

• Report on the results
• Sketches direction to go toward
• Follow up tickets are created in jira.