Avoid block stride loop on AMD GPUs to increase performance for FEM kernels

[artv3]

It has been observed that performing block stride loops on AMD decreases performance, to increase performance use a direct mapping. Please see FEM kernels under apps.

[MrBurmark]

This is specifically for block stride loops and not grid stride loops?

[artv3]

The kernel that prompted this had block stride loops, but yes I think we have seen lower performance with grid stride loops as well in other contexts as well.