Performance problem launching multiple target regions

What’s going on the host that requires the data to be copied back and forth? Are there additional sections of code that offloaded so the data movement is not necessary?

I have to investigate it.

Again, I’d first try to remove data movement, but then if there’s not a data dependency, try to overlap compute and data movement by adding “nowait” to the compute regions and update directives.

Understood. Thanks for your great help. Regards.