This was helpful to think about in terms of how programmers think about useful "abstractions"—in this case, multiple threads of control—and do not need to understand the low-level details of how implementation works. However, I think it's helpful to understand the implementation because that can affect how this is used. If there were multiple different threads of control, I would think it's fine to implement code with conditions (divergence really isn't a problem here). Knowing it's SIMD however affects when I would use this: running code with large amounts of divergence probably wouldn't create large speedups.
This was helpful to think about in terms of how programmers think about useful "abstractions"—in this case, multiple threads of control—and do not need to understand the low-level details of how implementation works. However, I think it's helpful to understand the implementation because that can affect how this is used. If there were multiple different threads of control, I would think it's fine to implement code with conditions (divergence really isn't a problem here). Knowing it's SIMD however affects when I would use this: running code with large amounts of divergence probably wouldn't create large speedups.