Are we able to dictate what each warp executes? Or do we only dictate the data and function that the whole GPU will execute?

How would you define a thread block? Is it constrained by the hardware i.e. each SM can only have a limited number of blocks? Or is it just a software abstraction for grouping threads into warp-aligned groups?

@endofmooore, from written assignment two, and the programming assignment now, we see that thread blocks are a software abstraction that the programmer determines to group threads together.