This looks like we are veering into declarative languages - that is, we are basically telling the compiler / scheduler what we want the code to do. Do DSLs tend to be declarative, or are they at least trending in that direction?

@rahulahoop DSL's are often declarative, since the DSL itself is just an interface to rewrite the code you give it. There will be imperative parts (at some point you have to specify the computation you want to perform), but by making it declarative, you allow the implementation of the DSL to do things like loop-fusion, auto-vectorization, and auto-parallelism with essentially no changes to the "meat" of your code. At least, that's my understanding of it.