Variable rotates or constant rotates?

Currently I only need constant rotates, but since you can build a rotate out of shifts, it shouldn't be difficult to also handle variable rotates.

Constant rotates are easy and uncontroversial. Variable rotates not so much because if the LHS is not a power of 2 and the RHS can represent an absolute value larger than the length of LHS, you get a (possibly signed) modulus operator in the expansion and that's likely going to result in a nasty synthesized netlist.

Excellent point, I'd forgotten about that (and this is why I filed an issue first). I suppose it would have to be constant right hand side then.

We can always extend it to variable rotates later if someone comes up with a good way to handle this issue. So let's stick to constant rotates for now.