Rachit Nigam

In general, this kind of reasoning would require understanding all of the possible values a register can take (or an overapproximation of that) at any given program point. Might be...

How would it be simpler?

That sounds reasonable. TODO item to this issue: - look into whether Verilator supports overflow checking - Change simulation generated to error on overflows by default.

@EclecticGriffin do you think that the interpreter would catch this problem? If so, we should probably close this issue and say the interpreter is the way to detect these problems.

Ah, I misremembered the problem; I thought interpreter's overflow detection is sufficient to catch this problem because the loop index will keep overflowing.

I'm curious about something: bignums seems to be a critical feature in pyret and yet seem to cause major bottlenecks like these. Would it be feasible to implement bignums as...

Here’s what the current implementation does: - When the type checking of a scope finishes, all let bound variables are returned to the parent context. - For combine blocks, the...

> - changing the AST to mark variables that are allowed to turn into combine registers The code for this would be somewhat ugly--while traversing a `for` loop body, if...

Addressed for 1-dimensional memories by https://github.com/cucapra/dahlia/pull/394

The two passes being the passes that hoist memory reads and "expensive" binary ops. See #315 for discussion on how to do it.