Mike Urbach

> TODO(cgyurgyik): Verify with PyTorch community this is still the bufferization method used. FYI this is definitely not the bufferization method they use, this is a newer method I wanted...

> Part of this issue is placing all the necessary steps in one location. For the record, this flow does exist in one place, wrapped up by a Python script...

> Curiously, do you have examples of bigger programs you've successfully lowered to Calyx native with desired behavior? Nothing more complex than a single loop nest, like what is checked...

> Yeah I can bring up more examples, but honestly it wasn't hard for me to cause the compiler to crash / produce semantically incorrect code (e.g., slicing bits that...

I agree, it seems best to make this explicit in the IR rather than deferring to an external toolchain. I was thinking of something similar, in the context of scheduling....

One thought I had about `if`: if we are able to handle back edges in the control graph (for `par`, in the presence of `while`), could a similar approach handle...

> So, the previous version of this pass actually did exactly that. If you’re curious, take a look at the static_timing.rs file under passes in the source tree. > >...

@rachitnigam I thought about this some more, and I think one possible point in the design space is to keep seq and par as-is (count up through the child states),...

Makes sense, I do like the conceptual simplicity of the 1 FSM per thread idea. I am mainly thinking about pipelined loops, which right now are represented with something like...

Logistically, how would this work? Would remove-comb-groups be removed from the pre-opt alias, and the tdcc and top-down-st passes updated accordingly?