P4 spec issue 1261 add bounded loops

[jafingerhut]

No description provided.

[jafingerhut]

This is NOT a complete implementation of for loops in p4c. It probably isn't even a complete or correct change to the syntax as of the first 2 commits. It is a work in progress that I have used to verify that with these proposed grammar changes, bison does not give an error message when it processes the grammar.

[asl]

@jafingerhut Some time ago we implemented bounded loops downstream. They employ for (variable in sequence) syntax as a way to enforce that the loop is indeed bounded (it seems quite a non-trivial task in general for C-style for (start; end; increment) loops).

Let me know if you'd like me to factor our this change and submit just to have some end-to-end implementation.

[fruffy]

Bounded loops are a top-level agenda item of the P4 language working group for this year:

https://docs.google.com/document/d/1XSgdXeG1UuF1FM_XAqxDrHeN4dHZWBnJPKVS6SnGNwM/edit#heading=h.q7bzcwglv61g

Upstreaming this is a great idea and very useful. How the syntax will end up looking like is a different problem though.

[asl]

Sure, I can see what could be done.

[jafingerhut]

@asl Thanks for the offer. Fabian gave the high order answer. There should be a public meeting focused on the topic of loops in P4 coming up soon-ish (within a month or so), and I can be sure to try to include you in the choice of meeting time so it fits in your schedule.

[ChrisDodd]

I've started looking at loops in the compiler, so I'm looking at what to add in the IR for these, and what is needed in other parts of the compiler.

[ChrisDodd]

What is a typeRef supposed to indicate here? Creating a temp of the type and then iterating over it?

[ChrisDodd]

I added an extra commit with the IR changes for this.

[asl]

I added an extra commit with the IR changes for this.

I've almost finished porting our changes for the loops. I will post a PR shortly

[asl]

I ported our downstream code at #4558

Currently use-def changes are missed, I am working on porting them asap.

Let me know if there are any questions / comments / suggestions

[ChrisDodd]

I've pulled together both changes in #4562

[asl]

I was not part of the original discussion, so do not know about all the details here. But my understanding is that if the intention is to support bounded loops, then I think it should iterate over collection (or as in my original PR, over explicit range like a ... b). Generic C-style loops for (start; cond; update) are much harder to prove that they are bounded.