sporky-loops

[svishnus]

• Added some comments on where spork gets added as a prim and lowered into IR
• Created a unified interface for unrolled and regular loops in ForkJoin
• Remove sources.mlb (used for millet-ls) in the end
• Spork_Prim now gets a boolean attribute isLoop which can be observed by the compilation pipeline
• We are instead going with spork_choose loopBody unrolledThunk() regularThunk()
• Need to make sure spork/spoin nesting depth isn't getting messed up
• Think of introducing management of spareHeartbeatTokens and .hasEnoughToSpawn checks in the compiler -> remove from Scheduler.sml

NOTE: Can we ever get a loopy spork from a non-loopy initial call?

• Answer is yes, examples in loopy sporks notion page

[svishnus]

Introduced the new spork_choose prim at this point, which I'm trying to handle during closure-convert. I still don't fully understand what's going on. Compile errors are fixed, but there is some obviously wrong abstract value management happening.

[svishnus]

We have a basic threshold-based logic that counts the number of sxml decs in a loop body and decides to apply the optimization if loopBodySize < threshold (currently at 100). NOTE: This does not account for nested sporks . We also don't do any sort of weighted sum based of distinct weights per specific type of sxml dec.