[moonchild]

> I've never heard anything about these sorts of graph algorithms being possible with good asymptotics in an array style

yeah—idk graph algos really, but have heard parallelising combinatorial search in general (eg sat) is hard because forks and joins happen heterogeneously and erratically. this 2001 vintage has a bunch of completely sequential graph colouring algorithms in j https://dl.acm.org/doi/pdf/10.1145/570406.570416 (and j at least has sparse matrices!)

> Constant lifting within the compiler is pretty cool, I'll have to look into that.

hrm, it seems to refer to 2 things: 1) constants allocated in a special space; 2) interning. 2 is obviously worthwhile; 1 i would guess is related to memory being weird on the gpu?