10 ms·
Have you watched all of the lectures? Your questions are answered in them. The Mill is not at all like other VLIW architectures.
by chongli 13y ago
Have you watched all of the lectures? Your questions are answered in them. The Mill is not at all like other VLIW architectures.
- DSingularity 13y agoHe is talking about the fundamental problem of VLIW architectures. They rely on the static scheduling of programs to enable instruction level parallelism -- which is critical to performance. Scheduling is an NP Complete problem.
- Dylan16807 13y agoThe harder the problem, the worse it's going to be to calculate it on the fly.
- foxhill 13y agonot true. an OoO scheduler has a fixed window size with which to re-order. and as soon as you have data-dependent branches or dependencies, your compiler cannot possibly schedule instructions effectively.
- Daniel_Newby 13y ago> They rely on the static scheduling ... Or run-time profiling.
- chongli 13y agoHe hasn't said how this would be relevant to the Mill.
- nraynaud 13y agohe proposes backwards scheduling from the sink to the sources inside each function. I don't know how much complexity is that (because np-complete is not that bad when n is low)
- igodard 13y agoScheduling is the bin-packing problem, which is indeed NP. However, long (40+ years) known heuristics get within low single digits % of perfect, and generally better than OOO hardware scheduling because static is not constrained by instruction window size. We didn't invent those heuristics, but they work for us too.
- cpleppert 13y agoI think the real problem is that it isn't clear what makes it different from VLIW architectures aside from esoteric features like this that I have major issues with (other comment:https://news.ycombinator.com/item?id=6874853 https://news.ycombinator.com/item?id=6874853)