the minimum schedule induces extra edges in the graph (corresponding to ordering in the schedule) that connect what would otherwise be parallel components of your graph.
-
-
I believe if your program is in SSA, then you can just assign registers first come first serve and it ends up optimal. Could be wrong.
2 replies 0 retweets 0 likes -
yes, but i don't have a *schedule*. that's the problem.
1 reply 0 retweets 1 like -
Replying to @FioraAeterna @pcwalton and
i know i can schedule first come first serve, and that works, but i'll get a dramatically inflated number because the code isn't scheduled yet.
1 reply 0 retweets 0 likes -
Ah, I see the problem. Um…you could try just visiting all live intervals and greedily splitting each one if possible
2 replies 0 retweets 0 likes -
Replying to @pcwalton @FioraAeterna and
(by “split” I mean “find overlapping intervals and move instructions around to make them non overlapping if possible”)
1 reply 0 retweets 0 likes -
okay so if i understand here right... you suspect the best way to do this is basically to *try to come up with the schedule*? that is, there's no way to do it from graph analysis, you basically just have to do the whole actual scheduling process
1 reply 0 retweets 0 likes
Dunno. There might well be a smarter way to do it. Seems hard because of complex interactions between scheduling and the interference graph
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.