Wilson Snyder
Wilson Snyder
BTW thought you were going to also split this into a pre-pull to add the branch to .infos.
Correct, thanks
Similar to the case of multiple statements on a line aggregating: ``` %000000 if (cyc[0]) $write(""); if (cyc==4) $write(""); if (cyc==9999) $write(""); -000001 point: comment=if +000089 point: comment=else -000000 point:...
Yes, I'd suggest treat them as branches. Yes, the comment is lost, when we do cover groups that's pretty critical.
Which do I merge first? I'll read through again tonight, then assuming good do the first one, then you'll need to merge in master & repush the other.
orderModuleVars is the entire algorithm in that stage, so that's expected. I don't see any obvious mistakes, so please try the profiler, and/or look at adding prints around the stages...
I suggest we add a -mno-variable-order-tsp flag to fall back to the nonmtask algorithm. @godblesszhouzhou perhaps you'd be willing to make a pull to do that?
Or maybe with that flag just group by mtask number then historical. False sharing will kill performance otherwise.
I believe there are three things to help both capacity and misses: 1. Make sure different mtask'ed variables are padded to different lines. (easy) 2. Make sure multiple variables pack...
While I'd ideally like to see a pre-pull to cleanup the checker before we merge this, but I re-reviewed this pull and seems correct, and so I'll merge.