Not sure what has been decided here, but if nothing has been decided, I'd like to say that ext is the wrong place for this code. Either gem5 adopts a true fork of mcpat, puts it in src/power, adds regressions for it, etc. or it goes in a separate repo with EXTRAS.
If you look at ext, it is really for external packages that never really change. They're there because the packages themselves aren't found in normal distributions and gathering the code would be prohibitive. Essentially, we drop things into ext and forget about them. If we want to keep a gem5 fork of mcpat up to date, then we can't forget about it and we shouldn't hide it. It should be in the main source tree and part of the normal build/regression process. I see only one major downside: it's just more code to maintain (we have a dram simulator in the tree that has basically been rotting for a decade). If nobody wants to step up to maintain the power model, then I'd argue that it should go into EXTRAS and if someone wants to use it on a more modern version of gem5, then it can be updated on demand. Nate > Thanks Tony for posting this initial patch. I know it has > been a few weeks, but want to restart this discussion. > We would like to include this version of McPAT directly > into gem5 so that we can keep it "in sync" with the > gem5 output. We fear that if we move it to a separate > source tree, it will become stale with the constantly > evolving gem5 statistics and configurations. There are > also secondary benefits from AMD's perspective in keeping > it the same respository that I'd rather not get into. _______________________________________________ gem5-dev mailing list [email protected] http://m5sim.org/mailman/listinfo/gem5-dev
