Hey, I looked through this patch and I think: (1) The new code is making use of a variable "pushedRAS" that is not being updated for the predictInOrder function (2) The code could be consolidated even more (the predict() functions are awful similar)
For (1), on line 191 the pushedRAS variable is getting updated. This line should also be placed on line 314. This matters because while the predict functions were separate, the squash function became consolidated and is evaluating the pushedRAS variable in a couple conditons. Basically, for the InOrder version, that pushedRAS variable is never getting updated. For (2), the extras that the InOrder adds to the predict() function is that the predictor will return FALSE if there is no RAS entry (lines 273-277) or no BTB entry (line 343). In both of these scenarios, the prediction of true will be invalid without a PC to go with it, so the InOrder model chooses to treat that as a "false". This could probably be toggled via some config parameter whenever someone chooses to consolidate further. Nilay, do you mind adjusting for (1) and then committing? Thanks! -Korey On Mon, Jan 21, 2013 at 7:42 AM, Nilay <[email protected]> wrote: > > On Tue, January 15, 2013 6:51 pm, Nilay wrote: > > On Tue, January 15, 2013 6:24 pm, Korey Sewell wrote: > >> Can I review this change again before commit? Or can you separate the > >> patch > >> into inorder/o3 portions? > >> If you give me through this weekend I can try to figure out why > >> combining > >> the code is causing stats differences. > > > > You are free to review the patch. I was intending to commit it in a day > or > > two. Since you want time till weekend, I will wait till then. Since the > > aim of the patch is to combine the predictors, separate patches does not > > make any sense. > > > > Korey, did you go through the patch? What's your verdict? > > > -- > Nilay > > _______________________________________________ > gem5-dev mailing list > [email protected] > http://m5sim.org/mailman/listinfo/gem5-dev > -- - Korey _______________________________________________ gem5-dev mailing list [email protected] http://m5sim.org/mailman/listinfo/gem5-dev
