I don't want to make this more complicated than necessary, but did you figure out why flagging hwrei as "isControl" breaks O3? Conceptually you are correct that it changes control flow, so if that breaks O3 I would argue that it's a bug in O3 and not a problem with the flag. Adding "dynamic tagging" to the in-order model seems like a very convoluted workaround; the whole point of these flags is that these attributes are determinable statically.
Steve On Thu, Jul 7, 2011 at 11:22 AM, Gabe Black <[email protected]> wrote: > Make sure you update the other regressions back again. I forget which > ones exactly changed, but I tried it briefly and you can't -just- remove > isControl. > > Gabe > > On 07/07/11 06:08, Korey Sewell wrote: > > Hi Ali, > > the fix for O3 is easy (just remove the isControl flag), but the inorder > fix > > became a little messier than I wanted and I didnt get a chance to look at > it > > again yesterday. > > > > What I'll do is update the fix for O3 so people can commit. > > > > For Inorder, it looks like the best strategy is to dynamically tag the > hwrei > > and any other pc-changing, non-branch instructions as "isControl". That > > general mechanism isn't quite working. > > > > On Thu, Jul 7, 2011 at 1:09 AM, Ali Saidi <[email protected]> wrote: > > > >> Any update? Until you get this fixed one one can really commit anything. > >> > >> Ali > >> > >> On Jul 3, 2011, at 11:13 AM, Korey Sewell wrote: > >> > >>> OK, I think I have the fix for this in the works, but it's a long > holiday > >>> weekend so please bear with me and I'll be able to get it in and tested > >> at > >>> some point on Monday. > >>> > >>> In the interim, if O3-ALPHA is failing, people can just take off the > >>> "isControl" flag on the hwrei instruction in the alpha isa decoder. > >>> > >>> On Mon, Jun 27, 2011 at 10:14 PM, Korey Sewell <[email protected]> > >> wrote: > >>>> I'll take another look at this, there always seems to be a regression > >> that > >>>> falls through the cracks. > >>>> > >>>> In theory, hwrei should be a serializing instruction anyway, so > marking > >> it > >>>> as "control" should be irrelevant other than for some branch > prediction > >>>> effects. > >>>> > >>>> We'll see. > >>>> > >>>> > >>>> On Mon, Jun 27, 2011 at 9:51 PM, Steve Reinhardt <[email protected]> > >> wrote: > >>>>> On Sun, Jun 26, 2011 at 2:31 PM, Cron Daemon < > >> [email protected] > >>>>>> wrote: > >>>>>> ***** > >> build/ALPHA_FS/tests/opt/long/10.linux-boot/alpha/linux/tsunami-o3 > >>>>>> FAILED! > >>>>>> ***** > >>>>>> > >> build/ALPHA_FS/tests/opt/long/10.linux-boot/alpha/linux/tsunami-o3-dual > >>>>>> FAILED! > >>>>>> > >>>>> So I tracked this down with hg bisect and got: > >>>>> > >>>>> % hg bisect -g > >>>>> The first bad revision is: > >>>>> changeset: 8398:d389b6ec0e2d > >>>>> user: Korey Sewell <[email protected]> > >>>>> date: Sun Jun 19 21:43:39 2011 -0400 > >>>>> summary: alpha: make hwrei a control inst > >>>>> _______________________________________________ > >>>>> gem5-dev mailing list > >>>>> [email protected] > >>>>> http://m5sim.org/mailman/listinfo/gem5-dev > >>>>> > >>>> > >>>> > >>>> -- > >>>> - Korey > >>>> > >>> > >>> > >>> -- > >>> - Korey > >>> _______________________________________________ > >>> gem5-dev mailing list > >>> [email protected] > >>> http://m5sim.org/mailman/listinfo/gem5-dev > >>> > >> _______________________________________________ > >> gem5-dev mailing list > >> [email protected] > >> http://m5sim.org/mailman/listinfo/gem5-dev > >> > > > > > > _______________________________________________ > gem5-dev mailing list > [email protected] > http://m5sim.org/mailman/listinfo/gem5-dev > _______________________________________________ gem5-dev mailing list [email protected] http://m5sim.org/mailman/listinfo/gem5-dev
