It looks like gzip and parser failed in the most recent nightlies too (March 11th). They both passed on Feb 27th and didn't run on the March 6th.
In that time, I changed how the pagetable walker works slightly, although that won't be used in SE mode, changed how PCI config space is handled which also shouldn't affect SE mode. I also marked prefetches as prefetches, marked IO reads and writes as non-speculative, and started decoding a unofficial and poorly documented floating point instruction. These -shouldn't- matter in SE, other than maybe the prefetches, but it's possible something was misspeculated that behaved differently and threw things off slightly. <looks at the statsdiffs on zizzer> Yeah, the differences are that more non-speculative instructions were floating around and then squashed, and some other small changes which are probably due to that. All those other differences show up as less than 0.00% change. I think there were just mispeculated IO reads and writes (IN and OUT instructions, basically) which changed execution slightly. It sounds like you've already rerun parser at least. If your stats differences sound like what I'm describing, please go ahead and update them. I should have run them again myself, but I must have either forgotten to or assumed they wouldn't affect SE. Gabe On 03/16/11 12:40, Ali Saidi wrote: > > Looks like the parser x86 o3 regression fails without any changes in > the head. Anyone seen this? > > Ali > > > _______________________________________________ > m5-dev mailing list > [email protected] > http://m5sim.org/mailman/listinfo/m5-dev _______________________________________________ m5-dev mailing list [email protected] http://m5sim.org/mailman/listinfo/m5-dev
