I'd like to follow up on this question a bit. Does anyone have tests for Ruby in FS mode? I'd like to move from the memtester CPU in SE mode and see if we can get some real benchmarks running in FS mode.
As stated before, the comments in ruby_fs.py dont sound too promising: "# currently ruby fs only works in simple timing mode because ruby does not # support atomic accesses by devices. Also ruby_fs currently assumes # that is running a checkpoints that were created by ALPHA_FS under atomic # mode. Since switch cpus are not defined in these checkpoints, we don't # fast forward with the atomic cpu and instead set the FutureClass to None. # Therefore the cpus resolve to the correct names and unserialize correctly. #" Has anyone gotten past this point where you can boot directly without a checkpoint or is that still future work for now? On Wed, Jan 26, 2011 at 1:19 PM, Malek Musleh <malek.mus...@gmail.com>wrote: > ---------- Forwarded message ---------- > From: Malek Musleh <malek.mus...@gmail.com> > Date: Tue, Jan 25, 2011 at 10:37 AM > Subject: Questions about FS Support using Ruby's Memory Model > To: M5 users mailing list <m5-us...@m5sim.org> > > > Hello, > > I am trying to figure out what is the current status of the ruby > memory model integration into M5, and its current functionality in FS > mode. > > 1) What topologies are currently working, how to specify them in the > command line, > > 2) I see that in ruby_fs.py that there is a comment that only the > TimingSimpleCPU Model is currently functional with ruby, when O3 is > expected to be functional? > > 3) What known issues there are that would prohibit someone from using > (in terms of making modifications to the coherence protocol), or is > its current status before suited for those that want to simualte a > Point to Point Network but want to make changes outside of the Memory > Model? > > 4) I have been able to boot up to 4 cpus, and I assume once I follow > the steps on the wiki page to apply the linux kernel patches for > BigTsuanmi, that I can support up to 64 cpus? > > I have been able to build ALPHA_SE_MOESI_hammer/m5.opt and > ALPHA_SE_MOESI_CMP_directory/m5.opt and subsequently boot the Alpha > kernel using the ruby_fs.py script file, but I assume if no additional > parameters are specified in the command line, and it just uses a > simple link with N cpus. > > I did not see anything on the wiki page regarding building/running > ruby, which made me think that its still not functional, but the fact > that I was recently able to build and boot the kernel made me think > otherwise. > > Thanks. > > Malek > _______________________________________________ > m5-dev mailing list > m5-dev@m5sim.org > http://m5sim.org/mailman/listinfo/m5-dev > -- - Korey
_______________________________________________ m5-dev mailing list m5-dev@m5sim.org http://m5sim.org/mailman/listinfo/m5-dev
