OK, if the current implementation is actually building a crossbar out of buses, then calling it a crossbar does seem reasonable to me... though there is some collateral renaming to be done for consistency, e.g., what about DPRINTF flags?
Steve On Mon, Mar 24, 2014 at 11:28 AM, Andreas Hansson <[email protected]>wrote: > Hi Steve, > > I've obviously not done a great job at updating the comments here. Guilty > as charged. > > What the current implementation does is to add a layer per port, i.e. each > input is fanning out to an implicit cross bar matrix, and each Layer > captures the convergence and fan in to the output ports. > > I suggest to update the comment, and then update the name if everyone > agrees. > > Andreas > > On 24/03/2014 11:13, "Steve Reinhardt" <[email protected]> wrote: > > >Thanks for the quick reply. Of course if you have enough independent > >buses > >then yes, you've built a crossbar out of buses... so I think I understand > >where you're coming from. > > > >Looking at the comment here: > > > > 76 <http://repo.gem5.org/gem5/file/4574d5882066/src/mem/bus.hh#l76> > > /** > > > > 77 <http://repo.gem5.org/gem5/file/4574d5882066/src/mem/bus.hh#l77> > > * A bus layer is an internal bus structure with its own flow > > > > 78 <http://repo.gem5.org/gem5/file/4574d5882066/src/mem/bus.hh#l78> > > * control and arbitration. Hence, a single-layer bus mimics a > > > > 79 <http://repo.gem5.org/gem5/file/4574d5882066/src/mem/bus.hh#l79> > > * traditional off-chip tri-state bus (like PCI), where only one > > > > 80 <http://repo.gem5.org/gem5/file/4574d5882066/src/mem/bus.hh#l80> > > * set of wires are shared. For on-chip buses, a good starting > > > > 81 <http://repo.gem5.org/gem5/file/4574d5882066/src/mem/bus.hh#l81> > > * point is to have three layers, for requests, responses, and > > > > 82 <http://repo.gem5.org/gem5/file/4574d5882066/src/mem/bus.hh#l82> > > * snoop responses respectively (snoop requests are instantaneous > > > > 83 <http://repo.gem5.org/gem5/file/4574d5882066/src/mem/bus.hh#l83> > > * and do not need any flow control or arbitration). This case is > > > > 84 <http://repo.gem5.org/gem5/file/4574d5882066/src/mem/bus.hh#l84> > > * similar to AHB and some OCP configurations. > > > > 85 <http://repo.gem5.org/gem5/file/4574d5882066/src/mem/bus.hh#l85> > > * > > > > 86 <http://repo.gem5.org/gem5/file/4574d5882066/src/mem/bus.hh#l86> > > * As a further extensions beyond the three-layer bus, a future > > > > 87 <http://repo.gem5.org/gem5/file/4574d5882066/src/mem/bus.hh#l87> > > * multi-layer bus has with one layer per connected slave port > > > > 88 <http://repo.gem5.org/gem5/file/4574d5882066/src/mem/bus.hh#l88> > > * provides a full or partial crossbar, like AXI, OCP, PCIe etc. > > > > > >So yes, if you have a layer per output port, it's a crossbar. If you only > >have layers separated by function (request/response, etc.) then it's a set > >of buses. So I suppose it boils down to whether we're talking about the > >configuration in the first paragraph of your comment or the second? Are > >you implying that the "further extensions" for a "future" crossbar-like > >configuration are now available in the present? > > > >Steve > > > >On Mon, Mar 24, 2014 at 11:02 AM, Andreas Hansson > ><[email protected]>wrote: > > > >> Hi Steve, > >> > >> Thanks for the input. > >> > >> I¹m tempted to engage on using some ASCII art here, but let me first try > >> and lay it out in words. > >> > >> For each point of convergence there can be one outstanding transaction. > >> For example, two independent masters can simultaneously use two > >>different > >> memory controllers (channels). On each input port, there is only > >> contention at the output port (for both requests and responses). This is > >> done by means of the layers, and there is one layer per output port. To > >> me, that makes it a crossbar. > >> > >> Andreas > >> > >> > >> > >> On 24/03/2014 10:52, "Steve Reinhardt" <[email protected]> wrote: > >> > >> >Hi Andreas, > >> > > >> >You've certainly done a lot of work on the code, so I'm interested in > >> >hearing more about your rationale for the name change. However, to > >>me, it > >> >looks like a set of parallel independent buses... unless things have > >> >changed more than I realize. For each bus (layer), you only have one > >> >sender at a given point in time, and---at least for snoop > >>requests---that > >> >sender's data is broadcast to all the receivers, right? > >> > > >> >I see that there's an isomorphism to a crossbar there, but to me it's > >> >still > >> >logically operating as a set of busses, since (for example) you can't > >>have > >> >independent responses (from different sources to different > >>destinations) > >> >transmitted at the same time---which is what a crossbar would imply to > >>me. > >> > Or is that possible now? I checked the code, and since each layer > >>has a > >> >single IDLE/BUSY/RETRY state, I'm guessing not, but I only glanced at > >>the > >> >code, so I may have missed something. > >> > > >> >Thanks, > >> > > >> >Steve > >> > > >> > > >> > > >> > > >> >On Sun, Mar 23, 2014 at 9:37 AM, Andreas Hansson > >> ><[email protected]>wrote: > >> > > >> >> Hi all, > >> >> > >> >> Would anyone mind if I go ahead and change all the Bus classes to > >> >>Crossbar > >> >> or Xbar? > >> >> > >> >> Thus: > >> >> > >> >> Bus -> Xbar > >> >> CoherentBus -> CoherentXbar > >> >> NonCoherentBus -> NoncoherentXbar > >> >> > >> >> The current name is rather misleading as the interconnect is truly a > >> >> non-blocking crossbar (with separate multiplexers for the request, > >> >> response, snoop request and snoop response). > >> >> > >> >> Thanks, > >> >> > >> >> Andreas > >> >> > >> >> > >> >> -- IMPORTANT NOTICE: The contents of this email and any attachments > >>are > >> >> confidential and may also be privileged. If you are not the intended > >> >> recipient, please notify the sender immediately and do not disclose > >>the > >> >> contents to any other person, use it for any purpose, or store or > >>copy > >> >>the > >> >> information in any medium. Thank you. > >> >> > >> >> ARM Limited, Registered office 110 Fulbourn Road, Cambridge CB1 9NJ, > >> >> Registered in England & Wales, Company No: 2557590 > >> >> ARM Holdings plc, Registered office 110 Fulbourn Road, Cambridge CB1 > >> >>9NJ, > >> >> Registered in England & Wales, Company No: 2548782 > >> >> _______________________________________________ > >> >> 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 > >> > > >> > >> > >> -- IMPORTANT NOTICE: The contents of this email and any attachments are > >> confidential and may also be privileged. If you are not the intended > >> recipient, please notify the sender immediately and do not disclose the > >> contents to any other person, use it for any purpose, or store or copy > >>the > >> information in any medium. Thank you. > >> > >> ARM Limited, Registered office 110 Fulbourn Road, Cambridge CB1 9NJ, > >> Registered in England & Wales, Company No: 2557590 > >> ARM Holdings plc, Registered office 110 Fulbourn Road, Cambridge CB1 > >>9NJ, > >> Registered in England & Wales, Company No: 2548782 > >> > >> _______________________________________________ > >> 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 > > > > > -- IMPORTANT NOTICE: The contents of this email and any attachments are > confidential and may also be privileged. If you are not the intended > recipient, please notify the sender immediately and do not disclose the > contents to any other person, use it for any purpose, or store or copy the > information in any medium. Thank you. > > ARM Limited, Registered office 110 Fulbourn Road, Cambridge CB1 9NJ, > Registered in England & Wales, Company No: 2557590 > ARM Holdings plc, Registered office 110 Fulbourn Road, Cambridge CB1 9NJ, > Registered in England & Wales, Company No: 2548782 > _______________________________________________ > 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
