Hi Nicholas, +1 to what Jason said. I just assigned the students in my class an assignment using many of the different replacement policies and none reported problems: https://pages.cs.wisc.edu/~sinclair/courses/cs752/fall2024/handouts/cs752-fall2024-hw5.pdf
My guess is you are thinking of my student's prior presentation on various bug fixes we made for the replacement policies (among other things): https://www.gem5.org/assets/files/workshop-isca-2023/slides/analyzing-the-benefits-of-more-complex-cache.pdf, unless there is something else I'm not aware of ... Matt On Mon, Oct 28, 2024 at 11:31 AM Jason Lowe-Power via gem5-users < gem5-users@gem5.org> wrote: > Replacement policies should be fully supported. > > On Mon, Oct 28, 2024, 8:44 AM Beser, Nicholas D. <nick.be...@jhuapl.edu> > wrote: > >> Jason, >> >> >> >> There was a comment in the charts that the replacement policy was broken. >> Is that still the case? >> >> >> >> I wanted to assign the students a task to experiment with different >> replacement policies. >> >> >> >> Nick >> >> >> >> *From:* Jason Lowe-Power <ja...@lowepower.com> >> *Sent:* Monday, October 28, 2024 11:24 AM >> *To:* The gem5 Users mailing list <gem5-users@gem5.org> >> *Cc:* Beser, Nicholas D. <nick.be...@jhuapl.edu> >> *Subject:* [EXT] Re: [gem5-users] Does GarnetPt2Pt and GarnetMesh still >> work in Gem5? >> >> >> >> *APL external email warning: *Verify sender ja...@lowepower.com before >> clicking links or attachments >> >> >> >> Hi Nick, >> >> >> >> The generic topologies in Ruby (e.g., the ones in configs/topologies) >> were always *very brittle*. This was especially so for topologies like Mesh >> where the topology is tightly coupled to the number of each type of >> controller (and even the coherence protocol!). >> >> >> >> Therefore, in the stdlib we have taken the approach that we will not >> provide generic topologies. Instead, we will provide specific instances of >> topologies paired with a protocol. I.e., we will provide "prebuilt" cache >> hierarchies. For example, the Octopi cache: >> gem5/src/python/gem5/components/cachehierarchies/ruby/caches/prebuilt/octopi_cache/octopi.py >> at stable · gem5/gem5 · GitHub >> <https://github.com/gem5/gem5/blob/stable/src/python/gem5/components/cachehierarchies/ruby/caches/prebuilt/octopi_cache/octopi.py#L58> >> >> >> >> To answer your question directly, the correct way to demonstrate garnet >> pt2pt or mesh would be to write your own cache hierarchy. Note that you >> don't need to put it in src/ or recompile gem5. Because it's pure python >> using the gem5 libraries you can simply write a python file and use it. If >> you create a `CacheHierarchy` then it will likely be compatible with all of >> our boards including both the test board (used with traffic generators) and >> all of the ISAs. >> >> >> >> Cheers, >> >> Jason >> >> >> >> On Sat, Oct 26, 2024 at 1:01 PM Beser, Nicholas D. via gem5-users < >> gem5-users@gem5.org> wrote: >> >> I have been looking over the example of Garnet in the 2024 bootcamp. The >> Ring-garnet does work as advertised. I was wondering if GarnetPt2Pt and >> GarnetMesh still works. The example from the older 2022 bootcamp used >> NULL/gem5.opt, but while the code is still in the materials/archive, the >> commands don’t seem to work. The examples in 2024 seem to use >> ALL_CHI/gem5.opt and not NULL. >> >> >> >> What is the right way to demonstrate Garnet Pt2Pt and Garnet_mesh? >> >> >> >> Nick >> >> _______________________________________________ >> gem5-users mailing list -- gem5-users@gem5.org >> To unsubscribe send an email to gem5-users-le...@gem5.org >> >> _______________________________________________ > gem5-users mailing list -- gem5-users@gem5.org > To unsubscribe send an email to gem5-users-le...@gem5.org >
_______________________________________________ gem5-users mailing list -- gem5-users@gem5.org To unsubscribe send an email to gem5-users-le...@gem5.org
