Hello,
I am trying to use AMD’s APU model for gem5 found at https://gem5.googlesource.com/amd/gem5/+/agutierr/master-gcn3-staging to perform a sensitivity analysis of cache behavior on various GPU workloads, but it doesn’t look like cache statistics are being tracked correctly. Looking in the stats file at system.ruby.tcp_cntrl0.L1cache.demand_accesses shows that there are zero cache accesses, and the same goes for the other CUs’ TCPs. Should these statistics be tracked? I am running the Floyd-Warshall benchmark from the Pannotia Benchmark set with the 512 vertex and 64k edge graph provided in the dataset, and I wouldn’t imagine that this entire dataset fits into the register file. Any ideas for what I’m doing wrong? I was directed by Matt Sinclair to ask Tony Gutierrez specifically about this. Thank you, Devin Lafford
_______________________________________________ gem5-users mailing list gem5-users@gem5.org http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
