Just wanted to bump this since it's been a month. Any comments? It would probably be good if someone at AMD who is familiar with the protocol could take a look.
Thanks, Lena On Mon, Jul 20, 2015 at 3:48 PM, Jason Power <[email protected]> wrote: > This is an automatically generated e-mail. To reply, visit: > http://reviews.gem5.org/r/2978/ > > Ship it! > > LGTM. > > > - Jason Power > > On July 18th, 2015, 9:20 p.m. UTC, Lena Olson wrote: > Review request for Default. > By Lena Olson. > > *Updated July 18, 2015, 9:20 p.m.* > *Repository: * gem5 > Description > > Changeset 10924:752540bf5605 > --------------------------- > ruby: Add missing block deallocations in MOESI_hammer > > Some blocks in MOESI hammer were not getting deallocated when they were set to > an idle state (e.g. by invalidate or other_getx/s messages). While > functionally > correct, this caused some bad effects on performance, such as blocks in I in > the > L1s getting sent to the L2 upon eviction, in turn evicting valid blocks. > Also, > if a valid block was in LRU, that block could be evicted rather than a block > in > I. This patch adds in the missing deallocations. > > Testing > > I ran the ruby random tester with 4 cores and did not encounter any errors > caused by this patch. I also ran some multiprogrammed workload mixes; > sim_ticks can be misleading for multiprogrammed workloads so take this with a > large grain of salt, but the simulated runtime was consistently 20-30% less. > For ruby tester (+increased cache size+assoc to cause the wrong blocks > getting evicted) I see more like 2% improvement. > > Diffs > > - src/mem/protocol/MOESI_hammer-cache.sm (5fe05690d03d) > > View Diff <http://reviews.gem5.org/r/2978/diff/> > _______________________________________________ gem5-dev mailing list [email protected] http://m5sim.org/mailman/listinfo/gem5-dev
