----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: http://reviews.gem5.org/r/3315/#review7991 -----------------------------------------------------------
It would be nice to get input on the buffer counting: It appears that the total buffering limit (buffersFull()) may be mixing different types/sizes of buffers. Does that check seem ok? - Joel Hestness On Feb. 8, 2016, 7:58 p.m., Joel Hestness wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > http://reviews.gem5.org/r/3315/ > ----------------------------------------------------------- > > (Updated Feb. 8, 2016, 7:58 p.m.) > > > Review request for Default. > > > Repository: gem5 > > > Description > ------- > > Changeset 11331:8492ca5ea301 > --------------------------- > mem: [DRAFT] Make DRAMCtrl response queue finite > > The DRAMCtrl had an effectively infinite response queue by using the > QueuedSlavePort to manage responses. Unfortunately, for high-bandwidth > applications, this queue causes unrealistic bloat, quickly filling with more > than 100 packets. > > To fix this issue, change the response port from a QueuedSlavePort back to a > standard SlavePort, and implement the appropriate control flow. This includes > adding a backend queue, in which slots are reserved when packets arrive at on > the DRAMCtrl's request path. > > > Diffs > ----- > > src/mem/dram_ctrl.cc 10647f5d0f7f > src/mem/DRAMCtrl.py 10647f5d0f7f > src/mem/dram_ctrl.hh 10647f5d0f7f > > Diff: http://reviews.gem5.org/r/3315/diff/ > > > Testing > ------- > > Tested running with gem5's memtest.py: > > ../build/X86/gem5.opt --debug-flag=DRAM --outdir=$outdir > ../configs/example/memtest.py -u 100 > > and a couple hundred varying GPU workloads in gem5-gpu. > > > Thanks, > > Joel Hestness > > _______________________________________________ gem5-dev mailing list [email protected] http://m5sim.org/mailman/listinfo/gem5-dev
