Hi Biswabandan, Is there any chance you could dig a bit deeper here?
My suspicion is that the DRAM read queue is filling up and forcing packets to wait. The controller only counts reads when they are actually serviced. Perhaps the prefetcher counts issues prefetches as it creates the packets, even if they are not successfully sent? Check numRdRetry and the rdQLenPdf. Thanks, Andreas From: biswabandan panda via gem5-users <[email protected]<mailto:[email protected]>> Reply-To: biswabandan panda <[email protected]<mailto:[email protected]>>, gem5 users mailing list <[email protected]<mailto:[email protected]>> Date: Monday, 30 June 2014 12:19 To: gem5 users mailing list <[email protected]<mailto:[email protected]>> Subject: [gem5-users] DRAM Controller and Prefetchers Hi All, For a 2 level cache hierarchy with DDR3 (11-11-11) as the DRAM specs, the number of prefetch requests issued by the L2 prefetcher is high (very high in some cases) as compared to the number of prefetch requests responded by the DRAM controller. Any thoughts on it. The difference becomes almost negligible, when I use an L3 cache and turn on the prefetchers at the L2. In case of 3 level cache hierarchy, this equation holds true: pf_issued at the L2s = Pf hits at the L3 + Pf requests responded by the DRAM Here are the actual numbers for a 4 core setup with PARSEC benchmarks: blackscholes_4c_simlarge_ckpts.rcS.txt:system.mem_ctrls.num_reads::l2.prefetcher 130935 # Number of read requests responded to by this memory blackscholes_4c_simlarge_ckpts.rcS.txt:system.l2.prefetcher.prefetcher.num_hwpf_issued 148226 # number of hwpf issued canneal_4c_simlarge_ckpts.rcS.txt:system.mem_ctrls.num_reads::l2.prefetcher 346384 # Number of read requests responded to by this memory canneal_4c_simlarge_ckpts.rcS.txt:system.l2.prefetcher.prefetcher.num_hwpf_issued 354459 # number of hwpf issued dedup_4c_simlarge_ckpts.rcS.txt:system.mem_ctrls.num_reads::l2.prefetcher 2937725 # Number of read requests responded to by this memory dedup_4c_simlarge_ckpts.rcS.txt:system.l2.prefetcher.prefetcher.num_hwpf_issued 3251216 # number of hwpf issued facesim_4c_simlarge_ckpts.rcS.txt:system.mem_ctrls.num_reads::l2.prefetcher 3443990 # Number of read requests responded to by this memory facesim_4c_simlarge_ckpts.rcS.txt:system.l2.prefetcher.prefetcher.num_hwpf_issued 3639481 # number of hwpf issued ferret_4c_simlarge_ckpts.rcS.txt:system.mem_ctrls.num_reads::l2.prefetcher 2141141 # Number of read requests responded to by this memory ferret_4c_simlarge_ckpts.rcS.txt:system.l2.prefetcher.prefetcher.num_hwpf_issued 12161061 # number of hwpf issued freqmine_4c_simlarge_ckpts.rcS.txt:system.mem_ctrls.num_reads::l2.prefetcher 115625 # Number of read requests responded to by this memory freqmine_4c_simlarge_ckpts.rcS.txt:system.l2.prefetcher.prefetcher.num_hwpf_issued 168290 # number of hwpf issued rtview_4c_simlarge_ckpts.rcS.txt:system.mem_ctrls.num_reads::l2.prefetcher 590098 # Number of read requests responded to by this memory rtview_4c_simlarge_ckpts.rcS.txt:system.l2.prefetcher.prefetcher.num_hwpf_issued 863745 # number of hwpf issued swaptions_4c_simlarge_ckpts.rcS.txt:system.mem_ctrls.num_reads::l2.prefetcher 20308 # Number of read requests responded to by this memory swaptions_4c_simlarge_ckpts.rcS.txt:system.l2.prefetcher.prefetcher.num_hwpf_issued 16689699 # number of hwpf issued vips_4c_simlarge_ckpts.rcS.txt:system.mem_ctrls.num_reads::l2.prefetcher 3361213 # Number of read requests responded to by this memory vips_4c_simlarge_ckpts.rcS.txt:system.l2.prefetcher.prefetcher.num_hwpf_issued 3542000 # number of hwpf issued x264_4c_simlarge_ckpts.rcS.txt:system.mem_ctrls.num_reads::l2.prefetcher 1060135 # Number of read requests responded to by this memory x264_4c_simlarge_ckpts.rcS.txt:system.l2.prefetcher.prefetcher.num_hwpf_issued 1805457 # number of hwpf issued [https://ssl.gstatic.com/ui/v1/icons/mail/images/cleardot.gif] -- thanks®ards BISWABANDAN http://www.cse.iitm.ac.in/~biswa/<http://www.cse.iitm.ac.in/%7Ebiswa/> “We might fall down, but we will never lay down. We might not be the best, but we will beat the best! We might not be at the top, but we will rise.” -- 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-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
