Hi all,
 
I have some questions about the tagged prefetcher. If I configure tagged 
prefetcher in my simulation system, will it increase the amount of memory 
accesses? If that is true, could I use a queue in PhysicalMemory to buffer lots 
of packets without sending a packet back immediately after it arrives? I have 
tried to buffer many packets after a long time execution, but the number of the 
buffered packets in queue is at most 4. Is that caused by the strong data 
dependency of benchmarks or the simulation mechanism of M5? By the way, I have 
tried many benchmarks and all the results are the same. I wonder if I could 
change the configuration file or some simulation mechanism to implement 
buffering lots of packets in PhysicalMemory.
 
Another question is about the configuring to use prefetcher. I use 
configs/example/se.py as my configuration file and I just modify the L1 cache 
model as follows:
 
if options.caches:
system.cpu[i].addPrivateSplitL1Caches(
L1Cache(size’32kB',prefetch_policy='tagged',prefetch_miss='true'), L1Cache(size 
= '64kB',prefetch_policy='tagged',prefetch_miss='true'))
 
Unfortunately, I get such information when I run M5 simulator:
  ………..    
 0: system.physmem: Write of size 8 on address 0x3802018 data 0x0
warn: Entering event queue @ 0. Starting simulation...
  1000: global: system.cpu0.icache:Found a pf canidate, inserting into prefetch 
queue
 ……………
 28500: global: system.cpu0.dcache:Found a pf canidate, inserting into prefetch 
queue
 28500: global: system.cpu0.dcache:Found a pf canidate, inserting into prefetch 
queue
 29000: system.physmem: Read of size 64 on address 0x9b140
 29000: global: system.cpu0.dcache:Marking a HW_PF in service
./run: line 17: 5371 Segmentation fault     build/ALPHA_SE/m5.opt 
--trace-flags="HWPrefetch,MemoryAccess" configs/example/se.py -d --caches -c 
/home/pj/M5/m5/build/ALPHA_SE/spec/CFP2000/172.mgrid/exe/mgrid_base.gcc33-high-opt
 -i /home/pj/M5/m5/build/ALPHA_SE/spec/CFP2000/172.mgrid/data/ref/input/mgrid.in
 
I have not modified source of M5, why does it get a segmentation faults? Is 
there anything wrong with my configuration file? Is the TAGGED prefetcher the 
only implemented prefetcher? Because when I use GHB and STRIDE and recompiled 
the source, I get lots of errors.
 
Could anybody help me?
 
Thank you very much in advance!
  
Lei Yang
_______________________________________________
m5-users mailing list
m5-users@m5sim.org
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users

Reply via email to