Apologies for replying To add, I also tried to add 1GB in DirectoryMemory.cc constructor as proposed by Lena Olson here to fix the issue https://www.mail-archive.com/gem5-users%40gem5.org/msg11106.html
However I get an assertion failure in mem/abstract_mem.cc void AbstractMemory::access(PacketPtr): Assertion `AddrRange(pkt->getAddr(), pkt->getAddr() + (pkt->getSize() - 1)).isSubset(range)' failed Regards, Adarsh On Fri, Dec 18, 2015 at 12:56 AM, Adarsh Patil <[email protected]> wrote: > Hello, > I am trying to simulate x86 system with memory greater than 3GB using > Ruby hierarchy and MOESI_hammer protocol. > As x86 has an I/O gap between 3 to 4GB, two memory controllers are created. > > However by default the num-dirs is taken as 1 and I get the following > assertion error during bootup in file mem/ruby/structures/ > DirectoryMemory.cc > Assertion `idx < m_num_entries' failed > > Specifying the --num-dirs=2, I run into the following error in > file base/addr_range.hh before bootup even begins > fatal: Got 1 ranges spanning 1 interleaving bits > > > I noticed that in MOESI_hammer.py the create_system() function splits the > phys_mem_size into mem_module_size without knowledge of the I/O gap as below > mem_module_size = phys_mem_size / options.num_dirs > > I fixed this to the right values but this still does not fix the > interleaving bits issue. > > Also, when running the system with 3GB memory and num-dirs=2 (which works > well), I observe that even though the interleaving and hashing bits are > initially set in MemConfig.py, when executing src/mem/xbar.cc these bits > are set to 0 (disabled). > > What is the right way to run Ruby with multiple directories and memory > greater than 3GB? > How are the interleaving and hashing (xor) bits used? > > Thanks in advance. > > -- > *Regards,* > *Adarsh Patil* > *Indian Institute of Science* > *Bangalore, India* > TimeWarp <http://timewarp.adarsh.space> > -- *Regards,* *Adarsh Patil*TimeWarp <http://timewarp.adarsh.space>
_______________________________________________ gem5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
