Joe,
There is currently a memory leak in the simulated memory system related to
packets and requests that are created but never destroyed. The problem
usually manifests itself slowly and for short simulation runs doesn't
usually cause a problem. If the simulated memory system is designed in a
way such that it has a large number of writebacks occuring (small L1 cache
for example) the problem becomes larger more quickly and even short
simulations run into memory limitations. We are currently trying to
locate the source of the memory leak.
-Ron
On Fri, 9 Mar 2007, Joe Gross wrote:
Hello,
I'm trying to run several benchmarks, including spec2000 and stream
(http://www.cs.virginia.edu/stream/) on the 2.0b2 release, using g++ 3.4.3 as
the cross-compiler and gcc3.4.3,4.0.2,4.1.1 as the compiler of m5 and all
combinations have the same problem of the m5 binary progressively using more
and more memory, until it fills all the memory on the system(up to 1.7gb
now). I'm executing the se.py script using --caches and --detailed as well.
Memcheck shows these problems as well as others when running
tests/test-progs/hello/bin/alpha/linux/hello:
==16789== 1047232 bytes in 32726 blocks are definitely lost in loss record
369 of 374
==16789== at 0x1B904C07: operator new(unsigned) (vg_replace_malloc.c:133)
==16789== by 0x8114C77: CoherenceProtocol::CoherenceProtocol(std::string
const&, std::string const&, bool) (statistics.hh:398)
==16789== by 0x8115219: CoherenceProtocolBuilder::create() (param.hh:237)
==16789== by 0x8164822: SimObjectClass::createObject(IniFile&, std::string
const&) (builder.cc:136)
==16789==
==16789==
==16789== 2290960 bytes in 28637 blocks are definitely lost in loss record
372 of 374
==16789== at 0x1B904C07: operator new(unsigned) (vg_replace_malloc.c:133)
==16789== by 0x806E36F: Stats::Formula::Formula() (statistics.cc:65)
==16789== by 0x8187875: _GLOBAL__I_hostInstRate (stat_control.cc:52)
==16789== by 0x82CFEE4: (within m5.opt)
I'm wondering if I need to configure something differently, if the
cross-compiled binary (stream) could be doing this (although it makes no
malloc calls), if this is some issue with the compiler or if this is a known
problem. Any help is appreciated.
Joe
_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
