Hi everyone,

I've been trying to run some benchmarks from Rodinia on GEM5, syscall
emulation mode. Those are applications in OpenMP. Whenever I run these
benchmarks with 2 threads (allocating 2 CPUs), everything works just fine.
One example of a command line I've been using is the following:

*gem5/build/X86/gem5.debug --debug-flags=Exec
--debug-file=trace_Backprop.out.gz gem5/configs/example/se.py
--cpu-type=DerivO3CPU --caches --mem-size=1024MB --num-cpus=2 -c
gem5/tests/benchmarks/parallel/backprop -o 2048*

However, using the same binary, when I try to increase the number of thread
to 4, 8,.. the simulation ends unexpectedly, in the middle of the
application execution. In all applications I've ran into this problem
(using other benchmarks from Rodinia, for example, and even applications
from other benchmarks) the simulation always ends when the following OpenMP
function is executing:

*@gomp_team_barrier_wait_end*

I'm under the impression that GEM5 simulations "thinks" that it got into a
deadlock, and it ends the simulation prematurely, because this functions
executes for a long time, specially in the case of 4 or more threads. Am I
right assuming that?

I would appreciate a lot if someone could shed some light in this issue.

Thanks in advance,

-- 

*Thiago Dadalt Souto*
Porto Alegre, RS. Brasil.


Engenheiro Eletricista | *Schneider Electric* | R&D Embedded Software
Developer

Phone: 55 91494247 | 51 2104-2877

E-mail: [email protected]
_______________________________________________
gem5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users

Reply via email to