Hi All,

 I am trying to get selective traces for few stages in ALPHA_FS O3 system. In 
order to do that, I 
have enabled --debug-flags=Exec (with turning-off the non-required 
switches) and getting the intended trace section after linux boot. I am running 
the benchmarks from PARSEC suite and for experimental purposes, the maxinsts is 
set to 200M. Further in order to attach each thread with a particular core, I 
am using the GOMP_CPU_AFFINITY option in my rcS script. As shown below, this 
rcS script attempts to run 4 threads of streamcluster benchmark in 4core cpu.

#!/bin/sh

# File to run the streamcluster benchmark

cd /parsec/install/bin

#/sbi/m5 switchcpu
export GOMP_CPU_AFFINITY="0 1 2 3"
/sbin/m5 dumpstats
/sbin/m5 resetstats
echo "Simulation Beginning :D"
./streamcluster 10 20 32 4096 4096 1000 none 
/parsec/install/inputs/streamcluster/output.txt 4
echo "Done :D"
/sbin/m5 exit
/sbin/m5 exit 

However, in my generated traces (deep into the ROI region), all the time I can 
see traces for only thread 0 running in different cores(cpu0, cpu1, cpu2, 
cpu3), as shown below, although in my streamcluster rcS script I set the 
threads count as 4. 

2394499819584: system.cpu0.BPredUnit: [tid:0] <some custom print msgs>
2394499820918: system.cpu1.BPredUnit: [tid:0] <some custom print msgs>
2394499824253: system.cpu0.BPredUnit: [tid:0] <some custom print msgs>
2394499827588: system.cpu2.BPredUnit: [tid:0] <some custom print msgs> 
2394499830923: system.cpu3.BPredUnit: [tid:0] <some custom print msgs>
2394499834258: system.cpu1.BPredUnit: [tid:0] <some custom print msgs>
2394499837593: system.cpu0.BPredUnit: [tid:0] <some custom print msgs>
2394499841595: system.cpu2.BPredUnit: [tid:0] <some custom print msgs>
2394499844930: system.cpu3.BPredUnit: [tid:0] <some custom print msgs>

I am observing the same for other benchmarks also. Is it due to the limited 
maxinsts (as I set that to 200M) or anything I am missing?

Thanks,
Dibakar Gope
PhD Student, UW-Madison
_______________________________________________
gem5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users

Reply via email to