Dear all
I simulated different benchmarks of Spec CPU2006 by giving different
frequencies to each of 8 CPUs.
By simply modifying the file 'se.py' as shown below, I could give each CPU
a different frequency.
-------------------------------------------------------------
i = 0
num = '1.27' # CPU0 freq
# All cpus belong to a common cpu_clk_domain, therefore running at a common
# frequency.
for cpu in system.cpu:
freq = num + 'GHz'
print freq
system.cpu_clk_domain_fs = SrcClockDomain(clock = freq,
voltage_domain =
system.cpu_voltage_domain)
if i == 0 :
num = '0.86' # CPU1 freq
if i == 1 :
num = '0.51' # CPU2 freq
if i == 2 :
num = '2.13' # CPU3 freq
if i == 3 :
num = '3.1' # CPU4 freq
if i == 4 :
num = '2.6' # CPU5 freq
if i == 5 :
num = '4' # CPU6 freq
if i == 6 :
num = '2.59' # CPU7 freq
i = i + 1
cpu.clk_domain = system.cpu_clk_domain_fs
----------------------------------------------------------------
I also simulated when 8 CPUs had the same frequency of 4 GHz.
A strange point was found when comparing the results of two simulations.
When the frequencies of 4GHz and 1.27GHz were applied to the same core, the
'numCycles' (number of cpu cycles simulated) was 352,523,502 and
115,626,606 respectively.
I think, the 'numCycles' seems to have an appropriate value proportional to
frequency.
However, the value of each 'committedInsts' (Number of Instructions
Simulated) was 26,958,308 and 52,761,444, and the lower frequency was
higher. (Int insts, fp_insts, commit.branches, branchMispredicts,
commit.loads .... etc. were also higher in the lower frequencies.)
Moreover, values of other int, fp instruction, or values of int,
fp_alu_accesses, etc. are higher in the lower frequency.
I think there is definitely a problem with the value coming out like this.
I checked the 'config.ini' file and thought that the frequency was entered
correctly.
I also checked the source files to see how the value of the problem
parameter increases.
But I could not figure out why this was happening.
If you are aware of the cause of these problems, please let me know how I
can solve this problem.
Thanks for reading.
_______________________________________________
gem5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
