Hi,

Is this a modified Cortex-A9? If not:

I'd recommend:

* setting the opLat of instructions as the ARM manual (forward cases),
specially SimdFloatMultAcc which is not 1 at all.
* MemRead and MemWrite are 1 in my opinion (1+Dcache hit latency ~= 2 for
A9 for basic ldr, str).
* Widths as being 2 can roughly estimate the perf of an A9 (in reality,
there's no ROB, but it may quad-issue in some cases, issueWidth=2 is then a
good average for the A9).
* wbDepth for me should be max(opLat/issueLat).
* L1 hit latencies are 1 or 2, L2 = 8 cycles (common values, as ARM says)


Hope it helps,

--
Fernando A. Endo, PhD student and researcher

Université de Grenoble, UJF
France



2014-03-24 12:14 GMT+01:00 Stian Hvatum <[email protected]>:

> Hi,
> I am trying to model a single Cortex-A9 as featured in the Exynos4412
> using this configuration: http://pastebin.com/t1AU4D7H
>
> I wonder if someone else has modelled a similar processor or has any
> knowledge about the parameters, and is able to help me with the
> configuration. Using the current configuration I get some performance
> discepancy compared to the real hardware. The configuration string used to
> run gem5 (stable) is the following:
>
> {..}gem5.opt --debug-flags=Cache,MemoryAccess,Exec,-ExecSymbol
> --remote-gdb-port=0 --debug-file=trace.out -d m5out/add-adc
> {..}configs/example/se.py -c bin/binary --cpu-type=exynos_4412p
> --mem-type=LPDDR2_S4_1066_x32 --sys-clock=400MHz --cpu-clock=1700MHz
> --num-l3caches=0 --caches --l2cache --l2_assoc=16 --l1d_size=32kB
> --l2_size=1MB --mem-size=2048MB
>
> Small loops with integer operations are slightly faster on hardware (~
> 25%), while general benchmarks are slightly faster in the simulator (~
> 25%). I wonder if this might be due to memory misconfiguration and that the
> Fast-loop-mode featured in the Cortex-A9 seems to be missing from gem5?
>
> Best regards,
> Stian Hvatum
> _______________________________________________
> gem5-users mailing list
> [email protected]
> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
>
_______________________________________________
gem5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users

Reply via email to