This discussion has been locked.
You can no longer post new replies to this discussion. If you have a question you can start a new discussion

Energy Profiling of an ARM7 core

Note: This was originally posted on 8th July 2013 at http://forums.arm.com

Hi

I want to make energy profiling and analysis of a ARM7 core. I have some doubts regarding this. May be my doubts are silly.

#1 :: I don't have a development board with myself and so, I intend to make a virtual simulation of the core. I have a DS-5 community edition installed in my windows 7 system.
     Please tell me whether I would be able to do a virtual Simulation with an external hardware connected?. I came to know that about the ARMulator. Can it be invoked in the DS-5 for a virtual simulation of a ARM7.? If it can be,what should I do to use the ARMulator.? Firstly , can the DS-5 be used for the above mentioned purpose?

#2 ::  How can I test the core by using Benchmark codes like Dhrystone marking in DS-5?

Regards,
RamPrasadh N.
  • Note: This was originally posted on 10th July 2013 at http://forums.arm.com

    I don't really understand what you mean by "virtual Simulation with an external hardware connected", so I may be missing the point.

    #1:  The full version of DS-5 comes with an FVP (Fixed Virtual Platform) model of a quad core Cortex-A9.  You can use it to run code built for ARM7 (although I'm not sure if the old BE8 is supported by Cortex-A9).  But the model does not do cycle counts nor power.  The model is not included in the Community Edition, you'll need to get the eval of the full version.  There is a separate Fast Models product that can produce custom models (different processors, different peripherals) that DS-5 can connect to.  DS-5 does not support ARMulator.

    If you have a real hardware running Linux or Android and you have (or can make) an appropriate power test point, you can use Streamline and the ARM Energy Probe (or a National Instruments DAQ) to make power measurements.

    #2: The full version of DS-5 includes a number of examples which you could use to as a basis for building a benchmark.

    Note that Community Edition does not include bare-metal (no OS) use; it only supports targets running Linux or Android.  The full version can do bare-metal, too.
  • Note: This was originally posted on 10th July 2013 at http://forums.arm.com

    I am sorry. It is "without an external hardware".

    I will actually be needing a model which can do a cycle count and energy profiling for the ARM7 core. I came to know that, an RVISS emulation can be run on a RVDS4.1 professional suite for ARM7 core. But the problem is RVDS does not have evaluation version.  I think if I can have a RVDS4.1 suite , them I may be able to do an energy profiling for an ARM7 core.

    Please Correct me if I am wrong.
  • Note: This was originally posted on 11th July 2013 at http://forums.arm.com

    For a simple enough system (ARM7TDMI, synchronous clocking, no cache, no MMU, no interrupts, perfect memory) you might be able to get cycle accurate simulation (without profiling) from RVDS 4.1; but RVDS 4.1 doesn't have any energy profiling features.
  • Note: This was originally posted on 11th July 2013 at http://forums.arm.com

    So , Can I take from your words that, If I simulate a code for ARM7 in RVDS,I can get the Number of cycles of instructions for the executed code?
    Is there any other possible way to do an energy profiling for ARM7?

    I do have an other query here.

    How can I calculate the current consumption of  an ARM core by simulation.? If I have a processor running on ARM core, I may be able to dis-able all the peripherals, in software and measure the supply current through a meter. It may be approximated to the core current consumption.Correct me If I am wrong. But, will I be able to do it through simulation?
  • Note: This was originally posted on 11th July 2013 at http://forums.arm.com

    [color=#222222][font=Arial, Helvetica, sans-serif][size=2]> Is there any other possible way to do an energy profiling for ARM7? [/size][/font][/color]

    Well, the ISSM won't tell you anything useful about what real implementations do. Not all instructions consume the same power. Arithmetic vs instructions which touch memory for example are totally different. Netlist-level  power simulation tools exist, but they are very slow and need good input test vectors. Failing that get a real SoC and a power meter like a Streamline power probe, or National Instruments DAQ, and measure power on a real device ...


    [color=#222222][font=Arial, Helvetica, sans-serif][size=2]> But, will I be able to do it through simulation? [/size][/font][/color]
    [color=#222222][font=Arial, Helvetica, sans-serif][size=2]
    [/size][/font][/color]
    Define "simulation". Transistor toggling in full logic level RTL simulation has some correlation with analogue power sim, but it is not easy, and not always accurate. Analogue level power simulation is useful, but very slow. It's not an easy job with a trivial "just use this tool" answer I'm afraid, it [size=2]really[/size] depends what you are trying to do.