MIPS Calculation on ARMv7

Hi,

It can't be 1 MIPS when running at 1 MHz on scalar CPU. For example, take ARM9E assembly code where there various types of instructions like load, multiply, add etc. Consider Multiply(SMULL) will take 3 cycles i.e. only one instruction can be executed in 3 cycles.

My intention of asking MIPS calculation is that to know how many instructions it can execute in one second(without considering memory latencies, interlock dependencies, dual issue etc). Is there any way that I can calculate this using small sub routines?

Can you explain me what is Dhrystone? How to obtain these numbers?

I have sample codes like this:

main(int argc, char *argv[])

{

     // Creation()

    // Initialization()

    while(1)

    {

          // memory/file read of data

         // Process_Data()

          // memory/file write of data

    }

    // DestroyMemory()

}

I want to find out how many instructions has been executed for the Process_Data() function in a second?

Thanks and Regards

Sridhar Artham

Hi,

It can't be 1 MIPS when running at 1 MHz on scalar CPU. For example, take ARM9E assembly code where there various types of instructions like load, multiply, add etc. Consider Multiply(SMULL) will take 3 cycles i.e. only one instruction can be executed in 3 cycles.

My intention of asking MIPS calculation is that to know how many instructions it can execute in one second(without considering memory latencies, interlock dependencies, dual issue etc). Is there any way that I can calculate this using small sub routines?

Can you explain me what is Dhrystone? How to obtain these numbers?

I have sample codes like this:

main(int argc, char *argv[])

{

     // Creation()

    // Initialization()

    while(1)

    {

          // memory/file read of data

         // Process_Data()

          // memory/file write of data

    }

    // DestroyMemory()

}

I want to find out how many instructions has been executed for the Process_Data() function in a second?

Thanks and Regards

Sridhar Artham

Hi,

to fulfill your intention any sort of benchmarks would be useless.
You had better measure the execution time by using the built-in performance counter.

Best regards,
Yasuhiko Koumoto.

Hi,

Do you have information which counter should be used to get the instruction count information?

If you see, ARM IDE, after running the code, we get analysis report. Basically it contains, Instructions, Average Instructions per Second, Cycles etc.

I want to calculate Cycles using some sub routines. Can you suggest which is the counter used to calculate that cycle information?

thanks and Regards

Sridhar Artham

Hi,

I think your expectations will be only realized when you will use ICE.
Almost ARM IDE would not have such flexible features.
Regarding Performance Monitoring Unit, I cannot believe other than the cycle counter.

Best regards,

Yasuhiko Koumoto.