Measure executing time on ARM Cortex-A8 using hardware counter

问题

I'm using a Exynos 3110 processor (1 GHz Single-core ARM Cortex-A8, e.g. used in the Nexus S) and try to measure execution times of particular functions. I have an Android 4.0.3 running on the Nexus S. I tried the method from

[1] How to measure program execution time in ARM Cortex-A8 processor?

I loaded the kernel module to allow reading the register values in user mode. I am using the following program to test the counter:

static inline unsigned int get_cyclecount (void)
{
    unsigned int value;
    // Read CCNT Register
    asm volatile ("MRC p15, 0, %0, c9, c13, 0\t\n": "=r"(value));
    return value;
}


static inline void init_perfcounters (int do_reset, int enable_divider)
{
    // in general enable all counters (including cycle counter)
    int value = 1;

    // peform reset:  
    if (do_reset)
    {
        value |= 2;     // reset all counters to zero.
        value |= 4;     // reset cycle counter to zero.
    } 

    if (enable_divider)
        value |= 8;     // enable "by 64" divider for CCNT.

    value |= 16;

    // program the performance-counter control-register:
    asm volatile ("MCR p15, 0, %0, c9, c12, 0\t\n" :: "r"(value));  

    // enable all counters:  
    asm volatile ("MCR p15, 0, %0, c9, c12, 1\t\n" :: "r"(0x8000000f));  

    // clear overflows:
    asm volatile ("MCR p15, 0, %0, c9, c12, 3\t\n" :: "r"(0x8000000f));
}


int main(int argc, char **argv)
{
    int i = 0;
    unsigned int start = 0;
    unsigned int end = 0;

    printf("Hello Counter\n");

    init_perfcounters(1,0);

    for(i=0;i<10;i++)
    {
        start = get_cyclecount();
        sleep(1); // sleep one second
        end = get_cyclecount();

        printf("%u %u %u\n", start, end, end - start);
    }

    return 0;
}

According to [1] the counter is incremented with each clock cycle. I switched the scaling_governor to userspace and set the CPU frequency to 1GHz to make sure that the clock frequency is not change by Android.

If I run the program the sleeps of 1 second are executed, but the counter values are in the range of ~200e6, instead of the expected 1e9. Is there anything processor specific I am missing here? Is the clock rate of the counters different to the clock rate of the processor ?

回答1:

Check out this professor's page: http://users.ece.utexas.edu/~valvano/arm/ He has multiple full example programs that have to do with time/periodic-timers/measuring-execution-time, they are developed for ARM Cortex-M3 based microcontrollers. I hope this isn't very different from what you are working on. I think you would be interested in Performance.c

回答2:

Are you sure governors are used in Android for performance management the same way that in standard Linux? And are you using custom Android image or one provided by manufacturer? I would assume there are lower level policies in place in manufacturer provided image (tied to sleeps or modem activity, etc). It could be also that sleep code directly scales voltage and frequency. It might be worthwhile to disable the whole CPUFreq not just the policies (or governors).

来源：https://stackoverflow.com/questions/9795132/measure-executing-time-on-arm-cortex-a8-using-hardware-counter