Measure CPU % (percent) usage?

Brian,

You are using CCS3.3, aren't you? CCS3.3 will tell you CPU cycle counts between two break points. I think you may be able to get total cycle count that you want by using this feature (called clock) .

How to enable it:

1) Under "Profile" menu, go to Clock -> Enable

2) Under "Profile" menu, go to Clock -> View

3) Under "Profile" menu, go to Clock -> Setup

-  Select: cycle.CPU

- Select: Auto (reset option)

You will see cycle count on bottom right of the CCS window.

Good luck!

Regards,

Peter Chung

Hi Peter,

I'm not sure where to put my breakpoints.  I have two ADS chips feeding two McBSP ports and two DMA callbacks (and have enabled both the HALF and BLOCK interrupts for each).  Then I have two USB endpoints, one for each ADS, serviced from one USB callback interrupt.  I'm not sure I can find one place in my code where I can guarantee all code will be executed between two points.  I could measure any one of four or five key routines, but I don't quite know where to measure everything.

I have a main loop where I check four global flags (HALF and BLOCK for each of the two DMA channels) and conditionally run service code if the global is set.  Then I have the USB callback for PSOF which runs outside the main loop (but perhaps this is negligible).  I think the key is that I have nothing to synchronize my main firmware loop to the USB frame cycle.

Would DSP/BIOS somehow make this easier to measure? ... or does DSP/BIOS still use flags and/or events to service DMA and USB?

P.S.  I'm going to follow your instructions above and try the cycle count inside my main loop, and see how that works.  I predict that I will get different CPU cycle counts, especially if my loop repeats frequently while waiting for a global event to fire.

Brian,

Yes, I see your frustration. I have checked some built-in DSP_BIOS features in CCS3.3 but they does not seem to work. Let me ask around.

Regards,

Peter Chung

Hi Peter,

I'm back to the question of calculating my total CPU % usage with multiple tasks running in parallel. Did you ever ask around on this topic?

Brian

Hi,

CCS tool does not provide a good estimated numbers. So we created using timer and provide cycle numbers of  cpu active or idle time.

//================================

#include "csl_gpt.h"

// Global variables

CSL_GptObj gptObj;

CSL_Handle    hGpt;

CSL_Status    status;

CSL_Config    hwConfig;

CSL_GptObj gptObj1;

CSL_Handle    hGpt1;

CSL_Config    hwConfig1;

CSL_GptObj gptObj2;

CSL_Handle    hGpt2;

CSL_Config    hwConfig2;

Uint32 timeCnt1,timeCnt2;

Uint32 timeCnt3=0;

Uint32 timeCnt4=0;

/**

 *  \brief  Function to calculate the system clock

 *

 *  \param    none

 *

 *  \return   System clock value in Hz

 */

void run_GPT(unsigned int flag);

void run_GPTidle(unsigned int flag);

void main(void) {

///////////////////////////////////////////////////////

// setup timer 0

///////////////////////////////////////////////////////

status = 0;

hGpt = GPT_open (GPT_0, &gptObj, &status);

GPT_reset(hGpt);

/* Configure GPT module */

hwConfig.autoLoad = GPT_AUTO_ENABLE;

hwConfig.ctrlTim = GPT_TIMER_ENABLE;

hwConfig.preScaleDiv = GPT_PRE_SC_DIV_0; // divide by 2

hwConfig.prdLow = 0xFFFF;

hwConfig.prdHigh = 0xFFFF;

GPT_config(hGpt, &hwConfig);

///////////////////////////////////////////////////////

// setup timer 1

///////////////////////////////////////////////////////

status = 0;

hGpt1 = GPT_open (GPT_1, &gptObj1, &status);

GPT_reset(hGpt1);

/* Configure GPT module */

hwConfig.autoLoad = GPT_AUTO_ENABLE;

hwConfig.ctrlTim = GPT_TIMER_ENABLE;

hwConfig.preScaleDiv = GPT_PRE_SC_DIV_0; // divide by 2

hwConfig.prdLow = 0xFFFF;

hwConfig.prdHigh = 0xFFFF;

GPT_config(hGpt1, &hwConfig);

run_GPT(1); // enable timer

    status = your_function1;

run_GPT(0); // disable timer

run_GPTidle(1);

 your_function2;

  run_GPTidle(0);

}

void run_GPT(unsigned int flag) {

  if (flag) {

    GPT_start(hGpt);

    IRQ_globalDisable();

    GPT_getCnt(hGpt, &timeCnt1);

    IRQ_globalEnable();

  } else {

    IRQ_globalDisable();

    GPT_getCnt(hGpt, &timeCnt2);

    IRQ_globalEnable();

    GPT_stop(hGpt);

    printf("exec cks = %ld   idle cks = %ld\n",2*(timeCnt1-timeCnt2),2*(timeCnt3-timeCnt4));

  }

}

void run_GPTidle(unsigned int flag) {

  if (flag) {

    GPT_start(hGpt1);

    IRQ_globalDisable();

    GPT_getCnt(hGpt1, &timeCnt3);

    IRQ_globalEnable();

  } else {

    IRQ_globalDisable();

    GPT_getCnt(hGpt1, &timeCnt4);

    IRQ_globalEnable();

    GPT_stop(hGpt1);

  }

}

I am working with the C5506, which does not have the GPT peripheral.

Thanks for the reply, Hyun. It looked very promising at first. I had never noticed the GPT Module before. I quickly looked in the CSL documentation and see that the GPT Module is available on the C5501/C5502 DSPs.