Some rtixxxx functions not always working

Hi Paul,

can you try this?

rtiResetCounter(rtiCOUNTER_BLOCK0);
rtiStartCounter(rtiCOUNTER_BLOCK0);

CodeToBenchmark();

rtiStopCounter(rtiCOUNTER_BLOCK0);

countervalue = rtiGetCurrentTick(rtiCOMPARE0);

BenchmarkTime=countervalue *(10E-6);  // 1 tick corresponds to 1 us.

I get similar 'random' results with this code suggestion.

that code sequences worked fine for us in the past. 

can you send your project in the private email? or in public forum?

You can replace your proprietary codes by a fixed for-loop inside.

Hi,

I basically used the demo that came with it.

Added these functions:

void start_time()

{

 rtiResetCounter(rtiCOUNTER_BLOCK0);

 rtiStartCounter(rtiCOUNTER_BLOCK0);

}

UINT32 elapsed_time()

{

 static UINT32 Elapsed_Count;

 rtiStopCounter(rtiCOUNTER_BLOCK0);

 Elapsed_Count = rtiGetCurrentTick(rtiCOMPARE0);

     return Elapsed_Count;

}

void main(void)

{

#define DEVICEID (*(volatile unsigned int *)0xFFFFFFF0) 

//unsigned int i;

/* USER CODE BEGIN (3) */

/** - Call RUN LED routine during the startup, signal PROC ready */

run_LED_StartUp();

rtiInit();

Test_Benchmark();

// /** - Enable IRQ and FIQ interrupt in Cortex R4 */

// swi_enable_fiq_irq();

//...

}

void delay_1ms(void) 

{

 static volatile unsigned int delayval;

 delayval = 10532; // about 1ms

 while(delayval--);

}

UINT32 Calibration_Time[MAX_CALIBRATION];

extern "C" void Test_Benchmark(void)

{

 int C, D;

 for (C = 0; C < MAX_CALIBRATION; C++)

 {

 start_time();

 for (D = 0; D < (C + 1); D++)

 {

 delay_1ms();

 }

 Calibration_Time[C] = elapsed_time();

 }

}

Hi,

Did you get a chance to use the code I submitted?

Thanks.

Paul,

Not yet, sorry, i have been busy working on other forum questions.  Will try out sometimes today and let you know...

Ok.  Hopefully you can find out why I am getting random results.  It is not pratical to do some benchmark using a stop watch...

have you consider using the internal PMU (Performance monitor Unit) built into Cortex R4?  This will give you the exact cycle counts.

We use this for our performance measurement.

Paul,

The cortex R4 includes an internal counter for timestamping that is more accurate than the RTI.

If you are using HalCoGen the file 'sys_pmu.asm' includes functions to start/stop/read/reset the counters.

The counters in the PMU can count quite a lot of different events but there's also a simple cycle counter for benchmarking.

You can also try this out through CCS by poking around in the register window to see how it works:

http://e2e.ti.com/support/microcontrollers/hercules/f/312/p/260549/911524.aspx#911524

-Anthony

Hi,

For your performance measurement,please take a look at this thread: https://e2e.ti.com/support/microcontrollers/hercules/int-hercules/f/365/t/236801.aspx

let me know if you plan to switch to PMU so that i wont try out the RTI codes (i am not trying RTI codes now until i got confirmation from you)

my recommendation is to use PMU because you can have more option regarding profiling your codes.

The link you gave me does not work for me.

I tried using the PMU function but with CCS I just got linker unresolved externals:

undefined first referenced

symbol in file

--------- ----------------

_pmuGetCycleCount_() <whole-program>

_pmuResetCycleCounter_() <whole-program>

_pmuEnableCountersGlobal_() <whole-program>

_pmuInit_() <whole-program>

I don't mind using either RTI or PMU, whichever works.

Paul,

Both links above worked for me, but maybe different inside TI. which one is broken from your view?

Regarding the PMU functions is the sys_pmu.asm part of your project?

-anthony

Paul,

the link Anthony mentioned should give you the interactive way of doing PMU, but if you want to do profiling using codes, please see the below comments.

i am attaching sys_pmu.asm and sys_pmu.h that you should have in your halcogen project.  I used these two files to profile cycle counts and they work for me.

One caveat to look for is that the earlier version of Halcogen generated sys_pmu.h has a bug in defining of a defined variable pmuCOUNTER0.

Correct definition should be:

#define pmuCOUNTER0 0x80000001U.   so please check yours or use the attached file if you want.

In your main.c, ensure these hooks are in:

#include "sys_pmu.h"

main function:

  _pmuInit_();

_pmuEnableCountersGlobal_();

Useful functions for cycle counts:

 // Reset and restart PMU 

_pmuResetCounters_();
_pmuStartCounters_(pmuCOUNTER0);

// get pmu cycle

_pmuGetCycleCount_();

you can do different kinds of profiling: cycle count, event count

let me know if this works for you.

4454.sys_pmu.asm

1513.sys_pmu.h

We strongly recommend you to use PMU.  

Ok it was called from cpp so I had to add  'extern "C" void ...' to link.

The funny thing is, I get random results with PMU too!

I am starting to wonder if the Code Composer Studio optimization settings are causing this. 

I chose 'Optimization level = 4' and 'Optimize for speed = 5'.

humm!  this is interesting.

i would recommend you to look at the dis-assembly codes for this nested for loops and see if it makes sense.

can you try compiler version 5.1 available for production release?

Thanks a lot, your suggestions seem to work. 

Although I did have to change the optimization settings to prevent 'random' values. 

Are you aware of what is the best settings for speed that won't cause problems?

Paul,

i have not run into this random values on optimization setting yet.

can you please confirm if you have the same issue on latest compiler version 5.1?  

I can forward this to the compiler team in TI that they can take a look if you can tar up your project, i am sure they can figure this out.

Hi Henry,

I only have version 4.2.4.

When I read the performance counter, is there an exact way to convert the count to microseconds? 

My demo ARM RM48L950 board software runs at 160MHz.

Hi Paul,

The PMU counted in term of CPU clock.  So, just multiply the cycles with the period to get your microsecond.

You can go to help tab in CCS and scroll down to update new software.

There is an "work with" tab and select code Generation Tool Updates.

Then, you will see below window appears TI Compiler Updates.  Click on that to expand it.

You will see ARM compiler version 5.1.  Check that box and follow instruction to install it...

Hi Henry,

Thanks again for your support.

I tried the software upgrade but it fails to see any upgrades because the web site access looks blocked by our network.

Paul,

Try this link: http://processors.wiki.ti.com/index.php/Compiler_Releases to download manually...

Hi Henry,

I could not find any Code Composer Studio to download from the page you sent me.

It says 'Go Here' but I can't see it on that page either: https://www-a.ti.com/downloads/sds_support/TICodegenerationTools/download.htm

Hi Henry,

Ok I found and downloaded it but it won't install for the same network limitations reasons....

http://processors.wiki.ti.com/index.php/Download_CCS

Hi Paul

the link you sent is the CCS download, but that CCS 5.4 "may not" come with compiler 5.1

if you want to download it, use offline install option (download more, but maybe the best option for your company firewall thing...)

I am checking with the compiler person in TI for you on compiler 5.1.

The link i sent to you yesterday has DSP, C2000 compiler on there, but i could not find ARM compiler.

Will get back to you...

Hi Henry,

I hope the new CCS solves a lot of issues.  Right now here is my list in case you have solutions for me:

When l click 'Debug Active Projet', the loading takes >3 minutes to complete.

Breakpoints do not always toggle on/off.  Perhaps there is a limit or they can toggle while the program is running?

Sometimes I hit breakpoint that should not.  Like I put an 'if iteration = 200  do something' and put a breakpoint on the do something.  But the debugger hits the breakpoint when iteration = 0?

Sometimes the program seems to get lost and stops at some interrupt handlers like: _c_int00, _prefetchAbort

Stepping in the debugger takes around 2 seconds per step, even on very simple assignment statements.

Using optimization (or thumb 16?), I get random results.

Paul,

i forwarded your issue to the software compiler team.  They said that for some legal reason, the ARM compiler from TI has to be updated from CCS, but they are checking again.

I have the single download version with me, but i can not send it to you even in private link.

Is there any possibility that you can use a guess network in your company to install newer version of compiler?

As far as the breakpoint, there are only around 6 supported.

I will forward this to our compiler support team to see if they can give any suggestion, but please try to download 5.1 from your network somehow first.