STM32H723: How to optimize summation of an array?

[yonatan]

Hi folks.

I am trying to optimize (by time) the following piece of code.

	for (uint32_t i = 6 + adc_data_index; i < 35 + adc_data_index; i++)
	{
		raw[0] += (adc_data[i]);
		raw[1] += (adc_data[i + 35]);
		raw[2] += (adc_data[i + 70]);
		raw[3] += (adc_data[i + 115]);
	}

For now it takes 3.5 micro-second at 250 MHz clock

I want to make it less by at least factor of 2.

Do you have any ideas?

What I tried:

1. Change the optimization to be -Ofast

2. Using pointer

3. Also, thought about FMAC and DFSDM

How can I achieve that?

Thanks

Yonatan

[LCE]

> It saved me ~250 nS

Oh my, that's disappointing... :(

[LCE]

Maybe it helps if you place at least the iteration variable i and the destination buffer raw[] into DTCM.

And / or using data cache might help.

[yonatan]

Hi @TDK 

This looks good.

I cannot use the DTCM because I use the DMA for the ADC samples.

However, I tried to put the time-consuming function in the ITCM.

However I an encountering very strange phenomena.

It works with the debugger (after flashing) but does not after reconnecting to power.

I did it with 3 steps:

1. Added this above the time-consuming function

__attribute__((section(".itcmram")))

2. Added to the linker script new section:

 _siitcmram = LOADADDR(.itcmram);  
  .itcmram :
  {
    . = ALIGN(4);
    _sitcmram = .;        /* create a global symbol at data start */
    *(.itcmram)           /* .data sections */
    *(.itcmram*)          /* .data* sections */	  
    . = ALIGN(4);
    _eitcmram = .;        /* define a global symbol at data end */
  } >ITCMRAM

3. Then added to startup.s:

CopyDataInitITCM:
  ldr r3, =_siitcmram
  ldr r3, [r3, r1]
  str r3, [r0, r1]
  adds r1, r1, #4
LoopCopyDataInitITCM:
  ldr r0, =_sitcmram
  ldr r3, =_eitcmram
  adds r2, r0, r1
  cmp r2, r3
  bcc CopyDataInitITCM
  ldr  r2, =_sbss
  b  LoopFillZerobss

What am I missing??

Again, it works after flashing the MCU with the SWD and also after disconnecting.

However, After power reset it looks like it is stuck (LED is not blinking)

I can't debug it to find where is it because when using the debugger it runs.

What you all think?

[AScha.3]

Just a question: you just summing the adc results , for average or better resolution ?

Why not use the ADC , let him make the sum : by oversampling ?

Or using the DFSDM , to get a filtered average ?

[TDK]

When it's stuck, to find out where it is, attach a debugger and debug without downloading the code or resetting the device. You can access these options within the debugger configuration.

[Pavel A.]

> I can't debug it to find where is it because when using the debugger it runs.

Debug without (re)downloading the program.

* First, download the program by any way (debugger or CubeProgramer).

* Next, configure the debugger to load symbols only, without writing to flash. 

* Power cycle the board.

* Start debugging (remember that the code is already flashed).

Check that you're stopping at main and your itcmram code has been copied properly (disassemble it, put a breakpoint there)

* Run, see where it gets stuck.

[yonatan]

You are right. my problem is that I don't want all the samples, just part of them.

regarding the DFSDM - Can I use it to sum just part of the samples?

[yonatan]

For who are wondering, I had a bug in the startup file.

This code worked for me:

// Above the c function
__attribute__((section(".itcmram")))

//In the linker script file:
  _siITCMRAM = LOADADDR(.ITCMRAM);
  .ITCMRAM :
  {
    . = ALIGN(4);
    _sITCMRAM = .;        /* create a global symbol at itcmram start */
    *(.itcmram)        /* .itcmram sections */
    *(.itcmram*)       /* .itcmram* sections */
    . = ALIGN(4);
    _eITCMRAM = .;        /* define a global symbol at itcmram end */
  } >ITCMRAM AT> FLASH

// In startup file: (after LoopFillZerobss)
  ldr r0, =_sITCMRAM
  ldr r1, =_eITCMRAM
  ldr r2, =_siITCMRAM
  movs r3, #0
  b LoopCopyITCMRAM

CopyITCMRAM:
  ldr r4, [r2, r3]
  str r4, [r0, r3]
  adds r3, r3, #4

LoopCopyITCMRAM:
  adds r4, r0, r3
  cmp r4, r1
  bcc CopyITCMRAM

View more

[LCE]

Thanks for coming back with the working code!

How's the timing with the function in ITCM RAM?

[yonatan]

it saved me something like another ~500 nS

(BTW, I also enabled the ICache so the "profit" is small)