Graduate

Question

improving progam executiion speed

Forum|Forum|2 years ago
February 7, 2024
10 replies
8910 views

To improve the execution time of a program being run on the board 'Nucleo-439ZI' from the current 20 ms period to less than 1 ms [more than 20 times reduction]. To reduce the same, I tried to switch the clock frequency source of the timer to HSE (8 MHz) instead of the internal clock (32.768KHz) [That means 24 times higher frequency to improve system latency ]. However, some issues are observed when tried by clicking 'Resolve Clock Issues' could not be resolved.

How can I improve execution time from 20ms to 1 ms?

This topic has been closed for replies.

A

Andrew Neil

Super User

@KumarTP wrote:
some issues are observed when tried by clicking 'Resolve Clock Issues' could not be resolved.

Well, nobody can help you there without knowing what, exactly, those issues were!

@KumarTP wrote:
How can I improve execution time from 20ms to 1 ms?

first you need to know what's taking all the time!

If it's in busy-wait loops, or similar, then no amount of speeding up the clock is going to help.

F

FBL

Technical Moderator

Hello @KumarTP

Would you share your IOC file in .text format? or maybe a screenshot of your clock tree would be sufficient.

K

KumarTPAuthor

Graduate

Please find the below screenshots

64 KB of 113 KB

M

mƎALLEm

Technical Moderator

Screenshots not clear.. Please share better image resolution.

T

Tesla DeLorean

Graduate II

Pretty sure it's not going to be running off the 32 KHz HSI source.

For order of magnitude improvements you might need to change your approach.

You have not articulated your problem well, or illustrated it clearly. Try harder.

M

mƎALLEm

Technical Moderator

Meanwhile, I need to check internally if this constraint on min HCLK value makes sense and it is not a bug in CubeMx

M

mƎALLEm

Technical Moderator

Hi guys,

Normally and at least the MCU could start with HSI which is 16Mhz. To me 25MHz as minimum does not make sense. So it smells a bug in CubeMx.

Moreover in the datasheet in Table 25. "Typical and maximum current consumption in Run mode" 2MHz is there.

I'm raising the behavior internally to confirm the bug and fix it. I will come back to you for any feedback.

T

Tesla DeLorean

Graduate II

MCU should clock down to DC.

Suspect the threshold instead relates to some other peripheral, perhaps clocked independently from 48 MHz, like USB, SDIO or CRYP

M

mƎALLEm

Technical Moderator

@Tesla DeLorean @waclawek.jan ,

Yes indeed .. I didn't notice that USB and ETH were enabled in the ioc file.

Thank you guys for pointing me this out :folded_hands:

@KumarTP You need to disable USB and ETH in your ioc file to reach lower frequencies, as these peripherals need at least 25MHz for HCLK to work correctly.

A

Andrew Neil

Super User

@mƎALLEm wrote:
@KumarTP You need to disable USB and ETH in your ioc file to reach lower frequencies, as these peripherals need at least 25MHz for HCLK to work correctly.

But the goal was to speed things up - so the lower limit shouldn't be a problem to @KumarTP ?

M

mƎALLEm

Technical Moderator

Yes but, in the middle of the discussion he raised an issue when he switches to HSE (@8MHz) which is related to min HCLK clock.

But frankly from my side I didn't catch yet the problem from his first post (confused with the wording "timer" and internal clock at 32.768KHz).

M

mƎALLEm

Technical Moderator

Again the pictures you are sharing have poor resolutions we can't see what is going on. If you can't please share again the ioc file showing these warnings.

Please share readable images.

Also you can click on the boxes in RED to show the warning(s).

K

KumarTPAuthor

Graduate

Please find the clk settings.

L

LCE

Graduate II

@KumarTP There are really some pros here at work to help you, but as they all pointed out, your basic problem description is still rather cryptic.

What do you exactly mean with "execution time", or of what?

So, which peripherals are you using?

Which clock speed do you need or expect, SYSCLK for the CPU, which for the peripherals?

Note: for a few peripherals / applications HSI might be too inaccurate or jittery.

T

Tesla DeLorean

Graduate II

I wish someone could clearly articulate what the problem is here, and where the 32.768 KHz LSE (32 KHz of the LSI) would be involved.

TIM5 can be used to measure LSI/LSE, but not as a source

TIM11 can be used related to HSE_RTC (1 MHz)

Still have no-idea "what" is taking 20 ms, its relationship to the MCU clock, or if 1 ms is remotely realistic.

If it takes 20 ms at 168-180 MHz the F439 normally clocks, there's going to need to be a better explanation of the task being performed, perhaps from the driving requirement rather than the current deficient solution.

L

LCE

Graduate II

Sigh...

Now you post a pdf including another low resolution picture.

Which still shows that there might be some bad clock settings.

Start reading the datasheet and reference manual about possible PLL settings, compare to the code that Cube throws out.

And what about "execution time 20 ms"?

K

KumarTPAuthor

Graduate

The idea is to reduce program compiling time. Now it is taking 20ms, trying to reduce compiling time from 20ms to 1ms.( need to speed up ) And there are no other peripherals used.