STM32G474: Performance issues (?) and freq() issues.

[nspsck]

I recently recieved my STM32G474CE boards made by WeAct Studio and started testing Micropython on it. Soon I have noticed that the freq() function does not work as expected and the performance from this thing is way worse than the F4 series.

Here are some results in the REPL using the freq() function:

>>> pyb.freq()
(170000000, 170000000, 42500000, 85000000) # 170MHz
>>> pyb.freq(96000000) # 96MHz
>>> pyb.freq()
(8000000, 8000000, 2000000, 4000000) # 8MHz
>>> pyb.freq(170000000) # 170MHz
>>> pyb.freq()
(16400000, 16400000, 4100000, 8200000) # 16.4MHz

It seems like the the clock is divided by 10 for some reasons. Also, according to the datasheet in the section about pll characteristics, it seems the MCU needs different voltage for different clock frequencies which I do not see covered anywhere in the powerctl.c. (This section.) Am not so sure if that would cause any problems.

I have also tried to modify the pllvalues.py by adding a component to the mcu_table list declared in line 34:

    "stm32g4": MCU(
        range_sysclk=range(2, 170 + 1),
        range_m=range(1, 8 + 1), # Can go 16, but in the Manual it changes PLLM to PLLSYSM
        range_n=range(8, 127 + 1),
        range_p=range(2, 31 + 1),
        range_q=range(2, 8 + 1, 2),
        range_vco_in=range(3, 16 + 1), # 2.66 - 16
        range_vco_out=range(96, 344 + 1),
    ),

But this results to a Fatal error: can't change freq. Could this also just be my chip?

For the performance part, I have used 2 simple tests:

import time

# Int
def test(p):
    r = range(p)
    c = 0
    for i in r:
        c += 1
    return c

# Float
# def test(p):
#     r = range(p)
#     c = 0.0001
#     for i in r:
#         c += c
#         c = c / 2
#     return c * 4 / p

def main():
    ms = time.ticks_ms
    diff = time.ticks_diff
    
    while True:
        s = ms()
        print(test(50000))
        d = diff(ms(), s)
        print("Time used %dms" % d)
        
main()

Here are some results running my "benchmark":

# STM32G474CE @170MHz
# Float
MPY: sync filesystems
MPY: soft reboot
8e-09
Time used 1074ms
8e-09
Time used 1076ms
8e-09
Time used 1076ms
# Int
MPY: sync filesystems
MPY: soft reboot
50000
Time used 292ms
50000
Time used 292ms
50000
Time used 293ms
50000

# STM32F411CE @120MHz
# Float
MPY: sync filesystems
MPY: soft reboot
8e-09
Time used 795ms
8e-09
Time used 796ms
8e-09
Time used 797ms
# Int
MPY: sync filesystems
MPY: soft reboot
50000
Time used 167ms
50000
Time used 167ms
50000
Time used 167ms
50000

The results is really odd for me, since at least the Integer operation should not differ that much...

Any help and suggestions are greatly appreciated!

[ajt2018]

Could it be related to the fact that the WeAct board costs £0.40 from AliExpress yet the STM32G431CBU6 on it's own would set you back £2.6 if you bought 1000's of them from digikey ?

[robert-hh]

Did you call pllvalues.py with a value of 8 for the HSE`?

[nspsck]

By pllvalues.py I am refering to this auxiliary script. Which I can not call directly. It directly reads the .h files with HSE informations. What I have tried was:

in mpconfigboard.h:

// The board has an 8MHz HSE, the following gives 170MHz CPU speed
#define MICROPY_HW_CLK_PLLM         (2)
#define MICROPY_HW_CLK_PLLN         (85)
#define MICROPY_HW_CLK_PLLP         (2)
#define MICROPY_HW_CLK_PLLQ         (8)
#define MICROPY_HW_CLK_PLLR         (2)

in stm32g4xx_hal_conf.h:

// Oscillator values in Hz
#define HSE_VALUE               (8000000)
#define LSE_VALUE               (32768)
#define EXTERNAL_CLOCK_VALUE    (24000)

As far as I know, the EXTERNAL_CLOCK_VALUE is only used for the i2s interface and should not affect the main cpu frequency. But I could be wrong on this one, since I have not done a deep research on this.

[robert-hh]

That should be sufficient.

[ajt2018]

I'd highly recommend downloading and running ST CubeMX and the Clock Configuration tab.
You will be able to pick the STM32G4 and configure the external clock, PLL multipliers, dividers and bus speeds, then generate some code which will show you the values for PLL's etc. I find it far easier to visualise all this inside CubeMX and just pull out the settings I need for my own code.
(See image of the clock configuration in this StackOverflow question: https://stackoverflow.com/questions/49481275/generating-system-clock-configuration-for-stm32f103c8-with-cubemx)
I would have thought that the Quad SPI didn't work because of clock setup (it's always related to clock settings when peripherals don't work!)

[nspsck]

Thanks for the recommendation! Actually, that is the software I am already using. And yes, the quad spi issue is likely caused by the clock settings, but if I continue to use micropython, unless I verify all the registers using the stm module and the user manual, there is (to me) no otherway to verify the actual clock signal, because freq() is kinda broken on the G4 series right now. Unless your experience on this particular chip is different, which implies that I must have done something wrong, badly.