Trying to understand the cycle count on my STM32G431 code

[treguy]

Hi,

I have some code running on an STM32G431 that does a loop in which it reads on a pin on port GPIOB, shifts the value and writes the result on GPIOC.

I have trouble really understanding the timing I see on the oscilloscope.

The code is:

   ldr     r1, =0x48000410
   ldr     r2, =0x48000814

1:
   ldr     r3, [r1]
   lsl     r4, r3, #3
   str     r4, [r2]
   b       1b

the delay between the signal going changing on B1 and the signal changing on C4 is between 40 and 80ns depending on when the change occurs during the loop. If I made no mistake, both the SYSCLK and HCLK run at 170MHz:

    let mut d: embassy_stm32::Config = Default::default();
    d.rcc.hsi = false;
    d.rcc.hse = Some(Hse { freq: Hertz::mhz(8), mode: HseMode::Oscillator });
    d.rcc.pll = Some(
        Pll { 
            source: PllSource::HSE,     // 8MHz
            prediv: PllPreDiv::DIV2,    // 4MHz        (min: 2.065MHz, max: 16MHz)
            mul: PllMul::MUL85,         // 340MHz      (max allowed: 344MHz)
            divp: None,
            divq: None,
            divr: Some(PllRDiv::DIV2),  // 170MHz      (max allowed: 170MHz)
        }
    );
    d.rcc.sys = Sysclk::PLL1_R;
    d.rcc.boost = true; 

The HAL I use has default of DIV1 for the AHB and APB prescalers:

            ahb_pre: AHBPrescaler::DIV1,
            apb1_pre: APBPrescaler::DIV1,
            apb2_pre: APBPrescaler::DIV1,

My code seems to be running in about (40/5.88=) 7 cycles, but I don't exactly know which instructions takes more than one cycle.

I know I can add a nop after the GPIO read and that make no different on the execution time. Code runs from flash and cache is activated.

The STM32G4 series reference manual don't says there are wait states for the register, but it also says in the GPIOs main features: "Fast toggle capable of changing every two clock cycles". I could not find why it can't be done every cycle.

Is there any simple answer to this question? Is there a good documentation to get in depth knowledge of cycle counting the stm32(g4) ?

Thank you,

[SofLit]

Hello @treguy and welcome to the community,

STM32G4 family is Cortex-M4 based MCUs. There is no document for instruction cycles number for this specific family. Need to refer to the Cortex-M4 Technical Reference Manual / Section 3.3 Instruction set summary

Or to PM0214 "STM32 Cortex®-M4 MCUs and MPUs programming manual"

[TDK]

> Is there any simple answer to this question? Is there a good documentation to get in depth knowledge of cycle counting the stm32(g4) ?

On the M4 cycle counts per instruction are relatively straightforward, though there can still be some variables. In your case it seems straightforward. Here is a page on cycle counts on the M4:

Cortex-M4 Technical Reference Manual r0p0

Your code has the following:

   ldr     r3, [r1]     ; 2 cycles
   lsl     r4, r3, #3   ; 1 cycle
   str     r4, [r2]     ; 2 cycles
   b       1b           ; 1 + P cycles, (P=1 probably)

So the page suggests the code takes 7 cycles (if P=1, which is the lowest possible value), which is what you're seeing.