MSP-EXP430FR2433: MSP430FR2433 SMCLK related issues confirmed.

As you say "but when the adc_init interface was called, the pin level was not changed according to the 1S time". Do you mean the pin will not change or the change time is longer than 1s.

In the actual test situation, when the adc_init interface is called, the change time exceeds 1S, which is about 1.6s.

I think it is disturbed by the interrupt. I advise you use a timer to generate a 1s clock

We have tried to annotate the adc_init interface, the mg_sleep time is normal, and the adc_init interface is turned on, the time is abnormal.
At the same time, we tried to modify the following configuration in the adc_init interface:
change
TA1CTL = TASSEL__SMCLK | MC__UP | TACLR;
to
TA1CTL = TASSEL__ACLK | MC__UP | TACLR;
The mg_sleep 1s time returns to normal, so "TA1CTL = TASSEL__SMCLK | MC__UP | TACLR;" will affect the SMCLK clock?

As Eason says, __delay_cycles is a simple spin loop, and doesn't take into account any background (interrupt) activity. In your case, that activity includes 80000 calls to ADC_ISR (and maybe some calls to TIMER0_A1_ISR). That adds up.

You're already using both timers, but consider that ADC_ISR is called at the same rate as the TA1 overflow (80kHz). if you count 80 of those, you get 1 millisecond. E.g.

if (++tick_count >= 80) {
    tick_count=0; 
    ++ms_clock;
}

would give you a simple software clock that you can use to delay for some number of milliseconds. A delay might look something like:

void ms_delay (unsigned ms) {
    unsigned start = ms_clock;
    while ((ms_clock - start) < ms) /*EMPTY*/;
    return;
}