What is preferred way to have an interrupt every second - setting timecmp to timecmp+1000 or time+1000

Question

me and friend were discussing what the most accurate way have an timer interrupt happen every second in RISC-V. We understand that the internal clock is generally inaccurate, but were still wondering what typically would be the most accurate way to do this, and we thought of reasons for both (assuming you set this immediately in your interrupt handler):

setting TIMECMP = TIMECMP + 1000

• you will incur minimal delay processing as it is simply a load+add+store

setting TIMECMP = TIME+1000

• if the interrupt is delayed but the clock stays true for some reason, this will drift less

Answer 1

When resetting a cyclic timer interrupt, you always take the timer register/capture value and use that for calculating the next interrupt. There are several reasons why:

• Interrupt latency. The time from where the MCU registers the interrupt to the point where the software enters the interrupt is partially a fixed, known delay, partially a potentially huge, unknown delay if some other ISR was already executing.
• Code execution overhead. The code in your ISR resetting the timer takes time to execute.

The above delays are far bigger than caused by any clock inaccuracy. If you calculate the time for the next interrupt based on the time when the timer set the interrupt flag, instead of the time when you entered the interrupt, you compensate for these problems.