I have tested this using the following code and an oscilloscope:

voidsetup() {// Increase TCA0 speed to F_CPUTCA0.SINGLE.CTRLA=TCA_SINGLE_CLKSEL_DIV1_gc |TCA_SINGLE_ENABLE_bm;// Timing test - Drive a pin high for one secondpinMode(4,OUTPUT);digitalWrite(4,HIGH);delay(1000);digitalWrite(4,LOW);}voidloop() {}

This is the output:

So I am confident that this maintains correct timing. This test, without this patch, produces the following output:

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign ourContributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let usrecheck it.}

Just curious - does "CPU clock / 2" mean that the ISR would be called at 8Mhz on a 16Mhz processor? Hope not, otherwise the ISR would only have two asm instructions before the next interrupt is generated...

another idea - it seems that MegaCoreX has a different implementation, seehttps://github.com/MCUdude/MegaCoreX/blob/master/megaavr/cores/coreX-corefiles/wiring.c. Their ISR is very small, several clock freqs are supported, and other fancy stuff.

MCUdude/MegaCoreX@298601b

Does MegaCoreX have the problem described by this PR, i.e. changing the PWM freq will make millis() and friends run faster/slower? Would it be an option to adopt wiring.c from MegaCoreX?