AVR Freaks

For the best performance to the Port I/O you should use the PORT_IOBUS path.

In your example you need to consider the execution time of the instructions. At 8MHz this 125nsec per clock so the toggle can occur every cycle but is not practical.

Extending your example and enabling the clock output to decode the instruction provide the following, that one cycle to toggle the port pin and two cycles for the branch.

while (true) {
    //port_pin_toggle_output_level(TEST_PIN);
    PORT_IOBUS->Group[0].OUTTGL.reg = PORT_PA10;
    PORT_IOBUS->Group[0].OUTTGL.reg = PORT_PA10;
    PORT_IOBUS->Group[0].OUTTGL.reg = PORT_PA10;
    PORT_IOBUS->Group[0].OUTTGL.reg = PORT_PA10;
}

The disassembly

    PORT_IOBUS->Group[0].OUTTGL.reg = PORT_PA10;
00000180   str	r2, [r3, #28]		 
    PORT_IOBUS->Group[0].OUTTGL.reg = PORT_PA10;
00000182   str	r2, [r3, #28]		 
    PORT_IOBUS->Group[0].OUTTGL.reg = PORT_PA10;
00000184   str	r2, [r3, #28]		 
    PORT_IOBUS->Group[0].OUTTGL.reg = PORT_PA10;
00000186   str	r2, [r3, #28]		 
00000188   b	#-12	

Blue trace is the CPU clock

Red is pin PA10

Hope that helps,

-Chris

The thing to remember is that on the Cortex, the gpio is a peripheral vs devices like the AVR where there are specific instructions and the i/o is tightly coupled. To read or write a memory address (gpio appears as memory as does everything else) you need to:

load a register with the address of the peripheral. The ARM Cortex M0 has no 32 bit immediate load, so this is done via a literal pool - the address is stored in memory close by and the load is done pc relative.

load another register with the value (to write)

do a load or store

This all takes a few cycles. Note that once you've loaded the peripheral address into a register, you can keep using this, so subsequent accesses are faster.

Just a comment for controlling up to 4 LED's on the SAMD10. You can use the TCC peripheral to perform this with hardware such that you can have 16-bit resolution using a 16MHz clock (48MHz/3) to provide a period of 244Hz. The TCC has 4 Independent shadowed compare channels to control the duty cycle for each LED.