Posted on January 14, 2017 at 19:27JW is right. Normal mcu have a data register. If need to modify few bits (gpios) very fast, the core would read DR, do a bit AND mask to clear gpios that needs to be, theb OR with gpios to set high, then write back to thw DR. And oh, manually wrap a interrupt disable to avoid corruption if interrupts do modify the same GPIO port....
The BSSR and BRR register simplify the job, just with 2 Write operations, multiple gpios can be set and cleared with minimum cycles. No need Read/Modify/write opcode. With proper C style wrapping and proper compiler optimisation, it is possible to get as good as assembly programming. I stopped developping in asm 10 years back. For asm specific opcodes such as nop(), C macros exists. Alternatively to HAL, LL low layer library is one way to get there.
Another scheme would be, if suitableto your needs, to use DMA to write DR from SRAM buffer (assuming the whole port as outputs are for speedy output toggling. [Not tried yet]
