Why do AVR microprocessors have two ways (paths) to access I/O ports?

Question

I've an ATmega328P.

Register Summary Page 275 ATmega328P datasheet.

The first address is the I/O address, and the second is the data memory address.

I'm going to set all (D ports) Data Direction Registers To 1 and all (D ports) Port Registers to 1.

Using I/O addresses

    int main(void)
    {
    
    asm("ldi 16,0xff"); //load 0xFF to the general purpose register 16
    asm("out 0x0A,16"); //load register 16 data to PORTD
    asm("ldi 16,0xFF"); //load 0xFF to the general purpose register 16
    asm("out 0x0B,16"); //load register 16 data to PORTD

    }

Next I'm going to set all (D ports) Data Direction Registers To 1 and half of (D ports) Port Registers to 1

Using memory addresses:

    volatile unsigned char* DDRD  =  (unsigned char*)(0x2A);
    volatile unsigned char* PORTD =  (unsigned char*)(0x2B);
    int main()
    {
     DDRD = 0xff;  
     PORTD = 0X0f;
     
     
    }

The code is compiled to:

    main:
    lds     r30, 0x102 #r30 = *(0x102) = 0x2A
    lds     r31, 0x103 #r31 = *(0x103) = 0x00
    ser     r24        #r24 = 0xff
    st      Z, r24     #*(0x002A) = r24 = 0xff    DDRD Memory Address Is 0x2A
    lds     r30, 0x100 #r30 = *(0x102) = 0x2B
    lds     r31, 0x101 #r31 = *(0x102) = 0x00
    ldi     r24, 0xF   #r24 = 0x0f
    st      Z, r24     #*(0x002B) = r24 = 0xff    PORTD Memory Address Is 0x2B
    ldi     r24, 0
    ldi     r25, 0
    ret
    ; End of function main

They both work properly. Why did they build the CPU to have instructions to deal with ports I/O addresses (like in, out instructions) while they already built a set of instructions that deal with memory space (like: lds, sts, ld, st?) Why are there still instructions like in and out? Why ddin't they just get rid of them as they are useless after they mapped the I/O ports in memory

Answer 1

Take a deeper look on the instruction set, and you will find:

1. The in and out instructions use an immediate address of 6 bits width. They are limited to addresses fixed at the compile time, you cannot change the address at run time.
2. The ld and st instructions with an immediate address need a 16-bit argument. They are limited to addresses fixed at the compile time, you cannot change the address at run time.
3. The ld and st instructions with an indirect address use the value in X, Y, or Z. You are able to provide the address at run time.

The ld and st instructions need more program space and/or more cycles to execute.

This gives you freedom to use the best instruction to reach your goal:

• You have a fixed address: Use in and out, respectively.
• You need a variable address: Use ld and st, respectively.

Note 1: Because of the memory layout, the IO registers have this offset, starting right after the 32 general purpose registers.

Note 2: Depending on the optimization level of your compiler, it recognizes the "fixed address" case and emits the shorter and/or faster instructions.

---

Final note:

Questions asking "Why?" are rarely satisfyingly answerable, if you ask anybody but the inventors. Therefore, we can just speculate.

Despite the design as a RISC machine, on the one hand they apparently wanted to optimize I/O accesses, as these are so common on microcontrollers. On the other hand all I/O registers and the general purpose registers are memory mapped for presumably easier compiler creation, which was a design goal.

Answer 2

They are accessible as registers so that you can do bit read/set/clear operations on them.

They are accessible as RAM memory locations. because everything is.

Also some of the AVR instructions have different names but assemble to the same machine code. in that case the multiple names are just a convenience for the programmer.