Addressing Modes

The operand can be in the instruction itself, in a register, or in memory reached in one of two ways. Every mode is an answer to where is the operand, and the trade never changes: fewer bytes and fewer cycles, against more flexibility.

Indirect addressing keeps the address in a register, so it can be computed at run time. Increment the register and the same instruction reaches the next location — which is a loop over an array.

Every pointer, string and data structure in every language rests on this one mode. int *p is a register holding an address; *p = 5 is an indirect store; arr[i] is a base plus an index, which is why array access costs about the same as reading a plain variable.

The 8085 has five modes. The 8086 adds based, indexed, based-indexed and combinations with displacement — so MOV AX, [BX + SI + 4] computes an address from three parts in one instruction.

That is exactly the shape of a struct field inside an array. The mode was designed for a common pattern, which is the CISC philosophy in one example: add modes to shorten code. RISC goes the other way and computes addresses with explicit arithmetic.

• Immediate and register — shortest, fastest, least flexible.
• Direct — longer instruction, one extra memory cycle, address fixed at assembly time.
• Indirect and indexed — most flexible, most cycles.

In a tight loop on a small processor the mode chosen can change the running time by a factor of two, which is why hand-written assembly still beats compilers there. On a modern out-of-order CPU with caches, memory locality dominates and addressing-mode cost is nearly invisible — the currency changed, and the reasoning did not.