Devin (dgoodman.delete@this.example.edu) on January 12, 2020 1:13 pm wrote:
> Ricardo B (ricardo.b.delete@this.xxxxx.xx) on January 11, 2020 7:42 pm wrote:
> > 8086 has a powerful addressing mode where the physical address is calculated
> > as: wrap_to_16(N*regA + regB + immediate) + (segmentReg <
> 8086 does not have scaled addressing modes; that came in the 80386 with the SIB byte.
>
> 8086 was limited to {BP,BX} + {SI,DI} + imm16 + (segreg << 4).
>
> > Handling this requires 16-bit hardware plus a simplified 20-bit adder for the last addition.
>
>

>
1. It's segmentReg << 4


Ah, I stand corrected.


>
2. It's a 16-bit adder (with one input limited to 12 bits), but a full 20-bit adder isn't
   > that much more expensive. Back then, integer ALUs were an appreciable part of the silicon
   > budget, but still a small part, and 25% wider wasn't going to be a deal-breaker.

>

Bigger and slower.
Not sure what adder architecture they used in 8086.

>
> The 8086 segment model also allows a range of programs to exceed 64kB of total space without
> > paying the downsides of larger addressing (eg memory space and bus cycles to store/load pointers)
> > in the form of programs which fit in 64 kB code + 64 kB stack + 64 kB data.
>
> Yes, you still want 16-bit addressing; that's why the 68000 supported it so well.
> But you could easily implement both 16- and 20-bit addressing, like the MSP430X programming
> model. It would just be a different sort of near/far pointer distinction.
>

I'm not familiar with the MSP 430X.
If it's like the M68K you have basically 2 options which run minimum overhead.
a) Total program size under 64K use 16-bit addresses.
b) otherwise use "full width" addresses (32-bit in M68K)at some memory/performance cost.

There are plenty of others possibilities but then you're spending cycles manipulating addresses including using intermediate width address (eg 24-bit in M68K).

In the x86 there's an intermediate option which runs at minimum overhead.
Up to 64 kB code plus up to 64 kB for data plus up to 64 kB for stack runs at no penalty.
Plus other intermediate possibilities (eg playing with the ES).