|
|
Instruction Set, Design Objectives
The two primary goals in defining the instruction set were to
provide enough functionality to write real programs, and to
keep the opcode format simple. Other common goals, such as fast
execution of common algorithms, minimal usage of code memory,
and support for compilers were not considered important.
The opcode format was kept simple, with the expectation that
a complex opcode format would make the project more difficult.
All instructions are encoded as a single 8-bit byte. This was
done in the hope that it would simplify the design of the
control circuitry, before I wound up creating the
microcode programming language.
CPU Architecture Decisions
After reviewing the architectures of many chips, including the
Intel 8047, 8051, 8086, Motorola 6502, 68000, and a couple
others, and the limitations of using only 8 bits for all
opcodes, many decisions were made, which ultimately lead to
the final instruction set definition.
A 16-bit address space was chosen, and of course the data path
would be 8-bits wide. A standard von Neumann
memory model was chosen, as it seemed simpler and would reduce
the number of pins required.
Addressing Modes
Direct addressing was the first sacrifice. Indirect and
immediate (constant load) addressing are both required, and
they can be used together to achieve "direct" access to
memory. Load and store, push, pop, relative program
branching also were considered absolutely necessary.
Registers
Two data registers, two address pointer registers, and one
stack pointer were chosen as the minimum "reasonable" set
of registers. Four status bits, Carry, Negative, Zero,
and Pointers Equal were chosen. These are the only
registers visible to the programmer.
Immediate Addressing
There must be instructions that can load the registers with
constants. Since it was decided at the beginning to use 8-bit
opcode for all instructions, 4 bits were alloced to be data
for these load instructions. 32 of the 256 opcode would be
needed to be able to load just on 8 bit register. Ultimately
I allocated half of all the 256 opcodes, to allow both of
the data registers and one of the 16 bit pointers to be loaded
with immediate data, without needing to move data between
other registers.
Register to Register Math Functions
Because 4 bits had been allocated for constant data, it seemed
a wise choice to define 16 math functions involving the data
registers. At least for a great number of the instructions,
4 bits would indicate the overall function, and the remaining
four would be treated as data or input to the ALU's control lines.
Program Branching
One of the most difficult decisions in the instruction set was
how to handle program branching. Without many bits to use, it
seemed impossible... at least at first. Unconditional branching
and subroutine calls would be handled by using the first 16 bit
memory pointer. Conditional branching was handled by requiring
the target of a branch to be aligned on a 32 bit boundry. This
would allow a branch, using 4 bits to specify the target of
the branch, to jump up to 68 bytes away. The downside is that
NOP instructions must be added before any instruction that will
be the target of such a branch, but the assembler can be written
to take care of this automatically. By requiring the branch
target to be on an 32-bit boundry, and not a multiple of 4 bytes
away from the branch instruction, the padded NOP instructions
would not end up inside of a loop, where a branch at the bottom
of the loop would branch back to the top. Two blocks of 16 opcodes
were allocated, to allow a wide branching range.
Moving Data Around
It seems like most assembly language code is composed mostly
of move instructions. Move instructions, push, pop, and bit move
instructions to transfer bits to and from the carry bit (the
only bit for condiction branch) were added.
Instruction Set
The original instruction set document
is still the only complete specification of the instruction set.
(someday I may add a more easy-to-read instruction set summary here)
|
Download Website
CPU Programming
Hardware Info