Chapter 13- The Instruction Set Architecture (ISA)

 

The instruction set architecture (ISA) of a computer is the structure of the computer as seen by an assembly language programmer.  In this chapter, we look at the computer hardware as seen at the assembly language level, discuss addressing modes, and briefly discuss assembly language.  We shall then present and discuss a very simple assembly language.

We specify that the computer to be studied is a stored program computer, as are all modern computers.  Such a computer executes a program that has been previously stored in the computer’s memory system, perhaps having been copied in from the disk.  Only very early computers, such as the ENIAC (1945), are not classified as stored program computers.  The program for the ENIAC was specified by a set of switches on one of its panel; in this design the memory and registers stored only data.  Some other early machines executed programs directly read from punch cards and not stored in memory.  At this point we insist that if the machine is not a stored program computer, it is ancient history and not to be studied.

The computer to be discussed in these notes is called the “Boz–7”.  It is the seventh version of a design by your favorite author, who could think of nothing better to do than name it after himself.  The Boz–7 is a synthetic computer, purposely designed along a “minimalist” style, in order to keep it as simple as possible.  It is a bit of an odd mix, partly following the principles of RISC (Reduced Instruction Set Computer) design and partly introducing some more complex design features for the sake of illustrating a few other concepts.

 

Address Space: 26-bit addressing with 32-bit data paths

The Boz–7 series is designed for 32-bit data paths, general-purpose registers and a 26-bit address space.  Such a design requires some explanation.

The memory comprises 226 (64M = 67,108,864) 32–bit words.  Were it byte-addressable, it would be sized at 256 megabytes, quite small for a modern computer, but not silly.  Memory is divided logically into 64 pages, each of 220 (1M = 1,048,576) 32–bit words.  Each program can access exactly one of these pages, the page corresponding to the page number in the Program Status Register at the time the program is running.  Obviously, the pages are numbered 0 through 63, with pages denoted by 6–bit binary numbers.  Page 0 will be reserved for the Operating System.  Only a program running with Memory Manager privileges can change the PSR, as this affects the page number.

The previous paragraph carries an implication that we now state explicitly.  Those machine instructions that generate an address do so by computing a 20–bit unsigned integer. This address is an offset into the page assigned to the process.  As seen below, the circuitry to load the Memory Address Register extends this 20–bit address into a full 26–bit address.

Later, we shall see that the physical organization of the memory does not match its logical organization.  The logical organization is a didactic trick to facilitate the introduction of a few issues related to operating system design.  The physical organization reflects the use of commercially available memory chips in the proposed memory.

 


The MAR is a 26–bit register that contains the address into physical memory.  This figure shows the mapping of 20-bit program addresses into 26–bit real addresses, using the six page number bits found in the PSR.  Bus B3 supplies the 20 low–order address bits to the MAR, these being copied from the 20 low–order bits of the IR (Instruction Register).

 

Figure: Conversion of Program Addresses

 

From the viewpoint of the program, all address registers are 20–bit registers.  This includes the Program Counter (PC) and the Stack Pointer (SP).  All address calculations, including indexed and indirect addressing are done modulo 220 and yield an unsigned 20–bit binary number that is passed to the 20 low–order bits of the MAR.  This odd arrangement does serve to force separation of process address spaces; no process can access the memory allocated to another.  This is a convenient security feature, although it is a bit rigid.

 


Figure: Program Addressing with the Page Structure

To summarize the situation, each program issues 20–bit addresses (representable as 5 hex digits) that are offsets into a memory page of size 220 (1,048,576) 32–bit words.  This is converted to a 26–bit address sent to the MAR for actual memory access.  For example, if the page number is 23 (hex 0x17) and the 20–bit address is 0x54321, the 26–bit address is 0x1754321.  As this is a 26–bit address, the high–order hexadecimal digit is less than 4, so that the address can be considered as 28–bits with the first two bits forced to 00.


Specifications for the Boz–7

Further specifications of the computer, called the Boz–7, are as follows.

      1)   It is a stored program computer; i.e., the computer memory is used to store
            both data and the machine language instructions of the program under execution.

      2)   It is a Load/Store machine; only register loads and stores access memory.

      3)   The Boz–7 is a 32–bit machine.  The basic unit of data is a 32–bit word.  This
            is in contrast to machines, such as the Pentium class, in which the basic data
            unit is the byte (8 bits) although the basic integer size is usually 32 bits.

      4)   This is a two’s-complement machine.  Negative numbers are stored in the
            two’s-complement form, so the arithmetic is said to be two’s-complement.  The
            range for integer values in from – 2,147,483,648 to 2,147,483,647 inclusive.

      5)   Real number arithmetic is not supported.  We may envision the computer as
            a typical RISC, with an attached floating point unit that we will not design.

      6)   The CPU uses a 26–bit Memory Address Register (MAR) to address memory.

      7)   The memory uses a 32–bit Memory Buffer Register (MBR) to transfer data to and
            from the Central Processing Unit.

      8)   The CPU uses a 16–bit I/O Address Register (IOA) to address I/O registers.

      9)   The CPU uses a 32–bit I/O Data Register (IOD) to put and get I/O data.

      10) The Boz–7 uses 20-bit addressing and the entire address space is occupied.  The
            memory is 32–bit word-addressable, for a total of 220 (1 048 576) words. It is not
            byte-addressable.  One advantage of this addressing scheme is that we may ignore
            the byte ordering problem known as Big Endian – Little Endian.

      11) The Boz–7 has a 5–bit op-code, allowing for a maximum of 25 = 32 different
            instructions.  By design, not all op-codes have been assigned.

      12) The Boz–7 uses isolated I/O with the dedicated instructions GET and PUT.

      13) The Boz–7 has four addressing modes: direct, indirect, indexed, and
            indexed-indirect.  In addition, two instructions allow immediate addressing.
            Indexed-indirect addressing is implemented as pre-indexed indirect.  This decision
            allows implementation of register indirect addressing, a fifth address mode.

      14) The Boz–7 has eight general purpose registers, denoted %R0 through %R7
             Each of these registers is a 32–bit register, able to hold a complete memory word.
                   %R0 is identically 0.  It is not used to store any number but the constant 0.
                   %R1 through %R7 is read/write registers, used to store results of computation.

Each of the eight registers can be used as an index register or as the source operand for an instruction.  Only registers %R1 – %R7 can be changed by arithmetic or register load operations.  Attempts to change %R0 are undertaken for side effects only.

 

NOTE: The reason for selection of eight registers and not more is that the 3–bit register select field fit neatly into the preferred instruction format, while a 4–bit field did not.


Program Status Register (PSR)

Here is the structure of the 32–bit processor status register (PSR), also called the program status register.  Note that not all bits are assigned.

 

31

30

29

28

27

26

25

24

23

22

21

20

19

18

17

16

Not presently assigned.

6–bit Page Number

 

15

14

13

12

11

10

9

8

7

6

5

4

3

2

1

0

Security Flags

V

C

Z

N

R = 00

I

CPU Priority

 

Bits 2 – 0 of the PSR specify a three-bit unsigned integer corresponding to the CPU priority.  This unsigned integer corresponds to a CPU priority in the range 0 to 7 inclusive.

I/O device priority is one of 4, 5, 6, or 7.  Levels 1, 2, and 3 are used for software interrupts, which are the preferred mechanism by which a user program will invoke services of the operating system. User programs almost always execute at priority 0.

Bit 3 is the interrupt bit, used to allow or disallow the raising of interrupts by input/output devices.  If this bit is zero, then interrupts are blocked.  Such a setting may be required by the operating system in the initial processing of an interrupt to block other interrupts.

Bits 5 – 4 of the PSR are reserved, with each bit hardwired to logic 0 in the current design.  It is common practice for computer designs to have reserved bits, as opposed to bits that are just not used.  In this design, reserving the bits allows for more priority levels in the future.

Bits 9 – 6 of the PSR reflect the effect of the last arithmetic operation.
            C         Carry bit                      the last operation generated a carry out.
                                                            Not a problem; useful for multi-precision arithmetic.
            N         Negative bit                the result of the last operation was negative.
            Z          Zero bit                       the result of the last operation was zero.
            V         Overflow Bit               the last operation caused a numeric overflow.

Question:        How to set the N, Z, and C bits based on the last ALU operation.
Answer:          The control unit will do this as a part of executing the arithmetic.
                        These bits cannot be set by loading the PSR.

Bits 15 – 10 of the PSR are used as security flags, allowing the operating system to assign privileges to other programs.  Specific privileges might include: access to I/O devices, memory management and process scheduling, and access to all files in the file system.

The current design is more of a reaction to the UNIX user/super–user model in which a program has either no privileges or has every privilege.  At present, we shall not be more specific on assignment of these bits to privilege levels.  When the operating system runs in the UNIX “super–user” mode, it has privilege 6310 = 1111112.

Bits 21 – 16 of the PSR determine which of the 64 memory pages is allocated to the process.  The memory is divided into pages of 220 words and the program can use only one of them.

Bits 31 – 22 of the PSR are presently not assigned any function and may serve any number of uses in the future.  Because they are not reserved, system software may use them.


General comments on 32–bit words

We shall use eight hexadecimal digits to represent the 32–bit binary values stored in the
Boz–7 memory words.  This notation is used for character data, integer data, and instructions.  We use bit numbering in which bit 31 is on the left and bit 0 is on the right, so that the bits as read from left to right are from the most significant to least significant.

 

Character Data Format

The Boz–7 will be viewed as storing character data in the 8–bit ASCII format or 16–bit UNICODE (if we are to be more modern).  Standard 8–bit ASCII data would be stored four characters to the memory word and manipulated four characters at a time.  Characters would be numbered in the word according to the following convention.

Bits

31 to 24

23 to 16

15 to 8

7 to 0

Character

3

2

1

0

This course will focus on integer data.  It is not that character data are unimportant, it is just that we need simple examples so that we can focus on the hardware and not on the data.

 

Integer Data Format

The Boz–7 stores signed integers as 32–bit two’s-complement numbers.  The range of integers that can be stored and processed directly by the CPU is – 2,147,483,648 ( – 231 )  to 2,147, 483, 647 ( 231 – 1), inclusive.  Other precision arithmetic ( 8–bit, 16–bit, and 64–bit) are not supported by this design, though they would be useful in a real computer.

 

Real Number Format

The Boz–7 is not designed to process real numbers, also called floating point numbers.  If it did, it would use IEEE–754 single-precision format and use an attached coprocessor to do the calculations.  In this regard, it would be typical of RISC–type processors in allocating floating point execution to an attached processor.  We shall ignore floating-point numbers.

 

 

The Assembly Language of the Boz–7

The assembly language of a computer represents the lowest level instructions that the computer can execute directly.  Some of us have to program computers in assembly language and most of us (thankfully) do not have that task.  The main issue in favoring a higher level language over assembly language is programmer productivity.  If a programmer can write only so many lines of code per day (there are good measures of this), then it is better that he or she write lines of code that translate into many assembly language instructions that if each line of code translates only into one such instruction.

 

In computer architecture, we view assembly language statements as the “functional specifications” of the computer, in that each such statement indicates a specific action that the computer must complete.  The assembly language of this computer has been designed to present a typical collection of functions typically found on a modern machine.  Once we have stated what the computer must do, we design the computer to do exactly that.


The instructions in the assembly language of the Boz–7 are listed below, in numeric order of the op-codes.  Note that not all 32 op–codes are used in this version of the design.  The reader will note an unexplained gap in the operation code sequence.  This gaps will facilitate the design of the CPU control unit by considerably simplifying its circuitry.

Op-Code

Mnemonic

Description

00000

HLT

Halt the Computer

00001

LDI

Load Register from Immediate Operand

00010

ANDI

Logical AND Register with Immediate Operand

00011

ADDI

Add Signed Immediate Operand to Register

00100

NOP

Not Yet Defined – At Present it is does nothing

00101

NOP

Not Yet Defined – At Present it is does nothing

00110

NOP

Not Yet Defined – At Present it is does nothing

00111

NOP

Not Yet Defined – At Present it is does nothing

01000

GET

Input to Register

01001

PUT

Output from Register

01010

RET

Return from Subroutine

01011

RTI

Return from Interrupt (Not Implemented)

01100

LDR

Load Register from Memory

01101

STR

Store Register into Memory

01110

JSR

Subroutine Call

01111

BR

Branch on Condition Code to Address

10000

LLS

Logical Left Shift

10001

LCS

Circular Left Shift

10010

RLS

Logical Right Shift

10011

RAS

Arithmetic Right Shift

10100

NOT

Logical NOT (One’s Complement)

10101

ADD

Addition

10110

SUB

Subtraction

10111

AND

Logical AND

11000

OR

Logical OR

11001

XOR

Logical Exclusive OR

 

Privileged Instructions
In a multi–user computer, some instructions must be reserved for use by the Operating System and its system programs.  These include access to I/O devices (our instructions are GET and PUT) to preclude the simultaneous use of such devices by more than one process.  Other privileged instructions would be those to manipulate the Program Status Register and directly access the Stack Pointer.  The stack pointer is changed by both PUSH and POP instructions, which are not privileged; but it required O/S privilege to initialize its value.

A modern assembler would convert a number of instructions to operating system calls, often called “traps”, “software traps”, or “software interrupts”.  These include:
            HLT    translated into a return to the Operating System (which terminates the
                        process, reallocates memory, and starts another process),
            GET    translated into a call to an Operating System routine to get input, and
            PUT     translated into a call to an Operating System routine to output data.


Addressing Modes

The Boz–7 computer may be said to support five addressing modes: immediate addressing and four true addressing modes, which are direct, indirect, indexed, and indexed-indirect.  As this is a Load/Store machine, these modes are limited to certain instructions, specifically the following four instructions that will be used to illustrate the addressing modes

             LDI            Load Register Immediate
             ADDI        Add Register Immediate

             LDR           Load Register from Memory
             STR            Store Register to Memory

 

Of these instructions, only the first two can use immediate addressing.  Only the second two instructions can use the other four addressing modes to address memory.  As an aside, we shall see that the I/O instructions (discussed below) can be considered to use direct addressing in that the argument specifies the address of the I/O register.  However, we note that these instructions do not address memory and so give them minimal coverage here.

 

One of the main differences between a RISC device, such as our computer, and a CISC device such as the VAX–11/780 (now obsolete) is that the latter can issue arithmetic commands that involve the memory directly; such as ADD X, Y to add directly the contents of the two memory locations and place the result into one of them.  In our computer, only a general–purpose register can be the target of an ADD instruction, and the operands must be either both registers or one register and an immediate operand.  This is the major design constraint of a load/store architecture.  It has been discovered that the increase in CPU performance more than pays for the inconvenience of this design constraint.

 

To differentiate the immediate address mode from other address modes, let’s consider a simple instruction set with two modes of addressing (direct and immediate) and a single accumulator, which is loaded by the instruction called LOAD.  What does the instruction LOAD 100 do?  In immediate mode, the register is loaded with the value 100.  Thus we see that the immediate mode should not be called an address mode, as no memory address is used; the argument is coded immediately in the instruction.  In direct mode, the register is loaded with the value of the memory word at address 100.

Immediate Addressing

Most computer architectures call for immediate instructions to have the argument encoded directly within the 32–bit machine word representing the instruction.  In these designs, immediate instructions do not reference computer memory to access arguments and thus differ from other addressing modes in which the machine instruction encodes an address for an argument in main memory.  One notable difference is found in the ISA (Instruction Set Architecture) for the IBM mainframe series (S/360, S.370, z/9, z/10, etc.) in which an immediate instruction has two operands, one of which is a memory reference and one of which is encoded within the instruction.  We shall not use that type of instruction here.

In the Boz–7, the lower order 20 bits of the machine instruction (bits IR19 – IR0) are used by many instructions to store either the argument (immediate addressing) or an address used to locate the argument in memory (other addressing modes).


I/O Device Register Addressing

The two I/O instructions, GET and PUT, reference I/O devices by a 16–bit address on the I/O bus, which is separate from the memory bus.  The MAR is not involved in this addressing, as it uses a register named IOA that accesses the I/O device registers.  As indicated above, this type of addressing can be viewed as direct addressing, except that the term is reserved for discussions of memory addressing.  The use of device registers to access I/O devices is explained in the chapter on Input/Output.

 

Memory Addressing

It should be understood that only four instructions actually compute memory addresses.  These four instructions can use any of the available addressing modes.
            LDR and STR             the address of the argument.
            BR and JSR                the address of the jump target.

 

The Boz–7 computer may be said to support five addressing modes, which are direct, indirect, register-indirect, indexed, and indexed–indirect.  These addressing modes are built around two primitive operations.

            Indirection:     This is similar to the use of pointers in some modern languages.
            Indexing:         This is similar to the use of indices in accessing arrays.

 

We note that there are two possible varieties of indexed–indirect addressing, depending on whether the indexing is done first or the indirection is done first.  Generally speaking, we may view these two in terms for describing higher–level languages as follows.
             Pre–indexed indirect              an array of pointers.
             Post–indexed indirect                        a pointer to an array.

Many computers will support both modes, but the Boz–7 computer supports only
pre-indexed indirect.  This is due to the design requirement that the CPU be simple.

 

The four true addressing modes are constructed from these two primitives, according to the following table.

 

 

Indexing Not Used

Indexing Used

Indirection Not Used

Direct Addressing

Indexed Addressing

Indirection Used

Indirect Addressing

Indexed-Indirect Addressing

 

All address calculations are performed modulo 220 (modulo 1048576), so that no addresses outside the permissible range of memory addresses are generated.  All addresses are interpreted as unsigned 20–bit integers.

 

When bits 19 through 0 of the IR form an immediate address, they are interpreted as an
20–bit integer; either unsigned (for the ANDI) or two’s–complement for the LDI and ADDI.  Specifically the ranges are:
            ANDI              0 to 220 – 1, representable as five hexadecimal digits, but best
                                    viewed as a collection of twenty Boolean bits with no numeric value.
            LDI, ADDI     – 219 to 219 – 1, or – 524, 288 to 524, 287.


Effective Address

Each of the four true addressing modes references a word in memory.  We say that each of these modes gives rise to an effective address, that is the address of the operand being indicated by the specified mode.  We use the following memory lay-out as an aid in our discussion of the addressing modes and the corresponding effective addresses.  For each of the examples, we consider the LDR instruction that loads the accumulator from the location indicated by the effective address.

 

In what is below, we let the symbol Z stand for 0x0A, or 10 in decimal.  We assume that memory in the vicinity of this address is laid out in the following map, and that the contents of register 3 are 7; denoted by (%R3) == 7.

 

Address

5

6

7

8

9

A

B

C

D

E

F

10

11

Contents

11

32

2C

1E

56

5

7A

10

3

F

E

D

8

Figure: Sample Address Map (All values are hexadecimal)

 

In immediate addressing, there is no access to memory and no effective address.  The effect of the instruction LDI %R1, Z is the same as LDI %R1, 10; register %R1 gets the value 10.

 

We use the idea of an effective address as a part of determining the effect of the instruction.  For register load instructions, the effect of the instruction is that the register has a given value stored into it.  For the true addressing modes, this value depends on the effective address.  If we use the term “EA” to represent effective address, what happens in the four true addressing modes is that the target register gets the contents of Memory[EA].

LDR:   Register ¬ M[EA]

A word of caution is now in order.  We shall discuss assembly language statements such as LDR %R1, Z, in which Z is considered as an address.  This is in contrast to high-level language statements such as X = Z, in which Z is the value stored at some address.  For a high-level language, the compiler associates a memory location with each variable, so that the variable Z is associated with an address, z, and we retrieve Memory[z].

 

Direct Addressing

LDR %R1,      Z

Recall that Z is an address, not the value stored at that address.

 

For direct addressing the effective address is EA = Z

 

The effect is %R1 ¬ M[ Z ]

Here %R1 gets M[0x0A] = 5, after the register load %R1 == 5.


Indirect Addressing

LDR %R1,      * Z

 

The effective address is EA = M[Z] = M[0x0A] = 5

The effect is %R1 ¬ M[ M[Z] ] = M[ M0x0A] ] = M[5] = 11

Here %R1 gets M[5], after the register load %R1 == 11

 

Indexed Addressing

LDR %R1,      Z, 3

The effective address is EA = Z + (%R3) = A + 7 = 11

The effect is %R1 ¬ M[ Z + (%R3)] = M[0xA + 7] = M[0x11] = 8

Here %R1 gets M[0x11]; after the register load %R1 == 8

 

Register-Indirect Addressing

LDR %R1, *(3)

This occurs when the address Z = 0.  In this case the register holds the effective address.

The effective address is EA = (%R3) = 7.
The effect is %R1 ¬ M[ (%R3) ] = M [7] = 2C.

Here %R1 gets M[0x11]; after the register load %R1 == 2C

 

Indexed-Indirect Addressing

LDR %R1,      * Z, 3

There are two types of indexed-indirect addressing, depending on the order of operations.

            In preindexed-indirect addressing, the indexing is done first, then the indirection.
            In postindexed-indirect addressing, the indirection is done first, then the indexing.

 

Preindexed-Indirect

The effective address is EA = M[Z + (%R3)] = M[0x0A + 0x07] = M[0x11] = 8

The effect is %R1 ¬ M[ M[ Z + (%R3)]] = M[0x08] = 0x1E.

Think of this as an array of pointers.  The address Z + (%R3) refers to an array entry that is used as a pointer to the addressed entry.

 

Postindexed-Indirect

The effective address is EA = M[Z] + (%R3) = M[0x0A] + 7 = 11 = 5 + 7 = C

The effect is %R1 ¬ M[ M[ Z] + (%R3)] = M[0x0C] = 10.

Think of this as a pointer to an array.  The address Z holds a pointer to an array at address M[Z].  This array is indexed by %R3.

 

It is important for the student to understand the difference between the two varieties of indexed-indirect addressing.  Many computers will implement both types of addressing, but the Boz–7 implements only one, in order to simplify the design.

 

The Boz–7 design implements more addressing modes than a typical RISC design.  This is due to its use as a teaching tool and the need to discuss these addressing modes.


Syntax of the Assembly Language

We now characterize the syntax of each type of assembly language statement as used in the Boz–7 computer.  One reason to do this is to keep the instructor from getting confused.

 

Immediate Operations

The syntax of the immediate operations is quite simple.

      Syntax                          Example

      LDI %Rn, value           LDI %R3, 100
      ANDI %Rn, value        ANDI %R5, 0xFFA08   -- Bit masking is easier in hexadecimal.
      ADDI %Rn, value        ADDI %R5, 200
      NOP                             NOP                                -- Not much to say on this one.

 

It is expected that the following operations will be common enough to warrant syntactic sugar in the assembly language.

            LDI %Rn,    0             Clear the register
            ADI %Rn,    1             Increment the register
            ADI %Rn, – 1             Decrement the register

 

Input / Output Operations
      Syntax                               Example

      GET %Rn, I/O_Register   GET %R2, XX           Load the general-purpose register from
                                                                                    the I/O device register.
      PUT %Rn, I/O_Register   PUT %R0, YY            Store the contents of register into the
                                                                                    I/O device register. 

 

Load/Store Operations

The syntax of these is fairly simple.  For direct addressing we have the following.

            Syntax                                     Example
            LDR %Rn, address                 LDR %R3, X              Loads %R3 from address X
            STR %Rn, address                  STR %R0, Z               Loads %R0 into address Z
                                                                                                This clears address Z.

Other variants of the syntax are illustrated in the discussion on addressing modes.

 

Branch

The syntax of this instruction, and its variants, is quite simple.

            Syntax                         Example

            BR address                 BRU W           Note that this can use all addressing modes.

Other variants of the syntax are illustrated in the discussion on addressing modes.  The use of condition codes for conditional and unconditional branching is explained in the section on syntactic sugar.  In this example BRU W is assembled as BR W with condition code = 000.

 


Subroutine Call and Return

            Syntax                         Example

            JSR address                JSR W             Note that this can use all addressing modes.

            RET                             RET                 The instruction does not take an argument.
            RTI                              RTI                  Not yet implemented, this takes no argument.

 

Unary Register          (These instructions use one source register, hence the name “unary”.)

            Syntax                                                 Example

            Op, Destination, Source, [Count]       LLS %R5, %R6, 3
                                                                        NOT %R2, %R1

 

The NOT operation does not take a count.  The shift operations take a count, with the shift count defaulting to one if it is not provided.  Shift by 0 is the same as a copy.

 

The restriction on this class of instruction is that the count, if used, must be a constant number known at the time that the assembler is run.  As the registers are 32–bit, the counts are evaluated modulo 32.  Constants and defined numbers are allowable in the instruction, but variables are not provided for in the syntax.  Shifting by a count stored in a variable would be implemented using a looping structure.

 

A side effect of this constraint is that the right circular shift by a fixed amount may be translated by the assembler into an equivalent left circular shift.

 

 

Binary Register         (These instructions use two source registers, hence the name “binary”)

            Syntax                                                             Example

            Op, Destination, Source_1, Source_2             ADD %R3, %R2, %R1

 

Note that the subtract operation is the only one for which the order of the source registers is important.  SUB %R3, %R2, %R1 causes %R3 ¬ (%R2) – (%R1).

 

%R0 as a Source Register

 

Register %R0 is often used as a “source of 0”, a way to place the constant value 0 in another register or into a memory location.

 

 

%R0 as a Destination Register

 

Any operation that used register zero (%R0) as a destination in effect just discards the results.  One use would be to force a arithmetic operation with the only goal of setting the sign bits.

As an example, we might say SUB %R0, %R1, %R2 to get the sign of (%R1) – (%R2) without storing the results of the subtraction.


Syntactic Sugar

Syntactic sugar, as applied to assembly language, refers to instructions that appear in the assembly language that are assembled as other instructions.  This translation is performed by the assembler, which is the software program that emits the machine code.  Our assembly language has a number of instructions that fall under this category.  Here are some examples.

 

Syntactic Sugar         Assembled As                        Comments

CLR %R2          LDI %R2, 0         Clears the register
CLW X            STR %R0, X         Clears the word at address X
INC %R2          ADI %R2,  1        Increments the register
DEC %R2          ADI %R2, -1        Decrements the register

NOP              LLS %R0, %R0, 0    No-Operation: Does Nothing

 

RCS %R3, %R1, 3  LCS %R3, %R1, 13   Right circular shift by N is
                                    the same as left circular by
                                    32 – N.  The shift count must
                                    be a constant number.

 

DBL %R2          LLS %R2, %R2, 1    Left shift by one is the
                                    same as a multiply by 2.

 

MOV %R2, %R3     LSH %R2, %R3, 0    Shift by 0 is a copy.

 

NEG %R4, %R5     SUB %R4, %R0, %R5  Subtract from %R0 º 0 is
                                    the same as negation.

 

TST %R1          SUB %R0, %R1, %R0  Compares %R1 to zero by

                                    subtracting %R0 from it and
                                    discarding the result.

 

CMP %R1, %R2     SUB %R0, %R1, %R2  Determines the sign of
                                    (%R1) – (%R2), discarding
                                    the result.

 

BRU              BR 000             Branch always

BLT              BR 001             Branch if negative
BEQ              BR 010             Branch if zero

BLE              BR 011             Branch if not positive
BCO              BR 100             Branch if carry out is 0
BGE              BR 101             Branch if not negative
BNE              BR 110             Branch if not zero
BGT              BR 111             Branch if positive

 

BNS              BR 001             Same as BLT.
                                    Used by I/O operations, in
                                    which a negative status
                                    indicates an error.


More Comments on the Assembly Language

We now discuss the specifics of the syntax of the assembly language and show how the fields in the IR (Instruction Register) are used to specify the precise operation.  The IR is a 32–bit register used to hold the instruction being executed.

This is a Load/Store RISC machine.  Only Load Register and Store Register instructions access the memory.  The only instructions that deal with memory addresses are the register load and store instructions, the branch (jump) instructions, and the subroutine call. 

Immediate Addressing

31

30

29

28

27

26

25

24

23

22

21

20

19 – 0

Op–Code

 

Destination Register

Source
Register

Immediate Argument

Op–Code        00000  NOP    Halt
                        00001  LDI     Load Immediate         (Does not use Source Register)
                        00010  ANDI  Immediate logical AND
                        00011  ADDI  Add Immediate

In these instructions, the source register most commonly will be the same as the destination register.  While there is some benefit to having a distinct source register, the true motivation for this design is that it simplifies the logic of the control unit.

The most common immediate instructions will probably be the following.
                        LDI     %RD, 0                       -- Load the register with a 0.
                        LDI     %RD, 1                       -- Load the register with a 1
                        ADDI  %RD, %RD,    1         -- Increment the register
                        ADDI  %RD, %RD, – 1         -- Decrement the register

Input/Output Instructions
This design calls for isolated I/O, so it has dedicated input and output instructions.

Input

Op-Code         01000  GET    Get a 32–bit word into a destination register from an input.

 

31

30

29

28

27

26

25

24

23

22

21

20

19

18

17

16

15 – 0

0

1

1

0

0

 

Destination
Register

Not Used

Not Used

I/O

Address

 

Output

Op-Code         01001  PUT     Put a 32–bit word from a source register to an output register.

 

31

30

29

28

27

26

25

24

23

22

21

20

19

18

17

16

15 – 0

0

1

0

0

1

 

Not Used

Source
Register

Not Used

I/O

Address

 

Note that these two instructions use different fields to denote the register affected.  This choice will simplify the control circuit.  All unused bits are assumed to be 0.


Memory Addressing

The next four instructions (LDR, STR, BR, and JSR) can use memory addressing.  The first two use the memory address for a data copy between a specific register and memory.  The next two use the memory address as the target location for a jump.

 

The generic structure of these instructions is as follows.

31

30

29

28

27

26

25

24

23

22

21

20

19 – 0

Op–Code

I bit

Register/Flags

Index

Address

The contents of bits 25 – 23 depends on the instruction.

 

The Real Reason for %R0 º 0

We now discuss an addressing trick that is one of the real reasons that we have included a general–purpose register that is identically 0.  What we are doing is simplifying the control unit by not having to process non-indexed addressing; that is, direct or indirect.  Note that bits 22 – 20 of the IR specify the index register to be used in address calculations.

 

When the I-bit (bit 26) is zero, we will call for indexed addressing, using the specified register.  Thus the effective address is given by EA = Address + (%Rn), where %Rn is the register specified in bits 22 – 20 of the IR.  But note the following
            If Bits 22 – 20 = 0, we have %R0 and EA = Address + 0, thus a direct address.

 

When the I-bit is 1, we have the same convention.  Indexed by %R0, we have indirect addressing, and indexed by another register, we have indexed-indirect addressing.

The “bottom line” on these addresses is shown in the table below.

 

 

IR22-20 = 000

IR22-20 ¹ 000

IR26 = 0

Indexed by %R0 (Direct)

Indexed

IR26 = 1

Indirect, indexed by %R0 (Indirect)

Indexed-Indirect

 

 

Load Register

31

30

29

28

27

26

25

24

23

22

21

20

19 – 0

0

1

1

0

0

I bit

Destination Register

Index Register

Address

 

Here the I bit can be considered part of the opcode, if desired.
            011000            Load the register using direct or indexed addressing
            011001            Load the register using indirect or indexed-indirect addressing

 

For a load register operation, bits 25 – 23 specify the destination register.  If the destination register is %R0, no register will change value.  While this seems to be a “no operation”, it does set the condition codes in the PSR and might be used solely for that effect.

 


Store Register

31

30

29

28

27

26

25

24

23

22

21

20

19 – 0

0

1

1

0

1

I bit

Source Register

Index Register

Address

For a store register operation, bits 25 – 23 specify the source register.  If the source register is %R0, the memory at the effective address will be cleared.

 

Subroutine Call

31

30

29

28

27

26

25

24

23

22

21

20

19 – 0

0

1

1

1

0

I bit

Not
Used

Index
Register

Address

 

Branch

31

30

29

28

27

26

25

24

23

22

21

20

19 – 0

0

1

1

1

1

I bit

Branch
Condition

Index
Register

Address

The condition code field determines under which conditions the Branch instruction is executed.  The eight possible options are.

                         Condition                Action
                              000                     Branch Always                    (Unconditional Jump)
                              001                     Branch on negative result
                              010                     Branch on zero result
                              011                     Branch if result not positive
                              100                     Branch if carry-out is 0
                              101                     Branch if result not negative
                              110                     Branch if result is not zero
                              111                     Branch on positive result

 

Return from Subroutine / Return from Interrupt.

31

30

29

28

27

26 – 0

Op–Code

Not Used

 

Op–Code =     01010  RET     Return from Subroutine
                        01011  RTI      Return from Interrupt (Not presently implemented)

 

Neither of these instructions takes an argument or uses an address, as the appropriate information is assumed to have been placed on the stack.


Register-to-Register Instructions

Unary Register-To-Register

31

30

29

28

27

26

25

24

23

22

21

20

19

18

17

16

15

14 – 0

Op Code

 

Destination Register

Source Register

Shift Count

Not Used

 

Opcode =        10000        LLS                 Logical Left Shift
                        10001        LCS                 Circular Left Shift
                        10010        RLS                 Logical Right Shift
                        10011        RAS                Arithmetic Right Shift
                        10100        NOT                Logical NOT (Shift count ignored)

NOTES:          1.               If (Count Field) = 0, a shift operation becomes a register move.
                        2.               If (Source Register = 0), the operation becomes a clear.
                        3.               Circular right shifts are not supported, because they may be
                                          implemented using circular left shifts.
                        4.               The shift count, being a 5 bit number, has values 0 to 31 inclusive.

 

The last topic for discussion is the binary register-to-register operations.  By “binary” we do not refer to binary arithmetic, but to arithmetic operators, such as addition; those take two arguments and produce one result.

 

Binary Register-To-Register

31

30

29

28

27

26

25

24

23

22

21

20

19

18

17

16 – 0

Op Code

 

Destination Register

Source Register 1

Source Register 2

Not used

 

Opcode =        10101        ADD               Addition
                        10110        SUB                Subtraction
                        10111        AND               Logical AND

                        11000        OR                  Logical OR
                        11001        XOR               Logical Exclusive OR

NOTES: Subtract with (Destination Register) = 0 becomes a compare to set condition codes.

The Other Op-Codes

The other op–codes are not implemented in this design.  The reason is simply the fact that the above selection of op–codes suffices to make the points required for this text.

Wasted Space

The reader might notice the label “Not Used” for many of the fields in the above instructions.  This is a clear indication that a CISC design, with varying instruction lengths would be more efficient in the use of memory.  The reason chosen by this author for the uniform instruction length is similar to that chosen by most designers of RISC: simplicity.  For the RISC designers, simplicity implies a more efficient design.  For this author, simplicity means a design that is easier to describe.  The Boz–7 is inherently inefficient.


A Word Of Caution to System Programmers

At this point in the chapter, we pause and present a cautionary tale in the form of a possible strategy for programming the Boz–7 computer.  This example might seem ridiculous, except for the fact that something like it happened to Apple Computers, when the company expanded the address space of its Macintosh line of computers to a full 32 bits.

 

Suppose it is the goal of a systems programmer to provide a device handler, written as a subroutine, and thus invoked by a JSR instruction.  The clever programmer will recall that only the 20 low order bits (represented by five hexadecimal digits) are used in forming addresses.  More specifically, the upper twelve bits (three hex digits) are ignored.

 

In indirect addressing as used with subroutine calls, the argument contains the address of subroutine to be called.  Specifically, the following instruction will cause the subroutine at address 0x12345 to be called.

            JSR *Z                        with M[Z] = 0x12345.

To be more specific in the example, note that M[Z] = 0x00012345, as it is properly represented by eight hexadecimal digits.  It is just that we suppress leading zeroes.

Now consider the instruction

            JSR *Z                        with M[Z] = 0x13612345.

Because only the 20 low order bits are used in forming the address of the subroutine, this is equivalent to the previous example, with the subroutine at address 0x12345 being called.  A clever programmer can make use of this, with code such as the following.

            LDR    %R1, Z                        // Get the address of the pointer into register 1.
            RLS     %R1, %R1, 20            // Shift out the address bits
            ANDI  %R2, %R1, 0x7          // R2 now has three low order bits of R1
            RLS     %R1, %R1, 3              // Shift these bits out
            ANDI  %R1, %R1, 0x3F        // Keep the six low order bits.

            JSR      *Z                                // Complete the subroutine jump.

Reading the code fragment above, one might see that R2 has a priority and R1 has a six–bit number that will serve as a set of security flags.  Let’s trace the execution of this clever code.

            LDR    %R1, Z                        // %R1 = = 0x13612345
            RLS     %R1, %R1, 20            // %R1 = = 0x00000136
            ANDI  %R2, %R1, 0x7          // %R2 = = 0x00000006
            RLS     %R1, %R1, 3              // %R1 = = 0x00000026
            ANDI  %R1, %R1, 0x3F        // %R1 = = 0x00000026 = = 001001102.
            JSR      *Z                                // Complete the subroutine jump.

So we have been clever and coded the 32–bit word describing the subroutine with three entries: its 20–bit address, the priority with which it runs, and the security flags with which it runs.  Now, what happens when we upgrade the system to run with true 32–bit addresses?  The answer is that the operating system, being full of these tricks, must be rewritten completely.  This is not an easy task.  As mentioned above, such has actually happened.


Organization of the Memory Unit

The design of the Boz–7 uses a 32–bit data word and 32–bit instruction word.  For memory reference instructions, the address is formed from a 20–bit address in the CPU converted to a 26–bit address by use of the 6–bit page register for the PSR associated with the program.  The process of generating a logical memory address is shown in a figure, copied from earlier in this chapter.

Figure: Memory Addressing in the Boz–7

Because the design uses a 26–bit address field to access memory, it was decided to have a memory requiring only 26 address bits; thus a memory with 64 Meg (226) of entries.  Each data and instruction register of this computer contains 32 bits, so it was decided that each addressable memory unit would also contain 32 bits.  Future revisions of this computer series might provide for byte addressing, but that has been postponed until this author can consider all of the consequence of providing such addressing.  For now we keep it simple.

 

The Boz–7 Memory as It Is

The memory contains 226 ( 64M = 67,108,864 ) addressable 32–bit words.  The memory is organized into sixty four logical banks, each holding 1M ( 1,048,576 ) 32–bit words.

 

The physical organization of the memory is based on memory chip technology that was current as of April, 2004.  The memory uses eight Micron MT47H16M16 memory chips, each a chip being organized as a 16 Meg x 16 chip.  These chips are organized into four memory banks, each bank containing a pair of 16 Meg x 16 chips and considered as a
16 Meg x 32 memory unit.  Since this memory holds 226 addressable units, it must be accessed by twenty–six address lines; these are called A25 .. A0.

 

Low order interleaving is used.  The 24 high–order bits of the address ( A25 .. A2 ) are sent to each of the memory banks and the 2 low–order bits of the address ( A1, A0 ) select the bank.  To speed memory access, all memory accesses are made to four banks at once.

 


The data bus on this design is the only feature that stretches the current state-of-the-art in computer design.  It is a 128–bit (four 32–bit word) bus, attached to the CPU via a 32KB SRAM cache with 128–bit cache lines.  The organization is as follows.

 

 

Recalling that 32KB = 215 bytes = 218 bits, we see that the cache memory is divided into 1024 (210) cache lines, each of 128 (27) bits or eight 32–bit words, the same size as the data bus between cache memory and main memory.  The goal is to transfer data one cache line at a time, facilitating fast access to main memory.  We illustrate the cache in terms of reading from memory, such as might be done for fetching an instruction.

      1)   A 26–bit address is sent to the cache memory and READ is asserted.

      2)   If the addressed unit is in the cache memory, the 32–bit word is transmitted
            to the MBR in the CPU within about 2-to-3 nanoseconds.

      3)   If the addressed unit is not in the cache memory, the 26–bit address is treated as
                        a)         A 3–bit offset within a cache line
                        b)         A 23–bit address associated with identifying the cache line

      4)   The 26–bit address is transmitted to the four–way interleaved memory, which
            sends the 24 high order bits of the address to all four memory banks.  This results
            in the main memory producing four 32–bit words.  These four words are
            transmitted simultaneously along the 128–bit data bus, filling the cache line.

      5)   The addressed word is then copied into the MBR of the CPU.  With any luck, the
            next few memory references will match entries in the cache.

 

Memory Interface

At this moment, the memory unit will have the following interface to the CPU
      MAR         the 20-bit Memory Address Register
      MBR         the 32–bit Memory Buffer Register
      READ       when asserted high, the CPU reads from memory
      WRITE     when asserted high, the CPU writes to memory

 

As an arbitrary tie breaker, we specify that READ takes precedence over WRITE, although the control unit should never assert the two signals at the same time.  Thus, we have

 

READ

WRITE

Action

0

0

Nothing happens

0

1

CPU writes to memory

1

0

CPU reads from memory

1

1

 


 

Solved Problems

1.   Calculate the effective address for each of the following instructions.  Assume that the contents of the index register 3 are (%R3) = – 7; W is the memory location 0x007D and that a partial memory map looks as follows:

 

Address

75

76

77

78

79

7A

7B

7C

7D

7E

7F

80

81

82

Contents

4D

7F

4F

BA

BA

77

8F

22

82

DD

E1

23

F0

0D

 

      a)   LDR, %R1      W                    . Remember that W refers to address 0x7D

      b)   BR *                W                    . Remember that this is indirect addressing

      c)   BR *        W, 3                                    . Handle this the way that the Boz-3b does

 

ANSWER:
      a)   Effective address is W, W = 0x7D.
      b)   Effective address is EA = M[W] = M[0x7D] = 0x82.
      c)   Pre-Indexed    EA = M[W + (%R3)] = M[0x7D – 7] = M[0x76] = 0x7F
            Post-Indexed  EA = M[W] + (%R3) = M[0x7D] – 7 = 0x82 – 7 = 0x7B

 

2.   In an assembly language program, the symbol Z refers to address 0x10B.  The index register, R3, has value –4 (negative 4).  The memory map (with 16–bit words) is:

Address

107

108

109

10A

10B

10C

10D

Contents

0113

0111

010F

010D

0108

0109

0107

What are the effective addresses for the following instructions:

      a)         Load        Z            Direct Addressing

      b)         Load        * Z         Indirect Addressing

      c)         Load        Z, 3        Indexed Addressing.

      d)         Load        * Z, 3     Indexed Indirect Addressing, Pre–Indexed

Answer:

      a)                 The address is Z itself                     Z =                              = 0x10B.

      b)                 The address is the contents of Z     M[Z] = M[0x10B]      = 0x108.

      c)                 The address is Z + (R3) = 0x10B + (–4) = 0x10B – 4        = 0x107

      d)                 The address is M[Z + (R3)] = M[0x10B – 4] = M[107]     = 0x113.


 

3.   Assemble the Boz–7 instruction LCS %R5, %R5, 9.
      Show the machine language as eight hexadecimal digits.

ANSWER:  The format of the instruction is as follows.

Unary Register-To-Register

31

30

29

28

27

26

25

24

23

22

21

20

19

18

17

16

15

14 – 0

Op Code

 

Destination Register

Source Register

Shift Count

Not Used

The op code for this instruction is binary 10001 or hexadecimal 11.
The shift count is 0x9 or binary 1001.

NOTE: The shift count must be a 5–bit number to fill bits 19 – 16; it is 01001.

The binary version of the instruction is first written as follows

31

30

29

28

27

26

25

24

23

22

21

20

19

18

17

16

15

14 – 0

10001

0

101

101

01001

0’s

Write this first as         10001 0 101 101 01001 00000
then group by fours    
1000 1010 1101 0100 1000 00
which is really            
1000 1010 1101 0100 1000 0000 0000 0000
or                                
0x8AD4 8000.

4.   Assemble the Boz–5 instruction XOR %R6, %R2, %R4.  Give the answer
      as a hexadecimal number with eight hexadecimal digits.

ANSWER:        The opcode for XOR is 0x19, or binary 11001.

      The template for the object code for this type of instruction is as follows.

Binary Register-To-Register

31

30

29

28

27

26

25

24

23

22

21

20

19

18

17

16 – 0

Op Code

 

Destination Register

Source 1

Source 2

Not used

XOR %R6, %R2, %R4 (Opcode is 11001)

31

30

29

28

27

26

25

24

23

22

21

20

19

18

17

16 – 0

Op Code

 

11001

 

 

0

Destination Register

110

Source 1

 

010

Source 2

 

100

Not used

Take the binary code and rearrange it left to right.

11001 0 110 010 100 0

1100 1011 0010 1000, which is 0xCB28 and expands to 0xCB28 0000.


 

5.   Assemble the Boz–5 instruction ADD %R3, %R2, %R1.
      Show the machine language as eight hexadecimal digits.

ANSWER:        There are two possible answers to this one.  I give both.
                           The opcode for ADD is 0x15 or 10101 in binary.

      The binary register–to–register instruction format is as follows,
      with
color added to reduce confusion.

31

30

29

28

27

26

25

24

23

22

21

20

19

18

17

16 – 0

Op Code

 

Destination

Source 1

Source 2

Not used

The destination register is clearly %R3 (011).  We have two options for the source registers.
      Source 1 is %R1 (001)      and source 2 is %R2 (010) or vice–versa.

Bits 15 – 0 of the word form 16 bits or four hexadecimal digits: 0X0000.
We now fill in bits 31 – 16, noting that each of bits 26 and 16 must be 0.

Option 1: Source 1 = 001, source 2 = 010

Bit

31

30

29

28

27

26

25

24

23

22

21

20

19

18

17

16

Binary

1

0

1

0

1

0

0

1

1

0

0

1

0

1

0

0

Hex

A

9

9

4

Option 1: Source 1 = 010, source 2 = 001

Bit

31

30

29

28

27

26

25

24

23

22

21

20

19

18

17

16

Binary

1

0

1

0

1

0

0

1

1

0

1

0

0

0

1

0

Hex

A

9

A

2

Append the four hexadecimal zeroes to either answer to get the following.

                           Answer 1:     0xA994 0000
                           Answer 2:     0xA9A2 0000

4.   The page and offset addressing scheme of the Boz–7 series is described earlier in this chapter.  Convert each of the page numbers and address offsets, given in hexadecimal to a full 26–bit address, also expressed in hexadecimal.

      a)   Page = 0X00               Offset = 0X00BAD
      b)   Page = 0X0A              Offset = 0X1CAFE
      c)   Page = 0X2F               Offset = 0X02030

ANSWER:
      The address scheme prefixes two page bytes to the five bytes of address.
      Note: The page number is a six–bit number, so the maximum page number is 0x3F.

      a)   Page = 0x00, Offset = 0x00BAD       00 00BAD      or 000 0BAD

      b)   Page = 0x0A, Offset = 0x1CAFE      0A 1CAFE      or 0A1 CAFE

      c)   Page = 0x2F, Offset = 0x02030         2F 02030         or 2F0 2030