Chapter 7 – The CPU and Memory


 

         This chapter covers the following main topics:

o        The components of the CPU

o        The concept of Registers

o        The memory unit and its operations

o        The fetch-execute instruction cycle

o        Instruction format

o        Classification of instructions


Introduction

         This chapter and the next few chapters extents the Little Man Computer model to discuss the real computer hardware architecture

         This chapter covers, in particular, both the Central Processing Unit (CPU) and Memory

         CPUs varies in their capabilities, complexity, and even applications

         The CPU model covered in this chapter is not based on any particular model

         Memory is separated both physically and functionally from the CPU

         The CPU and memory are very closely associated

o        Every instruction the CPU executes requires at least one, and possibly more, memory access


The Components of the CPU

         Figure 7.1 page 166 shows a simplified system block diagram of a typical general-purpose CPU and memory structures

         A simplified internal structure of a typical CPU can be viewed as consisting of 3 main components

o        Arithmetic/logic unit (ALU)

         Responsible for logical and arithmetic calculations

o        Control unit (CU)

         Responsible for managing and controlling the instruction execution

o        I/O interface

         Responsible for managing I/O operations

         Defines an interface between the CPU and the various I/O devices

         Figure 7.2 in page 167 shows the relationship between the LMC model and a real computer model

         This diagram shows that the 2 models are conceptually very similar

o        Little Man                     correspond to the CU function in the real computer

o        Calculator                     correspond to the ALU function in the real computer

o        Mailbox                          correspond to the memory function in the real computer

o        Program counter         correspond to the PC register function in the real computer

o        I/O baskets                   correspond to the I/O interface function in the real computer

 

The Concept of Registers

         Register is a hardware component which provides temporary data storage capability within the CPU

         Registers storage is required during instruction execution

         Registers provide very limited storage capability (typical register size: 16, 32, 64, 128 bits)

         Therefore, a register is typically used to store a single data unit (e.g. memory address, integer number)

         Register access is extremely fast (fastest storage device) as it sets at the top of the storage hierarchy

         Registers are accessed directly by the Control Unit (i.e. not addressed as a memory location)

         Modern CPUs would typically have few 100’s of registers

         Registers are referred to by their names where each has a unique name

         Different CPUs usually vary in their

o        Number of registers

o        Size of each register

o        The purpose of each register

o        Registers naming convention

         Each register has a particular and predefined purpose such as:

o        Hold data being processes

o        Hold instruction to be executed

o        Hold memory or I/O address to be accessed

o        Hold status data for tracking computer and execution statuses

o        Hold numbers in different numerical formats (unsigned integer, signed integer, floating point)

 

         The following are typical operations that can be performed on registers

o        Load it with data from another location (mostly from other registers or memory)

o        Add or subtract data from other location and store result in the register

o        Shift or rotate content left or right by one or more bits

o        Clear register by setting its content to zero

o        Increment or decrement content

o        Bit inversion

 

         Registers can be classified into 3 main categories:

o        General-purpose or accumulator or user-visible registers

         Are used to hold general-purpose data (e.g. integer number)

         Can be directly referenced by application programs

o        Privileged registers

         Can only be accessed by the operating system

         Used in the implementation of operating system services (e.g. memory management)

o        CPU internal usage registers

         Can only be accessed by the CPU

         Used to during instruction execution (e.g. hold instruction being executed)

 

         The control unit uses several registers for its internal use:

o        Program counter register (PC) holds the memory address of the instruction to be executed

o        Instruction register (IR) holds the actual instruction being executed

o        Memory address register (MAR) holds a memory address to be accessed

o        Memory data register (MDR) holds the data that has been retrieved from memory (in case of read) or to be stored into memory (in case of write)

o        Status register (SR) holds 1-bit flags for keeping track of special condition during instruction execution (e.g. carry, overflow, power failure, internal error, etc.)

 

         See Figure 7.3 in page 153 for registers available in the IBM zSeries computers


The Memory Unit

         Memory is responsible for temporarily holding instructions and data during program execution

         Data and instructions in memory are immediately available to the CPU

         Memory is connected to the CPU through special system bus

         Memory consists of linear storage cells

         Each cell is capable of storing a single data unit

         In most modern computers, the cell size is 1 byte (i.e. smallest addressable data unit)

         Memory cells are usually grouped to store large data unit (e.g. integer number)

         A memory cell is referenced by its address

         Each address is a unique unsigned integer number, starting at address 0

         All memory cells are identical in their purpose (e.g. store general purpose data)

 

The Operation of Memory

         The MAR and MDR registers acts as an interface between the CPU and memory

         MAR

o        Holds the address in memory which to be open for data access

o        Connected to address decoder that resolves the memory address to be accessed

         Address decoder

o        Interpret the address passed from MAR to identify the address to be accessed

o        Activate the appropriate address line in memory for access

         MDR

o        MDR is data holding register during memory access

o        Store the content of memory that is currently addressed by MAR in case of read

o        Store the data to be stored into memory in case of write

o        The word size that can be retrieved or stored into memory in a single operation is determined by the size of MDR

 

         There are 3 control lines that controls memory access, see Figures 7.4 - 7.7 in pages 154 - 157

o        Address line

         There is 1 address line for every memory cell

         An address line for a particular cell is turned only when addressing data within that cell

o        Read/write line

         Determines whether the memory access is a read or a write operation

         Turned on in case of read and off in case of write

o        Activation line

         Used to control the memory access operation

         Turned on to start a memory access operation (i.e. read or write)

 

         See Figure 7.6 in page 172 to see how memory access is done

 

         The following are the steps taken to load data from a particular memory location (read operation)

1.        CPU copies the address of memory to be accessed into MAR

2.        CPU sets the Read/Write switch on to indicate a read operation

3.        CPU sets Activation line on to start the data transfer

4.        Data transfer takes place retrieving data from the specified memory location and store it into the MDR register

5.        Data is copied from the MDR into another register depending on the type of read data

 

         The following are the steps taken to store data into a particular memory location (write operation)

1.        CPU copies the address of memory to be accessed into MAR

2.        CPU copies data into MDR

3.        CPU sets the Read/Write switch off to indicate a write operation

4.        CPU sets Activation line on to start the data transfer

5.        Data transfer takes place to store data into the specified memory location

 

         Only one memory location is addressed at any given time to prevent conflict

 

Memory Capacity

         The size of the MAR register is the main determining factors of the addressable physical memory

o        Memory Capacity = 2n where n is the MAR size in bits

         In today’s computer a typical MAR is at least 32-bits which allows 4GB of memory addressing

         Many modern CPUs supports 64-bits which allows 264 = 16 x 1018 bytes of memory addressing

         In these computers the actual limiting factor for memory capacity would be

o        Physical space for fitting large number of memory chips

o        Time requirement for decoding and accessing such a huge memory

 

Memory Implementations

         The most popular types of memory in use by computers are

o        Magnetic core memory

o        Static RAM

o        Dynamic RAM

o        Read Only Memory (ROM)

 

Magnetic Core Memory

         Magnetic core memory is an old technology supporting non-volatile storage (i.e. data is maintained after power is turned off)

         Magnetic core memory is expensive and slow and has been mostly replaced by RAM

         Few special purpose computers still uses this type of memory where non-volatile storage is required

 

RAM

         Most current computers uses either static or dynamic RAM

         Both static and dynamic RAM are volatile and writeable (i.e. data is lost when power is turned off)

         Dynamic RAM advantages over Static RAM:

o        Less Expensive

o        Requires less power

o        Can be made smaller (i.e. can be made to fit more memory in a single chip)

         Static RAM advantages over Dynamic RAM

o        Does not need periodic refreshing

o        Faster access time (Static memory is what cache memory is made of)

 

         Dynamic RAM is more popular than static RAM

         The size of memory in a single dynamic RAM chip has increased significantly in the last few years (from fewer than 64KB to over 64 MB)

 

ROM

         Supports non-volatile storage

         Read only (i.e. data never changes)

         Mostly used to store system programs and data used at computer starts up (i.e. boot time)

         Modern ROM is made so it can be erased and reprogrammed in the factory

         However within the computer ROM is still read only

 

EEPROM and Flash ROM

         Recent memory innovation supporting non-volatile and writeable memory

         Typically slower than ROM

 


The Fetch-Execute Instruction Cycle

         The fetch-execute cycle in a real computer works very similar to the one for LMC

         The fetch-execute instruction cycle is a 2 phase process

1.        Fetch phase                   fetch an instruction from memory for execution

2.        Execute phase               execute the instruction

         The fetch phase is consistent and follow the same steps for all types of instructions

         The fetch phase consists of the following steps:

1.        Copy content of PC to MAR – this result in transferring the instruction located at the specified address to the MDR register

2.        Copy content of MDR into IR

3.        Decode the instruction

 

         The execution phase varies between the different instructions

         The LOAD instruction execution steps

  1. Copy address specified in the instruction to MAR
  2. Copy content of MDR register into a general-purpose register
  3. Increment the PC register                  

 

         A special notation is used to describe the fetch-and-execute phase steps for instruction execution

REGa → REGb                              transfer data from one register to another

REG[address] → REGb               transfer the address part of the register content to another register

REGa + REGb → REGc                add content of two registers and store result into a third register

REGa + const → REGb                add content of register to constant and store result into a third register

 

Examples

         The LMC STORE instruction fetch-execute steps

1.        PC → MAR

2.        MDR → IR

3.        IR[address] → MAR

4.        A → MDR

5.        PC + 1 → PC

 

         The LMC ADD instruction fetch-execute steps

1.        PC → MAR

2.        MDR → IR

3.        IR[address] → MAR

4.        A + MDR → A

5.        PC + 1 → PC

 

         The LMC BR instruction fetch-execute steps

1.        PC → MAR

2.        MDR → IR

3.        IR[address] → PC


Buses

         A bus is the physical interface to interconnect the different components within the computer system

         This topic will be covered in more details in chapter 8


Instruction Word Formats

         The instruction set format in a typical real CPU is similar to the one supported by the LMC

         The instruction word format is divided into 2 parts

o        Op code

o        Address fields

         See Figure 7.14 in page 184 for an example 32-bit instruction format

         The address field may refer to register address, memory address or constant data

         Two types of addresses are defined

o        Source address

o        Destination address

         The source and destination addresses are also called operands

         The address may be expressed

o        Explicitly as an address field in the instruction word

o        Implicitly as part of the definition of the instruction (i.e. no address field is necessary)

         On some computers one or more of the addresses may be implicit

         In modern computers most address references are explicit

         Instructions with 1 address field are called unary instruction

         Instructions with 2 address fields are called binary instruction

         Instructions with 3 address fields are called ternary instruction

         The source(s) and destination addresses may be the same or may be different

         Example of instruction with the same source and destination is a complement of a register content

         The ADD instruction in LMC uses the accumulator register as both source for one of the numbers and as the destination for the result

         The MOVE instruction obviously uses different source and destination addresses


Instruction Word Requirements and Constrains

         The word size of instructions depends on

o        The CPU design of the instruction set

o        The usage of the address fields

         2 techniques are in common use for instruction word size design

o        Fixed length                  instruction size is the same for all instructions in the instruction set

o        Variable length            instruction size may vary between instructions within the instruction set

         In IBM S/390 systems most instructions are 32-bits long but few are 16-bits or 48-bits long

         x86 uses 8-bits and 16-bits long instructions and some Pentium instructions can be as long as 88-bits

         Figure 7.15 in page 172 shows the IBM mainframe and the SPARC RISC instruction formats

o        The IBM instruction set defines variable word size (16-bits to 48-bits)

o        The SPARC RISC defines fixed word size of 32-bits

 

Advantages of Fixed Length Instruction design

         Simple to design

         Simplifies the fetch operation

         It is easy to locate any instruction since every instruction takes the same amount of space in memory

         The program counter adjustment is simple (i.e. incremented by instruction word size - fixed)

         The Skip and Jump instructions (involves adjusting the program counter) are also simple

 

Advantages of Variable Length Instruction design

         Efficient use of memory since each instruction is only as long as it needs to be. Typically designed so that most frequently used instructions have the shortest word size

         The use of different lengths within the instruction set is used to provide flexibility

o        The ability to address large address space

o        Expand of instruction set to support complex instructions that specifies 2 or more address fields (e.g. instruction that add data from two memory location and put result in a third memory location)

 

         Many new CPU designs uses the fixed length instructions exclusively due to its simplicity


Classification of Instructions

         Most instructions manipulate data

         Very few do not operate on data like

o        Flow control instructions (e.g. jump)

o        Control and administration instructions  (e.g. halt)

o        No operation instruction (allows programs to create time delays)

         Most modern computers also provides instructions for the operating system

         These instructions are known as privileged instructions

         Computer manufacturers usually group instructions into categories

         Within a category instructions usually have

o        Similar instruction format

o        Support similar addressing modes

         Figure 7.16 in page 174 shows the instruction set for the Motorola 68000 CPU

         The Motorola 68000 CPU divides its instruction set into 8 categories:

1.        Data Movement instructions

2.        Integer Arithmetic instructions

3.        Boolean Logic instructions

4.        Shift and Rotate instructions

5.        Bit Manipulation instructions

6.        Bit Field instructions

7.        Binary Coded Decimal instructions

8.        Program Flow instructions

9.        Privileged instructions

10.     Exception handling instructions

11.     Floating point arithmetic instructions

Note: the Privileged, Exception handling, and the floating point arithmetic instructions are not listed

 

Data Movement Instructions

         The data movement category typically include instructions to

o        Move data from memory to general registers

o        Move data from general registers to memory

o        Move data between general registers

o        In some computers move data directly between memory locations

         Variations of these instructions are used to handle different data sizes (e.g. LOAD byte, LOAD half-word, LOAD word, and LOAD double-word)

 

Arithmetic Instructions

         Every CPU instruction set include instructions to perform integer addition and subtraction arithmetic

         With the exception of very few special-purpose CPUs, integer multiplication and division arithmetic are also supported

         Many instruction sets provides integer arithmetic for different data size (e.g. byte, half-word, word, double-word)

         Many instruction sets provides integer arithmetic in different addressing modes (i.e. memory and register combinations)

         Most modern CPUs also provide floating point arithmetic capabilities

         In some cases a separate match coprocessor is needed to enable the floating point capabilities (e.g. 80386 and later brands)

         This capability significantly enhances the performance of programs that uses extensive floating point operations (e.g. CAD/CAM programs)

         Floating point instructions usually operate on a separate set of registers with 64-bits or 128-bits word size

         Modern instruction sets also include instructions to convert between integer and floating point

         Most modern instruction sets also provide instructions to perform BCD arithmetic

 

Boolean Logic Instructions

         Most modern instruction sets provide instructions for performing Boolean algebra logic operations

         Commonly include NOT, AND, XOR, OR instructions

 

Single Operand Manipulation Instructions

         Most instruction sets also provide convenient single operand instructions (e.g. negate, increment, decrement a value, setting register to zero, etc.)

         Most of these instruction operate on registers but some can also operate on memory

 

Bit Manipulation Instructions

         Most instruction sets provide instructions for setting, resetting, and testing individual bits in a data word

         Some instruction sets also provide instructions to operate on multiple bits at once

 

Shift and Rotate Instructions

         Most instruction sets provide instructions for shifting and rotating data words

         Shifting is the process of moving data to left or right one or more bits

         2 different kinds of shifting are usually provided, see Figure 7.17 in page 177

o        Logical shift

o        Arithmetic shift

         Rotating takes the bits as they exists and rotate them into the other end

         Some instruction sets include carry and overflow bit as part of rotate and shifting instructions

 

Program Control Instructions

         Program control instructions control the flow of the program execution

         Program control instructions include

o        Jump conditionally (e.g. branch on zero, nonzero, negative, positive, carry, overflow, etc.)

o        Jump unconditionally

o        Subroutine call sometimes know as jump subroutine and return, see Figure 7.18 in page 179

         Used to implement subroutines, procedures, and function calls

         A returning address (next instruction after call) must be saved (typically save on the stack)

         A jump to the subroutine address is done after the save

         The return instruction reload the saved return address and jump back

 

Stack Instructions

         Stack constitute one of the most important data storage structures in programming

         Stack is implemented as last-in-first-out LIFO structure

         The stack pointer register is used to keep track of the top of the stack

         Two operations are typical in the stack, see figure 7.19 in page 194

o        Push               add a data element to the top of the stack

o        Pop                 remove a data element from the top of the stack

         Subroutine call typically use stack to save the returning address

         The stack allows for nested subroutine calls (returning is done in a reversing order), see Figure 7.21 in page 181

 

Multiple Data Instructions

         Specialized instructions to speed up and simplify multimedia processing

         Multimedia operations are characterized by simple instructions applied identically to every piece of the data in a set (e.g. move an image, scale an image)

         Multiple data instructions perform a single operation on multiple pieces of data concurrently

         Also called Single Instruction, Multiple Data (SIMD)

         Intel Pentium MMX support SIMD instructions

         Figure 7.23 in page 182, shows 4-wide SIMD add instruction

 

Other Instructions

         This category of instructions include I/O and machine control instructions

         These instructions, in most systems, are privileged instructions

         These instructions are used by the operating system to implement important operating system services (e.g. I/O device drivers, memory management, etc.)