EECE 321 Lecture Chapter 1 Lecture, Spring 2012

EECE 321 Computer Organization SPRING 2012 Instructor Information  Instructor: Dr. Lama Hamandi  Office: Bechtel 406 D  Extension: 3617  Office Hours: MF 11:30am -1:30pm, T 8:00am - 10:00am (by appointment)  E-Mail: lh13@aub.edu.lb Course Information  Prerequisite s  EECE 230 Computers and Programming  EECE 320 Digital Systems Design Course Information (2)  Textbook  David Patterson and John Hennessey, Computer Organization and Design: The Hardware/Software Interface, Fourth Edition, Morgan Kaufmann Publishers, 2009.  References  J. Bashkar, A VHDL Primer, Third Edition, Prentice-Hall, 1999.  P. Ashenden, The Student’s Guide to VHDL, MKP Publishers, 2nd Edition, 2008.  D. Perry, VHDL: Programming by Example, McGraw Hill Publishers, 4th Edition, 2002. Course Description  This course covers the organization of modern computer systems.  We will learn how to program computers at the assembly level.  We will also learn how to design the main components of a von Neumann computer system, including its instruction set architecture, datapath, control unit, memory system, input/output interfaces, and system buses.  To consolidate the material presented in class, we will work on several assembly-language programming and datapath design assignments, and a major computer interfacing project. Grading  Attendance/attitude 5 %  Assignments/Project 20 %  Quiz 1 20 %  Quiz 2 20 %  Final Exam 35 % Your grade depends on your WORK and not on your SITUATION. Exam Schedule  Quiz 1  Saturday, March 31, 2012.  Quiz 2  Saturday, May 12, 2012.  Final Exam  TBD by the Registrar’s Office. Course Policies  Lectures begin at 10:00. Late students may be refused entry to the classroom.  Students who miss a lecture are responsible for its contents.  Course withdrawal deadline is Friday, April 27, 2012.  Students who accumulate 6 or more absences by Thursday, April 26, 2012 will be withdrawn from the course.  Beyond April 26, students who accumulate 6 or more absences (excused or not) will receive a 10-point deduction on their final (course) grade. Course Policies (Continued)  All assignments are due on time.  Late assignments will be penalized (deduction of 20% per day).  Cheating on assignments or exams will not be tolerated.  A grade of zero will be given to those who cheat and those who help others to cheat.  No makeup exam will be offered for missed quizzes. Assignment #0  Please organize yourselves into groups of two.  You will have to work with the same partner for the whole semester, so choose wisely!  Email me the names in each group by Friday, February 24. Basic Structure of Computers Patterson and Hennessey; Sections 1.1 – 1.7 What is a Computer?  A fast electronic calculating machine that reads digitized input information, processes it according to a list of internally stored instructions, and produces the resulting output information.  The processed input information corresponds to data (numbers or characters) that are represented by binary code (sequences of 0’s and 1’s).  The sequence of machine instructions used to process the information is called a program. Machine instructions are also represented as binary code. What is a Computer?  But what do we mean by a computer?  Different types: desktop, servers, embedded computers…  Different uses: automobiles, graphics, finance, weather forecast…  Different manufacturers: Intel, Apple, IBM, Microsoft, Sun…  Different underlying technologies and different costs!  Best way to learn:  Focus on a specific instance and learn how it works  While learning general principles and historical perspectives General-Purpose Computers Specialized Computers Embedded Computers The NASA Mars Rover and the Honda CR- V – Computers on Wheels The Cockpit of an Airbus A321 – Where is the Computer? Computer Technology Has Advanced Very Rapidly Year Computer Technology Used 1951 Vacuum Tube 1965 Transistor 1975 Integrated Circuit 1995 Very Large Scale Integrated Circuit (100 million transistors) 2005 Ultra Large Scale Integrated Circuit (1 billion Transistors) Currently Multicore, GPUs Computers have achieved Huge performance improvement in 60 years at affordable prices. Moore’s Law and Intel Processors Picture Courtesy of Intel Corporation transistor capacity doubles every 18-24 months 2011: i7, 1.3 Billion transistors The Intel 4004 – circa 1971 Picture courtesy of Intel Corporation. The Intel Pentium 4 – circa 2000 Picture courtesy of Intel Corporation. Pentium 4 Chip Picture courtesy of Intel Corporation. Intel Core 2 Duo Picture courtesy of Intel Corporation. AMD Barcelona microprocessor The Five Basic Functional Units  In its simplest form, a computer can be divided into five basic functional units: input, output, memory, arithmetic and logic (ALU), and control. Memory Input Output I/O Datapath Control Processor The Stored Program Computer  Von Neumann proposed representing machine instructions as numbers and storing them along with the data they operated on in memory.  This led to the development of programming languages and tools (compilers, assemblers) for translating programs into machine instructions.  The stored program computer (also referred to as the von Neumann architecture) remains the guiding paradigm for all modern computer designs. Input and Output Units  Input Unit  Reads data from the external environment.  Examples: keyboard, mouse, joystick, electromechanical sensor.  Output Unit  Sends processed results to the external environment.  Example: display monitor, printer, transducer.  Some devices perform both input and output functions.  Example: modems or network cards. The Memory Unit  Used to store instructions and data.  Main Memory  Made of semiconductor material.  Operates at electronic speeds.  Secondary Memory  Made of magnetic/optical material.  Operates at electromechanical speeds (much slower than main memory). Main Memory  Inside the CPU, data is processed in units of words, multiples of words, or parts of words.  RAM = Random Access Memory.  Volatile. Once power is turned off, all data stored in the memory is lost.  Called “random access” since any memory word may be accessed in a fixed amount of time (called the memory access time) independent of its address. Typical memory access times = 10-100 nanoseconds.  ROM = Read-Only Memory  Non volatile. Data is preserved even after power is turned off.  Also random access. Secondary Memory  Much larger, cheaper (per bit), and slower than main memory.  Used to store large amounts of information, especially when it is not accessed frequently.  Typically built using magnetic and optical media.  Hard disk drives  Floppy disks  CD-RW disks  Tapes Processor Datapath  That part of the processor responsible for executing instructions.  Contains special arithmetic and logic units (e.g. adders, multipliers, shifters) for processing data.  Contains fast storage elements called registers to hold temporary values or intermediate results.  Registers can be accessed 5-10 times faster than memory.  General purpose.  Specialized. Arithmetic and Logic Unit (ALU)  Executes arithmetic and logic instructions.  ADD $R1,$R2,$R3 # $R1  $R2 + $R3  XORI $R5,$R6,100 # $R5  $R6  100  In modern processors, most operands are stored in registers. However, some operands may be stored in memory, and are accessed through special load and store instructions.  LW $R2,100($R7) # $R2  Memory[$R7 + 100]  SW $R1,20($R5) # Memory[$R5 + 20]  $R1 Processor Control Unit  Responsible for coordinating the fetching and execution of instructions.  Coordinates the operation of the input, output, memory, and arithmetic and logic units.  Sends control signals (e.g. timing, synchronization, read/write) to different units and senses their states.  Control circuits and control lines are typically distributed across the CPU and various units. Connecting the Various Hardware Components  The various hardware components are connected by collections of wires called buses.  Buses have different functions:  Data buses carry data.  Address buses carry memory or device addresses.  Control buses carry control signals. Input CPU Memory Output M-bits N-bits Data Bus Address Bus Computers = Hardware + Software  Computers rely on a number of specialized programs to manage the hardware and simplify the task of programming the computer (abstraction).  Operating systems.  Compilers.  Assemblers.  Loaders.  Linkers. An abstraction omits unneeded details and helps the user cope with complexity. The Operating System  Program responsible for managing and scheduling system resources to ensure efficient and safe hardware operation.  Provides access to hardware resources through system calls (e.g. cin, cout in C++).  Enables multiple programs to share common resources.  Allocates memory space and prevents programs from overwriting each others’ data. Operating System Device Drivers Application Program 1 Application Program 2 Application Program 3 Hardware Processor Compilers, Assemblers, Linkers, and Loaders a = b + 5; c = d * R[7]; ADDI R1,R2,5 MUL R3,R1,28(R4) 01101001101110101110101110001101 11101011101011100011010110100110 Compiler Assembler Linker User Code Code Libraries Relocatable Code Loader Compilers, Assemblers, Linkers, and Loaders (2)  A compiler is a program that translates high-level language application programs to assembly code.  An assembler is a program that translates assembly code to machine code.  Assemblers may be integrated with compilers, in which case the compiler can translate a HLL program into machine code directly.  A linker is a program that links application programs with code libraries.  #include <iostream>  A loader is a program that loads a program into memory and directs the processor to begin executing the program. Why learn this stuff?  You want to call yourself a “computer engineer”  You want to build software people use (need performance)  You need to make a purchasing decision or offer “expert” advice  Both Hardware and Software affect performance:  Algorithm determines number of source-level statements  Language/Compiler/Architecture determine machine instructions  Processor/Memory determine how fast instructions are executed  Assessing and Understanding Performance Computer Performance What is “Performance”?  There is no single definition. It depends on how you define it, and how it is measured.  Example: Which airplane has better “performance”?  Fastest: Concorde.  Longest range: DC-8.  Largest capacity: B747.  Highest passenger throughput: B747. Airplane Passenger Capacity Cruising Range (KM) Cruising Speed (KPH) Passenger Throughput (passengers × KPH) Boeing 777 375 7408 976 366000 Boeing 747 470 6640 976 458720 BAC/Sud Concorde 132 6400 2160 285120 Douglas DC-8-50 146 13952 870 127020 What is Computer Performance?  Roughly, computer performance can be measured in terms of execution time or execution throughput.  Execution time = time needed to complete a single task.  More important to an individual user.  Execution throughput = rate at which tasks can be completed.  More important to a service provider.  The two measures are interrelated.  Reducing execution time almost always improves throughput.  Improving throughput also improves the execution time of individual tasks since they would have to spend less time waiting for other tasks to finish first. Relative Performance  Example Problem:  Machine A runs a program in 10 seconds, and machine B runs the same program in 15 seconds. How much faster is A than B?  Performance = (Execution Time)-1.  The smaller the execution time, the higher the performance.  Machine A is 1.5 times faster than machine B. 1.5 10 15 Time Execution Time Execution e Performanc e Performanc A B B A    Measuring Performance  Since performance is related to execution time, it can be measured in seconds per program (wall clock time).  However, many factors contribute to total execution time.  Memory accesses.  Disk accesses.  I/O activities.  Operating system overhead.  Running other tasks.  CPU execution time = time spent executing a program without any overhead.  Can further be subdivided into user time (CPU time spent in the program) and system time (CPU time spent in the OS performing tasks on behalf of the program). Performance of the Hardware  Computers are digital systems whose operation is synchronized to a clock.  A clock cycle is the smallest time unit in which events take place in the hardware.  Typically, the speed of the clock is measured by the clock cycle time (period) or its inverse, the clock rate (frequency).  How is clock cycle time (or clock rate) related to CPU execution time?  CPU execution time = CPU clock cycles × clock cycle time.  CPU execution time = CPU clock cycles ÷ clock rate. How to Improve Performance?  A designer can help improve performance by:  Reducing the number of CPU clock cycles.  Reducing the clock cycle time (increasing clock rate).  This typically is a trade-off, since these are often contradictory goals.  Example Problem:  Computer A, running at 400 MHz, executes a program in 10 seconds. We want to design computer B so that it executes the same program in 6 seconds. The hardware design team indicates that a significant improvement in clock rate is possible, but that this will increase the number of clock cycles by a factor of 1.2. What should the target clock rate be?  Answer: CPU execution time = CPU clock cycles ÷ clock rate A: CPU clock cycles = 10  400 106 B: 6 = CPU clock cycles ÷ clock rate B: 6 = 400 107 1.2 ÷ clock rate B: clock rate = 800 MHz Performance also Depends on the Compiler  Compilers determine the number of instructions generated for a given program, which, in turn, determines the number of CPU clock cycles executed.  CPU clock cycles = Number of instructions × Avg. clock cycles per instruction.  Average clock cycles per instruction = CPI.  CPI depends on the type of instructions (complex instructions take longer to execute than simple instructions; they have larger CPI).  CPI is a good way to compare the performance of different implementations of the same instruction set architecture. The Impact of Implementation on Performance  Example Problem:  Assume there are two implementations of the same instruction- set architecture (ISA).  Machine A has a clock cycle time of 1 ns and, for a given program, CPI = 2.0; machine B has a clock cycle time of 2 ns and a CPI = 1.2 for the same program (i.e. same number of instr.)  Which machine is faster, and by how much?  Answer: machine A is 1.2 times faster than machine B for this program.  CPU execution time = CPU clock cycles × clock cycle time = instruction count × CPI × clock cycle time 2.1 *1 2 *2 2.1 * * execution time CPU execution time CPU A B     A A B B B A clockcycletime CPI clockcycletime CPI e Performanc Performance The Basic Performance Equation  CPU performance depends on instruction count, CPI, and clock cycle time.  CPU time = instruction count × CPI × clock cycle time.  CPU time = (instruction count × CPI) / clock rate.  Instruction count depends on the compiler and the ISA.  CPI depends on the ISA and the hardware implementation.  Clock cycle time depends on the hardware organization and implementation technology. Cycles Clock Seconds n Instructio Cycles Clock Program ns Time Instructio    Measuring Performance Components  CPU time = instruction count × CPI × clock cycle time  CPU time measured by running the program.  Clock cycle time (clock rate) published by hardware manufacturer.  Instruction count can be obtained through application profiling (dynamic program analysis), architecture simulation, or hardware counters.  CPI is the most difficult to measure since it depends on the ISA, the mix of instructions executed, and hardware organization and implementation. Typically, CPI is measured using a detailed architecture simulation, or by combining simulation with hardware counters. CPU time = instruction count × CPI × clock cycle time Performance = 1/CPUtime = f / (instr cnt * CPI) Since instr. Cnt is the same, compare f / CPI P2 has highest performance. 1.3 solution instr s instr cc cc s / / /  CPI in More Detail  If different instruction classes take different numbers of cycles     n 1 i i i Instructio n Count ) (CPI Cycles Clock  Weighted average CPI            n 1 i i i n Count Instructio Instructio n Count CPI n Count Instructio Clock Cycles CPI Relative frequency CPI Example  Alternative compiled code sequences using instructions in classes A, B, C Class A B C CPI for class 1 2 3 IC in sequence 1 2 1 2 IC in sequence 2 4 1 1  Sequence 1: IC = 5 instruction count  Clock Cycles = 2×1 + 1×2 + 2×3 = 10  Avg. CPI = 10/5 = 2.0  Sequence 2: IC = 6  Clock Cycles = 4×1 + 1×2 + 1×3 = 9  Avg. CPI = 9/6 = 1.5