Monday, August 28, 2017

Transcription

1 Monday, August 28, 2017 Topics for today Course in context. Course outline, requirements, grading. Administrivia: Tutoring: Department, PLTL, LRC Knowledge Survey The concept of a multi-level machine Motivations for knowing how to program at assembly level COMP 162 in context You can see the position of COMP 162 in both the Computer Science and the Information Technology degree programs in the on-line charts In each degree program it is the first hardware-oriented course. In the BSCS it leads into COMP262 and COMP 362. In the BSIT, it provides some foundation for COMP 221 though it is not a prerequisite for that course. Comp 162 is also a required course in the minor in Computer Science and the minor in Robotics Engineering. Comp 162 is an introduction to computer systems, low-level programming, and a brief introduction to topics covered in more depth in later courses. The official pre-requisite is Comp 150 or IT 151 or COMP 121 or IT 152 In practice, the pre-requisite is some knowledge of programming in a high-level language so that when we discuss loops and if statements, you know what we are talking about. Textbook Computer Systems (5th Edition), J. Stanley Warford, Jones and Bartlett (2016) An online copy of the book is accessible through the CS Safari Books page. Course outline The first third of the course is a general introduction to computer architecture, data representation, the von Neumann model The second third looks specifically at the Pep/8 system, low-level programming techniques and how high-level language features map onto assembly language. The final third is some language theory (relevant to compilers) and introductions to systems topics covered in later courses. Comp 162 Notes Page 1 of 9 August 28, 2017

2 Lab The lab sessions (3 hours each week) will introduce the C language and then the Pep/9 programming environment and give you supervised time to work on the programming assignments (first two in C, remaining 4 in Pep/9). Administrivia The course uses both CI Learn (Blackboard/Canvas) and the web. CI is transitioning to Canvas; this is the last semester for Blackboard. Key information such as grades will be posted to both systems this semester. Most information will be posted on the course web page to avoid duplication. My home page is: A link to the Comp 162 course page is about halfway down the page. Contact information (phone, , office, office hours etc) is at the top of the page. Course requirements (probable) Three homeworks, six programs (C/Pep), three mid-term exams and one final exam. Note that all the exams are open book, open notes. Grading In figuring the final course grade this semester, components are weighted approximately as follows Final 16% Homeworks 8% each Mid-terms 8% each Programs 6% each Plus-minus grading will be used in the course grade. Tutoring In SIE 1119, in El Dorado Hall and in the Learning Resource Center (Broome Library) PLTL Peer-led team learning - available for Comp 162 in Fall Recommended, but space is limited. Knowledge Survey An anonymous before-and-after survey of course-related knowledge Comp 162 Notes Page 2 of 9 August 28, 2017

3 Overview: The concept of a multi-level machine The course textbook is organized round the idea of a computer system being a multi-level machine. This way of approaching a system (think of an onion) does not originate with Warford but it is a good way to break up the course topics. In Warford s view, a system has 7 levels with the application layer (the outermost) being level 7 and the logic circuit layer (the innermost) being level 1. Somewhat simplified, we have: Level 7: Level 6: Applications, e.g., Word, Safari, computer games written in high-level language. High-level language, e.g., C, Java which a compiler translates to assembly language Typical statement: vector[j]=45; Level 5: Level 4: Level 3: Level 2: Level 1: Assembly language (symbolic) translated by assembler into numeric form Typical statements: ldwa 45,i ldwx J,d stwa vector,x Operating system. Some instructions in the output from the assembler are processed directly by level 3, others (e.g. input/output) require action by the operating system. The Operating System itself is held in the form of machine code instructions. Machine Code binary instructions. A program of very simple microinstructions interprets the machine code program. Circuits are activated according to the current microinstruction. This is where, finally, computation actually happens. Different courses in the degree programs cover different parts of this model. In Comp 162 the topics we look at include Level 6 programming in C How an assembler works (the mapping of level 5 onto levels 4 and 3) The relationship between elements of High-Level language programs (level 6) and their counterparts at level 5. Some aspects of Operating Systems (Level 4) The operations possible in Level 3. Comp 162 Notes Page 3 of 9 August 28, 2017

4 1: Circuit level 2: Microinstruction level 7: Application layer 6 High-level language level 5: Assembly language level 4: Operating System Level 3: Instruction Set Architecture level Comp 162 Notes Page 4 of 9 August 28, 2017

5 Implementing a multi-level machine How can we implement one level on the one below it? To see some possibilities, think about the problem of implementing a very high-level language on a system that can run a high-level language (for example C). Here are two examples of very highlevel languages. Example 1: APL. (a real language) The following APL program computes the mean of the numbers in an array {(+/ω) ρω} The following APL program (from Wikipedia) sorts a list of word in array X by word length Example 2: VECMAT (a made-up language) VECMAT is a high-level language for vector and matrix manipulation. vector A,B; matrix C,D; input A, C; B = C * A; output B; Two options for implementing a level (level N) on the level beneath (level N-1) are: translation and interpretation. Consider the problem of implementing VECMAT on a machine that can run C. (a) translation. Write a translator in C that translates a VECMAT program into an equivalent C program. We can run this translator on any system that has C then subsequently run the C program that the translator outputs. (b) interpretation. Write a program in C that interprets the VECMAT program. The interpreter does much the same actions that you would do when tracing a program keep track of contents of variables, remember which instruction you are tracing, take appropriate action for each instruction. These implementation methods (translation, interpretation) are found at various points in the multilevel machine when we implement level N on level N-1. Comp 162 Notes Page 5 of 9 August 28, 2017

6 Motivation for studying low-level programming Most users of computer systems are at the application level, they don t care how the system actually works as long as it runs their applications (Safari, PeopleSoft, Microsoft Office, etc.) correctly. So why study computer architecture and low-level (assembly language programming)? 1. Knowing how computers work lets you compare different systems. For example, what is RISC architecture and why might it be superior to non-risc? How are Intel and AMD processors similar and how are they different? 2. Assembly languages are very similar and once you have studied one, another is relatively easy to pick up. 3. If you are involved in writing compilers or operating systems or other types of system software or you are programming embedded systems there may be circumstances where a high-level language gets in the way of what you want to do. For example, it might not be possible to access some of the hardware from within a high-level language program. Then you may need to resort to embedding assembly code in your program. 4. The output from a compiler may be too big or run too slowly for your particular circumstance, e.g., writing code for a system with limited memory. Being able to tweak the assembly code output from a compiler is a valuable skill as is being able to tweak the highlevel language so that the compiler produces the program with the characteristics you want. Here are four example scenarios. Example 1 You are developing a control program in C for an embedded system, for example an MP6 player (it s only a matter of time). The output from the compiler currently occupies 4130 bytes. By knowing how the compiler translates various data and control structures, you make appropriate changes to the C source code, recompile and the new object program is only 4020 bytes. Now the code fits in a 4K memory (4096 bytes) rather than requiring an 8K memory (assuming memories are multiples of 4K bytes). You have reduced the cost by perhaps $1 per unit significant if you are producing millions of units. Example 2 Here is an example of how knowledge of assembly language lets us improve on compiler output. The source code in this example initializes, to zero, the first 100 elements of an array. for (i=0; i<100; i++) A[i] = 0; Comp 162 Notes Page 6 of 9 August 28, 2017

7 (a) Compiler output. Without any kind of optimizations, a compiler might produce the following code (somewhat generic assembly language) MOVL #0,I MOVL #0,R3 LAB: MOVL #0,A[R3] ADDL #1,R3 ADDL #1,I COMPL R3,#100 BLSS LAB (b) If we were to translate the loop by hand, we might come up with the following that uses more sophisticated programming techniques MOVL #A+100,R3 LAB: CLRL (R3)- COMPL R3,#A BGTR LAB (c) An expert assembly language programmer, with knowledge of the complete instruction set available, might be able to use instructions in a truly innovative way, leading to the following program CLRL A MOVC3 #396,A,A+4 We can compare these three assembly code programs (very approximately) under two headings: space and time: How many bytes of memory does each occupy? How long does each take to run? For the former, we approximate space requirements by simply counting the number of symbols in the program; for the latter we count the number of instruction executions as the program runs. This leads to the following table. Program Space Time A (Compiler) B (Hand-compile) C (Hand-compile: expert) 6 2 The time comparison is a little unfair because the MOVC3 instruction in the last program will take a lot longer to run than the average instruction. Nevertheless, we can see that substantial improvements are possible over the code output by the compiler. (We will look at optimization techniques later.) Comp 162 Notes Page 7 of 9 August 28, 2017

8 Example 3 In a certain program, 1% of the code accounts for 50% of the execution time. This is not unrealistic, a 600-line program might have a 6-line inner loop where the program spends most of its time. Assuming: It takes 100 person-months (pm) to write the entire program in C Assembly code takes 10 times longer to write than C Assembly code runs 4 times faster than C. Compare the following strategies with respect to programming time and execution time. (1) Write the entire program in C (2) Write the entire program in assembler (3) Write the entire program in C then rewrite the critical 1% in assembly code Strategy Programming Time Execution Time pm X pm 0.25X = 110pm 0.5X X = 0.625X So for a small (one-time?) extra investment in programming time, we have reduced execution time by 37%. Example 4 In most cases, one high-level language instruction is translated into a sequence of many assembly language instructions. However, consider the following fragment of a high-level language program F = 0 For ( I = N; I > = 0; I --) F = F * X + C[ I ] Did you recognize this as the evaluation of a polynomial user Horner s Method 1? If you did not recognize it, what are the chances of a compiler recognizing it and being able to translate it into the following single assembly code instruction (DEC VAX)? POLYD X,N,C A programmer on the other hand, knowing the problem to be solved (polynomial evaluation) and the tools available (POLYD) can put the two together. 1 See, for example, Comp 162 Notes Page 8 of 9 August 28, 2017

9 Why Pep/9? We use the Pep/9 virtual machine developed at Pepperdine University. It is an integrated environment that lets you create, edit, assemble and run assembly code or machine code programs. Pep/9 is the latest version of a long-running project that you can find at Click on the Fifth Edition link in the Editions tab. We use Pep/9 rather than a real system (e.g., Intel, AMD) because (a) versions are readily available for Mac and Windows and (coming soon?) Linux (b) It is just the right size (about 40 instructions) for a one-semester course. Unfortunately, there is only one textbook that describes the system but it is relatively straightforward Reading. Read the Preface and Chapter 1 (only 1.2 is really needed). Chapter 2 is relevant to the first few labs. We will begin covering Chapter 3 material in the next lecture. Comp 162 Notes Page 9 of 9 August 28, 2017