Department of Computer Science and Engineering Yonghong Yan

Transcription

1 Appendix A and Chapter 2.12: Compiler, Assembler, Linker and Program Execution CSCE 212 Introduction to Computer Architecture, Spring Department of Computer Science and Engineering Yonghong Yan yanyh@cse.sc.edu

2 Appendix A and Chapter 2.12: Compiler, Assembler, Linker and Program Execution Lecture 04 A.1 Introduction A.2 Assemblers A.3 Linkers 2.12 Translating and Starting a Program Hands-on Lecture 05 Dynamic linking, compilation process and preprocessing 2.12 Translating and Starting a Program Introducing memory Hands-on Lecture 06 Basic C programming A.4 Loading and A.5 Memory Usage To be covered as part of Chapter 2 2

3 Review of Lecture 03 and Chapter 01 3

4 Review: Power, Frequency and ILP Moore s Law to processor speed (frequency) CPU frequency increase was flattened around Two main reasons: 1. Limited ILP and 2. Power consumption and heat dissipation Note: Even Moore s Law is ending around 2021: conductors/devices/transistors -could-stop-shrinking-in w.com/s/601441/moores-lawis-dead-now-what/ imworstall/2016/07/26/econo mics-is-important-the-end-ofmoores-law 1.8 The Sea Change: The Switch to Multiprocessors 4

5 Multiprocessors Multicore microprocessors become main-stream starting from 2005 More than one processor per chip Requires explicitly parallel programming Compare with instruction level parallelism Hardware executes multiple instructions at once Hidden from the programmer Hard to do Programming for performance Load balancing Optimizing communication and synchronization 5

6 Review for Chapter 1: Three Most Important Topics Moore s Law: From 1970s, transistor density doubles every years CPU frequency (performance) doubles every 2 years till ~2005 From 2005, transistor density still double every years, CPU frequency flats Industry moves to multicore, manycore and multiprocessing Abstraction High-level language, assembly language/isa as HW/SW interface, binary for computer Software interface, method declaration and definition CPU Performance Power: Linearly proportional to the CPU frequency 6

7 Review: CPU Time Performance improved by Reducing number of clock cycles Increasing clock rate Hardware designer must often trade off clock rate against cycle count CPU Time = CPU Clock Cycles Clock Cycle Time = CPU Clock Cycles Clock Rate 7

8 Review: CPU Time, Instruction Count and CPI Hardware/CPU executes a program instruction by instructions Clock Cycles = Instruction Count Cycles per Instruction CPU Time = Instruction Count CPI Clock Cycle Time Instruction Count CPI = Clock Rate Instruction Count for a program Determined by program, ISA and compiler Average cycles per instruction Determined by CPU hardware If different instructions have different CPI Average CPI affected by instruction mix 8

9 Review: CPI Example Alternative compiled code sequences using instructions in classes A, B, C Class A B C CPI for class IC in sequence IC in sequence n Sequence 1: IC = 5 n Clock Cycles = å i= 1 Clock Cycles = = 10 n Avg. CPI = 10/5 = 2.0 n (CPIi Instruction Count i) n Sequence 2: IC = 6 n Clock Cycles = = 9 n Avg. CPI = 9/6 = 1.5 9

10 End of Review of Lecture 03 and Chapter 01 10

11 Appendix A and Chapter 2.12: Compiler, Assembler, Linker and Program Execution Lecture 04 A.1 Introduction A.2 Assemblers A.3 Linkers 2.12 Translating and Starting a Program Hands-on Lecture 05 Dynamic linking, compilation process and preprocessing 2.12 Translating and Starting a Program Introducing memory Hands-on Lecture 06 Basic C programming A.4 Loading and A.5 Memory Usage To be covered as part of Chapter 2 11

12 Von Neumann Computer Programs and data are stored in the memory Data needs to be loaded from memory to register for compute E.g. lw $R2, 20[$R4]; int a = A[5]; Memory are used to store data and it has address E.g. sw $R3, 40[$R4]; A[10] = 5; In general, the computer can only perform one instruction at a time. 12

13 Data, Files/Folders, and Memory Source code, executable, object are all files Files: Hello.c, sum_full.c, sum Folder:.,.., /home/yanyh, etc Compiler, OS kernel, etc are all stored as files gcc, vmlinuz generic Information about files/folders and data are also files Metadata Files need to be loaded to memory in order to be processed./hello: load the file hello and execute it ls: load the file ls, which is the command ls, and execute it 13

14 Levels of Program Code to Multiple Target Architectures Java C/C++ Python Javac and JIT gcc/clang python MIPS X86/X86_64 ARM 14

15 X86_64 Assembly Example Disassembly a machine binary code using objdump 15

16 Code in Assembly Language Assembly language either is written by a programmer or is the output of a compiler. 16

17 High-Level Program, Assembly Code and Binary 17

18 Hand-On, sum x86_ A method in assembly.globl: a global symbol.type.cfi_startproc.cfi_endproc ret: return for loop check i<n, if true, else goto end; loop body i++ end 18

19 Sum, MIPS Mainly different instructions 19

20 Sum, x86_64 Number of instructions per loop iteration Count it 20

21 When to Use Assembly Language Advantage: Speed, size and predictable No compiler middle-man Fit for mission-critical, embedded domain, e.g. space shuttle or car control Hybrid approach Non-critical part in high-level language Critical part in assembly language Explore special instructions E.g. those special-purpose instructions that can do more than one thing 21

22 Drawbacks of Assembly Language Assembly language has many (and more) disadvantages that strongly argue against its wide-spread use. Machine-specific code, i.e. assembly code are not portable Rewrite for new or different architectures Harder than high level language to write large code or software Harder to keep a high-level software structure Harder to read and debug Most compilers are good enough to convince that you do not need to write assembly code for general-purpose applications Except embedded or IoT domain 22

23 Assembler Translates file of assembly language statements into a file of binary machine instructions and binary data. Two main steps: Find memory address for symbols (e.g. functions). Translate each assembly statement by combining the numeric equivalents of opcodes, register specifiers, and labels into a legal instruction Binary Produce object files 23

24 Object File ELF Format: #include <stdio.h> int a[10]={0,1,2,3,4,5,6,7,8,9}; int b[10]; int main(int argc, char* argv[]){ int i; static int k = 3; } for(i = 0; i < 10; i++) { printf("%d\n",a[i]); b[i] = k*a[i]; } 24

25 Contents of Object File for the Sample C program Offset Contents Comment Header section number of bytes of Machine code section 4 44 number of bytes of initialized data section 8 40 number of bytes of Uninitialized data section (array b[]) (not part of this object module) number of bytes of Symbol table section number of bytes of Relocation information section Machine code section (124 bytes) 20 X code for the top of the for loop (36 bytes) 56 X code for call to printf() (22 bytes) 68 X code for the assignment statement (10 bytes) 88 X code for the bottom of the for loop (4 bytes) 92 X code for exiting main() (52 bytes) Initialized data section (44 bytes) beginning of array a[] : end of array a[] (40 bytes) variable k (4 bytes) Symbol table section (60 bytes) 188 X array a[] : offset 0 in Initialized data section (12 bytes) 200 X variable k : offset 40 in Initialized data section (10 bytes) 210 X array b[] : offset 0 in Uninitialized data section (12 bytes) 222 X main : offset 0 in Machine code section (12 bytes) 234 X printf : external, used at offset 56 of Machine code section (14 bytes) Relocation information section (44 bytes) 248 X relocation information 25

26 Some Terms Object file vs Executable Object file is the file for binary format of machine instructions, not linked with others, nor positioned (in memory) for execution Executable is binary format of object files that are linked and positioned ready for execution. Symbol Names, e.g. global function name, variable name Library Archive or package of multiple object files 26

27 Inspect an ELF Object File or Executable Executable and Linkable Format (ELF) readelf and objdump command in Linux to inspect object/executable file or disassembly Only objdump can do disassembly nm command to display symbol information Try sum_full.o and sum example sum_full.o is an object file sum is an executable 27

28 Linking Linker (ld command) searches a collection of object files and program libraries to find nonlocal routines used in a program, combines them into a single executable file, and resolves references between routines in different files. 28

29 Role of the Linker ELF Header Section 1 Data Section 2 Data Section n Data Section-Header Table Linkable File ELF Header Section 1 Data Section 2 Data Section n Data Section-Header Table ELF Header Program-Header Table Segment 1 Data Segment 2 Data Segment n Data Executable File Linkable File 29

30 Compile Multiple Files and Link to One Executable Split the sum_full.c into two files sum.c that only contains the definition of sum method Also the #define REAL float line on top Remove the sum definition from sum_full.c, but still keep sum method declaration (referred too as function signature) Compile both together and generate sum executable Compile in one step: gcc sum_full.c sum.c -o sum The command compiles each *.c file one by one into object files and then link the two object files into one executable Compile in multiple steps: compile each.c file one by one and link together 30

31 Compile in One Step 31

32 Compile in Multiple Steps 32

33 Try readelf 33

34 Try objdump for both object file and executable 34

35 objdump -D to disassembly: convert binary object code back to symbolic assembly code 35

36 nm: list symbols from object files T: define a symbol U: undefined symbol Linker to link Address are relative 36

37 End of Lecture 04 37

38 Appendix A and Chapter 2.12: Compiler, Assembler, Linker and Program Execution Lecture 04 A.1 Introduction A.2 Assemblers A.3 Linkers 2.12 Translating and Starting a Program Hands-on Lecture 05 Dynamic linking, compilation process and preprocessing 2.12 Translating and Starting a Program Introducing memory Hands-on Lecture 06 Basic C programming A.4 Loading and A.5 Memory Usage To be covered as part of Chapter 2 38

39 Review of Lecture 04 39

40 Review: Von Neumann Computer, How Array References are Translated to Load/Store Instructions Programs and data are stored in the memory Data needs to be loaded from memory to register for compute E.g. lw $R2, 20[$R4]; int a = A[5]; Right (of = ) value Memory are used to store data and it has address E.g. sw $R3, 40[$R4]; A[10] = 5; Left (of = ) value In general, the computer can only perform one instruction at a time. 40

41 Review: High-Level Program, Assembly Code and Binary, Loop is translated to use branch instruction Function call is translated to use jalinstruction bne: branch if not equal ble: branch if less than or equal jal: jump and link 41

42 Review: Pros and Cons of Assembly Code +: Speed and predictable Used for mission-critical application -: Machine and architecture-dependent -: Hard to program large software and debug Most compilers are good enough to convince that you do not need to write assembly code for general-purpose applications 42

43 Review: Compiling, Assembly and Disassembly Linking, e.g. gcc Compiling, e.g. gcc -s Assembly, e.g. gcc -s disassembly, e.g. objdump -d 43

44 Linking Linker (ld command) searches a collection of object files and program libraries to find nonlocal routines used in a program, combines them into a single executable file, and resolves references between routines in different files. 44

45 Review: Assembler and Linking Assembler: Convert symbolic machine instructions to binary instructions Linker: Link symbols (one is to define and the others are to use) from multiple object files Combine object files and create executables Know how to compile multiple source files into objects and then link them into one executable Static linking, hand-on exercises Dynamic linking Familiar with gcc, readelf, objdump, nm, ldd commands and their usage and purpose 45

46 Compile Multiple Files and Link to One Executable Split the sum_full.c into two files sum.c that only contains the definition of sum method Also the #define REAL float line on top Remove the sum definition from sum_full.c, but still keep sum method declaration (referred too as function signature) Compile both together and generate sum executable Compile in one step: gcc sum_full.c sum.c -o sum The command compiles each *.c file one by one into object files and then link the two object files into one executable Compile in multiple steps: compile each.c file one by one and link together 46

47 End of Review of Lecture 04 47

48 Appendix A and Chapter 2.12: Compiler, Assembler, Linker and Program Execution Lecture 04 A.1 Introduction A.2 Assemblers A.3 Linkers 2.12 Translating and Starting a Program Hands-on Lecture 05 Dynamic linking, compilation process and preprocessing 2.12 Translating and Starting a Program Introducing memory Hands-on Lecture 06 Basic C programming A.4 Loading and A.5 Memory Usage To be covered as part of Chapter 2 48

49 Static Linking If multiple program want to use read_timer functions They all include the full definition in their source code Duplicate: If the function changes, we need to change each file Separate reader_timer in a new file, compile and statically linked with other object files to create executables Duplicate the same object in multiple executables. Dynamic linking at the runtime Create a dynamic library that provides reader_timer implementation Tell ld to link the library at the runtime Runtime load and link them on the fly and execute 49

50 Static Library vs Shared (Dynamic) Library Static library needs to be duplicated in every executable Bigger code size, better optimized Shared library are loaded on the fly during the execution Smaller code size, performance hits of loading shared memory Combine both 50

51 Hands-On for dynamic linking Sum example for static and dynamic linking: from sum.c and sum_full.c created in the last exercise, Create a new file read_timer.c that includes the read_timer and read_timer_ms definition in the file Leave only the read_timer and read_timer_ms declaration in the sum_full.c They are the interface of the two methods. Compile read_timer.c into a dynamic library The library name is my_read_timer, and the library file is libmy_read_timer.so. You can choose any name. Compile sum.c and sum_full.c and link with lib my_read_timer gcc sum_full.c sum.c -o sum -L. -lmy_read_timer Use ldd command to list dependent libraries 51

52 Build Steps with Dynamic Library Linking error: cannot find reader_timer implementation when linking from sum_full.o Linking error: do not know where to find the libread_timer.so file. -L< >: to tell where to find the library file, in this case, the current folder (.) -l<...>: to tell the library file name, which will be expanded to lib< >.so file 52

53 ldd command to list the dependent libraries 53

54 Compilation Process in C Compilation process: gcc hello.c -o hello Constructing an executable image for an application FOUR stages Command: gcc <options> <source_file.c> Compiler Tool gcc (GNU Compiler) man gcc (on Linux m/c) icc (Intel C compiler) 54

55 4 Stages of Compilation Process Preprocessing gcc -E hello.c -o hello.i hello.c à hello.i Compilation (after preprocessing) gcc -S hello.i -o hello.s Assembling (after compilation) gcc -c hello.s -o hello.o Linking object files gcc hello.o -o hello Output à Executable (a.out) Run à./hello (Loader) 55

56 4 Stages of Compilation Process 1. Preprocessing (Those with # ) Expansion of Header files (#include ) Substitute macros and inline functions (#define ) 2. Compilation Generates assembly language,.s file Verification of functions usage using prototypes Header files: Prototypes declaration 3. Assembling Generates re-locatable object file (contains m/c instructions),.o file nm app.o T main U puts nm or objdump tool used to view object files 56

57 4 Stages of Compilation Process (contd..) 4. Linking Generates executable file (nm tool used to view exe file) Binds appropriate libraries Static Linking Dynamic Linking (default) Loading and Execution (of an executable file) Evaluate size of code and data segment Allocates address space in the user mode and transfers them into memory Load dependent libraries needed by program and links them Invokes Process Manager à Program registration 57

58 Compiling a C Program gcc <options> program_name.c Options: Four stages into one Wall: Shows all warnings -o output_file_name: By default a.out executable file is created when we compile our program with gcc. Instead, we can specify the output file name using "-o" option. -g: Include debugging information in the binary. man gcc 58

59 Linking Multiple files to make executable file Two programs, prog1.c and prog2.c for one single task To make single executable file using following instructions First, compile these two files with option "-c" gcc -c prog1.c gcc -c prog2.c -c: Tells gcc to compile and assemble the code, but not link. We get two files as output, prog1.o and prog2.o Then, we can link these object files into single executable file using below instruction. gcc -o prog prog1.o prog2.o Now, the output is prog executable file. We can run our program using./prog 59

60 Linking with other libraries Normally, compiler will read/link libraries from /usr/lib directory to our program during compilation process. Library are precompiled object files To link our programs with libraries like pthreads and realtime libraries (rt library). gcc <options> program_name.c -lpthread -lrt -lpthread: Link with pthread library à libpthread.so file -lrt: Link with rt library à librt.so file Option here is "-l<library>" Another option "-L<dir>" used to tell gcc compiler search for library file in given <dir> directory. 60

61 Preprocessing Things with # #include <stdio.h> #define REAL float Others Processes the C source files BEFORE handing it to compiler. `Pre`-process gcc E cpp 61

62 File Inclusion Recall : #include <filename> #include <foo.h> System directories #include foo.h Current directories gcc I/usr/include to specify where to search those header files gcc I/usr/include sum_full.c o sum Preprocessing replaces the line #include <foo.h> with the content of the file foo.h 62

63 Macros Define and replaced by preprocessing Every occurrence of REAL will be replaced with float before compilation. 63

64 About printf in C printf( format string,vars); Format string? This year is %d\n Your score is %d\n Conversion by % %d : int %f : float, double %c : char %s : char *, string %e : float, double in scientific form 64

65 Tools and Steps for Program Execution User-created files Makefile C/C++ C/C++ Source Source and and Header Header Files Files Assembly Assembly Source Source Files Files Linker Script File Make Utility preprocessor compiler assembler Archive Utility Library Library Files Files Object Object Files Files Linker and Locator Shared Object File Linkable Image File Executable Image File Link Map File 65

66 Loading and Introducing Memory To load a file from disk into memory Loading: To execute a file, e.g../ is to specify the path of sum file To execute any linux command, e.g. ls, cd, etc. Right-click an app icon to execute the app System search your file from folders specified in PATH env echo $PATH The runtime instance of an executable is called a process It occupies memory, It uses resources (files, sockets, driver, etc). It executes its threads (machine instructions). See the processes of the system using ps command, Windows task manager, and Mac OS X Activity Monitor 66

67 Computer Memory Addressing Memory is byte-addressed linearly A process can access 2 16 (64K), 2 32 (4G) or 2 64 bytes of memory depending on whether your computer is 16, 32 or 64 bit addressing 67

68 Data, Address and Pointer 68

69 End of Lecture 05 69