1. Languages and the Machine Chapter 5 CS221

2. Topics The Compilation Process The Assembly Process Linking and Loading Macros We will skip –Case Study: Extensions to the Instruction Set – The Intel MMX™ and Motorola AltiVec™ SIMD Instructions

3. Compilation Process Assembly to Machine code fairly straightforward, but compilation is not Translate a program written in a high level language into a functionally equivalent program in assembly language Consider a simple high-level language assignment statement: –Foo = Bar + Zot + 15; Steps involved in compiling this statement into assembly code: –Lexical analysis: separate into tokens, Foo, =, +, etc. –Syntactic Analysis / Parsing : Determine that we are performing an assignment, VAR = EXPRESSION –Semantic Analysis : Determine that Foo, Bar, Zot are names, 4 is an integer –Code Generation : Determine the proper assembly code to perform the action ld [Bar], %r0, %r1 ld [Zot], %r0, %r2 addcc %r1, %r2, %r1 addcc %r1, 15, %r2 st %r2, %r0, [Foo]

4. Compiler Issues Each compiler specific to a particular ISA –E.g., an int on one machine may be 32 bits, on another may be 64 bits Cause of error in networking library ported to Alpha Int issue not a problem in Java; JVM specifies 32 bits –E.g., in previous example, if the ISA allowed operands of addcc to be memory addresses, we could have done addcc [Bar], [Zot], %r1 addcc %r1, 15, [Foo] –Hopefully the compiler generates efficient code but optimization is a tough issue! Cross compiler: one that generates code for a different ISA (example, CodeWarrior)

5. Mapping Variables to Memory Global variables –Accessible from anywhere in the program, given a fixed address –E.g., global variable X at memory address 400 Local variables –Also called automatic variables –Defined inside a function or method, e.g. void foo() { int a,b; … } –These variables created when foo is invoked, destroyed when foo exits –These variables are created by pushing them on the stack when the function is invoked, and are popped off when the function exits

6. Local Variables and the Stack Recall that the stack typically grows downward in memory Here we start with 1234 stored on the top of the stack Mem 048…048… SP = 8 Push FFFF Mem 048…048… SP = 4 FFFF 1234

7. Local Variables and the Stack In our case, local variables are “pushed” on the stack upon entering the function –void foo() { int a; } Copy SP into Frame Pointer FP (also called the Base Pointer, or BP) Mem before Foo 048…048… SP = 8 Mem in Foo 048…048… SP = 4 Var a 1234 FP = 8

8. Accessing Stack Variables These variables are referenced as offsets from the frame pointer, called based addressing To access a : [%fp – 4] Mem in Foo 048…048… SP = 4 Var a 1234 FP = 8 Why not use [%sp] ? Consider pushing lots of stuff on the stack… Or data structures

9. C to ASM Example on x86 #include int c; int main() { int a,b; a=3; b=4; c=a*b; } pushl%ebp movl%esp, %ebp subl$8, %esp movl $3, -4(%ebp) movl $4, -8(%ebp) movl -4(%ebp),%eax imul1 -8(%ebp),%eax movl%eax, c ….comm c,4,4

10. Arrays in Memory Arrays may be allocated on the stack or allocated off the heap, a pool of memory where portions may be dynamically allocated. Access elements of an array a bit different than regular variables. int A[10];Array of 10 integers Mem allocated for A 0 4 8 … 40 A (Base) = 4A[0] A[1] … A[9] ElementAddr = A + (Index*Size) e.g. A[2] is at 4 + (2*4) = 12

11. If-Statements Conditional statements map to a comparison and a branch instruction C –if (x==y) statement1; else statement2; Assembly (assume X in r1, Y in r2) –subcc %r1, %r2 ! Zero flag set if res=0 –bne Statement2 ! Branch if zero flag is not set –! Statement1 code –ba StatementNext! Branch always Statement2: ! Statement2 code StatementNext:

12. Loops While, Do-While, For loops implemented using the same conditional check and branch as the if-then statement –The branch returns back to previous code instead of jumping forward over code

13. Production Level Assemblers Allow programmer to specify location of data and code Provide mnemonics for all instructions and addressing modes Permit the use of symbolic labels to represent addresses and constants Provide a means to specify the starting address of the program Include a way to share variables between different assembled programs Support macros

14. Assembly Example

15. Assembled Code

16. Two Pass Assemblers Most assemblers are “two-pass” –First pass Determine addresses of all data and instructions Perform any assembly-time arithmetic Put definitions and constants into the symbol table –Second pass Generate machine code Insert actual addresses and values of symbols which are known from the symbol table –Two passes useful for forward references, i.e. referencing later on in the program

17. Forward Reference

18. Symbol Table Generated during the first pass Maps identifiers to values, table filled in as values are encountered and the program is parsed from top to bottom.org 2048; Says assemble code starting at 2048 const.equ value ; Defines const equal to value

20. Assembled Program

21. Final Tasks of the Assembler Linking and Loading We need the following additional info –Module name and size –Address start symbol –Information about global and external symbols –Information about any library routines –Values of constants –Relocation information

22. Location of Programs in Memory We have been using.org to specify a fixed start location Typically we will want programs capable of running in arbitrary locations –If we are concatenating together different modules, the addresses for identifiers in the different modules must be relocated Linker : software that combines separately assembled modules Loader : software that loads another program into memory and may modify addresses if the program is loaded in a location different from the origin –Must also set appropriate registers, e.g. %SP

23. Linking:.global and.extern A.global is used in the module that a symbols is defined and.extern is used in every other module that refers to it

24. Linking and Loading Symbol tables for previous example Symbols whose address might change market relocatable (not all addresses! Some may be fixed)

25. DLL’s Windows uses Dynamic Link Libraries, or DLL’s Linking a common routine in many programs results in duplicate code from that common routine in each program In a DLL, commonly used routines (e.g. memory management, graphics) present in only one place, the DLL –Smaller program sizes, each program does not need to have its own copy –All programs share the exact same code while executing –Don’t need recompiling or relinking Disadvantages –Deletion of a shared DLL by mistake can cause problems –Versions must be the same –DLL code file can live in many places in Windows –“DLL Hell”

26. Macros An assembly macro looks kind of like defining a subroutine For example, there say that there is no PUSH instruction to push data on the stack. We can make a macro for push:

27. Macro Expansion Given the previous macro, we could now write the following code: push %r15! Push r15 on the stack push %r20! Push r20 on the stack Upon assembly, these macros are expanded to generate the following actual code: addcc %r14, -4, %r14 st %r15, %r14 addcc %r14, -4, %r14 st %r20, %r14

28. Macros vs. Subroutines Later we will see how to write actual subroutines we can call –Only one copy of the shared code in a subroutine Tradeoffs –Subroutines Takes up less memory since only one copy of the code But slower than macros; subroutines have overhead of invoking and returning –Macros Take up more space than subroutine call due to macro expansion for each occurrence of the macro Faster than subroutines; no overhead to invoke/return

29. Skipping for now Discussion on Pentium MMX We may return to this later if time permits