Implementing Subprograms

1. Implementing Subprograms Acknowledgement: Mehdi Emadi Dr. Abhik Roychoudhury And …

2. Objectives • To understand the implementation of subprogram calls • basic procedure of subprogram linkage • activation records • static and dynamic chaining

3. General issues • When calling: • Save state of calling procedure • Determine parameter-passing mechanism for each parameter and arrange access appropriately • Allocate storage for local variables • Initialize local variable values • Arrange transfer of control to procedure • Arrange return of control to calling procedure • Arrange access to visible non-local variables

4. General issues (cont.) • When Returning: • Move final parameter values back to calling arguments (when pass by result) • Deallocate storage used for local variables • Undo access to visible non-local variables • Return control to calling procedure

5. Activation Records • Programs consist of Code, Data, and book-keeping • Noncode part is organized in an Activation Record ( Code does not change by each call) • Activation record instances can include • Parameter definitions • Local variables • Functional value • Return address • Other data to be discussed

6. Fortran 77 • A simple situation: • Actual code is fixed at compile time • Local variables/data are also of fixed size • Subprograms cannot be recursive • A simple solution: • Local variables statically allocated (static activation records) • All referencing of non-local variables is through COMMON: a globally accessible static memory space

7. Fortran 77 • Procedure call: • Save the execution status of current procedure • Carry out parameter-passing • Pass return address to callee • Transfer control to callee

8. Fortran 77 (cont.) • Procedure return: • If using pass-by-value-result, final values of parameters are moved to arguments • If subprogram is a function, final value is moved back • Execution status of the caller is restored • Control is transferred back to caller

9. Common Local variables Main Code Return address Parameters Procedure Local variables A Code Fortran77 Activation Records • No more than one can exist for a procedure • AR Instances can be statically allocated • Either separate or interleave code and data

10. Algol-like Languages • New complexities over Fortran 77 • Parameters can be passed in different ways (value or reference) • Subprogram variables may be dynamically allocated • Recursion legal: multiple activation records may exist simultaneously • Scope rules (usually static) determine visibility of non-local variables

11. Algol-like Languages • Activation records • Format and size is known at compile time since all local variables and their size is statically fixed • Instances of activation records are dynamically generated at run-time

12. Contents of Activation Records • Return address • Pointer to code segment of caller and offset address of instruction • Static link: • A pointer to the activation record of static parent • Used to support non-local variable access • Dynamic link: • A pointer to activation record of caller • Used in static scoped languages to support procedure return (indicates how much of stack to remove) • Used in dynamic scoped languages for non-local variable access

13. Activation Records (continued) • Parameters: • Either values or addresses provided by caller • Local variables: • Scalars stored directly within activation record • Non-scalars (structures) stored elsewhere • References to locals can be via offset from beginning of record: known at compile time • Functional Value (if a function) • Place to put the returned value of the function • (In some languages the function name can be treated like a local variable)

14. sum Local } Parameter tot Dynamic Link Offset of sum Static Link Return Address Start of record Activation Record Example procedure sub (var tot: real); var sum : real; begin : end; • A new activation record is created each time sub is called • Run-time Stack holds activation records • Stack discipline is appropriate for conventional procedures • But not for co-routines

15. Functional: ? Functional: ? Functional: ? Functional: ? Functional: ? Functional: ? Parameter n=2 Parameter n=3 Parameter n=1 Parameter n=3 Parameter n=3 Parameter n=2 Dynamic Link Dynamic Link Dynamic Link Dynamic Link Dynamic Link Dynamic Link Static Link Static Link Static Link Static Link Static Link Static Link Return (fact) Return (main) Return (fact) Return (fact) Return (main) Return (main) Local value = ? Local value = ? Local value = ? Factorial Example

16. Functional: 1 Functional: 6 Functional: 2 Functional: ? Functional: ? Functional: ? Parameter n=2 Parameter n=3 Parameter n=1 Parameter n=3 Parameter n=3 Parameter n=2 Dynamic Link Dynamic Link Dynamic Link Dynamic Link Dynamic Link Dynamic Link Static Link Static Link Static Link Static Link Static Link Static Link Return (fact) Return (main) Return (fact) Return (fact) Return (main) Return (main) Local value = ? Local value = ? Local value = ? Factorial Example

17. Non-local references • Reference to a non-local variable requires a two step process: • Find the activation record instance where the variable was allocated • Use local offset to access the variable within the record • Finding the ARI (under static scoping) is not straightforward

18. Non-local references • program Prog; int i=0; procedure A; int i=1; B(); end; • procedure B; int i=2; • C(); end; procedure C; i=3; end; A(); • end; Show the set of activation records in existence when the statement ‘i=3’ is encountered. Which declaration of ‘i’ will be updated? Following the dynamic chain will find the wrong ‘i’!

19. Static Chains • A link from each procedure’s activation record to its parent’s activation record with respect to static scope • The static chain is different from the dynamic chain: • The static parent of C is Prog. • The dynamic parent of C is B.

20. Static Chains (continued) • No need to search for variable: number of static links is known at compile time (why?) • Static Depth: nesting level of scope • Chain Offset (Nesting Depth): difference between static depth of reference environment and static depth of declaration environment • References represented as (chain_offset, local_offset) • Locals can be handled with chain_offset=0

21. Maintaining Static Chains • At subprogram call: • Correct parent known at compile time • Must find most recent activation record of parent • Avoid search of dynamic chain by using existing static chain to find parent (treat procedure reference like variable • At return: just remove record from stack

22. Cost of Static Chains • Static chains allow references to non-local variables to be resolved, but at a cost: • References to non-local variables “above” parent are slow • Time to resolve a reference is difficult to predict, and sensitive to source code changes

23. Summary: MacLennan terminology • State of a procedure: • A fixed program part • A variable activation part • The ep (environment part), the context to be used for this activation of the procedure, local and nonlocal context • The ip (instruction part), the current instruction being (or to be) executed in this activation of the procedure

24. Review of scope rules • Q and T are declarations of procedures within P, so scope of names Q and T is same as scope of declaration a. • R and S are declarations of procedures in Q. • U is a declaration of a procedure in T. • Storage managed by adding activation records, when procedure invoked, on a stack.

25. Activation record stack

26. Scope rules • Scope rules: The scope of a variable are the set of statements where the variable may be accessed (i.e., named) in a program. • Static scope: Scope is dependent on the syntax of the program. • Dynamic scope: Scope is determined by the execution of the program.

27. Scope rules • Static nested scope: A variable is accessible in the procedure it is declared in, and all procedures internal to that procedure, except a new declaration of that variable name in an internal procedure will eliminate the new variable's scope from the scope of the outer variable. • A variable declared in a procedure is local in that procedure; otherwise it is global.

28. Activation record stack

29. Activation record stack • Problem is: How to manage this execution stack? • Two pointers perform this function: • 1. Dynamic link pointer points to activation record that called (invoked) the new activation record. It is used for returning from the procedure to the calling procedure. • 2. Static link pointer points to the activation record that is global to the current activation record (i.e., points to theactivation record of the procedure containing the declaration of this procedure).

30. Program a(…); • var N: integer; • procedure b(sum: real); • var i: integer; • avg: real; • Data: array[1..10] of real; • procedure c(val: real); • begin • … • writeln(Data[i]); • end {c}; • begin • … • end {b}; • begin … • end {a}

31. Accessing a variable • At run-time skip down the static chain (how many skips? Static distance) to get to the activation record of the variable. • The address of the variable is obtained by adding the fixed offset of the variable to the address obtained in step 1.

32. Static distance(static nesting level:snl) and offset of the variable are computed by the compiler at compile time. • Symbol table • ----------------------------------------- • Name Type Snl Offset • ----------------------------------------- • N int 1 1 • Sum real 2 1 • i int 2 2 • Avg real 2 3 • Data real array 2 4 • Val real 3 1

33. Implementation of subprogram storage: more examples • Each subprogram has a block of storage containing such information, called an activation record. • Consider the following C subprogram: • float FN( float X, int Y) • const initval=2; • #define finalval 10 • float M(10); int N; • N = initval; • if(N<finalval){ ... } • return (20 * X + M(N)); } • Information about procedureFN is contained in its activation record.

34. Activation records

35. Dynamic nature of activation records • Each invocation of FN causes a new activation record to be created. • Thus the static code generated by the compiler for FN will be associated with a new activation record, each time FN is called. • A stack structure is used for activation record storage.

36. Dynamic nature of activation records

37. Activation record structure

38. Example of act. record stack Declarations Var A in P B in Q C in R

39. Activation record example 1 • Ex 1. In R: C := B+A;  C local, A,B global • For each variable, get pointer to proper activation record. • Assume AR is current act.record pointer (R). • 1. B is one level back: • Follow AR.SL to get AR containing B. • Get R-value of B from fixed offset L-value • 2. A is two levels back: • Follow (AR.SL).SL to get activation record containing A. • Add R-value of A from fixed offset L-value • 3. C is local. AR points to correct act record. • Store sum of B+A into L-value of C

40. Activation record example 2 • Example 2. Execution in procedure Q: A := B •  B is local, A global • Assume AR is current activation record pointer (Q) • 1. B is now local. AR points to activation record • Get R-value from local activation record

41. Activation record example 2 • 2. A is now one level back • AR.SL is activation record of P • Store R-value of B into L-value of A • Compiler knows static structure, so it can generate the number of static link chains it has to access in order to access the correct activation record containing the data object. This is a compile time, not a runtime calculation.

42. Setting the static link • …

43. Displays: an alternative Static links are in a separate array • One element per static depth • Exactly two steps per reference: (display_offset, local_offset) • When a subprogram at a given static depth is called, save and replace the link at that depth Efficiency compared to Static Chains: • Deeply nested references more efficient • Maintenance of subprogram calls less efficient Yet most programs are not deeply nested (three in practice)

44. Sub5 Display Sub4 Static Depth 3 Display Sub3 Static Depth 2 Static Depth 3 Sub2 Sub2 Static Depth 1 Static Depth 2 Sub1 Sub1 Static Depth 0 Static Depth 1 MAIN MAIN Static Depth 0 Display Example

45. Dynamic Scoping • Deep Access (NOT same as deep binding): • References to non-local variables are be resolved by following dynamic links • Easy to implement • Search down call chain may be slow • Activation records must store variable names

46. Shallow Access(shallow binding method) • Like FORTRAN, each procedure is statically allocated one copy of its activation record. • Containing the most recent activation of that procedure. • Variable access is efficient. • In the case of recursive cal, push the activation record on the stack. • Activation records are allocated statically, their size is fixed (=> cannot be used for languages with dynamic arrays …)

47. Implementing Subprogram as Parameters Two major problems: • Static type checking Full specification for formal parameter shall be present: type of the function, the number, order, and type of each parameter. • Nonlocal references (free variables) what is the nonlocal environment used when a function is invoked?

48. Program Main; var X: integer; Procedure Q(var I:integer;function R(J:integer):integer); var X: integer; begin X:=4; I:=R(I); end; Procedure P; var I: integer; function FN(K:integer):integer; begin X:=X+K; FN:=I+K; end; begin I:=2; Q(X,FN); end; begin X:=7; P; End.

49. What should be the nonlocal environment used when FN is invoked within Q (Formal parameter R in Q)? • The nonlocal environment is the same as the one that would be used if the call R(I,X) were simply replaced by the call FN(I,X) in subprogram Q. • The nonlocal environment is the same as the one that would be used if the subprogram FN were invoked at the point where it appears as an actual parameter in the parameter list (in calling Q(X,FN)).

50. Procedural parameters are represented by closures. • The ip field: entry address of the actual parameter • The ep field: a pointer to the environment of definition of the actual procedure