CS39003-Assignment 5 : Machine-independent Code Generator for tinyC Solved

1         Preamble – tinyC
In this assignment you are required to write the semantic actions in Bison to translate a tinyC program into an array of 3-address quad’s, a supporting symbol table, and other auxiliary data structures. The translation should be machineindependent, yet it has to carry enough information so that you can later target it to a specific architecture (viz. x86-32 / x86-64).

2         Scope of Machine-Independent Translation
You may consider the following from the different phases to write actions for translation.

2.1        Expression Phase
You should support all arithmetic, shift, relational, bit, logical (boolean), and assignment expressions excluding:

1.   sizeof operator

2.   Comma (,) operator

3.   Compound assignment operators:

*= /= %= += -= <<= = &= ^= |=

Only simple assignment operator (=) needs to be supported.

4.   Structure component expression

2.2        Declarations Phase
Support for declarations should be provided as follows:

1.   Simple variable, pointer, array, and function declarations should be supported. For example, the following would be translated:

float d = 2.3; int i, w[10]; int a = 4, *p, b; void func(int i, float d); char c;

2.   Consider only void, char, int, and float type-specifiers. As specified in C, char and int are to be taken as signed.

 For computation of offset and storage mapping of variables, assume the following sizes[1] (in bytes) of types:

Type
Size
Remarks
void
undefined
 
char
1
 
int
4
 
float
8
 
void *
4
All pointers have same size
It may also help to support an implicit bool (boolean) type with constants 1 (TRUE) and 0 (FALSE). This type may be inferred for a logical expression or for an int expression in logical context. Note that the users cannot define, load, or store variables of bool type explicitly, hence it is not storable and does not have a size.

3.   Initialization of arrays may not be considered.

4.   storage-class-specifier, enum-specifier, type-qualifier, and function-specifier may not be considered.

5.   Function declaration with only parameter type list may not be considered.

Thus,

void func(int i, float d); should be supported while

void func(int, float); may not be supported.

2.3        Statement Phase
All statements excluding the following should be supported.

1.   Declaration within for.

2.   All Labelled statements (labeled-statement).

3.   switch in selection-statement.

4.   All Jump statements (jump-statement) except return.

2.4         External Definitions Phase

You should support function definitions, and skip external declarations.

3         The 3-Address Code
Use the 3-Address Code specification as discussed in the class. For easy reference the same is reproduced here. Every 3-Address Code:

•   Uses only up to 3 addresses.

•   Is represented by a quad comprising – opcode, argument 1, argument 2, and result; where argument 2 is optional.

3.1        Address Types
•   Name: Source program names appear as addresses in 3-Address Codes.

•   Constant: Many different types and their (implicit) conversions are allowed as deemed addresses.

•   Compiler-Generated Temporary: Create a distinct name each time a temporary is needed – good for optimization.

3.2        Instruction Types
For Addresses x, y, z, and Label L, the following instruction types should be supported:

•   Binary Assignment Instruction: For a binary op (including arithmetic, shift, relational, bit, or logical operators):

x = y op z

•   Unary Assignment Instruction: For a unary operator op (including unary minus or plus, logical negation, bit, and conversion operators):

x = op y

•   Copy Assignment Instruction:

x = y

•   Unconditional Jump: goto L

•   Conditional Jump:

a)   Value-based:

if x goto L ifFalse x goto L

b)   Comparison-based: For a relational operator op (including <, , ==, ! =, ≤, ≥):

if x relop y goto L

•   Function Call: A function call p(x1, x2, ..., xN) having N≥ 0 parameters is coded as (for addresses p, x1, x2, and xN):

param x1 param x2 ... param xN

y = call p, N

Note that N is not redundant as procedure calls can be nested.

•   Return Value: Returning a value and / or assigning it is optional. If there is a return value v it is returned from the procedure p as:

return v

•   Indexed Copy Instructions: It can be used as:

x = y[z] x[z] = y

•   Address and Pointer Assignment Instructions: These can be used as:

x = &y x = *y *x = y

4         Design of the Translator
a)   Lexer & Parser: Use the Flex and Bison specifications (if required you may correct your specifications) you had developed in Assignments 3 and 4 respectively, and write semantic actions for translating the subset of tinyC as specified in Section 2. Note that many grammar rules of your tinyC parser may not have any action or may just have propagate-only actions. Also, some of the lexical tokens may not be used.

b)   Augmentation: Augment the grammar rules with markers and add new grammar rules as needed for the intended semantic actions. Justify your augmentation decisions within comments of the rules.

c)    Attributes: Design the attributes for every grammar symbol (terminal as well as non-terminal). List the attributes against symbols (with brief justification) in comment on the top of your Bison specification file. Highlight the inherited attributes, if any.

d)   Symbol Table: Use symbol tables for user-defined (including arrays and pointers) variables, temporary variables and functions. The symbol table should have at least the following fields:

Name
Type
Initial
Size
Offset
Nested
 
 
Value
 
 
Table
...
...
...
...
...
...
For example, for

float d = 2.3; int i, w[10]; int a = 4, *p, b; void func(int i, float d); char c; the Symbol Tables will look like:

              ST(global)                                     This is the Symbol Table for global symbols

Name
Type
Initial
Size
Offset
Nested
 
 
Value
 
 
Table
d
float
2.3
8
0
null
i
int
null
4
8
null
w
array(10, int)
null
40
12
null
a
int
4
4
52
null
p
ptr(int)
null
4
56
null
b
int
null
4
60
null
func
function
null
0
64
ptr-to-ST(func)
c
char
null
1
64
null
                      ST(func)                    This is the Symbol Table for function func

Name
Type
Initial
Size
Offset
Nested
 
 
Value
 
 
Table
i
int
null
4
0
null
d
float
null
8
4
null
retVal
void
null
0
12
null
The Symbol Tables should support the following methods:

lookup(...)
A method to lookup an id (given its name or lexeme) in the Symbol Table. If the id exists, the entry is returned, otherwise a new entry is created.
gentemp(...)
A static method to generate a new temporary, insert it to the Symbol Table, and return a pointer to the entry.
update(...)
A method to update different fields of an existing entry.
print(...)
A method to print the Symbol Table in a suitable format.
Points to Note:

•   The fields and the methods are indicative. You may change their name, functionality and also add other fields and / or methods that you may need.

•   It should be easy to extend the Symbol Table as further features are supported and more functionality is added.

•   The global symbol table is unique.

•   Every function will have a symbol table of its parameters and automatic variables. This symbol table will be nested in the global symbol table.

•   Symbol definitions within blocks are naturally carried in separate symbol tables. Each such table will be nested in the symbol table of the enclosing scope. This will give rise to an implicit stack of symbol tables (global one being the bottom-most) the while symbols are processed during translation. The search for a symbol starts from the top-most (current) table and goes down the stack up to the global table.

e) Quad Array: The array to store the 3-address quad’s. Index of a quad in the array is the address of the 3-address code. The quad array will have the following fields (having usual meanings):

op
arg 1
arg 2
result
...
...
...
...
Note:

•   arg 1 and / or arg 2 may be a variable (address) or a constant.

•   result is variable (address) only.

•   arg 2 may be null.

For example, if

int i = 10, a[10], v = 5;

...

do i = i - 1; while (a[i] < v);

translates to

100: t1 = i - 1 101: i = t1

102: t2 = i * 4

103: t3 = a[t2]

104: if t3 < v goto 100

the quad’s are represented as:

Index
op
arg 1
arg 2
result
...
...
...
...
...
100
–
i
1
t1
101
=
t1
 
i
102
*
i
4
t2
103
=[ ]
a
t2
t3
104
< 
t3
v
100
The Quad Array should support the following methods:

emit(...)
An overloaded static method to add a (newly generated) quad of the form: result = arg1 op arg2 where op usually is a binary operator. If arg2 is missing, op is unary. If op also is missing, this is a copy instruction.
print(...)
A method to print the quad array in a suitable format.
For example the above state of the array may be printed (with the symbol information) as:

void main()

{

int i = 10; int a[10]; int v = 5; int t1; int t2; int t3;

L100: t1 = i - 1; L101: i = t1;

L102: t2 = i * 4;

L103: t3 = a[t2];

L104: if (t3 < v) goto L100;

}

Point to Note:

• The fields and the methods are indicative. You may change their name, functionality and also add other fields and / or methods that you may need.

f) Global Functions: Following (or similar) global functions and more may be needed to implement the semantic actions:

makelist(i)

A global function to create a new list containing only i, an index into the array of quad’s, and to return a pointer to the newly created list.
merge(p1, p2)

A global function to concatenate two lists pointed to by p1 and p2 and to return a pointer to the concatenated list.
backpatch(p, i)

A global function to insert i as the target label for each of the quad’s on the list pointed to by p.
typecheck(E1, E2)

A global function to check if E1 & E2 have same types (that is, if <type of E1 = <type  of E2). If not, then to check if they have compatible types (that is, one can be converted to the other), to use an appropriate conversion function conv<type of E12<type of E2(E) or conv<type of E22<type of E1(E) and to make the necessary changes in the Symbol Table entries. If not, that is, they are of incompatible types, to throw an exception during translation.
conv<type12<type2(E)

A global function to converta an expression E from its current type type1 to target type type2, to adjust the attributes of E accordingly, and finally to generate additional codes, if needed.
 

aIt is assumed that this function is called from typecheck(E1, E2) and hence the

conversion is possible.

Naturally, these are indicative and should be adopted as needed. For every function used clearly explain the input, the output, the algorithm, and the purpose with possible use at the top of the function.


5         The Assignment
1.   Write a 3-Address Code translator based on the Flex and Bison specifications of tinyC. Assume that the input tinyC file is lexically, syntactically, and semantically correct. Hence no error handling and / or recovery is expected.

2.   Prepare a Makefile to compile and test the project.

3.   Prepare test input files ass5 roll test<number.c to test the semantic actions and generate the translation output in ass5 roll quads<number.out.

4.   Name your files as follows:

File
Naming
Flex Specification
ass5 roll.l
Bison Specification
ass5 roll.y
Data Structures (Class Definitions) &

Global Function Prototypes
ass5 roll translator.h
Data Structures, Function Implementations & Translator main()
ass5 roll translator.cxx
Test Inputs
ass5 roll test<number.c
Test Outputs
ass5 roll quads<number.out

 
[1] Using hard-coded sizes for types does not keep the code machine-independent. Hence you may want to use constants like size of char, size  of int, size  of float, and size  of pointer for sizes that can be defined at the time of machine-dependent targeting.