Please submit your answers
In this homework, we are going to generate Java Virtual Machine codes by processing the abstract syntax tree constructed from the parsing process.
We are going to support all features of C-- language for integer data type. That is, for the purpose of this homework, float data type is not supported.
Please refer to the following documents about Java Virtual Machine codes:
- Instructions Set (limited to instructions that are related to integers only)
- Compiling Examples
The template for codes generation program is similar to that given for hw7. Also, click here for an executable for hw8.
One can obtain the Java Virtual Machine codes for the test data as follows: Take the example of data04. We can present it as a Java program as follows:
public class data04 {
public static
int
Factorial(int v)
{
int limit = 7;
if (v < 0 || v > limit) return -1;
{
int c = 0, fact = 1;
/* Loop invariant: fact == c! */
while (c < v) {
c = c+1; fact = fact*c;
}
return fact;
}
}
}
The commands are:
javac data04.java
javap -c data04 > data04.java.out
The command javap will produce the Java Virtual Machine Codes in readable format.
You can also compare the output data04.java.out by javap with the output of hw8 by running
hw8 < data04 > data04.out
NOTE: The codes produced by hw8 may differ from that by the Java compiler. But the codes are close. Sometimes, hw8 has a few more unnecessary branching statements which can be cleaned up by a simple pass over the generated codes.
Programming hints:
Previously, traverse_ast does not return any value. You may want to extend the function traverse_ast to return the number of lines of codes generated. In order that traverse_ast can generate the right line numbers for the Java Virtual Machine codes, the function traverse_ast takes one extra parameter next_line_number which is the line number of the first instruction generated in the processing of the subtree rooted at node.
The (unsophisticated) codes for AST_ADD in traverse_ast can be very simple as follows:
case AST_ADD:
size = traverse_ast( node->left , next_line_number );
size += traverse_ast( node->right , next_line_number + size );
printf("%4d: iadd\n", next_line_number+size);
size++;
return size;
You need to modify your symtbl module so that each ID symbol has an entry with a field local_var_index. Note that we do not make use of any temp variable for codes generation in this homework. We need a global variable local_var_num, which use is similar to that of temp_num, to keep track of the largest variable number that is currently in use. The function symtbl_dump_entry needs to be revised to print also the local_var_index associated with an identifier.
Also, modify hw5parser.y so that float data type is no longer supported.
type_spec : INT
{ data_type = TYPE_INT; }
| FLOAT
{ yyerror("float unsupported\n"); exit(-1); }
;
Java uses special comparison instructions when one of the operands is 0. They are ifeq, ifne, iflt, ifge, ifgt, ifle. For your first attempt, you can just use the general comparison instructions if_icmpeq, if_icmpne, if_icmplt, if_icmpge, if_icmpgt, if_icmple so that you can avoid testing if one of the operands is 0.
NOTE: the codes that your program generates do not have to match the quality of the codes generated by the Java (javac) compiler, or my hw8.
NOTE: You can assume that each integer constant in a C-- program is of small value, and can be represented in one byte so that there is no need to use the runtime constant pool. Also, you can assume that the number of local variables is small such that the index into the local variable array can be achieved in one byte.