TDT4205/exercises/06/vslc/src/generator.c

#include <vslc.h>

#define ERRPRT(format, args...)     fprintf(stderr, "[ERROR] "); fprintf(stderr ,format, ##args)

#define ASM(opcode, args...)        puts("\t"#opcode"\t"#args)
#define LABEL(label)                printf("_%s:\n", (char*)label)
#define COMMENT(format, args...)    printf("# "format"\n", ##args)

#define PUSH(param)                 printf("\tpushq\t%s\t\t\t\t# PUSH: %ld\n", #param, ++stack_depth)
#define POP(param)                  printf("\tpopq\t%s\t\t\t\t# POP:  %ld\n", #param, --stack_depth)

#define NO_REG_RECORD           6
//#define NO_CALLE_SAVED_REG      10

// Keep track of sequence of stack depth, ifs and whiles
static uint64_t 
    stack_depth,
    if_seq,
    while_seq,
    closest_while;

/**Generate table of strings in a rodata section. */
void generate_stringtable ( void );
/**Declare global variables in a bss section */
void generate_global_variables ( void );
/**Generate function entry code
 * @param function symbol table entry of function */
void generate_function ( symbol_t *function );
/**Generate code for a node in the AST, to be called recursively from
 * generate_function
 * @param node root node of current code block */
static void generate_node ( node_t *node );
/**Initializes program (already implemented) */
void generate_main ( symbol_t *first );

#define MIN(a,b) (((a)<(b)) ? (a):(b))

static const char *record[NO_REG_RECORD] = {
    "%rdi", "%rsi", "%rdx", "%rcx", "%r8", "%r9"
};

//static const char *calle_saved_reg[NO_CALLE_SAVED_REG] = {
//    "%rax", "%rcx", "%rdx", "%rdi", "%rsi", "%rsp", "%r8", "%r9", "%r10", "%r11"
//};

// Helper funcs for generating different nodes

/**
 * Generates assembly for printing
 * 
 * @param node print statement node with children to print
 */
static void generate_print(node_t* node);

/**
 * Generate identfier for a variable in memory
 * 
 * @param node identifier we want the addres of
 */
static void generate_var_ident(node_t *node);

/**
 * Main function to calculate and solve the expressions.
 * Based on a stack machine. Result is stored on stack.
 * 
 * @param node root node for expression
 */
static void solve_expressions(node_t *node);

/**
 * Generates a funciton call
 * 
 * @param node root node for function
 */
static void generate_function_call(node_t *node);

/**
 * Generates the return part of a function
 *  
 * @param node node containing the return statement
 */
static void generate_function_return(node_t *node);

/**
 * Used for calculating and evaluating the add/sub/mul/div statements.
 * Turns the statement into an expression, adds the result of rhs to lhs 
 * and stores the value back to the indentifier
 * 
 * @param node      node to the statement
 * @param operator  one of the following +, -, *, /
 */
static void solve_statements(node_t *node, char *operator);


static void generate_if_statement(node_t *node);

static void generate_while_statement(node_t *node);

static void solve_continue_statement();

/**
 * Generate assembly to fetch a variable on stack
 * 
 * @param node node to the variable to be fetched
 * @param dest where to put the value
 */
static void fetch_variable(node_t *node, const char* dest);

/**
 * Same as fetch_variable, but stores it back to memory.
 * 
 * @param node variable to be stored
 * @param src  from where should the data come from
 */
static void writeback_variable(node_t *node, char* src);

// Helper func for fetching all symbols in a table
static uint64_t fetch_symbols(tlhash_t* symbol_table, symbol_t*** symbol_list);

void
generate_program ( void )
{
    generate_stringtable();
    generate_global_variables();

    symbol_t **global_list;
    uint64_t no_globals = fetch_symbols(global_names, &global_list);

    bool main_generated = false;
    uint64_t seq0_index = -1;
    for (uint64_t g = 0; g < no_globals; g++)
    {
        if (global_list[g]->type != SYM_FUNCTION)
            continue;
        
        // If the name of the function is main
        if (!strcmp(global_list[g]->name, "main"))
        {
            generate_main(global_list[g]);
            main_generated = true;
        }

        if (!global_list[g]->seq)
            seq0_index = g;
    }

    // If no main was found, use the first function instead. 
    // That means the function with seq = 0
    if (!main_generated)
        generate_main(global_list[seq0_index]);

    for (uint64_t g = 0; g < no_globals; g++)
    {
        if (global_list[g]->type == SYM_FUNCTION)
            generate_function(global_list[g]);
    }

    free(global_list);
}

void
generate_stringtable ( void )
{
    /* These can be used to emit numbers, strings and a run-time
     * error msg. from main
     */
    puts("# DATA SECTION");
    puts(".section .data");
    puts(".intout:\t.asciz \"\%ld \"");
    puts(".strout:\t.asciz \"\%s \"");
    puts(".errout:\t.asciz \"Wrong number of arguments\"");

    for (uint64_t s = 0; s < stringc; s++) 
    {
        printf(".STR%03ld:\t.asciz %s\n", s, string_list[s]);
    }
    putchar('\n');
}

void
generate_global_variables ( void )
{
    symbol_t **global_list;
    uint64_t no_globals = fetch_symbols(global_names, &global_list);

    puts("# GLOBAL VARIABLES");
    puts(".bss");
    puts(".align 8");

    for (uint64_t g = 0; g < no_globals; g++) {
        if (global_list[g]->type == SYM_GLOBAL_VAR)
            printf(".%s:\n", global_list[g]->name);
    }
    putchar('\n');
    free(global_list);

}

void
generate_function ( symbol_t *function )
{
    // TODO: Generate code for declaring and entering function, then generate its body
    stack_depth = 0;

    printf("# func %s(nparams: %ld)\n", function->name, function->nparms);
    puts(".section .text");
    printf(".global _%s\n", function->name);
    LABEL(function->name);
    PUSH(%rbp);

    ASM(movq, %rsp, %rbp);

    // Push params to stack
    for (int arg = 0; arg < MIN(NO_REG_RECORD,function->nparms); arg++)
    {
        printf("\tpushq\t%s\t\t\t\t# PUSH: %ld\n", 
            record[arg],
            ++stack_depth
        );
    }

    // How many local variables are inside function
    uint64_t no_locals = function->locals->size - function->nparms;
    //stack_depth += no_locals;

    // Make room for the local vars
    while(no_locals--)
        PUSH($0);

    // IF the stack alignment is not 16 bytes, 
    // add one now as all local var also is 0
    if (stack_depth % 2)
        PUSH($0);


    // Now the stack ptr should be 16 byte aligned.

    generate_node(function->node);

    putchar('\n');
}


void
generate_node ( node_t *node)
{
    // TODO: Generate code corresponding to node

    // All statements have the same structure.
    // [0] is the lhs, needs to be identifier, parser ensures this
    // [1] is thr rhs
    switch (node->type)
    {
    case ASSIGNMENT_STATEMENT:
        // First solve the rhs
        solve_expressions(node->children[1]);
        // Then store in lhs
        writeback_variable(node->children[0], "%rax");
        break;

    case ADD_STATEMENT:
        // The following way is the naive way of doing an assignment
        /*
        fetch_variable(node->children[0], "%rax");
        ASM(pushq, %rax);
        solve_expressions(node->children[1]);
        ASM(popq, %r10);
        ASM(popq, %rax);
        ASM(addq, %r10, %rax);
        writeback_variable(node->children[0], "%rax");
        */
        /*  The thing is that add/sub/mul/div assignments 
            have the same structure as expressions.
            We can therefore just say that the assignment is 
            an expression, but remembering to do the writeback afterwards.
        */
        puts("# Add statement");
        solve_statements(node, "+");
        break;

    case SUBTRACT_STATEMENT:
        puts("# Subtract statement");
        solve_statements(node, "-");
        break;

    case MULTIPLY_STATEMENT:
        puts("# Multiply statement");
        solve_statements(node, "*");
        break;

    case DIVIDE_STATEMENT:
        puts("# Divide statement");
        solve_statements(node, "/");
        break;

    case PRINT_STATEMENT:
        puts("# Print statement");
        generate_print(node);
        break;

    case RETURN_STATEMENT:
        puts("# Return statement");
        generate_function_return(node);
        break;



    case IF_STATEMENT:
        generate_if_statement(node);
        break;

    case WHILE_STATEMENT:
        generate_while_statement(node);
        break;
    case NULL_STATEMENT:
        solve_continue_statement();
        break;

    
    case DECLARATION_LIST:
        /* List of blocks we dont need to traverse */
        break;

    default:
        for (int c = 0; c < node->n_children; c++)
            generate_node(node->children[c]);
        break;

    }
}


void
generate_print(node_t* node)
{
    // Push rdi and rsi to stack incase there are data in them
    for (uint64_t p = 0; p < node->n_children; p++)
    {
        node_t *curr_print = node->children[p];

        switch (curr_print->type)
        {
        case EXPRESSION:
            solve_expressions(curr_print);
            ASM(movq, $.intout, %rdi);
            ASM(movq, %rax, %rsi);
            break;

        case STRING_DATA:
            ASM(movq, $.strout, %rdi);
            printf("\tmovq\t$.STR%03ld, %%rsi\n", *(uint64_t*)curr_print->data);
            break;

        case IDENTIFIER_DATA:
            ASM(movq, $.intout, %rdi);
            fetch_variable(curr_print, "%rsi");
            break;
        default:
            break;
        }
        //ASM(movq, $0, %rax);
        COMMENT("printf call");
        ASM(call, printf);

    }
    // Adds a newline
    ASM(movq, $'\n', %rdi);
    //ASM(movq, $0, %rax);
    ASM(call, putchar);
}


// This will put the value of var in node in dest
void
fetch_variable(node_t *node, const char* dest)
{
    printf("\tmovq\t");
    generate_var_ident(node);
    printf(", %s\t\t# Fetched: %s\n", dest, node->entry->name);
}

// This will put the value in dest to the var in node
void
writeback_variable(node_t *node, char* src)
{
    printf("\tmovq\t%s,", src);
    generate_var_ident(node);
    printf("\t\t# Writeback: %s\n", node->entry->name);
}


void
generate_var_ident(node_t *node)
{
    symbol_t *ident_sym = node->entry;
    switch (ident_sym->type)
    {
    case SYM_GLOBAL_VAR:
        printf("$.%s", ident_sym->name);
        break;

    case SYM_PARAMETER:
        // If it is a paramter is one of the first 6, seacrch below bp
        if (ident_sym->seq < 6)
            printf("%ld(%%rbp)", -8 * (ident_sym->seq + 1));
        else
            // This requires that the parameters on 
            // stack is in reversed order... easier to implement
            printf("%ld(%%rbp)", 8 * (ident_sym->seq - 6 + 1 ));
        break;

    case SYM_LOCAL_VAR:
        printf("%ld(%%rbp)", -8 * (ident_sym->seq + 1));
        break;
    }
}


// This should allways push the result to stack
// no no no no, it should leave it in rax
void
solve_expressions(node_t *node)
{
    if (node->data)
    {   // Check if the expression is a function call
        bool is_function_call = !strcmp(node->data, "function_call");
        if (is_function_call)
        {
            generate_function_call(node);
            return;
        }
    }


    switch (node->n_children)
    {
    case 0:
        switch (node->type)
        {
        case IDENTIFIER_DATA:
            fetch_variable(node, "%rax");
            break;
        case NUMBER_DATA:
            printf("\tmovq\t$%ld,%%rax\n",*(int64_t*)node->data);
            break;
        }
        break;
    case 1:
        solve_expressions(node->children[0]);

        switch (*(char*)node->data)
        {
        case '-':
            ASM(negq, %rax);
            break;
        case '~':
            ASM(notq, %rax);
            break;
        }

        break;
    case 2:

        // First fetch lhs of expr and then rhs
        // Push results on stack
        for (int i = 0; i < 2; i++)
        {
            solve_expressions(node->children[i]);
            PUSH(%rax);
        }


        // Put rhs in %r10
        POP(%r10);
        // put lhs in %rax
        POP(%rax);

        // All operators below leaves result in %rax
        switch (*(char*)node->data)
        {
            /* Assignments */
            case '|': ASM(orq,  %r10, %rax);    break;  // Bitwise or of %rax and %r10
            case '^': ASM(xorq, %r10, %rax);    break;  // Bitwise xor of %rax and %r10
            case '&': ASM(andq, %r10, %rax);    break;  // Bitwise and of %rax and %r10
            case '+': ASM(addq, %r10, %rax);    break;  // Add %rax and %r10
            case '-': ASM(subq, %r10, %rax);    break;  // Subtract %r10 from %rax
            case '*': ASM(imulq, %r10);         break;  // Mulitply %rax with %r10
            case '/': 
                ASM(cqto);          // Convert rax to octaword, %rdx:%rax
                ASM(idivq, %r10);   // Divide %rdx:%rax by %r10
                break;
        }
        break;
    }
}

void
generate_function_call(node_t *node)
{
    printf("# Function call\n");

    bool isStack16ByteAligned = !(stack_depth % 2);

    // If the stack is 16 byte alligned here, offset 
    // by 1 because call pushes return addr to stack
    if (isStack16ByteAligned)
        PUSH($0);

    node_t *arg_list = node->children[1];
    if (arg_list->n_children)
        arg_list = arg_list->children[0];

    for (int arg = 0; arg < MIN(NO_REG_RECORD, arg_list->n_children); arg++)
    {
        switch (arg_list->children[arg]->type)
        {
        case NUMBER_DATA:
            printf("\tmovq\t$%ld, %s\n", 
                *(int64_t*)arg_list->children[arg]->data, 
                record[arg]
            );
            break;
        
        case EXPRESSION:
            solve_expressions(arg_list->children[arg]);
            printf("\tmovq\t%%rax, %s\n",record[arg]);
            break;
        
        default:
            fetch_variable(arg_list->children[arg], record[arg]);
            break;
        }
            
    }

    if (arg_list->n_children > NO_REG_RECORD)
    {
        // If there is an odd number of args to push to stack, add 1
        if (arg_list->n_children % 2)
            PUSH($0);

        for (int arg = arg_list->n_children - 1; arg >= NO_REG_RECORD; arg--)
        {
            switch (arg_list->children[arg]->type)
            {
            case NUMBER_DATA:
                printf("\tpushq\t$%ld\t\t\t\t# PUSH: %ld\n", 
                    *(int64_t*)arg_list->children[arg]->data,
                    ++stack_depth
                );
                break;
            
            // Arg can be an expression
            case EXPRESSION:
                solve_expressions(arg_list->children[arg]);
                PUSH(%rax);
                break;
            
            default:
                printf("\tpushq\t");
                generate_var_ident(arg_list->children[arg]);
                printf("\t\t\t\t# PUSH: %ld", ++stack_depth);
                putchar('\n');
                break;
            }
        }

    }

    printf("\tcall\t_%s\n", (char*)node->children[0]->data);

    // Aaaand pop the stack to return back to stack alignment
    if (isStack16ByteAligned)
        POP(%rcx);
    
    printf("# End of function call\n");

}

void
generate_function_return(node_t *node)
{
    solve_expressions(node->children[0]);
    ASM(leave);
    ASM(ret);
}

void
solve_statements(node_t *node, char *operator)
{
    node->type = EXPRESSION;
    node->data = strdup(operator);

    solve_expressions(node);

    writeback_variable(node->children[0], "%rax");
}

// Takes in a relation/number node and sets %rax to true if the statement is true
void
solve_relations(node_t *relation_root)
{
    switch (relation_root->type)
    {
    case NUMBER_DATA:
        solve_expressions(relation_root);
        break;
    
    case RELATION:
        if (relation_root->n_children != 2)
        {
            ERRPRT("Relation requires two expressions, one lhs and one rhs\n");
            exit(EXIT_FAILURE);
        }

        for (int i = 0; i < relation_root->n_children; i++)
        {
            solve_expressions(relation_root->children[i]);
            PUSH(%rax);
        }

        POP(%r10);
        POP(%rax);

        ASM(cmp, %r10, %rax);
        ASM(movq, $0, %rax);

        switch (*(char*)relation_root->data)
        {
        case '=':
            ASM(sete, %al);
            break;
        case '>':
            ASM(setg, %al);
            break;
        case '<':
            ASM(setl, %al);
            break;
        }
        break;
    }
}

void
generate_if_statement(node_t *node)
{
    uint64_t current_if_seq = if_seq++;

    COMMENT("Begin IF %ld", current_if_seq);

    // The realtion is allways in the first part of the IF
    solve_relations(node->children[0]);
    
    ASM(cmp, $1, %rax);
    printf("\tjne\t%s%03ld\n", (node->n_children > 2) ? "ELSE" : "ENDIF", current_if_seq);
    generate_node(node->children[1]);

    if (node->n_children > 2) {
        printf("\tjmp \tENDIF%03ld\n", current_if_seq);
        printf("ELSE%03ld:\n", current_if_seq);

        generate_node(node->children[2]);
    }

    COMMENT("End IF %ld", current_if_seq);
    printf("ENDIF%03ld:\n", current_if_seq);
}

void
generate_while_statement(node_t *node)
{
    uint64_t current_while_seq = while_seq++;
    
    uint64_t prev_closest_while = closest_while;
    closest_while = current_while_seq;

    COMMENT("Begin WHILE %ld", current_while_seq);   
    printf("WHILE%03ld:\n", current_while_seq);

    // Relation is allways the first entry in a while
    solve_relations(node->children[0]);

    ASM(cmp, $1, %rax);

    printf("\tjne\tENDWHILE%03ld\n", current_while_seq);

    generate_node(node->children[1]);
    closest_while = prev_closest_while;

    printf("\tjmp \tWHILE%03ld\n", current_while_seq);
    printf("ENDWHILE%03ld:\n", current_while_seq);

}

void
solve_continue_statement()
{
    COMMENT("Continue to WHILE%03ld", closest_while);
    printf("\tjmp \tWHILE%03ld\n", closest_while);
}

/**Generates the main function with argument parsing and calling of our
 * main function (first, if no function is named main)
 * @param first Symbol table entry of our main function */
void
generate_main ( symbol_t *first )
{
    puts("###### Entry point for GAS #####");
    puts ( ".globl main" );
    puts ( ".section .text" );
    puts ( "main:" );
    puts ( "\tpushq   %rbp" );  // Added this for stack alignment
    puts ( "\tpushq   %rbp" );
    puts ( "\tmovq    %rsp, %rbp" );

    printf ( "\tsubq\t$1,%%rdi\n" );
    printf ( "\tcmpq\t$%zu,%%rdi\n", first->nparms );
    printf ( "\tjne \tABORT\n" );
    printf ( "\tcmpq\t$0,%%rdi\n" );
    printf ( "\tjz  \tSKIP_ARGS\n" );

    printf ( "\tmovq\t%%rdi,%%rcx\n" );
    printf ( "\taddq\t$%zu, %%rsi\n", 8*first->nparms );

    // Modification to mail call, remove this if no params supplied
    if (first->nparms)
    {
        printf ( "PARSE_ARGV:\n" );
        printf ( "\tpushq\t%%rcx\n" );
        printf ( "\tpushq\t%%rsi\n" );

        printf ( "\tmovq\t(%%rsi),%%rdi\n" );
        printf ( "\tmovq\t$0,%%rsi\n" );
        printf ( "\tmovq\t$10,%%rdx\n" );
        printf ( "\tcall\tstrtol\n" );

    /*  Now a new argument is an integer in rax */

        printf ( "\tpopq\t%%rsi\n" );
        printf ( "\tpopq\t%%rcx\n" );
        printf ( "\tpushq\t%%rax\n" );
        printf ( "\tsubq\t$8, %%rsi\n" );
        printf ( "\tloop\tPARSE_ARGV\n" );

        /* Now the arguments are in order on stack */
        for (int arg = 0; arg < MIN(6,first->nparms); arg++)
            printf ( "\tpopq\t%s\n", record[arg] );

    }

    printf ( "SKIP_ARGS:\n" );
    printf ( "\tcall\t_%s\n", first->name );
    printf ( "\tjmp \tEND\n" );
    printf ( "ABORT:\n" );
    printf ( "\tmovq\t$.errout, %%rdi\n" );
    printf ( "\tcall\tputs\n" );

    printf ( "END:\n" );
    puts ( "\tmovq \t%rax, %rdi" );
    puts ( "\tcall \texit" );
    puts("###### FUNCTIONS FROM VSL BELOW #####");
    putchar('\n');

}

static uint64_t 
fetch_symbols(tlhash_t* symbol_table, symbol_t*** symbol_list)
{
    uint64_t no_symbols = tlhash_size(symbol_table);
    *symbol_list = malloc(no_symbols * sizeof(symbol_t));
    tlhash_values(symbol_table, (void **)*symbol_list );

    return no_symbols;
}