kraken/stdlib/bytecode_generator.krak

import io:*
import mem:*
import map:*
import hash_map:*
import stack:*
import string:*
import util:*
import tree:*
import symbol:*
import ast_nodes:*
// for error with syntax tree
import pass_common:*
import poset:*

fun type_size(t: *type): ulong
    return type_size_and_alignment(t).first
fun type_size_and_alignment(t: *type): pair<ulong,ulong> {
    if (t->indirection)
        return make_pair(#sizeof<*void>, #sizeof<*void>)
    match (t->base) {
        base_type::object() {
            var total_size: ulong = 0
            var max_size: ulong = 0
            var max_align: ulong = 0
            t->type_def->type_def.variables.for_each(fun(i: *ast_node) {
                var individual = type_size_and_alignment(i->declaration_statement.identifier->identifier.type)
                max_size = max(max_size, individual.first)
                max_align = max(max_align, individual.second)
                // increase total size by the individual size + padding to get alignment
                var padding = 0
                if (individual.second != 0)
                    padding = (individual.second - (total_size % individual.second)) % individual.second
                total_size += individual.first + padding
            })
            if (t->type_def->type_def.is_union)
                total_size = max_size
            // pad the end so that consecutive objects in memory are aligned
            if (max_align != 0)
                total_size += (max_align - (total_size % max_align)) % max_align
            return make_pair(total_size, max_align)
        }
        base_type::function() return make_pair(#sizeof<*void>, #sizeof<*void>)
        base_type::boolean() return make_pair(#sizeof<bool>, #sizeof<bool>)
        base_type::character() return make_pair(#sizeof<char>, #sizeof<char>)
        base_type::ucharacter() return make_pair(#sizeof<uchar>, #sizeof<uchar>)
        base_type::short_int() return make_pair(#sizeof<short>, #sizeof<short>)
        base_type::ushort_int() return make_pair(#sizeof<ushort>, #sizeof<ushort>)
        base_type::integer() return make_pair(#sizeof<int>, #sizeof<int>)
        base_type::uinteger() return make_pair(#sizeof<uint>, #sizeof<uint>)
        base_type::long_int() return make_pair(#sizeof<long>, #sizeof<long>)
        base_type::ulong_int() return make_pair(#sizeof<ulong>, #sizeof<ulong>)
        base_type::floating() return make_pair(#sizeof<float>, #sizeof<float>)
        base_type::double_precision() return make_pair(#sizeof<double>, #sizeof<double>)
    }
    error(string("Invalid type for type_size: ") + t->to_string())
}

fun offset_into_struct(struct_type: *type, ident: *ast_node): ulong {
    var offset: ulong = 0
    if (struct_type->type_def->type_def.is_union)
        return offset
    for (var i = 0; i < struct_type->type_def->type_def.variables.size; i++;) {
        var size_and_align = type_size_and_alignment(struct_type->type_def->type_def.variables[i]->declaration_statement.identifier->identifier.type)
        var align = size_and_align.second
        if (align != 0)
            offset += (align - (offset % align)) % align
        if (struct_type->type_def->type_def.variables[i]->declaration_statement.identifier == ident)
            break
        else
            offset += size_and_align.first
    }
    return offset
}

var register_size = #sizeof<*void>
adt operand_size {
    b8,
    b16,
    b32,
    b64
}
fun size_to_operand_size(size: ulong): operand_size {
    if (size == 1) return operand_size::b8()
    if (size == 2) return operand_size::b16()
    if (size == 4) return operand_size::b32()
    if (size == 8) return operand_size::b64()
    error("invalid operand size")
}
adt byte_inst {
    nop,
    imm:  imm,
    add:  add,
    ldr:  ldr,
    str:  str,
    jmp:  jmp,
    jz:   jz,
    call: call,
    ret
}
obj imm {
    var reg: int
    var val: long
}
obj add {
    var to_reg: int
    var a: int
    var b: int
}
obj ldr {
    var to_reg: int
    var from_reg: int
    var offset: long
    var size: operand_size
}
obj str {
    var to_reg: int
    var offset: long
    var from_reg: int
    var size: operand_size
}
obj jmp {
    var offset: long
}
obj jz {
    var reg: int
    var offset: long
}
obj call {
    var reg: int
}

fun to_string(s: operand_size): string {
    match (s) {
        operand_size::b8()  return string("8")
        operand_size::b16() return string("16")
        operand_size::b32() return string("32")
        operand_size::b64() return string("64")
    }
    return string("missed operand size")
}

fun to_string(b: byte_inst): string {
    match (b) {
        byte_inst::nop()     return string("nop")
        byte_inst::imm(i)    return string("r") + i.reg + " = imm " + i.val
        byte_inst::add(a)    return string("r") + a.to_reg + " = r" + a.a + " + r" + a.b
        byte_inst::ldr(l)    return string("r") + l.to_reg + " = ldr" + to_string(l.size) + " r" + l.from_reg + " (" + l.offset + ")"
        byte_inst::str(s)    return "str" + to_string(s.size) + " (r" + s.to_reg + "(" + s.offset + ") <= r" + s.from_reg + ")"
        byte_inst::jmp(j)    return string("jmp(pc += ") + j.offset + ")"
        byte_inst::jz(j)     return string("jmp(r") + j.reg + " == 0, pc += " + j.offset + ")"
        byte_inst::call(c)    return string("call pc = r") + c.reg
        byte_inst::ret()     return string("ret")
    }
    return string("Missed byte_inst case in to_string")
}

fun bytecode_to_string(functions: ref vector<bytecode_function>, instructions: ref vector<byte_inst>): string {
    return string("\n").join(functions.map(fun(bb: ref bytecode_function): string return bb.to_string(instructions);))
}

fun bytecode_function(name: ref string, start: int): bytecode_function {
    var to_ret.construct(name, start): bytecode_function
    return to_ret
}
obj bytecode_function (Object) {
    var name: string
    var instruction_start: int
    var instruction_end: int
    var var_to_frame_offset: map<*ast_node, int>
    var frame_size: int

    fun construct(): *bytecode_function {
        instruction_start = 0
        instruction_end = 0
        name.construct()
        var_to_frame_offset.construct()
        frame_size = 0
        return this
    }
    fun construct(name_in: ref string, instruction_start_in: int): *bytecode_function {
        instruction_start = instruction_start_in
        instruction_end = 0
        name.copy_construct(&name_in)
        var_to_frame_offset.construct()
        frame_size = 0
        return this
    }
    fun copy_construct(old: *bytecode_function) {
        instruction_start = old->instruction_start
        instruction_end = old->instruction_end
        name.copy_construct(&old->name)
        var_to_frame_offset.copy_construct(&old->var_to_frame_offset)
        frame_size = old->frame_size
    }
    fun operator=(other: ref bytecode_function) {
        destruct()
        copy_construct(&other)
    }
    fun destruct() {
        name.destruct()
        var_to_frame_offset.destruct()
    }
    fun to_string(instructions: ref vector<byte_inst>): string {
        var res = name + "(frame size " + frame_size + "):\n"
        res += "\t frame layout\n"
        res += "\t\tsaved RBP : RPB = 0\n"
        var_to_frame_offset.for_each(fun(n: *ast_node, o: int) {
            res += "\t\t" + n->identifier.name + ": RBP + " + o + "\n"
        })
        res += "\n\t bytecode\n"
        for (var i = instruction_start; i < instruction_end; i++;)
            res += string("\t\t") + i + string(": ") + to_string(instructions[i]) + "\n"
        return res
    }
}

obj bytecode_generator (Object) {
    var reg_counter: int
    var reg_max: int
    var id_counter: int
    var ast_name_map: hash_map<*ast_node, string>
    var functions: vector<bytecode_function>
    var node_function_idx: map<*ast_node, int>
    var instructions: vector<byte_inst>
    var fixup_function_addresses: vector<pair<int, *ast_node>>
    fun construct(): *bytecode_generator {
        id_counter = 0
        ast_name_map.construct()
        functions.construct()
        node_function_idx.construct()
        instructions.construct()
        fixup_function_addresses.construct()
        reg_counter = 3
        reg_max = 3

        return this
    }
    fun copy_construct(old: *bytecode_generator) {
        reg_counter = old->reg_counter
        reg_max = old->reg_max
        id_counter = old->id_counter
        ast_name_map.copy_construct(&old->ast_name_map)
        functions.copy_construct(&old->functions)
        node_function_idx.copy_construct(&old->node_function_idx)
        instructions.copy_construct(&old->instructions)
        fixup_function_addresses.copy_construct(&old->fixup_function_addresses)
    }
    fun operator=(other: ref bytecode_generator) {
        destruct()
        copy_construct(&other)
    }
    fun destruct() {
        ast_name_map.destruct()
        functions.destruct()
        node_function_idx.destruct()
        instructions.destruct()
        fixup_function_addresses.destruct()
    }
    fun get_id(): string return to_string(id_counter++);
    fun get_reg(): int return reg_counter++;
    fun reset_reg() {
        if (reg_counter > reg_max) {
            reg_max = reg_counter
        }
        reg_counter = 3
    }
    /*fun generate_bytecode(name_ast_map: map<string, pair<*tree<symbol>,*ast_node>>): pair<vector<bytecode_function>, vector<byte_inst>> {*/
    fun generate_bytecode(name_ast_map: map<string, pair<*tree<symbol>,*ast_node>>) {

        // iterate through asts
        name_ast_map.for_each(fun(name: string, tree_pair: pair<*tree<symbol>,*ast_node>) {
            // iterate through children for each ast
            // do lambdas seperatly, so we can reconstitute the enclosing object if it has one
            tree_pair.second->translation_unit.lambdas.for_each(fun(child: *ast_node) {
                generate_function_definition(child)
            })
            tree_pair.second->translation_unit.children.for_each(fun(child: *ast_node) {
                match (*child) {
                    ast_node::declaration_statement(backing)    generate_declaration_statement(child)
                    ast_node::compiler_intrinsic(backing)       generate_compiler_intrinsic(child)
                    ast_node::function(backing)                 generate_function_definition(child)
                    ast_node::template(backing) {
                        backing.instantiated.for_each(fun(node: *ast_node) {
                            match (*node) {
                                ast_node::function(backing) generate_function_definition(node)
                                ast_node::type_def(backing) {
                                    backing.methods.for_each(fun(method: *ast_node) {
                                        if (is_template(method))
                                            method->template.instantiated.for_each(fun(m: *ast_node) generate_function_definition(m);)
                                        else
                                            generate_function_definition(method)
                                    })
                                }
                            }
                        })
                    }
                    ast_node::type_def(backing) {
                        backing.methods.for_each(fun(method: *ast_node) {
                            if (is_template(method))
                                method->template.instantiated.for_each(fun(m: *ast_node) generate_function_definition(m);)
                            else
                                generate_function_definition(method)
                        })
                    }
                }
            })
        })
        fixup_function_addresses.for_each(fun(p: pair<int, *ast_node>) {
            instructions[p.first].imm.val = functions[node_function_idx[p.second]].instruction_start
        })
        for (var i = 0; i < functions.size - 1; i++;)
            functions[i].instruction_end = functions[i+1].instruction_start
        functions.last().instruction_end = instructions.size
        /*return make_pair(functions, instructions)*/
    }
    fun generate_function_definition(node: *ast_node): int {
        reset_reg()
        node_function_idx[node] = functions.size
        functions.add(bytecode_function(get_name(node), instructions.size))
        var parameter_offset = (register_size*2) cast int // have to pass saved RBP and return address
        node->function.parameters.for_each(fun(p: *ast_node) {
            functions.last().var_to_frame_offset[p] = parameter_offset
            parameter_offset += type_size(p->identifier.type)
        })
        emit_add(0, 0, emit_imm(-register_size))    // these two lines push rbp onto the stack, which grows towards negative
        emit_str(0, 0, 1, operand_size::b64())      // rsp[0] <= rbp
        emit_add(1, 0, emit_imm(0))                 // note that we start the frame size at register_size for this reason

        var push_frame_idx = instructions.size
        emit_add(0, 0, emit_imm(0))                 // this has to be fixed afterwards to be the -frame_size

        generate(node->function.body_statement)

        instructions[push_frame_idx].imm.val = -functions.last().frame_size

        return -1
    }
    fun generate_declaration_statement(node: *ast_node): int {
        var identifier = node->declaration_statement.identifier
        var ident_type = identifier->identifier.type
        functions.last().frame_size += type_size(ident_type)
        functions.last().var_to_frame_offset[identifier] = -functions.last().frame_size
        if (node->declaration_statement.expression) {
            emit_str(1, functions.last().var_to_frame_offset[identifier], generate(node->declaration_statement.expression), size_to_operand_size(type_size(get_ast_type(identifier))))
        }
        return -1
    }
    fun generate_assignment_statement(node: *ast_node): int {
        /*var to   = generate(node->assignment_statement.to, true)*/
        var from = generate(node->assignment_statement.from)
        var to   = generate(node->assignment_statement.to, true)
        emit_str(to, 0, from, size_to_operand_size(type_size(get_ast_type(node->assignment_statement.to))))
        return -1
    }
    fun generate_if_statement(node: *ast_node): int {
        var cond_reg = generate(node->if_statement.condition)
        var jz_index = instructions.size
        emit_jz(cond_reg,0)
        generate(node->if_statement.then_part)
        if (node->if_statement.else_part) {
            var jmp_index = instructions.size
            emit_jmp(0)
            instructions[jz_index].jz.offset = instructions.size - jz_index
            generate(node->if_statement.else_part)
            instructions[jmp_index].jmp.offset = instructions.size - jmp_index
        } else {
            instructions[jz_index].jz.offset = instructions.size - jz_index
        }
        return -1
    }
    fun generate_while_loop(node: *ast_node): int {
        generate(node->while_loop.condition)
        generate(node->while_loop.statement)
        return -1
    }
    fun generate_for_loop(node: *ast_node): int {
        if (node->for_loop.init)
            generate(node->for_loop.init)
        if (node->for_loop.condition)
            generate(node->for_loop.condition)
        if (node->for_loop.update)
            generate(node->for_loop.update)
        generate(node->for_loop.body)
        return -1
    }
    fun generate_identifier(node: *ast_node, lvalue: bool): int {
        if (lvalue) {
            return emit_add(1, emit_imm(functions.last().var_to_frame_offset[node]))
        } else {
            return emit_ldr(1, functions.last().var_to_frame_offset[node], size_to_operand_size(type_size(get_ast_type(node))))
        }
    }
    fun generate_return_statement(node: *ast_node): int {
        if (node->return_statement.return_value) {
            /*emit_str(1, register_size, generate(node->return_statement.return_value))*/
            emit_add(2, emit_imm(0), generate(node->return_statement.return_value))
            emit_add(0, 1, emit_imm(register_size))
            emit_ldr(1, 1, 0, operand_size::b64())
            emit_ret()
        } else {
            emit_ret()
        }
        return -1
    }
    fun generate_branching_statement(node: *ast_node): int {
        match(node->branching_statement.b_type) {
            branching_type::break_stmt()    instructions.add(byte_inst::nop())
            branching_type::continue_stmt() instructions.add(byte_inst::nop())
        }
        return -1
    }
    fun generate_cast(node: *ast_node): int {
        return generate(node->cast.value)
    }
    fun generate_value(node: *ast_node): int {
        if (node->value.value_type->is_bool())
            return emit_imm((node->value.string_value == "true") cast int)
        else
            return emit_imm(string_to_num<int>(node->value.string_value))
    }
    fun generate_code_block(node: *ast_node): int {
        node->code_block.children.for_each(fun(child: *ast_node) {
            // registers aren't used between statements (only stack reg)
            reset_reg()
            generate(child)
        })
        return -1
    }
    // this generates the function as a value, not the actual function
    fun generate_function(node: *ast_node): int {
        fixup_function_addresses.add(make_pair(instructions.size,node))
        return emit_imm(-2)
    }
    fun generate_function_call(node: *ast_node, lvalue: bool): int {
        var stack_offset = 0
        // reverse order
        node->function_call.parameters.reverse().for_each(fun(child: *ast_node) {
            // push param onto stack
            var param_size = type_size(get_ast_type(child))
            emit_add(0, 0, emit_imm(-param_size))
            stack_offset += param_size
            emit_str(0, 0, generate(child), size_to_operand_size(param_size))
        })
        var return_reg = emit_call(generate_function(node->function_call.func))
        emit_add(0, 0, emit_imm(stack_offset))
        return return_reg
    }

    fun generate_compiler_intrinsic(node: *ast_node): int {
        instructions.add(byte_inst::nop())
        return -1
    }

    fun generate(node: *ast_node): int return generate(node, false)
    fun generate(node: *ast_node, lvalue: bool): int {
        match (*node) {
            ast_node::declaration_statement(backing)    return generate_declaration_statement(node)
            ast_node::assignment_statement(backing)     return generate_assignment_statement(node)
            ast_node::if_statement(backing)             return generate_if_statement(node)
            ast_node::while_loop(backing)               return generate_while_loop(node)
            ast_node::for_loop(backing)                 return generate_for_loop(node)
            ast_node::function(backing)                 return generate_function(node)
            ast_node::function_call(backing)            return generate_function_call(node, lvalue)
            ast_node::compiler_intrinsic(backing)       return generate_compiler_intrinsic(node)
            ast_node::code_block(backing)               return generate_code_block(node)
            ast_node::return_statement(backing)         return generate_return_statement(node)
            ast_node::branching_statement(backing)      return generate_branching_statement(node)
            ast_node::cast(backing)                     return generate_cast(node)
            ast_node::value(backing)                    return generate_value(node)
            ast_node::identifier(backing)               return generate_identifier(node, lvalue)
        }
        error("Bad node")
    }
    fun get_name(node: *ast_node): string {
        var maybe_it = ast_name_map.get_ptr_or_null(node);
        if (maybe_it)
            return *maybe_it
        var result = get_ast_name(node) + get_id()
        if (is_function(node) && node->function.name == "main")
            result = "main"
        ast_name_map.set(node, result)
        return result
    }
    fun emit_imm(value: ulong): int { return emit_imm((value) cast int); }
    fun emit_imm(value: int): int {
        var i: imm
        i.reg = get_reg()
        i.val = value
        instructions.add(byte_inst::imm(i))
        return i.reg
    }
    fun emit_add(a: int, b: int): int {
        return emit_add(get_reg(), a, b)
    }
    fun emit_add(dest: int, a: int, b: int): int {
        var i: add
        i.to_reg = dest
        i.a = a
        i.b = b
        instructions.add(byte_inst::add(i))
        return i.to_reg
    }
    fun emit_ldr(reg: int, offset: int, size: operand_size): int { return emit_ldr(get_reg(), reg, offset, size); }
    fun emit_ldr(dest: int, reg: int, offset: int, size: operand_size): int {
        var l: ldr
        l.to_reg = dest
        l.from_reg = reg
        l.offset = offset
        l.size = size
        instructions.add(byte_inst::ldr(l))
        return l.to_reg
    }
    fun emit_str(to_reg: int, offset: int, from_reg: int, size: operand_size): int {
        var s: str
        s.to_reg = to_reg
        s.offset = offset
        s.from_reg = from_reg
        s.size = size
        instructions.add(byte_inst::str(s))
        return -1
    }
    fun emit_jmp(offset: int): int {
        var j: jmp
        j.offset = offset
        instructions.add(byte_inst::jmp(j))
        return -1
    }
    fun emit_jz(reg: int, offset: int): int {
        var j: jz
        j.reg = reg
        j.offset = offset
        instructions.add(byte_inst::jz(j))
        return -1
    }
    fun emit_ret(): int {
        instructions.add(byte_inst::ret())
        return -1
    }
    fun emit_call(reg: int): int {
        var c: call
        c.reg = reg
        instructions.add(byte_inst::call(c))
        return 2
    }

    // Stack ABI
    // it's system v x64, but all params passed on stack
    fun evaluate(): int {
        println("evaling main")
        println(bytecode_to_string(functions, instructions))
        var main_entry = functions.find_first_satisfying(fun(block: bytecode_function): bool return block.name == "main";)
        var registers.construct(reg_max): vector<long>
        registers.size = reg_max
        registers[0] = -register_size // with the stack being zeroed out, this makes it a return address of 0
        registers[1] = 0xdeadbeefcafebabe
        var stack_size = 8 * 1024 * 1024
        var stack = new<uchar>(stack_size) + stack_size
        for (var i = 0; i < stack_size; i++;)
            stack[-i + -1] = 0
        for (var i = main_entry.instruction_start; i < instructions.size; i++;) {
            println(string("evaling: ") + i + ": " + to_string(instructions[i]))
            match(instructions[i]) {
                byte_inst::nop()     {}
                byte_inst::imm(i)    registers[i.reg] = i.val
                byte_inst::add(a)    registers[a.to_reg] = registers[a.a] + registers[a.b]
                byte_inst::ldr(l)    match (l.size) {
                                        operand_size::b8()  registers[l.to_reg] = *(stack + registers[l.from_reg] + l.offset) cast *char
                                        operand_size::b16() registers[l.to_reg] = *(stack + registers[l.from_reg] + l.offset) cast *short
                                        operand_size::b32() registers[l.to_reg] = *(stack + registers[l.from_reg] + l.offset) cast *int
                                        operand_size::b64() registers[l.to_reg] = *(stack + registers[l.from_reg] + l.offset) cast *long
                                     }
                byte_inst::str(s)    match (s.size) {
                                        operand_size::b8()  *(stack + registers[s.to_reg] + s.offset) cast *uchar  = registers[s.from_reg]
                                        operand_size::b16() *(stack + registers[s.to_reg] + s.offset) cast *ushort = registers[s.from_reg]
                                        operand_size::b32() *(stack + registers[s.to_reg] + s.offset) cast *uint   = registers[s.from_reg]
                                        operand_size::b64() *(stack + registers[s.to_reg] + s.offset) cast *ulong  = registers[s.from_reg]
                                     }
                byte_inst::jmp(j)    i += j.offset - 1 // to counteract pc inc
                byte_inst::jz(j)     if (registers[j.reg] == 0)
                                        i += j.offset - 1 // to counteract pc inc
                byte_inst::call(c)    {
                    /*registers[0] -= register_size*/
                    registers[0] = registers[0] - register_size
                    *(stack + registers[0]) cast *long = i + 1
                    i = registers[c.reg] - 1
                    print("call!")
                    println("first part of memory is (after push)")
                    for (var i = 0; i < 8*8; i+=8;) {
                        print(string("-") + i + string(": "))
                        for (var j = 0; j < 8; j++;) {
                            if (j == 4)
                                print(" ")
                            print(*(stack - (i+j)*#sizeof<uchar> - 1) cast *uchar)
                            print(" ")
                        }
                        println()
                    }
                    println("Done")
                }
                byte_inst::ret()     {
                    var pc = *(stack + registers[0]) cast *long
                    /*registers[0] += register_size*/
                    registers[0] = registers[0] + register_size
                    print("returning! return value is\n\t")
                    var value = registers[2]
                    println(value)
                    println("first part of memory is")
                    for (var i = 0; i < 8*8; i+=8;) {
                        print(string("-") + i + string(": "))
                        for (var j = 0; j < 8; j++;) {
                            if (j == 4)
                                print(" ")
                            print(*(stack - (i+j)*#sizeof<uchar> - 1) cast *uchar)
                            print(" ")
                        }
                        println()
                    }
                    println("Done")
                    if (pc == 0) {
                        return value
                    } else {
                        i = pc - 1
                        println(string("returning to ") + pc)
                    }
                }
            }
        }
        return -1
    }
}