# This Source Code Form is subject to the terms of the Mozilla Public License,
# v. 2.0. If a copy of the MPL was not distributed with this file, You can
# obtain one at https://mozilla.org/MPL/2.0/.

# Stage 9 compiler.
#

const
	symbol_builtin_name_int := "Int";
	symbol_builtin_name_word := "Word";
	symbol_builtin_name_pointer := "Pointer";
	symbol_builtin_name_char := "Char";
	symbol_builtin_name_bool := "Bool";

	# Every type info starts with a word describing what type it is.
	#
	# PRIMITIVE_TYPE = 1
	#
	# Primitive types have only type size.
	symbol_builtin_type_int := S(1, 4);
	symbol_builtin_type_word := S(1, 4);
	symbol_builtin_type_pointer := S(1, 4);
	symbol_builtin_type_char := S(1, 1);
	symbol_builtin_type_bool := S(1, 1);

	# Info objects start with a word describing its type.
	#
	# INFO_TYPE = 1
	#
	# Type info has the type it belongs to.
	symbol_type_info_int := S(1, @symbol_builtin_type_int);
	symbol_type_info_word := S(1, @symbol_builtin_type_word);
	symbol_type_info_pointer := S(1, @symbol_builtin_type_pointer);
	symbol_type_info_char := S(1, @symbol_builtin_type_char);
	symbol_type_info_bool := S(1, @symbol_builtin_type_bool);

# When modifiying also change the read size in the entry point procedure.
.type source_code, @object
source_code: .zero 81920

.type compiler_strings, @object
compiler_strings: .zero 8192

var
	compiler_strings_position: Pointer := @compiler_strings;
	compiler_strings_length: Word := 0;
	source_code_position: Pointer := @source_code;

# Calculates and returns the string token length between quotes, including the
# escaping slash characters.
#
# Parameters:
# a0 - String token pointer.
#
# Returns the length in a0.
proc _string_length();
begin
	# Reset the counter.
	v0 := 0;

.string_length_loop:
	v88 := v88 + 1;

	lw t0, 88(sp)
	lb t0, (t0)

	li t1, '"'
	beq t0, t1, .string_length_end

	v0 := v0 + 1;
	goto .string_length_loop;

.string_length_end:
	return v0
end;

# Adds a string to the global, read-only string storage.
#
# Parameters:
# a0 - String token.
#
# Returns the offset from the beginning of the storage to the new string in a0.
proc _add_string();
begin
	v0 := v88 + 1;
	v4 := compiler_strings_length;

.add_string_loop:
	lw t0, 0(sp)
	lb t1, (t0)
	li t2, '"'

	beq t1, t2, .add_string_end

	la t2, compiler_strings_position
	lw t3, (t2)
	sb t1, (t3)

	addi t3, t3, 1
	sw t3, (t2)

	addi t0, t0, 1
	sw t0, 0(sp)

	li t2, '\\'
	bne t1, t2, .add_string_increment

	goto .add_string_loop;

.add_string_increment:
	la t2, compiler_strings_length
	lw t4, (t2)
	addi t4, t4, 1
	sw t4, (t2)

	goto .add_string_loop;

.add_string_end:
	return v4
end;

# Reads standard input into a buffer.
# a0 - Buffer pointer.
# a1 - Buffer size.
#
# Returns the amount of bytes written in a0.
proc _read_file();
begin
	mv a2, a1
	mv a1, a0
	# STDIN.
	li a0, 0
	li a7, 63 # SYS_READ.
	ecall
end;

# Writes to the standard output.
#
# Parameters:
# a0 - Buffer.
# a1 - Buffer length.
proc _write_s();
begin
	mv a2, a1
	mv a1, a0
	# STDOUT.
	li a0, 1
	li a7, 64 # SYS_WRITE.
	ecall
end;

# Writes a number to a string buffer.
#
# t0 - Local buffer.
# t1 - Constant 10.
# t2 - Current character.
# t3 - Whether the number is negative.
#
# Parameters:
# a0 - Whole number.
# a1 - Buffer pointer.
#
# Sets a0 to the length of the written number.
proc _print_i();
begin
	li t1, 10
	addi t0, s0, -9

	li t3, 0
	bgez a0, .print_i_digit10
	li t3, 1
	neg a0, a0

.print_i_digit10:
	rem t2, a0, t1
	addi t2, t2, '0'
	sb t2, 0(t0)
	div a0, a0, t1
	addi t0, t0, -1
	bne zero, a0, .print_i_digit10

	beq zero, t3, .print_i_write_call
	addi t2, zero, '-'
	sb t2, 0(t0)
	addi t0, t0, -1

.print_i_write_call:
	mv a0, a1
	addi a1, t0, 1
	sub a2, s0, t0
	addi a2, a2, -9
	sw a2, 0(sp)

	_memcpy();

	return v0
end;

# Writes a number to the standard output.
#
# Parameters:
# a0 - Whole number.
proc _write_i();
begin
	_print_i(v88, @v0);

	mv a1, a0
	addi a0, sp, 0
	_write_s();

end;

# Writes a character from a0 into the standard output.
proc _write_c();
begin
	_write_s(@v88, 1);
end;

# Write null terminated string.
#
# Parameters:
# a0 - String.
proc _write_z();
begin
.write_z_loop:
	# Check for 0 character.
	lw a0, 88(sp)
	lb a0, (a0)
	beqz a0, .write_z_end

	# Print a character.
	_write_c();

	# Advance the input string by one byte.
	v88 := v88 + 1;

	goto .write_z_loop;

.write_z_end:
end;

# Detects if a0 is an uppercase character. Sets a0 to 1 if so, otherwise to 0.
proc _is_upper();
begin
	v0 := v88 >= 'A';
	v4 := v88 <= 'Z';

	return v0 & v4

end;

# Detects if a0 is an lowercase character. Sets a0 to 1 if so, otherwise to 0.
proc _is_lower();
begin
	v0 := v88 >= 'a';
	v4 := v88 <= 'z';

	return v0 & v4

end;

# Detects if the passed character is a 7-bit alpha character or an underscore.
#
# Paramters:
# a0 - Tested character.
#
# Sets a0 to 1 if the character is an alpha character or underscore, sets it to 0 otherwise.
proc _is_alpha();
begin
	sw a0, 0(sp)

	_is_upper();
	sw a0, 4(sp)

	_is_lower(v0);

	lw t0, 0(sp)
	xori t1, t0, '_'
	seqz t1, t1

	lw t0, 4(sp)
	or a0, a0, t0
	or a0, a0, t1
end;

# Detects whether the passed character is a digit
# (a value between 0 and 9).
#
# Parameters:
# a0 - Exemined value.
#
# Sets a0 to 1 if it is a digit, to 0 otherwise.
proc _is_digit();
begin
	v0 := v88 >= '0';
	v4 := v88 <= '9';

	return v0 & v4
end;

proc _is_alnum();
begin
	sw a0, 4(sp)

	_is_alpha();
	sw a0, 0(sp)

	_is_digit(v4);

	lw a1, 0(sp)
	or a0, a0, a1
end;

# Reads the next token.
#
# Returns token length in a0.
proc _read_token();
begin
	# Current token position.
	v0 := source_code_position;
	# Token length.
	v4 := 0;

.read_token_loop:
	lw t0, 0(sp)
	lb t0, (t0) # Current character.

	# First we try to read a derictive.
	# A derictive can contain a dot and characters.
	li t1, '.'
	beq t0, t1, .read_token_next

	lw a0, 0(sp)
	lb a0, (a0)
	_is_alnum();
	bnez a0, .read_token_next

	goto .read_token_end;

.read_token_next:
	# Advance the source code position and token length.
	v4 := v4 + 1;
	v0 := v0 + 1;

	goto .read_token_loop;

.read_token_end:
	return v4
end;

# a0 - First pointer.
# a1 - Second pointer.
# a2 - The length to compare.
#
# Returns 0 in a0 if memory regions are equal.
proc _memcmp();
begin
	mv t0, a0
	li a0, 0

.memcmp_loop:
	beqz a2, .memcmp_end

	lbu t1, (t0)
	lbu t2, (a1)
	sub a0, t1, t2

	bnez a0, .memcmp_end

	addi t0, t0, 1
	addi a1, a1, 1
	addi a2, a2, -1

	goto .memcmp_loop;

.memcmp_end:
end;

# Copies memory.
#
# Parameters:
# a0 - Destination.
# a1 - Source.
# a2 - Size.
#
# Preserves a0.
proc _memcpy();
begin
	mv t0, a0

.memcpy_loop:
	beqz a2, .memcpy_end

	lbu t1, (a1)
	sb t1, (a0)

	addi a0, a0, 1
	addi a1, a1, 1
	addi a2, a2, -1

	goto .memcpy_loop

.memcpy_end:
	mv a0, t0
end;

# Advances the token stream by a0 bytes.
proc _advance_token();
begin
	la t0, source_code_position
	lw t1, (t0)
	add t1, t1, a0
	sw t1, (t0)
end;

# Prints the current token.
#
# Parameters:
# a0 - Token length.
#
# Returns a0 unchanged.
proc _write_token();
begin
	_write_s(source_code_position, v88);
	return v88
end;

proc _compile_section();
begin
	# Print and skip the ".section" (8 characters) directive and a space after it.
	_write_token(9);
	_advance_token();

	# Read the section name.
	_read_token();
	addi a0, a0, 1

	_write_token();
	_advance_token();
end;

# Prints and skips a line.
proc _skip_comment();
begin
	la t0, source_code_position
	lw t1, (t0)

.skip_comment_loop:
	# Check for newline character.
	lb t2, (t1)
	li t3, '\n'
	beq t2, t3, .skip_comment_end

	# Advance the input string by one byte.
	addi t1, t1, 1
	sw t1, (t0)

	goto .skip_comment_loop;

.skip_comment_end:
	# Skip the newline.
	addi t1, t1, 1
	sw t1, (t0)
end;

# Prints and skips a line.
proc _compile_line();
begin
.compile_line_loop:
	la a0, source_code_position
	lw a1, (a0)

	lb t0, (a1)
	li t1, '\n'
	beq t0, t1, .compile_line_end

	# Print a character.
	lw a0, (a1)
	_write_c();

	# Advance the input string by one byte.
	_advance_token(1);

	goto .compile_line_loop;

.compile_line_end:
	_write_c('\n');

	_advance_token(1);
end;

proc _compile_integer_literal();
begin
	_write_z("\tli t0, \0");

	_read_token();
	_write_token();
	_advance_token();

	_write_c('\n');
end;

proc _compile_character_literal();
begin
	_write_z("\tli t0, \0");

	_write_c('\'');
	_advance_token(1);

	la t0, source_code_position
	lw t0, (t0)
	lb a0, (t0)
	li t1, '\\'
	bne a0, t1, .compile_character_literal_end
	
	_write_c('\\');
	_advance_token(1);

.compile_character_literal_end:
	la t0, source_code_position
	lw t0, (t0)
	lb a0, (t0)
	_write_c();

	_write_c('\'');
	_write_c('\n');

	_advance_token(2);

end;

proc _compile_variable_expression();
begin
	_compile_designator();
	_write_z("\tlw t0, (t0)\n\0");
end;

proc _compile_address_expression();
begin
	# Skip the "@" sign.
	_advance_token(1);
	_compile_designator();

end;

proc _compile_negate_expression();
begin
	# Skip the "-" sign.
	_advance_token(1);
	_compile_term();

	_write_z("\tneg t0, t0\n\0");
end;

proc _compile_not_expression();
begin
	# Skip the "~" sign.
	_advance_token(1);
	_compile_term();

	_write_z("\tnot t0, t0\n\0");
end;

proc _compile_string_literal();
begin
	_string_length(source_code_position);
	sw a0, 0(sp)

	_add_string(source_code_position);
	sw a0, 4(sp)

	_advance_token(v0 + 2);
	_write_z("\tla t0, strings\n\0");

	_write_z("\tli t1, \0");
	_write_i(v4);
	_write_c('\n');

	_write_z("\tadd t0, t0, t1\n\0");
end;

proc _compile_term();
begin
	la t0, source_code_position
	lw t0, (t0)
	lb a0, (t0)
	sw a0, 0(sp)

	li t1, '\''
	beq a0, t1, .compile_term_character_literal

	li t1, '@'
	beq a0, t1, .compile_term_address

	li t1, '-'
	beq a0, t1, .compile_term_negation

	li t1, '~'
	beq a0, t1, .compile_term_not

	li t1, '"'
	beq a0, t1, .compile_term_string_literal

	_is_digit(v0);
	bnez a0, .compile_term_integer_literal

	goto .compile_term_variable;

.compile_term_character_literal:
	_compile_character_literal();
	goto .compile_term_end;

.compile_term_integer_literal:
	_compile_integer_literal();
	goto .compile_term_end;

.compile_term_address:
	_compile_address_expression();
	goto .compile_term_end;

.compile_term_negation:
	_compile_negate_expression();
	goto .compile_term_end;

.compile_term_not:
	_compile_not_expression();
	goto .compile_term_end;

.compile_term_string_literal:
	_compile_string_literal();
	goto .compile_term_end;

.compile_term_variable:
	_compile_variable_expression();
	goto .compile_term_end;

.compile_term_end:
end;

proc _compile_binary_rhs();
begin
	# Skip the whitespace after the binary operator.
	_advance_token(1);
	_compile_term();

	# Load the left expression from the stack;
	_write_z("\tlw t1, 24(sp)\n\0");
end;

proc _compile_expression();
begin
	_compile_term();

	la t0, source_code_position
	lw t0, (t0)
	lb a0, (t0)

	li t1, ' '
	bne a0, t1, .compile_expression_end

	# It is a binary expression.

	# Save the value of the left expression on the stack.
	_write_z("sw t0, 24(sp)\n\0");

	# Skip surrounding whitespace in front of the operator.
	_advance_token(1);
	la t0, source_code_position
	lw t0, (t0)
	lb t0, (t0)

	li t1, '+'
	beq t0, t1, .compile_expression_add

	li t1, '*'
	beq t0, t1, .compile_expression_mul

	li t1, '&'
	beq t0, t1, .compile_expression_and

	li t1, 'o'
	beq t0, t1, .compile_expression_or

	li t1, 'x'
	beq t0, t1, .compile_expression_xor

	li t1, '='
	beq t0, t1, .compile_expression_equals

	li t1, '<'
	beq t0, t1, .compile_expression_less

	li t1, '>'
	beq t0, t1, .compile_expression_greater

	# Unknown binary operator.
	unimp

.compile_expression_add:
	_advance_token(1);
	_compile_binary_rhs();

	# Execute the operation.
	_write_z("add t0, t0, t1\n\0");

	goto .compile_expression_end;

.compile_expression_mul:
	_advance_token(1);
	_compile_binary_rhs();

	# Execute the operation.
	_write_z("\tmul t0, t0, t1\n\0");

	goto .compile_expression_end;

.compile_expression_and:
	_advance_token(1);
	_compile_binary_rhs();

	# Execute the operation.
	_write_z("\tand t0, t0, t1\n\0");

	goto .compile_expression_end;

.compile_expression_or:
	_advance_token(2);
	_compile_binary_rhs();

	# Execute the operation.
	_write_z("or t0, t0, t1\n\0");

	goto .compile_expression_end;

.compile_expression_xor:
	_advance_token(3);
	_compile_binary_rhs();

	# Execute the operation.
	_write_z("xor t0, t0, t1\n\0");

	goto .compile_expression_end;

.compile_expression_equals:
	_advance_token(1);
	_compile_binary_rhs();

	# Execute the operation.
	_write_z("xor t0, t0, t1\nseqz t0, t0\n\0");

	goto .compile_expression_end;

.compile_expression_less:
	_advance_token(1);
	la t0, source_code_position
	lw t0, (t0)
	lb t0, (t0)

	li t1, '>'
	beq t0, t1, .compile_expression_not_equal

	li t1, '='
	beq t0, t1, .compile_expression_less_equal

	_compile_binary_rhs();

	# Execute the operation.
	_write_z("slt t0, t0, t1\n\0");

	goto .compile_expression_end;

.compile_expression_not_equal:
	_advance_token(1);
	_compile_binary_rhs();

	# Execute the operation.
	_write_z("\txor t0, t0, t1\nsnez t0, t0\n\0");

	goto .compile_expression_end;

.compile_expression_less_equal:
	_advance_token(1);
	_compile_binary_rhs();

	# Execute the operation.
	_write_z("\tslt t0, t0, t1\nxori t0, t0, 1\n\0");

	goto .compile_expression_end;

.compile_expression_greater:
	_advance_token(1);
	la t0, source_code_position
	lw t0, (t0)
	lb t0, (t0)

	li t1, '='
	beq t0, t1, .compile_expression_greater_equal

	_compile_binary_rhs();

	# Execute the operation.
	_write_z("\tslt t0, t1, t0\n\0");

	goto .compile_expression_end;

.compile_expression_greater_equal:
	_advance_token(1);
	_compile_binary_rhs();

	# Execute the operation.
	_write_z("\tslt t0, t1, t0\nxori t0, t0, 1\n\0");

	goto .compile_expression_end;

.compile_expression_end:
end;

proc _compile_call();
begin
	# Stack variables:
	# v0 - Procedure name length.
	# v4 - Procedure name pointer. 
	# v8 - Argument count.

	_read_token();
	sw a0, 0(sp)
	v4 := source_code_position;
	v8 := 0;

	# Skip the identifier and left paren.
	_advance_token(v0 + 1);

	la t0, source_code_position
	lw t0, (t0)
	lb t0, (t0)

	li t1, ')'
	beq t0, t1, .compile_call_finalize

.compile_call_loop:
	_compile_expression();

	# Save the argument on the stack.
	_write_z("\tsw t0, \0");

	# Calculate the stack offset: 116 - (4 * argument_counter)
	v12 := v8 * 4;
	v12 := 116 + -v12;
	_write_i(v12);

	_write_z("(sp)\n\0");

	# Add one to the argument counter.
	v8 := v8 + 1;

	la t0, source_code_position
	lw t0, (t0)
	lb t0, (t0)

	li t1, ','
	bne t0, t1, .compile_call_finalize

	_advance_token(2);
	goto .compile_call_loop;

.compile_call_finalize:
	# Load the argument from the stack.

	lw t0, 8(sp)
	beqz t0, .compile_call_end

	# Decrement the argument counter.
	v8 := v8 + -1;

	_write_z("\tlw a\0");
	_write_i(v8);

	_write_z(", \0");

	# Calculate the stack offset: 116 - (4 * argument_counter)
	v12 := v8 * 4;
	v12 := 116 + -v12;
	_write_i(v12);

	_write_z("(sp)\n\0");

	goto .compile_call_finalize;

.compile_call_end:
	_write_z("\tcall \0");
	_write_s(v4, v0);

	# Skip the right paren.
	_advance_token(1);
end;

proc _compile_goto();
begin
	_advance_token(5);

	_read_token();
	sw a0, 0(sp)

	_write_z("\tj \0");

	_write_token(v0);
	_advance_token();
end;

proc _compile_local_designator();
begin
	# Skip "v" in the local variable name.
	_advance_token(1);
	_write_z("\t addi t0, sp, \0");

	# Read local variable stack offset and save it.
	_read_token();
	_write_token();
	_advance_token();
	_write_c('\n');
end;

proc _compile_global_designator();
begin
	_write_z("\tla t0, \0");

	_read_token();
	_write_token();
	_advance_token();

	_write_c('\n');
end;

proc _compile_designator();
begin
	la t0, source_code_position
	lw t0, (t0)
	lb a0, (t0)

	li t1, 'v'
	beq a0, t1, .compile_designator_local

	goto .compile_designator_global;

.compile_designator_local:
	_compile_local_designator();
	goto .compile_designator_end;

.compile_designator_global:
	_compile_global_designator();
	goto .compile_designator_end;

.compile_designator_end:
end;

proc _compile_assignment();
begin
	_compile_designator();

	# Save the assignee address on the stack.
	_write_z("\tsw t0, 20(sp)\n\0");

	# Skip the assignment sign (:=) with surrounding whitespaces.
	_advance_token(4);

	# Compile the assignment.
	_compile_expression();

	_write_z("\tlw t1, 20(sp)\nsw t0, (t1)\n\0");
end;

proc _compile_return_statement();
begin
	# Skip "return" keyword and whitespace after it.
	_advance_token(7);
	_compile_expression();

	_write_z("mv a0, t0\n\0");
end;

proc _compile_statement();
begin
	# This is a call if the statement starts with an underscore.
	la t0, source_code_position
	lw t0, (t0)
	# First character after alignment tab.
	addi t0, t0, 1
	lb t0, (t0)
	
	li t1, '_'
	beq t0, t1, .compile_statement_call

	li t1, 'g'
	beq t0, t1, .compile_statement_goto

	li t1, 'v'
	beq t0, t1, .compile_statement_assignment

	# keyword_ret contains "\tret", so it's 4 bytes long.
	_memcmp(source_code_position, "\treturn", 7);
	beqz a0, .compile_statement_return

	_compile_line();
	goto .compile_statement_end;

.compile_statement_call:
	_advance_token(1);
	_compile_call();

	goto .compile_statement_semicolon;

.compile_statement_goto:
	_advance_token(1);
	_compile_goto();

	goto .compile_statement_semicolon;

.compile_statement_assignment:
	_advance_token(1);
	_compile_assignment();

	goto .compile_statement_semicolon;

.compile_statement_return:
	_advance_token(1);
	_compile_return_statement();
	_write_c('\n');

	goto .compile_statement_end;

.compile_statement_semicolon:
	_advance_token(2);
	_write_c('\n');

.compile_statement_end:
end;

proc _compile_procedure_body();
begin
.compile_procedure_body_loop:
	# 3 is "end" length.
	_memcmp(source_code_position, "end", 3);

	beqz a0, .compile_procedure_body_epilogue

	_compile_statement();
	goto .compile_procedure_body_loop;

.compile_procedure_body_epilogue:
end;

# Writes a regster name to the standard output.
#
# Parameters:
# a0 - Register character.
# a1 - Register number.
proc _write_register();
begin
	_write_c(v88);
	v84 := v84 + '0';
	_write_c(v84);
end;

proc _compile_procedure_prologue();
begin
	_write_z("\taddi sp, sp, -128\n\tsw ra, 124(sp)\n\tsw s0, 120(sp)\n\taddi s0, sp, 128\n\0");
	v0 := 0;

.compile_procedure_prologue_loop:
	_write_z("\tsw a\0");
	_write_i(v0);
	_write_z(", \0");

	# Calculate the stack offset: 88 - (4 * parameter_counter)
	v4 := v0 * 4;
	v4 := 88 + -v4;
	_write_i(v4);

	_write_z("(sp)\n\0");

	v0 := v0 + 1;
	lw a0, 0(sp)

	li t0, 8
	bne a0, t0, .compile_procedure_prologue_loop
end;

proc _compile_procedure();
begin
	# Skip "proc ".
	_advance_token(5);

	_read_token();
	sw a0, 0(sp) # Save the procedure name length.

	# Write .type _procedure_name, @function.
	_write_z(".type \0");

	_write_token(v0);
	_write_z(", @function\n\0");

	# Write procedure label, _procedure_name:
	_write_token(v0);
	_write_z(":\n\0");

	# Skip the function name and trailing parens, semicolon, "begin" and newline.
	_advance_token(v0 + 10);

	_compile_procedure_prologue();
	_compile_procedure_body();

	# Write the epilogue.
	_write_z("\tlw ra, 124(sp)\n\tlw s0, 120(sp)\n\taddi sp, sp, 128\n\tret\n\0");

	# Skip the "end" keyword, semicolon and newline.
	_advance_token(5);
end;

proc _skip_newlines();
begin
	# Skip newlines.
	la t0, source_code_position
	lw t1, (t0)

.skip_newlines_loop:
	lb t2, (t1)
	li t3, '\n'
	bne t2, t3, .skip_newlines_end
	beqz t2, .skip_newlines_end

	addi t1, t1, 1
	sw t1, (t0)

	goto .skip_newlines_loop;

.skip_newlines_end:
end;

# Skip newlines and comments.
proc _skip_empty_lines();
begin
.skip_empty_lines_loop:
	la t0, source_code_position
	lw t0, (t0)
	lb t0, (t0)

	li t1, '#'
	beq t0, t1, .skip_empty_lines_comment

	li t1, '\n'
	beq t0, t1, .skip_empty_lines_newline

	goto .skip_empty_lines_end;

.skip_empty_lines_comment:
	_skip_comment();
	goto .skip_empty_lines_loop;

.skip_empty_lines_newline:
	_advance_token(1);
	goto .skip_empty_lines_loop;

.skip_empty_lines_end:
end;

proc _compile_global_initializer();
begin
	la t0, source_code_position
	lw t0, (t0)
	lb t0, (t0)

	li t1, '"'
	beq t0, t1, .compile_global_initializer_string

	li t1, 'S'
	beq t0, t1, .compile_global_initializer_record

	li t1, '@'
	beq t0, t1, .compile_global_initializer_pointer

	la a0, source_code_position
	lw a0, (a0)
	lb a0, (a0)
	_is_digit();
	bnez a0, .compile_global_initializer_number

	unimp

.compile_global_initializer_pointer:
	# Skip @.
	_advance_token(1);
	_write_z("\n\t.word \0");
	_read_token();
	_write_token();
	_advance_token();

	goto .compile_global_initializer_end;

.compile_global_initializer_number:
	_write_z("\n\t.word \0");
	_read_token();
	_write_token();
	_advance_token(1);

	goto .compile_global_initializer_end;

.compile_global_initializer_record:
	# Skip "S(".
	_advance_token(2);

	la t0, source_code_position
	lw t0, (t0)
	lb t0, (t0)
	li t1, ')'
	beq t0, t1, .compile_global_initializer_closing

.compile_global_initializer_loop:
	_compile_global_initializer();

	la t0, source_code_position
	lw t0, (t0)
	lb t0, (t0)
	li t1, ')'
	beq t0, t1, .compile_global_initializer_closing

	# Skip comma and whitespace after it.
	_advance_token(2);

	goto .compile_global_initializer_loop;

.compile_global_initializer_closing:
	# Skip ")"
	_advance_token(1);

	goto .compile_global_initializer_end;

.compile_global_initializer_string:
	_write_z("\n\t.word strings + \0");
	_string_length(source_code_position);
	sw a0, 4(sp)

	_add_string(source_code_position);
	_write_i();

	# Skip the quoted string.
	_advance_token(v4 + 2);

	goto .compile_global_initializer_end;

.compile_global_initializer_end:
end;

proc _compile_constant_declaration();
begin
	_read_token();
	sw a0, 0(sp)

	_write_z(".type \0");
	_write_token(v0);
	_write_z(", @object\n\0");

	_write_token(v0);
	_write_c(':');

	# Skip the constant name with assignment sign and surrounding whitespaces.
	_advance_token(v0 + 4);
	_compile_global_initializer();
	# Skip semicolon and newline.
	_advance_token(2);
	_write_c('\n');
end;

proc _compile_const_part();
begin
	_skip_empty_lines();

	_memcmp(source_code_position, "const\0", 5);
	bnez a0, .compile_const_part_end

	# Skip "const" with the newline after it.
	_advance_token(6);
	_write_z(".section .rodata # Compiled from const section.\n\n\0");

.compile_const_part_loop:
	_skip_empty_lines();

	la t0, source_code_position
	lw t0, (t0)
	lb t0, (t0)

	# If the character at the line beginning is not indentation,
	# it is probably the next code section.
	li t1, '\t'
	bne t0, t1, .compile_const_part_end

	_advance_token(1);

	la t0, source_code_position
	lw t0, (t0)
	lb t0, (t0)
	li t1, '#'
	beq t0, t1, .compile_const_part_loop

	_compile_constant_declaration();
	goto .compile_const_part_loop;

.compile_const_part_end:
end;

proc _compile_variable_declaration();
begin
	_read_token();
	sw a0, 0(sp)

	_write_z(".type \0");
	_write_token(v0);
	_write_z(", @object\n\0");

	_write_token(v0);
	_write_c(':');

	# Skip the variable name and colon with space before the type.
	_advance_token(v0 + 2);

	# Skip the type name.
	_read_token();
	_advance_token();

	la t0, source_code_position
	lw t0, (t0)
	lb t0, (t0)
	li t1, ' '
	beq t0, t1, .compile_variable_declaration_initializer

	# Else we assume this is a zeroed 81920 bytes big array.
	_write_z(" .zero 81920\0");
	goto .compile_variable_declaration_finalize;

.compile_variable_declaration_initializer:
	# Skip the assignment sign with surrounding whitespaces.
	_advance_token(4);
	_compile_global_initializer();
	goto .compile_variable_declaration_finalize;

.compile_variable_declaration_finalize:
	# Skip semicolon and newline.
	_advance_token(2);
	_write_c('\n');
end;

proc _compile_var_part();
begin
	_memcmp(source_code_position, "var\0", 3);
	bnez a0, .compile_var_part_end

	# Skip "var" and newline.
	_advance_token(4);
	_write_z(".section .data\n\0");

.compile_var_part_loop:
	la t0, source_code_position
	lw t0, (t0)
	lb t0, (t0)

	li t1, 'p'
	beq t0, t1, .compile_var_part_end

	li t1, '\t'
	beq t0, t1, .compile_var_part_declaration

	_compile_line();
	goto .compile_var_part_loop;

.compile_var_part_declaration:
	_advance_token(1);
	_compile_variable_declaration();
	goto .compile_var_part_loop;

.compile_var_part_end:
end;

# Process the source code and print the generated code.
proc _compile_module();
begin
	_compile_const_part();
	_write_z(".section .bss\n\0");

.compile_module_bss:
	la t0, source_code_position
	lw t0, (t0)
	lb t0, (t0)
	li t1, 'v'
	beq t0, t1, .compile_module_code

	li t1, 'p'
	beq t0, t1, .compile_module_code

	_compile_line();
	goto .compile_module_bss;

.compile_module_code:
	_compile_var_part();
	_write_z(".section .text\n\0");
.compile_module_loop:
	_skip_newlines();

	la t0, source_code_position
	lw t0, (t0)
	lb t0, (t0)
	beqz t0, .compile_module_end
	li t1, '#'
	beq t0, t1, .compile_module_comment

	# 8 is ".section" length.
	_memcmp(source_code_position, ".section", 8);
	beqz a0, .compile_module_section

	# 5 is "proc " length. Space is needed to distinguish from "procedure".
	_memcmp(source_code_position, "proc ", 5);
	beqz a0, .compile_module_procedure

	# 6 is ".globl" length.
	_memcmp(source_code_position, ".globl", 6);
	beqz a0, .compile_module_global

	# Not a known token, exit.
	goto .compile_module_end;

.compile_module_section:
	_compile_section();

	goto .compile_module_loop;

.compile_module_global:
	_compile_line();

	goto .compile_module_loop;

.compile_module_comment:
	_skip_comment();

	goto .compile_module_loop;

.compile_module_procedure:
	_compile_procedure();

	goto .compile_module_loop;

.compile_module_end:
end;

proc _compile();
begin
	_write_z(".globl _start\n\n\0");
	_compile_module();

	_write_z(".section .rodata\n.type strings, @object\nstrings: .ascii \0");
	_write_c('"');

	la t0, compiler_strings
	sw t0, 0(sp)

.compile_loop:
	lw t0, 0(sp)
	la t1, compiler_strings_position
	lw t1, (t1)
	bge t0, t1, .compile_end

	lb a0, (t0)

	addi t0, t0, 1
	sw t0, 0(sp)

	_write_c();

	j .compile_loop

.compile_end:
	_write_c('"');
	_write_c('\n');
end;

# Terminates the program. a0 contains the return code.
#
# Parameters:
# a0 - Status code.
proc _exit();
begin
	li a7, 93 # SYS_EXIT
	ecall
end;

# Entry point.
proc _start();
begin
	# Read the source from the standard input.
	# Second argument is buffer size. Modifying update the source_code definition.
	_read_file(@source_code, 81920);
	_compile();

	_exit(0);

end;