def codegen(self, builder: IRBuilder, ctx: Context):
self.check_children()
body = ctx.function.append_basic_block()
condition = ctx.function.append_basic_block()
end = ctx.function.append_basic_block()
ctx.start_scope()
builder.branch(body)
builder.position_at_end(body)
ctx.label = body
body_node = self.children[1]
body_node.codegen(builder, ctx)
# Alla fine faccio il collect
collect_fun = ctx.get_function(fn.collect_function)
builder.call(collect_fun, [])
builder.branch(condition)
ctx.end_scope()
builder.position_at_end(condition)
ctx.label = condition
condition_node = self.children[0]
last = condition_node.codegen(builder, ctx)
cond_res = builder.icmp_unsigned("!=", last,
ir.Constant(ir.IntType(8), 0))
builder.cbranch(cond_res, body, end)
builder.position_at_end(end)
ctx.label = end
def codegen(self, builder: IRBuilder, ctx: Context):
self.check_children()
l = len(self.children)
condition = ctx.function.append_basic_block()
true = ctx.function.append_basic_block()
false = None
if l == 3:
false = ctx.function.append_basic_block()
end = ctx.function.append_basic_block()
if false is None:
false = end
builder.branch(condition)
builder.position_at_end(condition)
condition_node = self.children[0]
last = condition_node.codegen(builder, ctx)
cond_res = builder.icmp_unsigned("!=", last,
ir.Constant(ir.IntType(8), 0))
builder.cbranch(cond_res, true, false)
ctx.start_scope()
builder.position_at_end(true)
ctx.label = true
true_node = self.children[1]
true_node.codegen(builder, ctx)
builder.branch(end)
ctx.end_scope()
if l == 3:
ctx.start_scope()
builder.position_at_end(false)
ctx.label = false
false_node = self.children[2]
false_node.codegen(builder, ctx)
builder.branch(end)
ctx.end_scope()
builder.position_at_end(end)
ctx.label = end
# Alla fine faccio il collect
collect_fun = ctx.get_function(fn.collect_function)
builder.call(collect_fun, [])
class GenerateLLVM(object):
def __init__(self, name='module'):
# Perform the basic LLVM initialization. You need the following parts:
#
# 1. A top-level Module object
# 2. A Function instance in which to insert code
# 3. A Builder instance to generate instructions
#
# Note: For project 5, we don't have any user-defined
# functions so we're just going to emit all LLVM code into a top
# level function void main() { ... }. This will get changed later.
self.module = Module(name)
# self.function = Function(self.module,
# FunctionType(void_type, []),
# name='main')
# self.block = self.function.append_basic_block('entry')
self.block = None
# self.builder = IRBuilder(self.block)
self.builder = None
# Dictionary that holds all of the global variable/function
# declarations.
# Any declaration in the Gone source code is going to get an entry here
# self.vars = {}
self.globals = {}
self.locals = {}
# Dictionary that holds all of the temporary variables created in
# the intermediate code. For example, if you had an expression
# like this:
#
# a = b + c*d
#
# The corresponding intermediate code might look like this:
#
# ('load_int', 'b', 'int_1')
# ('load_int', 'c', 'int_2')
# ('load_int', 'd', 'int_3')
# ('mul_int', 'int_2','int_3','int_4')
# ('add_int', 'int_1','int_4','int_5')
# ('store_int', 'int_5', 'a')
#
# The self.temp dictionary below is used to map names such as 'int_1',
# 'int_2' to their corresponding LLVM values. Essentially, every time
# you make anything in LLVM, it gets stored here.
self.temps = {}
# Initialize the runtime library functions (see below)
self.declare_runtime_library()
self.last_branch = None
def start_function(self, name, rettypename, parmtypenames):
rettype = typemap[rettypename]
parmtypes = [typemap[pname] for pname in parmtypenames]
# Type.function(rettype, parmtypes, False)
func_type = FunctionType(rettype, parmtypes)
# Create the function for which we're generating code
# Function.new(self.module, func_type, name)
self.function = Function(self.module, func_type, name=name)
# Make the builder and entry block
self.block = self.function.append_basic_block("entry")
self.builder = IRBuilder(self.block)
# Make the exit block
self.exit_block = self.function.append_basic_block("exit")
# Clear the local vars and temps
self.locals = {}
self.temps = {}
# Make the return variable
if rettype is not void_type:
self.locals['return'] = self.builder.alloca(rettype, name="return")
# Put an entry in the globals
self.globals[name] = self.function
def new_basic_block(self, name=''):
self.builder = IRBuilder(self.block.instructions)
return self.function.append_basic_block(name)
def declare_runtime_library(self):
# Certain functions such as I/O and string handling are often easier
# to implement in an external C library. This method should make
# the LLVM declarations for any runtime functions to be used
# during code generation. Please note that runtime function
# functions are implemented in C in a separate file gonert.c
self.runtime = {}
# Declare printing functions
self.runtime['_print_int'] = Function(self.module,
FunctionType(
void_type, [int_type]),
name="_print_int")
self.runtime['_print_float'] = Function(self.module,
FunctionType(
void_type, [float_type]),
name="_print_float")
self.runtime['_print_bool'] = Function(self.module,
FunctionType(
void_type, [int_type]),
name="_print_bool")
def generate_code(self, ircode):
# Given a sequence of SSA intermediate code tuples, generate LLVM
# instructions using the current builder (self.builder). Each
# opcode tuple (opcode, args) is dispatched to a method of the
# form self.emit_opcode(args)
for opcode, *args in ircode:
if hasattr(self, 'emit_' + opcode):
getattr(self, 'emit_' + opcode)(*args)
else:
print('Warning: No emit_' + opcode + '() method')
# Add a return statement. Note, at this point, we don't really have
# user-defined functions so this is a bit of hack--it may be removed
# later.
# self.builder.ret_void()
def terminate(self):
# Add a return statement. This connects the last block to the exit
# block.
# The return statement is then emitted
if self.last_branch != self.block:
self.builder.branch(self.exit_block)
self.builder.position_at_end(self.exit_block)
if 'return' in self.locals:
self.builder.ret(self.builder.load(self.locals['return']))
else:
self.builder.ret_void()
def add_block(self, name):
# Add a new block to the existing function
return self.function.append_basic_block(name)
def set_block(self, block):
# Sets the current block for adding more code
self.block = block
self.builder.position_at_end(block)
def cbranch(self, testvar, true_block, false_block):
self.builder.cbranch(self.temps[testvar], true_block, false_block)
def branch(self, next_block):
if self.last_branch != self.block:
self.builder.branch(next_block)
self.last_branch = self.block
# ----------------------------------------------------------------------
# Opcode implementation. You must implement the opcodes. A few
# sample opcodes have been given to get you started.
# ----------------------------------------------------------------------
# Creation of literal values. Simply define as LLVM constants.
def emit_literal_int(self, value, target):
self.temps[target] = Constant(int_type, value)
def emit_literal_float(self, value, target):
self.temps[target] = Constant(float_type, value)
def emit_literal_bool(self, value, target):
self.temps[target] = Constant(bool_type, value)
# def emit_literal_string(self, value, target):
# self.temps[target] = Constant(string_type, value)
# Allocation of variables. Declare as global variables and set to
# a sensible initial value.
def emit_alloc_int(self, name):
var = self.builder.alloca(int_type, name=name)
var.initializer = Constant(int_type, 0)
self.locals[name] = var
def emit_alloc_float(self, name):
var = self.builder.alloca(float_type, name=name)
var.initializer = Constant(float_type, 0)
self.locals[name] = var
def emit_alloc_bool(self, name):
var = self.builder.alloca(bool_type, name=name)
var.initializer = Constant(bool_type, 0)
self.locals[name] = var
def emit_global_int(self, name):
var = GlobalVariable(self.module, int_type, name=name)
var.initializer = Constant(int_type, 0)
self.globals[name] = var
def emit_global_float(self, name):
var = GlobalVariable(self.module, float_type, name=name)
var.initializer = Constant(float_type, 0)
self.globals[name] = var
def emit_global_bool(self, name):
var = GlobalVariable(self.module, bool_type, name=name)
var.initializer = Constant(bool_type, 0)
self.globals[name] = var
# def emit_alloc_string(self, name):
# var = GlobalVariable(self.module, string_type, name=name)
# var.initializer = Constant(string_type, "")
# self.vars[name] = var
# Load/store instructions for variables. Load needs to pull a
# value from a global variable and store in a temporary. Store
# goes in the opposite direction.
def lookup_var(self, name):
if name in self.locals:
return self.locals[name]
else:
return self.globals[name]
def emit_load_int(self, name, target):
# print('LOADINT %s, %s' % (name, target))
# print('GLOBALS %s' % self.globals)
# print('LOCALS %s' % self.locals)
self.temps[target] = self.builder.load(self.lookup_var(name), target)
def emit_load_float(self, name, target):
self.temps[target] = self.builder.load(self.lookup_var(name), target)
def emit_load_bool(self, name, target):
self.temps[target] = self.builder.load(self.lookup_var(name), target)
def emit_store_int(self, source, target):
self.builder.store(self.temps[source], self.lookup_var(target))
def emit_store_float(self, source, target):
self.builder.store(self.temps[source], self.lookup_var(target))
def emit_store_bool(self, source, target):
self.builder.store(self.temps[source], self.lookup_var(target))
# Binary + operator
def emit_add_int(self, left, right, target):
self.temps[target] = self.builder.add(
self.temps[left], self.temps[right], target)
def emit_add_float(self, left, right, target):
self.temps[target] = self.builder.fadd(
self.temps[left], self.temps[right], target)
# Binary - operator
def emit_sub_int(self, left, right, target):
self.temps[target] = self.builder.sub(
self.temps[left], self.temps[right], target)
def emit_sub_float(self, left, right, target):
self.temps[target] = self.builder.fsub(
self.temps[left], self.temps[right], target)
# Binary * operator
def emit_mul_int(self, left, right, target):
self.temps[target] = self.builder.mul(
self.temps[left], self.temps[right], target)
def emit_mul_float(self, left, right, target):
self.temps[target] = self.builder.fmul(
self.temps[left], self.temps[right], target)
# Binary / operator
def emit_div_int(self, left, right, target):
self.temps[target] = self.builder.sdiv(
self.temps[left], self.temps[right], target)
def emit_div_float(self, left, right, target):
self.temps[target] = self.builder.fdiv(
self.temps[left], self.temps[right], target)
# Unary + operator
def emit_uadd_int(self, source, target):
self.temps[target] = self.builder.add(
Constant(int_type, 0),
self.temps[source],
target)
def emit_uadd_float(self, source, target):
self.temps[target] = self.builder.fadd(
Constant(float_type, 0.0),
self.temps[source],
target)
# Unary - operator
def emit_usub_int(self, source, target):
self.temps[target] = self.builder.sub(
Constant(int_type, 0),
self.temps[source],
target)
def emit_usub_float(self, source, target):
self.temps[target] = self.builder.fsub(
Constant(float_type, 0.0),
self.temps[source],
target)
# Binary < operator
def emit_lt_int(self, left, right, target):
self.temps[target] = self.builder.icmp_signed(
'<', self.temps[left], self.temps[right], target)
def emit_lt_float(self, left, right, target):
self.temps[target] = self.builder.fcmp_ordered(
'<', self.temps[left], self.temps[right], target)
# Binary <= operator
def emit_le_int(self, left, right, target):
self.temps[target] = self.builder.icmp_signed(
'<=', self.temps[left], self.temps[right], target)
def emit_le_float(self, left, right, target):
self.temps[target] = self.builder.fcmp_ordered(
'<=', self.temps[left], self.temps[right], target)
# Binary > operator
def emit_gt_int(self, left, right, target):
self.temps[target] = self.builder.icmp_signed(
'>', self.temps[left], self.temps[right], target)
def emit_gt_float(self, left, right, target):
self.temps[target] = self.builder.fcmp_ordered(
'>', self.temps[left], self.temps[right], target)
# Binary >= operator
def emit_ge_int(self, left, right, target):
self.temps[target] = self.builder.icmp_signed(
'>=', self.temps[left], self.temps[right], target)
def emit_ge_float(self, left, right, target):
self.temps[target] = self.builder.fcmp_ordered(
'>=', self.temps[left], self.temps[right], target)
# Binary == operator
def emit_eq_int(self, left, right, target):
self.temps[target] = self.builder.icmp_signed(
'==', self.temps[left], self.temps[right], target)
def emit_eq_bool(self, left, right, target):
self.temps[target] = self.builder.icmp_signed(
'==', self.temps[left], self.temps[right], target)
def emit_eq_float(self, left, right, target):
self.temps[target] = self.builder.fcmp_ordered(
'==', self.temps[left], self.temps[right], target)
# Binary != operator
def emit_ne_int(self, left, right, target):
self.temps[target] = self.builder.icmp_signed(
'!=', self.temps[left], self.temps[right], target)
def emit_ne_bool(self, left, right, target):
self.temps[target] = self.builder.icmp_signed(
'!=', self.temps[left], self.temps[right], target)
def emit_ne_float(self, left, right, target):
self.temps[target] = self.builder.fcmp_ordered(
'!=', self.temps[left], self.temps[right], target)
# Binary && operator
def emit_and_bool(self, left, right, target):
self.temps[target] = self.builder.and_(
self.temps[left], self.temps[right], target)
# Binary || operator
def emit_or_bool(self, left, right, target):
self.temps[target] = self.builder.or_(
self.temps[left], self.temps[right], target)
# Unary ! operator
def emit_not_bool(self, source, target):
self.temps[target] = self.builder.icmp_signed(
'==', self.temps[source], Constant(bool_type, 0), target)
# Print statements
def emit_print_int(self, source):
self.builder.call(self.runtime['_print_int'], [self.temps[source]])
def emit_print_float(self, source):
self.builder.call(self.runtime['_print_float'], [self.temps[source]])
def emit_print_bool(self, source):
self.builder.call(self.runtime['_print_bool'], [
self.builder.zext(self.temps[source], int_type)])
# Extern function declaration.
def emit_extern_func(self, name, rettypename, *parmtypenames):
rettype = typemap[rettypename]
parmtypes = [typemap[pname] for pname in parmtypenames]
func_type = FunctionType(rettype, parmtypes)
self.globals[name] = Function(self.module, func_type, name=name)
# Call an external function.
def emit_call_func(self, funcname, *args):
target = args[-1]
func = self.globals[funcname]
argvals = [self.temps[name] for name in args[:-1]]
self.temps[target] = self.builder.call(func, argvals)
# Function parameter declarations. Must create as local variables
def emit_parm_int(self, name, num):
var = self.builder.alloca(int_type, name=name)
self.builder.store(self.function.args[num], var)
self.locals[name] = var
def emit_parm_float(self, name, num):
var = self.builder.alloca(float_type, name=name)
self.builder.store(self.function.args[num], var)
self.locals[name] = var
def emit_parm_bool(self, name, num):
var = self.builder.alloca(bool_type, name=name)
self.builder.store(self.function.args[num], var)
self.locals[name] = var
# Return statements
def emit_return_int(self, source):
self.builder.store(self.temps[source], self.locals['return'])
self.branch(self.exit_block)
def emit_return_float(self, source):
self.builder.store(self.temps[source], self.locals['return'])
self.branch(self.exit_block)
def emit_return_bool(self, source):
self.builder.store(self.temps[source], self.locals['return'])
self.branch(self.exit_block)
def emit_return_void(self):
self.branch(self.exit_block)
class GenerateLLVM(object):
def __init__(self):
# Perform the basic LLVM initialization. You need the following parts:
#
# 1. A top-level Module object
# 2. A Function instance in which to insert code
# 3. A Builder instance to generate instructions
#
# Note: For project 5, we don't have any user-defined
# functions so we're just going to emit all LLVM code into a top
# level function void main() { ... }. This will get changed later.
self.module = Module('module')
# All globals variables and function definitions go here
self.globals = {}
self.blocks = {}
# Initialize the runtime library functions (see below)
self.declare_runtime_library()
def declare_runtime_library(self):
# Certain functions such as I/O and string handling are often easier
# to implement in an external C library. This method should make
# the LLVM declarations for any runtime functions to be used
# during code generation. Please note that runtime function
# functions are implemented in C in a separate file gonert.c
self.runtime = {}
# Declare printing functions
self.runtime['_print_int'] = Function(self.module,
FunctionType(
void_type, [int_type]),
name="_print_int")
self.runtime['_print_float'] = Function(self.module,
FunctionType(
void_type, [float_type]),
name="_print_float")
self.runtime['_print_byte'] = Function(self.module,
FunctionType(
void_type, [byte_type]),
name="_print_byte")
def generate_code(self, ir_function):
# Given a sequence of SSA intermediate code tuples, generate LLVM
# instructions using the current builder (self.builder). Each
# opcode tuple (opcode, args) is dispatched to a method of the
# form self.emit_opcode(args)
# print(ir_function)
self.function = Function(
self.module,
FunctionType(LLVM_TYPE_MAPPING[ir_function.return_type], [
LLVM_TYPE_MAPPING[ptype] for _, ptype in ir_function.parameters
]),
name=ir_function.name)
self.block = self.function.append_basic_block('entry')
self.builder = IRBuilder(self.block)
# Save the function as a global to be referenced later on another
# function
self.globals[ir_function.name] = self.function
# All local variables are stored here
self.locals = {}
self.vars = ChainMap(self.locals, self.globals)
# Dictionary that holds all of the temporary registers created in
# the intermediate code.
self.temps = {}
# Setup the function parameters
for n, (pname, ptype) in enumerate(ir_function.parameters):
self.vars[pname] = self.builder.alloca(LLVM_TYPE_MAPPING[ptype],
name=pname)
self.builder.store(self.function.args[n], self.vars[pname])
# Allocate the return value and return_block
if ir_function.return_type:
self.vars['return'] = self.builder.alloca(
LLVM_TYPE_MAPPING[ir_function.return_type], name='return')
self.return_block = self.function.append_basic_block('return')
for opcode, *args in ir_function.code:
if hasattr(self, 'emit_' + opcode):
getattr(self, 'emit_' + opcode)(*args)
else:
print('Warning: No emit_' + opcode + '() method')
if not self.block.is_terminated:
self.builder.branch(self.return_block)
self.builder.position_at_end(self.return_block)
self.builder.ret(self.builder.load(self.vars['return'], 'return'))
def get_block(self, block_name):
block = self.blocks.get(block_name)
if block is None:
block = self.function.append_basic_block(block_name)
self.blocks[block_name] = block
return block
# ----------------------------------------------------------------------
# Opcode implementation. You must implement the opcodes. A few
# sample opcodes have been given to get you started.
# ----------------------------------------------------------------------
# Creation of literal values. Simply define as LLVM constants.
def emit_MOV(self, value, target, val_type):
self.temps[target] = Constant(val_type, value)
emit_MOVI = partialmethod(emit_MOV, val_type=int_type)
emit_MOVF = partialmethod(emit_MOV, val_type=float_type)
emit_MOVB = partialmethod(emit_MOV, val_type=byte_type)
# Allocation of GLOBAL variables. Declare as global variables and set to
# a sensible initial value.
def emit_VAR(self, name, var_type):
var = GlobalVariable(self.module, var_type, name=name)
var.initializer = Constant(var_type, 0)
self.globals[name] = var
emit_VARI = partialmethod(emit_VAR, var_type=int_type)
emit_VARF = partialmethod(emit_VAR, var_type=float_type)
emit_VARB = partialmethod(emit_VAR, var_type=byte_type)
# Allocation of LOCAL variables. Declare as local variables and set to
# a sensible initial value.
def emit_ALLOC(self, name, var_type):
self.locals[name] = self.builder.alloca(var_type, name=name)
emit_ALLOCI = partialmethod(emit_ALLOC, var_type=int_type)
emit_ALLOCF = partialmethod(emit_ALLOC, var_type=float_type)
emit_ALLOCB = partialmethod(emit_ALLOC, var_type=byte_type)
# Load/store instructions for variables. Load needs to pull a
# value from a global variable and store in a temporary. Store
# goes in the opposite direction.
def emit_LOADI(self, name, target):
self.temps[target] = self.builder.load(self.vars[name], name=target)
emit_LOADF = emit_LOADI
emit_LOADB = emit_LOADI
def emit_STOREI(self, source, target):
self.builder.store(self.temps[source], self.vars[target])
emit_STOREF = emit_STOREI
emit_STOREB = emit_STOREI
# Binary + operator
def emit_ADDI(self, left, right, target):
self.temps[target] = self.builder.add(self.temps[left],
self.temps[right],
name=target)
def emit_ADDF(self, left, right, target):
self.temps[target] = self.builder.fadd(self.temps[left],
self.temps[right],
name=target)
# Binary - operator
def emit_SUBI(self, left, right, target):
self.temps[target] = self.builder.sub(self.temps[left],
self.temps[right],
name=target)
def emit_SUBF(self, left, right, target):
self.temps[target] = self.builder.fsub(self.temps[left],
self.temps[right],
name=target)
# Binary * operator
def emit_MULI(self, left, right, target):
self.temps[target] = self.builder.mul(self.temps[left],
self.temps[right],
name=target)
def emit_MULF(self, left, right, target):
self.temps[target] = self.builder.fmul(self.temps[left],
self.temps[right],
name=target)
# Binary / operator
def emit_DIVI(self, left, right, target):
self.temps[target] = self.builder.sdiv(self.temps[left],
self.temps[right],
name=target)
def emit_DIVF(self, left, right, target):
self.temps[target] = self.builder.fdiv(self.temps[left],
self.temps[right],
name=target)
# Print statements
def emit_PRINT(self, source, runtime_name):
self.builder.call(self.runtime[runtime_name], [self.temps[source]])
emit_PRINTI = partialmethod(emit_PRINT, runtime_name="_print_int")
emit_PRINTF = partialmethod(emit_PRINT, runtime_name="_print_float")
emit_PRINTB = partialmethod(emit_PRINT, runtime_name="_print_byte")
def emit_CMPI(self, operator, left, right, target):
tmp = self.builder.icmp_signed(operator, self.temps[left],
self.temps[right], 'tmp')
# LLVM compares produce a 1-bit integer as a result. Since our IRcode using integers
# for bools, need to sign-extend the result up to the normal int_type to continue
# with further processing (otherwise you'll get a LLVM type error).
self.temps[target] = self.builder.zext(tmp, int_type, target)
def emit_CMPF(self, operator, left, right, target):
tmp = self.builder.fcmp_ordered(operator, self.temps[left],
self.temps[right], 'tmp')
self.temps[target] = self.builder.zext(tmp, int_type, target)
emit_CMPB = emit_CMPI
# Logical ops
def emit_AND(self, left, right, target):
self.temps[target] = self.builder.and_(self.temps[left],
self.temps[right], target)
def emit_OR(self, left, right, target):
self.temps[target] = self.builder.or_(self.temps[left],
self.temps[right], target)
def emit_XOR(self, left, right, target):
self.temps[target] = self.builder.xor(self.temps[left],
self.temps[right], target)
def emit_LABEL(self, lbl_name):
self.block = self.get_block(lbl_name)
self.builder.position_at_end(self.block)
def emit_BRANCH(self, dst_label):
if not self.block.is_terminated:
self.builder.branch(self.get_block(dst_label))
def emit_CBRANCH(self, test_target, true_label, false_label):
true_block = self.get_block(true_label)
false_block = self.get_block(false_label)
testvar = self.temps[test_target]
self.builder.cbranch(self.builder.trunc(testvar, IntType(1)),
true_block, false_block)
def emit_RET(self, register):
self.builder.store(self.temps[register], self.vars['return'])
self.builder.branch(self.return_block)
def emit_CALL(self, func_name, *registers):
# print(self.globals)
args = [self.temps[r] for r in registers[:-1]]
target = registers[-1]
self.temps[target] = self.builder.call(self.globals[func_name], args)
class GenerateLLVM(object):
def __init__(self):
# Perform the basic LLVM initialization. You need the following parts:
#
# 1. A top-level Module object
# 2. A Function instance in which to insert code
# 3. A Builder instance to generate instructions
#
# Note: For project 5, we don't have any user-defined
# functions so we're just going to emit all LLVM code into a top
# level function void main() { ... }. This will get changed later.
self.module = Module('module')
self.globals = {}
# Initialize the runtime library functions (see below)
self.declare_runtime_library()
def generate_function(self, name, return_type, arg_types, arg_names,
ircode):
self.function = Function(self.module,
FunctionType(return_type, arg_types),
name=name)
self.globals[name] = self.function
self.block = self.function.append_basic_block('entry')
self.builder = IRBuilder(self.block)
self.locals = {}
self.vars = ChainMap(self.locals, self.globals)
# Have to declare local variables for holding function arguments
for n, (name, ty) in enumerate(zip(arg_names, arg_types)):
self.locals[name] = self.builder.alloca(ty, name=name)
self.builder.store(self.function.args[n], self.locals[name])
# Dictionary that holds all of the temporary registers created in
# the intermediate code.
self.temps = {}
# Make the exit block for function return
self.return_block = self.function.append_basic_block('return')
if return_type != void_type:
self.return_var = self.builder.alloca(return_type, name='return')
self.generate_code(ircode)
if not self.block.is_terminated:
self.builder.branch(self.return_block)
self.builder.position_at_end(self.return_block)
if return_type != void_type:
self.builder.ret(self.builder.load(self.return_var))
else:
self.builder.ret_void()
def declare_runtime_library(self):
# Certain functions such as I/O and string handling are often easier
# to implement in an external C library. This method should make
# the LLVM declarations for any runtime functions to be used
# during code generation. Please note that runtime function
# functions are implemented in C in a separate file gonert.c
self.runtime = {}
# Declare printing functions
self.runtime['_print_int'] = Function(self.module,
FunctionType(
void_type, [int_type]),
name="_print_int")
self.runtime['_print_float'] = Function(self.module,
FunctionType(
void_type, [float_type]),
name="_print_float")
self.runtime['_print_byte'] = Function(self.module,
FunctionType(
void_type, [byte_type]),
name="_print_byte")
def generate_code(self, ircode):
# Given a sequence of SSA intermediate code tuples, generate LLVM
# instructions using the current builder (self.builder). Each
# opcode tuple (opcode, args) is dispatched to a method of the
# form self.emit_opcode(args)
# Collect and create all basic blocks
labels = [instr[1] for instr in ircode if instr[0] == 'LABEL']
self.basicblocks = {
name: self.function.append_basic_block(name)
for name in labels
}
for opcode, *args in ircode:
if hasattr(self, 'emit_' + opcode):
getattr(self, 'emit_' + opcode)(*args)
else:
print('Warning: No emit_' + opcode + '() method')
# ----------------------------------------------------------------------
# Opcode implementation. You must implement the opcodes. A few
# sample opcodes have been given to get you started.
# ----------------------------------------------------------------------
# Creation of literal values. Simply define as LLVM constants.
def emit_MOVI(self, value, target):
self.temps[target] = Constant(int_type, value)
def emit_MOVF(self, value, target):
self.temps[target] = Constant(float_type, value)
pass # You must implement
def emit_MOVB(self, value, target):
self.temps[target] = Constant(byte_type, value)
# Allocation of variables. Declare as global variables and set to
# a sensible initial value.
def emit_VARI(self, name):
var = GlobalVariable(self.module, int_type, name=name)
var.initializer = Constant(int_type, 0)
self.globals[name] = var
def emit_VARF(self, name):
var = GlobalVariable(self.module, float_type, name=name)
var.initializer = Constant(float_type, 0.0)
self.globals[name] = var
def emit_VARB(self, name):
var = GlobalVariable(self.module, byte_type, name=name)
var.initializer = Constant(byte_type, 0)
self.globals[name] = var
def emit_ALLOCI(self, name):
self.locals[name] = self.builder.alloca(int_type, name=name)
def emit_ALLOCF(self, name):
self.locals[name] = self.builder.alloca(float_type, name=name)
def emit_ALLOCB(self, name):
self.locals[name] = self.builder.alloca(byte_type, name=name)
# Load/store instructions for variables. Load needs to pull a
# value from a global variable and store in a temporary. Store
# goes in the opposite direction.
def emit_LOADI(self, name, target):
self.temps[target] = self.builder.load(self.vars[name], target)
def emit_LOADF(self, name, target):
self.temps[target] = self.builder.load(self.vars[name], target)
def emit_LOADB(self, name, target):
self.temps[target] = self.builder.load(self.vars[name], target)
pass # You must implement
def emit_STOREI(self, source, target):
self.builder.store(self.temps[source], self.vars[target])
def emit_STOREF(self, source, target):
self.builder.store(self.temps[source], self.vars[target])
def emit_STOREB(self, source, target):
self.builder.store(self.temps[source], self.vars[target])
# Binary + operator
def emit_ADDI(self, left, right, target):
self.temps[target] = self.builder.add(self.temps[left],
self.temps[right], target)
def emit_ADDF(self, left, right, target):
self.temps[target] = self.builder.fadd(self.temps[left],
self.temps[right], target)
pass # You must implement
# Binary - operator
def emit_SUBI(self, left, right, target):
self.temps[target] = self.builder.sub(self.temps[left],
self.temps[right], target)
pass # You must implement
def emit_SUBF(self, left, right, target):
self.temps[target] = self.builder.fsub(self.temps[left],
self.temps[right], target)
pass # You must implement
# Binary * operator
def emit_MULI(self, left, right, target):
self.temps[target] = self.builder.mul(self.temps[left],
self.temps[right], target)
pass # You must implement
def emit_MULF(self, left, right, target):
self.temps[target] = self.builder.fmul(self.temps[left],
self.temps[right], target)
pass # You must implement
# Binary / operator
def emit_DIVI(self, left, right, target):
self.temps[target] = self.builder.sdiv(self.temps[left],
self.temps[right], target)
pass # You must implement
def emit_DIVF(self, left, right, target):
self.temps[target] = self.builder.fdiv(self.temps[left],
self.temps[right], target)
pass # You must implement
def emit_CMPI(self, op, left, right, target):
result = self.builder.icmp_signed(op, self.temps[left],
self.temps[right], '_temp')
self.temps[target] = self.builder.zext(result, int_type, target)
def emit_CMPF(self, op, left, right, target):
result = self.builder.fcmp_ordered(op, self.temps[left],
self.temps[right], '_temp')
self.temps[target] = self.builder.zext(result, int_type, target)
def emit_CMPB(self, op, left, right, target):
result = self.builder.icmp_signed(op, self.temps[left],
self.temps[right], '_temp')
self.temps[target] = self.builder.zext(result, int_type, target)
def emit_AND(self, left, right, target):
self.temps[target] = self.builder.and_(self.temps[left],
self.temps[right], target)
def emit_OR(self, left, right, target):
self.temps[target] = self.builder.or_(self.temps[left],
self.temps[right], target)
def emit_XOR(self, left, right, target):
self.temps[target] = self.builder.xor(self.temps[left],
self.temps[right], target)
# Print statements
def emit_PRINTI(self, source):
self.builder.call(self.runtime['_print_int'], [self.temps[source]])
def emit_PRINTF(self, source):
self.builder.call(self.runtime['_print_float'], [self.temps[source]])
pass # You must implement
def emit_PRINTB(self, source):
self.builder.call(self.runtime['_print_byte'], [self.temps[source]])
pass # You must implement
def emit_LABEL(self, name):
self.block = self.basicblocks[name]
self.builder.position_at_end(self.block)
def emit_CBRANCH(self, test, true_label, false_label):
true_block = self.basicblocks[true_label]
false_block = self.basicblocks[false_label]
self.builder.cbranch(self.builder.trunc(self.temps[test], IntType(1)),
true_block, false_block)
def emit_BRANCH(self, next_label):
target = self.basicblocks[next_label]
if not self.block.is_terminated:
self.builder.branch(target)
def emit_CALL(self, name, *extra):
*args, target = extra
args = [self.temps[a] for a in args]
func = self.vars[name]
self.temps[target] = self.builder.call(func, args)
def emit_RET(self, source):
self.builder.store(self.temps[source], self.return_var)
self.builder.branch(self.return_block)
class GeneratorVisitor(SmallCVisitor):
def __init__(self, output_file=None):
super(SmallCVisitor, self).__init__()
self.Module = Module(name=__file__)
self.Module.triple = "x86_64-pc-linux-gnu"
self.Builder = None
self.function = None
self.NamedValues = dict()
self.counter = 0
self.loop_stack = []
self.signal_stack = []
self.cond_stack = []
self.var_stack = []
self.cur_decl_type = None
self.indentation = 0
self.function_dict = dict()
self.error_queue = list()
self.output_file = output_file
def print(self):
if not self.output_file:
print(self.Module)
else:
f = open(self.output_file, "w+")
f.write(self.Module.__str__())
f.close()
def error(self, info):
print("Error: ", info)
return 0
def toBool(self, value):
zero = Constant(self.getType('bool'), 0)
return self.Builder.icmp_signed('!=', value, zero)
def getVal_of_expr(self, expr):
temp = self.visit(expr)
if isinstance(temp, Constant) or isinstance(
temp, CallInstr) or isinstance(temp, LoadInstr) or isinstance(
temp, Instruction) or isinstance(temp, GlobalVariable):
value = temp
else:
temp_val = self.getVal_local(temp.IDENTIFIER().getText())
temp_ptr = temp_val['ptr']
if temp.array_indexing():
index = self.getVal_of_expr(temp.array_indexing().expr())
temp_ptr = self.Builder.gep(temp_ptr,
[Constant(IntType(32), 0), index],
inbounds=True)
# if isinstance(temp_val['type'],ArrayType):
# if temp.array_indexing():
# index = self.getVal_of_expr(temp.array_indexing().expr())
# temp_ptr = self.Builder.gep(temp_ptr, [Constant(IntType(32), 0), index], inbounds=True)
# elif temp.AMPERSAND():
# Constant(PointerType(IntType(8)), temp_ptr.getText())
# elif temp.ASTERIKS():
# pass
# else: #返回数组地址
# temp_ptr = self.Builder.gep(temp_ptr, [Constant(IntType(32), 0), Constant(IntType(32), 0)], inbounds=True)
# return temp_ptr
value = self.Builder.load(temp_ptr)
return value
def getType(self, type):
if type == 'int':
return IntType(32)
elif type == 'char':
return IntType(8)
elif type == 'float':
return FloatType()
elif type == 'bool':
return IntType(1)
elif type == 'void':
return VoidType()
else:
self.error("type error in <getType>")
def getVal_local(self, id):
temp_maps = self.var_stack[::-1]
for map in temp_maps:
if id in map.keys():
return map[id]
self.error("value error in <getVal_local>")
return None
def visitFunction_definition(self,
ctx: SmallCParser.Function_definitionContext):
retType = self.getType(ctx.type_specifier().getText())
if ctx.identifier().ASTERIKS():
retType = retType.as_pointer()
argsType = []
argsName = []
# args
if ctx.param_decl_list():
args = ctx.param_decl_list()
var_arg = False
for t in args.getChildren():
if t.getText() != ',':
if t.getText() == '...':
var_arg = True
break
t_type = self.getType(t.type_specifier().getText())
if t.identifier().ASTERIKS():
t_type = t_type.as_pointer()
argsType.append(t_type)
argsName.append(t.identifier().IDENTIFIER().getText())
funcType = FunctionType(retType, tuple(argsType), var_arg=var_arg)
# no args
else:
funcType = FunctionType(retType, ())
# function
if ctx.identifier().IDENTIFIER().getText() in self.function_dict:
func = self.function_dict[ctx.identifier().IDENTIFIER().getText()]
else:
func = Function(self.Module,
funcType,
name=ctx.identifier().IDENTIFIER().getText())
self.function_dict[ctx.identifier().IDENTIFIER().getText()] = func
# blocks or ;
if ctx.compound_stmt():
self.function = ctx.identifier().IDENTIFIER().getText()
block = func.append_basic_block(
ctx.identifier().IDENTIFIER().getText())
varDict = dict()
self.Builder = IRBuilder(block)
for i, arg in enumerate(func.args):
arg.name = argsName[i]
alloca = self.Builder.alloca(arg.type, name=arg.name)
self.Builder.store(arg, alloca)
varDict[arg.name] = {
"id": arg.name,
"type": arg.type,
"value": None,
"ptr": alloca
}
self.var_stack.append(varDict)
self.visit(ctx.compound_stmt())
if isinstance(retType, VoidType):
self.Builder.ret_void()
self.var_stack.pop()
self.function = None
return
def visitFunctioncall(self, ctx: SmallCParser.FunctioncallContext):
var_map = self.var_stack[-1]
function = self.function_dict[ctx.identifier().getText()]
arg_types = function.args
index = 0
args = []
if ctx.param_list():
for param in ctx.param_list().getChildren():
if (param.getText() == ','):
continue
temp = self.getVal_of_expr(param)
arg_type = None
if index < len(arg_types):
arg_type = arg_types[index]
ptr_flag = False
if arg_type:
if isinstance(arg_type.type, PointerType):
ptr_flag = True
elif self.getVal_local(temp.name):
temp_type = self.getVal_local(temp.name)['type']
if isinstance(temp_type, PointerType):
ptr_flag = True
if not ptr_flag and not isinstance(temp, Constant):
temp = self.Builder.load(temp)
args.append(temp)
index += 1
return self.Builder.call(function, args)
# def visitVar_decl(self, ctx: SmallCParser.Var_declContext):
# type = self.getType(ctx.type_specifier())
# list = ctx.var_decl_list()
# for var in list.getChildren():
# if var.getText() != ',':
# if self.builder:
# alloca = self.builder.alloca(type, name=var.identifier().getText())
# self.builder.store(Constant(type, None), alloca)
# self.var_stack[-1][var.identifier().getText()] = alloca
# else:
# g_var = GlobalVariable(self.Module, type, var.identifier().getText())
# g_var.initializer = Constant(type, None)
# return
def visitStmt(self, ctx: SmallCParser.StmtContext):
if ctx.RETURN():
value = self.getVal_of_expr(ctx.expr())
if isinstance(value.type, PointerType):
value = self.Builder.load(value)
return self.Builder.ret(value)
elif ctx.CONTINUE():
self.signal_stack[-1] = 1
loop_blocks = self.loop_stack[-1]
self.Builder.branch(loop_blocks['continue'])
self.Builder.position_at_start(loop_blocks['buf'])
return None
elif ctx.BREAK():
self.signal_stack[-1] = -1
loop_blocks = self.loop_stack[-1]
self.Builder.branch(loop_blocks['break'])
self.Builder.position_at_start(loop_blocks['buf'])
else:
return self.visitChildren(ctx)
def visitCompound_stmt(self, ctx: SmallCParser.Compound_stmtContext):
# builder = IRBuilder(self.block_stack[-1])
# block = self.Builder.append_basic_block()
# self.block_stack.append(block)
# with self.Builder.goto_block(block):
result = self.visitChildren(ctx)
# self.block_stack.pop()
return result
def visitAssignment(self, ctx: SmallCParser.AssignmentContext):
value = self.getVal_of_expr(ctx.expr())
identifier = ctx.identifier()
identifier = self.getVal_local(identifier.IDENTIFIER().getText())
if isinstance(identifier['type'], ArrayType):
if ctx.identifier().array_indexing():
index = self.getVal_of_expr(
ctx.identifier().array_indexing().expr())
if isinstance(index.type, PointerType):
index = self.Builder.load(index)
else:
index = Constant(IntType(32), 0)
tempPtr = self.Builder.gep(identifier['ptr'],
[Constant(IntType(32), 0), index],
inbounds=True)
if isinstance(value.type, PointerType):
value = self.Builder.load(value)
return self.Builder.store(value, tempPtr)
if isinstance(value.type, PointerType):
value = self.Builder.load(value)
return self.Builder.store(value, identifier['ptr'])
def visitExpr(self, ctx: SmallCParser.ExprContext):
if ctx.condition():
return self.visit(ctx.condition())
elif ctx.assignment():
return self.visit(ctx.assignment())
elif ctx.functioncall():
return self.visit(ctx.functioncall())
def visitCondition(self, ctx: SmallCParser.ConditionContext):
if ctx.expr():
disjunction = self.getVal_of_expr(ctx.disjunction())
if isinstance(disjunction.type, PointerType):
disjunction = self.Builder.load(disjunction)
cond = self.Builder.icmp_signed('!=', disjunction,
Constant(disjunction.type, 0))
expr = self.getVal_of_expr(ctx.expr())
if isinstance(expr.type, PointerType):
expr = self.Builder.load(expr)
condition = self.getVal_of_expr(ctx.condition())
return self.Builder.select(cond, expr, condition)
else:
return self.getVal_of_expr(ctx.disjunction())
def visitDisjunction(self, ctx: SmallCParser.DisjunctionContext):
if ctx.disjunction():
disjunction = self.getVal_of_expr(ctx.disjunction())
if isinstance(disjunction.type, PointerType):
disjunction = self.Builder.load(disjunction)
conjunction = self.getVal_of_expr(ctx.conjunction())
if isinstance(conjunction.type, PointerType):
conjunction = self.Builder.load(conjunction)
left = self.Builder.icmp_signed('!=', disjunction,
Constant(disjunction.type, 0))
right = self.Builder.icmp_signed('!=', conjunction,
Constant(conjunction.type, 0))
return self.Builder.or_(left, right)
else:
return self.getVal_of_expr(ctx.conjunction())
def visitConjunction(self, ctx: SmallCParser.ConjunctionContext):
if ctx.conjunction():
conjunction = self.getVal_of_expr(ctx.conjunction())
if isinstance(conjunction.type, PointerType):
conjunction = self.Builder.load(conjunction)
comparison = self.getVal_of_expr(ctx.comparison())
if isinstance(comparison.type, PointerType):
comparison = self.Builder.load(comparison)
left = self.Builder.icmp_signed('!=', conjunction,
Constant(conjunction.type, 0))
right = self.Builder.icmp_signed('!=', comparison,
Constant(comparison.type, 0))
return self.Builder.and_(left, right)
else:
return self.getVal_of_expr(ctx.comparison())
def visitComparison(self, ctx: SmallCParser.ComparisonContext):
if ctx.EQUALITY():
relation1 = self.getVal_of_expr(ctx.relation(0))
if isinstance(relation1.type, PointerType):
relation1 = self.Builder.load(relation1)
relation2 = self.getVal_of_expr(ctx.relation(1))
if isinstance(relation2.type, PointerType):
relation2 = self.Builder.load(relation2)
return self.Builder.icmp_signed('==', relation1, relation2)
elif ctx.NEQUALITY():
relation1 = self.getVal_of_expr(ctx.relation(0))
if isinstance(relation1.type, PointerType):
relation1 = self.Builder.load(relation1)
relation2 = self.getVal_of_expr(ctx.relation(1))
if isinstance(relation2.type, PointerType):
relation2 = self.Builder.load(relation2)
return self.Builder.icmp_signed('!=', relation1, relation2)
else:
return self.getVal_of_expr(ctx.relation(0))
def visitRelation(self, ctx: SmallCParser.RelationContext):
if len(ctx.equation()) > 1:
equation1 = self.getVal_of_expr(ctx.equation(0))
if isinstance(equation1.type, PointerType):
equation1 = self.Builder.load(equation1)
equation2 = self.getVal_of_expr(ctx.equation(1))
if isinstance(equation2.type, PointerType):
equation2 = self.Builder.load(equation2)
if ctx.LEFTANGLE():
value = self.Builder.icmp_signed('<', equation1, equation2)
elif ctx.RIGHTANGLE():
value = self.Builder.icmp_signed('>', equation1, equation2)
elif ctx.LEFTANGLEEQUAL():
value = self.Builder.icmp_signed('<=', equation1, equation2)
elif ctx.RIGHTANGLEEQUAL():
value = self.Builder.icmp_signed('>=', equation1, equation2)
return value
else:
return self.getVal_of_expr(ctx.equation(0))
def visitFor_stmt(self, ctx: SmallCParser.For_stmtContext):
func = self.function_dict[self.function]
end_block = func.append_basic_block()
self.var_stack.append({})
decl_block = func.append_basic_block()
self.var_stack.append({})
cond_block = func.append_basic_block()
self.var_stack.append({})
stmt_block = func.append_basic_block()
self.var_stack.append({})
loop_block = func.append_basic_block()
# 1 -> continue, -1 -> break
self.signal_stack.append(0)
self.loop_stack.append({
'continue': cond_block,
'break': end_block,
'buf': loop_block
})
with self.Builder.goto_block(decl_block):
# self.Builder.position_at_start(end_block)
if ctx.var_decl():
self.visit(ctx.var_decl())
elif ctx.var_decl_list():
self.visit(ctx.var_decl_list())
else:
self.error("for error in <visitFor_stmt>")
self.Builder.branch(cond_block)
self.Builder.branch(decl_block)
with self.Builder.goto_block(cond_block):
# cond_expr
cond_expr = ctx.expr(0)
cond_expr = self.visit(cond_expr)
cond_expr = self.toBool(cond_expr)
self.Builder.cbranch(cond_expr, stmt_block, end_block)
with self.Builder.goto_block(stmt_block):
# expr
self.visit(ctx.stmt())
expr = ctx.expr(1)
self.visit(expr)
self.Builder.branch(cond_block)
if self.signal_stack[-1] == 0:
loop_blocks = self.loop_stack[-1]
self.Builder.position_at_start(loop_blocks['buf'])
self.Builder.branch(end_block)
self.Builder.position_at_start(end_block)
self.loop_stack.pop()
self.signal_stack.pop()
self.var_stack.pop()
self.var_stack.pop()
self.var_stack.pop()
self.var_stack.pop()
def visitWhile_stmt(self, ctx: SmallCParser.While_stmtContext):
func = self.function_dict[self.function]
end_block = func.append_basic_block()
self.var_stack.append({})
cond_block = func.append_basic_block()
self.var_stack.append({})
stmt_block = func.append_basic_block()
self.var_stack.append({})
loop_block = func.append_basic_block()
# 1 -> continue, -1 -> break
self.signal_stack.append(0)
self.loop_stack.append({
'continue': cond_block,
'break': end_block,
'buf': loop_block
})
self.Builder.branch(cond_block)
with self.Builder.goto_block(cond_block):
expr = self.getVal_of_expr(ctx.expr())
cond_expr = self.toBool(expr)
self.Builder.cbranch(cond_expr, stmt_block, end_block)
with self.Builder.goto_block(stmt_block):
self.visit(ctx.stmt())
self.Builder.branch(cond_block)
if self.signal_stack[-1] == 0:
loop_blocks = self.loop_stack[-1]
self.Builder.position_at_start(loop_blocks['buf'])
self.Builder.branch(end_block)
self.Builder.position_at_start(end_block)
self.loop_stack.pop()
self.signal_stack.pop()
self.var_stack.pop()
self.var_stack.pop()
self.var_stack.pop()
def visitCond_stmt(self, ctx: SmallCParser.Cond_stmtContext):
expr = self.getVal_of_expr(ctx.expr())
cond_expr = self.toBool(expr)
else_expr = ctx.ELSE()
if else_expr:
with self.Builder.if_else(cond_expr) as (then, otherwise):
with then:
self.var_stack.append({})
true_stmt = ctx.stmt(0)
self.visit(true_stmt)
self.var_stack.pop()
with otherwise:
self.var_stack.append({})
else_stmt = ctx.stmt(1)
self.visit(else_stmt)
self.var_stack.pop()
else:
with self.Builder.if_then(cond_expr):
self.var_stack.append({})
true_stmt = ctx.stmt(0)
self.visit(true_stmt)
self.var_stack.pop()
return None
def visitVar_decl(self, ctx: SmallCParser.Var_declContext):
self.cur_decl_type = self.getType(ctx.type_specifier().getText())
return self.visitChildren(ctx)
def visitVar_decl_list(self, ctx: SmallCParser.Var_decl_listContext):
ans = []
decls = ctx.variable_id()
for decl in decls:
ans.append(self.visit(decl))
return ans
def visitVariable_id(self, ctx: SmallCParser.Variable_idContext):
identifier = ctx.identifier()
type = self.cur_decl_type
if not self.function:
if identifier.array_indexing():
length = self.getVal_of_expr(
identifier.array_indexing().expr())
type = ArrayType(type, length.constant)
g_var = GlobalVariable(self.Module, type,
identifier.IDENTIFIER().getText())
else:
g_var = GlobalVariable(self.Module, type,
identifier.IDENTIFIER().getText())
g_var.initializer = Constant(type, None)
atomic = {
"id": identifier.IDENTIFIER().getText(),
"type": type,
"value": None,
"ptr": g_var
}
if not len(self.var_stack):
self.var_stack.append({})
self.var_stack[0][identifier.IDENTIFIER().getText()] = atomic
return g_var
var_map = self.var_stack[-1]
if identifier.array_indexing():
length = self.getVal_of_expr(identifier.array_indexing().expr())
type = ArrayType(type, length.constant)
ptr = self.Builder.alloca(typ=type,
name=identifier.IDENTIFIER().getText())
else:
ptr = self.Builder.alloca(typ=type,
name=identifier.IDENTIFIER().getText())
expr = ctx.expr()
if expr:
value = self.getVal_of_expr(expr)
else:
value = Constant(type, None)
if isinstance(value.type, PointerType):
value = self.Builder.load(value)
self.Builder.store(value, ptr)
var_map[identifier.IDENTIFIER().getText()] = {
"id": identifier.IDENTIFIER().getText(),
"type": type,
"value": value,
"ptr": ptr
}
return ptr
def visitPrimary(self, ctx: SmallCParser.PrimaryContext):
if ctx.BOOLEAN():
return Constant(IntType(1), bool(ctx.getText()))
elif ctx.INTEGER():
return Constant(IntType(32), int(ctx.getText()))
elif ctx.REAL():
return Constant(FloatType, float(ctx.getText()))
elif ctx.CHARCONST():
tempStr = ctx.getText()[1:-1]
tempStr = tempStr.replace('\\n', '\n')
tempStr = tempStr.replace('\\0', '\0')
if ctx.getText()[0] == '"':
tempStr += '\0'
temp = GlobalVariable(self.Module,
ArrayType(IntType(8), len(tempStr)),
name="str_" + tempStr[:-1] +
str(self.counter))
self.counter += 1
temp.initializer = Constant(
ArrayType(IntType(8), len(tempStr)),
bytearray(tempStr, encoding='utf-8'))
temp.global_constant = True
return self.Builder.gep(
temp, [Constant(IntType(32), 0),
Constant(IntType(32), 0)],
inbounds=True)
return Constant(IntType(8), ord(tempStr[0]))
elif ctx.identifier():
return self.visit(ctx.identifier())
elif ctx.functioncall():
return self.visit(ctx.functioncall())
elif ctx.expr():
return self.visit(ctx.expr())
else:
return self.error("type error in <visitPrimary>")
def visitFactor(self, ctx: SmallCParser.FactorContext):
if (ctx.MINUS()):
factor = self.getVal_of_expr(ctx.factor())
if isinstance(factor.type, PointerType):
factor = self.Builder.load(factor)
factor = self.Builder.neg(factor)
return factor
return self.visitChildren(ctx)
def visitTerm(self, ctx: SmallCParser.TermContext):
if (ctx.ASTERIKS()):
term = self.getVal_of_expr(ctx.term())
if isinstance(term.type, PointerType):
term = self.Builder.load(term)
factor = self.getVal_of_expr(ctx.factor())
if isinstance(factor.type, PointerType):
factor = self.Builder.load(factor)
return self.Builder.mul(term, factor)
if (ctx.SLASH()):
term = self.getVal_of_expr(ctx.term())
if isinstance(term.type, PointerType):
term = self.Builder.load(term)
factor = self.getVal_of_expr(ctx.factor())
if isinstance(factor.type, PointerType):
factor = self.Builder.load(factor)
return self.Builder.sdiv(term, factor)
return self.visitChildren(ctx)
def visitEquation(self, ctx: SmallCParser.EquationContext):
if (ctx.PLUS()):
equation = self.getVal_of_expr(ctx.equation())
if isinstance(equation.type, PointerType):
equation = self.Builder.load(equation)
if str(equation.type) != 'i32':
equation = self.Builder.zext(equation, IntType(32))
term = self.getVal_of_expr(ctx.term())
if isinstance(term.type, PointerType):
term = self.Builder.load(term)
if str(term.type) != 'i32':
term = self.Builder.zext(term, IntType(32))
return self.Builder.add(equation, term)
if (ctx.MINUS()):
equation = self.getVal_of_expr(ctx.equation())
if isinstance(equation.type, PointerType):
equation = self.Builder.load(equation)
if str(equation.type) != 'i32':
equation = self.Builder.zext(equation, IntType(32))
term = self.getVal_of_expr(ctx.term())
if isinstance(term.type, PointerType):
term = self.Builder.load(term)
if str(term.type) != 'i32':
term = self.Builder.zext(term, IntType(32))
return self.Builder.sub(equation, term)
return self.visitChildren(ctx)
def visitIdentifier(self, ctx: SmallCParser.IdentifierContext):
if (ctx.AMPERSAND() and ctx.array_indexing()):
return self.Builder.gep(self.getVal_of_expr(ctx.IDENTIFIER()),
self.getVal_of_expr(ctx.array_indexing()))
if (ctx.ASTERIKS() and ctx.array_indexing()):
return self.Builder.load(
self.Builder.gep(self.getVal_of_expr(ctx.IDENTIFIER()),
self.getVal_of_expr(ctx.array_indexing())))
if (ctx.AMPERSAND()):
return self.getVal_local(str(ctx.IDENTIFIER()))['ptr']
if (ctx.ASTERIKS()):
return self.getVal_local(str(ctx.IDENTIFIER()))['ptr']
temp = self.getVal_local(ctx.IDENTIFIER().getText())
temp_ptr = temp['ptr']
# if isinstance(temp_val['type'],ArrayType):
# if temp.array_indexing():
# index = self.getVal_of_expr(temp.array_indexing().expr())
# temp_ptr = self.Builder.gep(temp_ptr, [Constant(IntType(32), 0), index], inbounds=True)
# elif temp.AMPERSAND():
# Constant(PointerType(IntType(8)), temp_ptr.getText())
# elif temp.ASTERIKS():
# pass
# else: #返回数组地址
# temp_ptr = self.Builder.gep(temp_ptr, [Constant(IntType(32), 0), Constant(IntType(32), 0)], inbounds=True)
# return temp_ptr
if isinstance(temp['type'], ArrayType):
if ctx.array_indexing():
index = self.getVal_of_expr(ctx.array_indexing().expr())
if isinstance(index.type, PointerType):
index = self.Builder.load(index)
temp_ptr = self.Builder.gep(temp_ptr,
[Constant(IntType(32), 0), index],
inbounds=True)
value = self.Builder.load(temp_ptr)
self.var_stack[-1][temp_ptr.name] = {
'id': temp_ptr.name,
'type': temp_ptr.type,
'value': value,
'ptr': temp_ptr
}
return temp_ptr
else:
temp_ptr = self.Builder.gep(
temp_ptr,
[Constant(IntType(32), 0),
Constant(IntType(32), 0)],
inbounds=True)
value = self.Builder.load(temp_ptr)
self.var_stack[-1][temp_ptr.name] = {
'id': temp_ptr.name,
'type': temp_ptr.type,
'value': value,
'ptr': temp_ptr
}
return temp_ptr
temp_val = temp['ptr']
return temp_val
def visitArray_indexing(self, ctx: SmallCParser.Array_indexingContext):
return self.visit(ctx.expr())
# Perform the comparison
testvar = builder.icmp_signed('<', a_var, b_var)
# Make three blocks
then_block = f_func.append_basic_block('then')
else_block = f_func.append_basic_block('else')
merge_block = f_func.append_basic_block('merge')
# Emit the branch instruction
builder.cbranch(testvar, then_block, else_block)
# Generate code in the then-branch
builder.position_at_end(then_block)
result = builder.add(a_var, b_var)
builder.store(result, c_var)
builder.branch(merge_block)
# Generate code in the else-branch
builder.position_at_end(else_block)
result = builder.sub(a_var, b_var)
builder.store(result, c_var)
builder.branch(merge_block)
# Emit code after the if-else
builder.position_at_end(merge_block)
builder.ret(builder.load(c_var))
print(mod)
print(":::: RUNNING ::::")
class GenerateLLVM(object):
def __init__(self):
# Perform the basic LLVM initialization. You need the following parts:
#
# 1. A top-level Module object
# 2. A Function instance in which to insert code
# 3. A Builder instance to generate instructions
#
# Note: For project 5, we don't have any user-defined
# functions so we're just going to emit all LLVM code into a top
# level function void main() { ... }. This will get changed later.
self.module = Module('module')
# Dictionary that holds all of the global variable/function declarations.
# Any declaration in the Gone source code is going to get an entry here
self.globals = {}
self.vars = ChainMap(self.globals)
# Dictionary that holds all of the temporary registers created in
# the intermediate code.
# Initialize the runtime library functions (see below)
self.declare_runtime_library()
def declare_runtime_library(self):
# Certain functions such as I/O and string handling are often easier
# to implement in an external C library. This method should make
# the LLVM declarations for any runtime functions to be used
# during code generation. Please note that runtime function
# functions are implemented in C in a separate file gonert.c
self.runtime = {}
# Declare printing functions
self.runtime['_print_int'] = Function(self.module,
FunctionType(
void_type, [int_type]),
name="_print_int")
self.runtime['_print_float'] = Function(self.module,
FunctionType(
void_type, [float_type]),
name="_print_float")
self.runtime['_print_byte'] = Function(self.module,
FunctionType(
void_type, [byte_type]),
name="_print_byte")
def generate_functions(self, functions):
type_dict = {
'int': int_type,
'float': float_type,
'byte': byte_type,
'bool': int_type,
'void': void_type
}
for function in functions:
# register function
name = function.name if function.name != 'main' else '_gone_main'
return_type = type_dict[function.return_type]
param_types = [type_dict[t] for t in function.param_types]
function_type = FunctionType(return_type, param_types)
self.function = Function(self.module, function_type, name=name)
self.globals[name] = self.function
self.blocks = {}
self.block = self.function.append_basic_block('entry')
self.blocks['entry'] = self.block
self.builder = IRBuilder(self.block)
# local scope
self.vars = self.vars.new_child()
self.temps = {}
for n, (param_name, param_type) in enumerate(
zip(function.param_names, param_types)):
var = self.builder.alloca(param_type, name=param_name)
var.initializer = Constant(param_type, 0)
self.vars[param_name] = var
self.builder.store(self.function.args[n],
self.vars[param_name])
# alloc return var / block
if function.return_type != 'void':
self.vars['return'] = self.builder.alloca(return_type,
name='return')
self.return_block = self.function.append_basic_block('return')
# generate instructions
self.generate_code(function)
# return
if not self.block.is_terminated:
self.builder.branch(self.return_block)
self.builder.position_at_end(self.return_block)
if function.return_type != 'void':
self.builder.ret(
self.builder.load(self.vars['return'], 'return'))
else:
self.builder.ret_void()
self.vars = self.vars.parents
def generate_code(self, ircode):
# Given a sequence of SSA intermediate code tuples, generate LLVM
# instructions using the current builder (self.builder). Each
# opcode tuple (opcode, args) is dispatched to a method of the
# form self.emit_opcode(args)
for instr in ircode:
if instr[0] == 'LABEL':
self.blocks[instr[1]] = self.function.append_basic_block(
instr[1])
for opcode, *args in ircode:
if opcode == 'CALL':
self.emit_CALL(*args[:-1], target=args[-1])
elif hasattr(self, 'emit_' + opcode):
getattr(self, 'emit_' + opcode)(*args)
else:
print('Warning: No emit_' + opcode + '() method')
# ----------------------------------------------------------------------
# Opcode implementation. You must implement the opcodes. A few
# sample opcodes have been given to get you started.
# ----------------------------------------------------------------------
# Creation of literal values. Simply define as LLVM constants.
def emit_MOVI(self, value, target):
self.temps[target] = Constant(int_type, value)
def emit_MOVF(self, value, target):
self.temps[target] = Constant(float_type, value)
def emit_MOVB(self, value, target):
self.temps[target] = Constant(byte_type, value)
# Allocation of variables. Declare as global variables and set to
# a sensible initial value.
def emit_VARI(self, name):
var = GlobalVariable(self.module, int_type, name=name)
var.initializer = Constant(int_type, 0)
self.globals[name] = var
def emit_VARF(self, name):
var = GlobalVariable(self.module, float_type, name=name)
var.initializer = Constant(float_type, 0.0)
self.globals[name] = var
def emit_VARB(self, name):
var = GlobalVariable(self.module, byte_type, name=name)
var.initializer = Constant(byte_type, 0)
self.globals[name] = var
# Load/store instructions for variables. Load needs to pull a
# value from a global variable and store in a temporary. Store
# goes in the opposite direction.
def emit_LOADI(self, name, target):
self.temps[target] = self.builder.load(self.vars[name], target)
def emit_LOADF(self, name, target):
self.temps[target] = self.builder.load(self.vars[name], target)
def emit_LOADB(self, name, target):
self.temps[target] = self.builder.load(self.vars[name], target)
def emit_STOREI(self, source, target):
self.builder.store(self.temps[source], self.vars[target])
def emit_STOREF(self, source, target):
self.builder.store(self.temps[source], self.vars[target])
def emit_STOREB(self, source, target):
self.builder.store(self.temps[source], self.vars[target])
# Binary + operator
def emit_ADDI(self, left, right, target):
self.temps[target] = self.builder.add(self.temps[left],
self.temps[right], target)
def emit_ADDF(self, left, right, target):
self.temps[target] = self.builder.fadd(self.temps[left],
self.temps[right], target)
# Binary - operator
def emit_SUBI(self, left, right, target):
self.temps[target] = self.builder.sub(self.temps[left],
self.temps[right], target)
def emit_SUBF(self, left, right, target):
self.temps[target] = self.builder.fsub(self.temps[left],
self.temps[right], target)
# Binary * operator
def emit_MULI(self, left, right, target):
self.temps[target] = self.builder.mul(self.temps[left],
self.temps[right], target)
def emit_MULF(self, left, right, target):
self.temps[target] = self.builder.fmul(self.temps[left],
self.temps[right], target)
# Binary / operator
def emit_DIVI(self, left, right, target):
self.temps[target] = self.builder.sdiv(self.temps[left],
self.temps[right], target)
def emit_DIVF(self, left, right, target):
self.temps[target] = self.builder.fdiv(self.temps[left],
self.temps[right], target)
def emit_CMPI(self, op, left, right, target):
tmp = self.builder.icmp_signed(op, self.temps[left], self.temps[right],
'tmp')
self.temps[target] = self.builder.zext(tmp, int_type, target)
def emit_CMPF(self, op, left, right, target):
tmp = self.builder.fcmp_ordered(op, self.temps[left],
self.temps[right], 'tmp')
self.temps[target] = self.builder.zext(tmp, int_type, target)
def emit_CMPB(self, op, left, right, target):
tmp = self.builder.icmp_signed(op, self.temps[left], self.temps[right],
'tmp')
self.temps[target] = self.builder.zext(tmp, int_type, target)
# Logical ops
def emit_AND(self, left, right, target):
self.temps[target] = self.builder.and_(self.temps[left],
self.temps[right], target)
def emit_OR(self, left, right, target):
self.temps[target] = self.builder.or_(self.temps[left],
self.temps[right], target)
# control flow
def emit_LABEL(self, label):
self.block = self.blocks[label]
self.builder.position_at_end(self.blocks[label])
def emit_BRANCH(self, label):
if not self.block.is_terminated:
self.builder.branch(self.blocks[label])
def emit_CBRANCH(self, test, label1, label2):
tmp = self.builder.trunc(self.temps[test], IntType(1), 'tmp')
self.builder.cbranch(tmp, self.blocks[label1], self.blocks[label2])
# functions
def emit_ALLOCI(self, name):
var = self.builder.alloca(int_type, name=name)
var.initializer = Constant(int_type, 0)
self.vars[name] = var
def emit_ALLOCF(self, name):
var = self.builder.alloca(float_type, name=name)
var.initializer = Constant(float_type, 0)
self.vars[name] = var
def emit_ALLOCB(self, name):
var = self.builder.alloca(byte_type, name=name)
var.initializer = Constant(byte_type, 0)
self.vars[name] = var
def emit_RET(self, source):
self.builder.store(self.temps[source], self.vars['return'])
self.builder.branch(self.return_block)
def emit_CALL(self, func_name, *args, target):
func = self.vars[func_name]
args = [self.temps[arg] for arg in args]
self.temps[target] = self.builder.call(func, args)
# Print statements
def emit_PRINTI(self, source):
self.builder.call(self.runtime['_print_int'], [self.temps[source]])
def emit_PRINTF(self, source):
self.builder.call(self.runtime['_print_float'], [self.temps[source]])
def emit_PRINTB(self, source):
self.builder.call(self.runtime['_print_byte'], [self.temps[source]])
class GenerateLLVM(object):
def __init__(self):
self.module = Module('module')
self.globals = {}
self.blocks = {}
self.declare_runtime_library()
def declare_runtime_library(self):
self.runtime = {}
self.runtime['_print_int'] = Function(self.module,
FunctionType(
void_type, [int_type]),
name="_print_int")
self.runtime['_print_float'] = Function(self.module,
FunctionType(
void_type, [float_type]),
name="_print_float")
self.runtime['_print_byte'] = Function(self.module,
FunctionType(
void_type, [byte_type]),
name="_print_byte")
def generate_code(self, ir_function):
self.function = Function(
self.module,
FunctionType(LLVM_TYPE_MAPPING[ir_function.return_type], [
LLVM_TYPE_MAPPING[ptype] for _, ptype in ir_function.parameters
]),
name=ir_function.name)
self.block = self.function.append_basic_block('entry')
self.builder = IRBuilder(self.block)
self.globals[ir_function.name] = self.function
self.locals = {}
self.vars = ChainMap(self.locals, self.globals)
self.temps = {}
for n, (pname, ptype) in enumerate(ir_function.parameters):
self.vars[pname] = self.builder.alloca(LLVM_TYPE_MAPPING[ptype],
name=pname)
self.builder.store(self.function.args[n], self.vars[pname])
if ir_function.return_type:
self.vars['return'] = self.builder.alloca(
LLVM_TYPE_MAPPING[ir_function.return_type], name='return')
self.return_block = self.function.append_basic_block('return')
for opcode, *args in ir_function.code:
if hasattr(self, 'emit_' + opcode):
getattr(self, 'emit_' + opcode)(*args)
else:
print('Warning: No emit_' + opcode + '() method')
if not self.block.is_terminated:
self.builder.branch(self.return_block)
self.builder.position_at_end(self.return_block)
self.builder.ret(self.builder.load(self.vars['return'], 'return'))
def get_block(self, block_name):
block = self.blocks.get(block_name)
if block is None:
block = self.function.append_basic_block(block_name)
self.blocks[block_name] = block
return block
def emit_MOV(self, value, target, val_type):
self.temps[target] = Constant(val_type, value)
emit_MOVI = partialmethod(emit_MOV, val_type=int_type)
emit_MOVF = partialmethod(emit_MOV, val_type=float_type)
emit_MOVB = partialmethod(emit_MOV, val_type=byte_type)
def emit_VAR(self, name, var_type):
var = GlobalVariable(self.module, var_type, name=name)
var.initializer = Constant(var_type, 0)
self.globals[name] = var
emit_VARI = partialmethod(emit_VAR, var_type=int_type)
emit_VARF = partialmethod(emit_VAR, var_type=float_type)
emit_VARB = partialmethod(emit_VAR, var_type=byte_type)
def emit_ALLOC(self, name, var_type):
self.locals[name] = self.builder.alloca(var_type, name=name)
emit_ALLOCI = partialmethod(emit_ALLOC, var_type=int_type)
emit_ALLOCF = partialmethod(emit_ALLOC, var_type=float_type)
emit_ALLOCB = partialmethod(emit_ALLOC, var_type=byte_type)
def emit_LOADI(self, name, target):
self.temps[target] = self.builder.load(self.vars[name], name=target)
emit_LOADF = emit_LOADI
emit_LOADB = emit_LOADI
def emit_STOREI(self, source, target):
self.builder.store(self.temps[source], self.vars[target])
emit_STOREF = emit_STOREI
emit_STOREB = emit_STOREI
def emit_ADDI(self, left, right, target):
self.temps[target] = self.builder.add(self.temps[left],
self.temps[right],
name=target)
def emit_ADDF(self, left, right, target):
self.temps[target] = self.builder.fadd(self.temps[left],
self.temps[right],
name=target)
def emit_SUBI(self, left, right, target):
self.temps[target] = self.builder.sub(self.temps[left],
self.temps[right],
name=target)
def emit_SUBF(self, left, right, target):
self.temps[target] = self.builder.fsub(self.temps[left],
self.temps[right],
name=target)
def emit_MULI(self, left, right, target):
self.temps[target] = self.builder.mul(self.temps[left],
self.temps[right],
name=target)
def emit_MULF(self, left, right, target):
self.temps[target] = self.builder.fmul(self.temps[left],
self.temps[right],
name=target)
def emit_DIVI(self, left, right, target):
self.temps[target] = self.builder.sdiv(self.temps[left],
self.temps[right],
name=target)
def emit_DIVF(self, left, right, target):
self.temps[target] = self.builder.fdiv(self.temps[left],
self.temps[right],
name=target)
def emit_PRINT(self, source, runtime_name):
self.builder.call(self.runtime[runtime_name], [self.temps[source]])
emit_PRINTI = partialmethod(emit_PRINT, runtime_name="_print_int")
emit_PRINTF = partialmethod(emit_PRINT, runtime_name="_print_float")
emit_PRINTB = partialmethod(emit_PRINT, runtime_name="_print_byte")
def emit_CMPI(self, operator, left, right, target):
if operator == "=":
operator = "=="
tmp = self.builder.icmp_signed(operator, self.temps[left],
self.temps[right], 'tmp')
self.temps[target] = self.builder.zext(tmp, int_type, target)
def emit_CMPF(self, operator, left, right, target):
if operator == "=":
operator = "=="
tmp = self.builder.fcmp_ordered(operator, self.temps[left],
self.temps[right], 'tmp')
self.temps[target] = self.builder.zext(tmp, int_type, target)
emit_CMPB = emit_CMPI
def emit_AND(self, left, right, target):
self.temps[target] = self.builder.and_(self.temps[left],
self.temps[right], target)
def emit_OR(self, left, right, target):
self.temps[target] = self.builder.or_(self.temps[left],
self.temps[right], target)
def emit_XOR(self, left, right, target):
self.temps[target] = self.builder.xor(self.temps[left],
self.temps[right], target)
def emit_LABEL(self, lbl_name):
self.block = self.get_block(lbl_name)
self.builder.position_at_end(self.block)
def emit_BRANCH(self, dst_label):
if not self.block.is_terminated:
self.builder.branch(self.get_block(dst_label))
def emit_CBRANCH(self, test_target, true_label, false_label):
true_block = self.get_block(true_label)
false_block = self.get_block(false_label)
testvar = self.temps[test_target]
self.builder.cbranch(self.builder.trunc(testvar, IntType(1)),
true_block, false_block)
def emit_RET(self, register):
self.builder.store(self.temps[register], self.vars['return'])
self.builder.branch(self.return_block)
def emit_CALL(self, func_name, *registers):
args = [self.temps[r] for r in registers[:-1]]
target = registers[-1]
self.temps[target] = self.builder.call(self.globals[func_name], args)
class GenerateLLVM(object):
def __init__(self):
# Perform the basic LLVM initialization. You need the following parts:
#
# 1. A top-level Module object
# 2. A dictionary of global declarations
# 3. Initialization of runtime functions (for printing)
#
self.module = Module('module')
# Dictionary that holds all of the global variable/function declarations.
# Any declaration in the Wabbit source code is going to get an entry here
self.globals = {}
# Initialize the runtime library functions (see below)
self.declare_runtime_library()
def declare_runtime_library(self):
# Certain functions such as I/O and string handling are often easier
# to implement in an external C library. This method should make
# the LLVM declarations for any runtime functions to be used
# during code generation. Please note that runtime function
# functions are implemented in C in a separate file wabbitrt.c
self.runtime = {}
# Declare runtime functions
functions = [
('_print_int', void_type, [int_type]),
('_print_float', void_type, [float_type]),
('_print_byte', void_type, [int_type]),
('_grow', int_type, [int_type]),
('_peeki', int_type, [int_type]),
('_peekf', float_type, [int_type]),
('_peekb', int_type, [int_type]),
('_pokei', void_type, [int_type, int_type]),
('_pokef', void_type, [int_type, float_type]),
('_pokeb', void_type, [int_type, int_type]),
]
for name, rettype, args in functions:
self.runtime[name] = Function(self.module,
FunctionType(rettype, args),
name=name)
def declare_function(self, funcname, argtypes, rettype):
self.function = Function(self.module,
FunctionType(rettype, argtypes),
name=funcname)
# Insert a reference in global namespace
self.globals[funcname] = self.function
def generate_function(self, funcname, argnames, ircode):
# Generate code for a single Wabbit function. Each opcode
# tuple (opcode, args) is dispatched to a method of the form
# self.emit_opcode(args). Function should already be declared
# using declare_function.
self.function = self.globals[funcname]
self.block = self.function.append_basic_block('entry')
self.builder = IRBuilder(self.block)
# Stack of LLVM temporaries
self.stack = []
# Dictionary of local variables
self.locals = { }
# Combined symbol table
self.symbols = ChainMap(self.locals, self.globals)
# Have to declare local variables for holding function arguments
for n, (name, ty) in enumerate(zip(argnames, self.function.function_type.args)):
self.locals[name] = self.builder.alloca(ty, name=name)
self.builder.store(self.function.args[n], self.locals[name])
# Stack of blocks
self.blocks = [ ]
for opcode, *opargs in ircode:
if hasattr(self, 'emit_'+opcode):
getattr(self, 'emit_'+opcode)(*opargs)
else:
print('Warning: No emit_'+opcode+'() method')
# Add a return statement to void functions.
if self.function.function_type.return_type == void_type:
self.builder.ret_void()
# Helper methods for LLVM temporary stack manipulation
def push(self, value):
self.stack.append(value)
def pop(self):
return self.stack.pop()
def set_block(self, block):
self.block = block
self.builder.position_at_end(self.block)
# ----------------------------------------------------------------------
# Opcode implementation. You must implement the opcodes. A few
# sample opcodes have been given to get you started.
# ----------------------------------------------------------------------
# Creation of literal values. Simply define as LLVM constants.
def emit_CONSTI(self, value):
self.push(Constant(int_type, value))
def emit_CONSTF(self, value):
self.push(Constant(float_type, value))
# Allocation of variables. Declare as global variables and set to
# a sensible initial value.
def emit_VARI(self, name):
self.locals[name] = self.builder.alloca(int_type, name=name)
def emit_VARF(self, name):
self.locals[name] = self.builder.alloca(float_type, name=name)
# Allocation of globals
def emit_GLOBALI(self, name):
var = GlobalVariable(self.module, int_type, name=name)
var.initializer = Constant(int_type, 0)
self.globals[name] = var
def emit_GLOBALF(self, name):
var = GlobalVariable(self.module, float_type, name=name)
var.initializer = Constant(float_type, 0.0)
self.globals[name] = var
# Load/store instructions for variables. Load needs to pull a
# value from a global variable and store in a temporary. Store
# goes in the opposite direction.
def emit_LOAD(self, name):
self.push(self.builder.load(self.symbols[name], name))
def emit_STORE(self, target):
self.builder.store(self.pop(), self.symbols[target])
# Binary + operator
def emit_ADDI(self):
self.push(self.builder.add(self.pop(), self.pop()))
def emit_ADDF(self):
self.push(self.builder.fadd(self.pop(), self.pop()))
# Binary - operator
def emit_SUBI(self):
right = self.pop()
left = self.pop()
self.push(self.builder.sub(left, right))
def emit_SUBF(self):
right = self.pop()
left = self.pop()
self.push(self.builder.fsub(left, right))
# Binary * operator
def emit_MULI(self):
self.push(self.builder.mul(self.pop(), self.pop()))
def emit_MULF(self):
self.push(self.builder.fmul(self.pop(), self.pop()))
# Binary / operator
def emit_DIVI(self):
right = self.pop()
left = self.pop()
self.push(self.builder.sdiv(left, right))
def emit_DIVF(self):
right = self.pop()
left = self.pop()
self.push(self.builder.fdiv(left, right))
# Conversion
def emit_ITOF(self):
self.push(self.builder.sitofp(self.pop(), float_type))
def emit_FTOI(self):
self.push(self.builder.fptosi(self.pop(), int_type))
# Comparison operators
def emit_LEI(self):
right = self.pop()
left = self.pop()
result = self.builder.icmp_signed('<=', left, right)
self.push(self.builder.zext(result, int_type))
def emit_LTI(self):
right = self.pop()
left = self.pop()
result = self.builder.icmp_signed('<', left, right)
self.push(self.builder.zext(result, int_type))
def emit_GEI(self):
right = self.pop()
left = self.pop()
result = self.builder.icmp_signed('>=', left, right)
self.push(self.builder.zext(result, int_type))
def emit_GTI(self):
right = self.pop()
left = self.pop()
result = self.builder.icmp_signed('>', left, right)
self.push(self.builder.zext(result, int_type))
def emit_EQI(self):
right = self.pop()
left = self.pop()
result = self.builder.icmp_signed('==', left, right)
self.push(self.builder.zext(result, int_type))
def emit_NEI(self):
right = self.pop()
left = self.pop()
result = self.builder.icmp_signed('!=', left, right)
self.push(self.builder.zext(result, int_type))
# Comparison operators
def emit_LEF(self):
right = self.pop()
left = self.pop()
result = self.builder.fcmp_ordered('<=', left, right)
self.push(self.builder.zext(result, int_type))
def emit_LTF(self):
right = self.pop()
left = self.pop()
result = self.builder.fcmp_ordered('<', left, right)
self.push(self.builder.zext(result, int_type))
def emit_GEF(self):
right = self.pop()
left = self.pop()
result = self.builder.fcmp_ordered('>=', left, right)
self.push(self.builder.zext(result, int_type))
def emit_GTF(self):
right = self.pop()
left = self.pop()
result = self.builder.fcmp_ordered('>', left, right)
self.push(self.builder.zext(result, int_type))
def emit_EQF(self):
right = self.pop()
left = self.pop()
result = self.builder.fcmp_ordered('==', left, right)
self.push(self.builder.zext(result, int_type))
def emit_NEF(self):
right = self.pop()
left = self.pop()
result = self.builder.fcmp_ordered('!=', left, right)
self.push(self.builder.zext(result, int_type))
# Bitwise operations
def emit_ANDI(self):
right = self.pop()
left = self.pop()
self.push(self.builder.and_(left, right))
def emit_ORI(self):
right = self.pop()
left = self.pop()
self.push(self.builder.or_(left, right))
# Print statements
def emit_PRINTI(self):
self.builder.call(self.runtime['_print_int'], [self.pop()])
def emit_PRINTF(self):
self.builder.call(self.runtime['_print_float'], [self.pop()])
def emit_PRINTB(self):
self.builder.call(self.runtime['_print_byte'], [self.pop()])
# Memory statements
def emit_GROW(self):
self.push(self.builder.call(self.runtime['_grow'], [self.pop()]))
def emit_PEEKI(self):
self.push(self.builder.call(self.runtime['_peeki'], [self.pop()]))
def emit_PEEKF(self):
self.push(self.builder.call(self.runtime['_peekf'], [self.pop()]))
def emit_PEEKB(self):
self.push(self.builder.call(self.runtime['_peekb'], [self.pop()]))
def emit_POKEI(self):
value = self.pop()
addr = self.pop()
self.builder.call(self.runtime['_pokei'], [addr, value])
def emit_POKEF(self):
value = self.pop()
addr = self.pop()
self.builder.call(self.runtime['_pokef'], [addr, value])
def emit_POKEB(self):
value = self.pop()
addr = self.pop()
self.builder.call(self.runtime['_pokeb'], [addr, value])
# Control flow
def emit_IF(self):
then_block = self.function.append_basic_block()
else_block = self.function.append_basic_block()
exit_block = self.function.append_basic_block()
self.builder.cbranch(self.builder.trunc(self.pop(), IntType(1)), then_block, else_block)
self.set_block(then_block)
self.blocks.append([then_block, else_block, exit_block])
def emit_ELSE(self):
if not self.block.is_terminated:
self.builder.branch(self.blocks[-1][2])
self.set_block(self.blocks[-1][1])
def emit_ENDIF(self):
if not self.block.is_terminated:
self.builder.branch(self.blocks[-1][2])
self.set_block(self.blocks[-1][2])
self.blocks.pop()
def emit_LOOP(self):
top_block = self.function.append_basic_block()
exit_block = self.function.append_basic_block()
self.builder.branch(top_block)
self.set_block(top_block)
self.blocks.append([top_block, exit_block])
def emit_CBREAK(self):
next_block = self.function.append_basic_block()
self.builder.cbranch(self.builder.trunc(self.pop(), IntType(1)), self.blocks[-1][1], next_block)
self.set_block(next_block)
def emit_ENDLOOP(self):
if not self.block.is_terminated:
self.builder.branch(self.blocks[-1][0])
self.set_block(self.blocks[-1][1])
self.blocks.pop()
def emit_RETURN(self):
self.builder.ret(self.pop())
def emit_CALL(self, name):
func = self.globals[name]
args = [self.pop() for _ in range(len(func.args))][::-1]
self.push(self.builder.call(func, args))
class GenerateLLVM(object):
def __init__(self):
# Perform the basic LLVM initialization. You need the following parts:
#
# 1. A top-level Module object
# 2. A Function instance in which to insert code
# 3. A Builder instance to generate instructions
#
# Note: For project 5, we don't have any user-defined
# functions so we're just going to emit all LLVM code into a top
# level function void main() { ... }. This will get changed later.
self.module = Module('module')
self.function = Function(self.module,
FunctionType(void_type, []),
name='main')
self.block = self.function.append_basic_block('entry')
self.builder = IRBuilder(self.block)
# Dictionary that holds all of the global variable/function declarations.
# Any declaration in the Gone source code is going to get an entry here
self.vars = ChainMap()
# Dictionary that holds all of the temporary registers created in
# the intermediate code.
self.temps = {}
self.blocks = { }
self.funcs = { }
self.current_func = None
# Initialize the runtime library functions (see below)
self.declare_runtime_library()
def declare_runtime_library(self):
# Certain functions such as I/O and string handling are often easier
# to implement in an external C library. This method should make
# the LLVM declarations for any runtime functions to be used
# during code generation. Please note that runtime function
# functions are implemented in C in a separate file gonert.c
self.runtime = {}
# Declare printing functions
self.runtime['_print_int'] = Function(self.module,
FunctionType(void_type, [int_type]),
name="_print_int")
self.runtime['_print_float'] = Function(self.module,
FunctionType(void_type, [float_type]),
name="_print_float")
self.runtime['_print_byte'] = Function(self.module,
FunctionType(void_type, [byte_type]),
name="_print_byte")
def generate_code(self, ircode):
# Given a sequence of SSA intermediate code tuples, generate LLVM
# instructions using the current builder (self.builder). Each
# opcode tuple (opcode, args) is dispatched to a method of the
# form self.emit_opcode(args)
# first collect all the blocks
for instr in ircode:
if instr[0] == 'BLK':
self.blocks[instr[1]] = self.function.append_basic_block(instr[1])
for opcode, *args in ircode:
if hasattr(self, 'emit_'+opcode):
getattr(self, 'emit_'+opcode)(*args)
else:
print('Warning: No emit_'+opcode+'() method')
# Add a return statement. Note, at this point, we don't really have
# user-defined functions so this is a bit of hack--it may be removed later.
self.builder.ret_void()
# ----------------------------------------------------------------------
# Opcode implementation. You must implement the opcodes. A few
# sample opcodes have been given to get you started.
# ----------------------------------------------------------------------
def emit_FUNCTION_END(self, f_name):
if not self.block.is_terminated:
self.builder.branch(self.return_block)
self.builder.position_at_end(self.return_block)
self.builder.ret(self.builder.load(self.return_var))
def emit_FUNCTION(self, f_name, arguments, return_type):
arguments = eval(arguments)
argtypes = list(map(lambda z: z[0], arguments))
f_func = Function(self.module,
FunctionType(
type_lookup[return_type],
argtypes
),
name=f_name)
self.current_func = f_func
self.block = self.current_func.append_basic_block('entry')
self.return_block = self.current_func.append_basic_block('return')
self.return_var = self.builder.alloca(type_lookup[return_type], name='return')
self.funcs[f_name] = self.current_func
def emit_CALL(self, f_name, sources, target):
#import pdb
#pdb.set_trace()
f = self.funcs[f_name]
self.temps[target] = self.builder.call(f,
[self.temps[r] for r in eval(sources)]
)
def emit_RETURN(self, var):
self.builder.ret(self.vars[var])
# Creation of literal values. Simply define as LLVM constants.
def emit_MOVI(self, value, target):
self.temps[target] = Constant(int_type, value)
def emit_MOVF(self, value, target):
self.temps[target] = Constant(float_type, value)
def emit_MOVB(self, value, target):
self.temps[target] = Constant(byte_type, value)
# Allocation of variables. Declare as global variables and set to
# a sensible initial value.
def emit_VARI(self, name):
var = GlobalVariable(self.module, int_type, name=name)
var.initializer = Constant(int_type, 0)
self.vars[name] = var
def emit_VARF(self, name):
var = GlobalVariable(self.module, float_type, name=name)
var.initializer = Constant(float_type, 0)
self.vars[name] = var
def emit_VARB(self, name):
var = GlobalVariable(self.module, byte_type, name=name)
var.initializer = Constant(byte_type, 0)
self.vars[name] = var
def emit_ALLOCI(self, name):
var = self.builder.alloca(IntType(32), name=name)
self.builder.store(Constant(int_type, 0), var)
self.vars[name] = var
# Load/store instructions for variables. Load needs to pull a
# value from a global variable and store in a temporary. Store
# goes in the opposite direction.
def emit_LOADI(self, name, target):
self.temps[target] = self.builder.load(self.vars[name], target)
emit_LOADB = emit_LOADI
emit_LOADF = emit_LOADI
def emit_STOREI(self, source, target):
self.builder.store(self.temps[source], self.vars[target])
emit_STOREF = emit_STOREI
emit_STOREB = emit_STOREI
# Binary + operator
def emit_ADDI(self, left, right, target):
self.temps[target] = self.builder.add(self.temps[left], self.temps[right], target)
# Binary + operator
def emit_ADDF(self, left, right, target):
self.temps[target] = self.builder.fadd(self.temps[left], self.temps[right], target)
# Binary - operator
def emit_SUBI(self, left, right, target):
self.temps[target] = self.builder.sub(self.temps[left], self.temps[right], target)
def emit_SUBF(self, left, right, target):
self.temps[target] = self.builder.fsub(self.temps[left], self.temps[right], target)
# Binary * operator
def emit_MULI(self, left, right, target):
self.temps[target] = self.builder.mul(self.temps[left], self.temps[right], target)
def emit_MULF(self, left, right, target):
self.temps[target] = self.builder.fmul(self.temps[left], self.temps[right], target)
# Binary / operator
def emit_DIVI(self, left, right, target):
self.temps[target] = self.builder.sdiv(self.temps[left], self.temps[right], target)
def emit_DIVF(self, left, right, target):
self.temps[target] = self.builder.fdiv(self.temps[left], self.temps[right], target)
def emit_AND(self, left, right, target):
self.temps[target] = self.builder.and_(self.temps[left], self.temps[right], target)
def emit_OR(self, left, right, target):
self.temps[target] = self.builder.or_(self.temps[left], self.temps[right], target)
def emit_CMPF(self, op, left, right, target):
temp = self.builder.fcmp_ordered(op, self.temps[left], self.temps[right], 'temp')
self.temps[target] = self.builder.zext(temp, int_type)
def emit_CMPI(self, op, left, right, target):
temp = self.builder.icmp_signed(op, self.temps[left], self.temps[right], 'temp')
self.temps[target] = self.builder.zext(temp, int_type)
def cast_to_int(self, value):
return self.builder.zext(value, int_type)
def cast_to_byte(self, value):
return self.builder.trunc(value, IntType(1))
# Print statements
def emit_PRINTI(self, source):
self.builder.call(self.runtime['_print_int'], [self.cast_to_int(self.temps[source])])
def emit_PRINTF(self, source):
self.builder.call(self.runtime['_print_float'], [self.temps[source]])
def emit_PRINTB(self, source):
self.builder.call(self.runtime['_print_byte'], [self.temps[source]])
# conditionals and branches
def emit_CJMP(self, target, true_branch, false_branch):
self.builder.cbranch(self.cast_to_byte(self.temps[target]), self.blocks[true_branch], self.blocks[false_branch])
def emit_JMP(self, target):
self.builder.branch(self.blocks[target])
def emit_BLK(self, target):
self.builder.position_at_end(self.blocks[target])
# Copy the argument to a local variable. For reasons, that are
# complicated, we copy the function argument to a local variable
# allocated on the stack (this makes mutations of the variable
# easier when looping).
n_var = builder.alloca(IntType(32), name='n')
builder.store(f_func.args[0], n_var)
# Allocate the result variable
total_var = builder.alloca(IntType(32), name='total')
builder.store(Constant(IntType(32), 0), total_var)
# Make a new basic-block for the loop test
loop_test_block = f_func.append_basic_block('looptest')
builder.branch(loop_test_block)
builder.position_at_end(loop_test_block)
# Perform the comparison
testvar = builder.icmp_signed('>', builder.load(n_var), Constant(IntType(32), 0))
# Make three blocks
loop_body_block = f_func.append_basic_block('loopbody')
loop_exit_block = f_func.append_basic_block('loopexit')
# Branch based on the test var
builder.cbranch(testvar, loop_body_block, loop_exit_block)
builder.position_at_end(loop_body_block)
tmp = builder.add(builder.load(total_var), builder.load(n_var))
builder.store(tmp, total_var)
def _build_wrapper(self, library, name):
"""
The LLVM IRBuilder code to create the gufunc wrapper.
The *library* arg is the CodeLibrary to which the wrapper should
be added. The *name* arg is the name of the wrapper function being
created.
"""
intp_t = self.context.get_value_type(types.intp)
fnty = self._wrapper_function_type()
wrapper_module = library.create_ir_module('_gufunc_wrapper')
func_type = self.call_conv.get_function_type(self.fndesc.restype,
self.fndesc.argtypes)
fname = self.fndesc.llvm_func_name
func = ir.Function(wrapper_module, func_type, name=fname)
func.attributes.add("alwaysinline")
wrapper = ir.Function(wrapper_module, fnty, name)
# The use of weak_odr linkage avoids the function being dropped due
# to the order in which the wrappers and the user function are linked.
wrapper.linkage = 'weak_odr'
arg_args, arg_dims, arg_steps, arg_data = wrapper.args
arg_args.name = "args"
arg_dims.name = "dims"
arg_steps.name = "steps"
arg_data.name = "data"
builder = IRBuilder(wrapper.append_basic_block("entry"))
loopcount = builder.load(arg_dims, name="loopcount")
pyapi = self.context.get_python_api(builder)
# Unpack shapes
unique_syms = set()
for grp in (self.sin, self.sout):
for syms in grp:
unique_syms |= set(syms)
sym_map = {}
for syms in self.sin:
for s in syms:
if s not in sym_map:
sym_map[s] = len(sym_map)
sym_dim = {}
for s, i in sym_map.items():
sym_dim[s] = builder.load(
builder.gep(arg_dims,
[self.context.get_constant(types.intp, i + 1)]))
# Prepare inputs
arrays = []
step_offset = len(self.sin) + len(self.sout)
for i, (typ, sym) in enumerate(
zip(self.signature.args, self.sin + self.sout)):
ary = GUArrayArg(self.context, builder, arg_args, arg_steps, i,
step_offset, typ, sym, sym_dim)
step_offset += len(sym)
arrays.append(ary)
bbreturn = builder.append_basic_block('.return')
# Prologue
self.gen_prologue(builder, pyapi)
# Loop
with cgutils.for_range(builder, loopcount, intp=intp_t) as loop:
args = [a.get_array_at_offset(loop.index) for a in arrays]
innercall, error = self.gen_loop_body(builder, pyapi, func, args)
# If error, escape
cgutils.cbranch_or_continue(builder, error, bbreturn)
builder.branch(bbreturn)
builder.position_at_end(bbreturn)
# Epilogue
self.gen_epilogue(builder, pyapi)
builder.ret_void()
# Link
library.add_ir_module(wrapper_module)
library.add_linking_library(self.library)