phi_functions/lang.py

512 lines
16 KiB
Python

"""
This file contains the implementation of a simple interpreter of low-level
instructions. The interpreter takes a program, represented as its first
instruction, plus an environment, which is a stack of bindings. Bindings are
pairs of variable names and values. New bindings are added to the stack
whenever new variables are defined. Bindings are never removed from the stack.
In this way, we can inspect the history of state transformations caused by the
interpretation of a program. The difference between this file and the files of
same name in the previous lab is the presence of phi-functions. In other words,
this new language contains two extra instructions: phi-functions and phi-blocks.
The latter represents the set of phi-functions that exist at the beginning of
a basic block.
This file uses doctests all over. To test it, just run python 3 as follows:
"python3 -m doctest main.py". The program uses syntax that is excluive of
Python 3. It will not work with standard Python 2.
"""
from collections import deque
from abc import ABC, abstractmethod
class Env:
"""
A table that associates variables with values. The environment is
implemented as a stack, so that previous bindings of a variable V remain
available in the environment if V is overassigned.
Example:
>>> e = Env()
>>> e.set("a", 2)
>>> e.set("a", 3)
>>> e.get("a")
3
>>> e = Env({"b": 5})
>>> e.set("a", 2)
>>> e.get("a") + e.get("b")
7
"""
def __init__(s, initial_args={}):
s.env = deque()
for var, value in initial_args.items():
s.env.appendleft((var, value))
def get(self, var):
"""
Finds the first occurrence of variable 'var' in the environment stack,
and returns the value associated with it.
"""
val = next((value for (e_var, value) in self.env if e_var == var), None)
if val is not None:
return val
else:
raise LookupError(f"Absent key {var}")
def get_from_list(self, vars):
"""
Finds the first occurrence of any variable 'vr' in the list 'vars' that
has a binding in the environment, and returns the associated value.
Example:
>>> e = Env()
>>> e.set("b", 1)
>>> e.set("a", 2)
>>> e.set("b", 3)
>>> e.get_from_list(["b", "a"])
3
>>> e = Env()
>>> e.set("b", 1)
>>> e.set("a", 2)
>>> e.set("b", 3)
>>> e.set("a", 4)
>>> e.get_from_list(["b", "a"])
4
"""
# TODO: Implement this method
return 0
def set(s, var, value):
"""
This method adds 'var' to the environment, by placing the binding
'(var, value)' onto the top of the environment stack.
"""
s.env.appendleft((var, value))
def dump(s):
"""
Prints the contents of the environment. This method is mostly used for
debugging purposes.
"""
for var, value in s.env:
print(f"{var}: {value}")
class Inst(ABC):
"""
The representation of instructions. All that an instruction has, that is
common among all the instructions, is the next_inst attribute. This
attribute determines the next instruction that will be fetched after this
instruction runs. Also, every instruction has an index, which is always
different. The index is incremented whenever a new instruction is created.
"""
next_index = 0
def __init__(self):
self.nexts = []
self.preds = []
self.ID = Inst.next_index
Inst.next_index += 1
def add_next(self, next_inst):
self.nexts.append(next_inst)
next_inst.preds.append(self)
@classmethod
@abstractmethod
def definition(self):
raise NotImplementedError
@classmethod
@abstractmethod
def uses(self):
raise NotImplementedError
def get_next(self):
if len(self.nexts) > 0:
return self.nexts[0]
else:
return None
class Phi(Inst):
"""
A Phi-Function is an abstract notation used to facilitate the implementation
of static analyses. They were not really conceived to have a dynamic
semantics. Nevertheless, we can still interpret programs containing
phi-functions. A possible semantics of 'a = phi(a0, a1, a2)' is to
recover, from the environment, the first binding of either a0, a1 or a2.
If our program were in the so-called "Conventional-SSA Form", this
semantics would be perfect. But our program is not in such a format, and
we might have issues with swaps, for instance. That's why we shall use
phi-blocks to implement phi-functions. All the same, you can still write
programs using phi-functions without using phi-blocks, as long as variables
that are related by phi-functions do not have overlapping live ranges.
Example:
>>> a = Phi("a", ["b0", "b1", "b2"])
>>> e = Env()
>>> e.set("b0", 1)
>>> e.set("b1", 3)
>>> a.eval(e)
>>> e.get("a")
3
>>> a = Phi("a", ["b0", "b1"])
>>> e = Env()
>>> e.set("b1", 3)
>>> e.set("b0", 1)
>>> a.eval(e)
>>> e.get("a")
1
"""
def __init__(s, dst, args):
s.dst = dst
s.args = args
super().__init__()
def definition(s):
return s.dst
def uses(s):
return s.args
def eval(s, env):
"""
If the program were in Conventional-SSA form, then we could correctly
implement the semantics of phi-functions simply retrieving the first
occurrence of each variable in the list of uses. However, notice what
would happen with swaps:
>>> a0 = Phi("a0", ["a1", "a0"])
>>> a1 = Phi("a1", ["a0", "a1"])
>>> e = Env()
>>> e.set("a0", 1)
>>> e.set("a1", 3)
>>> a0.eval(e)
>>> a1.eval(e)
>>> e.get("a0") - e.get("a1")
0
In the example above, we would like to evaluate the two phi-functions in
parallel, e.g.: (a0, a1) = (a0:1, a1:3). In this way, after the
evaluation, we would like to have a0 == 3 and a1 == 1. However, there is
no way we can do it: our phi-functions are evaluated once at a time! The
problem is that variables a0 and a1 are defined by different
phi-functions, but they have overlapping live ranges. So, this
program is not in conventional SSA-form (as per Definition 1 in the
paper 'SSA Elimination after Register Allocation' - 2009).
"""
env.set(s.dst, env.get_from_list(s.uses()))
def __str__(self):
use_list = ", ".join(self.uses())
inst_s = f"{self.ID}: {self.dst} = phi[{use_list}]"
pred_s = f"\n P: {', '.join([str(inst.ID) for inst in self.preds])}"
next_s = f"\n N: {self.nexts[0].ID if len(self.nexts) > 0 else ''}"
return inst_s + pred_s + next_s
class PhiBlock(Inst):
"""
PhiBlocks implement a correct semantics for groups of phi-functions. A
phi-block groups a number of phi-functions as a matrix. Once a phi-block
is evaluated, all the values in a given column of this matrix are read and
saved, and then the definitions are updated --- all in parallel. To see a
more detailed explanation of this semantics, please, refer to Section 3 of
the paper 'SSA Elimination after Register Allocation'. In particular, take
a look into Figure 1 of that paper.
Example:
>>> a0 = Phi("a0", ["a0", "a1"])
>>> a1 = Phi("a1", ["a1", "a0"])
>>> aa = PhiBlock([a0, a1], [10, 31])
>>> e = Env()
>>> e.set("a0", 1)
>>> e.set("a1", 3)
>>> aa.eval(e, 10)
>>> e.get("a0") - e.get("a1")
-2
>>> a0 = Phi("a0", ["a0", "a1"])
>>> a1 = Phi("a1", ["a1", "a0"])
>>> aa = PhiBlock([a0, a1], [10, 31])
>>> e = Env()
>>> e.set("a0", 1)
>>> e.set("a1", 3)
>>> aa.eval(e, 31)
>>> e.get("a0") - e.get("a1")
2
"""
def __init__(self, phis, selector_IDs):
"""
A phi-block represents an M*N matrix, where each one of the M lines is
a phi-function, and each phi-function reads from N different parameters.
Each one of these N columns is associated with a 'selector', which is
the ID of the instruction that leads to that parallel assignment.
Examples:
>>> a0 = Phi("a0", ["a0", "a1"])
>>> a1 = Phi("a1", ["a1", "a0"])
>>> aa = PhiBlock([a0, a1], [10, 31])
>>> sorted(aa.selectors.items())
[(10, 0), (31, 1)]
>>> a0 = Phi("a0", ["a0", "a1"])
>>> a1 = Phi("a1", ["a1", "a0"])
>>> aa = PhiBlock([a0, a1], [10, 31])
>>> sorted([phi.definition() for phi in aa.phis])
['a0', 'a1']
"""
self.phis = phis
# TODO: implement the rest of this method
# here...
#########################################
super().__init__()
def definition(self):
"""
We consider that a phi-block defines multiple variables. These are the
variables assignment by the phi-functions that the phi-block contains.
Example:
>>> a0 = Phi("a0", ["a0", "a1"])
>>> a1 = Phi("a1", ["a1", "a0"])
>>> aa = PhiBlock([a0, a1], [10, 31])
>>> sorted(aa.definition())
['a0', 'a1']
"""
return [phi.definition() for phi in self.phis]
def uses(self):
"""
The uses of a phi-block are all the variables used by the phi-functions
that it contains. Notice that we don't need this method for anything; it
is here rather to help understand the structure of phi-blocks.
Example:
>>> a0 = Phi("a0", ["a0", "x"])
>>> a1 = Phi("a1", ["y", "a0"])
>>> aa = PhiBlock([a0, a1], [10, 31])
>>> sorted(aa.uses())
['a0', 'a0', 'x', 'y']
"""
return sum([phi.uses() for phi in self.phis], [])
def eval(self, env, PC):
# TODO: Read all the definitions
# TODO: Assign all the uses:
def __str__(self):
block_str = "\n".join([str(phi) for phi in self.phis])
return f"PHI_BLOCK [\n{block_str}\n]"
class BinOp(Inst):
"""
The general class of binary instructions. These instructions define a
value, and use two values. As such, it contains a routine to extract the
defined value, and the list of used values.
"""
def __init__(s, dst, src0, src1):
s.dst = dst
s.src0 = src0
s.src1 = src1
super().__init__()
@classmethod
@abstractmethod
def get_opcode(self):
raise NotImplementedError
def definition(s):
return set([s.dst])
def uses(s):
return set([s.src0, s.src1])
def __str__(self):
op = self.get_opcode()
inst_s = f"{self.ID}: {self.dst} = {self.src0}{op}{self.src1}"
pred_s = f"\n P: {', '.join([str(inst.ID) for inst in self.preds])}"
next_s = f"\n N: {self.nexts[0].ID if len(self.nexts) > 0 else ''}"
return inst_s + pred_s + next_s
class Add(BinOp):
"""
Example:
>>> a = Add("a", "b0", "b1")
>>> e = Env({"b0":2, "b1":3})
>>> a.eval(e)
>>> e.get("a")
5
>>> a = Add("a", "b0", "b1")
>>> a.get_next() == None
True
"""
def eval(self, env):
env.set(self.dst, env.get(self.src0) + env.get(self.src1))
def get_opcode(self):
return "+"
class Mul(BinOp):
"""
Example:
>>> a = Mul("a", "b0", "b1")
>>> e = Env({"b0":2, "b1":3})
>>> a.eval(e)
>>> e.get("a")
6
"""
def eval(s, env):
env.set(s.dst, env.get(s.src0) * env.get(s.src1))
def get_opcode(self):
return "*"
class Lth(BinOp):
"""
Example:
>>> a = Lth("a", "b0", "b1")
>>> e = Env({"b0":2, "b1":3})
>>> a.eval(e)
>>> e.get("a")
True
"""
def eval(s, env):
env.set(s.dst, env.get(s.src0) < env.get(s.src1))
def get_opcode(self):
return "<"
class Geq(BinOp):
"""
Example:
>>> a = Geq("a", "b0", "b1")
>>> e = Env({"b0":2, "b1":3})
>>> a.eval(e)
>>> e.get("a")
False
"""
def eval(s, env):
env.set(s.dst, env.get(s.src0) >= env.get(s.src1))
def get_opcode(self):
return ">="
class Bt(Inst):
"""
This is a Branch-If-True instruction, which diverts the control flow to the
'true_dst' if the predicate 'pred' is true, and to the 'false_dst'
otherwise.
Example:
>>> e = Env({"t": True, "x": 0})
>>> a = Add("x", "x", "x")
>>> m = Mul("x", "x", "x")
>>> b = Bt("t", a, m)
>>> b.eval(e)
>>> b.get_next() == a
True
"""
def __init__(s, cond, true_dst=None, false_dst=None):
super().__init__()
s.cond = cond
s.nexts = [true_dst, false_dst]
if true_dst != None:
true_dst.preds.append(s)
if false_dst != None:
false_dst.preds.append(s)
def definition(s):
return set()
def uses(s):
return set([s.cond])
def add_true_next(s, true_dst):
s.nexts[0] = true_dst
true_dst.preds.append(s)
def add_next(s, false_dst):
s.nexts[1] = false_dst
false_dst.preds.append(s)
def eval(s, env):
"""
The evaluation of the condition sets the next_iter to the instruction.
This value determines which successor instruction is to be evaluated.
Any values greater than 0 are evaluated as True, while 0 corresponds to
False.
"""
if env.get(s.cond):
s.next_iter = 0
else:
s.next_iter = 1
def get_next(s):
return s.nexts[s.next_iter]
def __str__(self):
inst_s = f"{self.ID}: bt {self.cond}"
pred_s = f"\n P: {', '.join([str(inst.ID) for inst in self.preds])}"
next_s = f"\n NT:{self.nexts[0].ID} NF:{self.nexts[1].ID}"
return inst_s + pred_s + next_s
def interp(instruction, environment, PC=0):
"""
This function evaluates a program until there is no more instructions to
evaluate. Notice that, in contrast to the previous labs, the interpreter
now receives three arguments. The third argument is necessary to implement
the correct semantics of phi-functions using phi-blocks. This argument can
be used to select the correct parallel copy that a PhiBlock implements.
Parameters:
-----------
instruction: the instruction that will be interpreted
environment: the list that associates variable names with their values
PC: the identifier of the last instruction that was interpreted.
Example:
>>> env = Env({"m": 3, "n": 2, "zero": 0})
>>> m_min = Add("answer", "m", "zero")
>>> n_min = Add("answer", "n", "zero")
>>> p = Lth("p", "n", "m")
>>> b = Bt("p", n_min, m_min)
>>> p.add_next(b)
>>> interp(p, env).get("answer")
2
"""
if instruction:
print("----------------------------------------------------------")
print(instruction)
environment.dump()
if isinstance(instruction, PhiBlock):
# TODO: implement this part:
pass
else:
# TODO: implement this part:
pass
return interp(instruction.get_next(), environment, instruction.ID)
else:
return environment