Assignment 2: Boa:   Adding new operators
1 The Boa Language
1.1 Concrete Syntax
1.2 Abstract Syntax
1.3 Semantics
2 Starter code for this assignment
3 Implementing a Compiler for Boa
3.1 Checking for scoping problems
3.2 Converting to Sequential Form
3.3 Compilation
4 Recommendations
5 Testing the Compiler
6 Running main
7 List of Deliverables
8 Grading Standards
9 Submission
10 Additional sequential form examples
8.6

Assignment 2: Boa: Adding new operators

Due: Thursday 09/30 at 9:00pm

git clone

In this compiler, you’ll enhance your existing compiler with Binary Operators and Arithmetic. Click here for scary snake picture!

1 The Boa Language

1.1 Concrete Syntax

The concrete syntax of Boa is:

‹expr›: | let ‹bindings› in ‹expr› | if ‹expr› : ‹expr› else: ‹expr› | ‹binop-expr› ‹binop-expr›: | NUMBER | IDENTIFIER | add1 ( ‹expr› ) | sub1 ( ‹expr› ) | ‹expr› + ‹expr› | ‹expr› - ‹expr› | ‹expr› * ‹expr› | ( ‹expr› ) ‹bindings›: | IDENTIFIER = ‹expr› | IDENTIFIER = ‹expr› , ‹bindings›

As in Adder, a Let can have one or more bindings.

1.2 Abstract Syntax

#[derive(Clone, Debug, PartialEq, Eq)]
pub enum Exp<Ann> {
    Num(i64, Ann),
    Var(String, Ann),
    Prim1(Prim1, Box<Exp<Ann>>, Ann),
    Prim2(Prim2, Box<Exp<Ann>>, Box<Exp<Ann>>, Ann),
    Let { bindings: Vec<(String, Exp<Ann>)>, // new binding declarations
          body: Box<Exp<Ann>>,  // the expression in which the new variables are bound
          ann: Ann
    },
    If { cond: Box<Exp<Ann>>,
         thn: Box<Exp<Ann>>,
         els: Box<Exp<Ann>>,
         ann: Ann
    },
}

#[derive(Copy, Clone, Debug, PartialEq, Eq)]
pub enum Prim1 {
    Add1,
    Sub1,
}

#[derive(Copy, Clone, Debug, PartialEq, Eq)]
pub enum Prim2 {
    Add,
    Sub,
    Mul,
}

1.3 Semantics

In addition to the semantics of Adder, we now have infix binary operators (addition, subtraction and multiplication), that are evaluated leftmost-innermost first (i.e., the standard left-to-right order that obeys parentheses), and conditional expressions. An Exp::If expression evaluates its condition, then evaluates its then-branch if the condition is non-zero, and evaluates its else-branch if the condition was zero.

To compile these expressions, we need a few more assembly instructions:

#[derive(Clone, Debug, PartialEq, Eq)]
pub enum Instr {
    Mov(MovArgs),

    Add(BinArgs),
    Sub(BinArgs),
    IMul(BinArgs),
    Cmp(BinArgs),

    Label(String),

    Jmp(String),
    Je(String),
    Jne(String),
    Jl(String),
    Jle(String),
    Jg(String),
    Jge(String),
}

Additionally, I have added another "work register" r15 to the registers:

#[derive(Copy, Clone, Debug, PartialEq, Eq)]
pub enum Reg {
    Rax,
    Rsp,
    R15,
}

You will likely find this extra register useful for implementing binary operations on constants, since you need to put 64-bit constants into a register before using add/sub/imul.

The sub and imul instructions are analogous to add, that take two arguments, apply their respective operations, and place their results in RAX. Be sure to use imul (signed multiplication) rather than mul (unsigned). Labels let us name the first of a sequence of instructions, akin to how we label start_here: to begin our code1Technically, labels are not instructions, so Instr is a bit mis-named.. The cmp instruction compares its two arguments, and sets some bits of the RFLAGS to tell if the arguments were equal, less than, greater than, etc. Rather than directly manipulating this register, we test the value of these flags with the jump instructions: jne will jump control to the named label if the flags mean NOT-EQUAL, and je will jump control to the named label when the flags mean EQUAL, etc.. Finally, jmp will unconditionally jump to the named label.

2 Starter code for this assignment

You’ve been given a starter codebase that has several pieces of infrastructure, mostly the same as before. I recommend you take a look at the following:

All of your edits — which will be to write the compiler for Boa, and test it — will happen in compile.rs and tests/examples.rs.

3 Implementing a Compiler for Boa

Again, the primary task of writing the Boa compiler is simple to state: take an instance of the Exp datatype and turn it into a list of assembly instructions. But since we now have more complicated expressions, we need to worry about simplifying the program, first.

3.1 Checking for scoping problems

  1. Extend the function check_prog<Span>(e: &Exp<Span>) -> Result<(), CompileErr<Span>> to support the new cases of expressions.

3.2 Converting to Sequential Form

Sequential Form asserts that throughout a program, any operator expression contains arguments that are immediate: that is, are numbers or identifiers, and therefore don’t perform any computation of their own. Additionally, we can think of the decision in an If expression to be an operation, so we also need to ensure that the condition is immediate.

  1. Design a function sequentialize(e: &Exp<u32>) -> SeqExp<()> that takes a tagged expression and produces a new expression that is in Sequential Form. You can assume that all of the u32 annotations in the input are unique. Because this program will include some new expressions, we’re willing to discard any decorations we had on the expression, which explains why the input could be decorated with any type of information, but the output will just be decorated with unit values.

When you need to generate fresh names (i.e., unique names that aren’t used anywhere in the expression), a useful strategy is to generate names of the form format!("#{}_{}", reason, tag), where reason is "prim1", "prim2_1", "prim2_2", "if", etc., and tag is the annotation on the expression. So if you need to generate a variable and your input expression is Exp::Prim1(op, e, 7), you can use the name "#prim1_7". This is only a suggestion, you may use whatever strategy you like in your compiler.

Additional examples of sequentialize are given below.

3.3 Compilation

{

  1. Adapt your compile_to_instrs function from Adder to compile the sequential expression forms in Boa. This means refactoring the similar cases as well as adding new support for If and Prim2. Remember that a simple invariant is for the code outputted by compile_to_instrs to always leave its answer in rax; this invariant is not the most efficient way to compile but will make it easier to get correct code

}

The starter code includes an extended compile_to_string function to invoke your functions and appropriately tag the ASTs with unique identifiers for you to use at the right moments.

4 Recommendations

Here’s an order in which you could consider tackling the implementation:

5 Testing the Compiler

As with Adder, we will have integration tests in tests/examples.rs with example files in the examples/ directory. You can (and should!) re-use your examples from the adder homework. Over the course of the semester you should accumulate a large amount of example programs that will help greatly when we start adding more complex features and optimizations.

Additionally, you may find it beneficial to unit test your sequentialize function in the provided submodule of compile.rs.

6 Running main

Running your own programs is the same as with Adder, except by convention we’ll use the .boa file extension.

7 List of Deliverables

8 Grading Standards

For this assignment, you will be graded on whether your code implements the specification (functional correctness).

9 Submission

Wait! Please read the assignment again and verify that you have not forgotten anything!

Please submit your homework on gradescope by the above deadline.

10 Additional sequential form examples

To address some recurring questions, here are some additional examples of sequentialization.

Given the boa program

let x = add1(2) in x

The most straightforward sequentialization algorithm will produce

let x = (let tmp = 2 in add1(tmp)) in x
where tmp can be whatever variable name you generate that is guaranteed to be different from all others.

However, notice that the original program was already in sequential form, so the temporary variable tmp is not truly necessary. So another valid sequentialization would be to return the program unchanged:

let x = add1(2) in x

Your sequentialization function can produce either one, just make sure your tests align with the strategy that you choose.

Here are some more examples.

let x1 = 1, x2 = 2 in
x1 + x2
can be sequentialized to
let x1 = 1 in
let x2 = 2 in
let tmp0 = x1 in
let tmp1 = x2 in
tmp0 + tmp1
or, without generating unnecessary temporaries:
let x1 = 1 in
let x2 = 2 in
x1 + x2

Next,

(2 * 9) + (18 - 3)
can be sequentialized to
let tmp0 = (let tmp1 = 2 in let tmp2 = 9 in tmp1 * tmp2) in
let tmp3 = (let tmp4 = 18 in let tmp5 = 3 in tmp4 - tmp5) in
tmp0 + tmp3
or, without generating unnecessary temporaries:
let tmp0 = 2 * 9 in
let tmp3 = 18 - 3 in
tmp0 + tmp3

Finally, an example with if:

3 + (if 5: add1(6) else: 7)
can be sequentialized to
let tmp0 = 3 in
let tmp1 = (let tmp2 = 5 in
            if tmp2:
              let tmp3 = 6 in add1(tmp3)
            else:
              7) in
tmp0 + tmp1
or without unnecessary temporaries:
let tmp1 = (if 5:
              add1(6)
            else:
              7) in
3 + tmp1

1Technically, labels are not instructions, so Instr is a bit mis-named.