shivansh / gogo

Go to MIPS compiler, written in Go. Course project for Compiler Design (CS335).

gogo

Go to MIPS compiler, written in Go.

Gopher vector imported from egonelbre/gophers .

---

Components

Component	Demo
Token generation / Lexer	test1.out
Parser	struct.go ➡️ struct.html
IR generation	scope.go ➡️ scope.ir
Code generation	pascalTriangle.ir ➡️ pascalTriangle.asm

Setting up

Run ./scripts/setup.sh from the root directory of the project to set up the pre-commit git hooks.

Dependencies

• gocc

Build

The following should generate relevant binaries inside the directory bin -

make

The generated binary bin/gogo can be used as follows -

Usage: gogo (-r | -r2s | -s) <filename>
  -p	Generates rightmost derivations used in bottom-up parsing
  -r	Generates IR instructions from go program
  -r2s  Generates the MIPS assembly from IR
  -s	Generates MIPS assembly from go program

NOTE: The generated MIPS assembly has been tested to work on SPIM MIPS32 simulator.

Testing

The tests can be built via -

make test

gogo's Issues

Figure out data types

Currently every variable and value is being treated as a string, and the type information is stored in SrcVar.Typ ; Instead of this resolve types when reading from IR file itself, and store for e.g. 1 as int instead of string (which is the case currently). This update should be done in the lexing step too where everything was also a string despite of the actual data type. This will be useful in the parsing step.

SrcVar should be renamed to something like SymInfo and each entity read from the IR file should point to a SymInfo object where the information about that entity (data type, scoping info etc.) will be located.

Handle function calls

The return value of a function should be stored somewhere for cases like -

val = func()

Something in lines of callee/caller saved registers.

parser: test failure when parsing slices

package parser

type Node struct {
	n     int
	child []*Node
}

func main() {
	fmt.Println("Hello, World!")
}

The offending line number is 5.

2018/03/19 11:56:02 Error in S1604: identifier(10,Node), Pos(offset=55, line=5, column=13), expected one of: ( * kwdFunc [ type kwdStruct 

Handle scopes

Support for comments in the (handwritten) IR file

What I have in mind is something like this, consider an example (handwritten) IR file -

0, #, This test demonstrates register re-use when a variable is already loaded.
1, label, main
2, =, v1, 1
3, =, v2, 2
3, =, v1, 4, #, should update the register in which v1 is already present instead of re-allocating
4, ret,

The following diff demonstrates the current and the expected behavior for the generated assembly (in term of comments which were introduced) -

diff --git a/current.asm b/expected.asm
index 7967333..85a4f71 100644
--- a/current.asm
+++ b/expected.asm
@@ -1,3 +1,4 @@
+; This test demonstrates register re-use when a variable is already loaded.
        .data
 v1:    .word   0
 v2:    .word   0
@@ -8,7 +9,8 @@ main:
        li, $t2, 1              ; v1 -> {reg: $t2, mem: $t1}
        la $t3, v2
        li, $t4, 2              ; v2 -> {reg: $t4, mem: $t3}
-       li, $t2, 4
+       li, $t2, 4              ; should update the register in which v1 is
+                               ; already present instead of re-allocating.
 
        ; Store variables back into memory
        sw $t2, ($t1)

Create symbol tables