filletofish / compiler-touch

Sample of building control flow graph, SSA form, phi nodes, LLVM IR for imagined language.

Compiler-touch

Simple compiler frontend to generate intermediate representation for simple language. The languange consists of integers, mutating variables, cycles and condition statements.

Two modes available:

• Generating my own intermediate presentation with features:

  • Building Control Flow Graph
  • Using SSA Form
  • Using Phi Nodes

• Generating LLVM intermediate presentation with power of LLVM Library.

Example:

Here is simple code written with our language.

Code:

a = 3
b = 0

for i = 1, b in
    a = a + 3

if b then
    a = a - 2
else
    a = 0

Control flow graph:

Its intermediate representation (using my own statements):

bb #0 entry
		preds: 
		succs: bb #1 loop_cond 
		dominatedBy: 
	a_0 = 3
	b_0 = 0
	i_0 = 1
	branch to: bb #1 loop_cond

bb #1 loop_cond
		preds: bb #0 entry bb #2 loop_body 
		succs: bb #2 loop_body bb #3 loop_cont 
		dominatedBy: bb #0 entry
	i_1 = [i_0 bb #0 entry; i_2 bb #2 loop_body; ]
	a_1 = [a_0 bb #0 entry; a_2 bb #2 loop_body; ]
	branch on: b_0 - i_1 to: bb #2 loop_body or: bb #3 loop_cont

bb #2 loop_body
		preds: bb #1 loop_cond 
		succs: bb #1 loop_cond 
		dominatedBy: bb #1 loop_cond
	a_2 = a_1 + 3
	i_2 = i_1 + 1
	branch to: bb #1 loop_cond

bb #3 loop_cont
		preds: bb #1 loop_cond 
		succs: bb #5 else bb #4 then 
		dominatedBy: bb #1 loop_cond
	branch on: b_0 to: bb #5 else or: bb #4 then

bb #4 then
		preds: bb #3 loop_cont 
		succs: bb #6 if_cont 
		dominatedBy: bb #3 loop_cont
	a_3 = a_1 - 2
	branch to: bb #6 if_cont

bb #5 else
		preds: bb #3 loop_cont 
		succs: bb #6 if_cont 
		dominatedBy: bb #3 loop_cont
	a_4 = 0
	branch to: bb #6 if_cont

bb #6 if_cont
		preds: bb #4 then bb #5 else 
		succs: 
		dominatedBy: bb #3 loop_cont
	a_5 = [a_3 bb #4 then; a_4 bb #5 else; ]

Its intermediate representation produced with LLVM Library:

; ModuleID = 'My Module'
source_filename = "My Module"

define common i32 @main() {
entrypoint:
  %a = alloca i32
  store i32 3, i32* %a
  %b = alloca i32
  store i32 0, i32* %b
  %i = alloca i32
  store i32 1, i32* %i
  br label %loopCoonditionBB

loopCoonditionBB:                                 ; preds = %loop, %entrypoint
  %b1 = load i32, i32* %b
  %i2 = load i32, i32* %i
  %loopcond = icmp slt i32 %i2, %b1
  br i1 %loopcond, label %loop, label %afterloop

loop:                                             ; preds = %loopCoonditionBB
  %a3 = load i32, i32* %a
  %addtmp = fadd i32 %a3, 3
  store i32 %addtmp, i32* %a
  %i4 = load i32, i32* %i
  %nextvar = fadd i32 %i4, 1
  store i32 %nextvar, i32* %i
  br label %loopCoonditionBB

afterloop:                                        ; preds = %loopCoonditionBB
  %b5 = load i32, i32* %b
  %ifcond = icmp ne i32 %b5, 0
  br i1 %ifcond, label %then, label %else

then:                                             ; preds = %afterloop
  %a6 = load i32, i32* %a
  %subtmp = fsub i32 %a6, 2
  store i32 %subtmp, i32* %a
  br label %ifcont

else:                                             ; preds = %afterloop
  store i32 0, i32* %a
  br label %ifcont

ifcont:                                           ; preds = %else, %then
  %iftmp = phi i32 [ %subtmp, %then ], [ 0, %else ]
  ret i32 %iftmp
}

What is IR?

Wiki article

Ok, then. With LLVM installed you can look at IR of C language by yourself. Just write simple C program.

Use flag -emit-llvm and level of optimization -O0 (from 0 to 3): clang -S -emit-llvm -O0 helloworld.c

Now hello.ll contains the IR.

Dependencies:

• llvm 4.0
• clang

Building:

./build.sh

Usage:

Printing intermediate representation:

./compiler <program.txt

Use cmd argument -gv for printing graph viz presentation

Use cmd argument -llvm for llvm mode

Sources:

• Documentation
• LLVM Tuturial
• Article by IBM