Local optimizator for 3-adress code that is in SSA format. Optimizator supports neutral elimination, constant folding, constant propagation and strenght reduction and runs those optimizations in a loop as long as there is something to optimize.
Local optimizator works for 3-address code that has the following syntax:
The language consists of declarations and IF / GOTO statements ( GOTO must go to some number ).
Code must be in SSA (Single Static Assigment) format ( 1 declaration per variable )
Operators that are supported :
+ - * / >> << - (unary)
x1 := 7
IF x1 < 5 GOTO 5
y := x1 * 8
x2 := y ^ 2
x3 := x2 + x2
z := 2 + 3
t := -x1 * 2
Run the following command
$ python3 blockCreator.py [path_to_file]
If path to file is not stated program uses test/test_examples/test.txt file that consists of instructions that demonstrate how each local optimization technique works and how they work together in a loop.
Local optimizator optimizes the code described above using the following steps
Optimizator splits the code into basic blocks by finding leader instructions
x1 := 7
IF x1 < 5 GOTO 5
--------------
y := x1 * 8
x2 := y ^ 2
--------------
x3 := x2 + x2
z := 2 + 3
t := -x1 * 2
Optimizator performs neutral elimination technique on each block
z := x + 0 => z := x
z := 0 + x => z := x
y := y - 0 => y := y
y := 0 - y => y := -y
z := 0 - (- z) => z := z
t := t * 0 => t := 0
t := 0 * t => t := 0
f := f * 1 => f := f
f := 1 * f => f := f
z := 0 / y => z := 0
t := t / 1 => t := t
z := z ^ 0 => z := 1
z := z ^ 1 => z := z
z := 1 ^ 5 => z := Optimizator performs constant folding technique on each instruction after neutral elimination
IF 3 < 2 GOTO 0 => (deleted instruction)
x := 2 + 3 => x := 5
x := 2 * 3 => x := 6
x := 6 / 3 => x := 2
x := 2 ^ 3 => x := 8
x := 2 - 3 => x := -1
IF 4 < 5 GOTO 1 => GOTO 1 ( conditionl jump => non conditional jump ) Optimizator performs strenght reduction technique on each instruction after constant folding and neutral elimination
y := y ^ 2 => y := y * y
t := y * 2 => y := y + y
t := 2 * (-t) => t := -t + -t
g := x * 16 => g << 4
g := x * 32 => g << 5
g := 7 * z => tmp_g := z << 3
g := tmp_g - z f := f * 33 => tmp_f := f << 5
f := tmp_f + f
Optimizator performs constant propagation technique on each block
x := 3 z := z + x => z := z + 3
x := 5
y := x + z => y := 5 + z
These optimizations are runnning in a loop until there is nothing more to optimize
This project is licensed under the MIT License - see the LICENSE.md file for details