Fixed merge conflict

.gitignore

@@ -12,3 +12,5 @@ parsetab.py
 coverage/
 build/
 src/Makefile_old
+in.s
+out.s

Makefile

 BUILD=build/
-CLEAN=src/*.pyc src/optimize/*.pyc
+CLEAN=*.pyc src/*.pyc src/optimize/*.pyc parser.out parsetab.py
 
 # Fix pdflatex search path
-TGT_DIR :=
-TGT_DOC :=
+TGT_DIR := report
 
 # Default target is 'all'. The 'build' target is defined here so that all
 # sub rules.mk can add prerequisites to the 'build' target.
 all:
 build:
 
-d := tests/
+d := report/
 include base.mk
 include $(d)/rules.mk
 
-.PHONY: doc
+d := tests/
+include base.mk
+include $(d)/rules.mk
 
-all: doc build
+all: report
 
 clean:
 	rm -rf $(CLEAN)

benchmarks/optimized/pi.s

-	.file	1 "pi.c"
-# GNU C 2.7.2.3 [AL 1.1, MM 40, tma 0.1] SimpleScalar running sstrix compiled by GNU C
-# Cc1 defaults:
-# -mgas -mgpOPT
-# Cc1 arguments (-G value = 8, Cpu = default, ISA = 1):
-# -quiet -dumpbase -o
-gcc2_compiled.:
-__gnu_compiled_c:
-	.rdata
-	.align	2
-$LC0:
-	.ascii	"Usage: %s <iterations>\n\000"
-	.sdata
-	.align	2
-$LC3:
-	.ascii	"%.10f\n\000"
-	.align	3
-$LC1:
-	.word	0xffc00000		# 2147483647
-	.word	0x41dfffff
-	.align	3
-$LC2:
-	.word	0x00000000		# 1
-	.word	0x3ff00000
-	.align	3
-$LC4:
-	.word	0x00000000		# 4
-	.word	0x40100000
-	.text
-	.align	2
-	.globl	main
-	.extern	stderr, 4
-	.text
-	.loc	1 5
-	.ent	main
-main:
-	.frame	$fp,56,$31		# vars= 32, regs= 2/0, args= 16, extra= 0
-	.mask	0xc0000000,-4
-	.fmask	0x00000000,0
-	subu	$sp,$sp,56
-	sw	$31,52($sp)
-	sw	$fp,48($sp)
-	move	$fp,$sp
-	sw	$4,56($fp)
-	sw	$5,60($fp)
-	jal	__main
-	sw	$0,24($fp)
-	lw	$2,56($fp)
-	li	$3,0x00000002		# 2
-	beq	$2,$3,$L2
-	lw	$2,60($fp)
-	lw	$4,stderr
-	la	$5,$LC0
-	lw	$6,0($2)
-	jal	fprintf
-	move	$4,$0
-	jal	exit
-$L2:
-	lw	$3,60($fp)
-	addu	$2,$3,4
-	lw	$4,0($2)
-	jal	atoi
-	sw	$2,20($fp)
-	li	$4,0x00000001		# 1
-	jal	srandom
-	sw	$0,16($fp)
-$L3:
-	lw	$2,16($fp)
-	lw	$3,20($fp)
-	slt	$2,$2,$3
-	bne	$2,$0,$L6
-	j	$L4
-$L6:
-	jal	random
-	mtc1	$2,$f0
-	#nop
-	cvt.d.w	$f0,$f0
-	l.d	$f2,$LC1
-	div.d	$f0,$f0,$f2
-	s.d	$f0,32($fp)
-	jal	random
-	mtc1	$2,$f0
-	#nop
-	cvt.d.w	$f0,$f0
-	l.d	$f2,$LC1
-	div.d	$f0,$f0,$f2
-	s.d	$f0,40($fp)
-	l.d	$f0,32($fp)
-	l.d	$f2,32($fp)
-	mul.d	$f0,$f0,$f2
-	l.d	$f2,40($fp)
-	l.d	$f4,40($fp)
-	mul.d	$f2,$f2,$f4
-	add.d	$f0,$f0,$f2
-	l.d	$f2,$LC2
-	c.le.d	$f0,$f2
-	bc1f	$L7
-	lw	$3,24($fp)
-	addu	$2,$3,1
-	move	$3,$2
-	sw	$3,24($fp)
-$L7:
-$L5:
-	lw	$3,16($fp)
-	addu	$2,$3,1
-	move	$3,$2
-	sw	$3,16($fp)
-	j	$L3
-$L4:
-	l.s	$f0,24($fp)
-	#nop
-	cvt.d.w	$f0,$f0
-	l.s	$f2,20($fp)
-	#nop
-	cvt.d.w	$f2,$f2
-	div.d	$f0,$f0,$f2
-	l.d	$f2,$LC4
-	mul.d	$f0,$f0,$f2
-	la	$4,$LC3
-	dmfc1	$6,$f0
-	jal	printf
-	li	$2,0x00000001		# 1
-	j	$L1
-$L1:
-	move	$sp,$fp			# sp not trusted here
-	lw	$31,52($sp)
-	lw	$fp,48($sp)
-	addu	$sp,$sp,56
-	j	$31
-	.end	main
\ No newline at end of file

benchmarks/wiki.c

+#include <stdio.h>
+
+int main(void)
+{
+    int a = 3, b = 5, d = 5, x = 100;
+    int c;
+    if (a > b)
+    {
+        int c = a + b;
+        d = 2;
+    }
+    
+    c = 4;
+    
+    return b * d + c;
+}

main.py

 #!/usr/bin/python
 from src.parser import parse_file
 from src.optimize import optimize
-from src.writer import write_statements
+
 
 if __name__ == '__main__':
     from sys import argv, exit
 
     if len(argv) < 2:
-        print 'Usage: python %s FILE' % argv[0]
+        print 'Usage: python %s SOURCE_FILE [ OUT_FILE [ SOURCE_OUT_FILE ] ]' \
+                % argv[0]
         exit(1)
 
-    # Parse File
-    original = parse_file(argv[1])
-    optimized = optimize(original, verbose=1)
+    # Parse file
+    program = parse_file(argv[1])
+    program.debug = True
+
+    if len(argv) > 3:
+        # Save input assembly in new file for easy comparison
+        program.save(argv[3])
+
+    optimize(program, verbose=1)
 
     if len(argv) > 2:
         # Save output assembly
-        out = write_statements(optimized)
-        f = open(argv[2], 'w+')
-        f.write(out)
-        f.close()
+        program.save(argv[2])

report/Makefile

 RM=rm -rf
 
-all: report.pdf
+report: report.pdf
 
 %.pdf: %.tex
 	pdflatex $^

report/rules.mk

+CLEAN := $(CLEAN) report/*.pdf report/*.aux report/*.log \
+	report/*.out report/*.toc report/*.snm report/*.nav
+
+report: report/report.pdf
+
+report/%.pdf: report/%.tex
+	cd report; \
+	pdflatex report.tex; \
+	pdflatex report.tex

run

+#!/bin/sh
+python main.py benchmarks/build/$1.s out.s in.s && meld in.s out.s

src/dag.py

+class Dag:
+    def __init__(self, block):
+        """Create the Directed Acyclic Graph of all binary operations in a
+        basic block."""
+        self.nodes = []
+
+        for s in block:
+            if s.is_command('move') or s.is_monop():
+                rd, rs = s
+                y = self.find_reg_node(rs)
+                self.find_op_node(s.name, rd, y)
+            elif s.is_binop():
+                rd, rs, rt = s
+                y = self.find_reg_node(rs)
+                z = self.find_reg_node(rt)
+                self.find_op_node(s.name, rd, y, z)
+
+    def find_reg_node(self, reg):
+        for n in self.nodes:
+            if reg in n.reg:
+                return n
+
+        node = DagLeaf(reg)
+        self.nodes.append(node)
+
+        return node
+
+    def find_op_node(self, op, rd, *args):
+        for n in self.nodes:
+            if not isinstance(n, DagLeaf) and n.op == op and n.nodes == args:
+                n.labels.append(rd)
+
+                return n
+
+        node = DagNode(op, rd, *args)
+        self.nodes.append(node)
+
+        return node
+
+
+class DagNode:
+    def __init__(self, op, label, *args):
+        self.op = op
+        self.labels = [label]
+        self.nodes = args
+
+
+class DagLeaf:
+    def __init__(self, reg):
+        self.reg = reg

src/dataflow.py

-from copy import copy
-
 from statement import Block
 
 
@@ -11,10 +9,6 @@ class BasicBlock(Block):
 
         self.dominates = []
         self.dominated_by = []
-        self.in_set = set([])
-        self.out_set = set([])
-        self.gen_set = set([])
-        self.kill_set = set([])
 
     def add_edge_to(self, block):
         if block not in self.edges_to:
@@ -26,84 +20,6 @@ class BasicBlock(Block):
             self.dominates.append(block)
             block.dominated_by.append(self)
 
-    def create_gen_kill(self, defs):
-        used = set()
-        self_defs = {}
-
-        # Get the last of each definition series and put in in the `def' set
-        self.gen_set = set()
-
-        for s in reversed(self):
-            for reg in s.get_def():
-                if reg not in self_defs:
-                    print 'Found def:', s
-                    self_defs[reg] = s.sid
-                    self.gen_set.add(s.sid)
-
-        # Generate kill set
-        self.kill_set = set()
-
-        for reg, statement_ids in defs.iteritems():
-            if reg in self_defs:
-                add = statement_ids - set([self_defs[reg]])
-            else:
-                add = statement_ids
-
-            self.kill_set |= add
-
-
-def defs(blocks):
-    # Collect definitions of all registers
-    defs = {}
-
-    for b in blocks:
-        for s in b:
-            for reg in s.get_def():
-                if reg not in defs:
-                    defs[reg] = set([s.sid])
-                else:
-                    defs[reg].add(s.sid)
-
-    return defs
-
-
-def reaching_definitions(blocks):
-    """Generate the `in' and `out' sets of the given blocks using the iterative
-    algorithm from the slides."""
-    defs = defs(blocks)
-
-    for b in blocks:
-        b.create_gen_kill(defs)
-        b.out_set = b.gen_set
-
-    change = True
-
-    while change:
-        change = False
-
-        for b in blocks:
-            b.in_set = set()
-
-            for pred in b.edges_from:
-                b.in_set |= pred.out_set
-
-            oldout = copy(p.out_set)
-            p.out_set = b.gen_set | (b.in_set - b.kill_set)
-
-            if b.out_set != oldout:
-                change = True
-
-
-def pred(n, known=[]):
-    """Recursively find all predecessors of a node."""
-    direct = filter(lambda b: b not in known, n.edges_from)
-    p = copy(direct)
-
-    for ancestor in direct:
-        p += pred(ancestor, direct)
-
-    return p
-
 
 def find_leaders(statements):
     """Determine the leaders, which are:
@@ -160,109 +76,10 @@ def generate_flow_graph(blocks):
                 if other[0].is_label(target):
                     b.add_edge_to(other)
 
-            # A branch instruction also creates an edge to the next block
-            if last_statement.is_branch() and i < len(blocks) - 1:
+            # A branch and jump-and-line instruction also creates an edge to
+            # the next block
+            if (last_statement.is_branch() or last_statement.name == 'jal') \
+                    and i < len(blocks) - 1:
                 b.add_edge_to(blocks[i + 1])
         elif i < len(blocks) - 1:
             b.add_edge_to(blocks[i + 1])
-
-
-#def generate_dominator_tree(nodes):
-#    """Add dominator administration to the given flow graph nodes."""
-#    # Dominator of the start node is the start itself
-#    nodes[0].dom = set([nodes[0]])
-#
-#    # For all other nodes, set all nodes as the dominators
-#    for n in nodes[1:]:
-#        n.dom = set(copy(nodes))
-#
-#    def pred(n, known=[]):
-#        """Recursively find all predecessors of a node."""
-#        direct = filter(lambda x: x not in known, n.edges_from)
-#        p = copy(direct)
-#
-#        for ancestor in direct:
-#            p += pred(ancestor, direct)
-#
-#        return p
-#
-#    # Iteratively eliminate nodes that are not dominators
-#    changed = True
-#
-#    while changed:
-#        changed = False
-#
-#        for n in nodes[1:]:
-#            old_dom = n.dom
-#            intersection = lambda p1, p2: p1.dom & p2.dom
-#            n.dom = set([n]) | reduce(intersection, pred(n), set([]))
-#
-#            if n.dom != old_dom:
-#                changed = True
-#
-#    def idom(d, n):
-#        """Check if d immediately dominates n."""
-#        for b in n.dom:
-#            if b != d and b != n and b in n.dom:
-#                return False
-#
-#        return True
-#
-#    # Build tree using immediate dominators
-#    for n in nodes:
-#        for d in n.dom:
-#            if idom(d, n):
-#                d.set_dominates(n)
-#                break
-
-
-class Dag:
-    def __init__(self, block):
-        """Create the Directed Acyclic Graph of all binary operations in a
-        basic block."""
-        self.nodes = []
-
-        for s in block:
-            if s.is_command('move') or s.is_monop():
-                rd, rs = s
-                y = self.find_reg_node(rs)
-                self.find_op_node(s.name, rd, y)
-            elif s.is_binop():
-                rd, rs, rt = s
-                y = self.find_reg_node(rs)
-                z = self.find_reg_node(rt)
-                self.find_op_node(s.name, rd, y, z)
-
-    def find_reg_node(self, reg):
-        for n in self.nodes:
-            if reg in n.reg:
-                return n
-
-        node = DagLeaf(reg)
-        self.nodes.append(node)
-
-        return node
-
-    def find_op_node(self, op, rd, *args):
-        for n in self.nodes:
-            if not isinstance(n, DagLeaf) and n.op == op and n.nodes == args:
-                n.labels.append(rd)
-
-                return n
-
-        node = DagNode(op, rd, *args)
-        self.nodes.append(node)
-
-        return node
-
-
-class DagNode:
-    def __init__(self, op, label, *args):
-        self.op = op
-        self.labels = [label]
-        self.nodes = args
-
-
-class DagLeaf:
-    def __init__(self, reg):
-        self.reg = reg

src/dominator.py

+from copy import copy
+
+
+def generate_dominator_tree(nodes):
+    """Add dominator administration to the given flow graph nodes."""
+    # Dominator of the start node is the start itself
+    nodes[0].dom = set([nodes[0]])
+
+    # For all other nodes, set all nodes as the dominators
+    for n in nodes[1:]:
+        n.dom = set(copy(nodes))
+
+    def pred(n, known=[]):
+        """Recursively find all predecessors of a node."""
+        direct = filter(lambda x: x not in known, n.edges_from)
+        p = copy(direct)
+
+        for ancestor in direct:
+            p += pred(ancestor, direct)
+
+        return p
+
+    # Iteratively eliminate nodes that are not dominators
+    changed = True
+
+    while changed:
+        changed = False
+
+        for n in nodes[1:]:
+            old_dom = n.dom
+            intersection = lambda p1, p2: p1.dom & p2.dom
+            n.dom = set([n]) | reduce(intersection, pred(n), set([]))
+
+            if n.dom != old_dom:
+                changed = True
+
+    def idom(d, n):
+        """Check if d immediately dominates n."""
+        for b in n.dom:
+            if b != d and b != n and b in n.dom:
+                return False
+
+        return True
+
+    # Build tree using immediate dominators
+    for n in nodes:
+        for d in n.dom:
+            if idom(d, n):
+                d.set_dominates(n)
+                break

src/liveness.py

+from copy import copy
+
+
+RESERVED_REGISTERS = ['$fp', '$sp', '$31']
+
+
+def is_reg_dead_after(reg, block, index):
+    """Check if a register is dead after a certain point in a basic block."""
+    if reg in RESERVED_REGISTERS:
+        return False
+
+    if index < len(block) - 1:
+        for s in block[index + 1:]:
+            # If used, the previous definition is live
+            if s.uses(reg):
+                return False
+
+            # If redefined, the previous definition is dead
+            if s.defines(reg):
+                return True
+
+    # If dead within the same block, check if the register is in the block's
+    # live_out set
+    return reg not in block.live_out
+
+
+def create_use_def(block):
+    used = set()
+    defined = set()
+
+    # Get the last of each definition series and put in in the `def' set
+    block.use_set = set()
+    block.def_set = set()
+
+    for s in block:
+        # use[B] is the set of variables whose values may be used in B prior to
+        # any definition of the variable
+        for reg in s.get_use():
+            used.add(reg)
+
+            if reg not in defined:
+                block.use_set.add(reg)
+
+        # def[B] is the set of variables assigned values in B prior to any use
+        # of that variable in B
+        for reg in s.get_def():
+            defined.add(reg)
+
+            if reg not in used:
+                block.def_set.add(reg)
+
+
+def succ(block, known=[]):
+    """Recursively find all successors of a node."""
+    direct = filter(lambda b: b != block and b not in known, block.edges_to)
+    s = copy(direct)
+
+    for successor in direct:
+        s += succ(successor, known + direct)
+        return s
+
+    return s
+
+
+def create_in_out(blocks):
+    for b in blocks:
+        create_use_def(b)
+
+        b.live_in = b.use_set
+        b.live_out = set()
+
+    change = True
+
+    while change:
+        change = False
+
+        for b in blocks:
+            # in[B] = use[B] | (out[B] - def[B])
+            new_in = b.use_set | (b.live_out - b.def_set)
+
+            # out[B] = union of in[S] for S in succ(B)
+            new_out = set()
+
+            for s in succ(b):
+                new_out |= s.live_in
+
+            # Check if either `in' or `out' changed
+            if new_in != b.live_in:
+                b.live_in = new_in
+                change = True
+
+            if new_out != b.live_out:
+                b.live_out = new_out
+                change = True

src/optimize/__init__.py

-from src.dataflow import find_basic_blocks
-
-from redundancies import remove_redundant_jumps, move_1, move_2, move_3, \
-        move_4, load, shift, add
+from src.dataflow import find_basic_blocks, generate_flow_graph
+from redundancies import remove_redundancies
 from advanced import eliminate_common_subexpressions, fold_constants, \
-        copy_propagation, algebraic_transformations, eliminate_dead_code
-
-
-def remove_redundancies(block):
-    """Execute all functions that remove redundant statements."""
-    callbacks = [move_1, move_2, move_3, move_4, load, shift, add]
-    old_len = -1
-    changed = False
-
-    while old_len != len(block):
-        old_len = len(block)
-
-        while not block.end():
-            s = block.read()
-
-            for callback in callbacks:
-                if callback(s, block):
-                    changed = True
-                    break
-
-    return changed
+        copy_propagation, eliminate_dead_code
+import src.liveness as liveness
+import src.reaching_definitions as reaching_definitions
 
-
-def optimize_block(block):
-    """Optimize a basic block."""
-    while remove_redundancies(block) \
-            | eliminate_common_subexpressions(block) \
-            | fold_constants(block) \
-            | copy_propagation(block)\
-            | algebraic_transformations(block) \
-            | eliminate_dead_code(block):
-        pass
-
-
-def optimize(statements, verbose=0):
+def optimize(program, verbose=0):
     """Optimization wrapper function, calls global and basic-block level
     optimization functions."""
+    # Remember original number of statements
+    o = program.count_instructions()
+
     # Optimize on a global level
-    o = len(statements)
-    remove_redundant_jumps(statements)
-    g = len(statements)
+    program.optimize_global()
+    g = program.count_instructions()
+
+    # Perform dataflow analysis
+    program.perform_dataflow_analysis()
 
     # Optimize basic blocks
-    blocks = find_basic_blocks(statements)
-    map(optimize_block, blocks)
-    block_statements = map(lambda b: b.statements, blocks)
-    opt_blocks = reduce(lambda a, b: a + b, block_statements)
-    b = len(opt_blocks)
+    program.optimize_blocks()
 
-    # - Common subexpression elimination
-    # - Constant folding
-    # - Copy propagation
-    # - Dead-code elimination
-    # - Temporary variable renaming
-    # - Interchange of independent statements
+    # Concatenate optimized blocks to obtain
+    b = program.count_instructions()
 
+    # Print results
     if verbose:
-        print 'Original statements:             %d' % o
-        print 'After global optimization:       %d' % g
-        print 'After basic blocks optimization: %d' % b
-        print 'Optimization:                    %d (%d%%)' \
+        print 'Original statements:            %d' % o
+        print 'After global optimization:      %d (%d removed)' % (g, o - g)
+        print 'After basic block optimization: %d (%d removed)' % (b, g - b)
+        print 'Statements removed:             %d (%d%%)' \
                 % (o - b, int((o - b) / float(b) * 100))
-
-    return opt_blocks

src/optimize/redundancies.py

 import re
 
 
-def move_1(mov, statements):
+def remove_redundancies(block):
+    """Execute all functions that remove redundant statements."""
+    callbacks = [move_aa, move_inst, instr_move_jal, move_move, sw_ld, shift, 
+                 add_lw]
+    old_len = -1
+    changed = False
+
+    while old_len != len(block):
+        old_len = len(block)
+
+        block.reset()
+
+        while not block.end():
+            s = block.read()
+
+            for callback in callbacks:
+                if callback(s, block):
+                    changed = True
+                    break
+
+    return changed
+
+
+def move_aa(mov, statements):
     """
-    mov $regA, $regA          ->  --- remove it
+    move $regA, $regA          ->  --- remove it
     """
     if mov.is_command('move') and mov[0] == mov[1]:
         statements.replace(1, [])
@@ -11,9 +34,9 @@ def move_1(mov, statements):
         return True
 
 
-def move_2(mov, statements):
+def move_inst(mov, statements):
     """
-    mov $regA, $regB          ->  instr $regA, $regB, ...
+    move $regA, $regB          ->  instr $regA, $regB, ...
     instr $regA, $regA, ...
     """
     if mov.is_command('move'):
@@ -26,10 +49,10 @@ def move_2(mov, statements):
             return True
 
 
-def move_3(ins, statements):
+def instr_move_jal(ins, statements):
     """
     instr $regA, ...          ->  instr $4, ...
-    mov $4, $regA                 jal XX
+    move $4, $regA                 jal XX
     jal XX
     """
     if ins.is_command() and len(ins):
@@ -47,10 +70,10 @@ def move_3(ins, statements):
                 return True
 
 
-def move_4(mov1, statements):
+def move_move(mov1, statements):
     """
-    mov $RegA, $RegB         ->  move $RegA, $RegB
-    mov $RegB, $RegA
+    move $RegA, $RegB         ->  move $RegA, $RegB
+    move $RegB, $RegA
     """
     if mov1.is_command('move'):
         mov2 = statements.peek()
@@ -62,7 +85,7 @@ def move_4(mov1, statements):
             return True
 
 
-def load(sw, statements):
+def sw_ld(sw, statements):
     """
     sw $regA, XX              ->  sw $regA, XX
     ld $regA, XX
@@ -86,7 +109,7 @@ def shift(shift, statements):
         return True
 
 
-def add(add, statements):
+def add_lw(add, statements):
     """
     add $regA, $regA, X       ->  lw ..., X($regA)
     lw ..., 0($regA)
@@ -124,3 +147,5 @@ def remove_redundant_jumps(statements):
                         s.name = 'bne' if s.is_command('beq') else 'beq'
                         s[2] = j[0]
                         statements.replace(3, [s, label])
+
+    statements.reset()

src/parser.py

 import ply.lex as lex
 import ply.yacc as yacc
 
-from statement import Statement as S, Block
+from statement import Statement as S
+from program import Program
 
 
 # Global statements administration
@@ -46,6 +47,7 @@ def t_offset_address(t):
 def t_int(t):
     r'-?[0-9]+'
     t.type = 'WORD'
+    t.value = int(t.value)
     return t
 
 def t_WORD(t):
@@ -79,11 +81,12 @@ def p_line_instruction(p):
 
 def p_line_comment(p):
     'line : COMMENT NEWLINE'
-    statements.append(S('comment', p[1], inline=False))
+    statements.append(S('comment', p[1]))
 
 def p_line_inline_comment(p):
     'line : instruction COMMENT NEWLINE'
-    statements.append(S('comment', p[2], inline=True))
+    # Add the inline comment to the last parsed statement
+    statements[-1].options['comment'] = p[2]
 
 def p_instruction_command(p):
     'instruction : command'
@@ -125,4 +128,4 @@ def parse_file(filename):
     except IOError:
         raise Exception('File "%s" could not be opened' % filename)
 
-    return Block(statements)
+    return Program(statements)

src/program.py

+from statement import Statement as S, Block
+from dataflow import find_basic_blocks, generate_flow_graph
+from optimize.redundancies import remove_redundant_jumps, remove_redundancies
+from optimize.advanced import eliminate_common_subexpressions, \
+        fold_constants, copy_propagation, eliminate_dead_code
+from writer import write_statements
+import liveness
+import reaching_definitions
+
+
+class Program(Block):
+    def __len__(self):
+        """Get the number of statements in the program."""
+        return len(self.statements) if hasattr(self, 'statements') \
+               else reduce(lambda a, b: len(a) + len(b), self.blocks, 0)
+
+    def get_statements(self, add_block_comments=False):
+        """Concatenate the statements of all blocks and return the resulting
+        list."""
+        if hasattr(self, 'statements'):
+            return self.statements
+
+        # Only add block start and end comments when in debug mode
+        if add_block_comments and self.debug:
+            get_id = lambda b: b.bid
+            statements = []
+
+            for b in self.blocks:
+                message = ' Block %d (%d statements), edges from: %s' \
+                          % (b.bid, len(b), map(get_id, b.edges_from))
+
+                if hasattr(b, 'live_in'):
+                    message += ', LIVE_in: %s' % list(b.live_in)
+
+                if hasattr(b, 'reach_in'):
+                    message += ', REACH_in: %s' % list(b.reach_in)
+
+                statements.append(S('comment', message, block=False))
+
+                statements += b.statements
+
+                message = ' End of block %d, edges to: %s' \
+                          % (b.bid, map(get_id, b.edges_to))
+
+                if hasattr(b, 'live_out'):
+                    message += ', LIVE_out: %s' % list(b.live_out)
+
+                if hasattr(b, 'reach_out'):
+                    message += ', REACH_out: %s' % list(b.reach_out)
+
+                statements.append(S('comment', message, block=False))
+
+            return statements
+
+        return reduce(lambda a, b: a + b,
+                      [b.statements for b in self.blocks])
+
+    def count_instructions(self):
+        """Count the number of statements that are commands or labels."""
+        return len(filter(lambda s: s.is_command() or s.is_label(),
+                          self.get_statements()))
+
+    def optimize_global(self):
+        """Optimize on a global level."""
+        remove_redundant_jumps(self)
+
+    def optimize_blocks(self):
+        """Optimize on block level. Keep executing all optimizations until no
+        more changes occur."""
+        self.program_iterations = self.block_iterations = 0
+        program_changed = True
+
+        while program_changed:
+            self.program_iterations += 1
+            program_changed = False
+
+            for block in self.blocks:
+                self.block_iterations += 1
+                block_changed = True
+
+                while block_changed:
+                    block_changed = False
+
+                    if remove_redundancies(block):
+                        block_changed = True
+
+                    if eliminate_common_subexpressions(block):
+                        block_changed = True
+
+                    if fold_constants(block):
+                        block_changed = True
+
+                    if copy_propagation(block):
+                        block_changed = True
+
+                    if eliminate_dead_code(block):
+                        block_changed = True
+
+                    if block_changed:
+                        program_changed = True
+
+    def find_basic_blocks(self):
+        """Divide the statement list into basic blocks."""
+        self.blocks = find_basic_blocks(self.statements)
+
+        for b in self.blocks:
+            b.debug = self.debug
+
+        # Remove the old statement list, since it will probably change
+        del self.statements
+
+    def perform_dataflow_analysis(self):
+        """Perform dataflow analysis:
+           - Divide the statement list into basic blocks
+           - Generate flow graph
+           - Create liveness sets: def, use, in, out
+           - Create reaching definitions sets: gen, kill, in, out"""
+        self.find_basic_blocks()
+        generate_flow_graph(self.blocks)
+        liveness.create_in_out(self.blocks)
+        reaching_definitions.create_in_out(self.blocks)
+
+    def save(self, filename):
+        """Save the program in the specified file."""
+        f = open(filename, 'w+')
+        f.write(write_statements(self.get_statements(True)))
+        f.close()

src/reaching_definitions.py

+from dataflow import BasicBlock as B
+
+
+def get_defs(blocks):
+    """Collect definitions of all registers."""
+    defs = {}
+
+    for b in blocks:
+        for s in b:
+            for reg in s.get_def():
+                if reg not in defs:
+                    defs[reg] = set([s.sid])
+                else:
+                    defs[reg].add(s.sid)
+
+    return defs
+
+
+def create_gen_kill(block, global_defs):
+    block_defs = {}
+
+    # Get the last of each definition series and put in in the `def' set
+    block.gen_set = set()
+
+    for s in reversed(block):
+        for reg in s.get_def():
+            if reg not in block_defs:
+                block_defs[reg] = s.sid
+                block.gen_set.add(s.sid)
+
+    # Generate kill set
+    block.kill_set = set()
+
+    for reg, statement_ids in global_defs.iteritems():
+        if reg in block_defs:
+            block.kill_set |= statement_ids - set([block_defs[reg]])
+
+
+def create_in_out(blocks):
+    """Generate the `in' and `out' sets of the given blocks using the iterative
+    algorithm from the lecture slides."""
+    # Create gen/kill sets
+    defs = get_defs(blocks)
+
+    for b in blocks:
+        create_gen_kill(b, defs)
+        b.reach_out = b.gen_set
+
+    change = True
+
+    while change:
+        change = False
+
+        for b in blocks:
+            b.reach_in = set()
+
+            for pred in b.edges_from:
+                b.reach_in |= pred.reach_out
+
+            new_out = b.gen_set | (b.reach_in - b.kill_set)
+
+            if new_out != b.reach_out:
+                b.reach_out = new_out
+                change = True

src/todo.txt

 * Apart python script maken die de generator genereert dmv config file
+* 'verbose' argument in main.py
+
+* Gaat fout:
+
+li  $5,0x00008000  ->  l.d  $f0,32768($4)
+addu    $4,$4,$5
+l.d $f0,0($4)

src/writer.py

 from math import ceil
 
 
+TABSIZE = 4                 # Size in spaces of a single tab
+INLINE_COMMENT_LEVEL = 6    # Number of tabs to inline commment level
+COMMAND_SIZE = 8            # Default length of a command name, used for
+                            # indenting
+ADD_COMMENT_BLOCKS = True   # Wether to add newlines before and after
+                            # non-inline comment
+ADD_ARGUMENT_SPACE = False  # Wether to add a space between command arguments
+                            # and the previous comma
+
+
 def write_statements(statements):
     """Write a list of statements to valid assembly code."""
     out = ''
     indent_level = 0
-    prevline = ''
+    prev_comment = False
 
     for i, s in enumerate(statements):
-        newline = '\n' if i else ''
+        current_comment = False
 
         if s.is_label():
             line = s.name + ':'
             indent_level = 1
         elif s.is_comment():
-            line = '#' + s.name
-
-            if s.is_inline_comment():
-                l = len(prevline.expandtabs(4))
-                tabs = int(ceil((24 - l) / 4.)) + 1
-                newline = '\t' * tabs
-            else:
-                line = '\t' * indent_level + line
+            line = '\t' * indent_level + '#' + s.name
+            current_comment = s.options.get('block', True)
         elif s.is_directive():
             line = '\t' + s.name
         elif s.is_command():
             line = '\t' + s.name
 
+            # If there are arguments, add tabs until the 8 character limit has
+            # been reached. If the command name is 8 or more characers long,
+            # add a single space
             if len(s):
-                if len(s.name) < 8:
-                    line += '\t'
+                l = len(s.name)
+
+                if l < COMMAND_SIZE:
+                    line += '\t' * int(ceil((COMMAND_SIZE - l)
+                                       / float(TABSIZE)))
                 else:
                     line += ' '
 
-                line += ','.join(s.args)
+                delim = ', ' if ADD_ARGUMENT_SPACE else ','
+                line += delim.join(map(str, s))
         else:
             raise Exception('Unsupported statement type "%s"' % s.stype)
 
-        out += newline + line
-        prevline = line
+        # Add the inline comment, if there is any
+        if s.has_inline_comment():
+            start = INLINE_COMMENT_LEVEL * TABSIZE
+            diff = start - len(line.expandtabs(TABSIZE))
 
-    # Add newline at end of file
-    out += '\n'
+            # The comment must not be directly adjacent to the command itself
+            if diff > 0:
+                tabs = '\t' * (int(ceil(diff / float(TABSIZE))) + 1)
+            else:
+                tabs = '  '
+
+            line += tabs + '#' + s.options['comment']
+
+        # Add newline at end of command
+        line += '\n'
+
+        if ADD_COMMENT_BLOCKS:
+            if prev_comment ^ current_comment:
+                out += '\n'
+
+        out += line
+        prev_comment = current_comment
 
     return out
 
+
 def write_to_file(filename, statements):
     """Convert a list of statements to valid assembly code and write it to a
     file."""

test.py

+#!/usr/bin/python
 from testrunner import main
 import sys
+
 main(sys.argv[1:])

tests/rules.mk

 TESTS=$(wildcard tests/test_*.py)
 COVERAGE_OUTPUT_DIR := coverage
 OMIT := /usr/share/pyshared/*,test*,*__init__.py
+CLEAN := $(CLEAN) tests/*.pyc
 
 ifeq ($(findstring python-coverage,$(wildcard /usr/bin/*)), python-coverage)
 COVERAGE=/usr/bin/python-coverage

tests/test_dag.py

+import unittest
+
+from src.statement import Statement as S
+from src.dataflow import BasicBlock as B
+from src.dag import Dag, DagNode, DagLeaf
+
+
+class TestDag(unittest.TestCase):
+
+    def setUp(self):
+        pass
+
+    def tearDown(self):
+        pass
+
+    def test_dag_unary(self):
+        dag = Dag(B([S('command', 'neg.d', '$rd', '$rs')]))
+        expect = Dag([])
+        expect.nodes = [DagLeaf('$rs'), DagNode('neg.d', '$rd', \
+                        DagLeaf('$rs'))]
+
+        self.assertEqualDag(dag, expect)
+
+    def test_dag_binary(self):
+        dag = Dag(B([S('command', 'addu', '$rd', '$r1', '$r2')]))
+        expect = Dag([])
+        expect.nodes = [DagLeaf('$r1'),
+                        DagLeaf('$r2'),
+                        DagNode('addu', '$rd', DagLeaf('$r1'), DagLeaf('$r2'))]
+
+        self.assertEqualDag(dag, expect)
+
+#    def test_dag_combinednode(self):
+#        dag = Dag(B([S('command', 'mult', '$rd1', '$r1', '$r2'),
+#                     S('command', 'mult', '$rd2', '$r1', '$r2')]))
+#        expect = Dag([])
+#        multnode = DagNode('mult',
+#                           DagLeaf('$r1'),
+#                           DagLeaf('$r2'))
+#        multnode.labels = ['$rd1', '$rd2']
+#        expect.nodes = [DagLeaf('$r1'),
+#                        DagLeaf('$r2'),
+#                        multnode]
+#
+#        self.assertEqualDag(dag, expect)
+
+    def assertEqualDag(self, dag1, dag2):
+        self.assertEqual(len(dag1.nodes), len(dag2.nodes))
+
+        for node1, node2 in zip(dag1.nodes, dag2.nodes):
+            self.assertEqualNodes(node1, node2)
+
+    def assertEqualNodes(self, node1, node2):
+        if isinstance(node1, DagLeaf):
+            self.assertIsInstance(node2, DagLeaf)
+            self.assertEqual(node1.reg, node2.reg)
+        elif isinstance(node2, DagLeaf):
+            raise AssertionError
+        else:
+            self.assertEqual(node1.op, node2.op)
+            self.assertEqual(node1.labels, node2.labels)
+            self.assertEqual(len(node1.nodes), len(node2.nodes))
+
+            for child1, child2 in zip(node1.nodes, node2.nodes):
+                self.assertEqualNodes(child1, child2)

tests/test_dataflow.py

@@ -2,7 +2,7 @@ import unittest
 
 from src.statement import Statement as S
 from src.dataflow import BasicBlock as B, find_leaders, find_basic_blocks, \
-        generate_flow_graph, Dag, DagNode, DagLeaf, defs, reaching_definitions
+        generate_flow_graph
 
 
 class TestDataflow(unittest.TestCase):
@@ -24,42 +24,6 @@ class TestDataflow(unittest.TestCase):
                 [B(s[:2]).statements, B(s[2:4]).statements, \
                  B(s[4:]).statements])
 
-#    def test_get_gen(self):
-#        b1 = B([S('command', 'add', '$1', '$2', '$3'), \
-#                S('command', 'add', '$2', '$3', '$4'), \
-#                S('command', 'add', '$1', '$4', '$5')])
-#
-#        self.assertEqual(b1.get_gen(), ['$1', '$2'])
-
-#    def test_get_out(self):
-#        b1 = B([S('command', 'add', '$1', '$2', '$3'), \
-#                S('command', 'add', '$2', '$3', '$4'), \
-#                S('command', 'add', '$1', '$4', '$5'), \
-#                S('command', 'j', 'b2')])
-#
-#        b2 = B([S('command', 'add', '$3', '$5', '$6'), \
-#                S('command', 'add', '$1', '$2', '$3'), \
-#                S('command', 'add', '$6', '$4', '$5')])
-#
-#        blocks = [b1, b2]
-#
-#        # initialize  out[B] = gen[B] for every block
-#        for block in blocks:
-#            block.out_set = block.get_gen()
-#            print 'block.out_set', block.out_set
-#
-#        generate_flow_graph(blocks)
-
-#        change = True
-#        while change:
-#            for i, block in enumerate(blocks):
-#                block.get_in()
-#                oldout = block.out_set
-#                newout = block.get_out()
-#                if (block.out_set == block.get_out()):
-#                    change = False
-
-
     def test_generate_flow_graph_simple(self):
         b1 = B([S('command', 'foo'), S('command', 'j', 'b2')])
         b2 = B([S('label', 'b2'), S('command', 'bar')])
@@ -80,94 +44,3 @@ class TestDataflow(unittest.TestCase):
         self.assertEqual(b2.edges_to, [b3])
         self.assertIn(b1, b3.edges_from)
         self.assertIn(b2, b3.edges_from)
-
-    def test_dag_unary(self):
-        dag = Dag(B([S('command', 'neg.d', '$rd', '$rs')]))
-        expect = Dag([])
-        expect.nodes = [DagLeaf('$rs'), DagNode('neg.d', '$rd', \
-                        DagLeaf('$rs'))]
-
-        self.assertEqualDag(dag, expect)
-
-    def test_dag_binary(self):
-        dag = Dag(B([S('command', 'addu', '$rd', '$r1', '$r2')]))
-        expect = Dag([])
-        expect.nodes = [DagLeaf('$r1'),
-                        DagLeaf('$r2'),
-                        DagNode('addu', '$rd', DagLeaf('$r1'), DagLeaf('$r2'))]
-
-        self.assertEqualDag(dag, expect)
-
-#    def test_dag_combinednode(self):
-#        dag = Dag(B([S('command', 'mult', '$rd1', '$r1', '$r2'),
-#                     S('command', 'mult', '$rd2', '$r1', '$r2')]))
-#        expect = Dag([])
-#        multnode = DagNode('mult',
-#                           DagLeaf('$r1'),
-#                           DagLeaf('$r2'))
-#        multnode.labels = ['$rd1', '$rd2']
-#        expect.nodes = [DagLeaf('$r1'),
-#                        DagLeaf('$r2'),
-#                        multnode]
-#
-#        self.assertEqualDag(dag, expect)
-
-    def test_defs(self):
-        s1 = S('command', 'addu', '$3', '$1', '$2')
-        s2 = S('command', 'addu', '$1', '$3', 10)
-        s3 = S('command', 'subu', '$3', '$1', 5)
-        s4 = S('command', 'li', '$4', '0x00000001')
-        block = B([s1, s2, s3, s4])
-        self.assertEqual(defs([block]), {
-            '$3': set([s1.sid, s3.sid]),
-            '$1': set([s2.sid]),
-            '$4': set([s4.sid])
-        })
-
-    #def test_defs(self):
-    #    s1 = S('command', 'add', '$3', '$1', '$2')
-    #    s2 = S('command', 'move', '$1', '$3')
-    #    s3 = S('command', 'move', '$3', '$2')
-    #    s4 = S('command', 'li', '$4', '0x00000001')
-    #    block = B([s1, s2, s3, s4])
-    #    self.assertEqual(defs([block]), {
-    #        '$3': set([s1.sid, s3.sid]),
-    #        '$1': set([s2.sid]),
-    #        '$4': set([s4.sid])
-    #    })
-
-    def test_create_gen_kill_gen(self):
-        s1 = S('command', 'addu', '$3', '$1', '$2')
-        s2 = S('command', 'addu', '$1', '$3', 10)
-        s3 = S('command', 'subu', '$3', '$1', 5)
-        s4 = S('command', 'li', '$4', '0x00000001')
-        block = B([s1, s2, s3, s4])
-        block.create_gen_kill(defs([block]))
-        self.assertEqual(block.gen_set, set([s2.sid, s3.sid, s4.sid]))
-
-    #def test_get_kill_used(self):
-    #    block = B([S('command', 'move', '$1', '$3'),
-    #               S('command', 'add', '$3', '$1', '$2'),
-    #               S('command', 'move', '$1', '$3'),
-    #               S('command', 'move', '$2', '$3')])
-    #    self.assertEqual(block.get_kill(), set())
-
-    def assertEqualDag(self, dag1, dag2):
-        self.assertEqual(len(dag1.nodes), len(dag2.nodes))
-
-        for node1, node2 in zip(dag1.nodes, dag2.nodes):
-            self.assertEqualNodes(node1, node2)
-
-    def assertEqualNodes(self, node1, node2):
-        if isinstance(node1, DagLeaf):
-            self.assertIsInstance(node2, DagLeaf)
-            self.assertEqual(node1.reg, node2.reg)
-        elif isinstance(node2, DagLeaf):
-            raise AssertionError
-        else:
-            self.assertEqual(node1.op, node2.op)
-            self.assertEqual(node1.labels, node2.labels)
-            self.assertEqual(len(node1.nodes), len(node2.nodes))
-
-            for child1, child2 in zip(node1.nodes, node2.nodes):
-                self.assertEqualNodes(child1, child2)

tests/test_liveness.py

+import unittest
+
+from src.statement import Statement as S
+from src.dataflow import BasicBlock as B, find_basic_blocks, \
+        generate_flow_graph
+from src.liveness import create_use_def, create_in_out
+
+
+class TestLiveness(unittest.TestCase):
+
+    def setUp(self):
+        pass
+
+    def tearDown(self):
+        pass
+
+    def test_create_gen_kill(self):
+        s1 = S('command', 'addu', '$3', '$1', '$2')
+        s2 = S('command', 'addu', '$1', '$3', 10)
+        s3 = S('command', 'subu', '$3', '$1', 5)
+        s4 = S('command', 'li', '$4', '0x00000001')
+        block = B([s1, s2, s3, s4])
+
+        create_use_def(block)
+
+        self.assertEqual(block.use_set, set(['$1', '$2']))
+        self.assertEqual(block.def_set, set(['$3', '$4']))
+
+    def test_create_in_out(self):
+        s11 = S('command', 'li', 'a', 3)
+        s12 = S('command', 'li', 'b', 5)
+        s13 = S('command', 'li', 'd', 4)
+        s14 = S('command', 'li', 'x', 100)
+        s15 = S('command', 'beq', 'a', 'b', 'L1')
+
+        s21 = S('command', 'addu', 'c', 'a', 'b')
+        s22 = S('command', 'li', 'd', 2)
+
+        s31 = S('label', 'L1')
+        s32 = S('command', 'li', 'c', 4)
+        s33 = S('command', 'mult', 'b', 'd')
+        s34 = S('command', 'mflo', 'temp')
+        s35 = S('command', 'addu', 'return', 'temp', 'c')
+
+        b1, b2, b3 = find_basic_blocks([s11, s12, s13, s14, s15, s21, s22, \
+                                        s31, s32, s33, s34, s35])
+
+        generate_flow_graph([b1, b2, b3])
+        create_in_out([b1, b2, b3])
+
+        self.assertEqual(b1.use_set, set())
+        self.assertEqual(b1.def_set, set(['a', 'b', 'd', 'x']))
+
+        self.assertEqual(b2.use_set, set(['a', 'b']))
+        self.assertEqual(b2.def_set, set(['c', 'd']))
+
+        self.assertEqual(b3.use_set, set(['b', 'd']))
+        self.assertEqual(b3.def_set, set(['c', 'temp', 'return']))
+
+        self.assertEqual(b1.live_in, set())
+        self.assertEqual(b1.live_out, set(['a', 'b', 'd']))
+        self.assertEqual(b2.live_in, set(['a', 'b']))
+        self.assertEqual(b2.live_out, set(['b', 'd']))
+        self.assertEqual(b3.live_in, set(['b', 'd']))
+        self.assertEqual(b3.live_out, set())

tests/test_optimize.py

 import unittest
 
-from src.optimize.redundancies import remove_redundant_jumps
-from src.optimize import optimize_block
+from src.optimize.redundancies import remove_redundancies, remove_redundant_jumps
+from src.program import Program
 from src.statement import Statement as S, Block as B
 
 
+def optimize_block(block):
+    """Optimize a basic block using a Program object."""
+#    program = Program([])
+
+#    program.blocks = [block]
+#    del program.statements
+    
+#   program.optimize_blocks()
+
+    remove_redundancies(block)
+    eliminate_common_subexpressions(block)
+    fold_constants(block)
+    copy_propagation(block)
+
+    return program.blocks
+
+
 class TestOptimize(unittest.TestCase):
 
     def setUp(self):
@@ -19,7 +36,7 @@ class TestOptimize(unittest.TestCase):
         block = B([self.foo,
                    S('command', 'move', '$regA', '$regA'),
                    self.bar])
-        optimize_block(block)
+        remove_redundancies(block)
         self.assertEquals(block.statements, [self.foo, self.bar])
 
     def test_optimize_block_movab(self):
@@ -27,7 +44,7 @@ class TestOptimize(unittest.TestCase):
         block = B([self.foo,
                    move,
                    self.bar])
-        optimize_block(block)
+        remove_redundancies(block)
         self.assertEquals(block.statements, [self.foo, move, self.bar])
 
     def test_optimize_block_movinst_true(self):
@@ -35,7 +52,7 @@ class TestOptimize(unittest.TestCase):
                    S('command', 'move', '$regA', '$regB'),
                    S('command', 'addu', '$regA', '$regA', 2),
                    self.bar])
-        optimize_block(block)
+        remove_redundancies(block)
         self.assertEquals(block.statements, [self.foo,
                    S('command', 'addu', '$regA', '$regB', 2),
                    self.bar])
@@ -43,11 +60,11 @@ class TestOptimize(unittest.TestCase):
     def test_optimize_block_movinst_false(self):
         statements = [self.foo, \
                       S('command', 'move', '$regA', '$regB'), \
-                      S('command', 'addu', '$regA', '$regC', 2), \
+                      S('command', 'addu', '$regD', '$regC', 2), \
                       self.bar]
 
         block = B(statements)
-        optimize_block(block)
+        remove_redundancies(block)
         self.assertEquals(block.statements, statements)
 
     def test_optimize_block_instr_mov_jal_true(self):
@@ -56,7 +73,7 @@ class TestOptimize(unittest.TestCase):
                    S('command', 'move', '$4', '$regA'),
                    S('command', 'jal', 'L1'),
                    self.bar])
-        optimize_block(block)
+        remove_redundancies(block)
 
         self.assertEquals(block.statements, [self.foo,
                    S('command', 'addu', '$4', '$regC', 2),
@@ -70,7 +87,7 @@ class TestOptimize(unittest.TestCase):
                      S('command', 'jal', 'L1'), \
                      self.bar]
         block = B(arguments)
-        optimize_block(block)
+        remove_redundancies(block)
 
         self.assertEquals(block.statements, arguments)
 
@@ -79,7 +96,7 @@ class TestOptimize(unittest.TestCase):
                    S('command', 'sw', '$regA', '$regB'),
                    S('command', 'lw', '$regA', '$regB'),
                    self.bar])
-        optimize_block(block)
+        remove_redundancies(block)
 
         self.assertEquals(block.statements, [self.foo,
                    S('command', 'sw', '$regA', '$regB'),
@@ -91,7 +108,7 @@ class TestOptimize(unittest.TestCase):
                      S('command', 'lw', '$regD', '$regC'), \
                      self.bar]
         block = B(arguments)
-        optimize_block(block)
+        remove_redundancies(block)
 
         self.assertEquals(block.statements, arguments)
 
@@ -99,7 +116,7 @@ class TestOptimize(unittest.TestCase):
         block = B([self.foo,
                    S('command', 'sll', '$regA', '$regA', 0),
                    self.bar])
-        optimize_block(block)
+        remove_redundancies(block)
 
         self.assertEquals(block.statements, [self.foo, self.bar])
 
@@ -108,7 +125,7 @@ class TestOptimize(unittest.TestCase):
                      S('command', 'sll', '$regA', '$regB', 0), \
                      self.bar]
         block = B(arguments)
-        optimize_block(block)
+        remove_redundancies(block)
 
         self.assertEquals(block.statements, arguments)
 
@@ -116,7 +133,7 @@ class TestOptimize(unittest.TestCase):
                      S('command', 'sll', '$regA', '$regA', 1), \
                      self.bar]
         block2 = B(arguments2)
-        optimize_block(block2)
+        remove_redundancies(block2)
 
         self.assertEquals(block2.statements, arguments2)
 
@@ -125,7 +142,7 @@ class TestOptimize(unittest.TestCase):
                    S('command', 'addu', '$regA', '$regA', 10),
                    S('command', 'lw', '$regB', '0($regA)'),
                    self.bar])
-        optimize_block(block)
+        remove_redundancies(block)
 
         self.assertEquals(block.statements, [self.foo,
                    S('command', 'lw', '$regB', '10($regA)'),
@@ -137,7 +154,7 @@ class TestOptimize(unittest.TestCase):
                      S('command', 'lw', '$regB', '0($regC)'), \
                      self.bar]
         block = B(arguments)
-        optimize_block(block)
+        remove_redundancies(block)
 
         arguments2 = [self.foo, \
                      S('command', 'addu', '$regA', '$regB', 10), \
@@ -150,7 +167,7 @@ class TestOptimize(unittest.TestCase):
                      S('command', 'lw', '$regB', '1($regA)'), \
                      self.bar]
         block3 = B(arguments3)
-        optimize_block(block3)
+        remove_redundancies(block3)
 
         self.assertEquals(block.statements, arguments)
         self.assertEquals(block2.statements, arguments2)
@@ -209,7 +226,7 @@ class TestOptimize(unittest.TestCase):
                    S('command', 'move', '$regA', '$regB'),
                    S('command', 'move', '$regB', '$regA'),
                    self.bar])
-        optimize_block(block)
+        remove_redundancies(block)
 
         self.assertEquals(block.statements, [self.foo,
                    S('command', 'move', '$regA', '$regB'),
@@ -221,6 +238,6 @@ class TestOptimize(unittest.TestCase):
                      S('command', 'move', '$regB', '$regC'), \
                      self.bar]
         block = B(arguments)
-        optimize_block(block)
+        remove_redundancies(block)
 
         self.assertEquals(block.statements, arguments)

tests/test_optimize_advanced.py

@@ -3,7 +3,9 @@ from copy import copy
 
 from src.optimize.advanced import eliminate_common_subexpressions, \
         fold_constants, copy_propagation, algebraic_transformations
-from src.statement import Statement as S, Block as B
+from src.statement import Statement as S
+from src.dataflow import BasicBlock as B, generate_flow_graph
+import src.liveness as liveness
 
 
 class TestOptimizeAdvanced(unittest.TestCase):
@@ -22,6 +24,7 @@ class TestOptimizeAdvanced(unittest.TestCase):
         e = [S('command', 'addu', '$8', '$regA', '$regB'), \
              S('command', 'move', '$regC', '$8'), \
              S('command', 'move', '$regD', '$8')]
+        liveness.create_in_out([b])
         eliminate_common_subexpressions(b)
         self.assertEqual(b.statements, e)
 
@@ -30,6 +33,7 @@ class TestOptimizeAdvanced(unittest.TestCase):
                S('command', 'li', '$regA', '0x00000001'),
                S('command', 'addu', '$regD', '$regA', '$regB')])
         e = copy(b.statements)
+        liveness.create_in_out([b])
         eliminate_common_subexpressions(b)
         self.assertEqual(b.statements, e)
 
@@ -49,7 +53,7 @@ class TestOptimizeAdvanced(unittest.TestCase):
                    self.foo,
                    S('command', 'addu', '$3', '$2', '$4'),
                    self.bar])
-                   
+
     def test_copy_propagation_other_arg(self):
         block = B([self.foo,
                    S('command', 'move', '$1', '$2'),

tests/test_reaching_definitions.py

+import unittest
+
+from src.statement import Statement as S
+from src.dataflow import BasicBlock as B, find_basic_blocks, \
+        generate_flow_graph
+from src.reaching_definitions import get_defs, create_gen_kill, create_in_out
+
+
+class TestReachingDefinitions(unittest.TestCase):
+
+    def setUp(self):
+        pass
+
+    def tearDown(self):
+        pass
+
+    def test_get_defs(self):
+        s1 = S('command', 'add', '$3', '$1', '$2')
+        s2 = S('command', 'move', '$1', '$3')
+        s3 = S('command', 'move', '$3', '$2')
+        s4 = S('command', 'li', '$4', '0x00000001')
+        block = B([s1, s2, s3, s4])
+
+        self.assertEqual(get_defs([block]), {
+            '$3': set([s1.sid, s3.sid]),
+            '$1': set([s2.sid]),
+            '$4': set([s4.sid])
+        })
+
+    def test_create_gen_kill(self):
+        s1 = S('command', 'addu', '$3', '$1', '$2')
+        s2 = S('command', 'addu', '$1', '$3', 10)
+        s3 = S('command', 'subu', '$3', '$1', 5)
+        s4 = S('command', 'li', '$4', '0x00000001')
+        block = B([s1, s2, s3, s4])
+
+        create_gen_kill(block, get_defs([block]))
+
+        self.assertEqual(block.gen_set, set([s2.sid, s3.sid, s4.sid]))
+        self.assertEqual(block.kill_set, set([s1.sid]))
+
+    def test_create_in_out(self):
+        s11 = S('command', 'li', 'a', 3)
+        s12 = S('command', 'li', 'b', 5)
+        s13 = S('command', 'li', 'd', 4)
+        s14 = S('command', 'li', 'x', 100)
+        s15 = S('command', 'beq', 'a', 'b', 'L1')
+
+        s21 = S('command', 'addu', 'c', 'a', 'b')
+        s22 = S('command', 'li', 'd', 2)
+
+        s31 = S('label', 'L1')
+        s32 = S('command', 'li', 'c', 4)
+        s33 = S('command', 'mult', 'b', 'd')
+        s34 = S('command', 'mflo', 'temp')
+        s35 = S('command', 'addu', 'return', 'temp', 'c')
+
+        b1, b2, b3 = find_basic_blocks([s11, s12, s13, s14, s15, s21, s22, \
+                                        s31, s32, s33, s34, s35])
+
+        generate_flow_graph([b1, b2, b3])
+        create_in_out([b1, b2, b3])
+
+        self.assertEqual(b1.gen_set, set([s11.sid, s12.sid, s13.sid,
+                                            s14.sid]))
+        self.assertEqual(b1.kill_set, set([s22.sid]))
+        self.assertEqual(b2.gen_set, set([s21.sid, s22.sid]))
+        self.assertEqual(b2.kill_set, set([s13.sid, s32.sid]))
+        self.assertEqual(b3.gen_set, set([s32.sid, s34.sid, s35.sid]))
+        self.assertEqual(b3.kill_set, set([s21.sid]))
+
+        self.assertEqual(b1.reach_in, set())
+        self.assertEqual(b1.reach_out, set([s11.sid, s12.sid, s13.sid,
+                                            s14.sid]))
+        self.assertEqual(b2.reach_in, set([s11.sid, s12.sid, s13.sid,
+                                            s14.sid]))
+        self.assertEqual(b2.reach_out, set([s21.sid, s22.sid, s11.sid, \
+                                            s12.sid, s14.sid]))
+        self.assertEqual(b3.reach_in, set([s21.sid, s22.sid, s11.sid, \
+                                            s12.sid, s13.sid, s14.sid]))
+        self.assertEqual(b3.reach_out, set([s32.sid, s34.sid, s35.sid, \
+                                            s22.sid, s11.sid, s12.sid, \
+                                            s13.sid, s14.sid]))

tests/test_statement.py

@@ -36,9 +36,10 @@ class TestStatement(unittest.TestCase):
         self.assertFalse(S('comment', 'foo', inline=False).is_label())
         self.assertFalse(S('directive', 'foo').is_command())
 
-    def test_is_inline_comment(self):
-        self.assertTrue(S('comment', 'foo', inline=True).is_inline_comment())
-        self.assertFalse(S('comment', 'foo', inline=False).is_inline_comment())
+    def test_has_inline_comment(self):
+        self.assertTrue(S('comment', 'foo', comment='a').has_inline_comment())
+        self.assertFalse(S('comment', 'foo', comment='').has_inline_comment())
+        self.assertFalse(S('comment', 'foo').has_inline_comment())
 
     def test_jump_target(self):
         self.assertEqual(S('command', 'j', 'foo').jump_target(), 'foo')
@@ -93,3 +94,79 @@ class TestStatement(unittest.TestCase):
         self.assertTrue(S('command', 'addu', '$1', '$2', '$3').is_arith())
         self.assertFalse(S('command', 'foo').is_arith())
         self.assertFalse(S('label', 'addu').is_arith())
+
+    def test_get_def_true(self):
+        a = ['a']
+
+        self.assertEqual(S('command', 'move', 'a', 'b').get_def(), a)
+        self.assertEqual(S('command', 'subu', 'a', 'b', 'c').get_def(), a)
+        self.assertEqual(S('command', 'addu', 'a', 'b', 'c').get_def(), a)
+        self.assertEqual(S('command', 'div', 'a', 'b', 'c').get_def(), a)
+        self.assertEqual(S('command', 'sll', 'a', 'b', 'c').get_def(), a)
+        self.assertEqual(S('command', 'srl', 'a', 'b', 'c').get_def(), a)
+        self.assertEqual(S('command', 'la', 'a', '16($fp)').get_def(), a)
+        self.assertEqual(S('command', 'li', 'a', '16($fp)').get_def(), a)
+        self.assertEqual(S('command', 'lw', 'a', 'b').get_def(), a)
+        self.assertEqual(S('command', 'l.d', 'a', 'b').get_def(), a)
+        self.assertEqual(S('command', 'add.d', 'a', 'b', 'c').get_def(), a)
+        self.assertEqual(S('command', 'neg.d', 'a', 'b').get_def(), a)
+        self.assertEqual(S('command', 'sub.d', 'a', 'b', 'c').get_def(), a)
+        self.assertEqual(S('command', 'slt', 'a', 'b').get_def(), a)
+        self.assertEqual(S('command', 'xori', 'a', 'b', '0x0000').get_def(), a)
+        self.assertEqual(S('command', 'mov.d', 'a', 'b').get_def(), a)
+        self.assertEqual(S('command', 'dmfc1', 'a', '$f0').get_def(), a)
+        self.assertEqual(S('command', 'mtc1', 'b', 'a').get_def(), a)
+        self.assertEqual(S('command', 'trunc.w.d', 'a', 'b', 'c').get_def(), a)
+
+    def test_get_def_false(self):
+        self.assertEqual(S('command', 'bne', 'a', 'b', 'L1').get_def(), [])
+        self.assertEqual(S('command', 'beq', 'a', 'b', 'L1').get_def(), [])
+
+    def test_get_use_true(self):
+        arg1 = ['$1']
+        arg2 = ['$1', '$2']
+
+        self.assertEqual(S('command', 'addu', '$3', '$1', '$2').get_use(), \
+                arg2)
+        self.assertEqual(S('command', 'subu', '$3', '$1', '$2').get_use(), \
+                arg2)
+        self.assertEqual(S('command', 'mult', '$1', '$2').get_use(), arg2)
+        self.assertEqual(S('command', 'div', '$3', '$1', '$2').get_use(), arg2)
+        self.assertEqual(S('command', 'move', '$2', '$1').get_use(), arg1)
+        self.assertEqual(S('command', 'beq', '$1', '$2', '$L1').get_use(), \
+                arg2)
+        self.assertEqual(S('command', 'bne', '$1', '$2', '$L1').get_use(), \
+                arg2)
+        self.assertEqual(S('command', 'sll', '$2', '$1', 2).get_use(), arg1)
+        self.assertEqual(S('command', 'lb', '$2', '10($1)').get_use(), arg1)
+        self.assertEqual(S('command', 'lw', '$2', '10($1)').get_use(), arg1)
+        self.assertEqual(S('command', 'la', '$2', '10($1)').get_use(), arg1)
+        self.assertEqual(S('command', 'lb', '$2', 'n.7').get_use(), ['n.7'])
+        self.assertEqual(S('command', 'lbu', '$2', '10($1)').get_use(), arg1)
+        self.assertEqual(S('command', 'l.d', '$2', '10($1)').get_use(), arg1)
+        self.assertEqual(S('command', 's.d', '$1', '10($2)').get_use(), \
+                arg2)
+        self.assertEqual(S('command', 's.s', '$1', '10($2)').get_use(), \
+                arg2)
+        self.assertEqual(S('command', 'sw', '$1', '10($2)').get_use(), \
+                arg2)
+        self.assertEqual(S('command', 'sb', '$1', '10($2)').get_use(), \
+                arg2)
+        self.assertEqual(S('command', 'mtc1', '$1', '$2').get_use(), arg1)
+        self.assertEqual(S('command', 'add.d', '$3', '$1', '$2').get_use(), \
+                arg2)
+        self.assertEqual(S('command', 'sub.d', '$3', '$1', '$2').get_use(), \
+                arg2)
+        self.assertEqual(S('command', 'div.d', '$3', '$1', '$2').get_use(), \
+                arg2)
+        self.assertEqual(S('command', 'mul.d', '$3', '$1', '$2').get_use(), \
+                arg2)
+        self.assertEqual(S('command', 'neg.d', '$2', '$1').get_use(), arg1)
+        self.assertEqual(S('command', 'abs.d', '$2', '$1').get_use(), arg1)
+        self.assertEqual(S('command', 'dsz', '10($1)', '$2').get_use(), arg1)
+        self.assertEqual(S('command', 'dsw', '$1', '10($2)').get_use(), arg2)
+        self.assertEqual(S('command', 'c.lt.d', '$1', '$2').get_use(), arg2)
+        self.assertEqual(S('command', 'bgez', '$1', '$2').get_use(), arg1)
+        self.assertEqual(S('command', 'bltz', '$1', '$2').get_use(), arg1)
+        self.assertEqual(S('command', 'trunc.w.d', '$3', '$1', '$2').get_use(),
+                         arg2)

tests/test_writer.py

+import unittest
+
+from src.writer import write_statements
+from src.statement import Statement as S, Block as B
+
+
+class TestWriter(unittest.TestCase):
+
+    def setUp(self):
+        self.foo = S('command', 'move', '$regA', '$regB')
+        self.bar = S('command', 'addu', '$regC', '$regA', '$regB')
+
+    def tearDown(self):
+        del self.foo
+        del self.bar
+        
+    def test_writer_one(self):
+        output = write_statements([self.foo])
+        expect = "\tmove\t$regA,$regB\n"
+        self.assertEqual(output, expect)
+        
+    def test_writer_longname(self):
+        command = S('command', 'movemovemove', '$regA', '$regB')
+        output = write_statements([command])
+        expect = "\tmovemovemove $regA,$regB\n"
+        self.assertEqual(output, expect)
+        
+    def test_writer_several(self):
+        output = write_statements([self.foo, self.bar, self.foo])
+        expect = "\tmove\t$regA,$regB\n" \
+                 + "\taddu\t$regC,$regA,$regB\n" \
+                 + "\tmove\t$regA,$regB\n"
+        self.assertEqual(output, expect)
+        
+    def test_writer_with_label(self):
+        label = S('label', '$L1')
+        output = write_statements([self.foo, label, self.bar])
+        expect = "\tmove\t$regA,$regB\n" \
+                 + "$L1:\n" \
+                 + "\taddu\t$regC,$regA,$regB\n"
+        self.assertEqual(output, expect)
+        
+    def test_writer_with_comment(self):
+        comment = S('comment', 'tralala')
+        output = write_statements([self.foo, comment, self.bar])
+        expect = "\tmove\t$regA,$regB\n" \
+                 + "\n#tralala\n\n" \
+                 + "\taddu\t$regC,$regA,$regB\n"
+        self.assertEqual(output, expect)
+        
+    def test_writer_with_comment_non_tabbed(self):
+        directive = S('comment', 'tralala')
+        output = write_statements([directive, self.foo, self.bar])
+        expect = "\n#tralala\n\n" \
+                 + "\tmove\t$regA,$regB\n" \
+                 + "\taddu\t$regC,$regA,$regB\n"
+        self.assertEqual(output, expect)
+        
+    def test_writer_with_inlinecomment(self):
+        self.foo.options['comment'] = 'tralala'
+        output = write_statements([self.foo, self.bar])
+        expect = "\tmove\t$regA,$regB" \
+                 + "\t\t#tralala\n" \
+                 + "\taddu\t$regC,$regA,$regB\n"
+        self.assertEqual(output, expect)
+        
+    def test_writer_with_directive(self):
+        directive = S('directive', '.tralala trololo')
+        output = write_statements([self.foo, directive, self.bar])
+        expect = "\tmove\t$regA,$regB\n" \
+                 + "\t.tralala trololo\n" \
+                 + "\taddu\t$regC,$regA,$regB\n"
+        self.assertEqual(output, expect)