Merge branch 'master' of github.com:taddeus/peephole

benchmarks/build/test.s

+	.file	1 "test.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 -O0 -o
+
+gcc2_compiled.:
+__gnu_compiled_c:
+	.text
+	.align	2
+	.globl	main
+
+	.text
+
+	.loc	1 3
+	.ent	main
+main:
+	.frame	$fp,48,$31		# vars= 24, regs= 2/0, args= 16, extra= 0
+	.mask	0xc0000000,-4
+	.fmask	0x00000000,0
+	subu	$sp,$sp,48
+	sw	$31,44($sp)
+	sw	$fp,40($sp)
+	move	$fp,$sp
+	jal	__main
+	li	$2,0x00000002		# 2
+	sw	$2,16($fp)
+	li	$2,0x00000005		# 5
+	sw	$2,20($fp)
+	lw	$2,16($fp)
+	lw	$3,20($fp)
+	mult	$2,$3
+	mflo	$2
+	sw	$2,24($fp)
+	lw	$2,16($fp)
+	move	$4,$2
+	sll	$3,$4,1
+	addu	$3,$3,$2
+	sll	$2,$3,1
+	sw	$2,28($fp)
+	li	$2,0x00000015		# 21
+	sw	$2,32($fp)
+	move	$2,$0
+	j	$L1
+$L1:
+	move	$sp,$fp			# sp not trusted here
+	lw	$31,44($sp)
+	lw	$fp,40($sp)
+	addu	$sp,$sp,48
+	j	$31
+	.end	main

benchmarks/test.c

+#include <stdio.h>
+
+int main(void) {
+    int a = 2, b = 5, c = a * b, d = a * 6, e = 3 * 7;
+
+    return 0;
+}

report/report.tex

@@ -35,7 +35,7 @@ the keywords in to an action.
 
 \section{Design}
 
-There are two general types of of optimizations of the assembly code, global
+There are two general types of optimizations of the assembly code, global
 optimizations and optimizations on a so-called basic block. These optimizations
 will be discussed separately
 
@@ -99,6 +99,16 @@ Appendix \ref{opt}.
 A more advanced optimization is common subexpression elimination. This means
 that expensive operations as a multiplication or addition are performed only
 once and the result is then `copied' into variables where needed.
+\begin{verbatim}
+
+addu	$2,$4,$3              addu = $t1, $4, $3
+...                        mov = $2, $t1
+...                   ->   ...
+...                        ...
+addu	$5,$4,$3		   mov = $4, $t1
+
+\end{verbatim}
+
 
 A standard method for doing this is the creation of a DAG or Directed Acyclic
 Graph. However, this requires a fairly advanced implementation. Our
@@ -112,27 +122,13 @@ We now add the instruction above the first use, and write the result in a new
 variable. Then all occurrences of this expression can be replaced by a move of
 from new variable into the original destination variable of the instruction.
 
-This is a less efficient method then the DAG, but because the basic blocks are
+This is a less efficient method then the dag, but because the basic blocks are
 in general not very large and the execution time of the optimizer is not a
 primary concern, this is not a big problem.
 
-\subsubsection*{Constant folding}
+\subsubsection*{Fold constants}
 
-Another optimization is to do constant folding. Constant folding is replacing
-a expensive step like addition with a more simple step like loading a constant.
-Of course, this is not always possible. It is possible in cases where you apply
-an operation on two constants, or a constant and a variable of which you know
-for sure that it always has a certain value at that point. For example:
-\begin{verbatim}
-li   $regA, 1               li $regA, 1
-addu $regB, $regA, 2    ->  li $regB, 3
-\end{verbatim}
-Of course, if \texttt{\$regA} is not used after this, it can be removed, which
-will be done by the dead code elimination.
 
-One problem we encountered with this is that the use of a \texttt{li} is that
-the program often also stores this in the memory, so we had to check whether
-this was necessary here as well.
 
 \subsubsection*{Copy propagation}
 
@@ -161,7 +157,8 @@ An example would be the following:
 \begin{verbatim}
 move $regA, $regB           move $regA, $regB
 ...                         ...
-Code not writing $regA, $regB   ->  ...
+Code not writing $regA, ->  ...
+$regB                       ...
 ...                         ...
 addu $regC, $regA, ...      addu $regC, $regB, ...
 \end{verbatim}
@@ -171,18 +168,7 @@ removed by the dead code elimination.
 
 \subsubsection*{Algebraic transformations}
 
-Some expression can easily be replaced with more simple once if you look at
-what they are saying algebraically. An example is the statement $x = y + 0$, or
-in Assembly \texttt{addu \$1, \$2, 0}. This can easily be changed into $x = y$
-or \texttt{move \$1, \$2}.
-
-Another case is the multiplication with a power of two. This can be done way
-more efficiently by shifting left a number of times. An example:
-\texttt{mult \$regA, \$regB, 4    ->  sll  \$regA, \$regB, 2}. We perform this
-optimization for any multiplication with a power of two.
 
-There are a number of such cases, all of which are once again stated in
-appendix \ref{opt}. 
 
 \section{Implementation}
 
@@ -206,7 +192,7 @@ languages like we should do otherwise since Lex and Yacc are coupled with C.
 
 The decision was made to not recognize exactly every possible instruction in
 the parser, but only if something is for example a command, a comment or a gcc
-directive. We then transform per line to a object called a Statement. A
+directive. We then transform per line to an object called a Statement. A
 statement has a type, a name and optionally a list of arguments. These
 statements together form a statement list, which is placed in another object
 called a Block. In the beginning there is one block for the entire program, but
@@ -219,7 +205,7 @@ The optimizations are done in two different steps. First the global
 optimizations are performed, which are only the optimizations on branch-jump
 constructions. This is done repeatedly until there are no more changes.
 
-After all possible global optimizations are done, the program is separated into
+After all possible global optimizations are done, the program is seperated into
 basic blocks. The algorithm to do this is described earlier, and means all
 jump and branch instructions are called leaders, as are their targets. A basic
 block then goes from leader to leader.
@@ -231,7 +217,8 @@ steps can be done to optimize something.
 \subsection{Writing}
 
 Once all the optimizations have been done, the IR needs to be rewritten into
-Assembly code, so the xgcc cross compiler can make binary code out of it.
+Assembly code. After this step the xgcc crosscompiler can make binary code from
+the generated Assembly code.
 
 The writer expects a list of statements, so first the blocks have to be
 concatenated again into a list. After this is done, the list is passed on to

src/optimize/advanced.py

@@ -147,10 +147,27 @@ def fold_constants(block):
         elif s.name == 'lw' and s[1] in constants:
             # Usage of variable with constant value
             register[s[0]] = constants[s[1]]
-        elif s.name in ['addu', 'subu', 'mult', 'div']:
-            # TODO: implement 'mult' optimization
-            # Calculation with constants
-            rd, rs, rt = s[0], s[1], s[2]
+        elif s.name == 'mflo':
+            # Move of `Lo' register to another register
+            register[s[0]] = register['Lo']
+        elif s.name == 'mfhi':
+            # Move of `Hi' register to another register
+            register[s[0]] = register['Hi']
+        elif s.name in ['mult', 'div'] \
+                and s[0] in register and s[1] in register:
+            # Multiplication/division with constants
+            rs, rt = s
+
+            if s.name == 'mult':
+                binary = bin(register[rs] * register[rt])[2:]
+                binary = '0' * (64 - len(binary)) + binary
+                register['Hi'] = int(binary[:32], base=2)
+                register['Lo'] = int(binary[32:], base=2)
+            elif s.name == 'div':
+                register['Lo'], register['Hi'] = divmod(rs, rt)
+        elif s.name in ['addu', 'subu']:
+            # Addition/subtraction with constants
+            rd, rs, rt = s
             rs_known = rs in register
             rt_known = rt in register
 
@@ -167,22 +184,16 @@ def fold_constants(block):
                 if s.name == 'subu':
                     result = to_hex(rs_val - rt_val)
 
-                if s.name == 'mult':
-                    result = to_hex(rs_val * rt_val)
-
-                if s.name == 'div':
-                    result = to_hex(rs_val / rt_val)
-
                 block.replace(1, [S('command', 'li', rd, result)])
                 register[rd] = result
                 changed = True
             elif rt_known:
-                # c = 10        ->  b = a + 10
+                # a = 10        ->  b = c + 10
                 # b = c + a
                 s[2] = register[rt]
                 changed = True
             elif rs_known and s.name == 'addu':
-                # a = 10        ->  b = c + 10
+                # c = 10        ->  b = a + 10
                 # b = c + a
                 s[1] = rt
                 s[2] = register[rs]