scale.score.trans
Class AASR

java.lang.Object
  scale.score.trans.Optimization
      scale.score.trans.AASR

public class AASR
extends Optimization

This class replaces array element address calculations in a loop with simple additions.

$Id: AASR.java,v 1.57 2007-10-04 19:58:35 burrill Exp $

Copyright 2008 by the Scale Compiler Group,
Department of Computer Science
University of Massachusetts,
Amherst MA. 01003, USA
All Rights Reserved.

The array address strength reduction optimization attempts to replace the complex array element address calculation with a simple addition. For example, consider the loop

   for (int i = 0; i < n; i++)
     a(i) =
 

AASR transforms this to

   int *add = &a[0];
   for (int i = 0; i < n; i++) {
     *add =
     add = ((char *) add) + sizeof(int);
   }
 

which eliminates the multiplication, of the loop induction variable by the size of the array element, which would normally occur on each iteration. Arrays of higher rank result in even more savings because their array element address calculation requires more multiplies and additions.

Note that if the array is referenced in the loop in some manner that can't be transformed, then register pressure has been increased by one register (e.g., add). And, even though the induction variable is no longer referenced for each array access, it is still needed for the loop termination test.

A transformation is legal if:

• the loop has an induction variable with loop independent initial value and increment,
• the subscript expression depends on the loop induction variable,
• all other elements of the subscript expression are loop independent, and
• the subscript expression is affine.

It is beneficial if:

• the loop is an inner-most loop,
• the loop is smaller than maxLoopSize, and
• the loop does not contain a function call.

Field Summary

static boolean	classTrace True if traces are to be performed.
static int	maxLoopSize Maximum loop size in CFG nodes.
static boolean	useHeuristics If true, use heuristics that prune the cases where the optimization is applied.
static boolean	useSizeof Set true to use the sizeof() instead of the machine specific array element size in addressable units.

Fields inherited from class scale.score.trans.Optimization

dChanged, fpReorder, hasDummyAliases, IN_SSA, minimumExecutionCost, NA_SSA, NO_SSA, rChanged, scribble, signedIntsWrapOnOverflow, trace, un, unsafe, unsignedIntsWrapOnOverflow, VALID_SSA

Constructor Summary

AASR(Scribble scribble)

Method Summary

static int	newCFGNodes() Return the number of new nodes created.
void	perform() Do strength reduction of array accesses.
static int	replaced() Return the current number of array loads replaced.

Methods inherited from class scale.score.trans.Optimization

assertTrace, assertTrace, assertTrace, assertTrace, assertTrace, genTemp, insertStores, requiresSSA, setTrace, sort

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

classTrace

public static boolean classTrace

True if traces are to be performed.

useHeuristics

public static boolean useHeuristics

If true, use heuristics that prune the cases where the optimization is applied.

useSizeof

public static boolean useSizeof

Set true to use the sizeof() instead of the machine specific array element size in addressable units.

maxLoopSize

public static int maxLoopSize

Maximum loop size in CFG nodes.

Constructor Detail

AASR

public AASR(Scribble scribble)

Method Detail

replaced

public static int replaced()

Return the current number of array loads replaced.

newCFGNodes

public static int newCFGNodes()

Return the number of new nodes created.

perform

public void perform()

Do strength reduction of array accesses.

Specified by:

perform in class Optimization