org.opencores.mapping
Class FlowMap

java.lang.Object
  |
  +--org.opencores.mapping.FlowMap

public final class FlowMap

extends java.lang.Object

Performs unit-delay optimal mapping for LUTs - FlowMap algorithm.

See Also:

,

Inner Class Summary

(package private) class

FlowMap.Variable Internal class used for building larger LUTs, representing directed acyclic subgraph input.

Field Summary

static boolean	OPTIMIZE_SIZE Whether FlowMap should optimize space
private int	tempCnt current temp flag

Constructor Summary

FlowMap()

Method Summary

private NodeLUT	buildLUT(NodeLUT child, java.util.Vector GPI) Collapses DAG of same nodes with temp==tempCnt-1 into child, using shortest-first joining.
private void	buildSame(NodeLUT child, int label, java.util.Vector LUTs) adds all childs predecessors with same label into LUTs Vector
private boolean	flow(java.util.Vector PI, Node t, int label) Calculates flow from source s to sink t or to node with label 'label'.
private boolean	flowR(Node sink, boolean top, int label) Tries to find flow from source s to sink t.
private NodeLUT	joinLUTs(NodeLUT child) Collapses tree of same labeled nodes into s, using shortest-first joining.
private int	label(Node t) Calculates label of node t.
void	labeling(Graph g) first step of algorithm - labeling
Graph	mapping(Graph g) Second step of the algorithm - mapping.
private void	mark(Node t, java.util.Vector PI, int mil) Mark predecessor nodes ot t with tempCnt.
private void	markAccessible(NodeLUT sink, boolean top) Similar to routine A of Ford Fulerson max-flow algorithm, marks visited nodes with .visited flag.

Methods inherited from class java.lang.Object

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

Field Detail

OPTIMIZE_SIZE

public static boolean OPTIMIZE_SIZE

Whether FlowMap should optimize space

tempCnt

private int tempCnt

current temp flag

Constructor Detail

FlowMap

public FlowMap()

Method Detail

labeling

public void labeling(Graph g)

first step of algorithm - labeling

Parameters:

g - graph to label. Labels are stored in node.y

label

private int label(Node t)

Calculates label of node t.

Parameters:

n - node to calculate label

Returns:

label of t

mark

private void mark(Node t,
                  java.util.Vector PI,
                  int mil)

Mark predecessor nodes ot t with tempCnt.

Parameters:

t - node to mark predecessors

PI - vector to put primary input nets in

flow

private boolean flow(java.util.Vector PI,
                     Node t,
                     int label)

Calculates flow from source s to sink t or to node with label 'label'. Also mark function should be called to set .temp and .flag = false for all predecessors

Parameters:

s - source node

t - sink node

label - minimum label to stop

Returns:

true if there is flow > Conf.K between source and sink

See Also:

mark

flowR

private boolean flowR(Node sink,
                      boolean top,
                      int label)

Tries to find flow from source s to sink t.

Each node is splitted into two parts - top and bottom. Top can access bottoms of predecessors, bottom can access successor tops. Top-bottom link is valid if node .flag == false.

Mark function should be called to set .temp and .flag = false for all predecessors

Parameters:

sink - source node

top - designates top access to sink node

label - minimum label to stop

Returns:

true if there is a flow between source and sink

See Also:

mark, markAccessible, Ford, Jr. and D.R. Fulkerson: Maximal Flow Trough a Network

mapping

public Graph mapping(Graph g)

Second step of the algorithm - mapping. Call labeling first.

Parameters:

g - labeled graph

Returns:

mapped graph (completely rebuilt)

See Also:

labeling

joinLUTs

private NodeLUT joinLUTs(NodeLUT child)

Collapses tree of same labeled nodes into s, using shortest-first joining.

Parameters:

child - sink node

Returns:

collapsed tree node

markAccessible

private void markAccessible(NodeLUT sink,
                            boolean top)

Similar to routine A of Ford Fulerson max-flow algorithm, marks visited nodes with .visited flag.

Each node is splitted into two parts - top and bottom. Top can access bottoms of predecessors, bottom can access successor tops. Top-bottom link is valid if node .flag == false.

Parameters:

sink - sink node

top - designates top access to sink node

See Also:

flowR,

buildSame

private void buildSame(NodeLUT child,
                       int label,
                       java.util.Vector LUTs)

adds all childs predecessors with same label into LUTs Vector

buildLUT

private NodeLUT buildLUT(NodeLUT child,
                         java.util.Vector GPI)

Collapses DAG of same nodes with temp==tempCnt-1 into child, using shortest-first joining. .temp is incremented, to disable looping.

Parameters:

child - sink node

Returns:

collapsed tree node