File: Mongoose_ImproveFM.cpp

package info (click to toggle)
suitesparse 1%3A5.4.0%2Bdfsg-1
  • links: PTS, VCS
  • area: main
  • in suites: buster
  • size: 138,928 kB
  • sloc: ansic: 389,614; cpp: 24,213; makefile: 5,965; fortran: 1,927; java: 1,808; csh: 1,750; ruby: 725; sh: 226; perl: 225; python: 209; sed: 164; awk: 60
file content (331 lines) | stat: -rw-r--r-- 12,269 bytes parent folder | download | duplicates (3)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
/* ========================================================================== */
/* === Source/Mongoose_ImproveFM.cpp ======================================== */
/* ========================================================================== */

/* -----------------------------------------------------------------------------
 * Mongoose Graph Partitioning Library  Copyright (C) 2017-2018,
 * Scott P. Kolodziej, Nuri S. Yeralan, Timothy A. Davis, William W. Hager
 * Mongoose is licensed under Version 3 of the GNU General Public License.
 * Mongoose is also available under other licenses; contact authors for details.
 * -------------------------------------------------------------------------- */

#include "Mongoose_ImproveFM.hpp"
#include "Mongoose_BoundaryHeap.hpp"
#include "Mongoose_Debug.hpp"
#include "Mongoose_Internal.hpp"
#include "Mongoose_Logger.hpp"

namespace Mongoose
{

//-----------------------------------------------------------------------------
// Wrapper for Fidducia-Mattheyes cut improvement.
//-----------------------------------------------------------------------------
void improveCutUsingFM(EdgeCutProblem *graph, const EdgeCut_Options *options)
{
    Logger::tic(FMTiming);

    if (!options->use_FM)
        return;

    double heuCost = INFINITY;
    for (Int i = 0;
         i < options->FM_max_num_refinements && graph->heuCost < heuCost; i++)
    {
        heuCost = graph->heuCost;
        fmRefine_worker(graph, options);
    }

    Logger::toc(FMTiming);
}

//-----------------------------------------------------------------------------
// Make a number of partition moves while considering the impact on problem
// balance.
//-----------------------------------------------------------------------------
void fmRefine_worker(EdgeCutProblem *graph, const EdgeCut_Options *options)
{
    double *Gw          = graph->w;
    double W            = graph->W;
    Int **bhHeap        = graph->bhHeap;
    Int *bhSize         = graph->bhSize;
    Int *externalDegree = graph->externalDegree;
    double *gains       = graph->vertexGains;
    bool *partition     = graph->partition;

    /* Keep a stack of moved vertices. */
    Int *stack = graph->matchmap;
    Int head = 0, tail = 0;

    /* create & initialize a working cost and a best cost. */
    struct CutCost workingCost, bestCost;
    workingCost.heuCost = bestCost.heuCost = graph->heuCost;
    workingCost.cutCost = bestCost.cutCost = graph->cutCost;
    workingCost.W[0] = bestCost.W[0] = graph->W0;
    workingCost.W[1] = bestCost.W[1] = graph->W1;
    workingCost.imbalance = bestCost.imbalance = graph->imbalance;

    /* Tolerance and the linear penalty to assess. */
    double tol = options->soft_split_tolerance;
    double H   = graph->H;

    Int fmSearchDepth   = options->FM_search_depth;
    Int fmConsiderCount = options->FM_consider_count;
    Int i               = 0;
    bool productive     = true;
    for (; i < fmSearchDepth && productive; i++)
    {
        productive = false;

        /* Look for the best vertex to swap: */
        struct SwapCandidate bestCandidate;
        for (Int h = 0; h < 2; h++)
        {
            Int *heap = bhHeap[h];
            Int size  = bhSize[h];
            for (Int c = 0; c < fmConsiderCount && c < size; c++)
            {
                /* Read the vertex, and if it's marked, try the next one. */
                Int v = heap[c];
                if (graph->isMarked(v))
                {
                    continue;
                }

                /* Read the gain for the vertex. */
                double gain = gains[v];

                /* The balance penalty is the penalty to assess for the move. */
                double vertexWeight = (Gw) ? Gw[v] : 1;
                double imbalance    = workingCost.imbalance
                                   + (h ? -1.0 : 1.0) * (vertexWeight / W);
                double absImbalance = fabs(imbalance);
                double imbalanceDelta
                    = absImbalance - fabs(workingCost.imbalance);

                /* If the move hurts the balance past tol, add a penalty. */
                double balPenalty = 0.0;
                if (imbalanceDelta > 0 && absImbalance > tol)
                {
                    balPenalty = absImbalance * H;
                }

                /* Heuristic cost is the cut cost reduced by the gain for making
                 * this move. The gain for the move is amplified by any impact
                 * to the balance penalty. */
                double heuCost = workingCost.cutCost - (gain - balPenalty);

                /* If our heuristic value is better than the running one: */
                if (heuCost < bestCandidate.heuCost)
                {
                    bestCandidate.vertex       = v;
                    bestCandidate.partition    = static_cast<bool>(h);
                    bestCandidate.vertexWeight = vertexWeight;
                    bestCandidate.gain         = gain;
                    bestCandidate.bhPosition   = c;
                    bestCandidate.imbalance    = imbalance;
                    bestCandidate.heuCost      = heuCost;
                }
            }
        }

        /* If we were able to find the best unmoved boundary vertex: */
        if (bestCandidate.heuCost < INFINITY)
        {
            productive = true;
            graph->mark(bestCandidate.vertex);

            /* Move the vertex from the boundary into the move set. */
            bhRemove(graph, options, bestCandidate.vertex, bestCandidate.gain,
                     bestCandidate.partition, bestCandidate.bhPosition);
            stack[tail++] = bestCandidate.vertex;

            /* Swap & update the vertex and its neighbors afterwards. */
            fmSwap(graph, options, bestCandidate.vertex, bestCandidate.gain,
                   bestCandidate.partition);

            /* Update the cut cost. */
            workingCost.cutCost -= 2.0 * bestCandidate.gain;
            workingCost.W[bestCandidate.partition]
                -= bestCandidate.vertexWeight;
            workingCost.W[!bestCandidate.partition]
                += bestCandidate.vertexWeight;
            workingCost.imbalance = bestCandidate.imbalance;
            double absImbalance   = fabs(bestCandidate.imbalance);
            workingCost.heuCost
                = workingCost.cutCost
                  + (absImbalance > tol ? absImbalance * H : 0.0);

            /* Commit the cut if it's better. */
            if (workingCost.heuCost < bestCost.heuCost)
            {
                bestCost = workingCost;
                head     = tail;
                i        = 0;
            }
        }
    }

    /* We've exhausted our search space, so undo all suboptimal moves. */
    for (Int u = tail - 1; u >= head; u--)
    {
        Int vertex           = stack[u];
        Int bhVertexPosition = graph->BH_getIndex(vertex);

        /* Unmark this vertex. */
        graph->unmark(vertex);

        /* It is possible, although rare, that a vertex may have gone
         * from not in the boundary to an undo state that places it in
         * the boundary. It is also possible that a previous swap added
         * this vertex to the boundary already. */
        if (bhVertexPosition != -1)
        {
            bhRemove(graph, options, vertex, gains[vertex], partition[vertex],
                     bhVertexPosition);
        }

        /* Swap the partition and compute the impact on neighbors. */
        fmSwap(graph, options, vertex, gains[vertex], partition[vertex]);
        if (externalDegree[vertex] > 0)
            bhInsert(graph, vertex);
    }

    // clear the marks from all the vertices
    graph->clearMarkArray();

    /* Re-add any vertices that were moved that are still on the boundary. */
    for (Int k = 0; k < head; k++)
    {
        Int vertex = stack[k];
        if (externalDegree[vertex] > 0 && !graph->BH_inBoundary(vertex))
        {
            bhInsert(graph, vertex);
        }
    }

    // clear the marks from all the vertices
    graph->clearMarkArray();

    /* Save the best cost back into the graph. */
    graph->heuCost   = bestCost.heuCost;
    graph->cutCost   = bestCost.cutCost;
    graph->W0        = bestCost.W[0];
    graph->W1        = bestCost.W[1];
    graph->imbalance = bestCost.imbalance;
}

//-----------------------------------------------------------------------------
// This function swaps the partition of a vertex
//-----------------------------------------------------------------------------
void fmSwap(EdgeCutProblem *graph, const EdgeCut_Options *options, Int vertex, double gain,
            bool oldPartition)
{
    Int *Gp             = graph->p;
    Int *Gi             = graph->i;
    double *Gx          = graph->x;
    bool *partition     = graph->partition;
    double *gains       = graph->vertexGains;
    Int *externalDegree = graph->externalDegree;
    Int **bhHeap        = graph->bhHeap;
    Int *bhSize         = graph->bhSize;

    /* Swap partitions */
    bool newPartition = !oldPartition;
    partition[vertex] = newPartition;
    gains[vertex]     = -gain;

    /* Update neighbors. */
    Int exD = 0;
    for (Int p = Gp[vertex]; p < Gp[vertex + 1]; p++)
    {
        Int neighbor           = Gi[p];
        bool neighborPartition = partition[neighbor];
        bool sameSide          = (newPartition == neighborPartition);

        /* Update the bestCandidate vertex's external degree. */
        if (!sameSide)
            exD++;

        /* Update the neighbor's gain. */
        double edgeWeight   = (Gx) ? Gx[p] : 1;
        double neighborGain = gains[neighbor];
        neighborGain += 2 * (sameSide ? -edgeWeight : edgeWeight);
        gains[neighbor] = neighborGain;

        /* Update the neighbor's external degree. */
        Int neighborExD = externalDegree[neighbor];
        neighborExD += (sameSide ? -1 : 1);
        externalDegree[neighbor] = neighborExD;
        Int position             = graph->BH_getIndex(neighbor);

        /* If the neighbor was in a heap: */
        if (position != -1)
        {
            /* If it had its externalDegree reduced to 0, remove it from the
             * heap. */
            if (neighborExD == 0)
            {
                bhRemove(graph, options, neighbor, neighborGain,
                         neighborPartition, position);
            }
            /* If the neighbor is in the heap, we touched its gain
             * so make sure the heap property is satisfied. */
            else
            {
                Int v = neighbor;
                heapifyUp(graph, bhHeap[neighborPartition], gains, v, position,
                          neighborGain);
                v = bhHeap[neighborPartition][position];
                heapifyDown(graph, bhHeap[neighborPartition],
                            bhSize[neighborPartition], gains, v, position,
                            gains[v]);
            }
        }
        /* Else the neighbor wasn't in the heap so add it. */
        else
        {
            if (!graph->isMarked(neighbor))
            {
                ASSERT(!graph->BH_inBoundary(neighbor));
                bhInsert(graph, neighbor);
            }
        }
    }

    externalDegree[vertex] = exD;
}

//-----------------------------------------------------------------------------
// This function computes the gain of a vertex
//-----------------------------------------------------------------------------
void calculateGain(EdgeCutProblem *graph, const EdgeCut_Options *options, Int vertex,
                   double *out_gain, Int *out_externalDegree)
{
    (void)options; // Unused variable

    Int *Gp         = graph->p;
    Int *Gi         = graph->i;
    double *Gx      = graph->x;
    bool *partition = graph->partition;

    bool vp = partition[vertex];

    double gain        = 0.0;
    Int externalDegree = 0;
    for (Int p = Gp[vertex]; p < Gp[vertex + 1]; p++)
    {
        double ew     = (Gx ? Gx[p] : 1.0);
        bool sameSide = (partition[Gi[p]] == vp);
        gain += (sameSide ? -ew : ew);

        if (!sameSide)
            externalDegree++;
    }

    /* Save outputs */
    *out_gain           = gain;
    *out_externalDegree = externalDegree;
}

} // end namespace Mongoose