- * Create the linear programming problem containing the constraints defined by
- * two half-hulls. The objective function and optimization directions are not
- * written.
- *
- * Args:
- * syncState: container for synchronization data
- * i: first trace number
- * j: second trace number, garanteed to be larger than i
- * Returns:
- * A new glp_prob*, this problem must be freed by the caller with
- * glp_delete_prob()
- */
-static glp_prob* lpCreateProblem(GQueue* const lowerHull, GQueue* const
- upperHull)
-{
- unsigned int it;
- const int zero= 0;
- const double zeroD= 0.;
- glp_prob* lp= glp_create_prob();
- unsigned int hullPointNb= g_queue_get_length(lowerHull) +
- g_queue_get_length(upperHull);
- GArray* iArray= g_array_sized_new(FALSE, FALSE, sizeof(int), hullPointNb +
- 1);
- GArray* jArray= g_array_sized_new(FALSE, FALSE, sizeof(int), hullPointNb +
- 1);
- GArray* aArray= g_array_sized_new(FALSE, FALSE, sizeof(double),
- hullPointNb + 1);
- struct {
- GQueue* hull;
- struct LPAddRowInfo rowInfo;
- } loopValues[2]= {
- {lowerHull, {lp, GLP_UP, iArray, jArray, aArray}},
- {upperHull, {lp, GLP_LO, iArray, jArray, aArray}},
- };
-
- // Create the LP problem
- glp_term_out(GLP_OFF);
- if (hullPointNb > 0)
- {
- glp_add_rows(lp, hullPointNb);
- }
- glp_add_cols(lp, 2);
-
- glp_set_col_name(lp, 1, "a0");
- glp_set_col_bnds(lp, 1, GLP_FR, 0., 0.);
- glp_set_col_name(lp, 2, "a1");
- glp_set_col_bnds(lp, 2, GLP_LO, 0., 0.);
-
- // Add row constraints
- g_array_append_val(iArray, zero);
- g_array_append_val(jArray, zero);
- g_array_append_val(aArray, zeroD);
-
- for (it= 0; it < sizeof(loopValues) / sizeof(*loopValues); it++)
- {
- g_queue_foreach(loopValues[it].hull, &gfLPAddRow,
- &loopValues[it].rowInfo);
- }
-
- g_assert_cmpuint(iArray->len, ==, jArray->len);
- g_assert_cmpuint(jArray->len, ==, aArray->len);
- g_assert_cmpuint(aArray->len - 1, ==, hullPointNb * 2);
-
- glp_load_matrix(lp, aArray->len - 1, &g_array_index(iArray, int, 0),
- &g_array_index(jArray, int, 0), &g_array_index(aArray, double, 0));
-
- glp_scale_prob(lp, GLP_SF_AUTO);
-
- g_array_free(iArray, TRUE);
- g_array_free(jArray, TRUE);
- g_array_free(aArray, TRUE);
-
- return lp;
-}
-
-
-/*
- * A GFunc for g_queue_foreach(). Add constraints and bounds for one row.
- *
- * Args:
- * data Point*, synchronization point for which to add an LP row
- * (a constraint)
- * user_data LPAddRowInfo*
- */
-static void gfLPAddRow(gpointer data, gpointer user_data)
-{
- Point* p= data;
- struct LPAddRowInfo* rowInfo= user_data;
- int indexes[2];
- double constraints[2];
-
- indexes[0]= g_array_index(rowInfo->iArray, int, rowInfo->iArray->len - 1) + 1;
- indexes[1]= indexes[0];
-
- if (rowInfo->boundType == GLP_UP)
- {
- glp_set_row_bnds(rowInfo->lp, indexes[0], GLP_UP, 0., p->y);
- }
- else if (rowInfo->boundType == GLP_LO)
- {
- glp_set_row_bnds(rowInfo->lp, indexes[0], GLP_LO, p->y, 0.);
- }
- else
- {
- g_assert_not_reached();
- }
-
- g_array_append_vals(rowInfo->iArray, indexes, 2);
- indexes[0]= 1;
- indexes[1]= 2;
- g_array_append_vals(rowInfo->jArray, indexes, 2);
- constraints[0]= 1.;
- constraints[1]= p->x;
- g_array_append_vals(rowInfo->aArray, constraints, 2);
-}
-
-
-/*
- * Calculate min or max correction factors (as possible) using an LP problem.
- *
- * Args:
- * lp: A linear programming problem with constraints and bounds
- * initialized.
- * direction: The type of factors desired. Use GLP_MAX for max
- * approximation factors (a1, the drift or slope is the
- * largest) and GLP_MIN in the other case.
- *
- * Returns:
- * If the calculation was successful, a new Factors struct. Otherwise, NULL.
- * The calculation will fail if the hull assumptions are not respected.
- */
-static Factors* calculateFactors(glp_prob* const lp, const int direction)
-{
- int retval, status;
- Factors* factors;
-
- glp_set_obj_coef(lp, 1, 0.);
- glp_set_obj_coef(lp, 2, 1.);
-
- glp_set_obj_dir(lp, direction);
- retval= glp_simplex(lp, NULL);
- status= glp_get_status(lp);
-
- if (retval == 0 && status == GLP_OPT)
- {
- factors= malloc(sizeof(Factors));
- factors->offset= glp_get_col_prim(lp, 1);
- factors->drift= glp_get_col_prim(lp, 2);
- }
- else
- {
- factors= NULL;
- }
-
- return factors;
-}
-
-
-/*
- * Calculate min, max and approx correction factors (as possible) using an LP
- * problem.
- *
- * Args:
- * lp: A linear programming problem with constraints and bounds
- * initialized.
- *
- * Returns:
- * Please note that the approximation type may be MIDDLE, INCOMPLETE or
- * ABSENT. Unlike in analysis_chull, ABSENT is also used when the hulls do
- * not respect assumptions.
- */
-static void calculateCompleteFactors(glp_prob* const lp, FactorsCHull* factors)
-{
- factors->min= calculateFactors(lp, GLP_MIN);
- factors->max= calculateFactors(lp, GLP_MAX);
-
- if (factors->min && factors->max)
- {
- factors->type= MIDDLE;
- calculateFactorsMiddle(factors);
- }
- else if (factors->min || factors->max)
- {
- factors->type= INCOMPLETE;
- factors->approx= NULL;
- }
- else
- {
- factors->type= ABSENT;
- factors->approx= NULL;
- }
-}
-
-
-/*
- * Create and initialize an array like AnalysisStatsCHull.allFactors
- *
- * Args:
- * traceNb: number of traces
- *
- * Returns:
- * A new array, which can be freed with freeAllFactors()
- */
-static FactorsCHull** createAllFactors(const unsigned int traceNb)
-{
- FactorsCHull** factorsArray;
- unsigned int i;
-
- factorsArray= malloc(traceNb * sizeof(FactorsCHull*));
- for (i= 0; i < traceNb; i++)
- {
- factorsArray[i]= calloc((i + 1), sizeof(FactorsCHull));
-
- factorsArray[i][i].type= EXACT;
- factorsArray[i][i].approx= malloc(sizeof(Factors));
- factorsArray[i][i].approx->drift= 1.;
- factorsArray[i][i].approx->offset= 0.;
- }
-
- return factorsArray;
-}
-#endif
-
-
-/*
- * Compute synchronization factors using a linear programming approach.
- * Compute the factors using analysis_chull. Compare the two.
- *
- * When the solver library, glpk, is not available at build time, only compute
- * the factors using analysis_chull. This is to make sure that module runs its
- * finalize function so that its graph functions can be called later.
- *
- * Args:
- * syncState: container for synchronization data
- */
-static void finalizeAnalysisEvalLP(SyncState* const syncState)
-{
- AnalysisDataEval* analysisData= syncState->analysisData;
-#ifdef HAVE_LIBGLPK
- unsigned int i, j;
- AnalysisDataCHull* chAnalysisData= analysisData->chullSS->analysisData;
- FactorsCHull** lpFactorsArray;
-
- if (!syncState->stats && !syncState->graphsStream)
- {
- return;
- }
-
- /* Because of matching_distributor, this analysis may be called twice.
- * Only run it once */
- if ((syncState->graphsStream && analysisData->graphs->lps != NULL) ||
- (syncState->stats && analysisData->stats->chFactorsArray != NULL))
- {
- return;
- }
-
- lpFactorsArray= createAllFactors(syncState->traceNb);
-
- if (syncState->stats)
- {
- analysisData->stats->chFactorsArray=
- calculateAllFactors(analysisData->chullSS);
- analysisData->stats->lpFactorsArray= lpFactorsArray;
- }
-
- if (syncState->graphsStream)
- {
- analysisData->graphs->lps= malloc(syncState->traceNb *
- sizeof(glp_prob**));
- for (i= 0; i < syncState->traceNb; i++)
- {
- analysisData->graphs->lps[i]= malloc(i * sizeof(glp_prob*));
- }
- analysisData->graphs->lpFactorsArray= lpFactorsArray;
- }
-
- for (i= 0; i < syncState->traceNb; i++)
- {
- for (j= 0; j < i; j++)
- {
- glp_prob* lp;
-
- // Create the LP problem
- lp= lpCreateProblem(chAnalysisData->hullArray[i][j],
- chAnalysisData->hullArray[j][i]);
-
- // Use the LP problem to find the correction factors for this pair of
- // traces
- calculateCompleteFactors(lp, &lpFactorsArray[i][j]);
-
- if (syncState->graphsStream)
- {
- analysisData->graphs->lps[i][j]= lp;
- }
- else
- {
- glp_delete_prob(lp);
- }
- }
- }
-#endif
-
- g_array_free(analysisData->chullSS->analysisModule->finalizeAnalysis(analysisData->chullSS),
- TRUE);
-}
-
-
-/*
- * Compute synchronization accuracy information using a linear programming
- * approach. Write the neccessary data files and plot lines in the gnuplot
- * script.
- *
- * When the solver library, glpk, is not available at build time nothing is
- * actually produced.