assume NTL has been compiled in ISO mode
[barvinok.git] / hilbert.c
blob3dc355f4d3bca01a1fee62ffa7bd55ca56db6b55
1 #include <assert.h>
2 #include <stdlib.h>
3 #define Vector ZSolveVector
4 #define Matrix ZSolveMatrix
5 #include "zsolve/libzsolve.h"
6 #undef Vector
7 #undef Matrix
8 #include <barvinok/options.h>
9 #include <barvinok/util.h>
10 #include "hilbert.h"
11 #include "normalization.h"
12 #include "polysign.h"
13 #include "remove_equalities.h"
15 static ZSolveMatrix Matrix2zsolve(Matrix *M)
17 int i, j;
18 ZSolveMatrix zmatrix;
20 zmatrix = createMatrix(M->NbColumns-2, M->NbRows);
21 for (i = 0; i < M->NbRows; ++i)
22 for (j = 0; j < M->NbColumns-2; ++j) {
23 assert(mpz_cmp_si(M->p[i][1+j], -MAXINT) > 0);
24 assert(mpz_cmp_si(M->p[i][1+j], MAXINT) < 0);
25 zmatrix->Data[i*zmatrix->Width+j] = mpz_get_si(M->p[i][1+j]);
28 return zmatrix;
31 static Matrix *VectorArray2Matrix(VectorArray array, unsigned cols)
33 int i, j;
34 Matrix *M = Matrix_Alloc(array->Size, cols+1);
36 for (i = 0; i < array->Size; ++i) {
37 for (j = 0; j < cols; ++j)
38 value_set_si(M->p[i][j], array->Data[i][j]);
39 value_set_si(M->p[i][cols], 1);
41 return M;
44 static void Polyhedron_Remove_Positivity_Constraint(Polyhedron *P)
46 int i;
48 for (i = 0; i < P->NbConstraints; ++i) {
49 if (First_Non_Zero(P->Constraint[i]+1, P->Dimension) != -1)
50 continue;
51 if (i < P->NbConstraints-1)
52 Vector_Exchange(P->Constraint[i],
53 P->Constraint[P->NbConstraints-1],
54 P->Dimension+2);
55 P->NbConstraints--;
56 --i;
60 /* Return
61 * T 0
62 * 0 1
64 static Matrix *homogenize(Matrix *T)
66 int i;
67 Matrix *H = Matrix_Alloc(T->NbRows+1, T->NbColumns+1);
69 for (i = 0; i < T->NbRows; ++i)
70 Vector_Copy(T->p[i], H->p[i], T->NbColumns);
71 value_set_si(H->p[T->NbRows][T->NbColumns], 1);
72 return H;
75 static Matrix *Polyhedron2standard_form(Polyhedron *P, Matrix **T)
77 int i, j;
78 int rows;
79 unsigned dim = P->Dimension;
80 Matrix *M2;
81 Matrix *H, *U;
82 Matrix M;
84 assert(P->NbEq == 0);
85 Polyhedron_Remove_Positivity_Constraint(P);
86 for (i = 0; i < P->NbConstraints; ++i)
87 assert(value_zero_p(P->Constraint[i][1+dim]));
89 Polyhedron_Matrix_View(P, &M, P->NbConstraints);
90 H = standard_constraints(&M, 0, &rows, &U);
91 *T = homogenize(U);
92 Matrix_Free(U);
94 M2 = Matrix_Alloc(rows, 2+dim+rows);
96 for (i = dim; i < H->NbRows; ++i) {
97 Vector_Copy(H->p[i], M2->p[i-dim]+1, dim);
98 value_set_si(M2->p[i-dim][1+i], -1);
100 for (i = 0, j = H->NbRows-dim; i < dim; ++i) {
101 if (First_Non_Zero(H->p[i], i) == -1)
102 continue;
103 Vector_Oppose(H->p[i], M2->p[j]+1, dim);
104 value_set_si(M2->p[j][1+j+dim], 1);
105 ++j;
107 Matrix_Free(H);
108 return M2;
111 /* Assumes C is a linear cone (i.e. with apex zero).
112 * All equalities are removed first to speed up the computation
113 * in zsolve.
115 Matrix *Cone_Hilbert_Basis(Polyhedron *C, unsigned MaxRays)
117 unsigned dim;
118 int i;
119 Matrix *M2, *M3, *T;
120 Matrix *CV = NULL;
121 LinearSystem initialsystem;
122 ZSolveMatrix matrix;
123 ZSolveVector rhs;
124 ZSolveContext ctx;
126 remove_all_equalities(&C, NULL, NULL, &CV, 0, MaxRays);
127 dim = C->Dimension;
129 for (i = 0; i < C->NbConstraints; ++i)
130 assert(value_zero_p(C->Constraint[i][1+dim]) ||
131 First_Non_Zero(C->Constraint[i]+1, dim) == -1);
133 M2 = Polyhedron2standard_form(C, &T);
134 matrix = Matrix2zsolve(M2);
135 Matrix_Free(M2);
137 rhs = createVector(matrix->Height);
138 for (i = 0; i < matrix->Height; i++)
139 rhs[i] = 0;
141 initialsystem = createLinearSystem();
142 setLinearSystemMatrix(initialsystem, matrix);
143 deleteMatrix(matrix);
145 setLinearSystemRHS(initialsystem, rhs);
146 deleteVector(rhs);
148 setLinearSystemLimit(initialsystem, -1, 0, MAXINT, 0);
149 setLinearSystemEquationType(initialsystem, -1, EQUATION_EQUAL, 0);
151 ctx = createZSolveContextFromSystem(initialsystem, NULL, 0, 0, NULL, NULL);
152 zsolveSystem(ctx, 0);
154 M2 = VectorArray2Matrix(ctx->Homs, C->Dimension);
155 deleteZSolveContext(ctx, 1);
156 Matrix_Transposition(T);
157 M3 = Matrix_Alloc(M2->NbRows, M2->NbColumns);
158 Matrix_Product(M2, T, M3);
159 Matrix_Free(M2);
160 Matrix_Free(T);
162 if (CV) {
163 Matrix *T, *M;
164 T = Transpose(CV);
165 M = Matrix_Alloc(M3->NbRows, T->NbColumns);
166 Matrix_Product(M3, T, M);
167 Matrix_Free(M3);
168 Matrix_Free(CV);
169 Matrix_Free(T);
170 Polyhedron_Free(C);
171 M3 = M;
174 return M3;
177 /* Assumes no two arrays of values are the same, so we can just
178 * look for the first different elements, without knowing the
179 * length of the arrays.
181 static int lex_cmp(const void *va, const void *vb)
183 int i;
184 const Value *a = *(const Value **)va;
185 const Value *b = *(const Value **)vb;
187 for (i = 0; ; ++i) {
188 int sign = mpz_cmp(a[i], b[i]);
189 if (sign)
190 return sign;
194 /* Compute integer hull of truncated linear cone C, i.e., of C with
195 * the origin removed.
196 * Here, we do this by first computing the Hilbert basis of C
197 * and then discarding elements from this basis that are rational
198 * overconvex combinations of other elements in the basis.
200 Matrix *Cone_Hilbert_Integer_Hull(Polyhedron *C,
201 struct barvinok_options *options)
203 int i, j, k;
204 Matrix *hilbert = Cone_Hilbert_Basis(C, options->MaxRays);
205 Matrix *rays, *hull;
206 unsigned dim = C->Dimension;
207 Value tmp;
208 unsigned MaxRays = options->MaxRays;
210 /* When checking for redundant points below, we want to
211 * check if there are any _rational_ solutions.
213 POL_UNSET(options->MaxRays, POL_INTEGER);
215 POL_ENSURE_VERTICES(C);
216 rays = Matrix_Alloc(C->NbRays-1, C->Dimension);
217 for (i = 0, j = 0; i < C->NbRays; ++i) {
218 if (value_notzero_p(C->Ray[i][1+C->Dimension]))
219 continue;
220 Vector_Copy(C->Ray[i]+1, rays->p[j++], C->Dimension);
223 /* We only sort the pointers into the big Value array */
224 qsort(rays->p, rays->NbRows, sizeof(Value *), lex_cmp);
225 qsort(hilbert->p, hilbert->NbRows, sizeof(Value *), lex_cmp);
227 /* Remove rays from Hilbert basis */
228 for (i = 0, j = 0, k = 0; i < hilbert->NbRows && j < rays->NbRows; ++i) {
229 if (Vector_Equal(hilbert->p[i], rays->p[j], C->Dimension))
230 ++j;
231 else
232 hilbert->p[k++] = hilbert->p[i];
234 hilbert->NbRows = k;
236 /* Now remove points that are overconvex combinations of other points */
237 value_init(tmp);
238 for (i = 0; hilbert->NbRows > 1 && i < hilbert->NbRows; ++i) {
239 Matrix *LP;
240 Vector *obj;
241 int nray = rays->NbRows;
242 int npoint = hilbert->NbRows;
243 enum lp_result result;
245 LP = Matrix_Alloc(dim + 1 + nray + (npoint-1), 2 + nray + (npoint-1));
246 for (j = 0; j < dim; ++j) {
247 for (k = 0; k < nray; ++k)
248 value_assign(LP->p[j][k+1], rays->p[k][j]);
249 for (k = 0; k < npoint; ++k) {
250 if (k == i)
251 value_oppose(LP->p[j][1+nray+npoint-1], hilbert->p[k][j]);
252 else
253 value_assign(LP->p[j][1+nray+k-(k>i)], hilbert->p[k][j]);
256 value_set_si(LP->p[dim][0], 1);
257 for (k = 0; k < nray+npoint-1; ++k)
258 value_set_si(LP->p[dim][1+k], 1);
259 value_set_si(LP->p[dim][LP->NbColumns-1], -1);
260 for (k = 0; k < LP->NbColumns-2; ++k) {
261 value_set_si(LP->p[dim+1+k][0], 1);
262 value_set_si(LP->p[dim+1+k][1+k], 1);
265 /* Somewhat arbitrary objective function. */
266 obj = Vector_Alloc(LP->NbColumns-1);
267 value_set_si(obj->p[0], 1);
268 value_set_si(obj->p[obj->Size-1], 1);
270 result = constraints_opt(LP, obj->p, obj->p[0], lp_min, &tmp,
271 options);
273 /* If the LP is not empty, the point can be discarded */
274 if (result != lp_empty) {
275 hilbert->NbRows--;
276 if (i < hilbert->NbRows)
277 hilbert->p[i] = hilbert->p[hilbert->NbRows];
278 --i;
281 Matrix_Free(LP);
282 Vector_Free(obj);
284 value_clear(tmp);
286 hull = Matrix_Alloc(rays->NbRows + hilbert->NbRows, dim+1);
287 for (i = 0; i < rays->NbRows; ++i) {
288 Vector_Copy(rays->p[i], hull->p[i], dim);
289 value_set_si(hull->p[i][dim], 1);
291 for (i = 0; i < hilbert->NbRows; ++i) {
292 Vector_Copy(hilbert->p[i], hull->p[rays->NbRows+i], dim);
293 value_set_si(hull->p[rays->NbRows+i][dim], 1);
295 Matrix_Free(rays);
296 Matrix_Free(hilbert);
298 options->MaxRays = MaxRays;
299 return hull;