2 * Copyright 2008-2009 Katholieke Universiteit Leuven
3 * Copyright 2013 Ecole Normale Superieure
5 * Use of this software is governed by the MIT license
7 * Written by Sven Verdoolaege, K.U.Leuven, Departement
8 * Computerwetenschappen, Celestijnenlaan 200A, B-3001 Leuven, Belgium
9 * and Ecole Normale Superieure, 45 rue d'Ulm, 75230 Paris, France
12 #include <isl_ctx_private.h>
13 #include <isl_mat_private.h>
14 #include "isl_map_private.h"
17 #include <isl_config.h>
20 * The implementation of tableaus in this file was inspired by Section 8
21 * of David Detlefs, Greg Nelson and James B. Saxe, "Simplify: a theorem
22 * prover for program checking".
25 struct isl_tab *isl_tab_alloc(struct isl_ctx *ctx,
26 unsigned n_row, unsigned n_var, unsigned M)
32 tab = isl_calloc_type(ctx, struct isl_tab);
35 tab->mat = isl_mat_alloc(ctx, n_row, off + n_var);
38 tab->var = isl_alloc_array(ctx, struct isl_tab_var, n_var);
41 tab->con = isl_alloc_array(ctx, struct isl_tab_var, n_row);
44 tab->col_var = isl_alloc_array(ctx, int, n_var);
47 tab->row_var = isl_alloc_array(ctx, int, n_row);
50 for (i = 0; i < n_var; ++i) {
51 tab->var[i].index = i;
52 tab->var[i].is_row = 0;
53 tab->var[i].is_nonneg = 0;
54 tab->var[i].is_zero = 0;
55 tab->var[i].is_redundant = 0;
56 tab->var[i].frozen = 0;
57 tab->var[i].negated = 0;
71 tab->strict_redundant = 0;
78 tab->bottom.type = isl_tab_undo_bottom;
79 tab->bottom.next = NULL;
80 tab->top = &tab->bottom;
92 isl_ctx *isl_tab_get_ctx(struct isl_tab *tab)
94 return tab ? isl_mat_get_ctx(tab->mat) : NULL;
97 int isl_tab_extend_cons(struct isl_tab *tab, unsigned n_new)
106 if (tab->max_con < tab->n_con + n_new) {
107 struct isl_tab_var *con;
109 con = isl_realloc_array(tab->mat->ctx, tab->con,
110 struct isl_tab_var, tab->max_con + n_new);
114 tab->max_con += n_new;
116 if (tab->mat->n_row < tab->n_row + n_new) {
119 tab->mat = isl_mat_extend(tab->mat,
120 tab->n_row + n_new, off + tab->n_col);
123 row_var = isl_realloc_array(tab->mat->ctx, tab->row_var,
124 int, tab->mat->n_row);
127 tab->row_var = row_var;
129 enum isl_tab_row_sign *s;
130 s = isl_realloc_array(tab->mat->ctx, tab->row_sign,
131 enum isl_tab_row_sign, tab->mat->n_row);
140 /* Make room for at least n_new extra variables.
141 * Return -1 if anything went wrong.
143 int isl_tab_extend_vars(struct isl_tab *tab, unsigned n_new)
145 struct isl_tab_var *var;
146 unsigned off = 2 + tab->M;
148 if (tab->max_var < tab->n_var + n_new) {
149 var = isl_realloc_array(tab->mat->ctx, tab->var,
150 struct isl_tab_var, tab->n_var + n_new);
154 tab->max_var += n_new;
157 if (tab->mat->n_col < off + tab->n_col + n_new) {
160 tab->mat = isl_mat_extend(tab->mat,
161 tab->mat->n_row, off + tab->n_col + n_new);
164 p = isl_realloc_array(tab->mat->ctx, tab->col_var,
165 int, tab->n_col + n_new);
174 struct isl_tab *isl_tab_extend(struct isl_tab *tab, unsigned n_new)
176 if (isl_tab_extend_cons(tab, n_new) >= 0)
183 static void free_undo_record(struct isl_tab_undo *undo)
185 switch (undo->type) {
186 case isl_tab_undo_saved_basis:
187 free(undo->u.col_var);
194 static void free_undo(struct isl_tab *tab)
196 struct isl_tab_undo *undo, *next;
198 for (undo = tab->top; undo && undo != &tab->bottom; undo = next) {
200 free_undo_record(undo);
205 void isl_tab_free(struct isl_tab *tab)
210 isl_mat_free(tab->mat);
211 isl_vec_free(tab->dual);
212 isl_basic_map_free(tab->bmap);
218 isl_mat_free(tab->samples);
219 free(tab->sample_index);
220 isl_mat_free(tab->basis);
224 struct isl_tab *isl_tab_dup(struct isl_tab *tab)
234 dup = isl_calloc_type(tab->mat->ctx, struct isl_tab);
237 dup->mat = isl_mat_dup(tab->mat);
240 dup->var = isl_alloc_array(tab->mat->ctx, struct isl_tab_var, tab->max_var);
243 for (i = 0; i < tab->n_var; ++i)
244 dup->var[i] = tab->var[i];
245 dup->con = isl_alloc_array(tab->mat->ctx, struct isl_tab_var, tab->max_con);
248 for (i = 0; i < tab->n_con; ++i)
249 dup->con[i] = tab->con[i];
250 dup->col_var = isl_alloc_array(tab->mat->ctx, int, tab->mat->n_col - off);
253 for (i = 0; i < tab->n_col; ++i)
254 dup->col_var[i] = tab->col_var[i];
255 dup->row_var = isl_alloc_array(tab->mat->ctx, int, tab->mat->n_row);
258 for (i = 0; i < tab->n_row; ++i)
259 dup->row_var[i] = tab->row_var[i];
261 dup->row_sign = isl_alloc_array(tab->mat->ctx, enum isl_tab_row_sign,
265 for (i = 0; i < tab->n_row; ++i)
266 dup->row_sign[i] = tab->row_sign[i];
269 dup->samples = isl_mat_dup(tab->samples);
272 dup->sample_index = isl_alloc_array(tab->mat->ctx, int,
273 tab->samples->n_row);
274 if (!dup->sample_index)
276 dup->n_sample = tab->n_sample;
277 dup->n_outside = tab->n_outside;
279 dup->n_row = tab->n_row;
280 dup->n_con = tab->n_con;
281 dup->n_eq = tab->n_eq;
282 dup->max_con = tab->max_con;
283 dup->n_col = tab->n_col;
284 dup->n_var = tab->n_var;
285 dup->max_var = tab->max_var;
286 dup->n_param = tab->n_param;
287 dup->n_div = tab->n_div;
288 dup->n_dead = tab->n_dead;
289 dup->n_redundant = tab->n_redundant;
290 dup->rational = tab->rational;
291 dup->empty = tab->empty;
292 dup->strict_redundant = 0;
296 tab->cone = tab->cone;
297 dup->bottom.type = isl_tab_undo_bottom;
298 dup->bottom.next = NULL;
299 dup->top = &dup->bottom;
301 dup->n_zero = tab->n_zero;
302 dup->n_unbounded = tab->n_unbounded;
303 dup->basis = isl_mat_dup(tab->basis);
311 /* Construct the coefficient matrix of the product tableau
313 * mat{1,2} is the coefficient matrix of tableau {1,2}
314 * row{1,2} is the number of rows in tableau {1,2}
315 * col{1,2} is the number of columns in tableau {1,2}
316 * off is the offset to the coefficient column (skipping the
317 * denominator, the constant term and the big parameter if any)
318 * r{1,2} is the number of redundant rows in tableau {1,2}
319 * d{1,2} is the number of dead columns in tableau {1,2}
321 * The order of the rows and columns in the result is as explained
322 * in isl_tab_product.
324 static struct isl_mat *tab_mat_product(struct isl_mat *mat1,
325 struct isl_mat *mat2, unsigned row1, unsigned row2,
326 unsigned col1, unsigned col2,
327 unsigned off, unsigned r1, unsigned r2, unsigned d1, unsigned d2)
330 struct isl_mat *prod;
333 prod = isl_mat_alloc(mat1->ctx, mat1->n_row + mat2->n_row,
339 for (i = 0; i < r1; ++i) {
340 isl_seq_cpy(prod->row[n + i], mat1->row[i], off + d1);
341 isl_seq_clr(prod->row[n + i] + off + d1, d2);
342 isl_seq_cpy(prod->row[n + i] + off + d1 + d2,
343 mat1->row[i] + off + d1, col1 - d1);
344 isl_seq_clr(prod->row[n + i] + off + col1 + d1, col2 - d2);
348 for (i = 0; i < r2; ++i) {
349 isl_seq_cpy(prod->row[n + i], mat2->row[i], off);
350 isl_seq_clr(prod->row[n + i] + off, d1);
351 isl_seq_cpy(prod->row[n + i] + off + d1,
352 mat2->row[i] + off, d2);
353 isl_seq_clr(prod->row[n + i] + off + d1 + d2, col1 - d1);
354 isl_seq_cpy(prod->row[n + i] + off + col1 + d1,
355 mat2->row[i] + off + d2, col2 - d2);
359 for (i = 0; i < row1 - r1; ++i) {
360 isl_seq_cpy(prod->row[n + i], mat1->row[r1 + i], off + d1);
361 isl_seq_clr(prod->row[n + i] + off + d1, d2);
362 isl_seq_cpy(prod->row[n + i] + off + d1 + d2,
363 mat1->row[r1 + i] + off + d1, col1 - d1);
364 isl_seq_clr(prod->row[n + i] + off + col1 + d1, col2 - d2);
368 for (i = 0; i < row2 - r2; ++i) {
369 isl_seq_cpy(prod->row[n + i], mat2->row[r2 + i], off);
370 isl_seq_clr(prod->row[n + i] + off, d1);
371 isl_seq_cpy(prod->row[n + i] + off + d1,
372 mat2->row[r2 + i] + off, d2);
373 isl_seq_clr(prod->row[n + i] + off + d1 + d2, col1 - d1);
374 isl_seq_cpy(prod->row[n + i] + off + col1 + d1,
375 mat2->row[r2 + i] + off + d2, col2 - d2);
381 /* Update the row or column index of a variable that corresponds
382 * to a variable in the first input tableau.
384 static void update_index1(struct isl_tab_var *var,
385 unsigned r1, unsigned r2, unsigned d1, unsigned d2)
387 if (var->index == -1)
389 if (var->is_row && var->index >= r1)
391 if (!var->is_row && var->index >= d1)
395 /* Update the row or column index of a variable that corresponds
396 * to a variable in the second input tableau.
398 static void update_index2(struct isl_tab_var *var,
399 unsigned row1, unsigned col1,
400 unsigned r1, unsigned r2, unsigned d1, unsigned d2)
402 if (var->index == -1)
417 /* Create a tableau that represents the Cartesian product of the sets
418 * represented by tableaus tab1 and tab2.
419 * The order of the rows in the product is
420 * - redundant rows of tab1
421 * - redundant rows of tab2
422 * - non-redundant rows of tab1
423 * - non-redundant rows of tab2
424 * The order of the columns is
427 * - coefficient of big parameter, if any
428 * - dead columns of tab1
429 * - dead columns of tab2
430 * - live columns of tab1
431 * - live columns of tab2
432 * The order of the variables and the constraints is a concatenation
433 * of order in the two input tableaus.
435 struct isl_tab *isl_tab_product(struct isl_tab *tab1, struct isl_tab *tab2)
438 struct isl_tab *prod;
440 unsigned r1, r2, d1, d2;
445 isl_assert(tab1->mat->ctx, tab1->M == tab2->M, return NULL);
446 isl_assert(tab1->mat->ctx, tab1->rational == tab2->rational, return NULL);
447 isl_assert(tab1->mat->ctx, tab1->cone == tab2->cone, return NULL);
448 isl_assert(tab1->mat->ctx, !tab1->row_sign, return NULL);
449 isl_assert(tab1->mat->ctx, !tab2->row_sign, return NULL);
450 isl_assert(tab1->mat->ctx, tab1->n_param == 0, return NULL);
451 isl_assert(tab1->mat->ctx, tab2->n_param == 0, return NULL);
452 isl_assert(tab1->mat->ctx, tab1->n_div == 0, return NULL);
453 isl_assert(tab1->mat->ctx, tab2->n_div == 0, return NULL);
456 r1 = tab1->n_redundant;
457 r2 = tab2->n_redundant;
460 prod = isl_calloc_type(tab1->mat->ctx, struct isl_tab);
463 prod->mat = tab_mat_product(tab1->mat, tab2->mat,
464 tab1->n_row, tab2->n_row,
465 tab1->n_col, tab2->n_col, off, r1, r2, d1, d2);
468 prod->var = isl_alloc_array(tab1->mat->ctx, struct isl_tab_var,
469 tab1->max_var + tab2->max_var);
472 for (i = 0; i < tab1->n_var; ++i) {
473 prod->var[i] = tab1->var[i];
474 update_index1(&prod->var[i], r1, r2, d1, d2);
476 for (i = 0; i < tab2->n_var; ++i) {
477 prod->var[tab1->n_var + i] = tab2->var[i];
478 update_index2(&prod->var[tab1->n_var + i],
479 tab1->n_row, tab1->n_col,
482 prod->con = isl_alloc_array(tab1->mat->ctx, struct isl_tab_var,
483 tab1->max_con + tab2->max_con);
486 for (i = 0; i < tab1->n_con; ++i) {
487 prod->con[i] = tab1->con[i];
488 update_index1(&prod->con[i], r1, r2, d1, d2);
490 for (i = 0; i < tab2->n_con; ++i) {
491 prod->con[tab1->n_con + i] = tab2->con[i];
492 update_index2(&prod->con[tab1->n_con + i],
493 tab1->n_row, tab1->n_col,
496 prod->col_var = isl_alloc_array(tab1->mat->ctx, int,
497 tab1->n_col + tab2->n_col);
500 for (i = 0; i < tab1->n_col; ++i) {
501 int pos = i < d1 ? i : i + d2;
502 prod->col_var[pos] = tab1->col_var[i];
504 for (i = 0; i < tab2->n_col; ++i) {
505 int pos = i < d2 ? d1 + i : tab1->n_col + i;
506 int t = tab2->col_var[i];
511 prod->col_var[pos] = t;
513 prod->row_var = isl_alloc_array(tab1->mat->ctx, int,
514 tab1->mat->n_row + tab2->mat->n_row);
517 for (i = 0; i < tab1->n_row; ++i) {
518 int pos = i < r1 ? i : i + r2;
519 prod->row_var[pos] = tab1->row_var[i];
521 for (i = 0; i < tab2->n_row; ++i) {
522 int pos = i < r2 ? r1 + i : tab1->n_row + i;
523 int t = tab2->row_var[i];
528 prod->row_var[pos] = t;
530 prod->samples = NULL;
531 prod->sample_index = NULL;
532 prod->n_row = tab1->n_row + tab2->n_row;
533 prod->n_con = tab1->n_con + tab2->n_con;
535 prod->max_con = tab1->max_con + tab2->max_con;
536 prod->n_col = tab1->n_col + tab2->n_col;
537 prod->n_var = tab1->n_var + tab2->n_var;
538 prod->max_var = tab1->max_var + tab2->max_var;
541 prod->n_dead = tab1->n_dead + tab2->n_dead;
542 prod->n_redundant = tab1->n_redundant + tab2->n_redundant;
543 prod->rational = tab1->rational;
544 prod->empty = tab1->empty || tab2->empty;
545 prod->strict_redundant = tab1->strict_redundant || tab2->strict_redundant;
549 prod->cone = tab1->cone;
550 prod->bottom.type = isl_tab_undo_bottom;
551 prod->bottom.next = NULL;
552 prod->top = &prod->bottom;
555 prod->n_unbounded = 0;
564 static struct isl_tab_var *var_from_index(struct isl_tab *tab, int i)
569 return &tab->con[~i];
572 struct isl_tab_var *isl_tab_var_from_row(struct isl_tab *tab, int i)
574 return var_from_index(tab, tab->row_var[i]);
577 static struct isl_tab_var *var_from_col(struct isl_tab *tab, int i)
579 return var_from_index(tab, tab->col_var[i]);
582 /* Check if there are any upper bounds on column variable "var",
583 * i.e., non-negative rows where var appears with a negative coefficient.
584 * Return 1 if there are no such bounds.
586 static int max_is_manifestly_unbounded(struct isl_tab *tab,
587 struct isl_tab_var *var)
590 unsigned off = 2 + tab->M;
594 for (i = tab->n_redundant; i < tab->n_row; ++i) {
595 if (!isl_int_is_neg(tab->mat->row[i][off + var->index]))
597 if (isl_tab_var_from_row(tab, i)->is_nonneg)
603 /* Check if there are any lower bounds on column variable "var",
604 * i.e., non-negative rows where var appears with a positive coefficient.
605 * Return 1 if there are no such bounds.
607 static int min_is_manifestly_unbounded(struct isl_tab *tab,
608 struct isl_tab_var *var)
611 unsigned off = 2 + tab->M;
615 for (i = tab->n_redundant; i < tab->n_row; ++i) {
616 if (!isl_int_is_pos(tab->mat->row[i][off + var->index]))
618 if (isl_tab_var_from_row(tab, i)->is_nonneg)
624 static int row_cmp(struct isl_tab *tab, int r1, int r2, int c, isl_int t)
626 unsigned off = 2 + tab->M;
630 isl_int_mul(t, tab->mat->row[r1][2], tab->mat->row[r2][off+c]);
631 isl_int_submul(t, tab->mat->row[r2][2], tab->mat->row[r1][off+c]);
636 isl_int_mul(t, tab->mat->row[r1][1], tab->mat->row[r2][off + c]);
637 isl_int_submul(t, tab->mat->row[r2][1], tab->mat->row[r1][off + c]);
638 return isl_int_sgn(t);
641 /* Given the index of a column "c", return the index of a row
642 * that can be used to pivot the column in, with either an increase
643 * (sgn > 0) or a decrease (sgn < 0) of the corresponding variable.
644 * If "var" is not NULL, then the row returned will be different from
645 * the one associated with "var".
647 * Each row in the tableau is of the form
649 * x_r = a_r0 + \sum_i a_ri x_i
651 * Only rows with x_r >= 0 and with the sign of a_ri opposite to "sgn"
652 * impose any limit on the increase or decrease in the value of x_c
653 * and this bound is equal to a_r0 / |a_rc|. We are therefore looking
654 * for the row with the smallest (most stringent) such bound.
655 * Note that the common denominator of each row drops out of the fraction.
656 * To check if row j has a smaller bound than row r, i.e.,
657 * a_j0 / |a_jc| < a_r0 / |a_rc| or a_j0 |a_rc| < a_r0 |a_jc|,
658 * we check if -sign(a_jc) (a_j0 a_rc - a_r0 a_jc) < 0,
659 * where -sign(a_jc) is equal to "sgn".
661 static int pivot_row(struct isl_tab *tab,
662 struct isl_tab_var *var, int sgn, int c)
666 unsigned off = 2 + tab->M;
670 for (j = tab->n_redundant; j < tab->n_row; ++j) {
671 if (var && j == var->index)
673 if (!isl_tab_var_from_row(tab, j)->is_nonneg)
675 if (sgn * isl_int_sgn(tab->mat->row[j][off + c]) >= 0)
681 tsgn = sgn * row_cmp(tab, r, j, c, t);
682 if (tsgn < 0 || (tsgn == 0 &&
683 tab->row_var[j] < tab->row_var[r]))
690 /* Find a pivot (row and col) that will increase (sgn > 0) or decrease
691 * (sgn < 0) the value of row variable var.
692 * If not NULL, then skip_var is a row variable that should be ignored
693 * while looking for a pivot row. It is usually equal to var.
695 * As the given row in the tableau is of the form
697 * x_r = a_r0 + \sum_i a_ri x_i
699 * we need to find a column such that the sign of a_ri is equal to "sgn"
700 * (such that an increase in x_i will have the desired effect) or a
701 * column with a variable that may attain negative values.
702 * If a_ri is positive, then we need to move x_i in the same direction
703 * to obtain the desired effect. Otherwise, x_i has to move in the
704 * opposite direction.
706 static void find_pivot(struct isl_tab *tab,
707 struct isl_tab_var *var, struct isl_tab_var *skip_var,
708 int sgn, int *row, int *col)
715 isl_assert(tab->mat->ctx, var->is_row, return);
716 tr = tab->mat->row[var->index] + 2 + tab->M;
719 for (j = tab->n_dead; j < tab->n_col; ++j) {
720 if (isl_int_is_zero(tr[j]))
722 if (isl_int_sgn(tr[j]) != sgn &&
723 var_from_col(tab, j)->is_nonneg)
725 if (c < 0 || tab->col_var[j] < tab->col_var[c])
731 sgn *= isl_int_sgn(tr[c]);
732 r = pivot_row(tab, skip_var, sgn, c);
733 *row = r < 0 ? var->index : r;
737 /* Return 1 if row "row" represents an obviously redundant inequality.
739 * - it represents an inequality or a variable
740 * - that is the sum of a non-negative sample value and a positive
741 * combination of zero or more non-negative constraints.
743 int isl_tab_row_is_redundant(struct isl_tab *tab, int row)
746 unsigned off = 2 + tab->M;
748 if (tab->row_var[row] < 0 && !isl_tab_var_from_row(tab, row)->is_nonneg)
751 if (isl_int_is_neg(tab->mat->row[row][1]))
753 if (tab->strict_redundant && isl_int_is_zero(tab->mat->row[row][1]))
755 if (tab->M && isl_int_is_neg(tab->mat->row[row][2]))
758 for (i = tab->n_dead; i < tab->n_col; ++i) {
759 if (isl_int_is_zero(tab->mat->row[row][off + i]))
761 if (tab->col_var[i] >= 0)
763 if (isl_int_is_neg(tab->mat->row[row][off + i]))
765 if (!var_from_col(tab, i)->is_nonneg)
771 static void swap_rows(struct isl_tab *tab, int row1, int row2)
774 enum isl_tab_row_sign s;
776 t = tab->row_var[row1];
777 tab->row_var[row1] = tab->row_var[row2];
778 tab->row_var[row2] = t;
779 isl_tab_var_from_row(tab, row1)->index = row1;
780 isl_tab_var_from_row(tab, row2)->index = row2;
781 tab->mat = isl_mat_swap_rows(tab->mat, row1, row2);
785 s = tab->row_sign[row1];
786 tab->row_sign[row1] = tab->row_sign[row2];
787 tab->row_sign[row2] = s;
790 static int push_union(struct isl_tab *tab,
791 enum isl_tab_undo_type type, union isl_tab_undo_val u) WARN_UNUSED;
792 static int push_union(struct isl_tab *tab,
793 enum isl_tab_undo_type type, union isl_tab_undo_val u)
795 struct isl_tab_undo *undo;
802 undo = isl_alloc_type(tab->mat->ctx, struct isl_tab_undo);
807 undo->next = tab->top;
813 int isl_tab_push_var(struct isl_tab *tab,
814 enum isl_tab_undo_type type, struct isl_tab_var *var)
816 union isl_tab_undo_val u;
818 u.var_index = tab->row_var[var->index];
820 u.var_index = tab->col_var[var->index];
821 return push_union(tab, type, u);
824 int isl_tab_push(struct isl_tab *tab, enum isl_tab_undo_type type)
826 union isl_tab_undo_val u = { 0 };
827 return push_union(tab, type, u);
830 /* Push a record on the undo stack describing the current basic
831 * variables, so that the this state can be restored during rollback.
833 int isl_tab_push_basis(struct isl_tab *tab)
836 union isl_tab_undo_val u;
838 u.col_var = isl_alloc_array(tab->mat->ctx, int, tab->n_col);
841 for (i = 0; i < tab->n_col; ++i)
842 u.col_var[i] = tab->col_var[i];
843 return push_union(tab, isl_tab_undo_saved_basis, u);
846 int isl_tab_push_callback(struct isl_tab *tab, struct isl_tab_callback *callback)
848 union isl_tab_undo_val u;
849 u.callback = callback;
850 return push_union(tab, isl_tab_undo_callback, u);
853 struct isl_tab *isl_tab_init_samples(struct isl_tab *tab)
860 tab->samples = isl_mat_alloc(tab->mat->ctx, 1, 1 + tab->n_var);
863 tab->sample_index = isl_alloc_array(tab->mat->ctx, int, 1);
864 if (!tab->sample_index)
872 struct isl_tab *isl_tab_add_sample(struct isl_tab *tab,
873 __isl_take isl_vec *sample)
878 if (tab->n_sample + 1 > tab->samples->n_row) {
879 int *t = isl_realloc_array(tab->mat->ctx,
880 tab->sample_index, int, tab->n_sample + 1);
883 tab->sample_index = t;
886 tab->samples = isl_mat_extend(tab->samples,
887 tab->n_sample + 1, tab->samples->n_col);
891 isl_seq_cpy(tab->samples->row[tab->n_sample], sample->el, sample->size);
892 isl_vec_free(sample);
893 tab->sample_index[tab->n_sample] = tab->n_sample;
898 isl_vec_free(sample);
903 struct isl_tab *isl_tab_drop_sample(struct isl_tab *tab, int s)
905 if (s != tab->n_outside) {
906 int t = tab->sample_index[tab->n_outside];
907 tab->sample_index[tab->n_outside] = tab->sample_index[s];
908 tab->sample_index[s] = t;
909 isl_mat_swap_rows(tab->samples, tab->n_outside, s);
912 if (isl_tab_push(tab, isl_tab_undo_drop_sample) < 0) {
920 /* Record the current number of samples so that we can remove newer
921 * samples during a rollback.
923 int isl_tab_save_samples(struct isl_tab *tab)
925 union isl_tab_undo_val u;
931 return push_union(tab, isl_tab_undo_saved_samples, u);
934 /* Mark row with index "row" as being redundant.
935 * If we may need to undo the operation or if the row represents
936 * a variable of the original problem, the row is kept,
937 * but no longer considered when looking for a pivot row.
938 * Otherwise, the row is simply removed.
940 * The row may be interchanged with some other row. If it
941 * is interchanged with a later row, return 1. Otherwise return 0.
942 * If the rows are checked in order in the calling function,
943 * then a return value of 1 means that the row with the given
944 * row number may now contain a different row that hasn't been checked yet.
946 int isl_tab_mark_redundant(struct isl_tab *tab, int row)
948 struct isl_tab_var *var = isl_tab_var_from_row(tab, row);
949 var->is_redundant = 1;
950 isl_assert(tab->mat->ctx, row >= tab->n_redundant, return -1);
951 if (tab->preserve || tab->need_undo || tab->row_var[row] >= 0) {
952 if (tab->row_var[row] >= 0 && !var->is_nonneg) {
954 if (isl_tab_push_var(tab, isl_tab_undo_nonneg, var) < 0)
957 if (row != tab->n_redundant)
958 swap_rows(tab, row, tab->n_redundant);
960 return isl_tab_push_var(tab, isl_tab_undo_redundant, var);
962 if (row != tab->n_row - 1)
963 swap_rows(tab, row, tab->n_row - 1);
964 isl_tab_var_from_row(tab, tab->n_row - 1)->index = -1;
970 int isl_tab_mark_empty(struct isl_tab *tab)
974 if (!tab->empty && tab->need_undo)
975 if (isl_tab_push(tab, isl_tab_undo_empty) < 0)
981 int isl_tab_freeze_constraint(struct isl_tab *tab, int con)
983 struct isl_tab_var *var;
988 var = &tab->con[con];
996 return isl_tab_push_var(tab, isl_tab_undo_freeze, var);
1001 /* Update the rows signs after a pivot of "row" and "col", with "row_sgn"
1002 * the original sign of the pivot element.
1003 * We only keep track of row signs during PILP solving and in this case
1004 * we only pivot a row with negative sign (meaning the value is always
1005 * non-positive) using a positive pivot element.
1007 * For each row j, the new value of the parametric constant is equal to
1009 * a_j0 - a_jc a_r0/a_rc
1011 * where a_j0 is the original parametric constant, a_rc is the pivot element,
1012 * a_r0 is the parametric constant of the pivot row and a_jc is the
1013 * pivot column entry of the row j.
1014 * Since a_r0 is non-positive and a_rc is positive, the sign of row j
1015 * remains the same if a_jc has the same sign as the row j or if
1016 * a_jc is zero. In all other cases, we reset the sign to "unknown".
1018 static void update_row_sign(struct isl_tab *tab, int row, int col, int row_sgn)
1021 struct isl_mat *mat = tab->mat;
1022 unsigned off = 2 + tab->M;
1027 if (tab->row_sign[row] == 0)
1029 isl_assert(mat->ctx, row_sgn > 0, return);
1030 isl_assert(mat->ctx, tab->row_sign[row] == isl_tab_row_neg, return);
1031 tab->row_sign[row] = isl_tab_row_pos;
1032 for (i = 0; i < tab->n_row; ++i) {
1036 s = isl_int_sgn(mat->row[i][off + col]);
1039 if (!tab->row_sign[i])
1041 if (s < 0 && tab->row_sign[i] == isl_tab_row_neg)
1043 if (s > 0 && tab->row_sign[i] == isl_tab_row_pos)
1045 tab->row_sign[i] = isl_tab_row_unknown;
1049 /* Given a row number "row" and a column number "col", pivot the tableau
1050 * such that the associated variables are interchanged.
1051 * The given row in the tableau expresses
1053 * x_r = a_r0 + \sum_i a_ri x_i
1057 * x_c = 1/a_rc x_r - a_r0/a_rc + sum_{i \ne r} -a_ri/a_rc
1059 * Substituting this equality into the other rows
1061 * x_j = a_j0 + \sum_i a_ji x_i
1063 * with a_jc \ne 0, we obtain
1065 * x_j = a_jc/a_rc x_r + a_j0 - a_jc a_r0/a_rc + sum a_ji - a_jc a_ri/a_rc
1072 * where i is any other column and j is any other row,
1073 * is therefore transformed into
1075 * s(n_rc)d_r/|n_rc| -s(n_rc)n_ri/|n_rc|
1076 * s(n_rc)d_r n_jc/(|n_rc| d_j) (n_ji |n_rc| - s(n_rc)n_jc n_ri)/(|n_rc| d_j)
1078 * The transformation is performed along the following steps
1080 * d_r/n_rc n_ri/n_rc
1083 * s(n_rc)d_r/|n_rc| -s(n_rc)n_ri/|n_rc|
1086 * s(n_rc)d_r/|n_rc| -s(n_rc)n_ri/|n_rc|
1087 * n_jc/(|n_rc| d_j) n_ji/(|n_rc| d_j)
1089 * s(n_rc)d_r/|n_rc| -s(n_rc)n_ri/|n_rc|
1090 * n_jc/(|n_rc| d_j) (n_ji |n_rc|)/(|n_rc| d_j)
1092 * s(n_rc)d_r/|n_rc| -s(n_rc)n_ri/|n_rc|
1093 * n_jc/(|n_rc| d_j) (n_ji |n_rc| - s(n_rc)n_jc n_ri)/(|n_rc| d_j)
1095 * s(n_rc)d_r/|n_rc| -s(n_rc)n_ri/|n_rc|
1096 * s(n_rc)d_r n_jc/(|n_rc| d_j) (n_ji |n_rc| - s(n_rc)n_jc n_ri)/(|n_rc| d_j)
1099 int isl_tab_pivot(struct isl_tab *tab, int row, int col)
1104 struct isl_mat *mat = tab->mat;
1105 struct isl_tab_var *var;
1106 unsigned off = 2 + tab->M;
1108 if (tab->mat->ctx->abort) {
1109 isl_ctx_set_error(tab->mat->ctx, isl_error_abort);
1113 isl_int_swap(mat->row[row][0], mat->row[row][off + col]);
1114 sgn = isl_int_sgn(mat->row[row][0]);
1116 isl_int_neg(mat->row[row][0], mat->row[row][0]);
1117 isl_int_neg(mat->row[row][off + col], mat->row[row][off + col]);
1119 for (j = 0; j < off - 1 + tab->n_col; ++j) {
1120 if (j == off - 1 + col)
1122 isl_int_neg(mat->row[row][1 + j], mat->row[row][1 + j]);
1124 if (!isl_int_is_one(mat->row[row][0]))
1125 isl_seq_normalize(mat->ctx, mat->row[row], off + tab->n_col);
1126 for (i = 0; i < tab->n_row; ++i) {
1129 if (isl_int_is_zero(mat->row[i][off + col]))
1131 isl_int_mul(mat->row[i][0], mat->row[i][0], mat->row[row][0]);
1132 for (j = 0; j < off - 1 + tab->n_col; ++j) {
1133 if (j == off - 1 + col)
1135 isl_int_mul(mat->row[i][1 + j],
1136 mat->row[i][1 + j], mat->row[row][0]);
1137 isl_int_addmul(mat->row[i][1 + j],
1138 mat->row[i][off + col], mat->row[row][1 + j]);
1140 isl_int_mul(mat->row[i][off + col],
1141 mat->row[i][off + col], mat->row[row][off + col]);
1142 if (!isl_int_is_one(mat->row[i][0]))
1143 isl_seq_normalize(mat->ctx, mat->row[i], off + tab->n_col);
1145 t = tab->row_var[row];
1146 tab->row_var[row] = tab->col_var[col];
1147 tab->col_var[col] = t;
1148 var = isl_tab_var_from_row(tab, row);
1151 var = var_from_col(tab, col);
1154 update_row_sign(tab, row, col, sgn);
1157 for (i = tab->n_redundant; i < tab->n_row; ++i) {
1158 if (isl_int_is_zero(mat->row[i][off + col]))
1160 if (!isl_tab_var_from_row(tab, i)->frozen &&
1161 isl_tab_row_is_redundant(tab, i)) {
1162 int redo = isl_tab_mark_redundant(tab, i);
1172 /* If "var" represents a column variable, then pivot is up (sgn > 0)
1173 * or down (sgn < 0) to a row. The variable is assumed not to be
1174 * unbounded in the specified direction.
1175 * If sgn = 0, then the variable is unbounded in both directions,
1176 * and we pivot with any row we can find.
1178 static int to_row(struct isl_tab *tab, struct isl_tab_var *var, int sign) WARN_UNUSED;
1179 static int to_row(struct isl_tab *tab, struct isl_tab_var *var, int sign)
1182 unsigned off = 2 + tab->M;
1188 for (r = tab->n_redundant; r < tab->n_row; ++r)
1189 if (!isl_int_is_zero(tab->mat->row[r][off+var->index]))
1191 isl_assert(tab->mat->ctx, r < tab->n_row, return -1);
1193 r = pivot_row(tab, NULL, sign, var->index);
1194 isl_assert(tab->mat->ctx, r >= 0, return -1);
1197 return isl_tab_pivot(tab, r, var->index);
1200 /* Check whether all variables that are marked as non-negative
1201 * also have a non-negative sample value. This function is not
1202 * called from the current code but is useful during debugging.
1204 static void check_table(struct isl_tab *tab) __attribute__ ((unused));
1205 static void check_table(struct isl_tab *tab)
1211 for (i = tab->n_redundant; i < tab->n_row; ++i) {
1212 struct isl_tab_var *var;
1213 var = isl_tab_var_from_row(tab, i);
1214 if (!var->is_nonneg)
1217 isl_assert(tab->mat->ctx,
1218 !isl_int_is_neg(tab->mat->row[i][2]), abort());
1219 if (isl_int_is_pos(tab->mat->row[i][2]))
1222 isl_assert(tab->mat->ctx, !isl_int_is_neg(tab->mat->row[i][1]),
1227 /* Return the sign of the maximal value of "var".
1228 * If the sign is not negative, then on return from this function,
1229 * the sample value will also be non-negative.
1231 * If "var" is manifestly unbounded wrt positive values, we are done.
1232 * Otherwise, we pivot the variable up to a row if needed
1233 * Then we continue pivoting down until either
1234 * - no more down pivots can be performed
1235 * - the sample value is positive
1236 * - the variable is pivoted into a manifestly unbounded column
1238 static int sign_of_max(struct isl_tab *tab, struct isl_tab_var *var)
1242 if (max_is_manifestly_unbounded(tab, var))
1244 if (to_row(tab, var, 1) < 0)
1246 while (!isl_int_is_pos(tab->mat->row[var->index][1])) {
1247 find_pivot(tab, var, var, 1, &row, &col);
1249 return isl_int_sgn(tab->mat->row[var->index][1]);
1250 if (isl_tab_pivot(tab, row, col) < 0)
1252 if (!var->is_row) /* manifestly unbounded */
1258 int isl_tab_sign_of_max(struct isl_tab *tab, int con)
1260 struct isl_tab_var *var;
1265 var = &tab->con[con];
1266 isl_assert(tab->mat->ctx, !var->is_redundant, return -2);
1267 isl_assert(tab->mat->ctx, !var->is_zero, return -2);
1269 return sign_of_max(tab, var);
1272 static int row_is_neg(struct isl_tab *tab, int row)
1275 return isl_int_is_neg(tab->mat->row[row][1]);
1276 if (isl_int_is_pos(tab->mat->row[row][2]))
1278 if (isl_int_is_neg(tab->mat->row[row][2]))
1280 return isl_int_is_neg(tab->mat->row[row][1]);
1283 static int row_sgn(struct isl_tab *tab, int row)
1286 return isl_int_sgn(tab->mat->row[row][1]);
1287 if (!isl_int_is_zero(tab->mat->row[row][2]))
1288 return isl_int_sgn(tab->mat->row[row][2]);
1290 return isl_int_sgn(tab->mat->row[row][1]);
1293 /* Perform pivots until the row variable "var" has a non-negative
1294 * sample value or until no more upward pivots can be performed.
1295 * Return the sign of the sample value after the pivots have been
1298 static int restore_row(struct isl_tab *tab, struct isl_tab_var *var)
1302 while (row_is_neg(tab, var->index)) {
1303 find_pivot(tab, var, var, 1, &row, &col);
1306 if (isl_tab_pivot(tab, row, col) < 0)
1308 if (!var->is_row) /* manifestly unbounded */
1311 return row_sgn(tab, var->index);
1314 /* Perform pivots until we are sure that the row variable "var"
1315 * can attain non-negative values. After return from this
1316 * function, "var" is still a row variable, but its sample
1317 * value may not be non-negative, even if the function returns 1.
1319 static int at_least_zero(struct isl_tab *tab, struct isl_tab_var *var)
1323 while (isl_int_is_neg(tab->mat->row[var->index][1])) {
1324 find_pivot(tab, var, var, 1, &row, &col);
1327 if (row == var->index) /* manifestly unbounded */
1329 if (isl_tab_pivot(tab, row, col) < 0)
1332 return !isl_int_is_neg(tab->mat->row[var->index][1]);
1335 /* Return a negative value if "var" can attain negative values.
1336 * Return a non-negative value otherwise.
1338 * If "var" is manifestly unbounded wrt negative values, we are done.
1339 * Otherwise, if var is in a column, we can pivot it down to a row.
1340 * Then we continue pivoting down until either
1341 * - the pivot would result in a manifestly unbounded column
1342 * => we don't perform the pivot, but simply return -1
1343 * - no more down pivots can be performed
1344 * - the sample value is negative
1345 * If the sample value becomes negative and the variable is supposed
1346 * to be nonnegative, then we undo the last pivot.
1347 * However, if the last pivot has made the pivoting variable
1348 * obviously redundant, then it may have moved to another row.
1349 * In that case we look for upward pivots until we reach a non-negative
1352 static int sign_of_min(struct isl_tab *tab, struct isl_tab_var *var)
1355 struct isl_tab_var *pivot_var = NULL;
1357 if (min_is_manifestly_unbounded(tab, var))
1361 row = pivot_row(tab, NULL, -1, col);
1362 pivot_var = var_from_col(tab, col);
1363 if (isl_tab_pivot(tab, row, col) < 0)
1365 if (var->is_redundant)
1367 if (isl_int_is_neg(tab->mat->row[var->index][1])) {
1368 if (var->is_nonneg) {
1369 if (!pivot_var->is_redundant &&
1370 pivot_var->index == row) {
1371 if (isl_tab_pivot(tab, row, col) < 0)
1374 if (restore_row(tab, var) < -1)
1380 if (var->is_redundant)
1382 while (!isl_int_is_neg(tab->mat->row[var->index][1])) {
1383 find_pivot(tab, var, var, -1, &row, &col);
1384 if (row == var->index)
1387 return isl_int_sgn(tab->mat->row[var->index][1]);
1388 pivot_var = var_from_col(tab, col);
1389 if (isl_tab_pivot(tab, row, col) < 0)
1391 if (var->is_redundant)
1394 if (pivot_var && var->is_nonneg) {
1395 /* pivot back to non-negative value */
1396 if (!pivot_var->is_redundant && pivot_var->index == row) {
1397 if (isl_tab_pivot(tab, row, col) < 0)
1400 if (restore_row(tab, var) < -1)
1406 static int row_at_most_neg_one(struct isl_tab *tab, int row)
1409 if (isl_int_is_pos(tab->mat->row[row][2]))
1411 if (isl_int_is_neg(tab->mat->row[row][2]))
1414 return isl_int_is_neg(tab->mat->row[row][1]) &&
1415 isl_int_abs_ge(tab->mat->row[row][1],
1416 tab->mat->row[row][0]);
1419 /* Return 1 if "var" can attain values <= -1.
1420 * Return 0 otherwise.
1422 * The sample value of "var" is assumed to be non-negative when the
1423 * the function is called. If 1 is returned then the constraint
1424 * is not redundant and the sample value is made non-negative again before
1425 * the function returns.
1427 int isl_tab_min_at_most_neg_one(struct isl_tab *tab, struct isl_tab_var *var)
1430 struct isl_tab_var *pivot_var;
1432 if (min_is_manifestly_unbounded(tab, var))
1436 row = pivot_row(tab, NULL, -1, col);
1437 pivot_var = var_from_col(tab, col);
1438 if (isl_tab_pivot(tab, row, col) < 0)
1440 if (var->is_redundant)
1442 if (row_at_most_neg_one(tab, var->index)) {
1443 if (var->is_nonneg) {
1444 if (!pivot_var->is_redundant &&
1445 pivot_var->index == row) {
1446 if (isl_tab_pivot(tab, row, col) < 0)
1449 if (restore_row(tab, var) < -1)
1455 if (var->is_redundant)
1458 find_pivot(tab, var, var, -1, &row, &col);
1459 if (row == var->index) {
1460 if (restore_row(tab, var) < -1)
1466 pivot_var = var_from_col(tab, col);
1467 if (isl_tab_pivot(tab, row, col) < 0)
1469 if (var->is_redundant)
1471 } while (!row_at_most_neg_one(tab, var->index));
1472 if (var->is_nonneg) {
1473 /* pivot back to non-negative value */
1474 if (!pivot_var->is_redundant && pivot_var->index == row)
1475 if (isl_tab_pivot(tab, row, col) < 0)
1477 if (restore_row(tab, var) < -1)
1483 /* Return 1 if "var" can attain values >= 1.
1484 * Return 0 otherwise.
1486 static int at_least_one(struct isl_tab *tab, struct isl_tab_var *var)
1491 if (max_is_manifestly_unbounded(tab, var))
1493 if (to_row(tab, var, 1) < 0)
1495 r = tab->mat->row[var->index];
1496 while (isl_int_lt(r[1], r[0])) {
1497 find_pivot(tab, var, var, 1, &row, &col);
1499 return isl_int_ge(r[1], r[0]);
1500 if (row == var->index) /* manifestly unbounded */
1502 if (isl_tab_pivot(tab, row, col) < 0)
1508 static void swap_cols(struct isl_tab *tab, int col1, int col2)
1511 unsigned off = 2 + tab->M;
1512 t = tab->col_var[col1];
1513 tab->col_var[col1] = tab->col_var[col2];
1514 tab->col_var[col2] = t;
1515 var_from_col(tab, col1)->index = col1;
1516 var_from_col(tab, col2)->index = col2;
1517 tab->mat = isl_mat_swap_cols(tab->mat, off + col1, off + col2);
1520 /* Mark column with index "col" as representing a zero variable.
1521 * If we may need to undo the operation the column is kept,
1522 * but no longer considered.
1523 * Otherwise, the column is simply removed.
1525 * The column may be interchanged with some other column. If it
1526 * is interchanged with a later column, return 1. Otherwise return 0.
1527 * If the columns are checked in order in the calling function,
1528 * then a return value of 1 means that the column with the given
1529 * column number may now contain a different column that
1530 * hasn't been checked yet.
1532 int isl_tab_kill_col(struct isl_tab *tab, int col)
1534 var_from_col(tab, col)->is_zero = 1;
1535 if (tab->need_undo) {
1536 if (isl_tab_push_var(tab, isl_tab_undo_zero,
1537 var_from_col(tab, col)) < 0)
1539 if (col != tab->n_dead)
1540 swap_cols(tab, col, tab->n_dead);
1544 if (col != tab->n_col - 1)
1545 swap_cols(tab, col, tab->n_col - 1);
1546 var_from_col(tab, tab->n_col - 1)->index = -1;
1552 static int row_is_manifestly_non_integral(struct isl_tab *tab, int row)
1554 unsigned off = 2 + tab->M;
1556 if (tab->M && !isl_int_eq(tab->mat->row[row][2],
1557 tab->mat->row[row][0]))
1559 if (isl_seq_first_non_zero(tab->mat->row[row] + off + tab->n_dead,
1560 tab->n_col - tab->n_dead) != -1)
1563 return !isl_int_is_divisible_by(tab->mat->row[row][1],
1564 tab->mat->row[row][0]);
1567 /* For integer tableaus, check if any of the coordinates are stuck
1568 * at a non-integral value.
1570 static int tab_is_manifestly_empty(struct isl_tab *tab)
1579 for (i = 0; i < tab->n_var; ++i) {
1580 if (!tab->var[i].is_row)
1582 if (row_is_manifestly_non_integral(tab, tab->var[i].index))
1589 /* Row variable "var" is non-negative and cannot attain any values
1590 * larger than zero. This means that the coefficients of the unrestricted
1591 * column variables are zero and that the coefficients of the non-negative
1592 * column variables are zero or negative.
1593 * Each of the non-negative variables with a negative coefficient can
1594 * then also be written as the negative sum of non-negative variables
1595 * and must therefore also be zero.
1597 static int close_row(struct isl_tab *tab, struct isl_tab_var *var) WARN_UNUSED;
1598 static int close_row(struct isl_tab *tab, struct isl_tab_var *var)
1601 struct isl_mat *mat = tab->mat;
1602 unsigned off = 2 + tab->M;
1604 isl_assert(tab->mat->ctx, var->is_nonneg, return -1);
1607 if (isl_tab_push_var(tab, isl_tab_undo_zero, var) < 0)
1609 for (j = tab->n_dead; j < tab->n_col; ++j) {
1611 if (isl_int_is_zero(mat->row[var->index][off + j]))
1613 isl_assert(tab->mat->ctx,
1614 isl_int_is_neg(mat->row[var->index][off + j]), return -1);
1615 recheck = isl_tab_kill_col(tab, j);
1621 if (isl_tab_mark_redundant(tab, var->index) < 0)
1623 if (tab_is_manifestly_empty(tab) && isl_tab_mark_empty(tab) < 0)
1628 /* Add a constraint to the tableau and allocate a row for it.
1629 * Return the index into the constraint array "con".
1631 int isl_tab_allocate_con(struct isl_tab *tab)
1635 isl_assert(tab->mat->ctx, tab->n_row < tab->mat->n_row, return -1);
1636 isl_assert(tab->mat->ctx, tab->n_con < tab->max_con, return -1);
1639 tab->con[r].index = tab->n_row;
1640 tab->con[r].is_row = 1;
1641 tab->con[r].is_nonneg = 0;
1642 tab->con[r].is_zero = 0;
1643 tab->con[r].is_redundant = 0;
1644 tab->con[r].frozen = 0;
1645 tab->con[r].negated = 0;
1646 tab->row_var[tab->n_row] = ~r;
1650 if (isl_tab_push_var(tab, isl_tab_undo_allocate, &tab->con[r]) < 0)
1656 /* Add a variable to the tableau and allocate a column for it.
1657 * Return the index into the variable array "var".
1659 int isl_tab_allocate_var(struct isl_tab *tab)
1663 unsigned off = 2 + tab->M;
1665 isl_assert(tab->mat->ctx, tab->n_col < tab->mat->n_col, return -1);
1666 isl_assert(tab->mat->ctx, tab->n_var < tab->max_var, return -1);
1669 tab->var[r].index = tab->n_col;
1670 tab->var[r].is_row = 0;
1671 tab->var[r].is_nonneg = 0;
1672 tab->var[r].is_zero = 0;
1673 tab->var[r].is_redundant = 0;
1674 tab->var[r].frozen = 0;
1675 tab->var[r].negated = 0;
1676 tab->col_var[tab->n_col] = r;
1678 for (i = 0; i < tab->n_row; ++i)
1679 isl_int_set_si(tab->mat->row[i][off + tab->n_col], 0);
1683 if (isl_tab_push_var(tab, isl_tab_undo_allocate, &tab->var[r]) < 0)
1689 /* Add a row to the tableau. The row is given as an affine combination
1690 * of the original variables and needs to be expressed in terms of the
1693 * We add each term in turn.
1694 * If r = n/d_r is the current sum and we need to add k x, then
1695 * if x is a column variable, we increase the numerator of
1696 * this column by k d_r
1697 * if x = f/d_x is a row variable, then the new representation of r is
1699 * n k f d_x/g n + d_r/g k f m/d_r n + m/d_g k f
1700 * --- + --- = ------------------- = -------------------
1701 * d_r d_r d_r d_x/g m
1703 * with g the gcd of d_r and d_x and m the lcm of d_r and d_x.
1705 * If tab->M is set, then, internally, each variable x is represented
1706 * as x' - M. We then also need no subtract k d_r from the coefficient of M.
1708 int isl_tab_add_row(struct isl_tab *tab, isl_int *line)
1714 unsigned off = 2 + tab->M;
1716 r = isl_tab_allocate_con(tab);
1722 row = tab->mat->row[tab->con[r].index];
1723 isl_int_set_si(row[0], 1);
1724 isl_int_set(row[1], line[0]);
1725 isl_seq_clr(row + 2, tab->M + tab->n_col);
1726 for (i = 0; i < tab->n_var; ++i) {
1727 if (tab->var[i].is_zero)
1729 if (tab->var[i].is_row) {
1731 row[0], tab->mat->row[tab->var[i].index][0]);
1732 isl_int_swap(a, row[0]);
1733 isl_int_divexact(a, row[0], a);
1735 row[0], tab->mat->row[tab->var[i].index][0]);
1736 isl_int_mul(b, b, line[1 + i]);
1737 isl_seq_combine(row + 1, a, row + 1,
1738 b, tab->mat->row[tab->var[i].index] + 1,
1739 1 + tab->M + tab->n_col);
1741 isl_int_addmul(row[off + tab->var[i].index],
1742 line[1 + i], row[0]);
1743 if (tab->M && i >= tab->n_param && i < tab->n_var - tab->n_div)
1744 isl_int_submul(row[2], line[1 + i], row[0]);
1746 isl_seq_normalize(tab->mat->ctx, row, off + tab->n_col);
1751 tab->row_sign[tab->con[r].index] = isl_tab_row_unknown;
1756 static int drop_row(struct isl_tab *tab, int row)
1758 isl_assert(tab->mat->ctx, ~tab->row_var[row] == tab->n_con - 1, return -1);
1759 if (row != tab->n_row - 1)
1760 swap_rows(tab, row, tab->n_row - 1);
1766 static int drop_col(struct isl_tab *tab, int col)
1768 isl_assert(tab->mat->ctx, tab->col_var[col] == tab->n_var - 1, return -1);
1769 if (col != tab->n_col - 1)
1770 swap_cols(tab, col, tab->n_col - 1);
1776 /* Add inequality "ineq" and check if it conflicts with the
1777 * previously added constraints or if it is obviously redundant.
1779 int isl_tab_add_ineq(struct isl_tab *tab, isl_int *ineq)
1788 struct isl_basic_map *bmap = tab->bmap;
1790 isl_assert(tab->mat->ctx, tab->n_eq == bmap->n_eq, return -1);
1791 isl_assert(tab->mat->ctx,
1792 tab->n_con == bmap->n_eq + bmap->n_ineq, return -1);
1793 tab->bmap = isl_basic_map_add_ineq(tab->bmap, ineq);
1794 if (isl_tab_push(tab, isl_tab_undo_bmap_ineq) < 0)
1801 isl_int_swap(ineq[0], cst);
1803 r = isl_tab_add_row(tab, ineq);
1805 isl_int_swap(ineq[0], cst);
1810 tab->con[r].is_nonneg = 1;
1811 if (isl_tab_push_var(tab, isl_tab_undo_nonneg, &tab->con[r]) < 0)
1813 if (isl_tab_row_is_redundant(tab, tab->con[r].index)) {
1814 if (isl_tab_mark_redundant(tab, tab->con[r].index) < 0)
1819 sgn = restore_row(tab, &tab->con[r]);
1823 return isl_tab_mark_empty(tab);
1824 if (tab->con[r].is_row && isl_tab_row_is_redundant(tab, tab->con[r].index))
1825 if (isl_tab_mark_redundant(tab, tab->con[r].index) < 0)
1830 /* Pivot a non-negative variable down until it reaches the value zero
1831 * and then pivot the variable into a column position.
1833 static int to_col(struct isl_tab *tab, struct isl_tab_var *var) WARN_UNUSED;
1834 static int to_col(struct isl_tab *tab, struct isl_tab_var *var)
1838 unsigned off = 2 + tab->M;
1843 while (isl_int_is_pos(tab->mat->row[var->index][1])) {
1844 find_pivot(tab, var, NULL, -1, &row, &col);
1845 isl_assert(tab->mat->ctx, row != -1, return -1);
1846 if (isl_tab_pivot(tab, row, col) < 0)
1852 for (i = tab->n_dead; i < tab->n_col; ++i)
1853 if (!isl_int_is_zero(tab->mat->row[var->index][off + i]))
1856 isl_assert(tab->mat->ctx, i < tab->n_col, return -1);
1857 if (isl_tab_pivot(tab, var->index, i) < 0)
1863 /* We assume Gaussian elimination has been performed on the equalities.
1864 * The equalities can therefore never conflict.
1865 * Adding the equalities is currently only really useful for a later call
1866 * to isl_tab_ineq_type.
1868 static struct isl_tab *add_eq(struct isl_tab *tab, isl_int *eq)
1875 r = isl_tab_add_row(tab, eq);
1879 r = tab->con[r].index;
1880 i = isl_seq_first_non_zero(tab->mat->row[r] + 2 + tab->M + tab->n_dead,
1881 tab->n_col - tab->n_dead);
1882 isl_assert(tab->mat->ctx, i >= 0, goto error);
1884 if (isl_tab_pivot(tab, r, i) < 0)
1886 if (isl_tab_kill_col(tab, i) < 0)
1896 static int row_is_manifestly_zero(struct isl_tab *tab, int row)
1898 unsigned off = 2 + tab->M;
1900 if (!isl_int_is_zero(tab->mat->row[row][1]))
1902 if (tab->M && !isl_int_is_zero(tab->mat->row[row][2]))
1904 return isl_seq_first_non_zero(tab->mat->row[row] + off + tab->n_dead,
1905 tab->n_col - tab->n_dead) == -1;
1908 /* Add an equality that is known to be valid for the given tableau.
1910 int isl_tab_add_valid_eq(struct isl_tab *tab, isl_int *eq)
1912 struct isl_tab_var *var;
1917 r = isl_tab_add_row(tab, eq);
1923 if (row_is_manifestly_zero(tab, r)) {
1925 if (isl_tab_mark_redundant(tab, r) < 0)
1930 if (isl_int_is_neg(tab->mat->row[r][1])) {
1931 isl_seq_neg(tab->mat->row[r] + 1, tab->mat->row[r] + 1,
1936 if (to_col(tab, var) < 0)
1939 if (isl_tab_kill_col(tab, var->index) < 0)
1945 static int add_zero_row(struct isl_tab *tab)
1950 r = isl_tab_allocate_con(tab);
1954 row = tab->mat->row[tab->con[r].index];
1955 isl_seq_clr(row + 1, 1 + tab->M + tab->n_col);
1956 isl_int_set_si(row[0], 1);
1961 /* Add equality "eq" and check if it conflicts with the
1962 * previously added constraints or if it is obviously redundant.
1964 int isl_tab_add_eq(struct isl_tab *tab, isl_int *eq)
1966 struct isl_tab_undo *snap = NULL;
1967 struct isl_tab_var *var;
1975 isl_assert(tab->mat->ctx, !tab->M, return -1);
1978 snap = isl_tab_snap(tab);
1982 isl_int_swap(eq[0], cst);
1984 r = isl_tab_add_row(tab, eq);
1986 isl_int_swap(eq[0], cst);
1994 if (row_is_manifestly_zero(tab, row)) {
1996 if (isl_tab_rollback(tab, snap) < 0)
2004 tab->bmap = isl_basic_map_add_ineq(tab->bmap, eq);
2005 if (isl_tab_push(tab, isl_tab_undo_bmap_ineq) < 0)
2007 isl_seq_neg(eq, eq, 1 + tab->n_var);
2008 tab->bmap = isl_basic_map_add_ineq(tab->bmap, eq);
2009 isl_seq_neg(eq, eq, 1 + tab->n_var);
2010 if (isl_tab_push(tab, isl_tab_undo_bmap_ineq) < 0)
2014 if (add_zero_row(tab) < 0)
2018 sgn = isl_int_sgn(tab->mat->row[row][1]);
2021 isl_seq_neg(tab->mat->row[row] + 1, tab->mat->row[row] + 1,
2028 sgn = sign_of_max(tab, var);
2032 if (isl_tab_mark_empty(tab) < 0)
2039 if (to_col(tab, var) < 0)
2042 if (isl_tab_kill_col(tab, var->index) < 0)
2048 /* Construct and return an inequality that expresses an upper bound
2050 * In particular, if the div is given by
2054 * then the inequality expresses
2058 static struct isl_vec *ineq_for_div(struct isl_basic_map *bmap, unsigned div)
2062 struct isl_vec *ineq;
2067 total = isl_basic_map_total_dim(bmap);
2068 div_pos = 1 + total - bmap->n_div + div;
2070 ineq = isl_vec_alloc(bmap->ctx, 1 + total);
2074 isl_seq_cpy(ineq->el, bmap->div[div] + 1, 1 + total);
2075 isl_int_neg(ineq->el[div_pos], bmap->div[div][0]);
2079 /* For a div d = floor(f/m), add the constraints
2082 * -(f-(m-1)) + m d >= 0
2084 * Note that the second constraint is the negation of
2088 * If add_ineq is not NULL, then this function is used
2089 * instead of isl_tab_add_ineq to effectively add the inequalities.
2091 static int add_div_constraints(struct isl_tab *tab, unsigned div,
2092 int (*add_ineq)(void *user, isl_int *), void *user)
2096 struct isl_vec *ineq;
2098 total = isl_basic_map_total_dim(tab->bmap);
2099 div_pos = 1 + total - tab->bmap->n_div + div;
2101 ineq = ineq_for_div(tab->bmap, div);
2106 if (add_ineq(user, ineq->el) < 0)
2109 if (isl_tab_add_ineq(tab, ineq->el) < 0)
2113 isl_seq_neg(ineq->el, tab->bmap->div[div] + 1, 1 + total);
2114 isl_int_set(ineq->el[div_pos], tab->bmap->div[div][0]);
2115 isl_int_add(ineq->el[0], ineq->el[0], ineq->el[div_pos]);
2116 isl_int_sub_ui(ineq->el[0], ineq->el[0], 1);
2119 if (add_ineq(user, ineq->el) < 0)
2122 if (isl_tab_add_ineq(tab, ineq->el) < 0)
2134 /* Check whether the div described by "div" is obviously non-negative.
2135 * If we are using a big parameter, then we will encode the div
2136 * as div' = M + div, which is always non-negative.
2137 * Otherwise, we check whether div is a non-negative affine combination
2138 * of non-negative variables.
2140 static int div_is_nonneg(struct isl_tab *tab, __isl_keep isl_vec *div)
2147 if (isl_int_is_neg(div->el[1]))
2150 for (i = 0; i < tab->n_var; ++i) {
2151 if (isl_int_is_neg(div->el[2 + i]))
2153 if (isl_int_is_zero(div->el[2 + i]))
2155 if (!tab->var[i].is_nonneg)
2162 /* Add an extra div, prescribed by "div" to the tableau and
2163 * the associated bmap (which is assumed to be non-NULL).
2165 * If add_ineq is not NULL, then this function is used instead
2166 * of isl_tab_add_ineq to add the div constraints.
2167 * This complication is needed because the code in isl_tab_pip
2168 * wants to perform some extra processing when an inequality
2169 * is added to the tableau.
2171 int isl_tab_add_div(struct isl_tab *tab, __isl_keep isl_vec *div,
2172 int (*add_ineq)(void *user, isl_int *), void *user)
2181 isl_assert(tab->mat->ctx, tab->bmap, return -1);
2183 nonneg = div_is_nonneg(tab, div);
2185 if (isl_tab_extend_cons(tab, 3) < 0)
2187 if (isl_tab_extend_vars(tab, 1) < 0)
2189 r = isl_tab_allocate_var(tab);
2194 tab->var[r].is_nonneg = 1;
2196 tab->bmap = isl_basic_map_extend_space(tab->bmap,
2197 isl_basic_map_get_space(tab->bmap), 1, 0, 2);
2198 k = isl_basic_map_alloc_div(tab->bmap);
2201 isl_seq_cpy(tab->bmap->div[k], div->el, div->size);
2202 if (isl_tab_push(tab, isl_tab_undo_bmap_div) < 0)
2205 if (add_div_constraints(tab, k, add_ineq, user) < 0)
2211 /* If "track" is set, then we want to keep track of all constraints in tab
2212 * in its bmap field. This field is initialized from a copy of "bmap",
2213 * so we need to make sure that all constraints in "bmap" also appear
2214 * in the constructed tab.
2216 __isl_give struct isl_tab *isl_tab_from_basic_map(
2217 __isl_keep isl_basic_map *bmap, int track)
2220 struct isl_tab *tab;
2224 tab = isl_tab_alloc(bmap->ctx,
2225 isl_basic_map_total_dim(bmap) + bmap->n_ineq + 1,
2226 isl_basic_map_total_dim(bmap), 0);
2229 tab->preserve = track;
2230 tab->rational = ISL_F_ISSET(bmap, ISL_BASIC_MAP_RATIONAL);
2231 if (ISL_F_ISSET(bmap, ISL_BASIC_MAP_EMPTY)) {
2232 if (isl_tab_mark_empty(tab) < 0)
2236 for (i = 0; i < bmap->n_eq; ++i) {
2237 tab = add_eq(tab, bmap->eq[i]);
2241 for (i = 0; i < bmap->n_ineq; ++i) {
2242 if (isl_tab_add_ineq(tab, bmap->ineq[i]) < 0)
2248 if (track && isl_tab_track_bmap(tab, isl_basic_map_copy(bmap)) < 0)
2256 __isl_give struct isl_tab *isl_tab_from_basic_set(
2257 __isl_keep isl_basic_set *bset, int track)
2259 return isl_tab_from_basic_map(bset, track);
2262 /* Construct a tableau corresponding to the recession cone of "bset".
2264 struct isl_tab *isl_tab_from_recession_cone(__isl_keep isl_basic_set *bset,
2269 struct isl_tab *tab;
2270 unsigned offset = 0;
2275 offset = isl_basic_set_dim(bset, isl_dim_param);
2276 tab = isl_tab_alloc(bset->ctx, bset->n_eq + bset->n_ineq,
2277 isl_basic_set_total_dim(bset) - offset, 0);
2280 tab->rational = ISL_F_ISSET(bset, ISL_BASIC_SET_RATIONAL);
2284 for (i = 0; i < bset->n_eq; ++i) {
2285 isl_int_swap(bset->eq[i][offset], cst);
2287 if (isl_tab_add_eq(tab, bset->eq[i] + offset) < 0)
2290 tab = add_eq(tab, bset->eq[i]);
2291 isl_int_swap(bset->eq[i][offset], cst);
2295 for (i = 0; i < bset->n_ineq; ++i) {
2297 isl_int_swap(bset->ineq[i][offset], cst);
2298 r = isl_tab_add_row(tab, bset->ineq[i] + offset);
2299 isl_int_swap(bset->ineq[i][offset], cst);
2302 tab->con[r].is_nonneg = 1;
2303 if (isl_tab_push_var(tab, isl_tab_undo_nonneg, &tab->con[r]) < 0)
2315 /* Assuming "tab" is the tableau of a cone, check if the cone is
2316 * bounded, i.e., if it is empty or only contains the origin.
2318 int isl_tab_cone_is_bounded(struct isl_tab *tab)
2326 if (tab->n_dead == tab->n_col)
2330 for (i = tab->n_redundant; i < tab->n_row; ++i) {
2331 struct isl_tab_var *var;
2333 var = isl_tab_var_from_row(tab, i);
2334 if (!var->is_nonneg)
2336 sgn = sign_of_max(tab, var);
2341 if (close_row(tab, var) < 0)
2345 if (tab->n_dead == tab->n_col)
2347 if (i == tab->n_row)
2352 int isl_tab_sample_is_integer(struct isl_tab *tab)
2359 for (i = 0; i < tab->n_var; ++i) {
2361 if (!tab->var[i].is_row)
2363 row = tab->var[i].index;
2364 if (!isl_int_is_divisible_by(tab->mat->row[row][1],
2365 tab->mat->row[row][0]))
2371 static struct isl_vec *extract_integer_sample(struct isl_tab *tab)
2374 struct isl_vec *vec;
2376 vec = isl_vec_alloc(tab->mat->ctx, 1 + tab->n_var);
2380 isl_int_set_si(vec->block.data[0], 1);
2381 for (i = 0; i < tab->n_var; ++i) {
2382 if (!tab->var[i].is_row)
2383 isl_int_set_si(vec->block.data[1 + i], 0);
2385 int row = tab->var[i].index;
2386 isl_int_divexact(vec->block.data[1 + i],
2387 tab->mat->row[row][1], tab->mat->row[row][0]);
2394 struct isl_vec *isl_tab_get_sample_value(struct isl_tab *tab)
2397 struct isl_vec *vec;
2403 vec = isl_vec_alloc(tab->mat->ctx, 1 + tab->n_var);
2409 isl_int_set_si(vec->block.data[0], 1);
2410 for (i = 0; i < tab->n_var; ++i) {
2412 if (!tab->var[i].is_row) {
2413 isl_int_set_si(vec->block.data[1 + i], 0);
2416 row = tab->var[i].index;
2417 isl_int_gcd(m, vec->block.data[0], tab->mat->row[row][0]);
2418 isl_int_divexact(m, tab->mat->row[row][0], m);
2419 isl_seq_scale(vec->block.data, vec->block.data, m, 1 + i);
2420 isl_int_divexact(m, vec->block.data[0], tab->mat->row[row][0]);
2421 isl_int_mul(vec->block.data[1 + i], m, tab->mat->row[row][1]);
2423 vec = isl_vec_normalize(vec);
2429 /* Update "bmap" based on the results of the tableau "tab".
2430 * In particular, implicit equalities are made explicit, redundant constraints
2431 * are removed and if the sample value happens to be integer, it is stored
2432 * in "bmap" (unless "bmap" already had an integer sample).
2434 * The tableau is assumed to have been created from "bmap" using
2435 * isl_tab_from_basic_map.
2437 struct isl_basic_map *isl_basic_map_update_from_tab(struct isl_basic_map *bmap,
2438 struct isl_tab *tab)
2450 bmap = isl_basic_map_set_to_empty(bmap);
2452 for (i = bmap->n_ineq - 1; i >= 0; --i) {
2453 if (isl_tab_is_equality(tab, n_eq + i))
2454 isl_basic_map_inequality_to_equality(bmap, i);
2455 else if (isl_tab_is_redundant(tab, n_eq + i))
2456 isl_basic_map_drop_inequality(bmap, i);
2458 if (bmap->n_eq != n_eq)
2459 isl_basic_map_gauss(bmap, NULL);
2460 if (!tab->rational &&
2461 !bmap->sample && isl_tab_sample_is_integer(tab))
2462 bmap->sample = extract_integer_sample(tab);
2466 struct isl_basic_set *isl_basic_set_update_from_tab(struct isl_basic_set *bset,
2467 struct isl_tab *tab)
2469 return (struct isl_basic_set *)isl_basic_map_update_from_tab(
2470 (struct isl_basic_map *)bset, tab);
2473 /* Given a non-negative variable "var", add a new non-negative variable
2474 * that is the opposite of "var", ensuring that var can only attain the
2476 * If var = n/d is a row variable, then the new variable = -n/d.
2477 * If var is a column variables, then the new variable = -var.
2478 * If the new variable cannot attain non-negative values, then
2479 * the resulting tableau is empty.
2480 * Otherwise, we know the value will be zero and we close the row.
2482 static int cut_to_hyperplane(struct isl_tab *tab, struct isl_tab_var *var)
2487 unsigned off = 2 + tab->M;
2491 isl_assert(tab->mat->ctx, !var->is_redundant, return -1);
2492 isl_assert(tab->mat->ctx, var->is_nonneg, return -1);
2494 if (isl_tab_extend_cons(tab, 1) < 0)
2498 tab->con[r].index = tab->n_row;
2499 tab->con[r].is_row = 1;
2500 tab->con[r].is_nonneg = 0;
2501 tab->con[r].is_zero = 0;
2502 tab->con[r].is_redundant = 0;
2503 tab->con[r].frozen = 0;
2504 tab->con[r].negated = 0;
2505 tab->row_var[tab->n_row] = ~r;
2506 row = tab->mat->row[tab->n_row];
2509 isl_int_set(row[0], tab->mat->row[var->index][0]);
2510 isl_seq_neg(row + 1,
2511 tab->mat->row[var->index] + 1, 1 + tab->n_col);
2513 isl_int_set_si(row[0], 1);
2514 isl_seq_clr(row + 1, 1 + tab->n_col);
2515 isl_int_set_si(row[off + var->index], -1);
2520 if (isl_tab_push_var(tab, isl_tab_undo_allocate, &tab->con[r]) < 0)
2523 sgn = sign_of_max(tab, &tab->con[r]);
2527 if (isl_tab_mark_empty(tab) < 0)
2531 tab->con[r].is_nonneg = 1;
2532 if (isl_tab_push_var(tab, isl_tab_undo_nonneg, &tab->con[r]) < 0)
2535 if (close_row(tab, &tab->con[r]) < 0)
2541 /* Given a tableau "tab" and an inequality constraint "con" of the tableau,
2542 * relax the inequality by one. That is, the inequality r >= 0 is replaced
2543 * by r' = r + 1 >= 0.
2544 * If r is a row variable, we simply increase the constant term by one
2545 * (taking into account the denominator).
2546 * If r is a column variable, then we need to modify each row that
2547 * refers to r = r' - 1 by substituting this equality, effectively
2548 * subtracting the coefficient of the column from the constant.
2549 * We should only do this if the minimum is manifestly unbounded,
2550 * however. Otherwise, we may end up with negative sample values
2551 * for non-negative variables.
2552 * So, if r is a column variable with a minimum that is not
2553 * manifestly unbounded, then we need to move it to a row.
2554 * However, the sample value of this row may be negative,
2555 * even after the relaxation, so we need to restore it.
2556 * We therefore prefer to pivot a column up to a row, if possible.
2558 struct isl_tab *isl_tab_relax(struct isl_tab *tab, int con)
2560 struct isl_tab_var *var;
2561 unsigned off = 2 + tab->M;
2566 var = &tab->con[con];
2568 if (var->is_row && (var->index < 0 || var->index < tab->n_redundant))
2569 isl_die(tab->mat->ctx, isl_error_invalid,
2570 "cannot relax redundant constraint", goto error);
2571 if (!var->is_row && (var->index < 0 || var->index < tab->n_dead))
2572 isl_die(tab->mat->ctx, isl_error_invalid,
2573 "cannot relax dead constraint", goto error);
2575 if (!var->is_row && !max_is_manifestly_unbounded(tab, var))
2576 if (to_row(tab, var, 1) < 0)
2578 if (!var->is_row && !min_is_manifestly_unbounded(tab, var))
2579 if (to_row(tab, var, -1) < 0)
2583 isl_int_add(tab->mat->row[var->index][1],
2584 tab->mat->row[var->index][1], tab->mat->row[var->index][0]);
2585 if (restore_row(tab, var) < 0)
2590 for (i = 0; i < tab->n_row; ++i) {
2591 if (isl_int_is_zero(tab->mat->row[i][off + var->index]))
2593 isl_int_sub(tab->mat->row[i][1], tab->mat->row[i][1],
2594 tab->mat->row[i][off + var->index]);
2599 if (isl_tab_push_var(tab, isl_tab_undo_relax, var) < 0)
2608 /* Remove the sign constraint from constraint "con".
2610 * If the constraint variable was originally marked non-negative,
2611 * then we make sure we mark it non-negative again during rollback.
2613 int isl_tab_unrestrict(struct isl_tab *tab, int con)
2615 struct isl_tab_var *var;
2620 var = &tab->con[con];
2621 if (!var->is_nonneg)
2625 if (isl_tab_push_var(tab, isl_tab_undo_unrestrict, var) < 0)
2631 int isl_tab_select_facet(struct isl_tab *tab, int con)
2636 return cut_to_hyperplane(tab, &tab->con[con]);
2639 static int may_be_equality(struct isl_tab *tab, int row)
2641 return tab->rational ? isl_int_is_zero(tab->mat->row[row][1])
2642 : isl_int_lt(tab->mat->row[row][1],
2643 tab->mat->row[row][0]);
2646 /* Check for (near) equalities among the constraints.
2647 * A constraint is an equality if it is non-negative and if
2648 * its maximal value is either
2649 * - zero (in case of rational tableaus), or
2650 * - strictly less than 1 (in case of integer tableaus)
2652 * We first mark all non-redundant and non-dead variables that
2653 * are not frozen and not obviously not an equality.
2654 * Then we iterate over all marked variables if they can attain
2655 * any values larger than zero or at least one.
2656 * If the maximal value is zero, we mark any column variables
2657 * that appear in the row as being zero and mark the row as being redundant.
2658 * Otherwise, if the maximal value is strictly less than one (and the
2659 * tableau is integer), then we restrict the value to being zero
2660 * by adding an opposite non-negative variable.
2662 int isl_tab_detect_implicit_equalities(struct isl_tab *tab)
2671 if (tab->n_dead == tab->n_col)
2675 for (i = tab->n_redundant; i < tab->n_row; ++i) {
2676 struct isl_tab_var *var = isl_tab_var_from_row(tab, i);
2677 var->marked = !var->frozen && var->is_nonneg &&
2678 may_be_equality(tab, i);
2682 for (i = tab->n_dead; i < tab->n_col; ++i) {
2683 struct isl_tab_var *var = var_from_col(tab, i);
2684 var->marked = !var->frozen && var->is_nonneg;
2689 struct isl_tab_var *var;
2691 for (i = tab->n_redundant; i < tab->n_row; ++i) {
2692 var = isl_tab_var_from_row(tab, i);
2696 if (i == tab->n_row) {
2697 for (i = tab->n_dead; i < tab->n_col; ++i) {
2698 var = var_from_col(tab, i);
2702 if (i == tab->n_col)
2707 sgn = sign_of_max(tab, var);
2711 if (close_row(tab, var) < 0)
2713 } else if (!tab->rational && !at_least_one(tab, var)) {
2714 if (cut_to_hyperplane(tab, var) < 0)
2716 return isl_tab_detect_implicit_equalities(tab);
2718 for (i = tab->n_redundant; i < tab->n_row; ++i) {
2719 var = isl_tab_var_from_row(tab, i);
2722 if (may_be_equality(tab, i))
2732 /* Update the element of row_var or col_var that corresponds to
2733 * constraint tab->con[i] to a move from position "old" to position "i".
2735 static int update_con_after_move(struct isl_tab *tab, int i, int old)
2740 index = tab->con[i].index;
2743 p = tab->con[i].is_row ? tab->row_var : tab->col_var;
2744 if (p[index] != ~old)
2745 isl_die(tab->mat->ctx, isl_error_internal,
2746 "broken internal state", return -1);
2752 /* Rotate the "n" constraints starting at "first" to the right,
2753 * putting the last constraint in the position of the first constraint.
2755 static int rotate_constraints(struct isl_tab *tab, int first, int n)
2758 struct isl_tab_var var;
2763 last = first + n - 1;
2764 var = tab->con[last];
2765 for (i = last; i > first; --i) {
2766 tab->con[i] = tab->con[i - 1];
2767 if (update_con_after_move(tab, i, i - 1) < 0)
2770 tab->con[first] = var;
2771 if (update_con_after_move(tab, first, last) < 0)
2777 /* Make the equalities that are implicit in "bmap" but that have been
2778 * detected in the corresponding "tab" explicit in "bmap" and update
2779 * "tab" to reflect the new order of the constraints.
2781 * In particular, if inequality i is an implicit equality then
2782 * isl_basic_map_inequality_to_equality will move the inequality
2783 * in front of the other equality and it will move the last inequality
2784 * in the position of inequality i.
2785 * In the tableau, the inequalities of "bmap" are stored after the equalities
2786 * and so the original order
2788 * E E E E E A A A I B B B B L
2792 * I E E E E E A A A L B B B B
2794 * where I is the implicit equality, the E are equalities,
2795 * the A inequalities before I, the B inequalities after I and
2796 * L the last inequality.
2797 * We therefore need to rotate to the right two sets of constraints,
2798 * those up to and including I and those after I.
2800 * If "tab" contains any constraints that are not in "bmap" then they
2801 * appear after those in "bmap" and they should be left untouched.
2803 * Note that this function leaves "bmap" in a temporary state
2804 * as it does not call isl_basic_map_gauss. Calling this function
2805 * is the responsibility of the caller.
2807 __isl_give isl_basic_map *isl_tab_make_equalities_explicit(struct isl_tab *tab,
2808 __isl_take isl_basic_map *bmap)
2813 return isl_basic_map_free(bmap);
2817 for (i = bmap->n_ineq - 1; i >= 0; --i) {
2818 if (!isl_tab_is_equality(tab, bmap->n_eq + i))
2820 isl_basic_map_inequality_to_equality(bmap, i);
2821 if (rotate_constraints(tab, 0, tab->n_eq + i + 1) < 0)
2822 return isl_basic_map_free(bmap);
2823 if (rotate_constraints(tab, tab->n_eq + i + 1,
2824 bmap->n_ineq - i) < 0)
2825 return isl_basic_map_free(bmap);
2832 static int con_is_redundant(struct isl_tab *tab, struct isl_tab_var *var)
2836 if (tab->rational) {
2837 int sgn = sign_of_min(tab, var);
2842 int irred = isl_tab_min_at_most_neg_one(tab, var);
2849 /* Check for (near) redundant constraints.
2850 * A constraint is redundant if it is non-negative and if
2851 * its minimal value (temporarily ignoring the non-negativity) is either
2852 * - zero (in case of rational tableaus), or
2853 * - strictly larger than -1 (in case of integer tableaus)
2855 * We first mark all non-redundant and non-dead variables that
2856 * are not frozen and not obviously negatively unbounded.
2857 * Then we iterate over all marked variables if they can attain
2858 * any values smaller than zero or at most negative one.
2859 * If not, we mark the row as being redundant (assuming it hasn't
2860 * been detected as being obviously redundant in the mean time).
2862 int isl_tab_detect_redundant(struct isl_tab *tab)
2871 if (tab->n_redundant == tab->n_row)
2875 for (i = tab->n_redundant; i < tab->n_row; ++i) {
2876 struct isl_tab_var *var = isl_tab_var_from_row(tab, i);
2877 var->marked = !var->frozen && var->is_nonneg;
2881 for (i = tab->n_dead; i < tab->n_col; ++i) {
2882 struct isl_tab_var *var = var_from_col(tab, i);
2883 var->marked = !var->frozen && var->is_nonneg &&
2884 !min_is_manifestly_unbounded(tab, var);
2889 struct isl_tab_var *var;
2891 for (i = tab->n_redundant; i < tab->n_row; ++i) {
2892 var = isl_tab_var_from_row(tab, i);
2896 if (i == tab->n_row) {
2897 for (i = tab->n_dead; i < tab->n_col; ++i) {
2898 var = var_from_col(tab, i);
2902 if (i == tab->n_col)
2907 red = con_is_redundant(tab, var);
2910 if (red && !var->is_redundant)
2911 if (isl_tab_mark_redundant(tab, var->index) < 0)
2913 for (i = tab->n_dead; i < tab->n_col; ++i) {
2914 var = var_from_col(tab, i);
2917 if (!min_is_manifestly_unbounded(tab, var))
2927 int isl_tab_is_equality(struct isl_tab *tab, int con)
2934 if (tab->con[con].is_zero)
2936 if (tab->con[con].is_redundant)
2938 if (!tab->con[con].is_row)
2939 return tab->con[con].index < tab->n_dead;
2941 row = tab->con[con].index;
2944 return isl_int_is_zero(tab->mat->row[row][1]) &&
2945 (!tab->M || isl_int_is_zero(tab->mat->row[row][2])) &&
2946 isl_seq_first_non_zero(tab->mat->row[row] + off + tab->n_dead,
2947 tab->n_col - tab->n_dead) == -1;
2950 /* Return the minimal value of the affine expression "f" with denominator
2951 * "denom" in *opt, *opt_denom, assuming the tableau is not empty and
2952 * the expression cannot attain arbitrarily small values.
2953 * If opt_denom is NULL, then *opt is rounded up to the nearest integer.
2954 * The return value reflects the nature of the result (empty, unbounded,
2955 * minimal value returned in *opt).
2957 enum isl_lp_result isl_tab_min(struct isl_tab *tab,
2958 isl_int *f, isl_int denom, isl_int *opt, isl_int *opt_denom,
2962 enum isl_lp_result res = isl_lp_ok;
2963 struct isl_tab_var *var;
2964 struct isl_tab_undo *snap;
2967 return isl_lp_error;
2970 return isl_lp_empty;
2972 snap = isl_tab_snap(tab);
2973 r = isl_tab_add_row(tab, f);
2975 return isl_lp_error;
2979 find_pivot(tab, var, var, -1, &row, &col);
2980 if (row == var->index) {
2981 res = isl_lp_unbounded;
2986 if (isl_tab_pivot(tab, row, col) < 0)
2987 return isl_lp_error;
2989 isl_int_mul(tab->mat->row[var->index][0],
2990 tab->mat->row[var->index][0], denom);
2991 if (ISL_FL_ISSET(flags, ISL_TAB_SAVE_DUAL)) {
2994 isl_vec_free(tab->dual);
2995 tab->dual = isl_vec_alloc(tab->mat->ctx, 1 + tab->n_con);
2997 return isl_lp_error;
2998 isl_int_set(tab->dual->el[0], tab->mat->row[var->index][0]);
2999 for (i = 0; i < tab->n_con; ++i) {
3001 if (tab->con[i].is_row) {
3002 isl_int_set_si(tab->dual->el[1 + i], 0);
3005 pos = 2 + tab->M + tab->con[i].index;
3006 if (tab->con[i].negated)
3007 isl_int_neg(tab->dual->el[1 + i],
3008 tab->mat->row[var->index][pos]);
3010 isl_int_set(tab->dual->el[1 + i],
3011 tab->mat->row[var->index][pos]);
3014 if (opt && res == isl_lp_ok) {
3016 isl_int_set(*opt, tab->mat->row[var->index][1]);
3017 isl_int_set(*opt_denom, tab->mat->row[var->index][0]);
3019 isl_int_cdiv_q(*opt, tab->mat->row[var->index][1],
3020 tab->mat->row[var->index][0]);
3022 if (isl_tab_rollback(tab, snap) < 0)
3023 return isl_lp_error;
3027 int isl_tab_is_redundant(struct isl_tab *tab, int con)
3031 if (tab->con[con].is_zero)
3033 if (tab->con[con].is_redundant)
3035 return tab->con[con].is_row && tab->con[con].index < tab->n_redundant;
3038 /* Take a snapshot of the tableau that can be restored by s call to
3041 struct isl_tab_undo *isl_tab_snap(struct isl_tab *tab)
3049 /* Undo the operation performed by isl_tab_relax.
3051 static int unrelax(struct isl_tab *tab, struct isl_tab_var *var) WARN_UNUSED;
3052 static int unrelax(struct isl_tab *tab, struct isl_tab_var *var)
3054 unsigned off = 2 + tab->M;
3056 if (!var->is_row && !max_is_manifestly_unbounded(tab, var))
3057 if (to_row(tab, var, 1) < 0)
3061 isl_int_sub(tab->mat->row[var->index][1],
3062 tab->mat->row[var->index][1], tab->mat->row[var->index][0]);
3063 if (var->is_nonneg) {
3064 int sgn = restore_row(tab, var);
3065 isl_assert(tab->mat->ctx, sgn >= 0, return -1);
3070 for (i = 0; i < tab->n_row; ++i) {
3071 if (isl_int_is_zero(tab->mat->row[i][off + var->index]))
3073 isl_int_add(tab->mat->row[i][1], tab->mat->row[i][1],
3074 tab->mat->row[i][off + var->index]);
3082 /* Undo the operation performed by isl_tab_unrestrict.
3084 * In particular, mark the variable as being non-negative and make
3085 * sure the sample value respects this constraint.
3087 static int ununrestrict(struct isl_tab *tab, struct isl_tab_var *var)
3091 if (var->is_row && restore_row(tab, var) < -1)
3097 static int perform_undo_var(struct isl_tab *tab, struct isl_tab_undo *undo) WARN_UNUSED;
3098 static int perform_undo_var(struct isl_tab *tab, struct isl_tab_undo *undo)
3100 struct isl_tab_var *var = var_from_index(tab, undo->u.var_index);
3101 switch (undo->type) {
3102 case isl_tab_undo_nonneg:
3105 case isl_tab_undo_redundant:
3106 var->is_redundant = 0;
3108 restore_row(tab, isl_tab_var_from_row(tab, tab->n_redundant));
3110 case isl_tab_undo_freeze:
3113 case isl_tab_undo_zero:
3118 case isl_tab_undo_allocate:
3119 if (undo->u.var_index >= 0) {
3120 isl_assert(tab->mat->ctx, !var->is_row, return -1);
3121 drop_col(tab, var->index);
3125 if (!max_is_manifestly_unbounded(tab, var)) {
3126 if (to_row(tab, var, 1) < 0)
3128 } else if (!min_is_manifestly_unbounded(tab, var)) {
3129 if (to_row(tab, var, -1) < 0)
3132 if (to_row(tab, var, 0) < 0)
3135 drop_row(tab, var->index);
3137 case isl_tab_undo_relax:
3138 return unrelax(tab, var);
3139 case isl_tab_undo_unrestrict:
3140 return ununrestrict(tab, var);
3142 isl_die(tab->mat->ctx, isl_error_internal,
3143 "perform_undo_var called on invalid undo record",
3150 /* Restore the tableau to the state where the basic variables
3151 * are those in "col_var".
3152 * We first construct a list of variables that are currently in
3153 * the basis, but shouldn't. Then we iterate over all variables
3154 * that should be in the basis and for each one that is currently
3155 * not in the basis, we exchange it with one of the elements of the
3156 * list constructed before.
3157 * We can always find an appropriate variable to pivot with because
3158 * the current basis is mapped to the old basis by a non-singular
3159 * matrix and so we can never end up with a zero row.
3161 static int restore_basis(struct isl_tab *tab, int *col_var)
3165 int *extra = NULL; /* current columns that contain bad stuff */
3166 unsigned off = 2 + tab->M;
3168 extra = isl_alloc_array(tab->mat->ctx, int, tab->n_col);
3171 for (i = 0; i < tab->n_col; ++i) {
3172 for (j = 0; j < tab->n_col; ++j)
3173 if (tab->col_var[i] == col_var[j])
3177 extra[n_extra++] = i;
3179 for (i = 0; i < tab->n_col && n_extra > 0; ++i) {
3180 struct isl_tab_var *var;
3183 for (j = 0; j < tab->n_col; ++j)
3184 if (col_var[i] == tab->col_var[j])
3188 var = var_from_index(tab, col_var[i]);
3190 for (j = 0; j < n_extra; ++j)
3191 if (!isl_int_is_zero(tab->mat->row[row][off+extra[j]]))
3193 isl_assert(tab->mat->ctx, j < n_extra, goto error);
3194 if (isl_tab_pivot(tab, row, extra[j]) < 0)
3196 extra[j] = extra[--n_extra];
3206 /* Remove all samples with index n or greater, i.e., those samples
3207 * that were added since we saved this number of samples in
3208 * isl_tab_save_samples.
3210 static void drop_samples_since(struct isl_tab *tab, int n)
3214 for (i = tab->n_sample - 1; i >= 0 && tab->n_sample > n; --i) {
3215 if (tab->sample_index[i] < n)
3218 if (i != tab->n_sample - 1) {
3219 int t = tab->sample_index[tab->n_sample-1];
3220 tab->sample_index[tab->n_sample-1] = tab->sample_index[i];
3221 tab->sample_index[i] = t;
3222 isl_mat_swap_rows(tab->samples, tab->n_sample-1, i);
3228 static int perform_undo(struct isl_tab *tab, struct isl_tab_undo *undo) WARN_UNUSED;
3229 static int perform_undo(struct isl_tab *tab, struct isl_tab_undo *undo)
3231 switch (undo->type) {
3232 case isl_tab_undo_empty:
3235 case isl_tab_undo_nonneg:
3236 case isl_tab_undo_redundant:
3237 case isl_tab_undo_freeze:
3238 case isl_tab_undo_zero:
3239 case isl_tab_undo_allocate:
3240 case isl_tab_undo_relax:
3241 case isl_tab_undo_unrestrict:
3242 return perform_undo_var(tab, undo);
3243 case isl_tab_undo_bmap_eq:
3244 return isl_basic_map_free_equality(tab->bmap, 1);
3245 case isl_tab_undo_bmap_ineq:
3246 return isl_basic_map_free_inequality(tab->bmap, 1);
3247 case isl_tab_undo_bmap_div:
3248 if (isl_basic_map_free_div(tab->bmap, 1) < 0)
3251 tab->samples->n_col--;
3253 case isl_tab_undo_saved_basis:
3254 if (restore_basis(tab, undo->u.col_var) < 0)
3257 case isl_tab_undo_drop_sample:
3260 case isl_tab_undo_saved_samples:
3261 drop_samples_since(tab, undo->u.n);
3263 case isl_tab_undo_callback:
3264 return undo->u.callback->run(undo->u.callback);
3266 isl_assert(tab->mat->ctx, 0, return -1);
3271 /* Return the tableau to the state it was in when the snapshot "snap"
3274 int isl_tab_rollback(struct isl_tab *tab, struct isl_tab_undo *snap)
3276 struct isl_tab_undo *undo, *next;
3282 for (undo = tab->top; undo && undo != &tab->bottom; undo = next) {
3286 if (perform_undo(tab, undo) < 0) {
3292 free_undo_record(undo);
3301 /* The given row "row" represents an inequality violated by all
3302 * points in the tableau. Check for some special cases of such
3303 * separating constraints.
3304 * In particular, if the row has been reduced to the constant -1,
3305 * then we know the inequality is adjacent (but opposite) to
3306 * an equality in the tableau.
3307 * If the row has been reduced to r = c*(-1 -r'), with r' an inequality
3308 * of the tableau and c a positive constant, then the inequality
3309 * is adjacent (but opposite) to the inequality r'.
3311 static enum isl_ineq_type separation_type(struct isl_tab *tab, unsigned row)
3314 unsigned off = 2 + tab->M;
3317 return isl_ineq_separate;
3319 if (!isl_int_is_one(tab->mat->row[row][0]))
3320 return isl_ineq_separate;
3322 pos = isl_seq_first_non_zero(tab->mat->row[row] + off + tab->n_dead,
3323 tab->n_col - tab->n_dead);
3325 if (isl_int_is_negone(tab->mat->row[row][1]))
3326 return isl_ineq_adj_eq;
3328 return isl_ineq_separate;
3331 if (!isl_int_eq(tab->mat->row[row][1],
3332 tab->mat->row[row][off + tab->n_dead + pos]))
3333 return isl_ineq_separate;
3335 pos = isl_seq_first_non_zero(
3336 tab->mat->row[row] + off + tab->n_dead + pos + 1,
3337 tab->n_col - tab->n_dead - pos - 1);
3339 return pos == -1 ? isl_ineq_adj_ineq : isl_ineq_separate;
3342 /* Check the effect of inequality "ineq" on the tableau "tab".
3344 * isl_ineq_redundant: satisfied by all points in the tableau
3345 * isl_ineq_separate: satisfied by no point in the tableau
3346 * isl_ineq_cut: satisfied by some by not all points
3347 * isl_ineq_adj_eq: adjacent to an equality
3348 * isl_ineq_adj_ineq: adjacent to an inequality.
3350 enum isl_ineq_type isl_tab_ineq_type(struct isl_tab *tab, isl_int *ineq)
3352 enum isl_ineq_type type = isl_ineq_error;
3353 struct isl_tab_undo *snap = NULL;
3358 return isl_ineq_error;
3360 if (isl_tab_extend_cons(tab, 1) < 0)
3361 return isl_ineq_error;
3363 snap = isl_tab_snap(tab);
3365 con = isl_tab_add_row(tab, ineq);
3369 row = tab->con[con].index;
3370 if (isl_tab_row_is_redundant(tab, row))
3371 type = isl_ineq_redundant;
3372 else if (isl_int_is_neg(tab->mat->row[row][1]) &&
3374 isl_int_abs_ge(tab->mat->row[row][1],
3375 tab->mat->row[row][0]))) {
3376 int nonneg = at_least_zero(tab, &tab->con[con]);
3380 type = isl_ineq_cut;
3382 type = separation_type(tab, row);
3384 int red = con_is_redundant(tab, &tab->con[con]);
3388 type = isl_ineq_cut;
3390 type = isl_ineq_redundant;
3393 if (isl_tab_rollback(tab, snap))
3394 return isl_ineq_error;
3397 return isl_ineq_error;
3400 int isl_tab_track_bmap(struct isl_tab *tab, __isl_take isl_basic_map *bmap)
3402 bmap = isl_basic_map_cow(bmap);
3407 bmap = isl_basic_map_set_to_empty(bmap);
3414 isl_assert(tab->mat->ctx, tab->n_eq == bmap->n_eq, goto error);
3415 isl_assert(tab->mat->ctx,
3416 tab->n_con == bmap->n_eq + bmap->n_ineq, goto error);
3422 isl_basic_map_free(bmap);
3426 int isl_tab_track_bset(struct isl_tab *tab, __isl_take isl_basic_set *bset)
3428 return isl_tab_track_bmap(tab, (isl_basic_map *)bset);
3431 __isl_keep isl_basic_set *isl_tab_peek_bset(struct isl_tab *tab)
3436 return (isl_basic_set *)tab->bmap;
3439 static void isl_tab_print_internal(__isl_keep struct isl_tab *tab,
3440 FILE *out, int indent)
3446 fprintf(out, "%*snull tab\n", indent, "");
3449 fprintf(out, "%*sn_redundant: %d, n_dead: %d", indent, "",
3450 tab->n_redundant, tab->n_dead);
3452 fprintf(out, ", rational");
3454 fprintf(out, ", empty");
3456 fprintf(out, "%*s[", indent, "");
3457 for (i = 0; i < tab->n_var; ++i) {
3459 fprintf(out, (i == tab->n_param ||
3460 i == tab->n_var - tab->n_div) ? "; "
3462 fprintf(out, "%c%d%s", tab->var[i].is_row ? 'r' : 'c',
3464 tab->var[i].is_zero ? " [=0]" :
3465 tab->var[i].is_redundant ? " [R]" : "");
3467 fprintf(out, "]\n");
3468 fprintf(out, "%*s[", indent, "");
3469 for (i = 0; i < tab->n_con; ++i) {
3472 fprintf(out, "%c%d%s", tab->con[i].is_row ? 'r' : 'c',
3474 tab->con[i].is_zero ? " [=0]" :
3475 tab->con[i].is_redundant ? " [R]" : "");
3477 fprintf(out, "]\n");
3478 fprintf(out, "%*s[", indent, "");
3479 for (i = 0; i < tab->n_row; ++i) {
3480 const char *sign = "";
3483 if (tab->row_sign) {
3484 if (tab->row_sign[i] == isl_tab_row_unknown)
3486 else if (tab->row_sign[i] == isl_tab_row_neg)
3488 else if (tab->row_sign[i] == isl_tab_row_pos)
3493 fprintf(out, "r%d: %d%s%s", i, tab->row_var[i],
3494 isl_tab_var_from_row(tab, i)->is_nonneg ? " [>=0]" : "", sign);
3496 fprintf(out, "]\n");
3497 fprintf(out, "%*s[", indent, "");
3498 for (i = 0; i < tab->n_col; ++i) {
3501 fprintf(out, "c%d: %d%s", i, tab->col_var[i],
3502 var_from_col(tab, i)->is_nonneg ? " [>=0]" : "");
3504 fprintf(out, "]\n");
3505 r = tab->mat->n_row;
3506 tab->mat->n_row = tab->n_row;
3507 c = tab->mat->n_col;
3508 tab->mat->n_col = 2 + tab->M + tab->n_col;
3509 isl_mat_print_internal(tab->mat, out, indent);
3510 tab->mat->n_row = r;
3511 tab->mat->n_col = c;
3513 isl_basic_map_print_internal(tab->bmap, out, indent);
3516 void isl_tab_dump(__isl_keep struct isl_tab *tab)
3518 isl_tab_print_internal(tab, stderr, 0);