Home | History | Annotate | Line # | Download | only in dist
      1 /*
      2  * Copyright 2008-2009 Katholieke Universiteit Leuven
      3  *
      4  * Use of this software is governed by the MIT license
      5  *
      6  * Written by Sven Verdoolaege, K.U.Leuven, Departement
      7  * Computerwetenschappen, Celestijnenlaan 200A, B-3001 Leuven, Belgium
      8  */
      9 
     10 #include <isl_ctx_private.h>
     11 #include <isl_map_private.h>
     12 #include "isl_sample.h"
     13 #include <isl/vec.h>
     14 #include <isl/mat.h>
     15 #include <isl_seq.h>
     16 #include "isl_equalities.h"
     17 #include "isl_tab.h"
     18 #include "isl_basis_reduction.h"
     19 #include <isl_factorization.h>
     20 #include <isl_point_private.h>
     21 #include <isl_options_private.h>
     22 #include <isl_vec_private.h>
     23 
     24 #include <bset_from_bmap.c>
     25 #include <set_to_map.c>
     26 
     27 static __isl_give isl_vec *isl_basic_set_sample_bounded(
     28 	__isl_take isl_basic_set *bset);
     29 
     30 static __isl_give isl_vec *empty_sample(__isl_take isl_basic_set *bset)
     31 {
     32 	struct isl_vec *vec;
     33 
     34 	vec = isl_vec_alloc(bset->ctx, 0);
     35 	isl_basic_set_free(bset);
     36 	return vec;
     37 }
     38 
     39 /* Construct a zero sample of the same dimension as bset.
     40  * As a special case, if bset is zero-dimensional, this
     41  * function creates a zero-dimensional sample point.
     42  */
     43 static __isl_give isl_vec *zero_sample(__isl_take isl_basic_set *bset)
     44 {
     45 	isl_size dim;
     46 	struct isl_vec *sample;
     47 
     48 	dim = isl_basic_set_dim(bset, isl_dim_all);
     49 	if (dim < 0)
     50 		goto error;
     51 	sample = isl_vec_alloc(bset->ctx, 1 + dim);
     52 	if (sample) {
     53 		isl_int_set_si(sample->el[0], 1);
     54 		isl_seq_clr(sample->el + 1, dim);
     55 	}
     56 	isl_basic_set_free(bset);
     57 	return sample;
     58 error:
     59 	isl_basic_set_free(bset);
     60 	return NULL;
     61 }
     62 
     63 static __isl_give isl_vec *interval_sample(__isl_take isl_basic_set *bset)
     64 {
     65 	int i;
     66 	isl_int t;
     67 	struct isl_vec *sample;
     68 
     69 	bset = isl_basic_set_simplify(bset);
     70 	if (!bset)
     71 		return NULL;
     72 	if (isl_basic_set_plain_is_empty(bset))
     73 		return empty_sample(bset);
     74 	if (bset->n_eq == 0 && bset->n_ineq == 0)
     75 		return zero_sample(bset);
     76 
     77 	sample = isl_vec_alloc(bset->ctx, 2);
     78 	if (!sample)
     79 		goto error;
     80 	if (!bset)
     81 		return NULL;
     82 	isl_int_set_si(sample->block.data[0], 1);
     83 
     84 	if (bset->n_eq > 0) {
     85 		isl_assert(bset->ctx, bset->n_eq == 1, goto error);
     86 		isl_assert(bset->ctx, bset->n_ineq == 0, goto error);
     87 		if (isl_int_is_one(bset->eq[0][1]))
     88 			isl_int_neg(sample->el[1], bset->eq[0][0]);
     89 		else {
     90 			isl_assert(bset->ctx, isl_int_is_negone(bset->eq[0][1]),
     91 				   goto error);
     92 			isl_int_set(sample->el[1], bset->eq[0][0]);
     93 		}
     94 		isl_basic_set_free(bset);
     95 		return sample;
     96 	}
     97 
     98 	isl_int_init(t);
     99 	if (isl_int_is_one(bset->ineq[0][1]))
    100 		isl_int_neg(sample->block.data[1], bset->ineq[0][0]);
    101 	else
    102 		isl_int_set(sample->block.data[1], bset->ineq[0][0]);
    103 	for (i = 1; i < bset->n_ineq; ++i) {
    104 		isl_seq_inner_product(sample->block.data,
    105 					bset->ineq[i], 2, &t);
    106 		if (isl_int_is_neg(t))
    107 			break;
    108 	}
    109 	isl_int_clear(t);
    110 	if (i < bset->n_ineq) {
    111 		isl_vec_free(sample);
    112 		return empty_sample(bset);
    113 	}
    114 
    115 	isl_basic_set_free(bset);
    116 	return sample;
    117 error:
    118 	isl_basic_set_free(bset);
    119 	isl_vec_free(sample);
    120 	return NULL;
    121 }
    122 
    123 /* Find a sample integer point, if any, in bset, which is known
    124  * to have equalities.  If bset contains no integer points, then
    125  * return a zero-length vector.
    126  * We simply remove the known equalities, compute a sample
    127  * in the resulting bset, using the specified recurse function,
    128  * and then transform the sample back to the original space.
    129  */
    130 static __isl_give isl_vec *sample_eq(__isl_take isl_basic_set *bset,
    131 	__isl_give isl_vec *(*recurse)(__isl_take isl_basic_set *))
    132 {
    133 	struct isl_mat *T;
    134 	struct isl_vec *sample;
    135 
    136 	if (!bset)
    137 		return NULL;
    138 
    139 	bset = isl_basic_set_remove_equalities(bset, &T, NULL);
    140 	sample = recurse(bset);
    141 	if (!sample || sample->size == 0)
    142 		isl_mat_free(T);
    143 	else
    144 		sample = isl_mat_vec_product(T, sample);
    145 	return sample;
    146 }
    147 
    148 /* Return a matrix containing the equalities of the tableau
    149  * in constraint form.  The tableau is assumed to have
    150  * an associated bset that has been kept up-to-date.
    151  */
    152 static struct isl_mat *tab_equalities(struct isl_tab *tab)
    153 {
    154 	int i, j;
    155 	int n_eq;
    156 	struct isl_mat *eq;
    157 	struct isl_basic_set *bset;
    158 
    159 	if (!tab)
    160 		return NULL;
    161 
    162 	bset = isl_tab_peek_bset(tab);
    163 	isl_assert(tab->mat->ctx, bset, return NULL);
    164 
    165 	n_eq = tab->n_var - tab->n_col + tab->n_dead;
    166 	if (tab->empty || n_eq == 0)
    167 		return isl_mat_alloc(tab->mat->ctx, 0, tab->n_var);
    168 	if (n_eq == tab->n_var)
    169 		return isl_mat_identity(tab->mat->ctx, tab->n_var);
    170 
    171 	eq = isl_mat_alloc(tab->mat->ctx, n_eq, tab->n_var);
    172 	if (!eq)
    173 		return NULL;
    174 	for (i = 0, j = 0; i < tab->n_con; ++i) {
    175 		if (tab->con[i].is_row)
    176 			continue;
    177 		if (tab->con[i].index >= 0 && tab->con[i].index >= tab->n_dead)
    178 			continue;
    179 		if (i < bset->n_eq)
    180 			isl_seq_cpy(eq->row[j], bset->eq[i] + 1, tab->n_var);
    181 		else
    182 			isl_seq_cpy(eq->row[j],
    183 				    bset->ineq[i - bset->n_eq] + 1, tab->n_var);
    184 		++j;
    185 	}
    186 	isl_assert(bset->ctx, j == n_eq, goto error);
    187 	return eq;
    188 error:
    189 	isl_mat_free(eq);
    190 	return NULL;
    191 }
    192 
    193 /* Compute and return an initial basis for the bounded tableau "tab".
    194  *
    195  * If the tableau is either full-dimensional or zero-dimensional,
    196  * the we simply return an identity matrix.
    197  * Otherwise, we construct a basis whose first directions correspond
    198  * to equalities.
    199  */
    200 static struct isl_mat *initial_basis(struct isl_tab *tab)
    201 {
    202 	int n_eq;
    203 	struct isl_mat *eq;
    204 	struct isl_mat *Q;
    205 
    206 	tab->n_unbounded = 0;
    207 	tab->n_zero = n_eq = tab->n_var - tab->n_col + tab->n_dead;
    208 	if (tab->empty || n_eq == 0 || n_eq == tab->n_var)
    209 		return isl_mat_identity(tab->mat->ctx, 1 + tab->n_var);
    210 
    211 	eq = tab_equalities(tab);
    212 	eq = isl_mat_left_hermite(eq, 0, NULL, &Q);
    213 	if (!eq)
    214 		return NULL;
    215 	isl_mat_free(eq);
    216 
    217 	Q = isl_mat_lin_to_aff(Q);
    218 	return Q;
    219 }
    220 
    221 /* Compute the minimum of the current ("level") basis row over "tab"
    222  * and store the result in position "level" of "min".
    223  *
    224  * This function assumes that at least one more row and at least
    225  * one more element in the constraint array are available in the tableau.
    226  */
    227 static enum isl_lp_result compute_min(isl_ctx *ctx, struct isl_tab *tab,
    228 	__isl_keep isl_vec *min, int level)
    229 {
    230 	return isl_tab_min(tab, tab->basis->row[1 + level],
    231 			    ctx->one, &min->el[level], NULL, 0);
    232 }
    233 
    234 /* Compute the maximum of the current ("level") basis row over "tab"
    235  * and store the result in position "level" of "max".
    236  *
    237  * This function assumes that at least one more row and at least
    238  * one more element in the constraint array are available in the tableau.
    239  */
    240 static enum isl_lp_result compute_max(isl_ctx *ctx, struct isl_tab *tab,
    241 	__isl_keep isl_vec *max, int level)
    242 {
    243 	enum isl_lp_result res;
    244 	unsigned dim = tab->n_var;
    245 
    246 	isl_seq_neg(tab->basis->row[1 + level] + 1,
    247 		    tab->basis->row[1 + level] + 1, dim);
    248 	res = isl_tab_min(tab, tab->basis->row[1 + level],
    249 		    ctx->one, &max->el[level], NULL, 0);
    250 	isl_seq_neg(tab->basis->row[1 + level] + 1,
    251 		    tab->basis->row[1 + level] + 1, dim);
    252 	isl_int_neg(max->el[level], max->el[level]);
    253 
    254 	return res;
    255 }
    256 
    257 /* Perform a greedy search for an integer point in the set represented
    258  * by "tab", given that the minimal rational value (rounded up to the
    259  * nearest integer) at "level" is smaller than the maximal rational
    260  * value (rounded down to the nearest integer).
    261  *
    262  * Return 1 if we have found an integer point (if tab->n_unbounded > 0
    263  * then we may have only found integer values for the bounded dimensions
    264  * and it is the responsibility of the caller to extend this solution
    265  * to the unbounded dimensions).
    266  * Return 0 if greedy search did not result in a solution.
    267  * Return -1 if some error occurred.
    268  *
    269  * We assign a value half-way between the minimum and the maximum
    270  * to the current dimension and check if the minimal value of the
    271  * next dimension is still smaller than (or equal) to the maximal value.
    272  * We continue this process until either
    273  * - the minimal value (rounded up) is greater than the maximal value
    274  *	(rounded down).  In this case, greedy search has failed.
    275  * - we have exhausted all bounded dimensions, meaning that we have
    276  *	found a solution.
    277  * - the sample value of the tableau is integral.
    278  * - some error has occurred.
    279  */
    280 static int greedy_search(isl_ctx *ctx, struct isl_tab *tab,
    281 	__isl_keep isl_vec *min, __isl_keep isl_vec *max, int level)
    282 {
    283 	struct isl_tab_undo *snap;
    284 	enum isl_lp_result res;
    285 
    286 	snap = isl_tab_snap(tab);
    287 
    288 	do {
    289 		isl_int_add(tab->basis->row[1 + level][0],
    290 			    min->el[level], max->el[level]);
    291 		isl_int_fdiv_q_ui(tab->basis->row[1 + level][0],
    292 			    tab->basis->row[1 + level][0], 2);
    293 		isl_int_neg(tab->basis->row[1 + level][0],
    294 			    tab->basis->row[1 + level][0]);
    295 		if (isl_tab_add_valid_eq(tab, tab->basis->row[1 + level]) < 0)
    296 			return -1;
    297 		isl_int_set_si(tab->basis->row[1 + level][0], 0);
    298 
    299 		if (++level >= tab->n_var - tab->n_unbounded)
    300 			return 1;
    301 		if (isl_tab_sample_is_integer(tab))
    302 			return 1;
    303 
    304 		res = compute_min(ctx, tab, min, level);
    305 		if (res == isl_lp_error)
    306 			return -1;
    307 		if (res != isl_lp_ok)
    308 			isl_die(ctx, isl_error_internal,
    309 				"expecting bounded rational solution",
    310 				return -1);
    311 		res = compute_max(ctx, tab, max, level);
    312 		if (res == isl_lp_error)
    313 			return -1;
    314 		if (res != isl_lp_ok)
    315 			isl_die(ctx, isl_error_internal,
    316 				"expecting bounded rational solution",
    317 				return -1);
    318 	} while (isl_int_le(min->el[level], max->el[level]));
    319 
    320 	if (isl_tab_rollback(tab, snap) < 0)
    321 		return -1;
    322 
    323 	return 0;
    324 }
    325 
    326 /* Given a tableau representing a set, find and return
    327  * an integer point in the set, if there is any.
    328  *
    329  * We perform a depth first search
    330  * for an integer point, by scanning all possible values in the range
    331  * attained by a basis vector, where an initial basis may have been set
    332  * by the calling function.  Otherwise an initial basis that exploits
    333  * the equalities in the tableau is created.
    334  * tab->n_zero is currently ignored and is clobbered by this function.
    335  *
    336  * The tableau is allowed to have unbounded direction, but then
    337  * the calling function needs to set an initial basis, with the
    338  * unbounded directions last and with tab->n_unbounded set
    339  * to the number of unbounded directions.
    340  * Furthermore, the calling functions needs to add shifted copies
    341  * of all constraints involving unbounded directions to ensure
    342  * that any feasible rational value in these directions can be rounded
    343  * up to yield a feasible integer value.
    344  * In particular, let B define the given basis x' = B x
    345  * and let T be the inverse of B, i.e., X = T x'.
    346  * Let a x + c >= 0 be a constraint of the set represented by the tableau,
    347  * or a T x' + c >= 0 in terms of the given basis.  Assume that
    348  * the bounded directions have an integer value, then we can safely
    349  * round up the values for the unbounded directions if we make sure
    350  * that x' not only satisfies the original constraint, but also
    351  * the constraint "a T x' + c + s >= 0" with s the sum of all
    352  * negative values in the last n_unbounded entries of "a T".
    353  * The calling function therefore needs to add the constraint
    354  * a x + c + s >= 0.  The current function then scans the first
    355  * directions for an integer value and once those have been found,
    356  * it can compute "T ceil(B x)" to yield an integer point in the set.
    357  * Note that during the search, the first rows of B may be changed
    358  * by a basis reduction, but the last n_unbounded rows of B remain
    359  * unaltered and are also not mixed into the first rows.
    360  *
    361  * The search is implemented iteratively.  "level" identifies the current
    362  * basis vector.  "init" is true if we want the first value at the current
    363  * level and false if we want the next value.
    364  *
    365  * At the start of each level, we first check if we can find a solution
    366  * using greedy search.  If not, we continue with the exhaustive search.
    367  *
    368  * The initial basis is the identity matrix.  If the range in some direction
    369  * contains more than one integer value, we perform basis reduction based
    370  * on the value of ctx->opt->gbr
    371  *	- ISL_GBR_NEVER:	never perform basis reduction
    372  *	- ISL_GBR_ONCE:		only perform basis reduction the first
    373  *				time such a range is encountered
    374  *	- ISL_GBR_ALWAYS:	always perform basis reduction when
    375  *				such a range is encountered
    376  *
    377  * When ctx->opt->gbr is set to ISL_GBR_ALWAYS, then we allow the basis
    378  * reduction computation to return early.  That is, as soon as it
    379  * finds a reasonable first direction.
    380  */
    381 __isl_give isl_vec *isl_tab_sample(struct isl_tab *tab)
    382 {
    383 	unsigned dim;
    384 	unsigned gbr;
    385 	struct isl_ctx *ctx;
    386 	struct isl_vec *sample;
    387 	struct isl_vec *min;
    388 	struct isl_vec *max;
    389 	enum isl_lp_result res;
    390 	int level;
    391 	int init;
    392 	int reduced;
    393 	struct isl_tab_undo **snap;
    394 
    395 	if (!tab)
    396 		return NULL;
    397 	if (tab->empty)
    398 		return isl_vec_alloc(tab->mat->ctx, 0);
    399 
    400 	if (!tab->basis)
    401 		tab->basis = initial_basis(tab);
    402 	if (!tab->basis)
    403 		return NULL;
    404 	isl_assert(tab->mat->ctx, tab->basis->n_row == tab->n_var + 1,
    405 		    return NULL);
    406 	isl_assert(tab->mat->ctx, tab->basis->n_col == tab->n_var + 1,
    407 		    return NULL);
    408 
    409 	ctx = tab->mat->ctx;
    410 	dim = tab->n_var;
    411 	gbr = ctx->opt->gbr;
    412 
    413 	if (tab->n_unbounded == tab->n_var) {
    414 		sample = isl_tab_get_sample_value(tab);
    415 		sample = isl_mat_vec_product(isl_mat_copy(tab->basis), sample);
    416 		sample = isl_vec_ceil(sample);
    417 		sample = isl_mat_vec_inverse_product(isl_mat_copy(tab->basis),
    418 							sample);
    419 		return sample;
    420 	}
    421 
    422 	if (isl_tab_extend_cons(tab, dim + 1) < 0)
    423 		return NULL;
    424 
    425 	min = isl_vec_alloc(ctx, dim);
    426 	max = isl_vec_alloc(ctx, dim);
    427 	snap = isl_alloc_array(ctx, struct isl_tab_undo *, dim);
    428 
    429 	if (!min || !max || !snap)
    430 		goto error;
    431 
    432 	level = 0;
    433 	init = 1;
    434 	reduced = 0;
    435 
    436 	while (level >= 0) {
    437 		if (init) {
    438 			int choice;
    439 
    440 			res = compute_min(ctx, tab, min, level);
    441 			if (res == isl_lp_error)
    442 				goto error;
    443 			if (res != isl_lp_ok)
    444 				isl_die(ctx, isl_error_internal,
    445 					"expecting bounded rational solution",
    446 					goto error);
    447 			if (isl_tab_sample_is_integer(tab))
    448 				break;
    449 			res = compute_max(ctx, tab, max, level);
    450 			if (res == isl_lp_error)
    451 				goto error;
    452 			if (res != isl_lp_ok)
    453 				isl_die(ctx, isl_error_internal,
    454 					"expecting bounded rational solution",
    455 					goto error);
    456 			if (isl_tab_sample_is_integer(tab))
    457 				break;
    458 			choice = isl_int_lt(min->el[level], max->el[level]);
    459 			if (choice) {
    460 				int g;
    461 				g = greedy_search(ctx, tab, min, max, level);
    462 				if (g < 0)
    463 					goto error;
    464 				if (g)
    465 					break;
    466 			}
    467 			if (!reduced && choice &&
    468 			    ctx->opt->gbr != ISL_GBR_NEVER) {
    469 				unsigned gbr_only_first;
    470 				if (ctx->opt->gbr == ISL_GBR_ONCE)
    471 					ctx->opt->gbr = ISL_GBR_NEVER;
    472 				tab->n_zero = level;
    473 				gbr_only_first = ctx->opt->gbr_only_first;
    474 				ctx->opt->gbr_only_first =
    475 					ctx->opt->gbr == ISL_GBR_ALWAYS;
    476 				tab = isl_tab_compute_reduced_basis(tab);
    477 				ctx->opt->gbr_only_first = gbr_only_first;
    478 				if (!tab || !tab->basis)
    479 					goto error;
    480 				reduced = 1;
    481 				continue;
    482 			}
    483 			reduced = 0;
    484 			snap[level] = isl_tab_snap(tab);
    485 		} else
    486 			isl_int_add_ui(min->el[level], min->el[level], 1);
    487 
    488 		if (isl_int_gt(min->el[level], max->el[level])) {
    489 			level--;
    490 			init = 0;
    491 			if (level >= 0)
    492 				if (isl_tab_rollback(tab, snap[level]) < 0)
    493 					goto error;
    494 			continue;
    495 		}
    496 		isl_int_neg(tab->basis->row[1 + level][0], min->el[level]);
    497 		if (isl_tab_add_valid_eq(tab, tab->basis->row[1 + level]) < 0)
    498 			goto error;
    499 		isl_int_set_si(tab->basis->row[1 + level][0], 0);
    500 		if (level + tab->n_unbounded < dim - 1) {
    501 			++level;
    502 			init = 1;
    503 			continue;
    504 		}
    505 		break;
    506 	}
    507 
    508 	if (level >= 0) {
    509 		sample = isl_tab_get_sample_value(tab);
    510 		if (!sample)
    511 			goto error;
    512 		if (tab->n_unbounded && !isl_int_is_one(sample->el[0])) {
    513 			sample = isl_mat_vec_product(isl_mat_copy(tab->basis),
    514 						     sample);
    515 			sample = isl_vec_ceil(sample);
    516 			sample = isl_mat_vec_inverse_product(
    517 					isl_mat_copy(tab->basis), sample);
    518 		}
    519 	} else
    520 		sample = isl_vec_alloc(ctx, 0);
    521 
    522 	ctx->opt->gbr = gbr;
    523 	isl_vec_free(min);
    524 	isl_vec_free(max);
    525 	free(snap);
    526 	return sample;
    527 error:
    528 	ctx->opt->gbr = gbr;
    529 	isl_vec_free(min);
    530 	isl_vec_free(max);
    531 	free(snap);
    532 	return NULL;
    533 }
    534 
    535 static __isl_give isl_vec *sample_bounded(__isl_take isl_basic_set *bset);
    536 
    537 /* Internal data for factored_sample.
    538  * "sample" collects the sample and may get reset to a zero-length vector
    539  * signaling the absence of a sample vector.
    540  * "pos" is the position of the contribution of the next factor.
    541  */
    542 struct isl_factored_sample_data {
    543 	isl_vec *sample;
    544 	int pos;
    545 };
    546 
    547 /* isl_factorizer_every_factor_basic_set callback that extends
    548  * the sample in data->sample with the contribution
    549  * of the factor "bset".
    550  * If "bset" turns out to be empty, then the product is empty too and
    551  * no further factors need to be considered.
    552  */
    553 static isl_bool factor_sample(__isl_keep isl_basic_set *bset, void *user)
    554 {
    555 	struct isl_factored_sample_data *data = user;
    556 	isl_vec *sample;
    557 	isl_size n;
    558 
    559 	n = isl_basic_set_dim(bset, isl_dim_set);
    560 	if (n < 0)
    561 		return isl_bool_error;
    562 
    563 	sample = sample_bounded(isl_basic_set_copy(bset));
    564 	if (!sample)
    565 		return isl_bool_error;
    566 	if (sample->size == 0) {
    567 		isl_vec_free(data->sample);
    568 		data->sample = sample;
    569 		return isl_bool_false;
    570 	}
    571 	isl_seq_cpy(data->sample->el + data->pos, sample->el + 1, n);
    572 	isl_vec_free(sample);
    573 	data->pos += n;
    574 
    575 	return isl_bool_true;
    576 }
    577 
    578 /* Compute a sample point of the given basic set, based on the given,
    579  * non-trivial factorization.
    580  */
    581 static __isl_give isl_vec *factored_sample(__isl_take isl_basic_set *bset,
    582 	__isl_take isl_factorizer *f)
    583 {
    584 	struct isl_factored_sample_data data = { NULL };
    585 	isl_ctx *ctx;
    586 	isl_size total;
    587 	isl_bool every;
    588 
    589 	ctx = isl_basic_set_get_ctx(bset);
    590 	total = isl_basic_set_dim(bset, isl_dim_all);
    591 	if (!ctx || total < 0)
    592 		goto error;
    593 
    594 	data.sample = isl_vec_alloc(ctx, 1 + total);
    595 	if (!data.sample)
    596 		goto error;
    597 	isl_int_set_si(data.sample->el[0], 1);
    598 	data.pos = 1;
    599 
    600 	every = isl_factorizer_every_factor_basic_set(f, &factor_sample, &data);
    601 	if (every < 0) {
    602 		data.sample = isl_vec_free(data.sample);
    603 	} else if (every) {
    604 		isl_morph *morph;
    605 
    606 		morph = isl_morph_inverse(isl_morph_copy(f->morph));
    607 		data.sample = isl_morph_vec(morph, data.sample);
    608 	}
    609 
    610 	isl_basic_set_free(bset);
    611 	isl_factorizer_free(f);
    612 	return data.sample;
    613 error:
    614 	isl_basic_set_free(bset);
    615 	isl_factorizer_free(f);
    616 	isl_vec_free(data.sample);
    617 	return NULL;
    618 }
    619 
    620 /* Given a basic set that is known to be bounded, find and return
    621  * an integer point in the basic set, if there is any.
    622  *
    623  * After handling some trivial cases, we construct a tableau
    624  * and then use isl_tab_sample to find a sample, passing it
    625  * the identity matrix as initial basis.
    626  */
    627 static __isl_give isl_vec *sample_bounded(__isl_take isl_basic_set *bset)
    628 {
    629 	isl_size dim;
    630 	struct isl_vec *sample;
    631 	struct isl_tab *tab = NULL;
    632 	isl_factorizer *f;
    633 
    634 	if (!bset)
    635 		return NULL;
    636 
    637 	if (isl_basic_set_plain_is_empty(bset))
    638 		return empty_sample(bset);
    639 
    640 	dim = isl_basic_set_dim(bset, isl_dim_all);
    641 	if (dim < 0)
    642 		bset = isl_basic_set_free(bset);
    643 	if (dim == 0)
    644 		return zero_sample(bset);
    645 	if (dim == 1)
    646 		return interval_sample(bset);
    647 	if (bset->n_eq > 0)
    648 		return sample_eq(bset, sample_bounded);
    649 
    650 	f = isl_basic_set_factorizer(bset);
    651 	if (!f)
    652 		goto error;
    653 	if (f->n_group != 0)
    654 		return factored_sample(bset, f);
    655 	isl_factorizer_free(f);
    656 
    657 	tab = isl_tab_from_basic_set(bset, 1);
    658 	if (tab && tab->empty) {
    659 		isl_tab_free(tab);
    660 		ISL_F_SET(bset, ISL_BASIC_SET_EMPTY);
    661 		sample = isl_vec_alloc(isl_basic_set_get_ctx(bset), 0);
    662 		isl_basic_set_free(bset);
    663 		return sample;
    664 	}
    665 
    666 	if (!ISL_F_ISSET(bset, ISL_BASIC_SET_NO_IMPLICIT))
    667 		if (isl_tab_detect_implicit_equalities(tab) < 0)
    668 			goto error;
    669 
    670 	sample = isl_tab_sample(tab);
    671 	if (!sample)
    672 		goto error;
    673 
    674 	if (sample->size > 0) {
    675 		isl_vec_free(bset->sample);
    676 		bset->sample = isl_vec_copy(sample);
    677 	}
    678 
    679 	isl_basic_set_free(bset);
    680 	isl_tab_free(tab);
    681 	return sample;
    682 error:
    683 	isl_basic_set_free(bset);
    684 	isl_tab_free(tab);
    685 	return NULL;
    686 }
    687 
    688 /* Given a basic set "bset" and a value "sample" for the first coordinates
    689  * of bset, plug in these values and drop the corresponding coordinates.
    690  *
    691  * We do this by computing the preimage of the transformation
    692  *
    693  *	     [ 1 0 ]
    694  *	x =  [ s 0 ] x'
    695  *	     [ 0 I ]
    696  *
    697  * where [1 s] is the sample value and I is the identity matrix of the
    698  * appropriate dimension.
    699  */
    700 static __isl_give isl_basic_set *plug_in(__isl_take isl_basic_set *bset,
    701 	__isl_take isl_vec *sample)
    702 {
    703 	int i;
    704 	isl_size total;
    705 	struct isl_mat *T;
    706 
    707 	total = isl_basic_set_dim(bset, isl_dim_all);
    708 	if (total < 0 || !sample)
    709 		goto error;
    710 
    711 	T = isl_mat_alloc(bset->ctx, 1 + total, 1 + total - (sample->size - 1));
    712 	if (!T)
    713 		goto error;
    714 
    715 	for (i = 0; i < sample->size; ++i) {
    716 		isl_int_set(T->row[i][0], sample->el[i]);
    717 		isl_seq_clr(T->row[i] + 1, T->n_col - 1);
    718 	}
    719 	for (i = 0; i < T->n_col - 1; ++i) {
    720 		isl_seq_clr(T->row[sample->size + i], T->n_col);
    721 		isl_int_set_si(T->row[sample->size + i][1 + i], 1);
    722 	}
    723 	isl_vec_free(sample);
    724 
    725 	bset = isl_basic_set_preimage(bset, T);
    726 	return bset;
    727 error:
    728 	isl_basic_set_free(bset);
    729 	isl_vec_free(sample);
    730 	return NULL;
    731 }
    732 
    733 /* Given a basic set "bset", return any (possibly non-integer) point
    734  * in the basic set.
    735  */
    736 static __isl_give isl_vec *rational_sample(__isl_take isl_basic_set *bset)
    737 {
    738 	struct isl_tab *tab;
    739 	struct isl_vec *sample;
    740 
    741 	if (!bset)
    742 		return NULL;
    743 
    744 	tab = isl_tab_from_basic_set(bset, 0);
    745 	sample = isl_tab_get_sample_value(tab);
    746 	isl_tab_free(tab);
    747 
    748 	isl_basic_set_free(bset);
    749 
    750 	return sample;
    751 }
    752 
    753 /* Given a linear cone "cone" and a rational point "vec",
    754  * construct a polyhedron with shifted copies of the constraints in "cone",
    755  * i.e., a polyhedron with "cone" as its recession cone, such that each
    756  * point x in this polyhedron is such that the unit box positioned at x
    757  * lies entirely inside the affine cone 'vec + cone'.
    758  * Any rational point in this polyhedron may therefore be rounded up
    759  * to yield an integer point that lies inside said affine cone.
    760  *
    761  * Denote the constraints of cone by "<a_i, x> >= 0" and the rational
    762  * point "vec" by v/d.
    763  * Let b_i = <a_i, v>.  Then the affine cone 'vec + cone' is given
    764  * by <a_i, x> - b/d >= 0.
    765  * The polyhedron <a_i, x> - ceil{b/d} >= 0 is a subset of this affine cone.
    766  * We prefer this polyhedron over the actual affine cone because it doesn't
    767  * require a scaling of the constraints.
    768  * If each of the vertices of the unit cube positioned at x lies inside
    769  * this polyhedron, then the whole unit cube at x lies inside the affine cone.
    770  * We therefore impose that x' = x + \sum e_i, for any selection of unit
    771  * vectors lies inside the polyhedron, i.e.,
    772  *
    773  *	<a_i, x'> - ceil{b/d} = <a_i, x> + sum a_i - ceil{b/d} >= 0
    774  *
    775  * The most stringent of these constraints is the one that selects
    776  * all negative a_i, so the polyhedron we are looking for has constraints
    777  *
    778  *	<a_i, x> + sum_{a_i < 0} a_i - ceil{b/d} >= 0
    779  *
    780  * Note that if cone were known to have only non-negative rays
    781  * (which can be accomplished by a unimodular transformation),
    782  * then we would only have to check the points x' = x + e_i
    783  * and we only have to add the smallest negative a_i (if any)
    784  * instead of the sum of all negative a_i.
    785  */
    786 static __isl_give isl_basic_set *shift_cone(__isl_take isl_basic_set *cone,
    787 	__isl_take isl_vec *vec)
    788 {
    789 	int i, j, k;
    790 	isl_size total;
    791 
    792 	struct isl_basic_set *shift = NULL;
    793 
    794 	total = isl_basic_set_dim(cone, isl_dim_all);
    795 	if (total < 0 || !vec)
    796 		goto error;
    797 
    798 	isl_assert(cone->ctx, cone->n_eq == 0, goto error);
    799 
    800 	shift = isl_basic_set_alloc_space(isl_basic_set_get_space(cone),
    801 					0, 0, cone->n_ineq);
    802 
    803 	for (i = 0; i < cone->n_ineq; ++i) {
    804 		k = isl_basic_set_alloc_inequality(shift);
    805 		if (k < 0)
    806 			goto error;
    807 		isl_seq_cpy(shift->ineq[k] + 1, cone->ineq[i] + 1, total);
    808 		isl_seq_inner_product(shift->ineq[k] + 1, vec->el + 1, total,
    809 				      &shift->ineq[k][0]);
    810 		isl_int_cdiv_q(shift->ineq[k][0],
    811 			       shift->ineq[k][0], vec->el[0]);
    812 		isl_int_neg(shift->ineq[k][0], shift->ineq[k][0]);
    813 		for (j = 0; j < total; ++j) {
    814 			if (isl_int_is_nonneg(shift->ineq[k][1 + j]))
    815 				continue;
    816 			isl_int_add(shift->ineq[k][0],
    817 				    shift->ineq[k][0], shift->ineq[k][1 + j]);
    818 		}
    819 	}
    820 
    821 	isl_basic_set_free(cone);
    822 	isl_vec_free(vec);
    823 
    824 	return isl_basic_set_finalize(shift);
    825 error:
    826 	isl_basic_set_free(shift);
    827 	isl_basic_set_free(cone);
    828 	isl_vec_free(vec);
    829 	return NULL;
    830 }
    831 
    832 /* Given a rational point vec in a (transformed) basic set,
    833  * such that cone is the recession cone of the original basic set,
    834  * "round up" the rational point to an integer point.
    835  *
    836  * We first check if the rational point just happens to be integer.
    837  * If not, we transform the cone in the same way as the basic set,
    838  * pick a point x in this cone shifted to the rational point such that
    839  * the whole unit cube at x is also inside this affine cone.
    840  * Then we simply round up the coordinates of x and return the
    841  * resulting integer point.
    842  */
    843 static __isl_give isl_vec *round_up_in_cone(__isl_take isl_vec *vec,
    844 	__isl_take isl_basic_set *cone, __isl_take isl_mat *U)
    845 {
    846 	isl_size total;
    847 
    848 	if (!vec || !cone || !U)
    849 		goto error;
    850 
    851 	isl_assert(vec->ctx, vec->size != 0, goto error);
    852 	if (isl_int_is_one(vec->el[0])) {
    853 		isl_mat_free(U);
    854 		isl_basic_set_free(cone);
    855 		return vec;
    856 	}
    857 
    858 	total = isl_basic_set_dim(cone, isl_dim_all);
    859 	if (total < 0)
    860 		goto error;
    861 	cone = isl_basic_set_preimage(cone, U);
    862 	cone = isl_basic_set_remove_dims(cone, isl_dim_set,
    863 					 0, total - (vec->size - 1));
    864 
    865 	cone = shift_cone(cone, vec);
    866 
    867 	vec = rational_sample(cone);
    868 	vec = isl_vec_ceil(vec);
    869 	return vec;
    870 error:
    871 	isl_mat_free(U);
    872 	isl_vec_free(vec);
    873 	isl_basic_set_free(cone);
    874 	return NULL;
    875 }
    876 
    877 /* Concatenate two integer vectors, i.e., two vectors with denominator
    878  * (stored in element 0) equal to 1.
    879  */
    880 static __isl_give isl_vec *vec_concat(__isl_take isl_vec *vec1,
    881 	__isl_take isl_vec *vec2)
    882 {
    883 	struct isl_vec *vec;
    884 
    885 	if (!vec1 || !vec2)
    886 		goto error;
    887 	isl_assert(vec1->ctx, vec1->size > 0, goto error);
    888 	isl_assert(vec2->ctx, vec2->size > 0, goto error);
    889 	isl_assert(vec1->ctx, isl_int_is_one(vec1->el[0]), goto error);
    890 	isl_assert(vec2->ctx, isl_int_is_one(vec2->el[0]), goto error);
    891 
    892 	vec = isl_vec_alloc(vec1->ctx, vec1->size + vec2->size - 1);
    893 	if (!vec)
    894 		goto error;
    895 
    896 	isl_seq_cpy(vec->el, vec1->el, vec1->size);
    897 	isl_seq_cpy(vec->el + vec1->size, vec2->el + 1, vec2->size - 1);
    898 
    899 	isl_vec_free(vec1);
    900 	isl_vec_free(vec2);
    901 
    902 	return vec;
    903 error:
    904 	isl_vec_free(vec1);
    905 	isl_vec_free(vec2);
    906 	return NULL;
    907 }
    908 
    909 /* Give a basic set "bset" with recession cone "cone", compute and
    910  * return an integer point in bset, if any.
    911  *
    912  * If the recession cone is full-dimensional, then we know that
    913  * bset contains an infinite number of integer points and it is
    914  * fairly easy to pick one of them.
    915  * If the recession cone is not full-dimensional, then we first
    916  * transform bset such that the bounded directions appear as
    917  * the first dimensions of the transformed basic set.
    918  * We do this by using a unimodular transformation that transforms
    919  * the equalities in the recession cone to equalities on the first
    920  * dimensions.
    921  *
    922  * The transformed set is then projected onto its bounded dimensions.
    923  * Note that to compute this projection, we can simply drop all constraints
    924  * involving any of the unbounded dimensions since these constraints
    925  * cannot be combined to produce a constraint on the bounded dimensions.
    926  * To see this, assume that there is such a combination of constraints
    927  * that produces a constraint on the bounded dimensions.  This means
    928  * that some combination of the unbounded dimensions has both an upper
    929  * bound and a lower bound in terms of the bounded dimensions, but then
    930  * this combination would be a bounded direction too and would have been
    931  * transformed into a bounded dimensions.
    932  *
    933  * We then compute a sample value in the bounded dimensions.
    934  * If no such value can be found, then the original set did not contain
    935  * any integer points and we are done.
    936  * Otherwise, we plug in the value we found in the bounded dimensions,
    937  * project out these bounded dimensions and end up with a set with
    938  * a full-dimensional recession cone.
    939  * A sample point in this set is computed by "rounding up" any
    940  * rational point in the set.
    941  *
    942  * The sample points in the bounded and unbounded dimensions are
    943  * then combined into a single sample point and transformed back
    944  * to the original space.
    945  */
    946 __isl_give isl_vec *isl_basic_set_sample_with_cone(
    947 	__isl_take isl_basic_set *bset, __isl_take isl_basic_set *cone)
    948 {
    949 	isl_size total;
    950 	unsigned cone_dim;
    951 	struct isl_mat *M, *U;
    952 	struct isl_vec *sample;
    953 	struct isl_vec *cone_sample;
    954 	struct isl_ctx *ctx;
    955 	struct isl_basic_set *bounded;
    956 
    957 	total = isl_basic_set_dim(cone, isl_dim_all);
    958 	if (!bset || total < 0)
    959 		goto error;
    960 
    961 	ctx = isl_basic_set_get_ctx(bset);
    962 	cone_dim = total - cone->n_eq;
    963 
    964 	M = isl_mat_sub_alloc6(ctx, cone->eq, 0, cone->n_eq, 1, total);
    965 	M = isl_mat_left_hermite(M, 0, &U, NULL);
    966 	if (!M)
    967 		goto error;
    968 	isl_mat_free(M);
    969 
    970 	U = isl_mat_lin_to_aff(U);
    971 	bset = isl_basic_set_preimage(bset, isl_mat_copy(U));
    972 
    973 	bounded = isl_basic_set_copy(bset);
    974 	bounded = isl_basic_set_drop_constraints_involving(bounded,
    975 						   total - cone_dim, cone_dim);
    976 	bounded = isl_basic_set_drop_dims(bounded, total - cone_dim, cone_dim);
    977 	sample = sample_bounded(bounded);
    978 	if (!sample || sample->size == 0) {
    979 		isl_basic_set_free(bset);
    980 		isl_basic_set_free(cone);
    981 		isl_mat_free(U);
    982 		return sample;
    983 	}
    984 	bset = plug_in(bset, isl_vec_copy(sample));
    985 	cone_sample = rational_sample(bset);
    986 	cone_sample = round_up_in_cone(cone_sample, cone, isl_mat_copy(U));
    987 	sample = vec_concat(sample, cone_sample);
    988 	sample = isl_mat_vec_product(U, sample);
    989 	return sample;
    990 error:
    991 	isl_basic_set_free(cone);
    992 	isl_basic_set_free(bset);
    993 	return NULL;
    994 }
    995 
    996 static void vec_sum_of_neg(__isl_keep isl_vec *v, isl_int *s)
    997 {
    998 	int i;
    999 
   1000 	isl_int_set_si(*s, 0);
   1001 
   1002 	for (i = 0; i < v->size; ++i)
   1003 		if (isl_int_is_neg(v->el[i]))
   1004 			isl_int_add(*s, *s, v->el[i]);
   1005 }
   1006 
   1007 /* Given a tableau "tab", a tableau "tab_cone" that corresponds
   1008  * to the recession cone and the inverse of a new basis U = inv(B),
   1009  * with the unbounded directions in B last,
   1010  * add constraints to "tab" that ensure any rational value
   1011  * in the unbounded directions can be rounded up to an integer value.
   1012  *
   1013  * The new basis is given by x' = B x, i.e., x = U x'.
   1014  * For any rational value of the last tab->n_unbounded coordinates
   1015  * in the update tableau, the value that is obtained by rounding
   1016  * up this value should be contained in the original tableau.
   1017  * For any constraint "a x + c >= 0", we therefore need to add
   1018  * a constraint "a x + c + s >= 0", with s the sum of all negative
   1019  * entries in the last elements of "a U".
   1020  *
   1021  * Since we are not interested in the first entries of any of the "a U",
   1022  * we first drop the columns of U that correpond to bounded directions.
   1023  */
   1024 static int tab_shift_cone(struct isl_tab *tab,
   1025 	struct isl_tab *tab_cone, struct isl_mat *U)
   1026 {
   1027 	int i;
   1028 	isl_int v;
   1029 	struct isl_basic_set *bset = NULL;
   1030 
   1031 	if (tab && tab->n_unbounded == 0) {
   1032 		isl_mat_free(U);
   1033 		return 0;
   1034 	}
   1035 	isl_int_init(v);
   1036 	if (!tab || !tab_cone || !U)
   1037 		goto error;
   1038 	bset = isl_tab_peek_bset(tab_cone);
   1039 	U = isl_mat_drop_cols(U, 0, tab->n_var - tab->n_unbounded);
   1040 	for (i = 0; i < bset->n_ineq; ++i) {
   1041 		int ok;
   1042 		struct isl_vec *row = NULL;
   1043 		if (isl_tab_is_equality(tab_cone, tab_cone->n_eq + i))
   1044 			continue;
   1045 		row = isl_vec_alloc(bset->ctx, tab_cone->n_var);
   1046 		if (!row)
   1047 			goto error;
   1048 		isl_seq_cpy(row->el, bset->ineq[i] + 1, tab_cone->n_var);
   1049 		row = isl_vec_mat_product(row, isl_mat_copy(U));
   1050 		if (!row)
   1051 			goto error;
   1052 		vec_sum_of_neg(row, &v);
   1053 		isl_vec_free(row);
   1054 		if (isl_int_is_zero(v))
   1055 			continue;
   1056 		if (isl_tab_extend_cons(tab, 1) < 0)
   1057 			goto error;
   1058 		isl_int_add(bset->ineq[i][0], bset->ineq[i][0], v);
   1059 		ok = isl_tab_add_ineq(tab, bset->ineq[i]) >= 0;
   1060 		isl_int_sub(bset->ineq[i][0], bset->ineq[i][0], v);
   1061 		if (!ok)
   1062 			goto error;
   1063 	}
   1064 
   1065 	isl_mat_free(U);
   1066 	isl_int_clear(v);
   1067 	return 0;
   1068 error:
   1069 	isl_mat_free(U);
   1070 	isl_int_clear(v);
   1071 	return -1;
   1072 }
   1073 
   1074 /* Compute and return an initial basis for the possibly
   1075  * unbounded tableau "tab".  "tab_cone" is a tableau
   1076  * for the corresponding recession cone.
   1077  * Additionally, add constraints to "tab" that ensure
   1078  * that any rational value for the unbounded directions
   1079  * can be rounded up to an integer value.
   1080  *
   1081  * If the tableau is bounded, i.e., if the recession cone
   1082  * is zero-dimensional, then we just use inital_basis.
   1083  * Otherwise, we construct a basis whose first directions
   1084  * correspond to equalities, followed by bounded directions,
   1085  * i.e., equalities in the recession cone.
   1086  * The remaining directions are then unbounded.
   1087  */
   1088 int isl_tab_set_initial_basis_with_cone(struct isl_tab *tab,
   1089 	struct isl_tab *tab_cone)
   1090 {
   1091 	struct isl_mat *eq;
   1092 	struct isl_mat *cone_eq;
   1093 	struct isl_mat *U, *Q;
   1094 
   1095 	if (!tab || !tab_cone)
   1096 		return -1;
   1097 
   1098 	if (tab_cone->n_col == tab_cone->n_dead) {
   1099 		tab->basis = initial_basis(tab);
   1100 		return tab->basis ? 0 : -1;
   1101 	}
   1102 
   1103 	eq = tab_equalities(tab);
   1104 	if (!eq)
   1105 		return -1;
   1106 	tab->n_zero = eq->n_row;
   1107 	cone_eq = tab_equalities(tab_cone);
   1108 	eq = isl_mat_concat(eq, cone_eq);
   1109 	if (!eq)
   1110 		return -1;
   1111 	tab->n_unbounded = tab->n_var - (eq->n_row - tab->n_zero);
   1112 	eq = isl_mat_left_hermite(eq, 0, &U, &Q);
   1113 	if (!eq)
   1114 		return -1;
   1115 	isl_mat_free(eq);
   1116 	tab->basis = isl_mat_lin_to_aff(Q);
   1117 	if (tab_shift_cone(tab, tab_cone, U) < 0)
   1118 		return -1;
   1119 	if (!tab->basis)
   1120 		return -1;
   1121 	return 0;
   1122 }
   1123 
   1124 /* Compute and return a sample point in bset using generalized basis
   1125  * reduction.  We first check if the input set has a non-trivial
   1126  * recession cone.  If so, we perform some extra preprocessing in
   1127  * sample_with_cone.  Otherwise, we directly perform generalized basis
   1128  * reduction.
   1129  */
   1130 static __isl_give isl_vec *gbr_sample(__isl_take isl_basic_set *bset)
   1131 {
   1132 	isl_size dim;
   1133 	struct isl_basic_set *cone;
   1134 
   1135 	dim = isl_basic_set_dim(bset, isl_dim_all);
   1136 	if (dim < 0)
   1137 		goto error;
   1138 
   1139 	cone = isl_basic_set_recession_cone(isl_basic_set_copy(bset));
   1140 	if (!cone)
   1141 		goto error;
   1142 
   1143 	if (cone->n_eq < dim)
   1144 		return isl_basic_set_sample_with_cone(bset, cone);
   1145 
   1146 	isl_basic_set_free(cone);
   1147 	return sample_bounded(bset);
   1148 error:
   1149 	isl_basic_set_free(bset);
   1150 	return NULL;
   1151 }
   1152 
   1153 static __isl_give isl_vec *basic_set_sample(__isl_take isl_basic_set *bset,
   1154 	int bounded)
   1155 {
   1156 	isl_size dim;
   1157 	if (!bset)
   1158 		return NULL;
   1159 
   1160 	if (isl_basic_set_plain_is_empty(bset))
   1161 		return empty_sample(bset);
   1162 
   1163 	dim = isl_basic_set_dim(bset, isl_dim_set);
   1164 	if (dim < 0 ||
   1165 	    isl_basic_set_check_no_params(bset) < 0 ||
   1166 	    isl_basic_set_check_no_locals(bset) < 0)
   1167 		goto error;
   1168 
   1169 	if (bset->sample && bset->sample->size == 1 + dim) {
   1170 		int contains = isl_basic_set_contains(bset, bset->sample);
   1171 		if (contains < 0)
   1172 			goto error;
   1173 		if (contains) {
   1174 			struct isl_vec *sample = isl_vec_copy(bset->sample);
   1175 			isl_basic_set_free(bset);
   1176 			return sample;
   1177 		}
   1178 	}
   1179 	isl_vec_free(bset->sample);
   1180 	bset->sample = NULL;
   1181 
   1182 	if (bset->n_eq > 0)
   1183 		return sample_eq(bset, bounded ? isl_basic_set_sample_bounded
   1184 					       : isl_basic_set_sample_vec);
   1185 	if (dim == 0)
   1186 		return zero_sample(bset);
   1187 	if (dim == 1)
   1188 		return interval_sample(bset);
   1189 
   1190 	return bounded ? sample_bounded(bset) : gbr_sample(bset);
   1191 error:
   1192 	isl_basic_set_free(bset);
   1193 	return NULL;
   1194 }
   1195 
   1196 __isl_give isl_vec *isl_basic_set_sample_vec(__isl_take isl_basic_set *bset)
   1197 {
   1198 	return basic_set_sample(bset, 0);
   1199 }
   1200 
   1201 /* Compute an integer sample in "bset", where the caller guarantees
   1202  * that "bset" is bounded.
   1203  */
   1204 __isl_give isl_vec *isl_basic_set_sample_bounded(__isl_take isl_basic_set *bset)
   1205 {
   1206 	return basic_set_sample(bset, 1);
   1207 }
   1208 
   1209 __isl_give isl_basic_set *isl_basic_set_from_vec(__isl_take isl_vec *vec)
   1210 {
   1211 	int i;
   1212 	int k;
   1213 	struct isl_basic_set *bset = NULL;
   1214 	struct isl_ctx *ctx;
   1215 	isl_size dim;
   1216 
   1217 	if (!vec)
   1218 		return NULL;
   1219 	ctx = vec->ctx;
   1220 	isl_assert(ctx, vec->size != 0, goto error);
   1221 
   1222 	bset = isl_basic_set_alloc(ctx, 0, vec->size - 1, 0, vec->size - 1, 0);
   1223 	dim = isl_basic_set_dim(bset, isl_dim_set);
   1224 	if (dim < 0)
   1225 		goto error;
   1226 	for (i = dim - 1; i >= 0; --i) {
   1227 		k = isl_basic_set_alloc_equality(bset);
   1228 		if (k < 0)
   1229 			goto error;
   1230 		isl_seq_clr(bset->eq[k], 1 + dim);
   1231 		isl_int_neg(bset->eq[k][0], vec->el[1 + i]);
   1232 		isl_int_set(bset->eq[k][1 + i], vec->el[0]);
   1233 	}
   1234 	bset->sample = vec;
   1235 
   1236 	return bset;
   1237 error:
   1238 	isl_basic_set_free(bset);
   1239 	isl_vec_free(vec);
   1240 	return NULL;
   1241 }
   1242 
   1243 __isl_give isl_basic_map *isl_basic_map_sample(__isl_take isl_basic_map *bmap)
   1244 {
   1245 	struct isl_basic_set *bset;
   1246 	struct isl_vec *sample_vec;
   1247 
   1248 	bset = isl_basic_map_underlying_set(isl_basic_map_copy(bmap));
   1249 	sample_vec = isl_basic_set_sample_vec(bset);
   1250 	if (!sample_vec)
   1251 		goto error;
   1252 	if (sample_vec->size == 0) {
   1253 		isl_vec_free(sample_vec);
   1254 		return isl_basic_map_set_to_empty(bmap);
   1255 	}
   1256 	isl_vec_free(bmap->sample);
   1257 	bmap->sample = isl_vec_copy(sample_vec);
   1258 	bset = isl_basic_set_from_vec(sample_vec);
   1259 	return isl_basic_map_overlying_set(bset, bmap);
   1260 error:
   1261 	isl_basic_map_free(bmap);
   1262 	return NULL;
   1263 }
   1264 
   1265 __isl_give isl_basic_set *isl_basic_set_sample(__isl_take isl_basic_set *bset)
   1266 {
   1267 	return isl_basic_map_sample(bset);
   1268 }
   1269 
   1270 __isl_give isl_basic_map *isl_map_sample(__isl_take isl_map *map)
   1271 {
   1272 	int i;
   1273 	isl_basic_map *sample = NULL;
   1274 
   1275 	if (!map)
   1276 		goto error;
   1277 
   1278 	for (i = 0; i < map->n; ++i) {
   1279 		sample = isl_basic_map_sample(isl_basic_map_copy(map->p[i]));
   1280 		if (!sample)
   1281 			goto error;
   1282 		if (!ISL_F_ISSET(sample, ISL_BASIC_MAP_EMPTY))
   1283 			break;
   1284 		isl_basic_map_free(sample);
   1285 	}
   1286 	if (i == map->n)
   1287 		sample = isl_basic_map_empty(isl_map_get_space(map));
   1288 	isl_map_free(map);
   1289 	return sample;
   1290 error:
   1291 	isl_map_free(map);
   1292 	return NULL;
   1293 }
   1294 
   1295 __isl_give isl_basic_set *isl_set_sample(__isl_take isl_set *set)
   1296 {
   1297 	return bset_from_bmap(isl_map_sample(set_to_map(set)));
   1298 }
   1299 
   1300 __isl_give isl_point *isl_basic_set_sample_point(__isl_take isl_basic_set *bset)
   1301 {
   1302 	isl_vec *vec;
   1303 	isl_space *space;
   1304 
   1305 	space = isl_basic_set_get_space(bset);
   1306 	bset = isl_basic_set_underlying_set(bset);
   1307 	vec = isl_basic_set_sample_vec(bset);
   1308 
   1309 	return isl_point_alloc(space, vec);
   1310 }
   1311 
   1312 __isl_give isl_point *isl_set_sample_point(__isl_take isl_set *set)
   1313 {
   1314 	int i;
   1315 	isl_point *pnt;
   1316 
   1317 	if (!set)
   1318 		return NULL;
   1319 
   1320 	for (i = 0; i < set->n; ++i) {
   1321 		pnt = isl_basic_set_sample_point(isl_basic_set_copy(set->p[i]));
   1322 		if (!pnt)
   1323 			goto error;
   1324 		if (!isl_point_is_void(pnt))
   1325 			break;
   1326 		isl_point_free(pnt);
   1327 	}
   1328 	if (i == set->n)
   1329 		pnt = isl_point_void(isl_set_get_space(set));
   1330 
   1331 	isl_set_free(set);
   1332 	return pnt;
   1333 error:
   1334 	isl_set_free(set);
   1335 	return NULL;
   1336 }
   1337