isl_map_gist: also perform gist in universe context

Originally, it didn't make sense to gist in a universe context,
because there were no constraints with respect to which the gist
could be computed.  However, since the more aggressive gist of
d5e1701 (isl_basic_map_gist: be more aggressive in removing constraints)
some contraints can get removed even in a universe context.

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_union_map: construct new isl_union_map on operations that change dimensions

Otherwise, the entries in the hashtable will end up hashed on the wrong
hash value (that of the original isl_dim).

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_union_map_is_empty: fix typo that resulted in memory leaks

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

add isl_basic_map_has_defining_equality

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

add isl_set_get_ctx

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

update .gitignore

isl_printer_print_map: fix printing in latex format

It was broken by e094707 (isl_output.c: extract out common print_dim).

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_set_wrap_facet: normalize resulting facet constraint

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_coalesce.c: fix typo in comment

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_pw_qpolynomial_intersect_domain: simplify polynomials using equalities

We now use any equalities we find in the intersected domain.
Perhaps we should check if the intersection had any effect
on the equalities first.

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_qpolynomial_move_dims: plug memory leak and fix reordering of divs

In particular, when divs got reordered, the new order would not
be reflected in the associated upoly.

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

add isl_pw_qpolynomial_read_from_str

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

add isl_qpolynomial_fold_substitute

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_pw_qpolynomial_intersect_domain: drop pieces that end up having empty domain

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_map_product: used nested spaces in result

That is, the product of A -> B and C -> D now has dimension
[A -> C] -> [B -> D].  This allows us to keep all information
about the spaces A, B, C, D, including their names and any
prior nestings.

Since it sometimes also useful to have concatenated spaces in the result,
we also introduct isl_map_flat_product, which produces what isl_map_product
used to produce.

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_pw_qpolynomial_bound: handle isl_pw_qpolynomials with wrapped domains

In particular, compute a bound over the range of the wrapped relation,
instead of over the entire relation.

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

support nested isl_dims

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

doc: fix typo

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

export isl_dim_tuple_match

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_output.c: extract out common print_dim

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_input.c: read_tuple: construct isl_dim instead of isl_basic_map

Since the tuples may contain affine expressions in terms of earlier
dimensions, it made sense to collect both the dimension information
and the possible equalities in a single object, i.e., an isl_basic_map.
However, when we introduce nested dimension specifications, we will
want to construct isl_dims recursively and then it makes more sense
to construct operate on isl_dims and to collect the equalities
in a separate data structure.

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

doc: remove description of how to get the piplib submodule

The submodule was actually removed way back in 309036c (add copyright
statements).

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_printer_print_pw_qpolynomial_fold: print dim as set instead of map

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

add isl_pw_qpolynomial_insert_dims

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

add isl_set_insert

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

add isl_basic_map_image_is_bounded

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

avoid NULL pointer exception in isl_basic_map_n_*

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

fix isl_pw_qpolynomial_reset_dim

The original implementation would only reset the dims of
the domains and not of the quasi-polynomials.

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

add isl_mat_add_rows

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

add isl_mat_add_zero_cols

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

export isl_basic_map_move_dims

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

hide isl_dim internals

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

add isl_dim_get_ctx

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_basic_map_reset_dim: finalize result

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_transitive_closure.c: anonymize input map during incremental computation

Most other operations inside the transitive closure computation
implicitly drop the tuple names, so we have to drop them explicitly
here for the maps to remain compatible.

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_transitive_closure.c: keep track of domains for Floyd-Warshall

When applying Floyd-Warshall to an isl_map, we don't care about
the domains anymore after we have grouped them.
However, when applying Floyd-Warshall to an isl_union_map, we
use the domains do initialize the grid with maps of the right
dimension.  We therefore need to order the domains according
to the group number so that the calling function has a chance
to find them.

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_dim_drop: always remove tuple name, even if number of dims to drop is zero

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_dim_set_tuple_name: allow explicit removal of tuple name

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

add support for union sets and relations

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_transitive_closure.c: extract out some reusable code

In particular, the code will also be useful for computing transitive
closures of union maps.

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

add isl_pw_qpolynomial_reset_dim

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

add isl_map_has_equal_dim

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

add isl_dim_get_hash

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_dim: allow specification of tuple names

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_map_from_domain_and_range: implement in terms of isl_map_apply_range

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_basic_map_domain: properly set dimension of result

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

privately export isl_set_reset_dim and isl_map_reset_dim

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_map_read: extract out read_var_def

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_hash_table_foreach: pass pointer to entry data

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_hash_table_foreach: add user argument

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_transitive_closure.c: fix double free on error path

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_map_gist: use simple hull instead of convex hull

isl is still not very good at computing convex hulls.

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_map.c: fix typo

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_cat: read generic objects

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_printer: add NULL checks

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

make isl_obj_none a constant expression

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

add isl_map_get_ctx

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_map_fast_is_universe: also detect redundant representations of the universe

During printing, we don't want to print any constraints if the given
map contains the universe.

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

add isl_map_reaching_path_lengths

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_map_power: delay equating path lengths to power parameter

It doesn't make sense for the isl_map_transitive_closure caller
to have to specify a dummy parameter.

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_map_power: apply Floyd-Warshall to compute powers as well

Before, Floyd-Warshall was only applied to compute transitive closure.
However, the power is essentially a transitive closure on an extended
relation, so we simply need to perform the nested call on such extended
relations.
In principle, we should be able to apply the incremental computation
to the power too, but changing the nested calls is a bit more tricky.

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_map_lexmax: produce less cryptic error message when result is unbounded

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

fix isl_basic_map_move_dims in presence of existentially quantified variables

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_test: drop debugging output

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_transitive_closure.c: fix typo in comment

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

fix isl_map_lex_ge and isl_map_lex_le in the zero-dimensional case

The first zero dimensions are always equal, so these functions
should return a universe map in the zero-dimensional case rather
than an empty map.

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_printer_print_map: replace variable by its definition in tuple

In particular, if a variable is equal to an integer affine combination
of parameters and earlier variables, then print the affine combination
instead of the name of the variable.

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

privately export isl_basic_map_fast_cmp and isl_basic_map_fast_is_equal

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_dim_set_name: don't accept names that look like numbers

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

Remove comma at the end of enumeration

This is invalid in C89. Some compilers warn or fail if they are set to
strict compilation options.

Signed-off-by: Tobias Grosser <grosser@fim.uni-passau.de>
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_convex_hull.c: avoid side-effects in calls to isl_int_* macros

add isl_hash_builtin

add isl_int_get_str

doc: fix some typos

isl 0.03

Makefile.am: really include include/isl_vertices.h

add isl_map_is_bijective

Tests cases from Tomofumi Yuki <tomofumi.yuki@gmail.com>

doc: fix typo

doc: drop duplicate booktitle from Kelly1996closure bib entry

doc: use more explicit notation to avoid confusion

add isl_constraint_is_div_constraint

isl_map_simplify.c: separate out isl_basic_map_is_div_constraint

isl_basic_map_from_constraint: keep all divs, even redundant ones

When this function is called by CLooG, it assumes that the position
of the divs will remain the same.  If the redundant divs are removed,
then this may not be the case.
We should probably try to come up with a more robust interface.

isl_basic_map_from_constraint: only return copy of bmap on equality constraints

Commit b39d157 (isl_basic_map_from_constraint: return copy of bmap in
constraint if possible) changed isl_basic_map_from_constraint to return
a copy when this bmap consists of a single constraint in order to handle
the case where a new constraint is created with unknown existentially
quantified variables.  However, if the divs _are_ known, then the div
definition is simply thrown away.  If the single constraint is an equality,
then it may be possible to recover the definition, but if it is an inequality,
then this is not possible and in fact it does not make sense to have unknown divs.
Therefore, only return a copy on equality constraints.

isl_constraint_add_div: correctly set line to first and only constraint

When isl_constraint_add_div was introduced in 3fcebd4 (isl_constraint:
add limited support for adding divs), the equalities were still placed
before the inequalities in the constraint matrix.
This changed in 9eb1488 (isl_basic_map: put inequalities before equalities),
but the implementation of isl_constraint_add_div was not adapted accordingly.
This has probably gone unnoticed because isl_constraint_add_div has only
been used on equality constraints.

isl_pip: rename pip_options to avoid conflict with PipLib's pip_options_free

isl_convex_hull.c: initial_facet_constraint: fix error handling

doc: describe backward incompatible changes

doc: mention discussion group

rename isl_basic_map_convex_hull to isl_basic_map_remove_redundancies

isl_affine_hull.c: outside_point: fix error handling

isl_tab_rollback: avoid double free on error path

isl_tab.c: tab_mat_product: avoid NULL pointer dereference

isl_tab_extend_cons: avoid NULL pointer dereference

implement antiquated isl_basic_set_n_* in terms of isl_basic_set_dim

isl_map_convex_hull: avoid NULL pointer dereference

isl_tab_add_valid_eq: return int instead of isl_tab *

isl_affine_hull.c: affine_hull_with_cone: avoid NULL pointer dereference

isl_equalities.c: parameter_compression_multi: avoid NULL pointer dereference

isl_basic_map_project_out: avoid NULL pointer dereference