implement isl_constraint on top of isl_vec

In e2a2317 (reimplement isl_constraint in terms of isl_aff,
Fri Jul 1 17:20:52 2011 +0200), were reimplemented on top of isl_aff
because both objects share many features.
There are also some difference, however, and it is therefore better
to implement isl_constraint directly on top of isl_vec.
In particular, an isl_aff has a denominator, whereas an isl_constraint
does not.  Although the denominator of an isl_aff is usually 1, it may
not always be and could therefore lead to subtle bugs.

Moreover, the space of an isl_constraint may be either that of
a set or a map, whereas, in principle, the domain of isl_aff is always
a set.  We allowed the domain of an isl_aff to be a map when it was
used to implement an isl_constraint, but this was not very clean.
This would become even more dramatic when we change the space of an isl_aff
to be that of a map.

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

add isl_vec_neg

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

isl_aff.c: extract isl_local_space_get_active

This function will also be useful for our new implementation of isl_constraints.

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

isl_output.c: drop redundant "set" argument

We used to have to keep track of whether a given isl_space was that
of a map or a set, but now the isl_space itself contains that information
so we no longer need this extra argument.

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

isl_union_*_alloc: project space onto parameters

Before, we were expecting the user to provide a parameter space,
but we might as well project whatever the user gives us onto
its parameter space.

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

explicitly differentiate between spaces of maps, sets and parameter sets

Before, no distinction was made internally between spaces of maps,
sets and parameter spaces.  In particular, an isl_space with zero
input and output dimensions could have been any of those.
When printing isl_sets and isl_maps, we could keep track of whether
the original object as a set or a map, but parametric domains and
zero-dimensional domains were still printed identically.

Now we explicitly keep track of whether a space is supposed to be
that of a map, set or parametric set.  This means for example that
"[N] -> { [] : N >= 0 }" and "[N] -> { : N >= 0 }" are no longer considered
to be identical.
This may break some code that was using one space in place of another,
especially since until recently there was no official way of creating
a parameter space.

The main advantage is that we can now recognize parameter spaces
and treat them appropriately.  In particular, the domains of affine
expressions can now be correctly identified as being either (possibly
zero-dimensional) sets or parameter domains.

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

allow isl_ids with negative reference counts

This will be useful for static isl_ids that are not supposed to change.

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

proper implementation of isl_union_map_from_range

The original implementation would just return the input
because isl_sets and isl_maps have the same internal representation.
This will change in the near future, however, and so we need
to actually compute the range of all the elements in the input.

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

isl_map_from_domain_and_range: drop explicit conversion to isl_map

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

isl_set_apply_pw_qpolynomial_fold: drop explicit conversion to isl_map

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

isl_tab_basic_set_non_trivial_lexmin: drop explicit conversion to isl_basic_map

And similarly for isl_tab_basic_set_non_neg_lexmin.

In future, the space of an isl_basic_map will be different from that
of an isl_basic_set.  Calling isl_basic_map_from_range may therefore
result in possibly costly changes.

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

isl_tab_pip.c: fix whitespace issue

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

isl_set_read_from_*: coerce set-like maps into sets

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

isl_input.c: map_read_polylib: convert set-like maps to sets

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

isl_input.c: obj_read: start off from parameter space

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

isl_input.c: represent nested tuple as isl_set

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

isl_input.c: obj_read_poly_or_fold: represent domain as isl_set

Although, internally, isl_sets and isl_maps are the same,
conceptually, the domain of an isl_pw_qpolynomial is an isl_set.

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

isl_basic_set_multiplicative_call: explicitly project on parameter space

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

isl_pw_qpolynomial_fold_bound: explicitly project on parameter space

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

isl_qpolynomial_bound_on_domain_bernstein: explicitly project on parameter space

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

isl_qpolynomial_bound_on_domain_range: explicitly project on parameter space

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

isl_union_map_from_map: explicitly project on parameter space

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

isl_union_map_align_params: explicitly project on parameter space

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

isl_polynomial.c: constant_on_domain: explicitly project on parameter space

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

basic_set_dim_max: explicitly project on parameter space

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

isl_basic_set_compute_vertices: explicitly turn chambers into parameter sets

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

bound.c: explicitly convert context into parameter set

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

pip.c: explicitly convert context into parameter set

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

add isl_pw_qpolynomial_fold_project_domain_on_params

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

add isl_qpolynomial_project_domain_on_params

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

add isl_pw_qpolynomial_project_out

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

add isl_morph_{dom,ran}_params

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

add isl_basic_set_params and isl_set_params

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

add isl_space_params_alloc

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

isl_stream_read_set: accept parameter domains

In particular, accept sets without a tuple, e.g.,

[N] -> { : N >= 1 }

At this moment, no distinction is made between a "parameter domain"
and a zero-dimensional set.  However, this will change in the near
future and so we need to have a different notation for these two
types of sets.  We start by allowing the new notation for parameter
domains.

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

isl_pw_qpolynomial_fold_bound: simplify construction of target space

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

isl_test.c: properly create set space

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

isl_map_range_map: fix space manipulation

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

isl_map_range: construct correct space for result

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

isl_basic_set_alloc: create basic set in proper set space

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

isl_space_wrap: properly create set space

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

add isl_set_from_pw_aff

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

rename PW_alloc_ to PW_alloc_size

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

isl_pw_*_add_disjoint: try and extend one of the pws instead of copying pieces

This is especially useful when isl_pw_*_add_disjoint is used
to repeatedly add a single piece.

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

isl_space_match: optimize trivial cases

A space is obviously equal to itself.

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

add isl_map_flat_domain_product

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

isl_access_info_compute_flow: align parameters of input access relations

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

align parameters of arguments to isl_map_{lexmin,lexmax}

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

isl_space_align_params: accept general isl_spaces

Before, we would only accept parameter spaces.

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

add isl_space_params

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

rename isl_dim to isl_space

The old name was confusing because the name suggested that the object
represents a single dimension, while in fact it represents an entire space.
The documented isl_dim based names are for backward compatibility.

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

document isl_dim_move

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

split off declarations from isl_polynomial.h to separate file

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

merge isl_*_dump functions

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

add isl_set_dim_min

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

add isl_basic_set_foreach_lexopt

add isl_pw_aff_union_opt

add isl_pw_aff_union_min

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

add isl_aff_le_basic_set

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

isl_qpolynomial_free: return NULL

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

Merge branch 'maint'

isl_qpolynomial_rat_cst: handle NULL space

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

bernstein_coefficients_cell: handle NULL poly

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

isl_printer_to_str: intialize buffer to empty string

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

isl_map_insert: properly handle named or nested spaces

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

isl_set_flat_product: call isl_map_range_flat_product

A set only has a "range", so we should only compute the product of
that range and not of the domain.

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

add some isl_map_plain_is_fixed tests

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

isl_basic_map_neg: gauss result

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

isl_dim_get_tuple_id: complain if tuple has no id

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

isl_dim_get_dim_id: complain if dim has no id

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

add isl_*_has_tuple_id

The isl_*_git_tuple_id functions cannot be used to find out
whether a dimension has an id as they may also return NULL
on failures.

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

add isl_*_has_dim_id

The isl_*_git_dim_id functions cannot be used to find out
whether a dimension has an id as they may also return NULL
on failures.

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

doc: warn about the limited use of identifiers of individual dimensions

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

document isl_map_dim

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

rename isl_map_insert to isl_set_insert_dims and add it to the docs

The new name is more consistent with other functions, especially
isl_map_add_dims.

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

Merge branch 'maint'

isl_dim_equal: handle NULL input

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

isl_dim_tuple_match: handle NULL input

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

isl_map_is_subset: a map cannot be as subset of a map in a different space

The only debatable case is that of an empty map, which in principle
could be considered to be a subset of any map.  This patch considers
empty maps not to be subsets of maps in other spaces.

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

document isl_map_plain_is_fixed

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

Merge branch 'maint'

isl_set_product: call isl_map_range_product instead of isl_map_product

A set only has a "range", so we should only compute the product of
that range and not of the domain.

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

isl_basic_map_remove_dims: properly handle named or nested spaces

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

isl_basic_map_drop: properly handle nested spaces

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

isl_access_info_compute_flow: plug memory leak on error path

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

doc: remove "struct" in front of isl_* types

Signed-off-by: Sebastian Pop <sebpop@gmail.com>
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

doc: update prototype of isl_constraint_free

It was changed in e2a2317 (reimplement isl_constraint in terms of isl_aff,
Fri Jul 1 17:20:52 2011 +0200).

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

isl_basic_map_remove_divs: handle NULL input

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

isl_schedule.c: handle basic maps individually in setup_carry_lp

Our dependence relations may be unions of basic maps.
When trying to carry dependences we may only be able to carry
some of them at a time.  We therefore need to consider them individually.

Reported-by: Tobias Grosser <tobias@grosser.es>
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

add isl_set_reset_tuple_id

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

Merge branch 'maint'

isl_dim_range_product: dimension names of domains don't need to match

It's sufficient for the domain tuples to match.

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

isl_aff.c: fix up copyright

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

doc: fix typos

Signed-off-by: Sebastian Pop <sebpop@gmail.com>
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>

isl_aff_floor: reduce coefficients of newly created div

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

add isl_aff_mod and isl_pw_aff_mod

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

add isl_int_mul_2exp

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

export isl_pw_aff_foreach_piece

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

add isl_set_dim_has_{lower,upper}_bound

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

isl_stream_read_map: accept assignments to tuple variables

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