Sven Verdoolaege [Mon, 3 Dec 2012 12:37:42 +0000 (13:37 +0100)]
isl_schedule.c: add_intra_constraints: avoid invalid access on error
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Mon, 3 Dec 2012 12:37:28 +0000 (13:37 +0100)]
isl_tab_basic_set_non_neg_lexmin: handle NULL input
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Mon, 3 Dec 2012 07:17:26 +0000 (08:17 +0100)]
isl 0.11
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 20:56:45 +0000 (21:56 +0100)]
isl_schedule.c: add_inter_validity_constraints: avoid invalid access on error
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 16:37:44 +0000 (17:37 +0100)]
isl_tab_basic_set_non_trivial_lexmin: handle NULL input
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 16:37:17 +0000 (17:37 +0100)]
isl_tab.c: push_union: handle NULL input
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 15:56:50 +0000 (16:56 +0100)]
isl_mat_transpose: handle NULL input
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 14:38:13 +0000 (15:38 +0100)]
isl_basic_map_foreach_lexopt: avoid invalid access on error
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 14:35:49 +0000 (15:35 +0100)]
isl_pw_*_on_shared_domain_in: avoid double free on error path
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 14:35:19 +0000 (15:35 +0100)]
isl_output.c: print_set_c: handle NULL input
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 14:34:09 +0000 (15:34 +0100)]
isl_basic_map_eliminate_vars: avoid invalid access on error
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 14:33:42 +0000 (15:33 +0100)]
isl_map_simplify.c: remove_dependent_vars: handle NULL input
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 13:17:31 +0000 (14:17 +0100)]
isl_tab_pip.c: sol_pop: avoid invalid access on error
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 12:29:05 +0000 (13:29 +0100)]
isl_basic_map_deltas: fix error handling
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 12:23:06 +0000 (13:23 +0100)]
isl_ast_codegen.c: generate_sorted_domains: avoid invalid access on error
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 12:17:16 +0000 (13:17 +0100)]
isl_test.c: test_ast: avoid invalid access on error
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 12:16:02 +0000 (13:16 +0100)]
isl_parameter_alignment_reordering: fix error handling
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 12:14:19 +0000 (13:14 +0100)]
isl_basic_map_normalize: avoid invalid access on error
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 12:13:26 +0000 (13:13 +0100)]
isl_map.c: basic_set_maximal_difference_at: fix error handling
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 12:11:01 +0000 (13:11 +0100)]
isl_convex_hull.c: map_simple_hull: avoid invalid access on error
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 12:10:31 +0000 (13:10 +0100)]
isl_ast_codegen.c: compute_domains: avoid invalid access on error
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 11:31:20 +0000 (12:31 +0100)]
isl_basic_map_is_empty: avoid invalid access on error
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 11:23:02 +0000 (12:23 +0100)]
isl_space_extend: avoid invalid access on error
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 11:22:48 +0000 (12:22 +0100)]
isl_map_subtract.c: basic_map_collect_diff: avoid invalid access on error
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 11:21:46 +0000 (12:21 +0100)]
isl_basic_map_overlying_set: avoid invalid access on error
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 11:20:44 +0000 (12:20 +0100)]
isl_map.c: basic_map_space_reset: handle NULL input
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 11:19:40 +0000 (12:19 +0100)]
isl_basic_set_swap_vars: handle NULL input
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 11:18:26 +0000 (12:18 +0100)]
isl_input.c: construct_constraints: respect memory management annotations
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 11:15:00 +0000 (12:15 +0100)]
isl_ast_codegen.c: create_node: avoid invalid access on error
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 11:13:57 +0000 (12:13 +0100)]
isl_ast_codegen.c: for_inc: handle NULL input
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 11:12:14 +0000 (12:12 +0100)]
isl_pw_aff_set_rational: avoid invalid access on error
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 10:00:22 +0000 (11:00 +0100)]
isl_vec_dup: avoid invalid accesses on error
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 2 Dec 2012 09:57:46 +0000 (10:57 +0100)]
isl_map.c: map_intersect_internal: properly plug memory leak on error path
The proposed fix in f85421c (isl_map.c: map_intersect_internal: plug memory
leak on error path, Fri Nov 30 18:10:08 2012 +0100) would only free "result",
but not "part". Revert that commit and make sure both are freed.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sat, 1 Dec 2012 14:17:15 +0000 (15:17 +0100)]
update for change in the constructor of the clang Driver
In particular, the Driver constructor no longer takes an IsProduction argument.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sat, 1 Dec 2012 09:40:49 +0000 (10:40 +0100)]
isl_vertices.c: vertices_from_list: plug memory leak on error path
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sat, 1 Dec 2012 09:39:37 +0000 (10:39 +0100)]
isl_basic_set_compute_vertices: plug memory leak on error path
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sat, 1 Dec 2012 09:38:51 +0000 (10:38 +0100)]
isl_test.c: fix typo
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Thu, 29 Nov 2012 21:53:30 +0000 (22:53 +0100)]
isl_test.c: test_output: avoid invalid access on error
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sat, 1 Dec 2012 09:34:41 +0000 (10:34 +0100)]
isl_test.c: before_for: return early on error
In particular, don't print any spurious error messages.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sat, 1 Dec 2012 09:34:41 +0000 (10:34 +0100)]
isl_test.c: test_sample: return early on error
In particular, don't print any spurious error messages.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Fri, 30 Nov 2012 23:30:18 +0000 (00:30 +0100)]
isl_tab_pip.c: find_in_pos: plug memory leak on error path
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Fri, 30 Nov 2012 23:29:56 +0000 (00:29 +0100)]
isl_tab_pip.c: add_parametric_cut: plug memory leak on error path
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Fri, 30 Nov 2012 23:28:49 +0000 (00:28 +0100)]
isl_tab_pip.c: sol_push_sol: plug memory leak on error path
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Fri, 30 Nov 2012 23:25:49 +0000 (00:25 +0100)]
isl_space_dup: avoid invalid accesses on error
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Fri, 30 Nov 2012 23:24:02 +0000 (00:24 +0100)]
isl_schedule.c: add_*_constraints: avoid invalid accesses on error
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Fri, 30 Nov 2012 23:22:27 +0000 (00:22 +0100)]
isl_reordering_extend_space: fix error handling
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Fri, 30 Nov 2012 23:21:40 +0000 (00:21 +0100)]
isl_multi_*_realign_domain: plug memory leak on error path
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Fri, 30 Nov 2012 23:19:27 +0000 (00:19 +0100)]
isl_map_subtract.c: map_is_subset: fix error handling
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Fri, 30 Nov 2012 22:23:29 +0000 (23:23 +0100)]
isl_map.c: update_dim_opt: return early on error
In particular, don't print any error messages based on the number
of elements in the list if the list doesn't exist.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Fri, 30 Nov 2012 17:27:22 +0000 (18:27 +0100)]
isl_basic_map_underlying_set: plug memory leak on error path
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Fri, 30 Nov 2012 17:10:08 +0000 (18:10 +0100)]
isl_map.c: map_intersect_internal: plug memory leak on error path
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Fri, 30 Nov 2012 17:01:42 +0000 (18:01 +0100)]
isl_ast_build_init_derived: check build->domain
If something has gone wrong during the construction of the isl_ast_build,
its domain field may be NULL. Since isl_ast_build_get_ctx obtains
the isl_ctx from this field, we should not continue if build->domain is NULL.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Fri, 30 Nov 2012 16:59:58 +0000 (17:59 +0100)]
isl_input.c: read_aff_domain: remove spurious vars_free on error path
The vars structure passed to read_aff_domain is owned by the caller.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Fri, 30 Nov 2012 16:57:23 +0000 (17:57 +0100)]
isl_input.c: read_map_tuple: return early on error
This avoids spurious error messages.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Fri, 30 Nov 2012 16:55:30 +0000 (17:55 +0100)]
isl_input.c: accept_affine_list: plug memory leak on error path
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Fri, 30 Nov 2012 16:50:37 +0000 (17:50 +0100)]
isl_affine_hull.c: equalities_in_underlying_set: plug memory leak on error path
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Fri, 30 Nov 2012 16:46:13 +0000 (17:46 +0100)]
isl_tab_detect_equalities: plug memory leak on error path
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Fri, 30 Nov 2012 16:52:50 +0000 (17:52 +0100)]
isl_ast_codegen.c: after_each_for: add missing return on error path
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Thu, 29 Nov 2012 23:18:19 +0000 (00:18 +0100)]
reintroduce isl_basic_set_add symbol
This function was renamed in 461a2cd (rename isl_basic_set_add to
isl_basic_set_add_dims, Mon Oct 1 10:55:08 2012 +0200), but it appears
that it is being used by CLooG. Reintroduce the symbol for backward
compatibility, but mark it "deprecated" so that it will hopefully
not be used by new users.
Reported-by: Jack Howarth <howarth@bromo.med.uc.edu>
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Fri, 23 Nov 2012 14:41:24 +0000 (15:41 +0100)]
add isl_union_map_uncurry
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Fri, 23 Nov 2012 20:43:29 +0000 (21:43 +0100)]
configure.ac: check if TargetInfo::CreateTargetInfo takes TargetOptions pointer
Newer versions of clang take a pointer, while older versions take a reference.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Fri, 23 Nov 2012 20:25:11 +0000 (21:25 +0100)]
interface/extract_interface.cc: take TargetOptions object from CompilerInstance
The memory management of TargetOptions has changed in recent clangs.
Taking TargetOptions from the CompilerInstance seems to work in both
older and newer versions of clang.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Thu, 22 Nov 2012 12:29:31 +0000 (13:29 +0100)]
AST printing callbacks: pass isl_ast_print_options
The callback may want to pass control back to isl_ast_node_*_print,
for which it needs an isl_ast_print_options object. Since the
caller of the callback already has a reference to such an object,
we might as well pass it along to the callback.
Since the callback may want to modifiy the isl_ast_print_options
object, we turn it into a properly reference counted isl object.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Thu, 22 Nov 2012 12:27:57 +0000 (13:27 +0100)]
doc: fix typo
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Tue, 20 Nov 2012 19:22:35 +0000 (20:22 +0100)]
add before_each_for/after_each_for callbacks
Sven Verdoolaege [Sat, 10 Nov 2012 15:46:06 +0000 (16:46 +0100)]
isl_ast_build_from_context: allow unnamed set dimensions
The original code would assume all dimensions have an associated isl_id.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 18 Nov 2012 08:24:50 +0000 (09:24 +0100)]
isl_basic_map_gist: drop irrelevant constraints from the context
The context may contain constraints on variables that do not even
appear in the input basic map. The constraints therefore cannot
be exploited to simplify the input and only serve to make some
of the internal computations more expensive, in particular the
computation of the affine hull of the intersection of input and context
and the emptiness checks used in the final determination of
the redundancy of a constraint.
It's therefore more efficient to remove those irrelevant constraints.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Tue, 20 Nov 2012 16:53:57 +0000 (17:53 +0100)]
isl_map_simplify.c: fix typo in comment
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Tue, 20 Nov 2012 14:39:39 +0000 (15:39 +0100)]
isl_ast_build_eliminate_divs: remove unknown divs before divs involving depth
Occasionally, eliminating unknown divs may turn other unknown divs
into known divs, which may depend on the current depth.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sat, 17 Nov 2012 12:28:18 +0000 (13:28 +0100)]
isl_affine_hull.c: affine_hull_with_cone: plug memory leak on error path
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sat, 17 Nov 2012 12:27:38 +0000 (13:27 +0100)]
change prototype of isl_factorizer_dump
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Mon, 19 Nov 2012 11:25:11 +0000 (12:25 +0100)]
isl_tab_sample: perform greedy search before performing basis reduction
While basis reduction is very useful on difficult problems, it is
overkill on easy problems. In fact, the basis reduction computation
can be quite expensive, especially in the presence of large coefficients.
This may happen in particular during the affine hull computation
when only a few sample points have been found so far. If these
points have large values, then the equalities that describe their
affine hull may have coefficients with larger values still.
Perform a greedy search before computing a reduced basis.
We also perform a greedy search after computing the reduced basis
just in case the problem has become easy through this basis reduction.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Tue, 20 Nov 2012 10:41:42 +0000 (11:41 +0100)]
isl_tab_sample: extract out compute_min and compute_max
We will be able to reuse them in the next commit.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Tue, 20 Nov 2012 10:19:00 +0000 (11:19 +0100)]
isl_tab_sample: treat emptiness of tableau as error
We only modify the tableau through calls to isl_tab_add_valid_eq,
which should never result in an empty tableau, given that
the tableau was initially not empty.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Fri, 16 Nov 2012 15:29:04 +0000 (16:29 +0100)]
AST generation: fix handling of separation classes
In particular, make sure that the classes remain separate.
After removing the previous classes from the current class,
we would intersect the result with the schedule domain
and then eliminate inner dimensions. This elimination process
may drop some constraints that are needed to ensure that the
current class is disjoint from the previous classes.
We therefore intersect the result of the elimination step
with the current separation class domain.
Reported-by: Tobias Grosser <tobias@grosser.es>
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Mon, 1 Oct 2012 09:24:49 +0000 (11:24 +0200)]
isl_printer_print_{basic_,}{set,map}: don't start new line in omega format
For the isl format, the new line was removed in 9a54ed5
(isl_printer_print_basic_{set,map}: don't start new line in isl format,
Sun Feb 13 09:05:48 2011 +0100).
Additionally, remove the parameter declaration.
Printing out the parameter declaration for each (basic) set or map
is not very convenient. Instead, we now support printing the parameter
declaration through isl_printer_print_space.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Mon, 1 Oct 2012 09:18:04 +0000 (11:18 +0200)]
isl_printer_print_space: support printing in "omega" format
In particular, print a declaration for the parameters in the space.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Mon, 1 Oct 2012 08:55:08 +0000 (10:55 +0200)]
rename isl_basic_set_add to isl_basic_set_add_dims
For consistency with isl_set_add_dims and isl_basic_set_insert_dims.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sat, 29 Sep 2012 18:11:32 +0000 (20:11 +0200)]
isl_aff_normalize: plug in divs with unit coefficient in other divs
The outer div expression can then be simplified and the inner div
expression may be removed if it was only used inside the outer div
expression.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sat, 29 Sep 2012 17:45:57 +0000 (19:45 +0200)]
isl_local_space_substitute_seq: take argument for the number of divs to modify
This will allow us to modify a single div in the next commit.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sat, 8 Sep 2012 13:04:45 +0000 (15:04 +0200)]
isl_aff_floor: normalize result
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sat, 8 Sep 2012 12:56:40 +0000 (14:56 +0200)]
isl_aff_ceil: compute ceil(e/m) as floor((e + m - 1)/m)
The original implementation would compute ceil(e/m) as -floor(-e/m).
Although both are correct, floor((e + m - 1)/m) is probably less
surprising.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Mon, 12 Nov 2012 11:13:15 +0000 (12:13 +0100)]
isl_aff_gist: add missing isl_vec_cow
We could end up modifying a different isl_aff that originated from
the same isl_aff as the one being modified.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 4 Nov 2012 15:51:30 +0000 (16:51 +0100)]
isl_union_set_compute_schedule: double check that we do not run out of rows
The previous commit fixed a case where we would generate more schedule
rows than anticipated. Make sure we do not run into this situation again.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 4 Nov 2012 15:30:54 +0000 (16:30 +0100)]
isl_union_set_compute_schedule: ensure carry_dependences makes progress
If solve_lp is unable to find a solution, then compute_schedule_wcc
calls carry_dependences to carry as many dependences as possible.
However, the reason that solve_lp was unable to find a solution
may be due to proximity dependences. In particular, if there are
an infinite number of dependences with the same starting or ending
itereration, then the way proximity dependences are currently handled
may force some of the schedule coefficients to be zero based solely
on those proximity dependences.
carry_dependences may then fail to carry any (validity) dependences
simply because there are no validity dependences.
More generally, it may produce a non-trivial solution on only some
of the nodes. This may result in a number of schedule rows (infinite
in the worst case) that is larger than the maximum computed in
compute_max_row, leading to out-of-bounds accesses.
We therefore check that the schedule row computed by carry_dependences
is non-trivial on all the nodes where it should be non-trivial.
If not, we try again on each dependence graph component separately
or fail if there is only one component.
Note that this commit does not fix the problem that the presence
of proximity dependences may eliminate some valid solutions,
but it should solve the out-of-bounds accesses.
Reported-by: Tobias Grosser <tobias@grosser.es>
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Tue, 30 Oct 2012 22:15:41 +0000 (23:15 +0100)]
update for change in clang's DiagnosticOptions
DiagnosticOptions was moved from Frontend to Basic and
createDiagnostics now takes a pointer to a DiagnosticOptions object.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 14 Oct 2012 14:38:21 +0000 (16:38 +0200)]
isl_map_plain_is_disjoint: handle inputs with different parameters
In particular, make sure we do not call isl_basic_map_plain_is_disjoint
if the input relations do not have the same parameters.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sat, 13 Oct 2012 12:04:19 +0000 (14:04 +0200)]
add README
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Andreas Kloeckner [Mon, 8 Oct 2012 17:36:50 +0000 (13:36 -0400)]
Introduce isl_int_free_str to remove GMP dependency in isl_int interface
Signed-off-by: Andreas Kloeckner <inform@tiker.net>
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Wed, 10 Oct 2012 10:36:37 +0000 (12:36 +0200)]
isl_union_set_compute_schedule: handle result with zero bands
The code in extract_schedule would assume that there is at least
one band in the resulting schedule. Skip the parts that depend
on this assumption when the number of bands is zero.
Reported-by: Sameer Sahasrabuddhe <Sameer.Sahasrabuddhe@amd.com>
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sun, 7 Oct 2012 10:18:45 +0000 (12:18 +0200)]
isl_ast_codegen.c: generate_domain: postpone single-valuedness test
In particular, perform the test on the map that will actually be used
to construct the call expression.
The original code would first perform the test and then construct
the call expression from a map with a simplified domain.
This construction may fail because after simplification of the domain,
the map may no longer be single valued.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Fri, 5 Oct 2012 12:15:20 +0000 (14:15 +0200)]
isl_ast_build_ast_from_schedule: make construction of else branches optional
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Tue, 2 Oct 2012 11:30:25 +0000 (13:30 +0200)]
isl_ast_build_ast_from_schedule: use "<" if upper bound has negative constant
This is a purely cosmetic change.
In particular, we prefer
for (int c1 = max(0, c0); c1 < min(N, N + c0); c1 += 1)
over
for (int c1 = max(0, c0); c1 <= min(N - 1, N + c0 - 1); c1 += 1)
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Tue, 2 Oct 2012 11:20:24 +0000 (13:20 +0200)]
isl_ast_expr: add isl_ast_op_lt and isl_ast_op_gt operations
isl_ast_op_lt will be useful in the next commit.
isl_ast_op_gt is added for consistency.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Mon, 1 Oct 2012 08:52:00 +0000 (10:52 +0200)]
add isl_basic_set_from_params
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Fri, 28 Sep 2012 18:19:01 +0000 (20:19 +0200)]
isl_ast_expr_from_aff: try harder to use isl_ast_op_pdiv_{q,r}
When generating C code, we prefer to generate "/" (or "%") rather
than "floord" because "floord" has an implicit condition on the
sign of the first argument.
We would already generate isl_ast_op_pdiv_q instead of isl_ast_op_fdiv_q
if we detect that the first argument is non-negative.
Now, we also check if the first argument is smaller than the second.
If so, we can rewrite
floor(e/d) = -ceil(-e/d) = -floor((-e + d - 1)/d)
and still use isl_ast_op_pdiv_q.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sat, 29 Sep 2012 08:18:55 +0000 (10:18 +0200)]
isl_ast_build_expr.c: separate out extract_modulo function
This will make it easier to improve the extraction of modulos
in a subsequent commit.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sat, 29 Sep 2012 10:02:01 +0000 (12:02 +0200)]
isl_ast_build_expr.c: fix typo in comment
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Sat, 29 Sep 2012 09:11:19 +0000 (11:11 +0200)]
isl_ast_build_expr.c: remove remnants of code removed before inclusion
The AST generation code has undergone many changes before it was
included in the master branch. Some references to old code got left in
by mistake.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
Sven Verdoolaege [Fri, 28 Sep 2012 17:36:50 +0000 (19:36 +0200)]
add extra AST generation test case from CLooG
The test case was not added before because the output had not been
tested yet.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>