This is the mail archive of the
gcc-regression@gcc.gnu.org
mailing list for the GCC project.
Regression for lno on i686-pc-linux-gnu
- From: Autocrasher <autocrasher at devphil dot com>
- To: gcc-regression at gcc dot gnu dot org, dberlin at dberlin dot org, dorit at il dot ibm dot com, kbesaw at us dot ibm dot com, olga at il dot ibm dot com, pop at cri dot ensmp dot fr, schwab at suse dot de
- Date: Sun, 5 Sep 2004 04:58:57 -0400
- Subject: Regression for lno on i686-pc-linux-gnu
3 new FAIL's (20 total) have appeared for i686-pc-linux-gnu:
gcc.sum gcc.dg/tree-ssa-chrec/ssa-chrec-53.c
gcc.sum gcc.dg/tree-ssa-chrec/ssa-chrec-60.c
gcc.sum gcc.dg/tree-ssa-chrec/ssa-chrec-62.c
These tests passed the last time they were ran.
Attached are the ChangeLogs since the previous run. Possibly a patch with
your name on it has caused the problem(s).
The full transcript is at http://www.devphil.com/build/
Specific problem logs at http://www.devphil.com/build/reports/tests-lno/
(explanations of each file are on the main page; you probably want summary*)
Host was an athlon-xp (i686) system. Target was specified as i686-pc-linux-gnu.
This particular run used standard (previously installed) binutils.
This email is generated by a machine, not sent from a person. If you think
it's been sent to you in error, please see the gcc-regression list archives
for possible followup discussion; this could be legitimate or it could be a
false positive. Also watch for email from the other regression testers and
check the logs from the known autobuilders.
If you write a response, please respect the Reply-To: header.
With love,
The Autocrasher (Phil's Builder Bot, version 2.1)
=========================================================================
diff -u1r previous/CL/gcc/ChangeLog.lno current/CL/gcc/ChangeLog.lno
--- previous/CL/gcc/ChangeLog.lno 2004-08-28 06:45:03.000000000 -0400
+++ current/CL/gcc/ChangeLog.lno 2004-09-04 06:45:03.000000000 -0400
@@ -1 +1,172 @@
+2004-09-03 Sebastian Pop <pop@cri.ensmp.fr>
+
+ * Makefile.in (tree-ssa-loop-niter.o): Depends on TREE_DATA_REF_H.
+ * cfgloop.h (struct nb_iter_bound): Moved here.
+ (record_estimate_niter): Declared extern here.
+ * tree-data-ref.c (compute_estimated_nb_iterations): New.
+ (estimate_niter_from_size_of_data): Pass in the statement containing
+ the data reference. Record the estimations in loop instead of
+ computing the minimum of the loop bounds. Check for null pointers
+ before computing the estimation.
+ (analyze_array_indexes): Pass in the statement containing the data
+ reference.
+ (analyze_array): Pass to analyze_array_indexes the statement.
+ (find_data_references_in_loop): Now extern. Compute and initialize
+ the estimated_nb_iterations.
+ * tree-data-ref.h (find_data_references_in_loop): Declared here.
+ * tree-ssa-loop-niter.c: Include tree-data-ref.h.
+ (struct nb_iter_bound): Moved to cfgloop.h.
+ (record_estimate): Renamed record_estimate_niter.
+ (estimate_numbers_of_iterations_loop): Add a call to the data
+ reference analyzer that finds the bounds based on the data size.
+
+ * tree-loop-linear.c (gather_interchange_stats): Don't compute the
+ inner loop, but instead, compute it. Pass in the offset of
+ the loop nest.
+ (try_interchange_loops): Pass to gather_interchange_stats the offset
+ of the loop nest.
+ (linear_transform_loops): Don't fail to transform a loop nest if the
+ loop nest itself has a sibling loop.
+
+2004-09-02 Sebastian Pop <pop@cri.ensmp.fr>
+
+ * tree-data-ref.c (array_base_name_differ_p): Fix comments
+ formating.
+
+2004-09-02 Sebastian Pop <pop@cri.ensmp.fr>
+
+ * tree-data-ref.c (build_classic_dist_vector,
+ build_classic_dir_vector): Don't record information on loops
+ outer that the analyzed nest.
+
+2004-09-02 Sebastian Pop <pop@cri.ensmp.fr>
+
+ * tree-loop-linear.c (gather_interchange_stats): Gather the
+ sum of the access functions only for the inner loops.
+
+2004-09-02 Olga Golovanevsky <olga@il.ibm.com>
+
+ * tree-vectorizer.c (get_phi_node_for_def): Removed.
+ (update_phi_nodes_for_guard, tree_duplicate_loop_to_edge_cfg): Updated.
+ (tree_duplicate_loop_to_edge): Updated.
+ (vect_generate_tmps_on_preheader): Treating also case of constant.
+ (vect_transform_loop): Treating also case of constant
+ non-divisibale by VF.
+ (vect_analyze_operations): Remove redundant check.
+
+2004-09-02 Daniel Berlin <dberlin@dberlin.org>
+
+ * lambda-code.c (gcc_loop_to_lambda_loop): Handle GT_EXPR and GE_EXPR.
+ Get rid of unhandled loop exit test stuff.
+ Handle induction variable being on right side of test.
+ (find_induction_var_from_exit_cond): Handle induction variable
+ being on right side of test.
+
+2004-09-02 Sebastian Pop <pop@cri.ensmp.fr>
+
+ * tree-data-ref.c (analyze_array_indexes): Determine the estimation
+ of number of iteration also when the number_of_iterations_in_loop
+ contains undetermined elements.
+
+2004-09-02 Sebastian Pop <pop@cri.ensmp.fr>
+
+ * tree-loop-linear.c (gather_interchange_stats): Add more comments.
+ Gather also strides of accessed data. Pass in the data references
+ array.
+ (try_interchange_loops): Add a new heuristic for handling the temporal
+ locality. Pass in the data references array.
+ (linear_transform_loops): Pass the data references array to
+ try_interchange_loops.
+
+2004-09-02 Andreas Schwab <schwab@suse.de>
+
+ * tree-loop-linear.c (linear_transform_loops): Fix call to
+ dump_dist_dir_vectors.
+
+2004-09-01 Sebastian Pop <pop@cri.ensmp.fr>
+
+ * cfgloop.h (struct loop): New fields parallel_p, and
+ estimated_nb_iterations.
+ * tree-chrec.c (nb_vars_in_chrec): New.
+ * tree-chrec.h (nb_vars_in_chrec): Declared here.
+ * tree-data-ref.c (tree_fold_divides_p): Don't test for a equals one,
+ this test is already in tree_fold_gcd.
+ (gcd): For integers, yet again a copy defined static.
+ (int_divides_p): Same for integers.
+ (dump_data_dependence_relation): Fix dumping. Dump dir and dist
+ vectors.
+ (dump_dist_dir_vectors): Don't pass in the size of the
+ vectors; it is now read from DDR_SIZE_VECT.
+ (dump_ddrs, estimate_niter_from_size_of_data, all_chrecs_equal_p): New.
+ (compute_distance_vector): Renamed compute_subscript_distance.
+ Fix comment.
+ (compute_distance_vector): Handle the case where the description of
+ a multivariate function is a TREE_VEC.
+ (analyze_ziv_subscript, analyze_siv_subscript_cst_affine,
+ analyze_siv_subscript, analyze_miv_subscript,
+ analyze_overlapping_iterations, ):
+ Pass in a pointer to last_conflicts, and initialize it.
+ (initialize_matrix_A, FLOOR, compute_overlap_steps_for_affine_univar,
+ compute_overlap_steps_for_affine_1_2): New.
+ (analyze_siv_subscript_affine_cst): Removed, replaced its uses with
+ analyze_siv_subscript_cst_affine.
+ (analyze_subscript_affine_affine): Pass in a pointer to
+ last_conflicts, and initialize it. Add a special case for avoiding
+ the computation of the Hermite normal form when there exist a trivial
+ solution: when gamma is zero. Rewrite the dependence tester for using
+ fewer tree nodes, it now uses integers for the computations.
+ (no_conflicts_p): Fix typo: return true when overlaps is chrec_known.
+ (subscript_dependence_tester): Compute the last conflict. Initialize
+ SUB_LAST_CONFLICT.
+ (build_classic_dist_vector, build_classic_dir_vector): Conflict
+ functions don't store the loop number that carry the dependence.
+ Retrieve this information from the array access function.
+ Initialize DDR_SIZE_VECT.
+ (find_data_references_in_loop): Detect parallel loops when all the
+ data accesses are in read only mode. Insert a single chrec_dont_know
+ node in the data dependence graph.
+ * tree-data-ref.h (struct data_dependence_relation): Add a field
+ size_vect.
+ (DDR_SIZE_VECT): New.
+ (dump_ddrs): Declared here.
+ (dump_dist_dir_vectors): Modify the declaration.
+ (gather_interchange_stats): Fix leading comment. Gather also the
+ access strides, and consequently pass in the datarefs array from which
+ this information is extracted. Correctly parenthesize the access to
+ nb_deps_not_carried_by_loop such that its value is incremented.
+ Modify the constants, such that in the unknown and known cases the
+ dependence steps are not increased.
+ (try_interchange_loops): Pass in the datarefs array. Add more
+ comments. Add a test for temporal locality: inner loops
+ should have smallest array access strides.
+ * tree-pretty-print.c (dump_generic_node): BINFO_TYPE has nothing to do
+ with the TREE_VEC node. Correctly dump the TREE_VEC node printing
+ all its elements.
+
+2004-08-31 Keith Besaw <kbesaw@us.ibm.com>
+
+ * tree-vectorizer.h (struct _loop_vec_info): Add fields ptr_mask,
+ may_misalign_stmts, and num_blocks
+ * tree-vectorizer.c (vect_compute_data_refs_alignment): Return
+ bool so caller knows when no vectorization is possible.
+ (vect_compute_data_ref_alignment): Return bool and rearrange code
+ so false returns, no vectorization possible, are performed first.
+ (vect_analyze_data_refs):
+ (vect_create_index_for_array_ref): Replaced with
+ vect_create_index_for_vector_ref.
+ (vect_create_addr_base_for_vector_ref): New function, contains code from
+ vect_create_data_ref so that it may be used by that procedure and
+ by vect_create_cond_for_align_checks.
+ (vect_create_cond_for_align_checks): New function.
+ (new_loop_vec_info): initialize added _loop_vec_info fields.
+ (destroy_loop_vec_info): Use/clear added _loop_vec_info fields.
+ (vect_create_data_ref): Changed to use vect_create_index_for_vector_ref
+ and vect_create_addr_base_for_vector_ref.
+ (vect_generate_tmps_on_preheader): Fix comment.
+ (vect_transform_loop): Add calls to tree_ssa_loop_version and
+ vect_create_cond_for_align_checks to implement the vector loop
+ versioning for alignment.
+ (vect_analyze_operations, vect_mark_stmts_to_be_vectorized,
+ vect_analyze_data_refs): Use num_blocks field from _loop_vec_info.
+
2004-08-25 Sebastian Pop <pop@cri.ensmp.fr>
diff -u1r previous/CL/gcc/testsuite/ChangeLog.lno current/CL/gcc/testsuite/ChangeLog.lno
--- previous/CL/gcc/testsuite/ChangeLog.lno 2004-08-24 06:45:19.000000000 -0400
+++ current/CL/gcc/testsuite/ChangeLog.lno 2004-09-04 06:45:20.000000000 -0400
@@ -1 +1,14 @@
+2004-09-02 Olga Golovanevsky <olga@il.ibm.com>
+ Dorit Naishlos <dorit@il.ibm.com>
+
+ * gcc.dg/vect/vect-69.c: Remove redundant test.
+ * gcc.dg/vect/vect-70.c: New test.
+ * gcc.dg/vect/vect-71.c: New test.
+
+2004-08-31 Keith Besaw <kbesaw@us.ibm.com>
+
+ * gcc.dg/vect/vect-33.c: Remove xfail *-*-*, now vectorized.
+ * gcc.dg/vect/vect-44.c: Remove xfail *-*-*, now vectorized.
+ * gcc.dg/vect/vect-48.c: Remove xfail *-*-*, now vectorized.
+
2004-08-23 Dorit Naishlos <dorit@il.ibm.com>