OSDN Git Service

* ira-color.c (allocno_reload_assign): Avoid accumulating
[pf3gnuchains/gcc-fork.git] / gcc / ChangeLog
index addc094..18374f0 100644 (file)
@@ -1,3 +1,144 @@
+2010-04-19  Vladimir Makarov  <vmakarov@redhat.com>
+
+       * ira-color.c (allocno_reload_assign): Avoid accumulating
+       reload registers in ALLOCNO_TOTAL_CONFLICT_HARD_REGS.
+
+2010-04-19  Martin Jambor  <mjambor@suse.cz>
+
+       * gimple.h (create_tmp_reg): Declare.
+       * gimplify.c (create_tmp_reg): New function.
+       (gimplify_return_expr): Use create_tmp_reg.
+       (gimplify_omp_atomic): Likewise.
+       (gimple_regimplify_operands): Likewise.
+       * tree-dfa.c (make_rename_temp): Likewise.
+       * tree-predcom.c (predcom_tmp_var): Likewise.
+       (reassociate_to_the_same_stmt): Likewise.
+       * tree-sra.c (replace_uses_with_default_def_ssa_name): Likewise.
+       (get_replaced_param_substitute): Likewise.
+       * tree-ssa-phiprop.c (phiprop_insert_phi): Likewise.
+       * tree-ssa-phiopt.c (cond_store_replacement): Likewise.
+       * tree-ssa-pre.c (get_representative_for): Likewise.
+       (create_expression_by_pieces): Likewise.
+       * tree-tailcall.c (adjust_return_value_with_ops): Likewise.
+       (create_tailcall_accumulator): Likewise.
+
+2010-04-19  Martin Jambor  <mjambor@suse.cz>
+
+       * cgraphunit.c (cgraph_redirect_edge_call_stmt_to_callee): Update
+       new_stmt.
+       (cgraph_materialize_all_clones): Assert !need_ssa_update_p.
+
+2010-04-19  Richard Guenther  <rguenther@suse.de>
+
+       PR tree-optimization/43796
+       * tree-vrp.c (adjust_range_with_scev): Lookup init and step
+       from SCEV in the lattice.
+       (vrp_visit_phi_node): Dump change.
+
+2010-04-19  Richard Guenther  <rguenther@suse.de>
+
+       * configure.ac: Fix quoting around elf_getshstrndx ABI check.
+       * configure: Re-generated.
+
+2010-04-19  Richard Guenther  <rguenther@suse.de>
+
+       PR tree-optimization/43783
+       * tree-ssa-pre.c (create_component_ref_by_pieces_1): Drop
+       constant ARRAY_REF operands two and three if possible.
+
+2010-04-19  Uros Bizjak  <ubizjak@gmail.com>
+
+       PR target/43766
+       * config/i386/i386.c (ix86_decompose_address): Handle ASHIFT addends.
+
+2010-04-19  Jie Zhang  <jie@codesourcery.com>
+
+       PR target/43662
+       * reginfo.c (reinit_regs): Set caller_save_initialized_p to false.
+
+2010-04-19 Ira Rosen <irar@il.ibm.com>
+
+       PR tree-optimization/37027
+       * tree-vectorizer.h (struct _loop_vec_info): Add new field reductions 
+       and macro to access it.
+       (vectorizable_reduction): Add argument.
+       (vect_get_slp_defs): Likewise.
+       * tree-vect-loop.c (vect_analyze_scalar_cycles_1): Collect reduction
+       statements for possible use in SLP.
+       (new_loop_vec_info): Initialize LOOP_VINFO_REDUCTIONS.
+       (destroy_loop_vec_info): Free LOOP_VINFO_REDUCTIONS.
+       (vect_create_epilog_for_reduction): Handle SLP. Modify documentation,
+       add new argument.
+       (vectorizable_reduction): Likewise.
+       * tree-vect-stmts.c (vect_get_vec_defs): Update call to 
+       vect_get_slp_defs.
+       (vectorizable_type_demotion, vectorizable_type_promotion,
+       vectorizable_store): Likewise.
+       (vect_analyze_stmt): Update call to vectorizable_reduction.
+       (vect_transform_stmt): Likewise.
+       * tree-vect-slp.c (vect_get_and_check_slp_defs): Handle reduction.
+       (vect_build_slp_tree): Fix indentation. Check that there are no loads
+       from different interleaving chains in same node.
+       (vect_slp_rearrange_stmts): New function.
+       (vect_supported_load_permutation_p): Allow load permutations for 
+       reductions. Call vect_slp_rearrange_stmts() to rearrange statements
+       inside SLP nodes if necessary.
+       (vect_analyze_slp_instance): Handle reductions.
+       (vect_analyze_slp): Try to build SLP instances originating from groups
+       of reductions.
+       (vect_detect_hybrid_slp_stmts): Skip reduction statements.
+       (vect_get_constant_vectors): Create initial vectors for reductions
+       according to reduction code. Add new argument.
+       (vect_get_slp_defs): Add new argument, pass it to 
+       vect_get_constant_vectors.
+       (vect_schedule_slp_instance): Remove SLP tree root statements.
+
+2010-04-19  Jakub Jelinek  <jakub@redhat.com>
+
+       * tree.h (ENUM_IS_SCOPED): Define.
+       * dwarf2out.c (gen_enumeration_type_die): Add DW_AT_enum_class
+       for ENUM_IS_SCOPED enums.
+
+2010-04-18  Eric Botcazou  <ebotcazou@adacore.com>
+
+       * fold-const.c (fold_comparison): Use ssizetype.
+       * gimple-fold.c (maybe_fold_offset_to_array_ref): Likewise.
+       * ipa-prop.c (ipa_modify_call_arguments): Use sizetype.
+       * tree-loop-distribution.c (build_size_arg_loc): Likewise.
+       * tree-object-size.c (compute_object_sizes): Use size_type_node.
+
+       * tree.h (initialize_sizetypes): Remove parameter.
+       (build_common_tree_nodes): Remove second parameter.
+       * stor-layout.c (initialize_sizetypes): Remove parameter.
+       Always create an unsigned type.
+       (set_sizetype): Assert that the passed type is unsigned and simplify.
+       * tree.c (build_common_tree_nodes): Remove second parameter.
+       Adjust call to initialize_sizetypes.
+       * c-decl.c (c_init_decl_processing): Remove second argument in call to
+       build_common_tree_nodes.
+
+2010-04-18  Matthias Klose  <doko@ubuntu.com>
+
+       * gcc.c (main): Search for liblto_plugin.so with mode R_OK.
+
+2010-04-18 Ira Rosen <irar@il.ibm.com> 
+
+       PR tree-optimization/43771
+       * tree-vect-slp.c (vect_supported_load_permutation_p): Check that
+       load permutation doesn't have gaps.
+
+2010-04-18  Jan Hubicka  <jh@suse.cz>
+
+       * i386.md (UNSPEC_SSE_PROLOGUE_SAVE_LOW): New.
+       (sse_prologue_save_insn expander): Use new pattern.
+       (sse_prologue_save_insn1): New pattern and splitter.
+       (sse_prologue_save_insn): Update to deal also with 64bit aligned
+       blocks.
+       * i386.c (setup_incoming_varargs_64): Do not compute jump
+       destination here.
+       (ix86_gimplify_va_arg): Update alignment needed.
+       (ix86_local_alignment): Do not align all local arrays to 128bit.
+
 2010-04-17  Jan Hubicka  <jh@suse.cz>
 
        * ipa-inline.c (cgraph_early_inlining): Handle flattening too.