re PR tree-optimization/49518 (ICE in vect_enhance_data_refs_alignment, at tree-vect-data-refs.c:1555)

2011-07-05  Richard Guenther  <rguenther@suse.de>

PR tree-optimization/49518
PR tree-optimization/49628
* tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Skip
irrelevant and invariant data-references.
(vect_analyze_data_ref_access): For invariant loads clear the
group association.

* g++.dg/torture/pr49628.C: New testcase.
* gcc.dg/torture/pr49518.c: Likewise.

From-SVN: r175847

run_acats (which): Extract last field from type -p, type output only if command succeeded.

* ada/acats/run_acats (which): Extract last field from type -p,
type output only if command succeeded.

From-SVN: r175843

Daily bump.

From-SVN: r175840

cp-demangle.c (d_expression): Handle 'this'.

* cp-demangle.c (d_expression): Handle 'this'.
(d_print_comp) [DEMANGLE_COMPONENT_FUNCTION_PARAM]: Likewise.

From-SVN: r175836

DR 1207 PR c++/49589

DR 1207
PR c++/49589
* mangle.c (write_expression): Handle 'this'.
* parser.c (cp_parser_postfix_dot_deref_expression): Allow
incomplete *this.
* semantics.c (potential_constant_expression_1): Check that
DECL_CONTEXT is set on 'this'.

From-SVN: r175835

error.c (dump_template_bindings): Don't print typenames for a partial instantiation.

* error.c (dump_template_bindings): Don't print typenames
for a partial instantiation.
(dump_function_decl): If we aren't printing function arguments,
print template arguments as <args> rather than [with ...].
(dump_expr): Don't print return type or template header.
[BASELINK]: Use BASELINK_FUNCTIONS rather than get_first_fn.
* pt.c (dependent_template_arg_p): Handle null arg.

From-SVN: r175834

* error.c (type_to_string): Avoid redundant akas.

From-SVN: r175833

re PR fortran/49623 (Segfault of "gfortran -L" if no argument follows the "-L")

PR fortran/49623
* gfortranspec.c (lang_specific_driver): Ignore options with
CL_ERR_MISSING_ARG errors.

From-SVN: r175830

re PR rtl-optimization/49619 (ICE in simplify_subreg, at simplify-rtx.c:5362)

PR rtl-optimization/49619
* combine.c (combine_simplify_rtx): In PLUS -> IOR simplification
pass VOIDmode as op0_mode to recursive call, and return temp even
when different from tor, just if it is not IOR of the original
PLUS arguments.

* gcc.dg/pr49619.c: New test.

From-SVN: r175825

re PR rtl-optimization/49472 (Compiler segfault on valid code)

PR rtl-optimization/49472
* simplify-rtx.c (simplify_unary_operation_1) <case NEG>: When
negating MULT, negate the second operand instead of first.
(simplify_binary_operation_1) <case MULT>: If one operand is
a NEG and the other is MULT, don't attempt to optimize by
negation of the MULT operand if it only moves the NEG operation
around.

* gfortran.dg/pr49472.f90: New test.

From-SVN: r175821

re PR debug/49602 (verify_ssa failed (definition does not dominate use) with "-O2 -g")

PR debug/49602
* tree-into-ssa.c (rewrite_debug_stmt_uses): Disregard
get_current_def return value if it can't be trusted to be
the current value of the variable in the current bb.

* gcc.dg/pr49602.c: New test.

From-SVN: r175818

re PR target/49600 (Bad SSE2 int->float split in i386.md)

PR target/49600
* config/i386/i386.md (SSE2 int->float split): Push operand 1 in
general register to memory for !TARGET_INTER_UNIT_MOVES.

From-SVN: r175817

re PR target/34734 ([avr] attribute((progmem)) not handled properly in C++)

PR target/34734
PR target/44643
* gcc.target/avr/avr.exp: Run over cpp files, too.
* gcc.target/avr/torture/avr-torture.exp: Ditto.
* gcc.target/avr/progmem.h: New file.
* gcc.target/avr/exit-abort.h: New file.
* gcc.target/avr/progmem-error-1.c: New file.
* gcc.target/avr/progmem-error-1.cpp: New file.
* gcc.target/avr/progmem-warning-1.c: New file.
* gcc.target/avr/torture/progmem-1.c: New file.
* gcc.target/avr/torture/progmem-1.cpp: New file.

From-SVN: r175811

re PR target/44643 ([avr] ICE in c-typeck.c)

PR target/44643
* config/avr/avr.c (avr_insert_attributes): Use TYPE_READONLY
instead of TREE_READONLY.

From-SVN: r175809

extend.texi (AVR Built-in Functions): Update documentation of __builtin_avr_fmul*.

* doc/extend.texi (AVR Built-in Functions): Update documentation
of __builtin_avr_fmul*.
* config/avr/avr.c (avr_init_builtins): Don't depend on
AVR_HAVE_MUL.
* config/avr/avr-c.c (avr_cpu_cpp_builtins): Ditto.
* config/avr/avr.md (fmul): Rename to fmul_insn.
(fmuls): Rename to fmuls_insn.
(fmulsu): Rename to fmulsu_insn.
(fmul,fmuls,fmulsu): New expander.
(*fmul.call,*fmuls.call,*fmulsu.call): New Insn.
* config/avr/t-avr (LIB1ASMFUNCS): Add _fmul, _fmuls, _fmulsu.
* config/avr/libgcc.S (__fmul): New function.
(__fmuls): New function.
(__fmulsu,__fmulsu_exit): New function.

From-SVN: r175807

re PR tree-optimization/49615 (internal compiler error: verify_stmts failed / LHS in noreturn call with pointer-to-never-returning-member)

2011-07-04  Richard Guenther  <rguenther@suse.de>

PR tree-optimization/49615
* tree-cfgcleanup.c (split_bbs_on_noreturn_calls): Fix
basic-block index check.

* g++.dg/torture/pr49615.C: New testcase.

From-SVN: r175803

t-avr (intfuncs16): Add _clrsbXX2.

libgcc/
* config/avr/t-avr (intfuncs16): Add _clrsbXX2.

gcc/
* longlong.h (count_leading_zeros, count_trailing_zeros,
COUNT_LEADING_ZEROS_0): Define for target avr if W_TYPE_SIZE is 16
resp. 64.

From-SVN: r175800

In gcc/objc/: 2011-07-04 Nicola Pero <nicola.pero@meta-innovation.com>

In gcc/objc/:
2011-07-04  Nicola Pero  <nicola.pero@meta-innovation.com>

Refactored encoding code into objc-encoding.h and objc-encoding.c.
* objc-act.c (util_obstack, util_firstobj, encode_type_qualifiers,
encode_type, generating_instance_variables, objc_method_parm_type,
objc_encoded_type_size, encode_method_prototype,
objc_build_encode_expr, pointee_is_readonly, encode_pointer,
encode_array, encode_vector, encode_aggregate_fields,
encode_aggregate_within, encode_next_bitfield,
encode_gnu_bitfield, encode_field_decl,
objc_v2_encode_property_attr): Moved to objc-encoding.h and
objc-encoding.c.  No change in the actual code.
Include objc-encoding.h.
(objc_init): Added TODO.
(objc_build_property_setter_name): Made non-static so it can be
called from objc-encoding.c.
* objc-act.h (OBJC_ENCODE_INLINE_DEFS,
OBJC_ENCODE_DONT_INLINE_DEFS): Moved to objc-encoding.h.
* objc-runtime-shared-support.h (objc_v2_encode_property_attr,
encode_method_prototype, encode_field_decl,
generating_instance_variables): Moved to objc-encoding.h.
(objc_build_property_setter_name): Declare.
* objc-encoding.c: New.
* objc-encoding.h: New.
* objc-gnu-runtime-abi-01.c: Include objc-encoding.h.
* objc-next-runtime-abi-01.c: Likewise.
* objc-next-runtime-abi-02.c: Likewise.
* objc-runtime-shared-support.c: Likewise.
* Make-lang.in (OBJC_OBJS): Added objc-encoding.o.
(objc/objc-lang.o): Reordered dependencies.
(objc/objc-runtime-shared-support.o): Reordered dependencies.
Added dependencies on objc-encoding.h and on $(GGC_H),
$(DIAGNOSTIC_CORE_H), $(FLAGS_H) and input.h.
(objc/objc-gnu-runtime-abi-01.o): Likewise.
(objc/objc-next-runtime-abi-01.o): Likewise.
(objc/objc-next-runtime-abi-02.o): Likewise.
(objc/objc-act.o): Reordered dependencies.  Added dependency on
objc-encoding.h.
(objc/objc-encoding.o): New rule.

* objc-encoding.c (encode_type): Use "%<%T%>" format when printing
warning "unknown type %<%T%> found during Objective-C encoding"
instead of using gen_type_name.

In gcc/objcp/:
2011-07-04  Nicola Pero  <nicola.pero@meta-innovation.com>

* Make-lang.in (OBJCXX_OBJS): Added objc-encoding.o.
(objcp/objcp-lang.o): Reordered dependencies.  Depend on GGC_H.
(objcp/objcp-decl.o): Reordered dependencies.
(objcp/objc-runtime-shared-support.o): Reordered dependencies.
Updated them to be identical to the corresponding new objc/ ones,
with the addition of objcp-decl.h.
(objcp/objc-runtime-shared-support.o): Likewise.
(objcp/objc-gnu-runtime-abi-01.o): Likewise.
(objcp/objc-next-runtime-abi-01.o): Likewise.
(objcp/objc-next-runtime-abi-02.o): Likewise.
(objcp/objcp-act.o): Reordered dependencies.  Added dependency on
objc-encoding.h.
(objcp/objc-encoding.o): New rule.

From-SVN: r175797

Daily bump.

From-SVN: r175793

PR 49296 Fix wrong PR number in ChangeLog

From-SVN: r175789

re PR tree-optimization/49610 (Segfault with -ftree-vectorize (or -O3))

        PR tree-optimization/49610
        * tree-vect-loop.c (vect_is_slp_reduction): Check that DEF_STMT has
        a basic block.

From-SVN: r175786

Daily bump.

From-SVN: r175785

tree.h (TYPE_ARTIFICIAL): New flag.

* tree.h (TYPE_ARTIFICIAL): New flag.
* dwarf2out.c (modified_type_die): Add a DW_AT_artificial attribute to
the DIE of the type if it is artificial.
(gen_array_type_die): Likewise.
(gen_enumeration_type_die): Likewise.
(gen_struct_or_union_type_die): Likewise.
* lto-streamer-in.c (unpack_ts_base_value_fields): Use TYPE_ARTIFICIAL.
* lto-streamer-out.c (pack_ts_base_value_fields): Likewise.
ada/
* gcc-interface/utils.c (record_builtin_type): Set TYPE_ARTIFICIAL on
the type according to the ARTIFICIAL_P parameter.
(create_type_decl): Likewise.
(create_type_stub_decl): Set TYPE_ARTIFICIAL on the type to 1.

Co-Authored-By: Nicolas Setton <setton@adacore.com>
Co-Authored-By: Olivier Hainque <hainque@adacore.com>
From-SVN: r175781

re PR fortran/49562 ([OOP] assigning value to type-bound function)

2011-07-02  Janus Weil  <janus@gcc.gnu.org>

PR fortran/49562
* expr.c (gfc_check_vardef_context): Handle type-bound procedures.

2011-07-02  Janus Weil  <janus@gcc.gnu.org>

PR fortran/49562
* gfortran.dg/typebound_proc_23.f90: New.

From-SVN: r175779

Daily bump.

From-SVN: r175775

re PR c++/49605 (-Wdelete-non-virtual-dtor is not picky enough)

2011-07-01  Jonathan Wakely  <jwakely.gcc@gmail.com>

PR c++/49605
* init.c (build_delete): Only warn for sfk_deleting_destructor.

From-SVN: r175771

Make-lang.in (gnat1): Prepend '+' to the command.

* gcc-interface/Make-lang.in (gnat1): Prepend '+' to the command.
(gnatbind): Likewise.

From-SVN: r175769

tree-object-size.c (pass_through_call): Handle BUILT_IN_ASSUME_ALIGNED.

* tree-object-size.c (pass_through_call): Handle
BUILT_IN_ASSUME_ALIGNED.

From-SVN: r175768

Make-lang.in (cp/decl.o): Depend on pointer-set.h.

* Make-lang.in (cp/decl.o): Depend on pointer-set.h.
(cp/class.o): Likewise.
(cp/error.o): Likewise.
(cp/name-lookup.o): Likewise.
(cp/decl2.o): Likewise.  Don't depend on $(POINTER_SET_H).

From-SVN: r175767

re PR c++/48261 (internal compiler error: in lookup_template_function, at cp/pt.c:6227)

PR c++/48261
* pt.c (lookup_template_function): Handle non-function.

From-SVN: r175766

re PR c++/48593 (Wrong return type when applying address operator to inherited, template dependend member, using a typedef)

PR c++/48593
* pt.c (tsubst_qualified_id): Check PTRMEM_OK_P.
* tree.c (build_qualified_name): Set PTRMEM_OK_P.
* semantics.c (finish_parenthesized_expr): Clear PTRMEM_OK_P on
SCOPE_REF, too.
* cp-tree.h (PTRMEM_OK_P): Apply to SCOPE_REF, too.
(QUALIFIED_NAME_IS_TEMPLATE): Switch to lang flag 1.

From-SVN: r175765

re PR c++/48883 (?: ternary operator fails in certain contexts - link error)

PR c++/48883
PR c++/49609
* pt.c (resolve_nondeduced_context): Call mark_used.

From-SVN: r175764

Darwin has case-insensitive filesystems

HFS+, the FS on Darwin, is case insensitive. So this patch adjusts
filename_cmp.c to ignore the casing when comparing filenames on Darwin.

include/ChangeLog:

        * filenames.h (HAVE_CASE_INSENSITIVE_FILE_SYSTEM): Define
        on Darwin, as well as on the systems that use a DOS-like
        filesystem.

libiberty/ChangeLog:

        * filename_cmp.c (filename_cmp, filename_ncmp): Add handling of
        HAVE_CASE_INSENSITIVE_FILE_SYSTEM.

From-SVN: r175762

re PR debug/49408 (member function template id not matching linkage name)

libiberty/
PR debug/49408
* cp-demangle.c (d_print_comp): Suppress argument list for function
references by the '&' unary operator.  Keep also already processed
variant without the argument list.  Suppress argument list types for
function call used in an expression.
* testsuite/demangle-expected: Fix excessive argument list types in
`test for typed function in decltype'.  New testcase for no argument
list types printed.  3 new testcases for function references by the
'&' unary operator..

From-SVN: r175761

tree-sra.c (tree_non_mode_aligned_mem_p): Also ignore MEM_REFs.

2011-07-01  Martin Jambor  <mjambor@suse.cz>

* tree-sra.c (tree_non_mode_aligned_mem_p): Also ignore MEM_REFs.

From-SVN: r175760

re PR c++/49085 (Crash with SIGSEGV during compilation.)

PR c++/49085
* semantics.c (finish_offsetof): Complain about incomplete type.

From-SVN: r175758

Use proper mode for stack save area.

2011-07-01  H.J. Lu  <hongjiu.lu@intel.com>

PR middle-end/48016
* explow.c (update_nonlocal_goto_save_area): Use proper mode
for stack save area.
* function.c (expand_function_start): Likewise.

From-SVN: r175756

config-list.mk (LIST): Append OPT-enable-obsolete to alpha-dec-osf5.1, mips-sgi-irix6.5.

* config-list.mk (LIST): Append OPT-enable-obsolete to
alpha-dec-osf5.1, mips-sgi-irix6.5.

From-SVN: r175754

re PR middle-end/49596 (FAIL: gcc.dg/torture/pr43879_1.c)

2011-07-01  Richard Guenther  <rguenther@suse.de>

PR middle-end/49596
* cgraph.h (varpool_all_refs_explicit_p): Not analyzed nodes
may have unknown refs.

From-SVN: r175753

tree-ssa-forwprop.c (simplify_bitwise_binary): Fix typo.

2011-07-01  Kai Tietz  <ktietz@redhat.com>

        * tree-ssa-forwprop.c (simplify_bitwise_binary): Fix typo.

2011-07-01  Kai Tietz  <ktietz@redhat.com>

        * gcc.dg/tree-ssa/bitwise-sink.c: New test.

From-SVN: r175752

config.gcc: Obsolete alpha*-dec-osf5.1, mips-sgi-irix6.5.

* config.gcc: Obsolete alpha*-dec-osf5.1, mips-sgi-irix6.5.
* doc/install.texi (Specific, alpha*-dec-osf5.1): Document it.
(Specific, mips-sgi-irix6): Likewise.

From-SVN: r175751

re PR libmudflap/49549 (Use of --noinhibit-exec is unportable)

libmudflap:
PR libmudflap/49549
* testsuite/lib/libmudflap.exp (load_gcc_lib): Load
target-supports.exp.
* testsuite/libmudflap.cth/cthfrags.exp: Only pass
--noinhibit-exec to GNU ld.

gcc:
PR libmudflap/49549
* doc/sourcebuild.texi (Effective-Target Keywords): Document gld.

gcc/testsuite:
PR libmudflap/49549
* lib/target-supports.exp (check_effective_target_gld): New proc.

From-SVN: r175749

tree-pretty-print.c (dump_generic_code): Print [idx]= and [idx1 ...

* tree-pretty-print.c (dump_generic_code) <case CONSTRUCTOR>: Print
[idx]= and [idx1 ... idx2]= before initializers if needed for
array initializers.

From-SVN: r175748

config.gcc (score-*-elf): Remove score7.o.

2011-07-01  Chen Liqin  <liqin.gcc@gmail.com>

        * config.gcc (score-*-elf): Remove score7.o.
        * config/score/t-score-elf: Likewise.
        * config/score/score.c: Merge score7 to score.c and remove forwarding functions.
        * config/score/score7.c: Deleted.
        * config/score/score7.h: Deleted.

From-SVN: r175747

re PR middle-end/49603 (177.mesa in SPEC CPU 2000 failed to build)

2011-07-01  Richard Guenther  <rguenther@suse.de>

PR tree-optimization/49603
* tree-vect-stmts.c (vectorizable_load): Remove unnecessary
assert.

* gcc.dg/torture/pr49603.c: New testcase.

From-SVN: r175746

correct changelog

From-SVN: r175744

re PR c++/49387 (t.cxx:140: error: too many initializers for ‘const __class_type_info_pseudo’)

PR c++/49387
* rtti.c (get_pseudo_ti_index): Call complete_type.

From-SVN: r175743

Daily bump.

From-SVN: r175742

re PR c++/49569 (-std=gnu++0x causes segmentation fault)

PR c++/49569
* method.c (implicitly_declare_fn): Set DECL_PARM_LEVEL and
DECL_PARM_INDEX on rhs parm.

From-SVN: r175738

* pt.c (iterative_hash_template_arg): Use cp_tree_operand_length.

From-SVN: r175737

re PR c++/49355 (new T({""}) crashes G++ when struct T { std::string foobar };)

PR c++/49355
* tree.c (stabilize_init): Handle aggregate initialization.

From-SVN: r175736

tree-sra.c (struct access): Rename total_scalarization to grp_total_scalarization

2011-06-30  Martin Jambor  <mjambor@suse.cz>

* tree-sra.c (struct access): Rename total_scalarization to
grp_total_scalarization
(completely_scalarize_var): New function.
(sort_and_splice_var_accesses): Set total_scalarization in the
representative access.
(analyze_access_subtree): Propagate total scalarization accross the
tree, no holes in totally scalarized trees, simplify coverage
computation.
(analyze_all_variable_accesses): Call completely_scalarize_var instead
of completely_scalarize_record.

* testsuite/gcc.dg/tree-ssa/sra-12.c: New test.

From-SVN: r175733

re PR c++/48481 (C++ overloading memory hog)

PR c++/48481
* name-lookup.c (struct arg_lookup): Add fn_set.
(add_function): Check it.
(lookup_arg_dependent_1): Initialize it.

From-SVN: r175732

* g++.dg/cpp0x/named.C: Test subobject of xvalue.

From-SVN: r175731

i386: Always use TARGET_DEEP_BRANCH_PREDICTION.

While it could be possible to output_set_got such that we can
individually annotate the instructions, it's simpler to simply
admit that all processors currently being manufactured do want
deep branch prediction.  At which point all of the complication
simply goes away.

        * config/i386/i386.h (X86_TUNE_DEEP_BRANCH_PREDICTION): Remove.
        (TARGET_DEEP_BRANCH_PREDICTION): Remove.
        * config/i386/i386.c: Don't include dwarf2out.h.
        (initial_ix86_tune_features): Remove X86_TUNE_DEEP_BRANCH_PREDICTION.
        (output_set_got): Don't test TARGET_DEEP_BRANCH_PREDICTION, delete
        all code dead thereafter.  Don't do dwarf2out_flush_queued_reg_saves.
        (ix86_expand_prologue): Set REG_CFA_FLUSH_QUEUE on set_got insn.
        (machopic_output_stub): Don't test TARGET_DEEP_BRANCH_PREDICTION.

From-SVN: r175730

dwarf2out: Add REG_CFA_FLUSH_QUEUE.

        * reg-notes.def (REG_CFA_FLUSH_QUEUE): New.
        * dwarf2out.c (dwarf2out_frame_debug): Handle it.
        * final.c (final_scan_insn): Look for it, and invoke
        dwarf2out_frame_debug before the insn if found.

From-SVN: r175729

dwarf2out: Handle pc_rtx as return column in REG_CFA_OFFSET too.

        * dwarf2out.c (dwarf2out_frame_debug_cfa_offset): Allow PC_RTX
        as a source, and interpret that as DWARF_FRAME_RETURN_COLUMN.

From-SVN: r175728

dwarf2out: Convert regs_saved_in_regs to VEC.

        * dwarf2out.c (struct reg_saved_in_data): Provide a typedef.
        Define a vector of this type.
        (regs_saved_in_regs): Use a VEC.
        (num_regs_saved_in_regs): Remove.
        (compare_reg_or_pc): New.
        (record_reg_saved_in_reg): Split out from...
        (dwarf2out_flush_queued_reg_saves): ... here.
        (clobbers_queued_reg_save): Update for VEC.
        (reg_saved_in): Likewise.
        (dwarf2out_frame_debug_init): Likewise.
        (dwarf2out_reg_save_reg): Use record_reg_saved_in_reg.
        (dwarf2out_frame_debug_cfa_register): Likewise.

From-SVN: r175727

re PR tree-optimization/49572 (gcc.dg/tree-ssa/20030709-2.c scan-tree-dump-times cddce2 ".rtmem" 0)

PR tree-optimization/49572
* tree-ssa-dom.c (initialize_hash_element) <GIMPLE_SINGLE_RHS>: Use the
type of the RHS instead of that of the LHS for the expression type.

From-SVN: r175721

df-scan.c (df_get_entry_block_def_set): Use INCOMING_REGNO macro unconditionally.

* df-scan.c (df_get_entry_block_def_set): Use INCOMING_REGNO macro
unconditionally.

From-SVN: r175718

opts.c (finish_options): Do not disable IPA-PTA during ltrans.

2011-06-30  Richard Guenther  <rguenther@suse.de>

* opts.c (finish_options): Do not disable IPA-PTA during ltrans.
* tree-ssa-structalias.c (create_variable_info_for): Do not
add initial constraints for non-var-decls.  Properly handle
globals in other ltrans partitions.
(intra_create_variable_infos): Manually create constraints for
the fake no-alias parameter.
(ipa_pta_execute): Dump the cgraph, handle ltrans partitions properly
and assert there are no clones.

From-SVN: r175707

re PR tree-optimization/46787 (Does not vectorize loop with load from scalar variable)

2011-06-30  Richard Guenther  <rguenther@suse.de>

PR tree-optimization/46787
* tree-data-ref.c (dr_address_invariant_p): Remove.
(find_data_references_in_stmt): Invariant accesses are ok now.
* tree-vect-stmts.c (vectorizable_load): Handle invariant
loads.
* tree-vect-data-refs.c (vect_analyze_data_ref_access): Allow
invariant loads.

* gcc.dg/vect/vect-121.c: New testcase.

From-SVN: r175704

re PR tree-optimization/49094 (ARM aligned(1) attribute is sometimes dropped)

2011-06-30  Martin Jambor  <mjambor@suse.cz>

PR tree-optimization/49094
* tree-sra.c (tree_non_mode_aligned_mem_p): New function.
(build_accesses_from_assign): Use it.

* testsuite/gcc.dg/tree-ssa/pr49094.c: New test.

From-SVN: r175703

* de.po: Update.

From-SVN: r175698

tree-ssa-structalias.c (find_func_aliases_for_builtin_call): Fix handling of BUILT_IN_ASSUME_ALIGNED.

* tree-ssa-structalias.c (find_func_aliases_for_builtin_call): Fix
handling of BUILT_IN_ASSUME_ALIGNED.

From-SVN: r175695

re PR debug/49364 ([4.7 regressions] Empty .debug_abbrev sections confuse SGI nm)

PR debug/49364
* dwarf2out.c (output_abbrev_section): Don't return early
if abbrev_die_table_in_use is 1.
(dwarf2out_finish): Instead don't call output_abbrev_section
nor emit abbrev_section_label in that case.

From-SVN: r175694

re PR fortran/49540 (Memory-hog with large DATA stmt)

PR fortran/49540
* gfortran.h (gfc_constructor): Add repeat field.
* trans-array.c (gfc_conv_array_initializer): Handle repeat > 1.
* array.c (current_expand): Add repeat field.
(expand_constructor): Copy repeat.
* constructor.c (node_free, node_copy, gfc_constructor_get,
gfc_constructor_lookup): Handle repeat field.
(gfc_constructor_lookup_next, gfc_constructor_remove): New functions.
* data.h (gfc_assign_data_value): Add mpz_t * argument.
(gfc_assign_data_value_range): Removed.
* constructor.h (gfc_constructor_advance): Removed.
(gfc_constructor_lookup_next, gfc_constructor_remove): New prototypes.
* data.c (gfc_assign_data_value): Add REPEAT argument, handle it and
also handle overwriting a range with a single entry.
(gfc_assign_data_value_range): Removed.
* resolve.c (check_data_variable): Adjust gfc_assign_data_value
call.  Use gfc_assign_data_value instead of
gfc_assign_data_value_expr.

* gfortran.dg/pr49540-1.f90: New test.
* gfortran.dg/pr49540-2.f90: New test.

From-SVN: r175693

re PR ada/49511 (acats test setup fails on HP-UX using posix shell)

PR ada/49511
* ada/acats/run_acats (which): Use last field of type -p output.

From-SVN: r175690

re PR libobjc/36610 (objc_msg_sendv is broken for targets which pass argument via registers)

2011-06-28  Dominique d'Humieres  <dominiq@lps.ens.fr>
    Iain Sandoe  <iains@gcc.gnu.org>

PR libobjc/36610
* objc.dg/torture/forward-1.m: Remove dg-xfail-run-if.
Only skip on 64-bit *-*-darwin8* && !objc2.

Co-Authored-By: Iain Sandoe <iains@gcc.gnu.org>
From-SVN: r175689

v850.h (CPP_SPEC): Define __v850e__ when compiling for the V850E.

* config/v850/v850.h (CPP_SPEC): Define __v850e__ when compiling
for the V850E.

From-SVN: r175688

re PR tree-optimization/38752 (ICE in set_uids_in_ptset, at tree-ssa-structalias.c:4817)

2011-06-30  Richard Guenther  <rguenther@suse.de>

PR tree-optimization/38752
* gcc.c-torture/compile/pr38752.c: New testcase.

From-SVN: r175684

tree-ssa-structalias.c (find_func_aliases_for_builtin_call): Fix handling of BUILT_IN_MEMPCPY_CHK and BUILT_IN_STPCPY_CHK.

* tree-ssa-structalias.c (find_func_aliases_for_builtin_call): Fix
handling of BUILT_IN_MEMPCPY_CHK and BUILT_IN_STPCPY_CHK.

From-SVN: r175682

tree-vect-loop.c (vect_determine_vectorization_factor): Handle both pattern and original statements if necessary.

        * tree-vect-loop.c (vect_determine_vectorization_factor): Handle
        both pattern and original statements if necessary.
        (vect_transform_loop): Likewise.
        * tree-vect-patterns.c (vect_pattern_recog): Update documentation.
        * tree-vect-stmts.c (vect_mark_relevant): Add new argument.
        Mark the pattern statement only if the original statement doesn't
        have its own uses.
        (process_use): Call vect_mark_relevant with additional parameter.
        (vect_mark_stmts_to_be_vectorized): Likewise.
        (vect_get_vec_def_for_operand): Use vectorized pattern statement.
        (vect_analyze_stmt): Handle both pattern and original statements
        if necessary.
        (vect_transform_stmt): Don't store vectorized pattern statement
        in the original statement.
        (vect_is_simple_use_1): Use related pattern statement only if the
        original statement is irrelevant.
        * tree-vect-slp.c (vect_get_and_check_slp_defs): Likewise.

From-SVN: r175681

Daily bump.

From-SVN: r175678

re PR c++/49216 ([C++0x] ICE on compiling new-expression with braced-init-list for arrays)

PR c++/49216
* init.c (build_new_1): Pass {} down to build_vec_init.
(build_vec_init): Handle it.

From-SVN: r175674

DR 1207 PR c++/49003

DR 1207
PR c++/49003
* cp-tree.h (struct saved_scope): Add x_current_class_ptr,
x_current_class_ref.
(current_class_ptr, current_class_ref): Use them.
* decl.c (build_this_parm): Handle getting the class type.
* parser.c (cp_parser_late_return_type_opt): Set up 'this'
for use within the trailing return type.

From-SVN: r175671

pt.c (tsubst_decl): In unevaluated operand, don't tsubst DECL_INITIAL unless our type use auto.

* pt.c (tsubst_decl) [VAR_DECL]: In unevaluated operand,
don't tsubst DECL_INITIAL unless our type use auto.

From-SVN: r175670

set.h, [...]: Remove base class default constructor calls.

2011-06-29  François Dumont  <francois.cppdevs@free.fr>

* include/debug/set.h, unordered_map, multiset.h, forward_list,
unordered_set, vector, deque, string, list, multimap.h: Remove base
class default constructor calls.
* include/debug/map.h: Likewise and cleanup several redefinition of
base iterator typedef.

From-SVN: r175665

Auto-vectorizer generates 128-bit AVX insns by default for bdver1.

* config/i386/i386.opt (mprefer-avx128): Redefine the flag as a Mask option.
* config/i386/i386.h (ix86_tune_indices): Add X86_TUNE_AVX128_OPTIMAL entry.
(TARGET_AVX128_OPTIMAL): New definition.
* config/i386/i386.c (initial_ix86_tune_features): Initialize
X86_TUNE_AVX128_OPTIMAL entry.
(ix86_option_override_internal): Enable the generation
of the 128-bit instructions when TARGET_AVX128_OPTIMAL is set.
(ix86_preferred_simd_mode): Use TARGET_PREFER_AVX128.
(ix86_autovectorize_vector_sizes): Use TARGET_PREFER_AVX128.

From-SVN: r175661

re PR c++/49272 (FAIL: g++.old-deja/g++.jason/rfg10.C  (test for errors, line 11))

PR c++/49272
* g++.old-deja/g++.jason/rfg10.C: Accept other overflow error too.

From-SVN: r175659

re PR c++/49520 ([C++0x] using-declaration and operator&& confuses constexpr)

PR c++/49520
* semantics.c (constexpr_fn_retval): Handle CLEANUP_POINT_EXPR here.
(massage_constexpr_body): Not here.

From-SVN: r175658

re PR c++/49554 ([C++0x] lambda capture causes "cannot call member function ... without object")

PR c++/49554
* semantics.c (lambda_proxy_type): New.
(build_capture_proxy): Use it.
* cp-tree.h (DECLTYPE_FOR_LAMBDA_PROXY): New.
* pt.c (tsubst) [DECLTYPE_TYPE]: Use them.

From-SVN: r175657

re PR rtl-optimization/49114 ([x32] Reload failed to handle (set reg:X (plus:X (subreg:X (reg:Y) 0) (const_int))))

PR rtl-optimization/49114
* reload.c (push_reload): Define in_subreg_loc and out_subreg_loc
only if LIMIT_RELOAD_CLASS to avoid -Werror build breaks.

From-SVN: r175652

re PR tree-optimization/49539 (ICE building gnattools)

PR tree-optimization/49539
* tree-ssa-forwprop.c (can_propagate_from): Check for abnormal SSA
names by means of stmt_references_abnormal_ssa_name.
(associate_plusminus): Call can_propagate_from before propagating
from definition statements.
(ssa_forward_propagate_and_combine): Remove superfluous newline.

From-SVN: r175650

re PR c++/45923 (constexpr diagnostics, more more)

PR c++/45923
* class.c (explain_non_literal_class): New.
(finalize_literal_type_property): Call it.
* cp-tree.h: Declare it.
* semantics.c (ensure_literal_type_for_constexpr_object): Call it.
(is_valid_constexpr_fn): Likewise.
(massage_constexpr_body): Split out from...
(register_constexpr_fundef): ...here.
(is_instantiation_of_constexpr): New.
(expand_or_defer_fn_1): Leave DECL_SAVED_TREE alone in that case.
(explain_invalid_constexpr_fn): New.
(cxx_eval_call_expression): Call it.
(potential_constant_expression_1): Likewise.  Avoid redundant errors.
* method.c (process_subob_fn): Diagnose non-constexpr.
(walk_field_subobs): Likewise.
(synthesized_method_walk): Don't shortcut if we want diagnostics.
(explain_implicit_non_constexpr): New.
(defaulted_late_check): Use it.
* call.c (build_cxx_call): Remember location.

From-SVN: r175646

method.c (maybe_explain_implicit_delete): Use pointer_set instead of htab.

* method.c (maybe_explain_implicit_delete): Use pointer_set
instead of htab.

From-SVN: r175645

class.c (finalize_literal_type_property): Update conditions.

* class.c (finalize_literal_type_property): Update conditions.
* method.c (defaulted_late_check): Set TYPE_HAS_CONSTEXPR_CTOR.

From-SVN: r175644

tree.c (build_vec_init_expr): Don't add TARGET_EXPR.

* tree.c (build_vec_init_expr): Don't add TARGET_EXPR.
* typeck2.c (digest_init_r): Handle VEC_INIT_EXPR.
* semantics.c (cxx_eval_vec_init_1): Correct type.

From-SVN: r175643

invoke.texi: Document -scev dump modifier.

2011-06-29  Richard Guenther  <rguenther@suse.de>

* doc/invoke.texi: Document -scev dump modifier.
* tree-pass.h (TDF_SCEV): New dump flag.
* tree-dump.c (dump_option_value_in): Add scev.
* tree-chrec.c: Replace all TDF_DETAILS checks with TDF_SCEV.
* tree-scalar-evolution.c: Likewise.

* gcc.dg/tree-ssa/loop-17.c: Adjust.

From-SVN: r175642

unwind-arm.c (enum __cxa_type_match_result): New.

gcc/
* config/arm/unwind-arm.c (enum __cxa_type_match_result): New.
(cxa_type_match): Correct declaration.
(__gnu_unwind_pr_common): Reconstruct
additional indirection when __cxa_type_match returns
succeeded_with_ptr_to_base.

libstdc++/
* libsupc++/eh_arm.c (__cxa_type_match): Construct address of
thrown object here.  Return succeded_with_ptr_to_base for all
pointer cases.

From-SVN: r175641

init.c (build_value_init): Decide whether or not to zero-initialize based on user-providedness of default...

* init.c (build_value_init): Decide whether or not to zero-initialize
based on user-providedness of default ctor, not any ctor.
(build_value_init_noctor): Adjust assert.

From-SVN: r175640

DR 990

DR 990
* call.c (convert_like_real) [ck_user]: Handle value-initialization.
(build_new_method_call_1): Likewise.
* init.c (expand_default_init): Handle direct list-initialization
of aggregates.

From-SVN: r175639

cls_double_va.c: Move PR number to comment.

* testsuite/libffi.call/cls_double_va.c: Move PR number to comment.
* testsuite/libffi.call/cls_longdouble_va.c: Likewise.

From-SVN: r175632

re PR rtl-optimization/49114 ([x32] Reload failed to handle (set reg:X (plus:X (subreg:X (reg:Y) 0) (const_int))))

PR rtl-optimization/49114
* reload.c (struct replacement): Remove SUBREG_LOC member.
(push_reload): Do not set it.
(push_replacement): Likewise.
(subst_reload): Remove dead code.
(copy_replacements): Remove assertion.
(copy_replacements_1): Do not handle SUBREG_LOC.
(move_replacements): Likewise.
(find_replacement): Remove dead code.  Use reload_adjust_reg_for_mode.
Detect subregs via recursive descent instead of via SUBREG_LOC.

From-SVN: r175631

avr.c (avr_encode_section_info): Dispatch to default_encode_section_info.

* config/avr/avr.c (avr_encode_section_info): Dispatch to
default_encode_section_info.

From-SVN: r175630

avr.c (avr_encode_section_info): Dispatch to default_encode_section_info.

* config/avr/avr.c (avr_encode_section_info): Dispatch to
default_encode_section_info.

From-SVN: r175629

* MAINTAINERS (RTL optimizers): Add self.

From-SVN: r175628

re PR middle-end/49545 (New C++ test failures)

gcc/
PR tree-optimization/49545
* builtins.c (get_object_alignment_1): Update function comment.
Do not use DECL_ALIGN for functions, but test
TARGET_PTRMEMFUNC_VBIT_LOCATION instead.
* fold-const.c (get_pointer_modulus_and_residue): Don't check
for functions here.
* tree-ssa-ccp.c (get_value_from_alignment): Likewise.

gcc/testsuite/
* gcc.dg/torture/pr49169.c: Restrict to ARM and MIPS targets.

From-SVN: r175627

inline_1.c: Remove dg-excess-errors.

* gcc.dg/inline_1.c: Remove dg-excess-errors.
* gcc.dg/inline_2.c: Likewise.
* gcc.dg/unroll_1.c: Likewise.

From-SVN: r175625

* gcc-interface/Makefile.in (TOOLS_LIBS): Add $(LIBINTL).

From-SVN: r175624