constraints.md (wH constraint): Add new constraints for allowing 32-bit integers (and eventually 8/16-bit...

[gcc]
2016-10-27  Michael Meissner  <meissner@linux.vnet.ibm.com>

* config/rs6000/constraints.md (wH constraint): Add new
constraints for allowing 32-bit integers (and eventually 8/16-bit
integers) into the vector registers.
(wI constraint): Likewise.
(wJ constraint): Likewise.
(wK constraint): Likewise.
* config/rs6000/rs6000-cpus.def (ISA_2_7_MASKS_SERVER): Add
-mvsx-small-integer as a default option for ISA 2.07
(i.e. power8).
(POWERPC_MASKS): Likewise.
* config/rs6000/rs6000.opt (-mvsx-small-integer): Add new debug
switch to turn off small integer support in vector registers.
* config/rs6000/rs6000.c (rs6000_hard_regno_mode_ok): Eliminate
test for -mupper-regs-di, since it is already done with the
reg_add[mode].scalar_in_vsx_p.  Add support for the switch
-mvsx-small-integer.
(rs6000_debug_reg_global): Add support for wH, wI, wJ, and wK
constraints.
(rs6000_setup_reg_addr_masks): Likewise.
(rs6000_init_hard_regno_mode_ok): Likewise.
(rs6000_option_override_internal): Add consistency checks for
-mvsx-small-integer.
(rs6000_secondary_reload_simple_move): SImode is a simple move if
-mvsx-small-integer.
(rs6000_secondary_reload): Use std::swap.
(rs6000_preferred_reload_class): Don't prefer FLOAT_REGS over
VSX_REGS for small integers in vector registers, since there is no
D-FORM address mode for such types.
(rs6000_register_move_cost): Use FIRST_FPR_REGNO instead of 32.
(rs6000_opt_masks): Add -mvsx-small-integer.
* config/rs6000/vsx.md (VSINT_84): Add SImode for small integer
support.
(VSX_EXTRACT_I2): Clone VSX_EXTRACT_I, but drop V4SI since SImode
extracts can be done on ISA 2.07.
(vsx_extract_<mode>): Add support for small integers in vsx
registers.
(vsx_extract_<mode>_p9): Use 'v' instead of VSX_EX, since we no
longer support V4SImode in this pattern.
(vsx_extract_si): New insn to support extraction of SImode in ISA
2.07 using either xxextractuw or vspltw.
(vsx_extract_<mode>_p8): Use 'v' instead of VSX_EX, since we no
longer support V4SImode in this pattern.
* config/rs6000/rs6000.h (enum rs6000_reg_class_enum): Add wH, wI,
wJ, and wK constraints.
* config/rs6000/rs6000.md (f32_sv): Use correct instruction for
storing SDmode with VSX instructions.
(zero_extendsi<mode>2): Reorder pattern, so RLDICL comes after the
GPR load and before the FPR and VSX loads.  Remove ??, ! from the
constraints.  Add MFVSRWZ and XXEXTRACTUW instructions to support
small integers in vector registers.
(extendsi<mode>2): Reorder pattern, so EXTSW comes after the GPR
load and before the FPR and VSX loads.  Remove ??, ! from the
constraints.  Add VEXTSW2D support for small integers in vector
registers.
(lfiwax): Remove ! constraint.  Add VEXTSW2D support for small
integers in vector registers.
(floatsi<mode>2_lfiwax): If -mvsx-small-integer issue a normal
move instead of using an UNSPEC.
(lfiwzx): Remove ! constraint.  Add XXEXTRACTUW support for small
integers in vector registers.
(floatunssi<mode>2_lfiwzx): If -mvsx-small-integer issue a normal
move instead of using an UNSPEC.
(movsi_internal1): Add support for -mvsx-small-integer.  Align
columns so that it is more readable.
(SImode splitter for ISA 3.0 constants): Add splitter for
-128..127 constants that can easily be constructed on ISA 3.0.
* doc/md.texi (PowerPC Constraints): Document wH, wI, wJ, and wK
constraints.

[gcc/testsuite]
2016-10-27  Michael Meissner  <meissner@linux.vnet.ibm.com>

* gcc.target/powerpc/vsx-simode.c: New test.
* gcc.target/powerpc/vsx-simode2.c: Likewise.
* gcc.target/powerpc/vsx-simode3.c: Likewise.

From-SVN: r241631

re PR fortran/78026 (ICE in gfc_resolve_omp_declare_simd, at fortran/openmp.c:5190)

PR fortran/78026
* parse.c (decode_statement): Don't create namespace for possible
select type here and destroy it afterwards.
(parse_select_type_block): Set gfc_current_ns to new_st.ext.block.ns.
(parse_executable, gfc_parse_file): Formatting fixes.
* match.c (gfc_match_select_type): Create namespace for select type
here, only after matching select type.  Formatting fixes.  Free that
namespace if not returning MATCH_YES, after gfc_undo_symbols,
otherwise remember it in new_st.ext.block.ns and switch to parent
namespace anyway.

* gfortran.dg/gomp/pr78026.f03: New test.
* gfortran.dg/select_type_38.f03: New test.

From-SVN: r241630

PR70975 Pass valid offset argument to sendfile

PR libstdc++/70975
* src/filesystem/ops.cc (do_copy_file) [_GLIBCXX_USE_SENDFILE]:
Pass non-null pointer to sendfile for offset argument.

From-SVN: r241629

re PR middle-end/78025 (ICE in simd_clone_adjust, at omp-simd-clone.c:1126)

PR middle-end/78025
* omp-simd-clone.c (simd_clone_adjust): Handle noreturn declare simd
functions.

* g++.dg/gomp/declare-simd-7.C: New test.

From-SVN: r241628

oacc-init.c (goacc_new_thread): Use sizeof of the appropriate size when allocating new thread.

* oacc-init.c (goacc_new_thread): Use sizeof of the appropriate
size when allocating new thread.

From-SVN: r241627

Fix initialization of UNIONs with -finit-derived.

gcc/fortran/
* expr.c (generate_union_initializer, get_union_initializer): New.
* expr.c (component_initializer): Consider BT_UNION specially.
* resolve.c (resolve_structure_cons): Hack for BT_UNION.
* trans-expr.c (gfc_trans_subcomponent_assign): Ditto.
* trans-expr.c (gfc_conv_union_initializer): New.
* trans-expr.c (gfc_conv_structure): Replace UNION handling code with
new function gfc_conv_union_initializer.

gcc/testsuite/gfortran.dg/
* dec_init_1.f90, dec_init_2.f90: Remove -fdump-tree-original.
* dec_init_3.f90, dec_init_4.f90: New tests.

From-SVN: r241626

builtins.c (expand_builtin_nonlocal_goto): Avoid evaluating PIC_OFFSET_TABLE_REGNUM twice.

* builtins.c (expand_builtin_nonlocal_goto): Avoid evaluating
PIC_OFFSET_TABLE_REGNUM twice.

From-SVN: r241625

match.pd ((convert (op:s (convert@2 @0) (convert?@3 @1)))): Add support for constant operand for OP.

* match.pd ((convert (op:s (convert@2 @0) (convert?@3 @1)))): Add
support for constant operand for OP.

gcc/testsuite
* gcc.dg/fold-narrowbopcst-1.c: New test.

From-SVN: r241624

Fix some DEC I/O testcases.

gcc/testsuite/gfortran.dg/
* dec_io_5.f90, dec_io_6.f90: Don't use "test.txt", and use
dg-shouldfail/dg-output instead of XFAIL.

From-SVN: r241623

dwarf2out.c (gen_member_die): Only reparent_child instead of splice_child_die if...

* dwarf2out.c (gen_member_die): Only reparent_child instead of
splice_child_die if child doesn't have DW_AT_specification attribute.

From-SVN: r241622

Fix target selectors in uniform_inside_sphere_distribution tests

* testsuite/ext/random/uniform_inside_sphere_distribution/cons/
default.cc: Fix effective target selector.
* testsuite/ext/random/uniform_inside_sphere_distribution/cons/
parms.cc: Likewise.
* testsuite/ext/random/uniform_inside_sphere_distribution/operators/
equal.cc: Likewise.
* testsuite/ext/random/uniform_inside_sphere_distribution/operators/
generate.cc: Likewise.
* testsuite/ext/random/uniform_inside_sphere_distribution/operators/
inequal.cc: Likewise.
* testsuite/ext/random/uniform_inside_sphere_distribution/operators/
serialize.cc: Likewise.

From-SVN: r241621

* class.c (add_method): Allow using-declarations to coexist.

From-SVN: r241620

Add myself to the MAINTAINERS file

* MAINTAINERS (Reviewers): Add myself.
(Write After Approval): Add myself.

From-SVN: r241619

Adjust precision of filesystem::last_write_time tests

* testsuite/experimental/filesystem/iterators/directory_iterator.cc:
Use end() function to get end iterator.
* testsuite/experimental/filesystem/iterators/pop.cc: Remove printf
statements that were present for debugging.
* testsuite/experimental/filesystem/iterators/
recursive_directory_iterator.cc: Use end() function to get end
iterator.
* testsuite/experimental/filesystem/operations/last_write_time.cc:
Only require file timestamps to be accurate to one second.

From-SVN: r241616

Enable ARMv8-M atomic and synchronization support for ARMv8-M Baseline

2016-10-27  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    gcc/
    * config/arm/arm.h (TARGET_HAVE_LDREX): Define for ARMv8-M Baseline.
    (TARGET_HAVE_LDREXBH): Likewise.
    (TARGET_HAVE_LDACQ): Likewise.

    gcc/testsuite/
    * gcc.target/arm/atomic-comp-swap-release-acquire-3.c: New test.
    * gcc.target/arm/atomic-op-acq_rel-3.c: Likewise.
    * gcc.target/arm/atomic-op-acquire-3.c: Likewise.
    * gcc.target/arm/atomic-op-char-3.c: Likewise.
    * gcc.target/arm/atomic-op-consume-3.c: Likewise.
    * gcc.target/arm/atomic-op-int-3.c: Likewise.
    * gcc.target/arm/atomic-op-relaxed-3.c: Likewise.
    * gcc.target/arm/atomic-op-release-3.c: Likewise.
    * gcc.target/arm/atomic-op-seq_cst-3.c: Likewise.
    * gcc.target/arm/atomic-op-short-3.c: Likewise.

From-SVN: r241615

Adapt other atomic operations to ARMv8-M Baseline

2016-10-27  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    gcc/
    * config/arm/arm.c (arm_split_atomic_op): Add function comment.  Add
    logic to to decide whether to copy over old value to register for new
    value.
    * config/arm/sync.md: Add comments explaning why mode and code
    attribute are not defined in iterators.md
    (thumb1_atomic_op_str): New code attribute.
    (thumb1_atomic_newop_str): Likewise.
    (thumb1_atomic_fetch_op_str): Likewise.
    (thumb1_atomic_fetch_newop_str): Likewise.
    (thumb1_atomic_fetch_oldop_str): Likewise.
    (atomic_exchange<mode>): Add new ARMv8-M Baseline only alternatives to
    mirror the more restrictive constraints of the Thumb-1 insns after
    split compared to Thumb-2 counterpart insns.
    (atomic_<sync_optab><mode>): Likewise.  Add comment to keep constraints
    in sync with non atomic version.
    (atomic_nand<mode>): Likewise.
    (atomic_fetch_<sync_optab><mode>): Likewise.
    (atomic_fetch_nand<mode>): Likewise.
    (atomic_<sync_optab>_fetch<mode>): Likewise.
    (atomic_nand_fetch<mode>): Likewise.
    * config/arm/thumb1.md (thumb1_addsi3): Add comment to keep contraint
    in sync with atomic version.
    (thumb1_subsi3_insn): Likewise.
    (thumb1_andsi3_insn): Likewise.
    (thumb1_iorsi3_insn): Likewise.
    (thumb1_xorsi3_insn): Likewise.

From-SVN: r241614

plugin.c (register_plugin_info): Produce an error message if the plugin is not found in the hash table.

* plugin.c (register_plugin_info): Produce an error message if the
plugin is not found in the hash table.

From-SVN: r241613

match.pd ((convert1 (minmax ((convert2 (x) c)))) -> minmax (x c)): New pattern.

* match.pd ((convert1 (minmax ((convert2 (x) c)))) -> minmax (x c)):
New pattern.

gcc/testsuite
* gcc.dg/fold-convmaxconv-1.c: New test.
* gcc.dg/fold-convminconv-1.c: New test.

From-SVN: r241612

re PR fortran/78092 (ICE when calling SIZEOF on CLASS(*) entry)

2016-10-26  Steven G. Kargl <kargl@gcc.gnu.org>

PR fortran/78092
* trans-intrinsic.c (gfc_conv_intrinsic_sizeof):  Fix reference to an
array element of type CLASS.

2016-10-26  Steven G. Kargl <kargl@gcc.gnu.org>

PR fortran/78092
* gfortran.dg/pr78092.f90: New test.

From-SVN: r241610

Daily bump.

From-SVN: r241609

stl_map.h (map()): Make default.

2016-10-26  François Dumont  <fdumont@gcc.gnu.org>

* include/bits/stl_map.h (map()): Make default.
* include/bits/stl_multimap.h (multimap()): Likewise.
* include/bits/stl_multiset.h (multiset()): Likewise.
* include/bits/stl_set.h (set()): Likewise.
* include/bits/stl_tree.h (_Rb_tree_impl()): Add conditional noexcept.
(_Rb_tree()): Make default.

From-SVN: r241601

sh.c (output_branch): Add missing fallthru comments.

* config/sh/sh.c (output_branch): Add missing fallthru comments.
(gen_shl_and): Likewise.
* config/sh/sh.md (movsicc): Add missing fallthru comments.

From-SVN: r241600

re PR target/78056 (build failure on Power7)

gcc/ChangeLog:

2016-10-26  Kelvin Nilsen  <kelvin@gcc.gnu.org>

PR target/78056
* config/rs6000/rs6000.c (spe_init_builtins): Modify loops to not
define builtin functions from the bdesc_spe_predicates or
bdesc_spe_evsel arrays if the builtin mask is not compatible with
the current compiler configuration.
(paired_init_builtins): Modify loop to not define define builtin
functions from the bdesc_paried_preds array if the builtin mask is
not compatible with the current compiler configuration.
(altivec_init_builtins): Modify loops to not define the
__builtin_altivec_stxvl function nor the builtin functions from
the bdesc_dst or bdesc_altivec_preds, or bdesc_abs arrays if the
builtin mask is not compatible with the current compiler
configuration.

gcc/testsuite/ChangeLog:

2016-10-26  Kelvin Nilsen  <kelvin@gcc.gnu.org>

PR target/78056
* gcc.target/powerpc/vsu/vec-any-eqz-7.c (test_any_equal): Change
expected error message.
* gcc.target/powerpc/vsu/vec-xst-len-12.c (store_data): Change
expected error message.
* gcc.target/powerpc/vsu/vec-all-nez-7.c
(test_all_not_equal_and_not_zero): Change expected error message.

From-SVN: r241599

mips.c (mips16_constant_cost): Add missing fallthru comments.

* config/mips/mips.c (mips16_constant_cost): Add missing
fallthru comments.
(mips16_build_call_stub): Increase buffer size.  Adjust
fallthru comment.

From-SVN: r241597

Show INSN_UIDs in compact mode

gcc/ChangeLog:
* print-rtl.c (rtx_writer::print_rtx_operand_code_u): Print
INSN_UIDs for all insns in compact mode.
(rtx_writer::print_rtx): Likewise.
* print-rtl.h (rtx_writer::flag_compact): Update comment.
* rtl-tests.c (selftest::test_dumping_insns): Update expected
output to include INSN_UID.
(selftest::test_uncond_jump): Likewise.

From-SVN: r241593

re PR libstdc++/78110 (freestanding libstdc++ fails to compile)

2016-10-26  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        PR libstdc++/78110
        * libsupc++/new_opa.cc: Don't include <malloc.h> in a free standing
        environment.  Declare memalign directly in that case.

From-SVN: r241591

haifa-sched.c (call_used_regs_num): Rename to...

* haifa-sched.c (call_used_regs_num): Rename to...
(call_saved_regs_num): ...this.
(fixed_regs_num): New variable.
(sched_pressure_start_bb): Subtract out fixed_regs. Scale call_saved
regs not call_used.
(alloc_global_sched_pressure_data): Compute call_saved and fixed regs.

From-SVN: r241590

microblaze.c (tls_mentioned_p): Avoid fallthru.

* config/microblaze/microblaze.c (tls_mentioned_p): Avoid
fallthru.

From-SVN: r241587

Introduce class rtx_writer

gcc/ChangeLog:
* print-rtl-function.c (print_rtx_function): Rewrite in terms of
class rtx_writer.
* print-rtl.c (outfile): Delete global.
(sawclose): Likewise.
(indent): Likewise.
(in_call_function_usage): Likewise.
(flag_compact): Likewise.
(flag_simple): Likewise.
(rtx_writer::rtx_writer): New ctor.
(print_rtx_operand_code_0): Convert to...
(rtx_writer::print_rtx_operand_code_0): ...this.
(print_rtx_operand_code_e): Convert to...
(rtx_writer::print_rtx_operand_code_e): ...this.
(print_rtx_operand_codes_E_and_V): Convert to...
(rtx_writer::print_rtx_operand_codes_E_and_V): ...this.
(print_rtx_operand_code_i): Convert to...
(rtx_writer::print_rtx_operand_code_i): ...this.
(print_rtx_operand_code_r): Convert to...
(rtx_writer::print_rtx_operand_code_r): ...this.
(print_rtx_operand_code_u): Convert to...
(rtx_writer::print_rtx_operand_code_u): ...this.
(print_rtx_operand): Convert to...
(rtx_writer::print_rtx_operand): ...this.
(print_rtx): Convert to...
(rtx_writer::print_rtx): ...this.
(print_inline_rtx): Rewrite in terms of class rtx_writer.
(debug_rtx): Likewise.
(print_rtl): Convert to...
(rtx_writer::print_rtl): ...this.
(print_rtl): Reimplement in terms of class rtx_writer.
(print_rtl_single): Rewrite in terms of class rtx_writer.
(print_rtl_single_with_indent): Convert to..
(rtx_writer::print_rtl_single_with_indent): ...this.
(print_simple_rtl): Rewrite in terms of class rtx_writer.
* print-rtl.h (flag_compact): Delete decl.
(class rtx_writer): New class.
* rtl-tests.c (selftest::assert_rtl_dump_eq): Rewrite in terms of
class rtx_writer.

From-SVN: r241586

arc.c (acr_print_operand): Adjust fallthru comment.

* config/arc/arc.c (acr_print_operand): Adjust fallthru comment.
(check_if_valid_sleep_operand): Add missing fallthru comment.
(arc_register_move_cost): Increase buffer size.
* config/arc/arc.md (cbranch4si_scratch): Add missing fallthru
comment.
* config/arc/predicates.md (move_str_operand): Avoid fallthru.

From-SVN: r241585

cr16.c (cr16_print_operand): Add missing fallthru comment.

* config/cr16/cr16.c (cr16_print_operand): Add missing fallthru
comment.  Add gcc_unreachable for path that should never happen.

From-SVN: r241584

Fix test for recursive_directory_iterator::pop

* testsuite/experimental/filesystem/iterators/pop.cc: Remove
unreliable dependency on directory order.

From-SVN: r241583

epiphany.c (epiphany_print_operand): Adjust fallthru comment.

* config/epiphany/epiphany.c (epiphany_print_operand): Adjust
fallthru comment.

From-SVN: r241582

re PR fortran/77973 (ICE in scan_omp_1_op, at omp-low.c:3841)

PR fortran/77973
* gimplify.c (gimplify_adjust_omp_clauses_1): For all added map
clauses with OMP_CLAUSE_SIZE being a decl, call omp_notice_variable
on outer context if any.

* gfortran.dg/gomp/pr77973.f90: New test.

Co-Authored-By: Martin Liska <mliska@suse.cz>
From-SVN: r241581

gen-pass-instances.awk (adjust_linenos): INcrement pass_lines[p] by increment rather than double it.

* gen-pass-instances.awk (adjust_linenos): INcrement pass_lines[p]
by increment rather than double it.
(insert_remove_pass): Strip leading whitespace from args[3].  Don't
emit a space before args[4].
(END): Don't emit a space before with_arg.

From-SVN: r241580

* config/fr30/fr30.c (fr30_print_operand): Adjust fallthru comment.

From-SVN: r241579

Adapt atomic compare and swap to ARMv8-M Baseline

2016-10-26  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    gcc/
    * config/arm/sync.md (atomic_compare_and_swap<mode>_1): Add new ARMv8-M
    Baseline only alternatives to (i) hold store atomic success value in a
    return register rather than a scratch register, (ii) use a low register
    for it and to (iii) ensure the cbranchsi insn generated by the split
    respect the constraints of Thumb-1 cbranchsi4_insn and
    cbranchsi4_scratch.
    * config/arm/thumb1.md (cbranchsi4_insn): Add comment to indicate
    constraints must match those in atomic_compare_and_swap.
    (cbranchsi4_scratch): Likewise.

From-SVN: r241578

Refactor atomic compare_and_swap to make it fit for ARMv8-M Baseline

2016-10-26  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    gcc/
    * config/arm/arm.c (arm_expand_compare_and_swap): Add new bdst local
    variable.  Add the new parameter to the insn generator.  Set that
    parameter to be CC flag for 32-bit targets, bval otherwise.  Set the
    return value from the negation of that parameter for Thumb-1, keeping
    the logic unchanged otherwise except for using bdst as the destination
    register of the compare_and_swap insn.
    (arm_split_compare_and_swap): Add explanation about how is the value
    returned to the function comment.  Rename scratch variable to
    neg_bval.  Adapt initialization of variables holding operands to the
    new operand numbers.  Use return register to hold result of store
    exclusive for Thumb-1, scratch register otherwise.  Construct the
    appropriate cbranch for Thumb-1 targets, keeping the logic unchanged
    for 32-bit targets.  Guard Z flag setting to restrict to 32bit targets.
    Use gen_cbranchsi4 rather than hand-written conditional branch to loop
    for strongly ordered compare_and_swap.
    * config/arm/predicates.md (cc_register_operand): New predicate.
    * config/arm/sync.md (atomic_compare_and_swap<mode>_1): Use a
    match_operand with the new predicate to accept either the CC flag or a
    destination register for the boolean return value, restricting it to
    CC flag only via constraint.  Adapt operand numbers accordingly.

From-SVN: r241577

frv.c (comparison_string): Do not fall through after an error.

* config/frv/frv.c (comparison_string): Do not fall through after
an error.

From-SVN: r241576

iq2000.c (iq2000_function_arg): Adjust fallthru comment.

* config/iq2000/iq2000.c (iq2000_function_arg): Adjust fallthru
comment.
(expand_one_builtin): Add missing break.

From-SVN: r241573

m32c.c (encode_pattern_1): Add fallthru comment.

* config/m32c/m32c.c (encode_pattern_1): Add fallthru comment.
(m32c_legitimate_address_p): Likewise.

From-SVN: r241572

* config/m32r/m32r.c (m32r_print_operand): Adjust fallthru comment.

From-SVN: r241571

* config/mcore/mcore.c (mcore_gen_compare): Adjust fallthru comments.

From-SVN: r241569

microblaze.c (microblaze_function_arg): Adjust fallthru comment.

* config/microblaze/microblaze.c (microblaze_function_arg): Adjust
fallthru comment.

From-SVN: r241568

msp430.c (msp430_legitimate_address_p): Adjust fallthru comment.

* config/msp430/msp430.c (msp430_legitimate_address_p): Adjust
fallthru comment.

From-SVN: r241567

* config/nios2/nios2.c (nios2_rtx_costs): Avoid fallthru.

From-SVN: r241566

rl78.c (rl78_calculate_death_notes): Add fallthru comment.

* config/rl78/rl78.c (rl78_calculate_death_notes): Add fallthru
comment.
(rl78_asm_ctor_dtor): Increase buffer size.

From-SVN: r241565

stormy16.c (xstormy16_asm_output_destrutor): Increase buffer size.

* config/stormy16/stormy16.c (xstormy16_asm_output_destrutor): Increase
buffer size.
(xstormy16_asm_output_constructor): Likewise.

From-SVN: r241564

Implement uniform_inside_sphere_distribution extension.

2016-10-26  Edward Smith-Rowland  <3dw4rd@verizon.net>

Implement uniform_inside_sphere_distribution extension.
* include/ext/random: Implement uniform_inside_sphere_distribution.
* include/ext/random.tcc: Ditto.
* testsuite/ext/random/uniform_inside_sphere_distribution/
cons/default.cc: New.
* testsuite/ext/random/uniform_inside_sphere_distribution/
cons/parms.cc: New.
* testsuite/ext/random/uniform_inside_sphere_distribution/
operators/equal.cc: New.
* testsuite/ext/random/uniform_inside_sphere_distribution/
operators/generate.cc: New.
* testsuite/ext/random/uniform_inside_sphere_distribution/
operators/inequal.cc: New.
* testsuite/ext/random/uniform_inside_sphere_distribution/
operators/serialize.cc: New.

From-SVN: r241562

pa.c (pa_asm_output_mi_thunk): Increase buffer size.

* config/pa/pa.c (pa_asm_output_mi_thunk): Increase buffer
size.

From-SVN: r241561

h8300.c (h8300_print_operand): Adjust FALLTHRU comment to silence warning.

* config/h8300/h8300.c (h8300_print_operand): Adjust FALLTHRU
comment to silence warning.

From-SVN: r241560

Add recursive_directory_iterator::pop(error_code&)

* include/experimental/bits/fs_dir.h (recursive_directory_iterator):
Overload pop (LWG 2706).
* src/filesystem/dir.cc (recursive_directory_iterator::pop): Define
new overload.
* testsuite/experimental/filesystem/iterators/pop.cc: New test.

From-SVN: r241559

spu.c (spu_sched_reorder): Add missing fallthru comment.

* config/spu/spu.c (spu_sched_reorder): Add missing fallthru comment.
(spu_legitimate_address_p): Fix logic error and add missing fallthru
comment.

From-SVN: r241557

re PR fortran/78108 (Generic type-bound operator conflicts)

2016-10-26  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/78108
* resolve.c (resolve_typebound_intrinsic_op): For submodules
suppress the error and return if the same procedure symbol
is added more than once to the interface.

2016-10-26  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/78108
* gfortran.dg/submodule_18.f08: New test.
* gfortran.dg/submodule_19.f08: New test.

From-SVN: r241555

Fix error handling in recursive_directory_iterator::increment

* src/filesystem/dir.cc (recursive_directory_iterator::increment):
Reset state on error.
* testsuite/experimental/filesystem/iterators/
recursive_directory_iterator.cc: Check state after increment error.

From-SVN: r241552

fix pr78060 pr78061 pr78088

PR tree-optimization/78060
PR tree-optimization/78061
PR tree-optimization/78088
* tree-ssa-loop-split.c (easy_exit_values): New function.
(tree_ssa_split_loops): Use it.
(compute_new_first_bound): Change order of operations,
fix invalid use of types.

testsuite/
* g++.dg/pr78060.C: New test.
* gfortran.dg/pr78061.f: New test.
* g++.dg/pr78088.C: New test.

From-SVN: r241551

New I/O specifiers CARRIAGECONTROL, READONLY, SHARE with -fdec.

gcc/fortran/
* gfortran.texi: Document.
* frontend-passes.c (gfc_code_walker): Add SHARE and CARRIAGECONTROL.
* io.c (gfc_free_open, gfc_resolve_open, gfc_match_open): Ditto.
* gfortran.h (gfc_open): Add SHARE, CARRIAGECONTROL, and READONLY.
* io.c (io_tag, match_open_element): Ditto.
* ioparm.def: Ditto.
* trans-io.c (gfc_trans_open): Ditto.
* io.c (match_dec_etag, match_dec_ftag): New functions.

libgfortran/io/
* libgfortran.h (IOPARM_OPEN_HAS_READONLY, IOPARM_OPEN_HAS_SHARE,
IOPARM_OPEN_HAS_CC): New for READONLY, SHARE, and CARRIAGECONTROL.
* close.c (st_close): Support READONLY.
* io.h (st_parameter_open, unit_flags): Support SHARE, CARRIAGECONTROL,
and READONLY.
* open.c (st_open): Ditto.
* transfer.c (data_transfer_init): Ditto.
* io.h (st_parameter_dt): New member 'cc' for CARRIAGECONTROL.
* write.c (write_check_cc, write_cc): New functions for CARRIAGECONTROL.
* transfer.c (next_record_cc): Ditto.
* file_pos.c (st_endfile): Support SHARE and CARRIAGECONTROL.
* io.h (st_parameter_inquire): Ditto.
* open.c (edit_modes, new_unit): Ditto.
* inquire.c (inquire_via_unit, inquire_via_filename): Ditto.
* io.h (unit_share, unit_cc, cc_fortran, IOPARM_INQUIRE_HAS_SHARE,
IOPARM_INQUIRE_HAS_CC): New for SHARE and CARRIAGECONTROL.
* open.c (share_opt, cc_opt): Ditto.
* read.c (read_x): Support CARRIAGECONTROL.
* transfer.c (read_sf, next_record_r, next_record_w): Ditto.
* write.c (list_formatted_write_scalar, write_a): Ditto.
* unix.h (close_share): New prototype.
* unix.c (open_share, close_share): New functions to handle SHARE.
* unix.c (open_external): Handle READONLY. Call open_share.
* close.c (st_close): Call close_share.

gcc/testsuite/
* dec_io_1.f90: New test.
        * dec_io_2.f90: New test.
        * dec_io_3.f90: New test.
        * dec_io_4.f90: New test.
        * dec_io_5.f90: New test.
        * dec_io_6.f90: New test.

From-SVN: r241550

PR78111 fix fallback code for filesystem::canonical

PR libstdc++/78111
* src/filesystem/ops.cc (canonical): Set error for non-existent path.

From-SVN: r241549

Add missing header in Filesystem TS configure tests

2016-10-26  Uros Bizjak  <ubizjak@gmail.com>

* acinclude.m4 (GLIBCXX_CHECK_FILESYSTEM_DEPS): Include <limits.h>
for PATH_MAX in realpath test.
* configure: Regenerate.

From-SVN: r241548

gen-pass-instances.awk is sensitive to the order in which passes are added...

gen-pass-instances.awk is sensitive to the order in which
passes are added; passes that appear later have to be added first.
PR target/71676
PR target/71678
* config/avr/avr-passes.def: Swap order of directives for
gen-pass-instances.awk.

From-SVN: r241547

Daily bump.

From-SVN: r241545

vax.c (vad_address_cost_1): Add missing FALLTHRU comment.

* config/vax/vax.c (vad_address_cost_1): Add missing FALLTHRU comment.
(vax_notice_update_cc): Likewise.

From-SVN: r241542

MAINTAINERS (Write After Approval): Add myself.

2016-10-25  Will Schmidt  <will_schmidt@vnet.ibm.com>

* MAINTAINERS (Write After Approval): Add myself.

From-SVN: r241541

re PR fortran/45516 ([F08] allocatable compontents of recursive type)

2016-10-25  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/45516
* class.c (gfc_find_derived_vtab): Detect recursive allocatable
derived type components. If present, add '_deallocate' field to
the vtable and build the '__deallocate' function.
* decl.c (build_struct): Allow recursive allocatable derived
type components for -std=f2008 or more.
(gfc_match_data_decl): Accept these derived types.
* expr.c (gfc_has_default_initializer): Ditto.
* resolve.c (resolve_component): Make sure that the vtable is
built for these derived types.
* trans-array.c(structure_alloc_comps) : Use the '__deallocate'
function for the automatic deallocation of these types.
* trans-expr.c : Generate the deallocate accessor.
* trans.h : Add its prototype.
* trans-types.c (gfc_get_derived_type): Treat the recursive
allocatable components in the same way as the corresponding
pointer components.

2016-10-25  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/45516
* gfortran.dg/class_2.f03: Set -std=f2003.
* gfortran.dg/finalize_21.f90: Modify tree-dump.
* gfortran.dg/recursive_alloc_comp_1.f08: New test.
* gfortran.dg/recursive_alloc_comp_2.f08: New test.
* gfortran.dg/recursive_alloc_comp_3.f08: New test.
* gfortran.dg/recursive_alloc_comp_4.f08: New test.

From-SVN: r241539

config.gcc (sparc*-*-solaris2*): Adjust.

* config.gcc (sparc*-*-solaris2*): Adjust.
(sparc64-*-linux*): Likewise.
* config/sparc/default-64.h: Rename to...
* config/sparc/default64.h: ...this.
* config/sparc/sparc.c (sparc_option_override): Replace TARGET_64BIT
with TARGET_ARCH64.
(sparc_mangle_type): Replace !TARGET_64BIT with TARGET_ARCH32.
* config/sparc/sparc.h: Minor tweaks.
* config/sparc/sparc.md: Replace !TARGET_64BIT and !TARGET_ARCH64 with
TARGET_ARCH32 throughout.  Minor various tweaks throughout.

From-SVN: r241538

re PR sanitizer/78106 (ASAN can't find heap-use-after-free with -O1)

PR sanitizer/78106
* gcc.dg/asan/pr78106.c: Add -ffat-lto-objects to dg-options.

From-SVN: r241537

input.c/libcpp: fix lifetimes of path buffers

Running "make selftest-valgrind" showed various leaks of the form:

408 bytes in 24 blocks are definitely lost in loss record 572 of 679
   at 0x4A0645D: malloc (in /usr/lib64/valgrind/vgpreload_memcheck-amd64-linux.so)
   by 0x1B0D057: xmalloc (xmalloc.c:148)
   by 0x1ACCAA1: append_file_to_dir(char const*, cpp_dir*) [clone .isra.3] (files.c:1567)
   by 0x1ACD56F: _cpp_find_file (files.c:390)
   by 0x1ACF8FB: cpp_read_main_file(cpp_reader*, char const*) (init.c:632)
   by 0x1AB3D97: selftest::lexer_test::lexer_test(selftest::line_table_case const&, char const*, selftest::lexer_test_options*) (input.c:2014)
   by 0x1AB792B: selftest::test_lexer_string_locations_u8(selftest::line_table_case const&) (input.c:2713)
   by 0x1ABA22A: selftest::for_each_line_table_case(void (*)(selftest::line_table_case const&)) (input.c:3227)
   by 0x1ABA381: selftest::input_c_tests() (input.c:3260)
   by 0x1A295F1: selftest::run_tests() (selftest-run-tests.c:62)
   by 0xF20DC4: toplev::run_self_tests() (toplev.c:2076)
   by 0xF20FCD: toplev::main(int, char**) (toplev.c:2153)

Fix the leak by freeing the file->path in destroy_cpp_file.
However, doing so would lead to a use-after-free in input.c's file cache
since the filenames in this cache are the libcpp file->path buffers.

Hence we need to ensure that any references to the file in the input.c
cache are purged before cleaning up file->path.  This is normally done
by the temp_source_file dtor.  Hence we need to reorder things to that
the temp_source_file dtor runs before cleaning up the cpp_parser.  The
patch does this by introducing a wrapper class around cpp_parser *, so
that the dtor can run after the dtor for temp_source_file.

gcc/ChangeLog:
* input.c (fcache::file_patch): Add comment about lifetime.
(selftest::cpp_reader_ptr): New class.
(selftest::lexer_test): Convert m_parser from cpp_reader *
to a cpp_reader_ptr, and move m_tempfile to after it.
(selftest::lexer_test::lexer_test): Update for above reordering.
(lexer_test::~lexer_test): Move cleanup of m_parser to
cpp_reader_ptr's dtor.

libcpp/ChangeLog:
* files.c (destroy_cpp_file): Free file->path.

From-SVN: r241536

Convert logical ops on integers to bitwise equivalent with -fdec.

gcc/fortran/
* gfortran.texi: Document.
* resolve.c (logical_to_bitwise): New function.
* resolve.c (resolve_operator): Wrap operands with logical_to_bitwise.

gcc/testsuite/gfortran.dg/
* dec_bitwise_ops_1.f90, dec_bitwise_ops_2.f90: New testcases.

From-SVN: r241534

Implement ~line_maps ()

line_maps instances such as the global line_table are
GC-managed, but the htab within location_adhoc_data_map.htab
is not GC-managed.

Previously this was deleted manually by a call to
location_adhoc_data_fini within toplev::main.

However, on adding a call to forcibly_ggc_collect after the
selftests, all of the htabs for the various line_tables
created during the selftests start showing up as leaks
in "make selftest-valgrind", e.g.:

13,536 (1,344 direct, 12,192 indirect) bytes in 12 blocks are definitely lost in loss record 1,065 of 1,086
   at 0x4A081D4: calloc (in /usr/lib64/valgrind/vgpreload_memcheck-amd64-linux.so)
   by 0x16DB3B0: xcalloc (xmalloc.c:163)
   by 0x16D8D34: htab_create_typed_alloc (hashtab.c:358)
   by 0x16D8DBD: htab_create_alloc (hashtab.c:286)
   by 0x16A2CCC: linemap_init(line_maps*, unsigned int) (line-map.c:353)
   by 0x1685605: selftest::line_table_test::line_table_test(selftest::line_table_case const&) (input.c:1624)
   by 0x167D09C: selftest::test_applying_fixits_modernize_named_init(selftest::line_table_case const&) (edit-context.c:1430)
   by 0x1686827: selftest::for_each_line_table_case(void (*)(selftest::line_table_case const&)) (input.c:3227)
   by 0x167F067: selftest::edit_context_c_tests() (edit-context.c:1658)
   by 0x1616E67: selftest::run_tests() (selftest-run-tests.c:71)
   by 0xC0DB25: toplev::run_self_tests() (toplev.c:2076)
   by 0x618EB4: toplev::main(int, char**) (toplev.c:2153)

This patch removes the manual one-time cleanup in favor of
adding a destructor to class line_maps, which cleans up
the non-GC-managed htab.

Doing so improves "make selftest-valgrind" from:

==61118== LEAK SUMMARY:
==61118==    definitely lost: 121,248 bytes in 1,515 blocks
==61118==    indirectly lost: 974,344 bytes in 959 blocks
==61118==      possibly lost: 0 bytes in 0 blocks
==61118==    still reachable: 1,332,599 bytes in 3,684 blocks
==61118==         suppressed: 0 bytes in 0 blocks

to:

==57182== LEAK SUMMARY:
==57182==    definitely lost: 13,840 bytes in 556 blocks
==57182==    indirectly lost: 0 bytes in 0 blocks
==57182==      possibly lost: 0 bytes in 0 blocks
==57182==    still reachable: 1,355,703 bytes in 3,684 blocks
==57182==         suppressed: 0 bytes in 0 blocks

gcc/ChangeLog:
* toplev.c (toplev::main): Remove call to
location_adhoc_data_fini.

libcpp/ChangeLog:
* include/line-map.h (line_maps::~line_maps): New dtor.
(location_adhoc_data_fini): Delete decl.
* line-map.c (line_maps::~line_maps): New dtor.
(location_adhoc_data_fini): Delete.

From-SVN: r241533

constexpr.c (maybe_constant_init): Pull out TARGET_EXPR_INITIAL.

* constexpr.c (maybe_constant_init): Pull out TARGET_EXPR_INITIAL.

(cxx_eval_outermost_constant_expr): Don't return a CONSTRUCTOR
with CONSTRUCTOR_NO_IMPLICIT_ZERO.
(cxx_eval_call_expression): Clear CONSTRUCTOR_NO_IMPLICIT_ZERO.

From-SVN: r241531

tree.h (wi::fits_to_tree_p): Accept only 0 and 1 for boolean types.

* tree.h (wi::fits_to_tree_p): Accept only 0 and 1 for boolean types.
* tree.c (int_fits_type_p): Likewise.  Adjust head comment.

From-SVN: r241529

re PR fortran/72770 (ICE in make_ssa_name_fn, at tree-ssanames.c:263)

gcc/testsuite/ChangeLog:

2016-10-25  Andre Vehreschild  <vehre@gcc.gnu.org>

PR fortran/72770
* gfortran.dg/alloc_comp_class_5.f03: Added test again that caused
this pr.

gcc/fortran/ChangeLog:

2016-10-25  Andre Vehreschild  <vehre@gcc.gnu.org>

PR fortran/72770
* class.c (find_intrinsic_vtab): No longer encode the string length
into vtype's name and use the char's kind for the size instead of
the string_length time the size.
* trans-array.c (gfc_conv_ss_descriptor): For deferred length char
arrays the dynamically sized type needs to be declared.
(build_class_array_ref): Address the i-th array element by multiplying
it with the _vptr->_size and the _len to make sure char arrays are
addressed correctly.
* trans-expr.c (gfc_conv_intrinsic_to_class): Made comment more
precise.

From-SVN: r241528

Expose forcibly_ggc_collect and run it after all selftests

gcc/ChangeLog:
* ggc-tests.c (forcibly_ggc_collect): Rename to...
(selftest::forcibly_ggc_collect): ...this, and remove "static".
(test_basic_struct): Update for above renaming.
(test_length): Likewise.
(test_union): Likewise.
(test_finalization): Likewise.
(test_deletable_global): Likewise.
(test_inheritance): Likewise.
(test_chain_next): Likewise.
(test_user_struct): Likewise.
(test_tree_marking): Likewise.
* selftest-run-tests.c (selftest::run_tests): Call
selftest::forcibly_ggc_collect at the end of the selftests.
* selftest.h (selftest::forcibly_ggc_collect): New decl.

From-SVN: r241527

re PR target/78102 (GCC refuses to generate PCMPEQQ instruction for SSE4.1)

PR target/78102
* optabs.def (vcondeq_optab, vec_cmpeq_optab): New optabs.
* optabs.c (expand_vec_cond_expr): For comparison codes
EQ_EXPR and NE_EXPR, attempt vcondeq_optab as fallback.
(expand_vec_cmp_expr): For comparison codes
EQ_EXPR and NE_EXPR, attempt vec_cmpeq_optab as fallback.
* optabs-tree.h (expand_vec_cmp_expr_p, expand_vec_cond_expr_p):
Add enum tree_code argument.
* optabs-query.h (get_vec_cmp_eq_icode, get_vcond_eq_icode): New
inline functions.
* optabs-tree.c (expand_vec_cmp_expr_p): Add CODE argument.  For
CODE EQ_EXPR or NE_EXPR, attempt to use vec_cmpeq_optab as
fallback.
(expand_vec_cond_expr_p): Add CODE argument.  For CODE EQ_EXPR or
NE_EXPR, attempt to use vcondeq_optab as fallback.
* tree-vect-generic.c (expand_vector_comparison,
expand_vector_divmod, expand_vector_condition): Adjust
expand_vec_cmp_expr_p and expand_vec_cond_expr_p callers.
* tree-vect-stmts.c (vectorizable_condition,
vectorizable_comparison): Likewise.
* tree-vect-patterns.c (vect_recog_mixed_size_cond_pattern,
check_bool_pattern, search_type_for_mask_1): Likewise.
* expr.c (do_store_flag): Likewise.
* doc/md.texi (@code{vec_cmpeq@var{m}@var{n}},
@code{vcondeq@var{m}@var{n}}): Document.
* config/i386/sse.md (vec_cmpeqv2div2di, vcondeq<VI8F_128:mode>v2di):
New expanders.
testsuite/
* gcc.target/i386/pr78102.c: New test.

From-SVN: r241525

intrinsic.texi (cosd): New mathop.

gcc/fortran/
* intrinsic.texi (cosd): New mathop.

From-SVN: r241523

Handle negative times in filesystem::last_write_time

* src/filesystem/ops.cc
(last_write_time(const path&, file_time_type, error_code&)): Handle
negative times correctly.
* testsuite/experimental/filesystem/operations/last_write_time.cc:
Test writing file times.

From-SVN: r241522

Fix error handling in copy_file and equivalent

* src/filesystem/ops.cc (do_copy_file): Report an error if source or
destination is not a regular file (LWG 2712).
(equivalent): Fix error handling and result when only one file exists.
* testsuite/experimental/filesystem/operations/copy.cc: Remove files
created by tests. Test copying directories.
* testsuite/experimental/filesystem/operations/copy_file.cc: Remove
files created by tests.
* testsuite/experimental/filesystem/operations/equivalent.cc: New.
* testsuite/experimental/filesystem/operations/is_empty.cc: New.
* testsuite/experimental/filesystem/operations/read_symlink.cc: Remove
file created by test.
* testsuite/experimental/filesystem/operations/remove_all.cc: New.
* testsuite/util/testsuite_fs.h (~scoped_file): Only try to remove
file if path is non-empty, to support removal by other means.

From-SVN: r241521

Enable .XOR. operator with -std=legacy.

gcc/fortran/
* match.c (gfc_match_intrinsic_op): Match ".XOR." with -std=legacy.
* gfortran.texi: Document.

gcc/testsuite/gfortran.dg/
* dec_logical_xor_1.f90: New.
        * dec_logical_xor_2.f90: New.
        * dec_logical_xor_3.f03: New.

From-SVN: r241520

Enable %LOC as an rvalue with -std=legacy.

gcc/fortran/
        * primary.c (gfc_match_rvalue): Match %LOC as LOC with -std=legacy.
        * gfortran.texi: Document.

gcc/testsuite/gfortran.dg/
* dec_loc_rval_1.f90: New test.
        * dec_loc_rval_2.f90: New test.
        * dec_loc_rval_3.f90: New test.

From-SVN: r241519

Support TYPE as alias for PRINT with -fdec.

gcc/fortran/
* decl.c (gfc_match_type): New function.
* match.h (gfc_match_type): New function.
* match.c (gfc_match_if): Special case for one-line IFs.
* gfortran.texi: Update documentation.
* parse.c (decode_statement): Invoke gfc_match_type.

gcc/testsuite/gfortran.dg/
* dec_type_print.f90: New testcase.

From-SVN: r241518

Treat form feed as whitespace.

gcc/fortran/
* gfortran.texi: Document.
* gfortran.h (gfc_is_whitespace): Include form feed ('\f').

gcc/testsuite/gfortran.dg/
* feed_1.f90, feed_2.f90: New testcases.

From-SVN: r241517

Cleanup -fdec.

gcc/fortran/
* invoke.texi, gfortran.texi: Touch up documentation of -fdec.
* gfortran.h (gfc_option): Move flag_dec_structure out of gfc_option.
* decl.c (match_record_decl, gfc_match_decl_type_spec,
gfc_match_structure_decl): Ditto.
* match.c (gfc_match_member_sep): Ditto.
* options.c (gfc_handle_option): Ditto.
* lang.opt (fdec-structure): Use Fortran Var for flag_dec_structure.
* lang.opt (fdec): Use Fortran Var to create flag_dec.
* options.c (set_dec_flags): With -fdec enable -fcray-pointer,
-fd-lines-as-comments (default), -fdollar-ok, and legacy std flags.

From-SVN: r241516

Add BZ marker to recent commit

From-SVN: r241515

go-lang.c: remove a redundant cast

gcc/go/ChangeLog:
* go-lang.c (go_langhook_type_for_mode): Remove redundant cast
from result of GET_MODE_CLASS.  Minor formatting fixes.

From-SVN: r241514

v850.c (v850_handle_data_area_attribute): Fix fallthru comment.

* config/v850/v850.c (v850_handle_data_area_attribute): Fix fallthru
comment.
(v850_output_aligned_bss): Add missing break.

From-SVN: r241513

* config/m68k/m68k.c (m68k_get_reloc_decoration): Add gcc_unreachable.

From-SVN: r241512

Fix not caught use-after-scope with -O1 (PR sanitize/78106)

PR sanitizer/78106
* sanopt.c (imm_dom_path_with_freeing_call): Handle gasm
statements as they can also contain possibly a freeing call.
PR sanitizer/78106
* gcc.dg/asan/pr78106.c: New test.

From-SVN: r241511

Fix 2 typos in IPA ICF pass

* gcc.dg/ipa/ipa-icf-32.c: Removed one scanned pattern.
    Martin Liska  <mliska@suse.cz>

PR ipa/78099
* common.opt: Mark flag_ipa_icf_variables as Optimization flag.
* ipa-icf.c (sem_function::get_hash): Add target optimization
node to hash.

From-SVN: r241510

* parser.c (cp_parser_postfix_expression): Adding missing break;.

From-SVN: r241509

With -fpu=neon DI mode shifts are expanded after reload.

With -fpu=neon DI mode shifts are expanded after reload.  DI mode registers can
either fully or partially overlap on both ARM and Thumb-2.  However the shift
expansion code can only deal with the full overlap case, and generates incorrect
code for partial overlaps.  The fix is to add new variants that support either
full overlap or no overlap.

    gcc/
PR target/78041
* config/arm/neon.md (ashldi3_neon): Add "r 0 i" and "&r r i" variants.
Remove partial overlap check for shift by 1.
(ashldi3_neon): Likewise.
    testsuite/
* gcc.target/arm/pr78041.c: New test.

From-SVN: r241508

constraints.md (Q constraint): Document its use for Thumb-1.

2016-10-25  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    gcc/
    * config/arm/constraints.md (Q constraint): Document its use for
    Thumb-1.
    (Pf constraint): New constraint for relaxed, consume or relaxed memory
    models.
    * config/arm/sync.md (atomic_load<mode>): Add new ARMv8-M Baseline only
    alternatives to allow any register when memory model matches Pf and
    thus lda is used, but only low registers otherwise.  Use unpredicated
    output template for Thumb-1 targets.
    (atomic_store<mode>): Likewise for stl.
    (arm_load_exclusive<mode>): Add new ARMv8-M Baseline only alternative
    whose output template does not have predication.
    (arm_load_acquire_exclusive<mode>): Likewise.
    (arm_load_exclusivesi): Likewise.
    (arm_load_acquire_exclusivesi): Likewise.
    (arm_store_release_exclusive<mode>): Likewise.
    (arm_store_exclusive<mode>): Use unpredicated output template for
    Thumb-1 targets.

From-SVN: r241507

internal-fn.def (LAUNDER): New internal function.

* internal-fn.def (LAUNDER): New internal function.
* internal-fn.c (expand_LAUNDER): New function.
c-family/
* c-common.h (enum rid): Add RID_BUILTIN_LAUNDER.
* c-common.c (c_common_reswords): Add __builtin_launder.
cp/
* cp-tree.h (finish_builtin_launder): Declare.
* parser.c (cp_parser_postfix_expression): Handle RID_BUILTIN_LAUNDER.
* semantics.c (finish_builtin_launder): New function.
* pt.c (tsubst_copy_and_build): Handle instantiation of IFN_LAUNDER.
* constexpr.c (cxx_eval_internal_function): Handle IFN_LAUNDER.
(potential_constant_expression_1): Likewise.
testsuite/
* g++.dg/cpp1z/launder1.C: New test.
* g++.dg/cpp1z/launder2.C: New test.

From-SVN: r241506

Typo.

From-SVN: r241505

New avt target pass to work around performance loss by PR fix.

gcc/
New avt target pass to work around performance loss by PR fix.
PR target/71676
PR target/71678
* config/avr/avr.md (casesi_<mode>_sequence) [qi,hi]: New insn.
(*cmp<mode>) [qi,qq,uqq,hi,hq,uhq,ha,uha]: Rename to cmp<mode>3.
* config/avr/predicates.md (extend_operator): New.
* config/avr/avr-passes.def (avr_pass_casesi): Register new pass.
* config/avr/avr-protos.h (avr_casei_sequence_check_operands)
(make_avr_pass_casesi): New prototypes.
* config/avr/avr.c (print-rtl.h): Include it.
(pass_data avr_pass_data_casesi): Data for new pass.
(avr_pass_casesi): New class implementing rtl_opt_pass .avr-casesi.
(make_avr_pass_casesi, avr_parallel_insn_from_insns)
(avr_is_casesi_sequence, avr_casei_sequence_check_operands)
(avr_optimize_casesi): New functions.
gcc/testsuite/
PR target/71676
PR target/71678
* gcc.target/avr/pr71676-2.c: New test.

From-SVN: r241504

re PR target/71676 ([avr] casesi won't handle switch values larger than 16 bits)

gcc/
PR target/71676
PR target/71678
* config/avr/avr.md (casesi): Rewrite avoiding subregs of SI.
gcc/testsuite/
PR target/71676
PR target/71678
* gcc.target/avr/pr71676-1.c: New test.
* gcc.target/avr/pr71676-3.c: New test.
* gcc.target/avr/pr71676.c: New test.
* gcc.target/avr/pr71678.c: New test.

Co-Authored-By: Pitchumani Sivanupandi <pitchumani.sivanupandi@microchip.com>
From-SVN: r241503

Daily bump.

From-SVN: r241502

test_frame_common.h: Add noclone attribute such that cloned verions of tested functions are not created.

gcc/testsuite/ChangeLog:

2016-10-24  Kugan Vivekanandarajah  <kuganvi@linaro.org>

* gcc.target/aarch64/test_frame_common.h: Add noclone attribute
such that cloned verions of tested functions are not created.

From-SVN: r241499

re PR libfortran/77828 (Linking gfortran-7 compiled program with libgfortran of 5.x allowed but crashes when containing write to string)

2016-10-24  Jerry DeLisle  <jvdelisle@gcc.gnu.org>

PR fortran/77828
* io/io.h (st_parameter_dt): Reorder for readability and sanity.
* io/transfer.c (data_transfer_init): Remove TODO and enable the
runtime error message, rec= specifier not allowed in STREAM
access.
* libtool-version: Bump major version of libgfortran to 4.

* ioparm.def: Reorder dt parameters to match libgfortran.
* libgfortran.h: Swap definitions of GFC_INTERNAL_UNIT and
GFC_INTERNAL_UNIT4.

From-SVN: r241497

re PR fortran/71895 (ICE in gfc_compare_derived_types, at fortran/interface.c:520)

2016-10-24  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/71895
* interface.c (gfc_compare_derived_types):  Convert gcc_assert()
to a gfc_internal_error() to prevent an ICE.

2016-10-24  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/71895
* gfortran.dg/pr71895.f90: New test.

From-SVN: r241493

dwarf2out.c (gen_subprogram_die): Add DW_AT_reference or DW_AT_rvalue_reference attributes.

* dwarf2out.c (gen_subprogram_die): Add DW_AT_reference or
DW_AT_rvalue_reference attributes.

* cp-objcp-common.c (cp_decl_dwarf_attribute): Handle DW_AT_reference
and DW_AT_rvalue_reference.

* g++.dg/debug/dwarf2/ref-2.C: New test.

From-SVN: r241492