[PATCH AArch64 2/2] Replace temporary inline assembler for vget_high

* config/aarch64/arm_neon.h (__GET_HIGH): New macro.
(vget_high_f32, vget_high_f64, vget_high_p8, vget_high_p16,
vget_high_s8, vget_high_s16, vget_high_s32, vget_high_s64,
vget_high_u8, vget_high_u16, vget_high_u32, vget_high_u64):
Remove temporary __asm__ and reimplement.

From-SVN: r214952

[PATCH AArch64 1/2] Add execution tests of vget_low and vget_high

* gcc.target/aarch64/vget_high_1.c: New test.
* gcc.target/aarch64/vget_low_1.c: Likewise.

From-SVN: r214950

[PATCH AArch64 2/2] Remove vector compare/tst __builtins

* config/aarch64/aarch64-builtins.c (aarch64_fold_builtin): Remove code
handling cmge, cmgt, cmeq, cmtst.

* config/aarch64/aarch64-simd-builtins.def (cmeq, cmge, cmgt, cmle,
cmlt, cmgeu, cmgtu, cmtst): Remove.

* config/aarch64/arm_neon.h (vceq_*, vceqq_*, vceqz_*, vceqzq_*,
vcge_*, vcgeq_*, vcgez_*, vcgezq_*, vcgt_*, vcgtq_*, vcgtz_*,
vcgtzq_*, vcle_*, vcleq_*, vclez_*, vclezq_*, vclt_*, vcltq_*,
vcltz_*, vcltzq_*, vtst_*, vtstq_*): Use gcc vector extensions.

From-SVN: r214949

[PATCH AArch64 1/2] Improve codegen of vector compares inc. tst instruction

gcc/:

* config/aarch64/aarch64-builtins.c (aarch64_types_cmtst_qualifiers,
TYPES_TST): Define.
(aarch64_fold_builtin): Update pattern for cmtst.

* config/aarch64/aarch64-protos.h (aarch64_const_vec_all_same_int_p):
Declare.

* config/aarch64/aarch64-simd-builtins.def (cmtst): Update qualifiers.

* config/aarch64/aarch64-simd.md (aarch64_vcond_internal<mode><mode>):
Switch operands, separate out more cases, refactor.

(aarch64_cmtst<mode>): Rewrite pattern to match (plus ... -1).

* config/aarch64.c (aarch64_const_vec_all_same_int_p): Take single
argument; rename old version to...
(aarch64_const_vec_all_same_in_range_p): ...this.
(aarch64_print_operand, aarch64_simd_shift_imm_p): Follow renaming.

* config/aarch64/predicates.md (aarch64_simd_imm_minus_one): Define.

gcc/testsuite/:

* gcc.target/aarch64/simd/int_comparisons.x: New file.
* gcc.target/aarch64/simd/int_comparisons_1.c: New test.
* gcc.target/aarch64/simd/int_comparisons_2.c: Ditto.

From-SVN: r214948

[PATCH][AArch64] Tidy: remove unused qualifier_const_pointer

* config/aarch64/aarch64-builtins.c (enum aarch64_type_qualifiers):
Remove qualifier_const_pointer, update comment.

From-SVN: r214947

[PATCH][AArch64] One-liner: fix type of an add in SIMD registers

* config/aarch64/aarch64.md (adddi3_aarch64): set type to neon_add.

From-SVN: r214946

[PATCH AArch64] Remove varargs from aarch64_simd_expand_args

        * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Replace
        varargs with pointer parameter.
        (aarch64_simd_expand_builtin): pass pointer into previous.

From-SVN: r214945

[ARM/AArch64] Schedule alu_ext for Cortex-A53.

      * config/arm/cortex-a53.md (cortex_a53_alu_shift): Add alu_ext,
      alus_ext.

From-SVN: r214944

[PATCH AArch64] Add a builtin for rbit(q?)_p8; add intrinsics and tests.

gcc/:

* config/aarch64/aarch64-simd.md (aarch64_rbit<mode>): New pattern.
* config/aarch64/aarch64-simd-builtins.def (rbit): New builtin.
* config/aarch64/arm_neon.h (vrbit_s8, vrbit_u8, vrbitq_s8, vrbitq_u8):
Replace temporary asm with call to builtin.
(vrbit_p8, vrbitq_p8): New functions.

gcc/testsuite/:

* gcc.target/aarch64/simd/vrbit_1.c: New test.

From-SVN: r214943

cfgloop.c (mark_loop_for_removal): New function.

2014-09-05  Richard Biener  <rguenther@suse.de>

* cfgloop.c (mark_loop_for_removal): New function.
* cfgloop.h (mark_loop_for_removal): Declare.
* cfghooks.c (delete_basic_block): Use mark_loop_for_removal.
(merge_blocks): Likewise.
(duplicate_block): Likewise.
* except.c (sjlj_emit_dispatch_table): Likewise.
* tree-eh.c (cleanup_empty_eh_merge_phis): Likewise.
* tree-ssa-threadupdate.c (ssa_redirect_edges): Likewise.
(thread_through_loop_header): Likewise.

From-SVN: r214942

re PR tree-optimization/63148 (r187042 causes auto-vectorization failure for X86 for -m32.)

2014-09-05  Richard Biener  <rguenther@suse.de>

PR middle-end/63148
* fold-const.c (try_move_mult_to_index): Remove.
(fold_binary_loc): Do not call it.
* tree-data-ref.c (dr_analyze_indices): Strip conversions
from the base object again.

c-family/
* c-format.c (check_format_arg): Properly handle
effectively signed POINTER_PLUS_EXPR offset.

* gcc.dg/vect/pr63148.c: New testcase.
* c-c++-common/pr19807-1.c: Likewise.
* g++.dg/tree-ssa/pr19807.C: Adjust.
* g++.dg/tree-ssa/tmmti-2.C: Remove.

From-SVN: r214941

[AArch64 Obvious] Add a mode to operand 1 of sibcall_value_insn

gcc/

* config/aarch64/aarch64.md (sibcall_value_insn): Give operand 1
a mode.

From-SVN: r214940

re PR plugins/59335 (Plugin doesn't build on trunk)

2014-09-05  Joey Ye  <joey.ye@arm.com>

    PR plugin/59335
    * Makefile.in (PLUGIN_HEADERS): Add wide-int.h, signop.h, hash-map.h,

From-SVN: r214938

re PR target/55701 (Inline some instances of memset for ARM)

PR target/55701
* config/arm/arm.md (setmem): New pattern.
* config/arm/arm-protos.h (struct tune_params): New fields.
(arm_gen_setmem): New prototype.
* config/arm/arm.c (arm_slowmul_tune): Initialize new fields.
(arm_fastmul_tune, arm_strongarm_tune, arm_xscale_tune): Ditto.
(arm_9e_tune, arm_v6t2_tune, arm_cortex_tune): Ditto.
(arm_cortex_a8_tune, arm_cortex_a7_tune): Ditto.
(arm_cortex_a15_tune, arm_cortex_a53_tune): Ditto.
(arm_cortex_a57_tune, arm_cortex_a5_tune): Ditto.
(arm_cortex_a9_tune, arm_cortex_a12_tune): Ditto.
(arm_v7m_tune, arm_v6m_tune, arm_fa726te_tune): Ditto.
(arm_const_inline_cost): New function.
(arm_block_set_max_insns): New function.
(arm_block_set_non_vect_profit_p): New function.
(arm_block_set_vect_profit_p): New function.
(arm_block_set_unaligned_vect): New function.
(arm_block_set_aligned_vect): New function.
(arm_block_set_unaligned_non_vect): New function.
(arm_block_set_aligned_non_vect): New function.
(arm_block_set_vect, arm_gen_setmem): New functions.

testsuite
* gcc.target/arm/memset-inline-1.c: New test.
* gcc.target/arm/memset-inline-2.c: New test.
* gcc.target/arm/memset-inline-3.c: New test.
* gcc.target/arm/memset-inline-4.c: New test.
* gcc.target/arm/memset-inline-5.c: New test.
* gcc.target/arm/memset-inline-6.c: New test.
* gcc.target/arm/memset-inline-7.c: New test.
* gcc.target/arm/memset-inline-8.c: New test.
* gcc.target/arm/memset-inline-9.c: New test.
* gcc.target/arm/memset-inline-10.c: New test.

From-SVN: r214937

arm.md (arm_movqi_insn): Use Uh instead of m constraint.

* config/arm/arm.md (arm_movqi_insn): Use Uh instead of m constraint.

From-SVN: r214936

arm.c (output_move_neon): Handle REG explicitly.

* config/arm/arm.c (output_move_neon): Handle REG explicitly.

From-SVN: r214935

make dead_debug_insert_temp take an rtx_insn *

gcc/ChangeLog:

2014-09-04  Trevor Saunders  <tsaunders@mozilla.com>

* valtrack.c (dead_debug_insert_temp): Take an rtx_insn * instead of
an rtx.
* valtrack.h: Adjust.

From-SVN: r214934

make several emit functions take an rtx_insn *

gcc/ChangeLog:

2014-09-04  Trevor Saunders  <tsaunders@mozilla.com>

* emit-rtl.c (emit_insn_before_noloc): Take an rtx_insn * instead of
an rtx.
(emit_jump_insn_before_noloc): Likewise.
(emit_call_insn_before_noloc): Likewise.
(emit_label_before): Likewise.
(emit_label_after): Likewise.
(emit_insn_before_setloc): Likewise.
(emit_jump_insn_before_setloc): Likewise.
(emit_call_insn_before_setloc): Likewise.
(emit_call_insn_before): Likewise.
* rtl.h: Adjust.

From-SVN: r214933

Remove a cast in cse.c

gcc/
2014-09-05  David Malcolm  <dmalcolm@redhat.com>

* cse.c (cse_insn): Strengthen local "new_rtx" from rtx to
rtx_insn *, eliminating a checked cast.

From-SVN: r214932

Params to modified_between_p

gcc/
2014-09-05  David Malcolm  <dmalcolm@redhat.com>

* rtl.h (modified_between_p): Strengthen params 2 and 3 from
const_rtx to const rtx_insn *.
* rtlanal.c (modified_between_p): Likewise, eliminating a checked
cast.

From-SVN: r214931

fixup_args_size_notes takes a pair of rtx_insn

gcc/
2014-09-05  David Malcolm  <dmalcolm@redhat.com>

* emit-rtl.c (try_split): Update NULL_RTX to NULL in call to
fixup_args_size_notes.
* expr.c (fixup_args_size_notes): Strengthen first two params from
rtx to rtx_insn *, eliminating a checked cast.
* rtl.h (fixup_args_size_notes): Strengthen first two params from
rtx to rtx_insn *.

From-SVN: r214930

Return type of get_ready_element

gcc/
2014-09-05  David Malcolm  <dmalcolm@redhat.com>

* haifa-sched.c (get_ready_element): Strengthen return type from
rtx to rtx_insn *.
* sched-int.h (get_ready_element): Likewise.

From-SVN: r214929

Remove dg-options for sh*-*-* from gcc.c-torture/execute/pr44683.c.

From-SVN: r214928

Daily bump.

From-SVN: r214927

re PR target/63165 (r214816 breaks Java build and 435.gromacs)

2014-09-04  Segher Boessenkool  <segher@kernel.crashing.org>

PR target/63165
* config/rs6000/rs6000.md (floatsi<mode>2_lfiwax_mem): Use
indexed_or_indirect_operand instead of memory_operand.
(floatsi<mode>2_lfiwzx_mem): Ditto.

From-SVN: r214924

use rtx_insn * a little more

gcc/ChangeLog:

2014-09-04  Trevor Saunders  <tsaunders@mozilla.com>

* config/bfin/bfin.c, config/c6x/c6x.c, config/m32c/m32c.c,
config/mn10300/mn10300.c, config/s390/s390.c, config/sh/sh.c,
ifcvt.c, reorg.c: Change types of variables from rtx to rtx_insn *.

From-SVN: r214923

Change the types of arguments and return values for several functions from rtx to rtx_insn *

gcc/ChangeLog:

2014-09-04  Trevor Saunders  <tsaunders@mozilla.com>

* emit-rtl.c (get_first_nonnote_insn): Return rtx_insn * instead of
rtx.
(get_last_nonnote_insn): Likewise.
(next_nonnote_insn_bb): Take rtx_insn * instead of rtx.
* resource.c (find_basic_block): Likewise.
* rtl.h: Adjust.
* rtlanal.c (keep_with_call_p): Take const rtx_insn * instead of
const_rtx.

From-SVN: r214922

insn_latency takes a pair of rtx_insn

gcc/
2014-09-04  David Malcolm  <dmalcolm@redhat.com>

* genattr.c (main): Within the prototype of insn_latency written
out to insn-attr.h, strengthen both params from rtx to rtx_insn *.
* genautomata.c (output_internal_maximal_insn_latency_func):
Within the implementation of insn_latency written out to
insn-automata.c, strengthen both params from rtx to rtx_insn *,
eliminating a pair of checked casts.

From-SVN: r214921

eh_returnjump_p takes an rtx_insn

2014-09-04  David Malcolm  <dmalcolm@redhat.com>

gcc/
* jump.c (eh_returnjump_p): Strengthen param "insn" from rtx to
rtx_insn *.

* rtl.h (eh_returnjump_p): Likewise.

From-SVN: r214919

* Makefile.in (TAGS): Handle constructs in timevar.def.

From-SVN: r214918

Revert mistakenly committed change.

From-SVN: r214917

Remove file that was mistakenly added.

From-SVN: r214916

Merge remote-tracking branch 'origin/aldyh/debug-early' into debug-early

Conflicts:
gcc/ChangeLog.debug-early
gcc/dwarf2out.c

From-SVN: r214915

Revert incorrect merging of debug-early branch.

From-SVN: r214914

Merge remote-tracking branch 'origin/aldyh/debug-early' into debug-early

Conflicts:
gcc/ChangeLog.debug-early
gcc/dwarf2out.c

From-SVN: r214912

+ * c/c-decl.c (write_global_declarations_1): Call global_decl() + with early=true.

+       * c/c-decl.c (write_global_declarations_1): Call global_decl()
+       with early=true.
+       (write_global_declarations_2): Call global_decl() with
+       early=false.
+       * dbxout.c (dbxout_global_decl): New argument.
+       * debug.c (do_nothing_debug_hooks): Use debug_nothing_tree_bool
+       for global_decl hook.
+       (debug_nothing_tree_bool): New.
+       (struct gcc_debug_hooks): New argument to global_decl.
+       * dwarf2out.c (output_die): Add misc debugging information.
+       (gen_variable_die): Do not reparent children.
+       (dwarf2out_global_decl): Add new documentation.  Add EARLY
+       argument.
+       (dwarf2out_decl): Make sure we don't generate new DIEs if we
+       already have a DIE.
+       * cp/name-lookup.c (do_namespace_alias): New argument to
+       global_decl debug hook.
+       * fortran/trans-decl.c (gfc_emit_parameter_debug_info): Same.
+       * godump.c (go_global_decl): Same.
+       * lto/lto-lang.c (lto_write_globals): Same.
+       * sdbout.c (sdbout_global_decl): Same.
+       * toplev.c (emit_debug_global_declarations): Same.
+       * vmsdbgout.c (vmsdbgout_global_decl): Same.
+       * tree.c (free_lang_data_in_decl): Do not call
+       dwarf2out_early_decl from here.

From-SVN: r214911

Add missing ChangeLog entries for last commit.

From-SVN: r214910

dwarf2out.c (struct die_struct): Add dumped_early field.

* dwarf2out.c (struct die_struct): Add dumped_early field.
(reparent_child): New.
(splice_child_die): Use reparent_child.
(gen_subprogram_die): Do not regenerate parameters if previously
dumped.
(gen_variable_die): Fix parent of decls that have been dumped
early to reflect correct context.
Do not regenerate decls if previously dumped.
(dwarf2out_decl): Add return value.
(dwarf2out_early_decl): Set dumped_early bit.

From-SVN: r214909

Make output_die() output the parent DIE for each DIE.

From-SVN: r214908

Make gen_formal_parameter_die() merge DIEs from a previous execution.

Do not set DW_AT_inline incorrectly on all functions.

Do not remove formal parameters when a previous DIE has been found.
Instead, augment the missing bits later.

From-SVN: r214907

dearly: Emit DIEs for decls early in the compilation process.

This is the original patch from Michael Matz, ported to a more recent
mainline.

From-SVN: r214906

re PR target/62040 (internal compiler error: in simplify_const_unary_operation, at simplify-rtx.c:1555)

PR target/62040
* config/aarch64/iterators.md (VQ_NO2E, VQ_2E): New iterators.
* config/aarch64/aarch64-simd.md (move_lo_quad_internal_<mode>): Split
it into two patterns.
(move_lo_quad_internal_be_<mode>): Likewise.

* gcc.target/aarch64/pr62040.c: New test.

From-SVN: r214905

options.texi: Document that Var and Init are required if CPP is given.

gcc/ChangeLog:

2014-09-04  Manuel López-Ibáñez  <manu@gcc.gnu.org>

* doc/options.texi: Document that Var and Init are required if CPP
is given.
* optc-gen.awk: Require Var and Init if CPP is given.
* common.opt (Wpedantic): Use Init.

libcpp/ChangeLog:

2014-09-04  Manuel López-Ibáñez  <manu@gcc.gnu.org>

* macro.c (replace_args): Use cpp_pedwarning, cpp_warning and
CPP_W flags.
* include/cpplib.h: Add CPP_W_C90_C99_COMPAT and CPP_W_PEDANTIC.
* init.c (cpp_create_reader): Do not init to -1 here.
* expr.c (num_binary_op): Use cpp_pedwarning.

gcc/c-family/ChangeLog:

2014-09-04  Manuel López-Ibáñez  <manu@gcc.gnu.org>

* c.opt (Wc90-c99-compat,Wc++-compat,Wcomment,Wendif-labels,
Winvalid-pch,Wlong-long,Wmissing-include-dirs,Wmultichar,Wpedantic,
(Wdate-time,Wtraditional,Wundef,Wvariadic-macros): Add CPP, Var
and Init.
* c-opts.c (c_common_handle_option): Do not handle here.
(sanitize_cpp_opts): Likewise.
* c-common.c (struct reason_option_codes_t): Handle
CPP_W_C90_C99_COMPAT and CPP_W_PEDANTIC.

gcc/testsuite/ChangeLog:

2014-09-04  Manuel López-Ibáñez  <manu@gcc.gnu.org>

* gcc.dg/cpp/endif-pedantic2.c: More general options do not
override specific ones, but specific ones do.

From-SVN: r214904

rs6000.c (special_handling_values): Add SH_EXTRACT.

[gcc]

2014-09-04  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

* config/rs6000/rs6000.c (special_handling_values): Add
SH_EXTRACT.
(rtx_is_swappable_p): Look for patterns with a VEC_SELECT, perhaps
wrapped in a VEC_DUPLICATE, representing an extract.  Mark these
as swappable with special handling SH_EXTRACT.  Remove
UNSPEC_VSX_XXSPLTW from the list of disallowed unspecs for the
optimization.
(adjust_extract): New function.
(handle_special_swappables): Add default to case statement; add
case for SH_EXTRACT that calls adjust_extract.
(dump_swap_insn_table): Handle SH_EXTRACT.

[gcc/testsuite]

2014-09-04  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

* gcc.target/powerpc/swaps-p8-13.c: New test.
* gcc.target/powerpc/swaps-p8-14.c: New test.
* gcc.target/powerpc/swaps-p8-15.c: New test.

From-SVN: r214903

vsx.md (*vsx_extract_<mode>_load): Always match selection of 0th memory doubleword, regardless of endianness.

[gcc]

2014-09-04  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

* config/rs6000/vsx.md (*vsx_extract_<mode>_load): Always match
selection of 0th memory doubleword, regardless of endianness.

[gcc/testsuite]

2014-09-04  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

* gcc.target/powerpc/vsx-extract-1.c:  Test 0th doubleword
regardless of endianness.

From-SVN: r214902

rx.h (HARD_REGNO_MODE_OK): Add braces.

2014-09-04  Jan-Benedict Glaw  <jbglaw@lug-owl.de>

* config/rx/rx.h (HARD_REGNO_MODE_OK): Add braces.

From-SVN: r214901

re PR ipa/61659 (Extra undefined symbol because of devirtualization)

PR ipa/61659
* decl.c (duplicate_decls): Check DECL_DECLARED_INLINE_P on
newdecl, not olddecl.

From-SVN: r214900

re PR debug/60655 (ICE: output_operand: invalid expression as operand)

PR debug/60655
* dwarf2out.c (mem_loc_descriptor <PLUS>): Return NULL if addend
can't be output.

From-SVN: r214899

Add target hook to override DWARF2 frame register size

gcc/

* target.def (TARGET_DWARF_FRAME_REG_MODE): New target hook.
* targhooks.c (default_dwarf_frame_reg_mode): New function.
* targhooks.h (default_dwarf_frame_reg_mode): New prototype.
* doc/tm.texi.in (TARGET_DWARF_FRAME_REG_MODE): Document.
* doc/tm.texi: Regenerate.
* dwarf2cfi.c (expand_builtin_init_dwarf_reg_sizes): Abstract mode
selection logic to default_dwarf_frame_reg_mode.

From-SVN: r214898

Daily bump.

From-SVN: r214897

compiler: Add precise type information on the heap.

* go-gcc.cc (Gcc_backend::implicit_variable): Remove init
parameter.  Add is_hidden parameter.
(Gcc_backend::implicit_variable_set_init): New method.
(Gcc_backend::implicit_variable_reference): New method.

From-SVN: r214894

invoke.texi: Document that -Wlogical-not-parentheses is enabled by -Wall.

* doc/invoke.texi: Document that -Wlogical-not-parentheses is enabled
by -Wall.
c-family/
* c.opt (Wlogical-not-parentheses): Enable by -Wall.

From-SVN: r214893

[multiple changes]

2014-09-03  Fritz Reese  <Reese-Fritz@zai.com>

        PR fortran/62174
        * decl.c (variable_decl): Don't overwrite typespecs of Cray
        * pointees
        when matching a component declaration.

2014-09-02  Fritz Reese  <Reese-Fritz@zai.com>

        PR fortran/62174
        * gfortran.dg/cray_pointers_11.f90: New.

From-SVN: r214891

rtl.texi (RTX_AUTOINC): Document that the first operand is the automodified register.

gcc/
* doc/rtl.texi (RTX_AUTOINC): Document that the first operand is
the automodified register.

From-SVN: r214890

output.h (get_some_local_dynamic_name): Declare.

gcc/
* output.h (get_some_local_dynamic_name): Declare.
* final.c (some_local_dynamic_name): New variable.
(get_some_local_dynamic_name): New function.
(final_end_function): Clear some_local_dynamic_name.
* config/alpha/alpha.c (machine_function): Remove some_ld_name.
(get_some_local_dynamic_name, get_some_local_dynamic_name_1): Delete.
(print_operand): Report an error if '%&' is used inappropriately.
* config/i386/i386.c (get_some_local_dynamic_name): Delete.
(get_some_local_dynamic_name_1): Delete.
* config/rs6000/rs6000.c (machine_function): Remove some_ld_name.
(rs6000_get_some_local_dynamic_name): Delete.
(rs6000_get_some_local_dynamic_name_1): Delete.
(print_operand): Report an error if '%&' is used inappropriately.
* config/s390/s390.c (machine_function): Remove some_ld_name.
(get_some_local_dynamic_name, get_some_local_dynamic_name_1): Delete.
(print_operand): Assert that get_some_local_dynamic_name is nonnull.
* config/sparc/sparc.c: Include rtl-iter.h.
(machine_function): Remove some_ld_name.
(sparc_print_operand): Report an error if '%&' is used inappropriately.
(get_some_local_dynamic_name, get_some_local_dynamic_name_1): Delete.

From-SVN: r214889

aarch64: Improve epilogue unwind info

        * config/aarch64/aarch64.c (aarch64_popwb_single_reg): Remove.
        (aarch64_popwb_pair_reg): Remove.
        (aarch64_set_frame_expr): Remove.
        (aarch64_restore_callee_saves): Add CFI_OPS argument; fill it with
        the restore ops performed by the insns generated.
        (aarch64_expand_epilogue): Attach CFI_OPS to the stack deallocation
        insn.  Perform the calls_eh_return addition later; do not attempt to
        preserve the CFA in that case.  Don't use aarch64_set_frame_expr.
        (aarch64_expand_prologue): Use REG_CFA_ADJUST_CFA directly, or no
        special markup at all.  Load cfun->machine->frame.hard_fp_offset
        into a local variable.
        (aarch64_frame_pointer_required): Don't check calls_alloca.

From-SVN: r214886

opts.c (default_options_optimization): Adjust max-combine-insns to 2 for -Og.

2014-09-03  Richard Biener  <rguenther@suse.de>

* opts.c (default_options_optimization): Adjust
max-combine-insns to 2 for -Og.

From-SVN: r214879

re PR ipa/62015 (ipa-cp-clone uses a clone that is too specialized for the call context)

2014-09-03  Martin Jambor  <mjambor@suse.cz>

PR ipa/62015
* ipa-cp.c (intersect_aggregates_with_edge): Handle impermissible
pass-trough jump functions correctly.

testsuite/
* g++.dg/ipa/pr62015.C: New test.

From-SVN: r214878

re PR ipa/61986 (ICE on valid code at -O3 on x86_64-linux-gnu indecide_about_value, at ipa-cp.c:3480)

2014-09-03  Martin Jambor  <mjambor@suse.cz>

PR ipa/61986
* ipa-cp.c (find_aggregate_values_for_callers_subset): Chain
created replacements in ascending order of offsets.
(known_aggs_to_agg_replacement_list): Likewise.

gcc/testsuite/
* gcc.dg/ipa/pr61986.c: New test.

From-SVN: r214877

re PR c/62294 (Missing "passing argument [...] from incompatible pointer type" warning.)

PR c/62294
* gcc.dg/pr62294.c: New test.
* gcc.dg/pr62294.h: New file.

From-SVN: r214876

Fix for tree-ssa-pre

* tree-ssa-sccvn.c (vn_reference_lookup_call): default (NULL) value
is set to preserve uninitialized value for vnresult.

From-SVN: r214873

pr39228.c: Use dg-add-options instead of dg-options.

* gcc.c-torture/execute/pr39228.c: Use dg-add-options instead
of dg-options.  Add "inline" keyword to test functions.

From-SVN: r214872

re PR c/62024 (__atomic_always_lock_free is not a constant expression)

PR c/62024
* c-parser.c (c_parser_static_assert_declaration_no_semi): Strip no-op
conversions.

* g++.dg/cpp0x/pr62024.C: New test.
* gcc.dg/pr62024.c: New test.

From-SVN: r214871

[NDS32] No need to take padding into consideration in Andes ABI2 because we can...

[NDS32] No need to take padding into consideration in Andes ABI2 because
        we can pass arguments in registers for variadic function.

  -- By default GCC uses must_pass_in_stack_var_size_or_pad for TARGET_MUST_PASS_IN_STACK.
     For Andes ABI2, the data layout in memory will be incorrect when calling variadic function
     under big-endian configuration.

* config/nds32/nds32.c (nds32_must_pass_in_stack): New implementation
for TARGET_MUST_PASS_IN_STACK.

From-SVN: r214869

[NDS32] Provide TARGET_ARG_PARTIAL_BYTES so that we can have structure argument
        partially in registers and stack.

  -- This is needed if we would like to pass variadic arguments via registers
     that are pused on the stack by callee.  So that va_arg() can access
     arguments continously in memory.

* config/nds32/nds32.c (nds32_arg_partial_bytes): New implementation
for TARGET_ARG_PARTIAL_BYTES.

From-SVN: r214868

[NDS32] Emit stack push instructions for varargs at prologue stage and emit...

[NDS32]  Emit stack push instructions for varargs at prologue stage and
         emit stack adjustment instructions for varargs at epilogue stage.

  -- Note that we do not need to handle it in v3push/v3pop prologue/epilogue
     expanding because we do not apply v3push/v3pop for variadic function.

* config/nds32/nds32.c (nds32_expand_prologue): Emit stack push
instructions for varargs implementation.
(nds32_expand_epilogue): Emit stack adjustment instructions for
varargs implementation.

From-SVN: r214867

[NDS32] Since we are going to provide new heuristic for fp-as-gp optimization,

        we better not to set cfun->machine->fp_as_gp_p at prologue expanding.

* config/nds32/nds32.c (nds32_expand_prologue): Suppress fp-as-gp
optimization detection.

From-SVN: r214866

[NDS32] Deal with nameless arguments in nds32_function_arg and nds32_function_arg_advance.

  1. We have caller prepare nameless arguments in the way of calling a normal function.

     e.g. Assume a function: void va_test (int n, ...);
          When calling "va_test (6, 11, 22, 33, 44, 55, 66);",
               r0    <- 6
               r1    <- 11
               r2    <- 22
               r3    <- 33
               r4    <- 44
               r5    <- 55
            [sp + 0] <- 66

  2. Note that we DO NOT handle it for TARGET_HARD_FLOAT case.
     The TARGET_HARD_FLOAT should push all the nameless arguments into
     stack and GCC is able to take care of them itself.
     In addition, we have not implemented hard float on trunk yet.

* config/nds32/nds32.c (nds32_function_arg): Deal with nameless
arguments.
(nds32_function_arg_advance): Deal with nameless arguments.
* config/nds32/nds32.h (NDS32_ARG_PASS_IN_REG_P): Split it into ...
(NDS32_ARG_ENTIRE_IN_GPR_REG_P): ... this one and ...
(NDS32_ARG_PARTIAL_IN_GPR_REG_P): ... this one.

From-SVN: r214865

tree-ssa-pre.c (alloc_expression_id): Use quick_grow_cleared.

2014-09-03  Richard Biener  <rguenther@suse.de>

* tree-ssa-pre.c (alloc_expression_id): Use quick_grow_cleared.
(struct bb_bitmap_sets): Remove deferred member.
(BB_DEFERRED): Remove.
(defer_or_phi_translate_block): Remove.
(compute_antic_aux): Remove deferring of blocks, assert
proper iteration order.
(compute_antic): Do not set BB_DEFERRED.
(eliminate): Allocate el_avail of proper size initially.

From-SVN: r214864

[NDS32] Since argument pointer must be 8-byte alignment...

[NDS32] Since argument pointer must be 8-byte alignment, our first parameter may
        require 4-byte offset if pushed varargs is not 8-byte aligned.

* config/nds32/nds32.h (FIRST_PARM_OFFSET): Set proper location
according to the value of crtl->args.pretend_args_size.

From-SVN: r214863

[NDS32] Prepare necessary varargs information in compute_stack_frame().

* config/nds32/nds32.c (nds32_compute_stack_frame): Prepare necessary
varargs information.

From-SVN: r214862

[NDS32] Implement TARGET_SETUP_INCOMING_VARARGS target hook.

* config/nds32/nds32.c (nds32_setup_incoming_varargs): New
implementation for TARGET_SETUP_INCOMING_VARARGS.
(nds32_strict_argument_naming): Refine comment.
* config/nds32/nds32.h (TARGET_SOFT_FLOAT, TARGET_HARD_FLOAT):
Define for future implementation.

From-SVN: r214861

adxintrin.h (_subborrow_u32): New.

gcc/
2014-09-02  Ilya Tocar  <ilya.tocar@intel.com>

        * config/i386/adxintrin.h (_subborrow_u32): New.
        (_addcarry_u32): Ditto.
        (_subborrow_u64): Ditto.
        (_addcarry_u64): Ditto.
        * config/i386/i386.c (ix86_builtins): Add IX86_BUILTIN_SBB32,
        IX86_BUILTIN_SBB64.
        (ix86_init_mmx_sse_builtins): Handle __builtin_ia32_sbb_u32,
        __builtin_ia32_sbb_u64

testsuite/
2014-09-02  Ilya Tocar  <ilya.tocar@intel.com>

        * gcc.target/i386/adx-addcarryx32-1.c: Test addcarry, subborrow.
        * gcc.target/i386/adx-addcarryx32-2.c: Ditto.
        * gcc.target/i386/adx-addcarryx32-3.c: Ditto.
        * gcc.target/i386/adx-addcarryx64-1.c: Ditto.
        * gcc.target/i386/adx-addcarryx64-2.c: Ditto.
        * gcc.target/i386/adx-addcarryx64-3.c: Ditto.

From-SVN: r214860

[NDS32] Define and rename some GPR-specific stuff.

* config/nds32/nds32.c (nds32_function_arg): Define and rename some
GPR-specific stuff.
(nds32_function_arg_advance): Likewise.
(nds32_init_cumulative_args): Likewise.
* config/nds32/nds32.h (NDS32_MAX_GPR_REGS_FOR_ARGS): Define.
(NDS32_FIRST_GPR_REGNUM): Define.
(NDS32_LAST_GPR_REGNUM): Define.
(NDS32_AVAILABLE_REGNUM_FOR_GPR_ARG): Define.
(NDS32_ARG_PASS_IN_REG_P): Use NDS32_MAX_GPR_REGS_FOR_ARGS.
(FUNCTION_ARG_REGNO_P): Use NDS32_MAX_GPR_REGS_FOR_ARGS.
(machine_function): Use GRP-specific stuff.

From-SVN: r214859

[NDS32] Remove unused variables.

* config/nds32/nds32.c (nds32_expand_prologue): Remove unused variables.
(nds32_expand_epilogue): Likewise.
(nds32_expand_prologue_v3push): Likewise.
(nds32_expand_epilogue_v3pop): Likewise.

From-SVN: r214858

[NDS32] We need to make sure it is NOT a variadic function so that some v3push/v3pop preparation can be applied.

  -- The rationale of such avoidance is that the 'pop25' instruction also
     contains return behavior and we have no chance to further adjust stack
     pointer before returning to caller, which is very important to
     variadic function.
  -- So we do not setup v3push/v3pop (push25/pop25) information for a
     variadic function.  It will use normal stack push/pop instead.

* config/nds32/nds32.c (nds32_compute_stack_frame): Do not use
v3push/v3pop for variadic function.
* config/nds32/nds32.md (prologue, epilogue): Likewise.

From-SVN: r214857

[NDS32] Consider varargs situation when creating stack operation assembly code.

  -- In fact, we only need to take care of 'push.s' situation,
     because we have to push all the nameless arguments into stack.
  -- We do not have to use 'pop.s' to restore them back to registers,
     because we can just adjust stack pointer to set frame location.
     This operation should be done in epilogue expanding.
  -- Because 'pop25' instruction also performs return operation, we will
     not use v3push/v3pop on isr and vararg functions.
     Therefore, their instruction length should be 4 bytes.

* config/nds32/nds32-md-auxiliary.c (nds32_output_stack_push):
Check rtx for varargs implementation.
(nds32_output_stack_pop): Likewise.
* config/nds32/nds32-protos.h: Have a rtx argument for
nds32_output_stack_push and nds32_output_stack_pop.
* config/nds32/nds32.md: Likewise.

From-SVN: r214856

[NDS32] Add a function to indentify if FUNC is an interrupt service routine.

* config/nds32/nds32-isr.c (nds32_isr_function_p): Define new function
to check if FUNC is an interrupt service routine.
* config/nds32/nds32-protos.h (nds32_isr_function_p): Declaration.

From-SVN: r214855

[NDS32] Preparation in nds32.h of using registers to save varargs.

  -- The varargs area may also need padding bytes so that we can ensure
     that argument pointer is 8-byte alignment.

* config/nds32/nds32.h (machine_function): Add some fields for variadic
arguments implementation.

From-SVN: r214854

[NDS32] In nds32_valid_stack_push_pop_p(), we look into OP rtx to see if we indeed save $fp/$gp/$lp registers.

* config/nds32/nds32-predicates.c
(nds32_valid_stack_push_pop): Rename to ...
(nds32_valid_stack_push_pop_p): ... this.
* config/nds32/nds32-protos.h: Likewise.
* config/nds32/predicates.md: Likewise.

From-SVN: r214853

[NDS32] Refine the implementation and consider CFA restore information for stack v3push/v3pop.

* config/nds32/nds32.c (nds32_gen_stack_v3push): Rename to ...
(nds32_emit_stack_v3push): ... this.
(nds32_gen_stack_v3pop): Rename to ...
(nds32_emit_stack_v3pop): ... this and consider CFA restore
information.

From-SVN: r214852

[NDS32] Refine the implementation and consider CFA restore information for stack push/pop multiple.

* config/nds32/nds32.c (nds32_gen_stack_push_multiple): Rename to ...
(nds32_emit_stack_push_multiple): ... this.
(nds32_gen_stack_pop_multiple): Rename to ...
(nds32_emit_stack_pop_multiple): ... this and consider CFA restore
information.

From-SVN: r214851

re PR bootstrap/61078 (ESA mode bootstrap failure since r209897)

2014-09-03  Andreas Krebbel  <Andreas.Krebbel@de.ibm.com>

PR target/61078
* config/s390/s390.md ("*negdi2_31"): Add s390_split_ok_p check
and add a second splitter to handle the remaining cases.

2014-09-03  Andreas Krebbel  <Andreas.Krebbel@de.ibm.com>

PR target/61078
* gcc.target/s390/pr61078.c: New testcase.

From-SVN: r214850

[NDS32] Define PIC_OFFSET_TABLE_REGNUM to $gp register.

* config/nds32/nds32.h (PIC_OFFSET_TABLE_REGNUM): Define.

From-SVN: r214849

20111227-2.c: Compile only for x86 targets.

* gcc.dg/20111227-2.c: Compile only for x86 targets.
* gcc.dg/20111227-3.c: Ditto.

From-SVN: r214848

gcc-dg.exp (set-target-env-var): Change env to ::env.

gcc/testsuite/ChangeLog:

2014-09-03  Max Ostapenko  <m.ostapenko@partner.samsung.com>

* lib/gcc-dg.exp (set-target-env-var): Change env to ::env.

From-SVN: r214846

Missed that file in r213079 of 2014-07-26

2014-09-03  Tobias Burnus  <burnus@net-b.de>

        PR fortran/61881
        PR fortran/61888
        PR fortran/57305
        * gfortran.dg/sizeof_4.f90: New.

From-SVN: r214843

use rtx_code_label in lab_rtx_for_bb hash map

gcc/ChangeLog:

* cfgexpand.c (label_rtx_for_bb): Change type to
hash_map<basic_block, rtx_code_label *> *.
(expand_gimple_basic_block): Adjust.
(pass_expand::execute): Likewise.

From-SVN: r214842

use rtx_code_label more

gcc/

* asan.c, cfgexpand.c, config/alpha/alpha.md, config/arm/arm.c,
config/epiphany/epiphany.md, config/h8300/h8300.c, config/i386/i386.md,
config/m32r/m32r.c, config/mcore/mcore.md, config/mips/mips.c,
config/mips/mips.md, config/nios2/nios2.c, config/pa/pa.c,
config/s390/s390.c, config/s390/s390.md, config/sh/sh-mem.cc,
config/sh/sh.c, config/sparc/sparc.c, dojump.c, function.c, optabs.c,
stmt.c: Assign the result of gen_label_rtx to rtx_code_label * instead
of rtx.

From-SVN: r214841

libgcov-interface.c (STRONG_ALIAS): New.

* libgcov-interface.c (STRONG_ALIAS): New.
(__gcov_flush): Call __gcov_reset_int.
(__gcov_reset): Strong alias for ...
(__gcov_reset_ing): ... this renamed hidden version.
* libgcov.h (__gcov_reset_int): New declaration.

From-SVN: r214840

Daily bump.

From-SVN: r214839

support ggc hash_map  and hash_set

gcc/ChangeLog:

* alloc-pool.c: Include coretypes.h.
* cgraph.h, dbxout.c, dwarf2out.c, except.c, except.h, function.c,
function.h, symtab.c, tree-cfg.c, tree-eh.c: Use hash_map and
hash_set instead of htab.
* ggc-page.c (in_gc): New variable.
(ggc_free): Do nothing if a collection is taking place.
(ggc_collect): Set in_gc appropriately.
* ggc.h (gt_ggc_mx(const char *)): New function.
(gt_pch_nx(const char *)): Likewise.
(gt_ggc_mx(int)): Likewise.
(gt_pch_nx(int)): Likewise.
* hash-map.h (hash_map::hash_entry::ggc_mx): Likewise.
(hash_map::hash_entry::pch_nx): Likewise.
(hash_map::hash_entry::pch_nx_helper): Likewise.
(hash_map::hash_map): Adjust.
(hash_map::create_ggc): New function.
(gt_ggc_mx): Likewise.
(gt_pch_nx): Likewise.
* hash-set.h (default_hashset_traits::ggc_mx): Likewise.
(default_hashset_traits::pch_nx): Likewise.
(hash_set::hash_entry::ggc_mx): Likewise.
(hash_set::hash_entry::pch_nx): Likewise.
(hash_set::hash_entry::pch_nx_helper): Likewise.
(hash_set::hash_set): Adjust.
(hash_set::create_ggc): New function.
(hash_set::elements): Likewise.
(gt_ggc_mx): Likewise.
(gt_pch_nx): Likewise.
* hash-table.h (hash_table::hash_table): Adjust.
(hash_table::m_ggc): New member.
(hash_table::~hash_table): Adjust.
(hash_table::expand): Likewise.
(hash_table::empty): Likewise.
(gt_ggc_mx): New function.
(hashtab_entry_note_pointers): Likewise.
(gt_pch_nx): Likewise.

From-SVN: r214834

rs6000-builtin.def (XVCVSXDDP_SCALE): New built-in definition.

[gcc]

2014-09-02  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

* config/rs6000/rs6000-builtin.def (XVCVSXDDP_SCALE):  New
built-in definition.
(XVCVUXDDP_SCALE): Likewise.
(XVCVDPSXDS_SCALE): Likewise.
(XVCVDPUXDS_SCALE): Likewise.
* config/rs6000/rs6000-c.c (altivec_overloaded_builtins):  Add
entries for VSX_BUILTIN_XVCVSXDDP_SCALE,
VSX_BUILTIN_XVCVUXDDP_SCALE, VSX_BUILTIN_XVCVDPSXDS_SCALE, and
VSX_BUILTIN_XVCVDPUXDS_SCALE.
* config/rs6000/rs6000-protos.h (rs6000_scale_v2df): New
prototype.
* config/rs6000/rs6000.c (real.h): New include.
(rs6000_scale_v2df): New function.
* config/rs6000/vsx.md (UNSPEC_VSX_XVCVSXDDP): New unspec.
(UNSPEC_VSX_XVCVUXDDP): Likewise.
(UNSPEC_VSX_XVCVDPSXDS): Likewise.
(UNSPEC_VSX_XVCVDPUXDS): Likewise.
(vsx_xvcvsxddp_scale): New define_expand.
(vsx_xvcvsxddp): New define_insn.
(vsx_xvcvuxddp_scale): New define_expand.
(vsx_xvcvuxddp): New define_insn.
(vsx_xvcvdpsxds_scale): New define_expand.
(vsx_xvcvdpsxds): New define_insn.
(vsx_xvcvdpuxds_scale): New define_expand.
(vsx_xvcvdpuxds): New define_insn.
* doc/extend.texi (vec_ctf): Add new prototypes.
(vec_cts): Likewise.
(vec_ctu): Likewise.
(vec_splat): Likewise.
(vec_div): Likewise.
(vec_mul): Likewise.

[gcc/testsuite]

2014-09-02  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

* gcc.target/powerpc/builtins-1.c: Add tests for vec_ctf,
vec_cts, and vec_ctu.
* gcc.target/powerpc/builtins-2.c: Likewise.

From-SVN: r214831

badalloc1.C (arena_size): Scale according to target pointer size.

* g++.old-deja/g++.eh/badalloc1.C [!STACK_SIZE && !__FreeBSD__]
[!__sun__ && !__hpux__] (arena_size): Scale according to
target pointer size.

From-SVN: r214828

re PR fortran/62270 (-Wlogical-not-parentheses warnings)

PR fortran/62270
* interface.c (compare_parameter): Fix condition.
* trans-expr.c (gfc_conv_procedure_call): Likewise.

* gfortran.dg/pointer_intent_7.f90: Adjust dg-error.

From-SVN: r214827

[2/2] Vectorise lroundf, lfloorf, lceilf using the new ARMv8-A vcvt* instructions.

PR target/62275
* config/arm/neon.md
(neon_vcvt<NEON_VCVT:nvrint_variant><su_optab><VCVTF:mode>
<v_cmp_result>): New pattern.
* config/arm/iterators.md (NEON_VCVT): New int iterator.
* config/arm/arm_neon_builtins.def (vcvtav2sf, vcvtav4sf, vcvtauv2sf,
vcvtauv4sf, vcvtpv2sf, vcvtpv4sf, vcvtpuv2sf, vcvtpuv4sf, vcvtmv2sf,
vcvtmv4sf, vcvtmuv2sf, vcvtmuv4sf): New builtin definitions.
* config/arm/arm.c (arm_builtin_vectorized_function): Handle
BUILT_IN_LROUNDF, BUILT_IN_LFLOORF, BUILT_IN_LCEILF.

PR target/62275
* gcc.target/arm/vect-lceilf_1.c: New test.
* gcc.target/arm/vect-lfloorf_1.c: Likewise.
* gcc.target/arm/vect-lroundf_1.c: Likewise.

From-SVN: r214826

[ARM][1/2] Implement lceil, lfloor, lround optabs with new ARMv8-A instructions.

      PR target/62275
      * config/arm/iterators.md (FIXUORS): New code iterator.
      (VCVT): New int iterator.
      (su_optab): New code attribute.
      (su): Likewise.
      * config/arm/vfp.md (l<vrint_pattern><su_optab><mode>si2): New pattern.

      PR target/62275
      * gcc.target/arm/lceil-vcvt_1.c: New test.
      * gcc.target/arm/lfloor-vcvt_1.c: Likewise.
      * gcc.target/arm/lround-vcvt_1.c: Likewise.

From-SVN: r214825

[AArch64] Use CC_Z and CC_NZ with csinc and similar instructions.

* config/aarch64/predicates.md (aarch64_comparison_operation):
New special predicate.
* config/aarch64/aarch64.md (*csinc2<mode>_insn): Use
aarch64_comparison_operation instead of matching an operator.
Update operand numbers.
(csinc3<mode>_insn): Likewise.
(*csinv3<mode>_insn): Likewise.
(*csneg3<mode>_insn): Likewise.
(ffs<mode>2): Update gen_csinc3<mode>_insn callsite.
* config/aarch64/aarch64.c (aarch64_get_condition_code):
Return -1 instead of aborting on invalid condition codes.
(aarch64_print_operand): Update aarch64_get_condition_code callsites
to assert that the returned condition code is valid.
* config/aarch64/aarch64-protos.h (aarch64_get_condition_code): Export.

From-SVN: r214824

DR 1453

/cp
2014-09-02  Paolo Carlini  <paolo.carlini@oracle.com>

DR 1453
* class.c (check_field_decls): A class of literal type cannot have
volatile non-static data members and base classes.
(explain_non_literal_class): Update.

/testsuite
2014-09-02  Paolo Carlini  <paolo.carlini@oracle.com>

DR 1453
* g++.dg/cpp0x/constexpr-volatile.C: New.
* g++.dg/ext/is_literal_type2.C: Likewise.

From-SVN: r214823

Makefile.in (TAGS): Handle constructs in common.opt, rtl.def, tree.def, and gimple.def

* Makefile.in (TAGS): Handle constructs in common.opt, rtl.def,
tree.def, and gimple.def

From-SVN: r214822