opts.c (finish_options): Disable aggressive opts for sanitizer.

gcc

2014-11-18  Marat Zakirov  <m.zakirov@samsung.com>

* opts.c (finish_options): Disable aggressive opts for sanitizer.
(common_handle_option): Move code to finish_options.

gcc/testsuite

2014-11-18  Marat Zakirov  <m.zakirov@samsung.com>

* c-c++-common/asan/aggressive-opts.c: New test.

From-SVN: r217690

re PR sanitizer/63802 (UBSan doesn't catch misaligned access if address is 16-bytes (or more) aligned)

2014-11-18  Yury Gribov  <y.gribov@samsung.com>

PR sanitizer/63802

gcc/
* stor-layout.c (min_align_of_type): Respect user alignment
more.

gcc/testsuite/
* c-c++-common/ubsan/pr63802.c: New test.

From-SVN: r217689

passes.c (remove_cgraph_node_from_order): New.

gcc/

* passes.c (remove_cgraph_node_from_order): New.
(do_per_function_toporder): Register cgraph removal
hook.

gcc/testsuite/

* g++.dg/pr63766.C: New.

From-SVN: r217688

arm.c (arm_issue_rate): Return 2 for cortex-m7.

2014-11-17  Terry Guo  <terry.guo@arm.com>

* config/arm/arm.c (arm_issue_rate): Return 2 for cortex-m7.
* config/arm/arm.md (generic_sched): Exclude cortex-m7.
(generic_vfp): Likewise.
* config/arm/cortex-m7.md: Pipeline description for cortex-m7.

From-SVN: r217687

Daily bump.

From-SVN: r217686

re PR rtl-optimization/63906 (lra_remat miscompiles glibc on aarch64)

2014-11-17  Vladimir Makarov  <vmakarov@redhat.com>

PR rtl-optimization/63906
* lra-remat.c (operand_to_remat): Check SP and
frame_pointer_required.

From-SVN: r217683

Support for unroll and jam optimization.

From-SVN: r217682

thunderx.md: Remove copyright which should not have been there.

2014-11-17  Andrew Pinski  <apinski@cavium.com>

* config/aarch64/thunderx.md: Remove copyright which should not
have been there.

From-SVN: r217680

rs6000.c (RELOAD_REG_AND_M16): Add support for Altivec style vector loads that ignore the bottom 3 bits of the...

[gcc]
2014-11-17  Michael Meissner  <meissner@linux.vnet.ibm.com>
    Ulrich Weigand  <Ulrich.Weigand@de.ibm.com>

* config/rs6000/rs6000.c (RELOAD_REG_AND_M16): Add support for
Altivec style vector loads that ignore the bottom 3 bits of the
address.
(rs6000_debug_addr_mask): New function to print the addr_mask
values if debugging.
(rs6000_debug_print_mode): Call rs6000_debug_addr_mask to print
out addr_mask.
(rs6000_setup_reg_addr_masks): Add support for Altivec style
vector loads that ignore the bottom 3 bits of the address.  Allow
pre-increment and pre-decrement on floating point, even if the
-mupper-regs-{sf,df} options were used.
(rs6000_init_hard_regno_mode_ok): Rework DFmode support if
-mupper-regs-df.  Add support for -mupper-regs-sf.  Rearrange code
placement for direct move support.
(rs6000_option_override_internal): Add checks for -mupper-regs-df
requiring -mvsx, and -mupper-regs-sf requiring -mpower8-vector.
If -mupper-regs, set both -mupper-regs-sf and -mupper-regs-df,
depending on the underlying cpu.
(rs6000_secondary_reload_fail): Add ATTRIBUTE_NORETURN.
(rs6000_secondary_reload_toc_costs): Helper function to identify
costs of a TOC load for secondary reload support.
(rs6000_secondary_reload_memory): Helper function for secondary
reload, to determine if a particular memory operation is directly
handled by the hardware, or if it needs support from secondary
reload to create a valid address.
(rs6000_secondary_reload): Rework code, to be clearer.  If the
appropriate -mupper-regs-{sf,df} is used, use FPR registers to
reload scalar values, since the FPR registers have D-form
addressing. Move most of the code handling memory to the function
rs6000_secondary_reload_memory, and use the reg_addr structure to
determine what type of address modes are supported.  Print more
debug information if -mdebug=addr.
(rs6000_secondary_reload_inner): Rework entire function to be more
general.  Use the reg_addr bits to determine what type of
addressing is supported.
(rs6000_preferred_reload_class): Rework.  Move constant handling
into a single place.  Prefer using FLOAT_REGS for scalar floating
point.
(rs6000_secondary_reload_class): Use a FPR register to move a
value from an Altivec register to a GPR, and vice versa.  Move VSX
handling above traditional floating point.

* config/rs6000/rs6000.md (mov<mode>_hardfloat, FMOVE32 case):
Delete some spaces in the constraints.
(DF->DF move peephole2): Disable if -mupper-regs-{sf,df} to
allow using FPR registers to load/store an Altivec register for
scalar floating point types.
(SF->SF move peephole2): Likewise.
(DFmode splitter): Add a define_split to move floating point
constants to the constant pool before register allocation.
Normally constants are put into the pool immediately, but
-ffast-math delays putting them into the constant pool for the
reciprocal approximation support.
(SFmode splitter): Likewise.

* config/rs6000/rs6000.opt (-mupper-regs-df): Make option public.
(-mupper-regs-sf): Likewise.

* config/rs6000/rs6000-c.c (rs6000_target_modify_macros): Define
__UPPER_REGS_DF__ if -mupper-regs-df.  Define __UPPER_REGS_SF__ if
-mupper-regs-sf.
(-mupper-regs): New combination option that sets -mupper-regs-sf
and -mupper-regs-df by default if the cpu supports the instructions.

* doc/invoke.texi (RS/6000 and PowerPC Options): Document
-mupper-regs, -mupper-regs-sf, and -mupper-regs-df.

* config/rs6000/predicates.md (memory_fp_constant): New predicate
to return true if the operand is a floating point constant that
must be put into the constant pool, before register allocation
occurs.

* config/rs6000/rs6000-cpus.def (ISA_2_6_MASKS_SERVER): Enable
-mupper-regs-df by default.
(ISA_2_7_MASKS_SERVER): Enable -mupper-regs-sf by default.
(POWERPC_MASKS): Add -mupper-regs-{sf,df} as options set by the
various -mcpu=... options.
(power7 cpu): Enable -mupper-regs-df by default.

* doc/invoke.texi (RS/6000 and PowerPC Options): Document
-mupper-regs.

[gcc/testsuite]
2014-11-17  Michael Meissner  <meissner@linux.vnet.ibm.com>

* gcc.target/powerpc/p8vector-ldst.c: Rewrite to use 40 live
floating point variables instead of using asm to test allocating
values to the Altivec registers.

* gcc.target/powerpc/upper-regs-sf.c: New -mupper-regs-sf and
-mupper-regs-df tests.
* gcc.target/powerpc/upper-regs-df.c: Likewise.

* config/rs6000/predicates.md (memory_fp_constant): New predicate

Co-Authored-By: Ulrich Weigand <uweigand@de.ibm.com>
From-SVN: r217679

Export "detect_leaks=0"

PR bootstrap/63888
* bootstrap-asan.mk (ASAN_OPTIONS): Export "detect_leaks=0".

From-SVN: r217678

re PR c++/33911 (attribute deprecated vs. templates)

PR c++/33911
gcc/cp/
* call.c (build_call_a): Don't warn_deprecated_use here.
(build_over_call): Or here.
* decl2.c (mark_used): Do it here.
(is_late_template_attribute): Attribute deprecated is not deferred.
(cplus_decl_attributes): Propagate TREE_DEPRECATED out to the template.
* parser.c (cp_parser_template_name): Warn about deprecated template.
(cp_parser_template_argument): Likewise.
libstdc++-v3/
* include/backward/binders.h: Suppress -Wdeprecated-declarations.
* include/ext/array_allocator.h: Likewise.

From-SVN: r217677

ira-conflicts.c (build_conflict_bit_table): Add the current object to OBJECTS_LIVE after traversing OBJECTS_LIVE.

       * ira-conflicts.c (build_conflict_bit_table): Add the current
        object to OBJECTS_LIVE after traversing OBJECTS_LIVE.

From-SVN: r217676

ipa-cp.c (ipa_get_indirect_edge_target_1): Handle speculation.

* ipa-cp.c (ipa_get_indirect_edge_target_1): Handle speculation.
(ipa_get_indirect_edge_target): Add SPECULATIVE argument.
(devirtualization_time_bonus): Use it.
(ipcp_discover_new_direct_edges): Likewise.
* ipa-inline-analysis.c (estimate_edge_devirt_benefit): Update.
* ipa-prop.h (ipa_get_indirect_edge_target): Update prototype.

From-SVN: r217675

Add -ftree-tail-merge to tail-merge testcases

2014-11-17  Tom de Vries  <tom@codesourcery.com>

* gcc.dg/pr43864-2.c: Add -ftree-tail-merge to dg-options.
* gcc.dg/pr43864-3.c: Same.
* gcc.dg/pr43864-4.c: Same.
* gcc.dg/pr43864.c: Same.
* gcc.dg/pr50763.c: Same.
* gcc.dg/pr51879-12.c: Same.
* gcc.dg/pr51879-16.c: Same.
* gcc.dg/pr51879-17.c: Same.
* gcc.dg/pr51879-18.c: Same.
* gcc.dg/pr51879-2.c: Same.
* gcc.dg/pr51879-3.c: Same.
* gcc.dg/pr51879-4.c: Same.
* gcc.dg/pr51879-6.c: Same.
* gcc.dg/pr51879-7.c: Same.
* gcc.dg/pr51879.c: Same.

From-SVN: r217674

Fix scan patterns for pr43864-{2,3,4].c

2014-11-17  Tom de Vries  <tom@codesourcery.com>

* gcc.dg/pr43864-2.c: Fix scan-tree-dump-times scan pattern.
* gcc.dg/pr43864-3.c: Same.
* gcc.dg/pr43864-4.c: Same.

From-SVN: r217673

re PR c++/50473 ([C++0x] ICE in type_has_nontrivial_copy_init, at cp/tree.c:2574)

PR c++/50473
* decl.c (cp_finish_decl): Don't try to process a non-dependent
constant initializer for a reference.
* pt.c (value_dependent_expression_p): A reference is always
dependent.
* call.c (extend_ref_init_temps_1): Also clear TREE_SIDE_EFFECTS
on any NOP_EXPRs.

From-SVN: r217672

tree.c (free_lang_data_in_decl): Set DECL_FUNCTION_SPECIFIC_OPTIMIZATION to optimization_default_node.

* tree.c (free_lang_data_in_decl): Set DECL_FUNCTION_SPECIFIC_OPTIMIZATION
to optimization_default_node.

From-SVN: r217671

Handle C++14 constexpr flow control.

* constexpr.c (cxx_eval_loop_expr, cxx_eval_switch_expr): New.
(cxx_eval_statement_list): New.
(cxx_eval_constant_expression): Handle LABEL_EXPR,
CASE_LABEL_EXPR, GOTO_EXPR, LOOP_EXPR, SWITCH_EXPR.  Handle jump
semantics of RETURN_EXPR.
(many functions): Add jump_target parameter.
(returns, breaks, continues, switches, label_matches): New.
* cp-tree.h (LABEL_DECL_BREAK, LABEL_DECL_CONTINUE): New.
* cp-gimplify.c (begin_bc_block): Set them.

From-SVN: r217670

cp-gimplify.c (genericize_cp_loop): Use LOOP_EXPR.

* cp-gimplify.c (genericize_cp_loop): Use LOOP_EXPR.
(genericize_for_stmt): Handle null statement-list.

From-SVN: r217669

cgraphunit.c (analyze_functions): Use opt_for_fn.

* cgraphunit.c (analyze_functions): Use opt_for_fn.
* cgraph.h (cgraph_node::optimize_for_size_p): Likewise.

From-SVN: r217668

cgraph.c (symbol_table::create_edge): Use opt_for_fn.

* cgraph.c (symbol_table::create_edge): Use opt_for_fn.
(cgraph_node::cannot_return_p): Likewise.
(cgraph_edge::cannot_lead_to_return_p): Likewise.
(cgraph_edge::maybe_hot_p): Likewise.

From-SVN: r217667

predict.c (maybe_hot_frequency_p): Use opt_for_fn.

* predict.c (maybe_hot_frequency_p): Use opt_for_fn.
(optimize_function_for_size_p): Likewise.
(probably_never_executed): Likewise; replace cfun by fun.

From-SVN: r217666

[AArch64] Extend aarch64_simd_vec_set pattern, replace asm for vld1_lane

gcc/:

* config/aarch64/aarch64-simd.md (aarch64_simd_vec_set<mode>): Add
variant reading from memory and assembling to ld1.

* config/aarch64/arm_neon.h (vld1_lane_f32, vld1_lane_f64, vld1_lane_p8,
vld1_lane_p16, vld1_lane_s8, vld1_lane_s16, vld1_lane_s32,
vld1_lane_s64, vld1_lane_u8, vld1_lane_u16, vld1_lane_u32,
vld1_lane_u64, vld1q_lane_f32, vld1q_lane_f64, vld1q_lane_p8,
vld1q_lane_p16, vld1q_lane_s8, vld1q_lane_s16, vld1q_lane_s32,
vld1q_lane_s64, vld1q_lane_u8, vld1q_lane_u16, vld1q_lane_u32,
vld1q_lane_u64): Replace asm with vset_lane and pointer dereference.

gcc/testsuite/:

* gcc.target/aarch64/vld1_lane.c: New test.

From-SVN: r217665

* constexpr.c (use_new_call): Always use new call handling.

From-SVN: r217664

C++14 constexpr support (minus loops and multiple returns)

C++14 constexpr support (minus loops and multiple returns)
gcc/
* tree-inline.c (copy_fn): New.
* tree-inline.h: Declare it.
gcc/cp/
* constexpr.c (use_new_call): New macro.
(build_data_member_initialization): Ignore non-mem-inits.
(check_constexpr_bind_expr_vars): Remove C++14 checks.
(constexpr_fn_retval): Likewise.
(check_constexpr_ctor_body): Do nothing in C++14.
(massage_constexpr_body): In C++14 only collect mem-inits.
(get_function_named_in_call): Handle null CALL_EXPR_FN.
(cxx_bind_parameters_in_call): Build bindings in same order as
parameters.  Don't treat iniviref parms specially in new call mode.
(cxx_eval_call_expression): If use_new_call, do constexpr expansion
based on DECL_SAVED_TREE rather than the massaged constexpr body.
Set up ctx->object from AGGR_INIT_EXPR_SLOT if we don't have one.
(is_sub_constant_expr): Don't mess with ctx.ctor here.
(cxx_eval_component_reference): A null element means we're mid-
initialization.
(cxx_eval_store_expression, cxx_eval_increment_expression): New.
(cxx_eval_constant_expression): Handle RESULT_DECL, DECL_EXPR,
MODIFY_EXPR, STATEMENT_LIST, BIND_EXPR, USING_STMT,
PREINCREMENT_EXPR, POSTINCREMENT_EXPR, PREDECREMENT_EXPR,
POSTDECREMENT_EXPR.  Don't look into DECL_INITIAL of variables in
constexpr functions.  In new-call mode find parms in the values table.
(potential_constant_expression_1): Handle null CALL_EXPR_FN.
Handle STATEMENT_LIST, MODIFY_EXPR, MODOP_EXPR, IF_STMT,
PREINCREMENT_EXPR, POSTINCREMENT_EXPR, PREDECREMENT_EXPR,
POSTDECREMENT_EXPR, BIND_EXPR, WITH_CLEANUP_EXPR,
CLEANUP_POINT_EXPR, MUST_NOT_THROW_EXPR, TRY_CATCH_EXPR,
EH_SPEC_BLOCK, EXPR_STMT, DECL_EXPR, CASE_LABEL_EXPR, BREAK_STMT,
CONTINUE_STMT, USING_STMT, IF_STMT, DO_STMT, FOR_STMT, WHILE_STMT,
SWITCH_STMT, ASM_EXPR.
(cxx_eval_vec_init_1): Call build_aggr_init_expr.
(cxx_eval_indirect_ref): Don't return a CONSTRUCTOR when the
caller wants an lvalue.
(cxx_eval_outermost_constant_expr): Pull object out of AGGR_INIT_EXPR.
(maybe_constant_init): Look through INIT_EXPR.
(ensure_literal_type_for_constexpr_object): Set
cp_function_chain->invalid_constexpr.
* cp-tree.h (struct language_function): Add invalid_constexpr bitfield.
* decl.c (start_decl): Set cp_function_chain->invalid_constexpr.
(check_for_uninitialized_const_var): Likewise.
(maybe_save_function_definition): Check it.
* parser.c (cp_parser_jump_statement): Set
cp_function_chain->invalid_constexpr.
(cp_parser_asm_definition): Likewise.

From-SVN: r217663

aarch64-builtins.c (TYPES_CREATE): Remove.

gcc/:

* config/aarch64/aarch64-builtins.c (TYPES_CREATE): Remove.
* config/aarch64/aarch64-simd-builtins.def (create): Remove.
* config/aarch64/aarch64-simd.md (aarch64_create<mode>): Remove.
* config/aarch64/arm_neon.h (vcreate_f64, vreinterpret_f64_s64,
vreinterpret_f64_u64): Replace __builtin_aarch64_createv1df with C casts.
* config/aarch64/iterators.md (VD1): Remove.

gcc/testsuite/:

* gcc.target/aarch64/simd/vfma_f64.c: Add asm volatile memory.
* gcc.target/aarch64/simd/vfms_f64.c: Likewise.

From-SVN: r217662

[AArch64] Remove crypto extension from default for cortex-a53, cortex-a57

* config/aarch64/aarch64-cores.def (cortex-a53): Remove
AARCH64_FL_CRYPTO from feature flags.
(cortex-a57): Likewise.
(cortex-a57.cortex-a53): Likewise.

From-SVN: r217661

re PR c++/52282 ([C++0x] rejects-valid issues with decltype/constexpr)

PR c++/52282
* decl.c (build_ptrmemfunc_type): Don't build a different
RECORD_TYPE for a qualified PMF.
* cp-tree.h (TYPE_PTRMEMFUNC_FN_TYPE): Merge cv-quals.
(TYPE_PTRMEMFUNC_FN_TYPE_RAW): New.
* decl2.c (cplus_decl_attributes): Use TYPE_PTRMEMFUNC_FN_TYPE_RAW.
* tree.c (cp_walk_subtrees): Likewise.
(cp_build_qualified_type_real): Remove special PMF handling.

From-SVN: r217660

tree.c (free_lang_data_in_decl): Annotate all functio nbodies with DECL_FUNCTION_SPECIFIC_TARGET.

* tree.c (free_lang_data_in_decl): Annotate all functio nbodies with
DECL_FUNCTION_SPECIFIC_TARGET.
* i386.c (ix86_set_current_function): Handle explicit default options.

* lto.c (lto_read_decls): Do not rebuild DECL_FUNCTION_SPECIFIC_TARGET.

From-SVN: r217659

builtins.c (expand_builtin_memcpy_with_bounds): Use target hook instead of BNDmode.

* builtins.c (expand_builtin_memcpy_with_bounds): Use target hook
instead of BNDmode.
(expand_builtin_mempcpy_with_bounds): Likewise.
(expand_builtin_memset_with_bounds): Likewise.

From-SVN: r217658

tree-ssa-strlen.c: include ipa-chkp.h, cgraph.h, ipa-ref.h, plugin-api.h.

gcc/

* tree-ssa-strlen.c: include ipa-chkp.h, cgraph.h,
ipa-ref.h, plugin-api.h.
(get_string_length): Handle calls with bounds.
(adjust_last_stmt): Likewise.
(handle_builtin_strchr): Likewise.
(handle_builtin_strcpy): Likewise.
(handle_builtin_memcpy): Likewise.
(handle_builtin_strcat): Likewise.

gcc/testsuite/

* gcc.target/i386/chkp-strlen-1.c: New.
* gcc.target/i386/chkp-strlen-2.c: New.
* gcc.target/i386/chkp-strlen-3.c: New.
* gcc.target/i386/chkp-strlen-4.c: New.
* gcc.target/i386/chkp-strlen-5.c: New.

From-SVN: r217657

tree-chkp-opt.c (chkp_get_nobnd_fndecl): New.

gcc/

* tree-chkp-opt.c (chkp_get_nobnd_fndecl): New.
(chkp_get_nochk_fndecl): New.
(chkp_optimize_string_function_calls): New.
(chkp_opt_execute): Call chkp_optimize_string_function_calls.
* tree-cfg.h (insert_cond_bb): New.
* tree-cfg.c (insert_cond_bb): New.

gcc/testsuite/

* gcc.target/i386/chkp-stropt-1.c: New.
* gcc.target/i386/chkp-stropt-2.c: New.
* gcc.target/i386/chkp-stropt-3.c: New.
* gcc.target/i386/chkp-stropt-4.c: New.
* gcc.target/i386/chkp-stropt-5.c: New.
* gcc.target/i386/chkp-stropt-6.c: New.
* gcc.target/i386/chkp-stropt-7.c: New.
* gcc.target/i386/chkp-stropt-8.c: New.
* gcc.target/i386/chkp-stropt-9.c: New.
* gcc.target/i386/chkp-stropt-10.c: New.
* gcc.target/i386/chkp-stropt-11.c: New.
* gcc.target/i386/chkp-stropt-12.c: New.
* gcc.target/i386/chkp-stropt-13.c: New.
* gcc.target/i386/chkp-stropt-14.c: New.
* gcc.target/i386/chkp-stropt-15.c: New.
* gcc.target/i386/chkp-stropt-16.c: New.

From-SVN: r217656

tree-core.h (built_in_class): Add builtin codes to be used by Pointer Bounds Checker for instrumented builtin...

* tree-core.h (built_in_class): Add builtin codes to be used
by Pointer Bounds Checker for instrumented builtin functions.
* tree-streamer-in.c: Include ipa-chkp.h.
(streamer_get_builtin_tree): Created instrumented decl if
required.
* ipa-chkp.h (chkp_maybe_clone_builtin_fndecl): New.
* ipa-chkp.c (chkp_build_instrumented_fndecl): Support builtin
function decls.
(chkp_maybe_clone_builtin_fndecl): New.
(chkp_maybe_create_clone): Support builtin function decls.
(chkp_versioning): Clone builtin functions.
* tree-chkp.c (chkp_instrument_normal_builtin): New.
(chkp_add_bounds_to_call_stmt): Support builtin functions.
(chkp_replace_function_pointer): Likewise.
* builtins.c (expand_builtin_memcpy_args): New.
(expand_builtin_memcpy): Call expand_builtin_memcpy_args.
(expand_builtin_memcpy_with_bounds): New.
(expand_builtin_mempcpy_with_bounds): New.
(expand_builtin_mempcpy_args): Add orig_exp arg. Support
BUILT_IN_CHKP_MEMCPY_NOBND_NOCHK
(expand_builtin_memset_with_bounds): New.
(expand_builtin_memset_args): Support BUILT_IN_CHKP_MEMSET_NOBND_NOCHK.
(expand_builtin_with_bounds): New.
* builtins.h (expand_builtin_with_bounds): New.
* expr.c (expand_expr_real_1): Support instrumented builtin calls.

From-SVN: r217655

Replace unsigned long with __SIZE_TYPE__

* g++.dg/ipa/pr63894.C (new): Replace unsigned long with
__SIZE_TYPE__.

From-SVN: r217654

Add more comments to some gimple accessors

gcc/ChangeLog:

* gimple.h (gimple_set_visited, gimple_visited_p)
(gimple_set_plf, gimple_plf, gimple_set_uid, gimple_uid): Add more
comments to these accessors.

Signed-off-by: Dodji Seketeli <dodji@redhat.com>
From-SVN: r217653

re PR tree-optimization/63898 (r217560 caused segfault building 462.libquantum from cpu2006)

2014-11-17  Richard Biener  <rguenther@suse.de>

PR middle-end/63898
PR middle-end/63883
* gfortran.dg/pr63883.f90: New testcase.

From-SVN: r217652

avr-log.c (avr_log_set_avr_log): Set avr_log_details to "all".

* config/avr/avr-log.c (avr_log_set_avr_log) [TARGET_ALL_DEBUG]:
Set avr_log_details to "all".

From-SVN: r217651

re PR tree-optimization/63898 (r217560 caused segfault building 462.libquantum from cpu2006)

2014-11-17  Richard Biener  <rguenther@suse.de>

PR middle-end/63898
* match.pd: Guard X / CST -> X * CST' transform against
zero CST.

From-SVN: r217650

Added Dodji Seketeli as line map maintainer

* MAINTAINERS (Various Maintainers): Added myself as line map
maintainer.
* ChangeLog: Update this.

Signed-off-by: Dodji Seketeli <dodji@redhat.com>
From-SVN: r217649

Add testcase for PR 63894

2014-11-17  Markus Trippelsdorf  <markus@trippelsdorf.de>

  PR ipa/63894
  * g++.dg/ipa/pr63894.C: New test.

From-SVN: r217648

thumb1.md (*addsi3_cbranch_scratch): Updated to UAL format.

gcc/
2014-11-17  Terry Guo  <terry.guo@arm.com>

* config/arm/thumb1.md (*addsi3_cbranch_scratch): Updated to UAL
format.

gcc/testsuite/
2014-11-17  Terry Guo  <terry.guo@arm.com>

* gcc.target/arm/thumb1-ual-1.c: New test.

From-SVN: r217647

ifcvt.c (HAVE_cbranchcc4): Define.

2014-11-17  Zhenqiang Chen  <zhenqiang.chen@arm.com>

* ifcvt.c (HAVE_cbranchcc4): Define.
(noce_emit_cmove, noce_get_alt_condition, noce_get_condition):
Use HAVE_cbranchcc4.

From-SVN: r217646

aarch64.c (aarch64_code_to_ccmode, [...]): New functions.

2014-11-17  Zhenqiang Chen  <zhenqiang.chen@linaro.org>

* config/aarch64/aarch64.c (aarch64_code_to_ccmode,
aarch64_convert_mode, aarch64_gen_ccmp_first,
aarch64_gen_ccmp_next): New functions.
(TARGET_GEN_CCMP_FIRST, TARGET_GEN_CCMP_NEXT): Define.

From-SVN: r217645

aarch64-protos.h (aarch64_ccmp_mode_to_code): New.

2014-11-17  Zhenqiang Chen  <zhenqiang.chen@linaro.org>

* config/aarch64/aarch64-protos.h (aarch64_ccmp_mode_to_code): New.
* aarch64.c (aarch64_nzcv_codes): New data.
(aarch64_ccmp_mode_to_code): New.
(aarch64_print_operand): Output nzcv.
config/aarch64/aarch64.md (cbranchcc4, *ccmp_and, *ccmp_ior, cstorecc4):
New patterns.
(cstore<mode>4): Handle ccmp_cc_register.
* config/aarch64/predicates.md (const0_operand): New.

From-SVN: r217644

aarch64-modes.def: Define ccmp CC mode.

2014-11-17  Zhenqiang Chen  <zhenqiang.chen@linaro.org>

* config/aarch64/aarch64-modes.def: Define ccmp CC mode.
* config/aarch64/aarch64.c (aarch64_get_condition_code_1): New function
extacted from aarch64_get_condition_code.
(aarch64_get_condition_code): Call aarch64_get_condition_code_1.
config/aarch64/predicates.md (ccmp_cc_register): New predicate.

From-SVN: r217643

constraints.md (Usn, [...]): New constraints.

2014-11-17  Zhenqiang Chen  <zhenqiang.chen@linaro.org>

* config/aarch64/constraints.md (Usn, aarch64_ccmp_immediate,
aarch64_ccmp_operand): New constraints.

From-SVN: r217642

Makefile.in: Add ccmp.o.

2014-11-17  Zhenqiang Chen  <zhenqiang.chen@linaro.org>

* Makefile.in: Add ccmp.o.
* ccmp.c: New file.
* ccmp.h: New file.
* expr.c: include "ccmp.h"
(expand_cond_expr_using_cmove): Handle VOIDmode.
(expand_expr_real_1): Try to expand ccmp.

From-SVN: r217641

cfgexpand.c (expand_gimple_cond): Check ccmp.

2014-11-17  Zhenqiang Chen  <zhenqiang.chen@linaro.org>

* cfgexpand.c (expand_gimple_cond): Check ccmp.
* expmed.c (emit_cstore): Make it global.
* expmed.h: #include "insn-codes.h"
(emit_cstore): New prototype.
* expr.c (expand_operands): Make it global.
* expr.h (expand_operands): New prototype.
* optabs.c (get_rtx_code): Make it global.
* optabs.h (get_rtx_code): New prototype.

From-SVN: r217640

target.def (gen_ccmp_first, [...]): Add two new hooks.

2014-11-17  Zhenqiang Chen  <zhenqiang.chen@linaor.org>

* target.def (gen_ccmp_first, gen_ccmp_first): Add two new hooks.
* doc/tm.texi.in (TARGET_GEN_CCMP_FIRST, TARGET_GEN_CCMP_NEXT): New.
* doc/tm.texi (TARGET_GEN_CCMP_FIRST, TARGET_GEN_CCMP_NEXT): New.

From-SVN: r217639

Always combine comparisons or conversions from booleans.

2014-11-16  Patrick Palka  <ppalka@gcc.gnu.org>

gcc/
PR middle-end/63790
* tree-ssa-forwprop.c (forward_propagate_into_comparison_1):
Always combine comparisons or conversions from booleans.

gcc/testsuite/
PR middle-end/63790
* gcc.dg/tree-ssa/pr21031.c: Drop XFAIL.
* gcc.dg/tree-ssa/forwprop-29.c: New test.

From-SVN: r217638

Daily bump.

From-SVN: r217637

* ipa-polymorphic-call.c
(ipa_polymorphic_call_context::speculation_consistent_p): Constify.
(ipa_polymorphic_call_context::meet_speculation_with): New function.
(ipa_polymorphic_call_context::combine_with): Handle types in construction
better.
(ipa_polymorphic_call_context::equal_to): Do not bother about useless
speculation.
(ipa_polymorphic_call_context::meet_with): New function.
* cgraph.h (class ipa_polymorphic_call_context): Add
meet_width, meet_speculation_with; constify speculation_consistent_p.
* ipa-cp.c (ipa_context_from_jfunc): Handle speculation; combine with incomming
context.
(propagate_context_accross_jump_function): Likewise; be more cureful.
about set_contains_variable.
(ipa_get_indirect_edge_target_1): Fix handling of dynamic type changes.
(find_more_scalar_values_for_callers_subset): Fix.
(find_more_contexts_for_caller_subset): Perform meet operation.

From-SVN: r217634

passes.c (execute_one_pass): Do not apply all transforms prior every simple IPA pass.

* passes.c (execute_one_pass): Do not apply all transforms prior
every simple IPA pass.
* cgraphunit.c: Do not include fibheap.h
(expand_thunk): Use get_untransformed_body.
(cgraph_node::expand): Likewise.
* tree-ssa-structalias.c (ipa_pta_execute): Skip inline clones.
* cgraph.c (release_function_body): Do not push cfun when CFG is not there.
(cgraph_node::get_untransformed_body): Break out from ...
(cgraph_node::get_body): ... here; add code to apply all transforms.
* cgraph.h (cgraph_node): Add get_untransformed_body.
* ipa-icf.c (sem_function::init): Use get_untransformed_body.
* cgraphclones.c (duplicate_thunk_for_node): Likewise.
* tree-inline.c (expand_call_inline): LIkewise.
* i386.c (ix86_reset_to_default_globals): Break out from ...
(ix86_set_current_function): ... here;
(ix86_reset_previous_fndecl): Use it.
(ix86_simd_clone_adjust): Use ix86_reset_previous_fndecl.

From-SVN: r217633

* gcc.dg/vect/pr63605.c (dg-final): Cleanup vect tree dump.

From-SVN: r217632

Update move-if-change from gnulib

2014-11-16  Jan-Benedict Glaw  <jbglaw@lug-owl.de>

* move-if-change: Sync from upstream gnulib.

From-SVN: r217631

tm.texi.in (TARGET_FLAGS_REGNUM): Move around.

* doc/tm.texi.in (TARGET_FLAGS_REGNUM): Move around.
* doc/tm.texi: Regenerate.

From-SVN: r217628

Update from upstream Automake files.

2014-11-16  Jan-Benedict Glaw  <jbglaw@lug-owl.de>

* compile: Sync with upstream Automake.
* depcomp: Ditto.
* install-sh: Ditto.
* missing: Ditto.
* mkinstalldirs: Ditto.
* ylwrap: Ditto.

From-SVN: r217627

sh.c: Do not include algorithm.

* config/sh/sh.c: Do not include algorithm.
(sh_emit_scc_to_t): Replace open-coded swap with std::swap
to swap values.
(sh_emit_compare_and_branch): Ditto.
(sh_emit_compare_and_set): Ditto.
* config/sh/sh.md (replacement peephole2): Ditto.
(cstore4_media): Ditto.
(*fmasf4): Ditto.

From-SVN: r217626

memset-4.c: New test.

2014-11-16  Andrew Pinski  <apinski@cavium.com>

* gcc.c-torture/execute/memset-4.c: New test.
* gcc.c-torture/execute/20110418-1.c: New test.
* gcc.c-torture/execute/20141022-1.c: New test.
* gcc.c-torture/execute/strcpy-2.c: New test.
* gcc.c-torture/execute/20140212-2.c: New test.
* gcc.c-torture/compile/20120913-1.c: New test.
* gcc.c-torture/compile/20121010-1.c: New test.
* gcc.c-torture/compile/20120917-1.c: New test.
* gcc.c-torture/compile/20140110-1.c: New test.
* gcc.c-torture/compile/20121220-1.c: New test.
* gcc.c-torture/compile/20120822-1.c: New test.
* gcc.c-torture/compile/20121027-1.c: New test.
* gcc.c-torture/compile/20120830-2.c: New test.

From-SVN: r217625

lra-remat.c (cand_transf_func): Process regno for rematerialization too.

2014-11-15  Vladimir Makarov  <vmakarov@redhat.com>

* lra-remat.c (cand_transf_func): Process regno for
rematerialization too.
* lra.c (lra): Switch on rematerialization pass.

From-SVN: r217624

PR 60324 VLA related fixes to random number generator.

2014-11-16  Janne Blomqvist  <jb@gcc.gnu.org>

PR libfortran/60324
* intrinsics/random.c (kiss_size): Rename to KISS_SIZE, make it a
macro instead of a variable.
(random_seed_i4): Make seed correct size, remove assert, KISS_SIZE
related changes.
(random_seed_i8): KISS_SIZE related changes.

From-SVN: r217623

Daily bump.

From-SVN: r217622

Update config.{sub,guess} from upstream repo

2014-11-16  Jan-Benedict Glaw  <jbglaw@lug-owl.de>

* config.sub: Update from upstream config repo.
* config.guess: Ditto.

From-SVN: r217619

parser.c (cp_parser_omp_declare_reduction_exprs): A block is not an expression.

* parser.c (cp_parser_omp_declare_reduction_exprs): A block is not
an expression.

From-SVN: r217618

constexpr.c (cxx_eval_builtin_function_call): Use fold_builtin_call_array.

* constexpr.c (cxx_eval_builtin_function_call): Use
fold_builtin_call_array.

From-SVN: r217617

* constexpr.c (cx_check_missing_mem_inits): Clarify error message.

From-SVN: r217616

guality.exp (check_guality): Fix `test_counts' restoration.

* g++.dg/guality/guality.exp (check_guality): Fix `test_counts'
restoration.

From-SVN: r217615

2014-11-15  Adhemerval Zanella  <azanella@linux.vnet.ibm.com>

* gcc.dg/atomic/c11-atomic-exec-5.c
(test_main_long_double_add_overflow): Only actually define if
LDBL_MANT_DIG != 106.
(test_main_complex_long_double_add_overflow): Likewise.
(test_main_long_double_sub_overflow): Likewise.
(test_main_complex_long_double_sub_overflow): Likewise.

(NOT_LDBL_EPSILON_2): Likewise.
(NOT_MINUS_LDBL_EPSILON_2): Likewise.

From-SVN: r217614

Don't run gcc.target/i386/sibcall-1.c on x32 targets

* gcc.target/i386/sibcall-1.c: Don't run on x32 targets,
instead of run on ia32 targets.

From-SVN: r217613

re PR middle-end/63884 (ICE: SIGSEGV in is_sec_implicit_index_fn with -fcilkplus and __builtin_sadd_overflow())

PR middle-end/63884
c-family/
* array-notation-common.c (is_sec_implicit_index_fn): Return false
for NULL fndecl.
(extract_array_notation_exprs): Return for NULL node.
testsuite/
* c-c++-common/cilk-plus/AN/pr63884.c: New test.

From-SVN: r217612

re PR target/60104 (load not folded into indirect branch on x86-64)

PR target/60104
* gcc.target/i386/sibcall-1.c: Don't run on pic targets.

From-SVN: r217611

lra.c (lra): Switch off rematerialization pass.

2014-11-15  Vladimir Makarov  <vmakarov@redhat.com>

* lra.c (lra): Switch off rematerialization pass.

From-SVN: r217609

xmmintrin.h (_mm_add_ps, [...]): Use vector extensions instead of builtins.

2014-11-15  Marc Glisse  <marc.glisse@inria.fr>

gcc/
* config/i386/xmmintrin.h (_mm_add_ps, _mm_sub_ps, _mm_mul_ps,
_mm_div_ps, _mm_store_ss, _mm_cvtss_f32): Use vector extensions
instead of builtins.
* config/i386/emmintrin.h (__v2du, __v4su, __v8hu, __v16qu): New
typedefs.
(_mm_sqrt_sd): Fix comment.
(_mm_add_epi8, _mm_add_epi16, _mm_add_epi32, _mm_add_epi64,
_mm_sub_epi8, _mm_sub_epi16, _mm_sub_epi32, _mm_sub_epi64,
_mm_mullo_epi16, _mm_cmpeq_epi8, _mm_cmpeq_epi16, _mm_cmpeq_epi32,
_mm_cmplt_epi8, _mm_cmplt_epi16, _mm_cmplt_epi32, _mm_cmpgt_epi8,
_mm_cmpgt_epi16, _mm_cmpgt_epi32, _mm_and_si128, _mm_or_si128,
_mm_xor_si128, _mm_store_sd, _mm_cvtsd_f64, _mm_storeh_pd,
_mm_cvtsi128_si64, _mm_cvtsi128_si64x, _mm_add_pd, _mm_sub_pd,
_mm_mul_pd, _mm_div_pd, _mm_storel_epi64, _mm_movepi64_pi64):
Use vector extensions instead of builtins.
* config/i386/smmintrin.h (_mm_cmpeq_epi64, _mm_cmpgt_epi64,
_mm_mullo_epi32): Likewise.
* config/i386/avxintrin.h (__v4du, __v8su, __v16hu, __v32qu):
New typedefs.
(_mm256_add_pd, _mm256_add_ps, _mm256_div_pd, _mm256_div_ps,
_mm256_mul_pd, _mm256_mul_ps, _mm256_sub_pd, _mm256_sub_ps):
Use vector extensions instead of builtins.
* config/i386/avx2intrin.h (_mm256_cmpeq_epi8, _mm256_cmpeq_epi16,
_mm256_cmpeq_epi32, _mm256_cmpeq_epi64, _mm256_cmpgt_epi8,
_mm256_cmpgt_epi16, _mm256_cmpgt_epi32, _mm256_cmpgt_epi64,
_mm256_and_si256, _mm256_or_si256, _mm256_xor_si256, _mm256_add_epi8,
_mm256_add_epi16, _mm256_add_epi32, _mm256_add_epi64,
_mm256_mullo_epi16, _mm256_mullo_epi32, _mm256_sub_epi8,
_mm256_sub_epi16, _mm256_sub_epi32, _mm256_sub_epi64): Likewise.
* config/i386/avx512fintrin.h (__v8du, __v16su, __v32hu, __v64qu):
New typedefs.
(_mm512_or_si512, _mm512_or_epi32, _mm512_or_epi64, _mm512_xor_si512,
_mm512_xor_epi32, _mm512_xor_epi64, _mm512_and_si512,
_mm512_and_epi32, _mm512_and_epi64, _mm512_mullo_epi32,
_mm512_add_epi64, _mm512_sub_epi64, _mm512_add_epi32,
_mm512_sub_epi32, _mm512_add_pd, _mm512_add_ps, _mm512_sub_pd,
_mm512_sub_ps, _mm512_mul_pd, _mm512_mul_ps, _mm512_div_pd,
_mm512_div_ps): Use vector extensions instead of builtins.
* config/i386/avx512bwintrin.h (_mm512_mullo_epi16, _mm512_add_epi8,
_mm512_sub_epi8, _mm512_sub_epi16, _mm512_add_epi16): Likewise.
* config/i386/avx512dqintrin.h (_mm512_mullo_epi64): Likewise.
* config/i386/avx512vldqintrin.h (_mm256_mullo_epi64, _mm_mullo_epi64):
Likewise.

gcc/testsuite/
* gcc.target/i386/intrinsics_opt-1.c: New testcase.
* gcc.target/i386/intrinsics_opt-2.c: Likewise.
* gcc.target/i386/intrinsics_opt-3.c: Likewise.
* gcc.target/i386/intrinsics_opt-4.c: Likewise.

From-SVN: r217608

lto-streamer-out.c (hash_tree): Use cl_optimization_hash.

* lto-streamer-out.c (hash_tree): Use cl_optimization_hash.
* lto-streamer.h (cl_optimization_stream_out, cl_optimization_stream_in): Declare.
* optc-save-gen.awk: Generate cl_optimization LTO streaming and hashing routines.
* opth-gen.awk: Add prototype of cl_optimization_hash.
* tree-streamer-in.c (unpack_ts_optimization): Remove.
(streamer_unpack_tree_bitfields): Use cl_optimization_stream_in.
* tree-streamer-out.c (pack_ts_optimization): Remove.
(streamer_pack_tree_bitfields): Use cl_optimization_stream_out.

From-SVN: r217607

* gcc.dg/tree-ssa/pr61144.c: Add dg-require-alias.

From-SVN: r217606

pubtypes-3.c: Include <string.h>.

* gcc.dg/pubtypes-3.c: Include <string.h>.
* gcc.dg/pubtypes-4.c: Likewise.

From-SVN: r217605

New unroll and jam option in Graphite.

From-SVN: r217604

* gcc.dg/darwin-cfstring-format-1.c: Adjust dg-error.

From-SVN: r217603

tree-cfg.c (replace_loop_annotate_in_block): New function extracted from...

* tree-cfg.c (replace_loop_annotate_in_block): New function extracted
from...
(replace_loop_annotate): ...here.  Call it on the header and on the
latch block, if any.  Restore proper behavior of final cleanup.

From-SVN: r217602

tree-ssa-loop-ivcanon.c (try_unroll_loop_completely): Add log message for max-completely-peeled-insns limit.

* tree-ssa-loop-ivcanon.c (try_unroll_loop_completely): Add log message
for max-completely-peeled-insns limit.

From-SVN: r217601

error.c (gfc_fatal_error_1): Renamed from gfc_fatal_error.

2014-11-15  Tobias Burnus  <burnus@net-b.de>

gcc/fortran/
        * error.c (gfc_fatal_error_1): Renamed from gfc_fatal_error.
        (gfc_fatal_error): Add; uses common diagnostics.
        * array.c (gfc_match_array_ref, gfc_match_array_spec): Use
        %< %>.
        * check.c (check_co_collective, gfc_check_lcobound,
        gfc_check_image_index, gfc_check_num_images,
        gfc_check_this_image, gfc_check_ucobound): Ditto.
        * cpp.c (gfc_cpp_post_options): Ditto.
        (gfc_cpp_init_0, gfc_cpp_done): Change %s to %qs.
        * gfc-diagnostic.def (DK_FATAL): Capitalize first letter.
        * gfortran.h (gfc_fatal_error_1): Add.
        * match.c (gfc_match_name, gfc_match_critical,
        lock_unlock_statement, sync_statement): Add %< %>.
        * module.c (bad_module, gfc_dump_module, gfc_use_module): Change
        %s to %qs.
        * options.c (gfc_handle_module_path_options,
        * gfc_handle_fpe_option,
        gfc_handle_coarray_option, gfc_handle_runtime_check_option,
        gfc_handle_option): Add %< %>.
        * simplify.c (gfc_simplify_num_images): Ditto.
        * trans-stmt.c (gfc_trans_sync): Use gfc_fatal_error_1.
        * trans-array.c (gfc_conv_array_initializer): Ditto.
        * trans-types.c (gfc_init_kinds): Use gfc_fatal_error instead
        of fatal_error; add %< %> quotations.

gcc/testsuite/
        * gfortran.dg/binding_label_tests_4.f03: Add dg-excess-errors.
        * gfortran.dg/coarray_9.f90: Ditto.
        * gfortran.dg/empty_label.f: Ditto.
        * gfortran.dg/empty_label.f90: Ditto.

From-SVN: r217600

Daily bump.

From-SVN: r217599

ipa-prop.h (ipa_known_type_data): Remove.

* ipa-prop.h (ipa_known_type_data): Remove.
(ipa_binfo_from_known_type_jfunc): Remove.

From-SVN: r217596

[AARCH64] Add scheduler for ThunderX

This adds the schedule model for ThunderX. There are a few TODOs in that
not all of the SIMD is model currently.  Also the idea of a simple
shift/extend is not modeled and all cases where there is a shift/extend
is considered as non simple and take up two cycles rather than correct
value of one cycle.  Also the 32bit divide and the 64bit divide
have different cycle counts but there is no way to model that currently.
Also multiply high takes one cycle more than the normal multiply but
there is no way to model that currently either.

Build and tested for aarch64-elf with no regressions.

ChangeLog:
* config/aarch64/aarch64-cores.def (thunderx): Change the scheduler
over to thunderx.
* config/aarch64/aarch64.md: Include thunderx.md.
(generic_sched): Set to no for thunderx.
* config/aarch64/thunderx.md: New file.

From-SVN: r217593

predicates.md (easy_fp_constant): Delete redunant tests for 0.0.

[gcc]
2014-11-14  Michael Meissner  <meissner@linux.vnet.ibm.com>

* config/rs6000/predicates.md (easy_fp_constant): Delete redunant
tests for 0.0.

* config/rs6000/vector.md (VEC_R): Move secondary reload support
insns to rs6000.md from vector.md.
(reload_<VEC_R:mode>_<P:mptrsize>_store): Likewise.
(reload_<VEC_R:mode>_<P:mptrsize>_load): Likewise.
(vec_reload_and_plus_<mptrsize>): Likewise.

* config/rs6000/rs6000.md (Fa): New mode attribute to give
constraint for the Altivec registers for a type.
(RELOAD): New mode iterator for all of the types that have
secondary reload address support to load up a base register.
(extendsfdf2_fpr): Use correct constraint.
(copysign<mode>3_fcpsgn): For SFmode, use correct xscpsgndp
instruction.
(floatsi<mode>2_lfiwax): Add support for -mupper-regs-{sf,df}.
Generate the non-VSX instruction if all registers were FPRs.  Do
not use the patterns in vsx.md for scalar operations.
(floatsi<mode>2_lfiwax_mem): Likewise.
(floatunssi<mode>2_lfiwzx): Likewise.
(floatunssi<mode>2_lfiwzx_mem): Likewise.
(fix_trunc<mode>di2_fctidz): Likewise.
(fixuns_trunc<mode>di2_fctiduz): Likewise.
(fctiwz_<mode>): Likewise.
(fctiwuz_<mode>): Likewise.
(friz): Likewise.
(floatdidf2_fpr): Likewise.
(floatdidf2_mem): Likewise.
(floatunsdidf2): Likewise.
(floatunsdidf2_fcfidu): Likewise.
(floatunsdidf2_mem): Likewise.
(floatdisf2_fcfids): Likewise.
(floatdisf2_mem): Likewise.
(floatdisf2_internal1): Add explicit test for not FCFIDS to make
it more obvious that the code is for pre-ISA 2.06 machines.
(floatdisf2_internal2): Likewise.
(floatunsdisf2_fcfidus): Add support for -mupper-regs-{sf,df}.
Generate the non-VSX instruction if all registers were FPRs.  Do
not use the patterns in vsx.md for scalar operations.
(floatunsdisf2_mem): Likewise.
(reload_<RELOAD:mode>_<P:mptrsize>_store): Move the reload
handlers here from vector.md, and expand the types we generate
reload handlers for.
(reload_<RELOAD:mode>_<P:mptrsize>_load): Likewise.
(vec_reload_and_plus_<mptrsize>): Likewise.

* config/rs6000/vsx.md (vsx_float<VSi><mode>2): Only provide the
vector forms of the instructions.  Move VSX scalar forms to
rs6000.md, and add support for -mupper-regs-sf.
(vsx_floatuns<VSi><mode>2): Likewise.
(vsx_fix_trunc<mode><VSi>2): Likewise.
(vsx_fixuns_trunc<mode><VSi>2): Likewise.
(vsx_float_fix_<mode>2): Delete DF version, rename to
vsx_float_fix_v2df2.
(vsx_float_fix_v2df2): Likewise.

[gcc/testsuite]
2014-11-14  Michael Meissner  <meissner@linux.vnet.ibm.com>

* gcc.target/powerpc/ppc-fpconv-1.c: Adjust for -mupper-regs-df
changes.
* gcc.target/powerpc/ppc-fpconv-2.c: Likewise.
* gcc.target/powerpc/ppc-fpconv-3.c: Likewise.
* gcc.target/powerpc/ppc-fpconv-4.c: Likewise.
* gcc.target/powerpc/ppc-fpconv-5.c: Likewise.
* gcc.target/powerpc/ppc-fpconv-6.c: Likewise.
* gcc.target/powerpc/ppc-fpconv-7.c: Likewise.
* gcc.target/powerpc/ppc-fpconv-8.c: Likewise.
* gcc.target/powerpc/ppc-fpconv-9.c: Likewise.
* gcc.target/powerpc/ppc-fpconv-10.c: Likewise.
* gcc.target/powerpc/ppc-round.c: Likewise.

From-SVN: r217590

ipa-prop.h (jump_func_type): Removed value IPA_JF_KNOWN_TYPE.

2014-11-14  Martin Jambor  <mjambor@suse.cz>

* ipa-prop.h (jump_func_type): Removed value IPA_JF_KNOWN_TYPE.
(ipa_pass_through_data): Removed field type_preserved.
(ipa_ancestor_jf_data): removed fields type and type_preserved.
(ipa_jump_func): Removed field known_type.
(ipa_get_jf_known_type_offset): Removed.
(ipa_get_jf_known_type_base_type): Likewise.
(ipa_get_jf_known_type_component_type): Likewise.
(ipa_get_jf_ancestor_type): Likewise.
* ipa-cp.c (print_ipcp_constant_value): Removed BINFO handling.
(ipa_get_jf_pass_through_result): Likewise.
(ipa_get_jf_ancestor_result): Always build ptr_node_type accesses.
(values_equal_for_ipcp_p): Removed BINFO handling.
(ipa_get_indirect_edge_target_1): Updated comment.
* ipa-prop.c (ipa_print_node_jump_functions_for_edge): Removed handling
of IPA_JF_KNOWN_TYPE jump functions.  Do not print removed fields.
(ipa_set_jf_known_type): Removed.
(ipa_set_jf_simple_pass_through): Do not set removed fields.  Update
all callers.
(ipa_set_jf_arith_pass_through): Likewise.
(ipa_set_ancestor_jf): Likewise.
(ipa_binfo_from_known_type_jfunc): Removed.
(prop_type_change_info): Removed fields known_current_type and
multiple_types_encountered.
(extr_type_from_vtbl_ptr_store): Removed.
(check_stmt_for_type_change): Do not attempt to identify changed type.
(detect_type_change_from_memory_writes): Do not set the removed fields,
always set jfunc to unknown.
(compute_complex_assign_jump_func): Do not detect dynamic type change.
(compute_complex_ancestor_jump_func): Likewise.
(compute_known_type_jump_func): Removed.
(ipa_compute_jump_functions_for_edge): Do not detect dynamic type
change.  Do not comute known type jump functions.
(combine_known_type_and_ancestor_jfs): Removed.
(update_jump_functions_after_inlining): Removed handling of
IPA_JF_KNOWN_TYPE jump functions.  Do not set removed fields.
(ipa_write_jump_function): Do not stream removed fields or known type
jump functions.
(ipa_read_jump_function): Likewise.

From-SVN: r217589

lra-int.h (lra_create_live_ranges): Add parameter.

2014-11-14  Vladimir Makarov  <vmakarov@redhat.com>

* lra-int.h (lra_create_live_ranges): Add parameter.
* lra-lives.c (temp_bitmap): Move higher.
(initiate_live_solver): Move temp_bitmap initialization into
lra_live_ranges_init.
(finish_live_solver): Move temp_bitmap clearing into
live_ranges_finish.
(process_bb_lives): Add parameter.  Use it to control live info
update and dead insn elimination.  Pass it to mark_regno_live and
mark_regno_dead.
(lra_create_live_ranges): Add parameter.  Pass it to
process_bb_lives.
(lra_live_ranges_init, lra_live_ranges_finish): See changes in
initiate_live_solver and finish_live_solver.
* lra-remat.c (do_remat): Process insn non-operand hard regs too.
Use temp_bitmap to update avail_cands.
* lra.c (lra): Pass new parameter to lra_create_live_ranges.  Move
check with lra_need_for_spill_p after live range pass.  Switch on
rematerialization pass.

From-SVN: r217588

ipa-prop.h (ipa_get_jf_pass_through_type_preserved): use agg_preserved flag instead.

2014-11-14  Martin Jambor  <mjambor@suse.cz>

* ipa-prop.h (ipa_get_jf_pass_through_type_preserved): use
agg_preserved flag instead.
(ipa_get_jf_ancestor_type_preserved): Likewise.
(ipa_node_params): Rename known_vals to known_csts, update all users.
New field known_contexts.
(ipa_get_indirect_edge_target): Update prototype.
(ipcp_poly_ctx_values_pool): Declare.
(ipa_context_from_jfunc): Likewise.
* ipa-inline.h (estimate_ipcp_clone_size_and_time): Updated prototype.
* cgraph.h (ipa_polymorphic_call_context): New method equal_to.  New
parameter newline of method dump.
* ipa-cp.c (ctxlat): New field.
(ipcp_values_pool): Renamed to ipcp_cst_values_pool, updated all users.
(ipcp_poly_ctx_values_pool):New variable.
(ipa_get_poly_ctx_lat): New function.
(print_ipcp_constant_value): New overloaded function for contexts.
(print_all_lattices): Also print contexts.
(ipa_topo_info): New field contexts;
(set_all_contains_variable): Also set the flag in the context lattice.
(initialize_node_lattices): Likewise for flag bottom.
(ipa_get_jf_ancestor_result): Removed BINFO handling.
(ipa_value_from_jfunc): Likewise.
(ipa_context_from_jfunc): New function.
(values_equal_for_ipcp_p): New overloaded function for contexts.
(allocate_and_init_ipcp_value): Construct the value.
(allocate_and_init_ipcp_value): New overloaded function for contexts.
(propagate_scalar_accross_jump_function): Removed handling of
KNOWN_TYPE jump functions.
(propagate_context_accross_jump_function): New function.
(propagate_constants_accross_call): Also propagate contexts.
(ipa_get_indirect_edge_target_1): Work on contexts rather than BINFOs.
(ipa_get_indirect_edge_target): Likewise.
(devirtualization_time_bonus): Likewise.
(gather_context_independent_values): Create and populate known_contexts
vector rather than known_binfos.
(perform_estimation_of_a_value): Work on contexts rather than BINFOs.
(estimate_local_effects): Likewise.
(add_all_node_vals_to_toposort): Also add contexts to teir topological
sort.
(ipcp_propagate_stage): Also propagate effects of contexts.
(ipcp_discover_new_direct_edges): Receive and pass known_contexts to
ipa_get_indirect_edge_target_1.
(cgraph_edge_brings_value_p): New overloaded function for contexts.
(create_specialized_node): Work on contexts rather than BINFOs.
(find_more_contexts_for_caller_subset): New function.
(known_contexts_useful_p): New function.
(copy_useful_known_contexts): Likewise.
(modify_known_vectors_with_val): Likewise.
(ipcp_val_in_agg_replacements_p): Renamed to
ipcp_val_agg_replacement_ok_p, return true for all offset indicating
non-aggregate.
(ipcp_val_agg_replacement_ok_p): New overloaded function for contexts.
(decide_about_value): Work on contexts rather than BINFOs.
(decide_whether_version_node): Likewise.
(ipcp_driver): Initialize the new alloc pool.
* ipa-prop.c (ipa_print_node_jump_functions_for_edge): Prettify
printing of edge contexts.
(ipa_set_ancestor_jf): Replace assert with conditional setting of
type_preserved to false.
(update_jump_functions_after_inlining): Use access function instead of
reading agg_preserved directly.  Store combined context in the ancestor
case.
(try_make_edge_direct_virtual_call): Work on contexts rather than
BINFOs.
(update_indirect_edges_after_inlining): Get context from
ipa_context_from_jfunc.
(ipa_free_node_params_substructures): Free also known_contexts.
(ipa_free_all_structures_after_ipa_cp): Free the new alloc pool.
(ipa_free_all_structures_after_iinln): Likewise.
* ipa-inline-analysis.c (evaluate_properties_for_edge): Work on
contexts rather than BINFOs.
(estimate_edge_devirt_benefit): Likewise.
(estimate_edge_size_and_time): Likewise.
(estimate_calls_size_and_time): Likewise.
(estimate_node_size_and_time): Likewise.
(estimate_ipcp_clone_size_and_time): Likewise.
(do_estimate_edge_time): Likewise.
(do_estimate_edge_size): Likewise.
(do_estimate_edge_hints): Likewise.
* ipa-polymorphic-call.c (ipa_polymorphic_call_context::dump): New
parameter newline, ouput newline only when it is set.
(ipa_polymorphic_call_context::equal_to): New method.

testsuite/
* g++.dg/ipa/devirt-11.C: Dont't run ipa-cp, remove times constraint
from the dump scan.
* g++.dg/ipa/devirt-21.C: Xfail.
* g++.dg/ipa/devirt-24.C: Likewise.
* g++.dg/ipa/devirt-10.C: Removed times constraint from the dump scan.
* g++.dg/ipa/devirt-41.C: Updated the dump scan.
* g++.dg/ipa/devirt-44.C: Likewise.
* g++.dg/ipa/devirt-43.C: Xfail.

From-SVN: r217587

ipa-cp.c (ipcp_value_source): Converted to a template class.

2014-11-14  Martin Jambor  <mjambor@suse.cz>

* ipa-cp.c (ipcp_value_source): Converted to a template class.  All
users converted to the same specialization as the using class/function
or specialization on tree.
(ipcp_value): Likewise.
(ipcp_lattice): Likewise.
(ipcp_agg_lattice): Now derived from tree specialization of
ipcp_lattice.
(values_topo): Moved to new class value_topo_info.
(ipa_lat_is_single_const): Turned into ipcp_lattice::is_single_const.
Updated all callers.
(print_lattice): Turned into ipcp_lattice::print.  Updated all
callers.
(value_topo_info): New class template.
(ipa_topo_info): New field constants.  New constructor.
(build_toporder_info): Do not clear stack_top, only checkign assert
it.
(set_lattice_to_bottom): Turned into ipcp_lattice::set_to_bottom.
Updated all callers.
(set_lattice_contains_variable): Turned into
ipcp_lattice::set_contains_variable.  Updated all callers.
(add_value_source): Turned into ipcp_value::add_source.  Updated all
callers.
(allocate_and_init_ipcp_value): New function.
(add_value_to_lattice): Turned into ipcp_lattice::add_value.  Last
parameter got default a value.  Updated all callers.
(add_scalar_value_to_lattice): Removed, users converted to using
ipcp_lattice::add_value with default value of the last parameter.
(add_val_to_toposort): Turned to value_topo_info::add_val.  Updated
all callers.
(propagate_effects): Made method of value_topo_info.
(cgraph_edge_brings_value_p): Now a template function.
(get_info_about_necessary_edges): Likewise.
(gather_edges_for_value): Likewise.
(perhaps_add_new_callers): Likewise.
(decide_about_value): Likewise.
* ipa-prop.h (ipcp_lattice): Remove fowrward declaration.

From-SVN: r217586

* doc/install.texi (--with-diagnostics-color=): Document.

From-SVN: r217585

tree-ssa.dce.c (eliminate_unnecessary_stmts): Eliminate IFN_GOMP_SIMD_LANE without lhs as useless.

* tree-ssa.dce.c (eliminate_unnecessary_stmts): Eliminate
IFN_GOMP_SIMD_LANE without lhs as useless.

From-SVN: r217584

runtime: set library name based on compiler name

Different compilers may have different release cadences or ABI
incompatibilities, so it does not make sense to use the same library
name for runtime libraries intended for different compilers.

This CL causes a libgo built by llgo to receive the name libgo-llgo.
Likewise, libgobegin is named libgobegin-llgo.

From-SVN: r217583

ipa-pure-const.c (struct funct_state_d): Add can_free field.

* ipa-pure-const.c (struct funct_state_d): Add can_free field.
(varying_state): Add true for can_free.
(check_call): For builtin or internal !nonfreeing_call_p set
local->can_free.
(check_stmt): For asm volatile and asm with "memory" set
local->can_free.
(analyze_function): Clear local->can_free initially, continue
calling check_stmt until all flags are computed, dump can_free
flag.
(pure_const_write_summary): Write can_free flag.
(pure_const_read_summary): Read it back.
(propagate_pure_const): Propagate also can_free flag, set
w->nonfreeing_fn if it is false after propagation.
* cgraph.h (cgraph_node): Add nonfreeing_fn member.
* gimple.c: Include ipa-ref.h, lto-streamer.h and cgraph.h.
(nonfreeing_call_p): Return cgraph nonfreeing_fn flag if set.
Also return true for IFN_ABNORMAL_DISPATCHER.
* cgraph.c (cgraph_node::dump): Dump nonfreeing_fn flag.
* lto-cgraph.c (lto_output_node): Write nonfreeing_fn flag.
(input_overwrite_node): Read it back.

From-SVN: r217582

sanopt.c: Include tree-ssa-operands.h.

* sanopt.c: Include tree-ssa-operands.h.
(struct sanopt_info): Add has_freeing_call_p,
has_freeing_call_computed_p, imm_dom_path_with_freeing_call_p,
imm_dom_path_with_freeing_call_computed_p, freeing_call_events,
being_visited_p fields.
(struct sanopt_ctx): Add asan_check_map field.
(imm_dom_path_with_freeing_call, maybe_optimize_ubsan_null_ifn,
maybe_optimize_asan_check_ifn): New functions.
(sanopt_optimize_walker): Use them, optimize even ASAN_CHECK
internal calls.
(pass_sanopt::execute): Call sanopt_optimize even for
-fsanitize=address.
* gimple.c (nonfreeing_call_p): Return true for non-ECF_LEAF
internal calls.

Co-Authored-By: Marek Polacek <polacek@redhat.com>
From-SVN: r217581

[Vectorizer] Tidy up vect_create_epilog / use_scalar_result

* tree-vect-loop.c (vect_create_epilog_for_reduction): Move code for
'if (extract_scalar_result)' to the only place that it is true.

From-SVN: r217580

e.54.2.c (main): Use N / 8 instead of 32 as block_size.

* libgomp.c/examples-4/e.54.2.c (main): Use N / 8 instead
of 32 as block_size.
* libgomp.fortran/examples-4/e.54.2.f90 (e_54_1): Use n / 8
instead of 32 as block_size.

From-SVN: r217579

* dwarf2.def (DW_AT_APPLE_optimized, DW_AT_APPLE_flags)
(DW_AT_APPLE_isa, DW_AT_APPLE_block)
(DW_AT_APPLE_major_runtime_vers, DW_AT_APPLE_runtime_class)
(DW_AT_APPLE_omit_frame_ptr, DW_AT_APPLE_property_name)
(DW_AT_APPLE_property_getter, DW_AT_APPLE_property_setter)
(DW_AT_APPLE_property_attribute, DW_AT_APPLE_objc_complete_type)
(DW_AT_APPLE_property): New macros.

From-SVN: r217578

semantics.c (end_maybe_infinite_loop): Use fold_non_dependent_expr.

2014-11-14  Paolo Carlini  <paolo.carlini@oracle.com>

* semantics.c (end_maybe_infinite_loop): Use fold_non_dependent_expr.
* parser.c (cp_parser_omp_clause_collapse): Likewise.
(cp_parser_enumerator_definition): Don't call
instantiate_non_dependent_expr...
* decl.c (build_enumerator): ... call fold_non_dependent_expr here.
* typeck2.c (massage_init_elt): Use fold_non_dependent_expr.
* constexpr.c (maybe_constant_value): Allow VIEW_CONVERT_EXPR in
the final gcc_assert.

* constexpr.c (fold_non_dependent_expr): Add.
* cp-tree.h (fold_non_dependent_expr): Declare it.
* call.c (null_ptr_cst_p): Use it.
* pt.c (tsubst_copy_and_build, build_non_dependent_expr): Likewise.
* semantics.c (begin_maybe_infinite_loop): Likewise.
* typeck.c (cp_build_binary_op): Likewise.
* typeck2.c (check_narrowing): Likewise.

* pt.c (fold_non_dependent_expr): Rename to
instantiate_non_dependent_expr.
(fold_non_dependent_expr_sfinae): Rename to
instantiate_non_dependent_expr_sfinae.
(convert_nontype_argument, build_non_dependent_expr): Adjust.
* decl.c (compute_array_index_type): Likewise.
* parser.c (cp_parser_parenthesized_expression_list,
cp_parser_enumerator_definition, cp_parser_omp_clause_collapse):
Likewise.
* semantics.c (end_maybe_infinite_loop, finish_static_assert):
Likewise.
* typeck.c (cxx_alignas_expr): Likewise.
* typeck2.c (store_init_value, massage_init_elt): Likewise.
* call.c: Adjust comments.
* class.c: Likewise.
* constexpr.c: Likewise.
* decl2.c: Likewise.
* tree.c: Likewise.

From-SVN: r217577

mt_allocator.cc: Fix assumption that sizeof(void *) is equal to sizeof(size_t).

2014-11-14  Joel Sherrill <joel.sherrill@oarcorp.com>

* src/c++98/mt_allocator.cc: Fix assumption that sizeof(void *) is
equal to sizeof(size_t). The m32c breaks this assumption.

From-SVN: r217576

Don't enable IFUNC by default for Android and uclibc

* config.gcc (default_gnu_indirect_function): Set to yes
for i[34567]86-*-linux* and x86_64-*-linux* if not targeting
Android nor uclibc.

From-SVN: r217575