re PR tree-optimization/69595 (Bogus -Warray-bound warning due to missed optimization)

2016-02-02  Richard Biener  <rguenther@suse.de>

PR tree-optimization/69595
* match.pd: Add range test simplifications to true/false.

* gcc.dg/Warray-bounds-17.c: New testcase.

From-SVN: r233076

Merge BUILT_IN_GOACC_HOST_DATA into BUILT_IN_GOACC_DATA_START

gcc/
* omp-builtins.def (BUILT_IN_GOACC_HOST_DATA): Remove.
* omp-low.c (expand_omp_target): Use BUILT_IN_GOACC_DATA_START
instead.
libgomp/
* libgomp.map (GOACC_2.0): Remove GOACC_host_data.
* oacc-parallel.c (GOACC_host_data): Remove function definition.

From-SVN: r233074

libgomp: Skip hsa offloading for OpenACC test cases

libgomp/
* testsuite/lib/libgomp.exp: Skip hsa offloading for OpenACC test
cases.

From-SVN: r233073

libgomp: Use HSA_RUNTIME_LIB, HSA_KMT_LIB in the testsuite

libgomp/
* plugin/configfrag.ac (HSA_KMT_LIB, HSA_KMT_LDFLAGS): New
variables.
* testsuite/libgomp-test-support.exp.in (hsa_runtime_lib)
(hsa_kmt_lib): Set variables.
* testsuite/lib/libgomp.exp (libgomp_init): Use them to amend
always_ld_library_path.
* Makefile.in: Regenerate.
* configure: Likewise.
* testsuite/Makefile.in: Likewise.

From-SVN: r233072

libgomp: For hsa offloading, compilation is all handled by the target compiler

libgomp/
* plugin/configfrag.ac (offload_additional_options)
(offload_additional_lib_paths): Don't amend for hsa offloading.
* configure: Regenerate.

From-SVN: r233071

libgomp: Don't configure for offloading target if we don't build the corresponding plugin

libgomp/
* plugin/configfrag.ac: Don't configure for offloading target if
we don't build the corresponding plugin.
* configure: Regenerate.

From-SVN: r233070

re PR tree-optimization/69606 (wrong code at -Os and above on x86_64-linux-gnu)

2016-02-02  Richard Biener  <rguenther@suse.de>

PR tree-optimization/69606
* tree-ssa-math-opts.c (bswap_replace): Clear flow sensitive
info on the result before moving a stmt.

* gcc.dg/torture/pr69606.c: New testcase.

From-SVN: r233069

re PR middle-end/68542 (10% 481.wrf performance regression)

gcc/

2016-02-02  Yuri Rumyantsev  <ysrumyan@gmail.com>

PR middle-end/68542
* config/i386/i386.c (ix86_expand_branch): Add support for conditional
branch with vector comparison.
* config/i386/sse.md (VI48_AVX): New mode iterator.
(define_expand "cbranch<mode>4): Add support for conditional branch
with vector comparison.
* tree-vect-loop.c (optimize_mask_stores): New function.
* tree-vect-stmts.c (vectorizable_mask_load_store): Initialize
has_mask_store field of vect_info.
* tree-vectorizer.c (vectorize_loops): Invoke optimaze_mask_stores for
vectorized loops having masked stores after vec_info destroy.
* tree-vectorizer.h (loop_vec_info): Add new has_mask_store field and
correspondent macros.
(optimize_mask_stores): Add prototype.

gcc/testsuite

2016-02-02  Yuri Rumyantsev  <ysrumyan@gmail.com>

PR middle-end/68542
* gcc.dg/vect/vect-mask-store-move-1.c: New test.
* gcc.target/i386/avx2-vect-mask-store-move1.c: New test.

From-SVN: r233068

[RS6000] lqarx and stqcx. registers

lqarx RT and stqcx. RS are valid only with even numbered gprs.  The
predicate to enforce this happens to allow a loophole, closed by this
patch.

PR target/69548
gcc/
* config/rs6000/predicates.md (quad_int_reg_operand): Don't
allow subregs.
gcc/testsuite/
* gcc.target/powerpc/pr69548.c: New test.

From-SVN: r233065

Daily bump.

From-SVN: r233064

[RS6000] ABI_V4 init of toc section

Since 4c4a180d lto has turned off flag_pic when linking a fixed
position executable.  So flag_pic is zero in rs6000_file_start.
However, when we get to actually emitting code, flag_pic may be on
again.  This results in undefined references to ".LCTOC1".

PR target/68662
* config/rs6000/rs6000.c (need_toc_init): New var, set it
whenever toc_label_name used.
(rs6000_file_start): Don't set up toc section here,
(rs6000_output_function_epilogue): do so here instead,
(rs6000_xcoff_file_start): and here.
* config/rs6000/rs6000.md (load_toc_aix_si): Set need_toc_init.
(load_toc_aix_di): Likewise.

From-SVN: r233061

re PR rtl-optimization/69592 (Compile-time and memory-use hog in combine)

PR rtl-optimization/69592
* rtlanal.c (nonzero_bits_binary_arith_p): New inline function.
(cached_nonzero_bits): Use it instead of ARITHMETIC_P.
(num_sign_bit_copies_binary_arith_p): New inline function.
(cached_num_sign_bit_copies): Use it instead of ARITHMETIC_P.

* gcc.dg/pr69592.c: New test.

From-SVN: r233059

re PR preprocessor/69543 (_Pragma does not apply within macro)

PR preprocessor/69543
PR c/69558
* c-pragma.c (handle_pragma_diagnostic): Pass input_location
instead of loc to control_warning_option.

* gcc.dg/pr69543.c: New test.
* gcc.dg/pr69558.c: New test.

From-SVN: r233058

* Add PR number.

From-SVN: r233057

anon5.C (dg-opetions): Use -gdwarf-2 instead of -g.

* g++.dg/other/anon5.C (dg-opetions): Use -gdwarf-2 instead of -g.

From-SVN: r233056

Fix PR # in last change

From-SVN: r233054

re PR testsuite/68580 (FAIL: c-c++-common/tsan/pr65400-1.c -O0 execution test)

PR tree-optimization/68580
* params.def (FSM_MAXIMUM_PHI_ARGUMENTS): New param.
* tree-ssa-threadbackward.c
(fsm_find_control_statement_thread_paths): Do not try to walk
through large PHI nodes.

From-SVN: r233053

ifcvt.c (bb_ok_for_noce_convert_multiple_sets): Return false when count is incremented above limit...

* ifcvt.c (bb_ok_for_noce_convert_multiple_sets): Return false
when count is incremented above limit, don't analyze further
insns afterwards.

From-SVN: r233050

Remove duplicate copy of the test body inadvertently inserted by patch.

From-SVN: r233045

omp-low.c (oacc_parse_default_dims): Avoid -Wsign-compare warning...

* omp-low.c (oacc_parse_default_dims): Avoid
-Wsign-compare warning, make sure value fits into int
rather than just unsigned int.

From-SVN: r233044

re PR tree-optimization/67921 ("internal compiler error: in build_polynomial_chrec, at tree-chrec.h:147" when using -fsanitize=undefined)

PR tree-optimization/67921
* fold-const.c (split_tree): New parameters.  Convert pointer
type variable part to proper type before negating.
(fold_binary_loc): Pass new arguments to split_tree.

gcc/testsuite/ChangeLog
PR tree-optimization/67921
* c-c++-common/ubsan/pr67921.c: New test.

From-SVN: r233042

nvptx.c (PTX_GANG_DEFAULT): New.

gcc/
* config/nvptx/nvptx.c (PTX_GANG_DEFAULT): New.
(nvptx_goacc_validate_dims): Extend to handle global defaults.
* target.def (OACC_VALIDATE_DIMS): Extend documentation.
* doc/tm.texti: Rebuilt.
* doc/invoke.texi (fopenacc-dim): Document.
* lto-wrapper.c (merge_and_complain): Add OPT_fopenacc_dim_ case.
(append_compiler_options): Likewise.
* omp-low.c (oacc_default_dims, oacc_min_dims): New.
(oacc_parse_default_dims): New.
(oacc_validate_dims): Add USED arg.  Select non-unity default when
possible.
(oacc_loop_fixed_partitions): Return mask of used partitions.
(oacc_loop_auto_partitions): Emit dump info.
(oacc_loop_partition): Return mask of used partitions.
(execute_oacc_device_lower): Parse default dimension arg.  Adjust
loop partitioning and validation calls.

gcc/c-family/
* c.opt (fopenacc-dim=): New option.

gcc/fortran/
* lang.opt (fopenacc-dim=): New option.

libgomp/
* testsuite/libgomp.oacc-c-c++-common/loop-dim-default.c: New.
* testsuite/libgomp.oacc-fortran/routine-7.f90: Serialize loop.

From-SVN: r233041

re PR tree-optimization/69556 (forwprop4/match.pd undoing work from recip)

2016-02-01  Richard Biener  <rguenther@suse.de>

PR middle-end/69556
* match.pd: Guard (C1/X)*C2 -> (C1*C2)/X with single_use.

* gcc.dg/tree-ssa/recip-8.c: New testcase.

From-SVN: r233040

re PR tree-optimization/69574 (gcc ICE at -O2 and -O3 on x86_64-linux-gnu in hide_evolution_in_other_loops_than_loop)

2016-02-01  Richard Biener  <rguenther@suse.de>

PR tree-optimization/69574
* tree-chrec.c (hide_evolution_in_other_loops_than_loop): Instead
of asserting return chrec_dont_know.

* gcc.dg/torture/pr69574.c: New testcase.

From-SVN: r233039

Missing copyright for mem-stats header files.

* mem-stats-traits.h: Add copyright header.
* mem-stats.h: Likewise.

From-SVN: r233037

re PR tree-optimization/69579 (gcc ICE at -O3 and __sigsetjmp with “tree check: expected ssa_name, have integer_cst in compute_optimized_partition_bases”)

2016-02-01  Richard Biener  <rguenther@suse.de>

PR tree-optimization/69579
* tree-ssa-loop-ivcanon.c (propagate_constants_for_unrolling):
Do not propagate through abnormal PHI results.

* gcc.dg/setjmp-6.c: New testcase.

From-SVN: r233036

Don't define guard macros when doing #include_next in math.h and stdlib.h

2016-02-01  Bernd Edlinger  <bernd.edlinger@hotmail.de>

PR libstdc++/69581
* include/c_compatibility/math.h: Move header guards.
* include/c_compatibility/stdlib.h: Likewise.

From-SVN: r233035

* postreload.c (reload_cse_simplify): Remove dead code.

From-SVN: r233034

re PR rtl-optimization/69570 (if-conversion bug on i?86)

PR rtl-optimization/69570
* ifcvt.c (bb_ok_for_noce_convert_multiple_sets): Return true only
if there is more than one set, not if there is a single set.

* g++.dg/opt/pr69570.C: New test.

From-SVN: r233033

re PR rtl-optimization/69535 (wrong code with -O -fno-tree-bit-ccp -fno-tree-reassoc due to use of uninitialised value)

PR rtl-opt/69535

  * combine.c (make_compound_operation): When looking through a
  subreg, make sure to re-extend to the width of the outer mode.

From-SVN: r233032

re PR testsuite/69584 (FAIL: gcc.dg/pr67964.c (test for excess errors))

PR testsuite/69584
* gcc.dg/tree-ssa/sra-17.c: Fix regexps to work on hpux.
* gcc.dg/tree-ssa/sra-18.c: Likewise.

From-SVN: r233030

re PR target/68741 (FAIL: tr1/8_c_compatibility/cstdio/functions.cc (test for excess errors))

PR target/68741
* inclhack.def (hpux_vsscanf): New fix.
* fixincl.x: Regenerated.
* tests/base/stdio.h [HPUX_VSSCANF_CHECK]: New test.

From-SVN: r233029

re PR debug/68244 (FAIL: g++.dg/parse/parens3.C  -std=gnu++98 (internal compiler error))

PR debug/68244
* g++.dg/parse/parens3.C: Use register "4" on hppa.

From-SVN: r233028

vector-5.c: Add -fno-common option on hppa*-*-hpux*.

* gcc.dg/tree-ssa/vector-5.c: Add -fno-common option on hppa*-*-hpux*.

From-SVN: r233027

Daily bump.

From-SVN: r233026

Regenerate .pot files.

gcc/po:
* gcc.pot: Regenerate.

libcpp/po:
* cpplib.pot: Regenerate.

From-SVN: r233023

* fi.po: Update.

From-SVN: r233021

re PR c++/68763 (ICE: in verify_unstripped_args, at cp/pt.c:1132)

PR c++/68763

* tree.c (strip_typedefs) [FUNCTION_TYPE]: Avoid building a new
function type if nothing is changing.

Co-Authored-By: Jason Merrill <jason@redhat.com>
From-SVN: r233020

re PR c++/69009 (ICE in instantiate_decl, at cp/pt.c:21511)

PR c++/69009

* pt.c (partial_specialization_p, impartial_args): New.
(instantiate_decl): Call impartial_args.

From-SVN: r233019

Fix abi-tag16.C.

* mangle.c (maybe_check_abi_tags): New.
(write_guarded_var_name): Call it.
(mangle_ref_init_variable): Call check_abi_tags.

From-SVN: r233018

Fix abi-tag17.C.

* pt.c (lookup_template_class_1): Don't share TYPE_ATTRIBUTES
between template and instantiation.

From-SVN: r233017

re PR fortran/67564 (Segfault on sourced allocattion statement with class(*) arrays)

2016-01-31  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/67564
* trans-expr.c (gfc_conv_procedure_call): For the vtable copy
subroutines, add a string length argument, when the actual
argument is an unlimited polymorphic class object.

2016-01-31  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/67564
* gfortran.dg/allocate_with_source_17.f03: New test.

From-SVN: r233016

Daily bump.

From-SVN: r233015

re PR middle-end/69546 (wrong code with -O and simple int128 arithmetics)

PR tree-optimization/69546
* wide-int.cc (wi::divmod_internal): For unsigned division
where both operands fit into uhwi, if o1 is 1 and o0 has
msb set, if divident_prec is larger than bits per hwi,
clear another quotient word and return 2 instead of 1.
Similarly for remainder with msb in HWI set, if dividend_prec
is larger than bits per hwi.

* gcc.dg/torture/pr69546.c: New test.

From-SVN: r233012

re PR fortran/69566 ([OOP] Failure of SELECT TYPE with unlimited polymorphic function result)

2016-01-30  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/69566
* trans-expr.c (gfc_conv_procedure_call): Correct expression
for 'ulim_copy', which was missing a test for 'comp'.

2016-01-30  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/69566
* gfortran.dg/unlimited_polymorphic_25.f03: New test.

From-SVN: r233011

PR r++/68490 - error initializing a structure with a flexible array membe

gcc/testsuite/ChangeLog:
2016-01-30  Martin Sebor  <msebor@redhat.com>

        PR c++/68490
        * g++.dg/ext/flexary10.C: New test.

From-SVN: r233010

re PR fortran/66707 (Endless compilation on wrong usage of common)

2016-01-30  Dominique d'Humieres  <dominiq@lps.ens.fr>

PR fortran/66707
gfortran.dg/common_23.f90: New test.

From-SVN: r233008

re PR target/65546 (FAIL: gcc.dg/vect/costmodel/ppc/costmodel-vect-31a.c)

2016-01-29  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

PR target/65546
* gcc.dg/vect/costmodel/ppc/costmodel-vect-31a.c: Correct
condition being checked, and disable it when the target supports
misaligned loads and stores.

From-SVN: r233006

Daily bump.

From-SVN: r233005

[hsa] Atomic assess memory model fixes

2016-01-29  Martin Jambor  <mjambor@suse.cz>

* hsa-gen.c (get_memory_order_name): Mask with MEMMODEL_BASE_MASK.
Use short lowercase names.
(get_memory_order): Mask with MEMMODEL_BASE_MASK.  Support
MEMMODEL_CONSUME with acquire semantics and MEMMODEL_SEQ_CST with
acq_rel one.  Protect warning agains segfaults if
get_memory_order_name returns NULL.
(gen_hsa_ternary_atomic_for_builtin): Support with MEMMODEL_SEQ_CST
with release semantics.  Do not warn if get_memory_order already did.
(gen_hsa_insns_for_call): Support with MEMMODEL_SEQ_CST with acquire
semantics.  Fix check for relaxed or acquire semantics.  Do not warn
if get_memory_order already did.

From-SVN: r233000

re PR debug/69518 (Flag -g causes "error: type variant has different TYPE_VFIELD")

PR debug/69518
* c-decl.c (finish_struct): Clear C_TYPE_INCOMPLETE_VARS in
all type variants, not just TYPE_MAIN_VARIANT.

* gcc.dg/torture/pr69518.c: New test.

From-SVN: r232998

document that isl-0.16 is supported

* config/isl.m4: Add comments about isl-0.16.
* configure: Regenerate.

gcc/
* doc/install.texi: Document that isl-0.16 is supported.

From-SVN: r232997

re PR target/69299 (-mavx performance degradation with r232088)

2016-01-29  Vladimir Makarov  <vmakarov@redhat.com>

PR target/69299
* config/i386/constraints.md (Bm): Describe as special memory
constraint.
* doc/md.texi (DEFINE_SPECIAL_MEMORY_CONSTRAINT): Describe it.
* genoutput.c (main): Process DEFINE_SPECIAL_MEMORY_CONSTRAINT.
* genpreds.c (struct constraint_data): Add is_special_memory.
(have_special_memory_constraints, special_memory_start): New
static vars.
(special_memory_end): Ditto.
(add_constraint): Add new arg is_special_memory.  Add code to
process its true value.  Update have_special_memory_constraints.
(process_define_constraint): Pass the new arg.
(process_define_register_constraint): Ditto.
(choose_enum_order): Process special memory.
(write_tm_preds_h): Generate enum const CT_SPECIAL_MEMORY and
function insn_extra_special_memory_constraint.
(main): Process DEFINE_SPECIAL_MEMORY_CONSTRAINT.
* gensupport.c (process_rtx): Process
DEFINE_SPECIAL_MEMORY_CONSTRAINT.
* ira-costs.c (record_reg_classes): Process CT_SPECIAL_MEMORY.
* ira-lives.c (single_reg_class): Use
insn_extra_special_memory_constraint.
* ira.c (ira_setup_alts): Process CT_SPECIAL_MEMORY.
* lra-constraints.c (process_alt_operands): Ditto.
(curr_insn_transform): Use insn_extra_special_memory_constraint.
* recog.c (asm_operand_ok, preprocess_constraints): Process
CT_SPECIAL_MEMORY.
* reload.c (find_reloads): Ditto.
* rtl.def (DEFINE_SPECIFAL_MEMORY_CONSTRAINT): New.
* stmt.c (parse_input_constraint): Use
insn_extra_special_memory_constraint.

From-SVN: r232993

Revert revsion 229087 changes in lra-spills.c

r229087, which caused PR 69530, was supposed to fix PR 67609.  r229458
has made r229087 unnecessary.

gcc/

PR target/69530
* lra-splill.c (lra_final_code_change): Revert r229087 by
removing all sub-registers.

gcc/testsuite/

PR target/69530
* gcc.target/i386/pr69530.c: New test.

From-SVN: r232991

re PR target/65604 (MIPS -fno-delayed-branch generates incorrect code with -mcheck-zero-division)

PR target/65604
* gcc.target/mips/div-delay.c: New test.

From-SVN: r232986

re PR target/65604 (MIPS -fno-delayed-branch generates incorrect code with -mcheck-zero-division)

PR target/65604
* config/mips/mips.c (mips_output_division): Check flag_delayed_branch.

From-SVN: r232985

re PR target/69551 (Wrong code with single element vector insert)

PR target/69551
* config/i386/i386.c (ix86_expand_vector_set) <case V4SImode>: For
SSE1, copy target into the temporary reg first before recursing
on it.

* gcc.target/i386/pr69551.c: New test.

From-SVN: r232982

re PR target/66137 (ICE with -ffixed-ebp)

PR target/66137
* gcc.target/i386/pr66137.c: New test.

From-SVN: r232981

p5600-bonding.c (dg-options): Force the test to be always built for p5600.

testsuite/
2016-01-29  Andrew Bennett  <andrew.bennett@imgtec.com>

       * gcc.target/mips/p5600-bonding.c (dg-options): Force the test to be
       always built for p5600.
       * gcc.target/mips/mips.exp (mips-dg-options): Add support for the
       isa=p5600 dg-option.

From-SVN: r232980

Use vm in sse2_cvtps2pd<mask_name>

sse2_cvtps2pd<mask_name> has

(define_insn "sse2_cvtps2pd<mask_name>"
  [(set (match_operand:V2DF 0 "register_operand" "=v")
        (float_extend:V2DF
          (vec_select:V2SF
            (match_operand:V4SF 1 "vector_operand" "vBm")
            (parallel [(const_int 0) (const_int 1)]))))]

The memory operand size is 8 bytes (2 floats).  We should use vm instead
of vBm.

* config/i386/sse.md (sse2_cvtps2pd<mask_name>): Replace vBm
with vm.

From-SVN: r232979

Add myself as a hsa maintainer

2016-01-29  Martin Jambor  <mjambor@suse.cz>

* MAINTAINERS (hsa maintainers): Add myself.

From-SVN: r232978

Test __cplusplus instead of __GXX_EXPERIMENTAL_CXX0X__

* ginclude/stdarg.h: Test __cplusplus instead of
__GXX_EXPERIMENTAL_CXX0X__.

From-SVN: r232977

re PR middle-end/69547 (no-op array initializer emits an empty loop)

2016-01-29  Richard Biener  <rguenther@suse.de>

PR tree-optimization/69547
* tree-ssa-dce.c (mark_aliased_reaching_defs_necessary_1):
Do not mark clobbers necessary.
(mark_all_reaching_defs_necessary_1): Likewise.

* g++.dg/tree-ssa/pr69547.C: New testcase.

From-SVN: r232976

re PR debug/66869 (-Wunused-function no longer warns for static declarations without definition)

PR debug/66869
* decl.c (wrapup_globals_for_namespace): Warn about unused static
function declarations.

* g++.dg/warn/Wunused-function2.C: New test.

From-SVN: r232975

Fix Cygwin bootstrap error due to TM symbols

PR libstdc++/69506
* config/os/newlib/os_defines.h (_GLIBCXX_USE_WEAK_REF): Define.

From-SVN: r232974

S/390: Use %qs in error messages.

gcc/ChangeLog

* config/s390/s390-c.c (s390_resolve_overloaded_builtin): Format
declaration name with %qs and print it in both error messages.  Also
fix indentation.

From-SVN: r232973

PR other/69006: S/390: Fix extra newlines after diagnostics.

gcc/ChangeLog

PR other/69006
* config/s390/s390-c.c (s390_resolve_overloaded_builtin): Remove
trailing blank line from error message.

From-SVN: r232972

S/390: Require a hardware vector support for test to succeed.

The test case works on S/390 too, but only with -march=z13 or later
(i.e. if Gcc can make use of hardware vector support).  Otherwise the
optimization gets too complex.  The attached patch forces Gcc to use
-march=z13 instead of xfail'ing the test on S/390.

gcc/testsuite/ChangeLog

* gcc.dg/tree-ssa/ssa-dom-cse-2.c: Require a hardware vector support for
test to succeed.

From-SVN: r232971

PR c++/69462: Provide FLT_EVAL_METHOD and DECIMAL_DIG in float.h.

gcc/ChangeLog

PR c++/69462
* ginclude/float.h: Also provide FLT_EVAL_METHOD and DECIMAL_DIG
for C++-11.

From-SVN: r232970

re PR c++/69509 (infinite loop compiling a VLA in a recursive constexpr function)

PR c++/69509
PR c++/69516
* constexpr.c (cxx_eval_array_reference): Give the "array subscript
out of bound" error earlier.
* init.c (build_vec_init): Change NE_EXPR into GT_EXPR.  Update the
commentary.

* g++.dg/ext/constexpr-vla2.C: New test.
* g++.dg/ext/constexpr-vla3.C: New test.
* g++.dg/ubsan/vla-1.C: Remove dg-shouldfail.

From-SVN: r232969

re PR middle-end/69537 (Incorrect -Wmaybe-uninitialized warning with enum variable)

2016-01-29  Richard Biener  <rguenther@suse.de>

PR middle-end/69537
* match.pd: Allow all integral types when simplifying a
widening or sign-changing conversion.

* gcc.dg/uninit-21.c: New testcase.

From-SVN: r232968

PR69545: revert patch

Revert 'assert instead of silently failing code gen' that produces an ICE with isl-0.14.

From-SVN: r232966

Fix cp_binding_level reuse logic

gcc/cp/ChangeLog:

* name-lookup.c (begin_scope): After reusing a cp_binding_level
structure, update free_binding_level before the structure's
level_chain field gets cleared, not after.

From-SVN: r232965

Daily bump.

From-SVN: r232963

re PR pch/68176 (all pch tests fail on eglibc systems (with bits/predefs.h))

PR pch/68176
* files.c (_cpp_find_file): Set file->implicit_preinclude even if
included from file->implicit_preinclude header.

From-SVN: r232956

re PR target/69459 (wrong code with -O2 and vector arithmetics @ x86_64)

PR target/69459
* config/i386/constraints.md (C): Only accept constant zero operand.
(BC): New constraint.
* config/i386/sse.md (*mov<mode>_internal): Use BC constraint
instead of C constraint.
* doc/md.texi (Machine Constraints): Update description
of C constraint.

testsuite/ChangeLog:

PR target/69459
* gcc.target/i386/pr69459.c: New test.

From-SVN: r232955

re PR target/68400 (ICE in change_address_1, at emit-rtl.c:2125)

PR target/68400
* gcc.target/mips/mips.exp (mips_option_groups): Add stack-protector.
* gcc.target/mips/pr68400.c: New test.

From-SVN: r232954

re PR target/68400 (ICE in change_address_1, at emit-rtl.c:2125)

2016-01-28  Steve Ellcey  <sellcey@imgtec.com>

PR target/68400
* config/mips/mips.c (and_operands_ok): Add MIPS16 check.

From-SVN: r232952

PR target/17381 - Unnecessary register move for float extend

2016-01-28  Martin Sebor  <msebor@redhat.com>

PR target/17381
* gcc.target/powerpc/pr17381.c: New test.

From-SVN: r232947

re PR middle-end/69542 (-fcompare-debug failure in simplify-rtx.c on i?86 since r232905)

PR middle-end/69542
* lra-remat.c (calculate_local_reg_remat_bb_data): Only consider
non-debug insns.

* gcc.dg/torture/pr69542.c: New test.

From-SVN: r232946

rs6000.c (output_cbranch): Don't statically predict branches if using guessed profile.

* config/rs6000/rs6000.c (output_cbranch): Don't statically predict
branches if using guessed profile.

From-SVN: r232945

Fix optimize_isl dump

* graphite-optimize-isl.c (optimize_isl): Fix dump.

From-SVN: r232939

re PR target/69305 (wrong code with -O and int128 @ aarch64)

PR target/69305

  * config/aarch64/aarch64-modes.def (CC_Cmode): New
  * config/aarch64/aarch64-protos.h: Update.
  * config/aarch64/aarch64.c (aarch64_zero_extend_const_eq): New.
  (aarch64_select_cc_mode): Add check for use of CC_Cmode.
  (aarch64_get_condition_code_1): Handle CC_Cmode.
  * config/aarch64/aarch64.md (addti3): Use adddi3_compareC.
  (*add<mode>3_compareC_cconly_imm): New.
  (*add<mode>3_compareC_cconly): New.
  (*add<mode>3_compareC_imm): New.
  (add<mode>3_compareC): New.
  (add<mode>3_carryin, *addsi3_carryin_uxtw): Sort compare operand
  to be first.  Use aarch64_carry_operation.
  (*add<mode>3_carryin_alt1, *addsi3_carryin_alt1_uxtw): Remove.
  (*add<mode>3_carryin_alt2, *addsi3_carryin_alt2_uxtw): Remove.
  (*add<mode>3_carryin_alt3, *addsi3_carryin_alt3_uxtw): Remove.
  (subti3): Use subdi3_compare1.
  (*sub<mode>3_compare0): Rename from sub<mode>3_compare0.
  (sub<mode>3_compare1): New.
  (*sub<mode>3_carryin0, *subsi3_carryin_uxtw): New.
  (*sub<mode>3_carryin): Use aarch64_borrow_operation.
  (*subsi3_carryin_uxtw): Likewise.
  (*ngc<mode>, *ngcsi_uxtw): Likewise.
  (*sub<mode>3_carryin_alt, *subsi3_carryin_alt_uxtw): New.
  * config/aarch64/iterators.md (DWI): New.
  * config/aarch64/predicates.md (aarch64_carry_operation): New.
  (aarch64_borrow_operation): New.

From-SVN: r232936

make debug comment more explicit

2016-01-28  Abderrazek Zaafrani  <a.zaafrani@samsung.com>

* graphite-optimize-isl.c (optimize_isl): Print a different debug
message when isl does not return a valid schedule.

From-SVN: r232935

remove out of sync comments

* graphite-isl-ast-to-gimple.c (class translate_isl_ast_to_gimple):
Remove comments from class declarations: they are already in the code
close by the defs.

From-SVN: r232934

use codegen_error_p

* graphite-isl-ast-to-gimple.c (binary_op_to_tree): Call codegen_error_p.
(ternary_op_to_tree): Same.
(unary_op_to_tree): Same.
(nary_op_to_tree): Same.
(gcc_expression_from_isl_expr_op): Same.
(gcc_expression_from_isl_expression): Same.
(graphite_create_new_loop): Same.
(graphite_create_new_loop_guard): Same.
(build_iv_mapping): Same.
(graphite_create_new_guard): Same.
(translate_isl_ast_to_gimple::copy_loop_phi_nodes): Same.
(translate_isl_ast_to_gimple::translate_pending_phi_nodes): Same.

From-SVN: r232933

assert instead of silently failing code gen

* graphite-isl-ast-to-gimple.c (get_rename_from_scev): Assert instead of
setting codegen_error to fail codegen.

From-SVN: r232932

* tree.h (BINFO_FLAG_0): Rename from BINFO_MARKED.

From-SVN: r232931

A recent change added UNSPEC to the CCMP patterns to stop combine optimizing the immediate in a rare case.

A recent change added UNSPEC to the CCMP patterns to stop combine optimizing
the immediate in a rare case.  This requires a fix to the CCMP cost
calculation as the CCMP instruction with unspec is no longer recognized.
Fix the ccmp_1.c test by adding -ffinite-math-only so FCCMPE is emitted on
relational compares.

2016-01-28  Wilco Dijkstra  <wdijkstr@arm.com>

    gcc/
* config/aarch64/aarch64.c (aarch64_if_then_else_costs):
Remove CONST_INT_P check in CCMP cost calculation.

    gcc/testsuite/
* gcc.target/aarch64/ccmp_1.c: Fix test issue.

From-SVN: r232930

re PR c++/67407 (ice in friend_accessible_p)

PR c++/67407

* search.c (dfs_walk_once, dfs_walk_once_r)
(dfs_walk_once_accessible_r, dfs_walk_once_accessible): Use
hash_set instead of BINFO_MARKED.
(dfs_unmark_r): Remove.

From-SVN: r232929

* directives.c (destringize_and_run): Adjust prototype.

From-SVN: r232928

pr65980.c: Skip on AIX.

        * gcc.dg/pr65980.c: Skip on AIX.
        * gcc.dg/pr66688.c: Same.

From-SVN: r232927

re PR target/63679 ([AArch64] Failure to constant fold.)

        PR target/63679
        * gcc.dg/tree-ssa/sra-17.c: Adjust regex for more label styles.
        * gcc.dg/tree-ssa/sra-18.c: Same.

From-SVN: r232926

re PR libstdc++/69450 (libstdc++-v3/include/math.h:66:1 2: error: 'constexpr bool std::isnan(double)' conflicts with a previous declaration)

PR libstdc++/69450
* acinclude.m4 (GLIBCXX_CHECK_MATH11_PROTO): Split check for obsolete
isinf and isnan functions into two independent checks.  Check on hpux.
* config.h.in: Regenerate.
* configure: Regenerate.
* include/c_global/cmath (isinf(double), isnan(double)): Use
_GLIBCXX_HAVE_OBSOLETE_ISINF and _GLIBCXX_HAVE_OBSOLETE_ISNAN,
respectively.

From-SVN: r232925

gigi.h (enum attr_type): Rename into...

* gcc-interface/gigi.h (enum attr_type): Rename into...
(enum attrib_type): ...this.
(struct attrib): Adjust.
* gcc-interface/decl.c (prepend_one_attribute): Likewise.

From-SVN: r232923

Add support for vector permute cost since various permutes can expand into a complex sequence of instructions.

Add support for vector permute cost since various permutes can expand
into a complex sequence of instructions.  This fixes major performance
regressions due to recent changes in SLP vectorizer (which now vectorizes
more aggressively and emits many complex permutes).  Set the cost to > 1
for all microarchitectures so that the number of permutes is usually zero
and regressions disappear.

2016-01-28  Wilco Dijkstra  <wdijkstr@arm.com>

* config/aarch64/aarch64.c (generic_vector_cost):
Set vec_permute_cost.
(cortexa57_vector_cost): Likewise.
(exynosm1_vector_cost): Likewise.
(xgene1_vector_cost): Likewise.
(aarch64_builtin_vectorization_cost): Use vec_permute_cost.
* config/aarch64/aarch64-protos.h (cpu_vector_cost):
Add vec_permute_cost entry.

From-SVN: r232922

Several instructions disassemble a zero immediate as wzr/xzr due to using a register operand in the disassembly.

Several instructions disassemble a zero immediate as wzr/xzr due to
using a register operand in the disassembly.  Avoid this by removing
the register operand.

2016-01-28  Wilco Dijkstra  <wdijkstr@arm.com>

* config/aarch64/aarch64.md (ccmp<mode>): Disassemble
immediate as %1.
(add<mode>3_compare0): Likewise.
(addsi3_compare0_uxtw): Likewise.
(add<mode>3nr_compare0): Likewise.
(compare_neg<mode>): Likewise.
(<optab><mode>3): Likewise.

From-SVN: r232921

Added missing testcase of r232919.

From-SVN: r232920

re PR fortran/62536 (ICE (segfault) for invalid END BLOCK statement)

gcc/fortran/ChangeLog:

2016-01-28  Andre Vehreschild  <vehre@gcc.gnu.org>

PR fortran/62536
* decl.c (gfc_match_end): Only unnest and remove BLOCK namespaces
when the END encountered does not match a BLOCK's end.

gcc/testsuite/ChangeLog:

2016-01-28  Andre Vehreschild  <vehre@gcc.gnu.org>

PR fortran/62536
* gfortran.dg/block_15.f08: New test.
* gfortran.dg/block_end_error_1.f90: Need to catch additional error
on incorrectly closed BLOCK.

From-SVN: r232919

tree-vect-stmts.c (vectorizable_comparison): Add NULL check for vectype.

gcc/

* tree-vect-stmts.c (vectorizable_comparison): Add
NULL check for vectype.

gcc/testsuite/

* gcc.dg/declare-simd.c: New test.

From-SVN: r232917