[Patch Testsuite, obvious] Workaround fragility in gcc.dg/tree-ssa/foldconst-6.c

gcc/testsuite/

* gcc.dg/tree-ssa/foldconst-6.c: Change expected pattern for
tree dump scanning.

From-SVN: r220440

Fix failures on AIX (PR libobjc/63765)

PR libobjc/63765
* thr.c (_XOPEN_SOURCE): Remove.

From-SVN: r220438

re PR ada/64349 (Bootstrapping Ada fails on darwin(9|10).)

2015-02-05  Tristan Gingold  <gingold@adacore.com>

PR ada/64349
* env.c: Fix thinko: handle Darwin case before default one.

From-SVN: r220437

re PR fortran/64757 (ICE in fold_convert_loc, at fold-const.c:2353)

2015-02-05  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/64757
* resolve.c (resolve_structure_cons): Obtain the rank of class
components.
* trans-expr.c (gfc_trans_alloc_subarray_assign): Do the
assignment to allocatable class array components.
(alloc_scalar_allocatable_for_subcomponent_assignment): If comp
is a class component, allocate to the _data field.
(gfc_trans_subcomponent_assign): If a class component with a
derived type expression set the _vptr field and for array
components, call gfc_trans_alloc_subarray_assign. For scalars,
the assignment is performed here.

2015-02-05  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/64757
* gfortran.dg/type_to_class_2.f90: New test
* gfortran.dg/type_to_class_3.f90: New test

From-SVN: r220436

2015-02-05  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/640757
* resolve.c (resolve_structure_cons): Obtain the rank of class
components.
* trans-expr.c (gfc_trans_alloc_subarray_assign): Do the
assignment to allocatable class array components.
(alloc_scalar_allocatable_for_subcomponent_assignment): If comp
is a class component, allocate to the _data field.
(gfc_trans_subcomponent_assign): If a class component with a
derived type expression set the _vptr field and for array
components, call gfc_trans_alloc_subarray_assign. For scalars,
the assignment is performed here.

2015-02-05  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/640757
* gfortran.dg/type_to_class_2.f90: New test
* gfortran.dg/type_to_class_3.f90: New test

From-SVN: r220435

re PR ipa/61548 (FAIL: gcc.dg/tls/alias-1.c)

PR ipa/61548
* tree-emultls.c (new_emutls_decl): Resolve alias after creating it.

Co-Authored-By: Trevor Saunders <tsaunders@mozilla.com>
From-SVN: r220434

re PR ipa/61548 (FAIL: gcc.dg/tls/alias-1.c)

PR ipa/61548
* ipa.c (symbol_table::remove_unreachable_nodes): Fix ordering issue
when removing varpool nodes.

From-SVN: r220433

re PR ipa/61548 (FAIL: gcc.dg/tls/alias-1.c)

PR ipa/61548
* varpool.c (varpool_node::remove): Fix order of variables.

From-SVN: r220432

Daily bump.

From-SVN: r220430

re PR ipa/64686 (ICE: in edge_badness, at ipa-inline.c:912 during Firefox LTO build with enabled checking)

PR ipa/64686
* ipa-inline.c (inline_small_functions): Fix ordering issue between
speculation resolution and key updates.
* g++.dg/torture/pr64686.C: New testcase.

From-SVN: r220429

ipa-prop.c (update_indirect_edges_after_inlining): By more careful about not letting any speculative edges unupdated.

* ipa-prop.c (update_indirect_edges_after_inlining): By more careful
about not letting any speculative edges unupdated.

From-SVN: r220425

re PR gcov-profile/64123 (Instrumented Firefox segfaults on start)

PR gcov/64123
* gcov-io.c (gcov_var): Export.

From-SVN: r220424

re PR middle-end/64922 (runtime error: member call on misaligned address for type 'struct _Rep')

PR middle-end/64922
* ipa-prop.c (update_indirect_edges_after_inlining): Correctly update
edges that become speculative.

From-SVN: r220423

dwarf2.h (enum dwarf_source_language): Add DW_LANG_Fortran03 and DW_LANG_Fortran08.

include/
* dwarf2.h (enum dwarf_source_language): Add DW_LANG_Fortran03
and DW_LANG_Fortran08.
gcc/
* dwarf2out.c (is_fortran): Also return true for DW_LANG_Fortran03
or DW_LANG_Fortran08.
(lower_bound_default): Return 1 for DW_LANG_Fortran03 or
DW_LANG_Fortran08.
(gen_compile_unit_die): Handle "GNU Fortran2003" and
"GNU Fortran2008" language strings.
* dbxout.c (get_lang_number): Use lang_GNU_Fortran.
* langhooks.h (lang_GNU_Fortran): New prototype.
* langhooks.c (lang_GNU_Fortran): New function.
* config/rs6000/rs6000.c (rs6000_output_function_epilogue): Use
lang_GNU_Fortran.
gcc/fortran/
* options.c: Include langhooks.h.
(gfc_post_options): Change lang_hooks.name based on
selected -std= mode.

From-SVN: r220422

sparc.c (RTX_OK_FOR_OFFSET_P): Fix off-by-one error.

* config/sparc/sparc.c (RTX_OK_FOR_OFFSET_P): Fix off-by-one error.
(RTX_OK_FOR_OLO10_P): Likewise.

From-SVN: r220421

re PR middle-end/64824 (ICE in gimple verification)

PR c/64824
PR c/64868
gcc/c/
* c-parser.c (c_parser_omp_atomic): Handle RDIV_EXPR.
gcc/cp/
* parser.c (cp_parser_omp_atomic): Handle RDIV_EXPR.
gcc/c-family/
* c-omp.c (c_finish_omp_atomic): Use TRUNC_DIV_EXPR
instead of RDIV_EXPR.  Use build_binary_op instead of
build2_loc.
libgomp/
* testsuite/libgomp.c/pr64824.c: New test.
* testsuite/libgomp.c/pr64868.c: New test.
* testsuite/libgomp.c++/pr64824.C: New test.
* testsuite/libgomp.c++/pr64868.C: New test.

From-SVN: r220420

* tree-ssa-loop-ivopts.c (get_address_cost): Use right mode for offset.

From-SVN: r220419

re PR middle-end/64922 (runtime error: member call on misaligned address for type 'struct _Rep')

PR middle-end/64922
* gimple.c: Include gimple-ssa.h.
(maybe_remove_unused_call_args): New function.
* gimple.h (maybe_remove_unused_call_args): Declare.
* cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Use it.
* tree-ssa-pre.c (eliminate_dom_walker::before_dom_children): Likewise.
* gimple-fold.c (gimple_fold_call): Likewise.

From-SVN: r220417

Actually check in the fix

From-SVN: r220416

Clear hard frame pointer alignment if not needed

When hard frame pointer isn't needed, the register for hard frame pointer
may be reused.  This patch clears alignment on hard frame pointer if hard
frame pointer isn't needed.

gcc/

PR rtl-optimization/64905
* lra-eliminations.c (setup_can_eliminate): Clear hard frame
pointer alignment if it isn't needed.

gcc/testsuite/

PR rtl-optimization/64905
* gcc.target/i386/pr64905.c: New file.

From-SVN: r220414

[AArch64] Add support for -mcpu=cortex-a72

2015-02-04  Matthew Wahab  <matthew.wahab@arm.com>

  gcc/
    * config/aarch64/aarch64-cores.def: Add cortex-a72 and
    cortex-a72.cortex-a53.
    * config/aarch64/aarch64-tune.md: Regenerate.
    * doc/invoke.texi (AArch64 Options/-mtune): Add "cortex-a72".

From-SVN: r220413

More fixes for update_web_docs_svn for jit docs (PR jit/64257)

gcc/jit/ChangeLog:
PR jit/64257
* docs/conf.py (html_theme): Change from 'pyramid'
to 'sphinxdoc'.

maintainer-scripts/ChangeLog:
PR jit/64257
* update_web_docs_svn: Update build of jit docs to use EPEL6
variant install of sphinx 1.0.8, rather than 0.6.6.

From-SVN: r220412

pr41447-1.c: Remove xfail.

* gcc.dg/guality/pr41447-1.c: Remove xfail.
* gcc.dg/guality/pr41616-1.c: Ditto.

From-SVN: r220411

fpmath-sf.S (__rl78_int_pack_a_r8): Fix edge case rounding up the fraction.

* config/rl78/fpmath-sf.S (__rl78_int_pack_a_r8): Fix edge case
rounding up the fraction.

* config/rl78/rl78.c (rl78_note_reg_set): Note the use of REGs
inside a MEM.

From-SVN: r220410

jit: fix typo in docs

gcc/jit/ChangeLog:
* docs/topics/contexts.rst (gcc_jit_context_acquire): Fix
typo.
* docs/_build/texinfo/libgccjit.texi: Regenerate.

From-SVN: r220408

builtins.def (DEF_BUILTIN_CHKP): Define if not defined.

* builtins.def (DEF_BUILTIN_CHKP): Define if not defined.
(DEF_LIB_BUILTIN_CHKP, DEF_EXT_LIB_BUILTIN_CHKP): Redefine.
(DEF_CHKP_BUILTIN): Define using DEF_BUILTIN_CHKP instead
of DEF_BUILTIN.
(BUILT_IN_MEMCPY, BUILT_IN_MEMMOVE, BUILT_IN_MEMSET, BUILT_IN_STRCAT,
BUILT_IN_STRCHR, BUILT_IN_STRCPY, BUILT_IN_STRLEN): Use
DEF_LIB_BUILTIN_CHKP macro instead of DEF_LIB_BUILTIN.
(BUILT_IN_MEMCPY_CHK, BUILT_IN_MEMMOVE_CHK, BUILT_IN_MEMPCPY_CHK,
BUILT_IN_MEMPCPY, BUILT_IN_MEMSET_CHK, BUILT_IN_STPCPY_CHK,
BUILT_IN_STPCPY, BUILT_IN_STRCAT_CHK, BUILT_IN_STRCPY_CHK): Use
DEF_EXT_LIB_BUILTIN_CHKP macro instead of DEF_EXT_LIB_BUILTIN.
* tree-core.h (enum built_in_function): In between
BEGIN_CHKP_BUILTINS and END_CHKP_BUILTINS only define enum values
for builtins that use DEF_BUILTIN_CHKP macro.

From-SVN: r220407

Fix signedness of compares expanding debug exprs.

for  gcc/ChangeLog

PR debug/64817
* cfgexpand.c (expand_debug_expr): Compute unsignedp from
operands for tcc_comparison exprs.  Fix typos.

for  gcc/testsuite/ChangeLog

PR debug/64817
* gcc.dg/pr64817-3.c: New.

From-SVN: r220406

Simplify XOR of (AND or IOR) of XOR.

for  gcc/ChangeLog

PR debug/64817
* simplify-rtx.c (simplify_binary_operation_1): Simplify one
of two XORs that have an intervening AND or IOR.

From-SVN: r220405

Avoid allocating memory when trying but failing to simplify XOR of AND.

for  gcc/ChangeLog

PR debug/64817
* simplify-rtx.c (simplify_binary_operation_1): Rewrite
simplification of XOR of AND to not allocate new rtx before
committing to a simplification.

From-SVN: r220404

[AArch64] Use std::swap instead of manually swapping in aarch64-ldpstp.md

* config/aarch64/aarch64-ldpstp.md: Use std::swap instead of
manual swaps in all peepholes.

From-SVN: r220403

[AArch64] Use std::swap instead of manually swapping

* config/aarch64/aarch64.c (aarch64_evpc_ext): Use std::swap instead
of manual swapping implementation.
(aarch64_expand_vec_perm_const_1): Likewise.

From-SVN: r220402

[AArch64] Rid the world of NAMED_PARAM

gcc/

* config/aarch64/aarch64.c (NAMED_PARAM): Delete it.
(generic_addrcost_table): Remove NAMED_PARAM.
(cortexa57_addrcost_table): Likewise.
(xgene1_addrcost_table): Likewise.
(generic_regmove_table): Likewise.
(cortexa53_regmove_table): Likewise.
(xgene1_regmove_table): Likewise.
(generic_vector_table): Likewise.
(cortexa57_vector_table): Likewise.
(xgene1_vector_table): Likewise.
(generic_tunings): Likewise.
(cortexa53_tunings): Likewise.
(cortexa57_tunings): Likewise.
(xgene1_tunings): Likewise.

From-SVN: r220401

re PR target/64408 (fr30-elf ICE in extract_insn, at recog.c:2202)

PR target/64408
* config/fr30/predicates.md (di_operand): Add SUBREG to the list
of accepted codes.
(nonimmediate_di_operand): Likewise.

From-SVN: r220400

[ARM] Add support for -mcpu=cortex-a72 and -mcpu=cortex-a72.cortex-a53

gcc/

* config/arm/arm-cores.def: Add cortex-a72 and
cortex-a72.cortex-a53.
* config/arm/bpabi.h (BE8_LINK_SPEC): Likewise.
* config/arm/t-aprofile (MULTILIB_MATCHES): Likewise.
* config/arm/arm-tune.md: Regenerate.
* config/arm/arm-tables.opt: Add entries for "cortex-a72" and
"cortex-a72.cortex-a53".
* doc/invoke.texi (ARM Options/-mtune): Likewise.

From-SVN: r220399

Adjust pr36728-*.c for targets that pass parameters in registers.

Co-Authored-By: Ramana Radhakrishnan <ramana.radhakrishnan@arm.com>
From-SVN: r220398

Minor corrections to gcc.dg/combine_ashiftrt_[12].c

2015-02-04  Alan Lawrence  <alan.lawrence@arm.com>
    Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>

* gcc.dg/combine_ashiftrt_1.c: Sort, complete and explain target
list, allow for multilibed targets.
* gcc.dg/combine_ashiftrt_2.c: Likewise.

Co-Authored-By: Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
From-SVN: r220397

msp430.c (msp430_use_f5_series_hwmult): Add more prefixes of known F5 using MSP430 MCUs.

* config/msp430/msp430.c (msp430_use_f5_series_hwmult): Add more
prefixes of known F5 using MSP430 MCUs.

From-SVN: r220396

[AArch64] Use target builtin instead of __builtin_sqrt for vsqrt_f64

* config/aarch64/aarch64-simd-builtins.def (sqrt): Use BUILTIN_VDQF_DF.
* config/aarch64/arm_neon.h (vsqrt_f64): Use __builtin_aarch64_sqrtdf
instead of __builtin_sqrt.

From-SVN: r220395

builtins-58.c: Check for pow at the end of words only.

* gcc.dg/builtins-58.c: Check for pow at the end of words only.
* gcc.dg/pr46728-6.c: Ditto.

From-SVN: r220394

loongson-simd.c: Update comment to clarify the need for mips_nanlegacy target.

2015-02-04  Robert Suchanek  <robert.suchanek@imgtec.com>

* gcc.target/mips/loongson-simd.c: Update comment to clarify the need
for mips_nanlegacy target.

From-SVN: r220393

[PATCH][libstdc++][Testsuite] isctype test fails for newlib.

libstdc++-v3/
2015-02-02  Matthew Wahab  <matthew.wahab@arm.com>

PR libstdc++/64467
* testsuite/28_regex/testsuiteraits/char/isctype.cc (test01): Add newlib
special case for '\n'.
* test01estsuite/28_regex/traits/wchar_t/isctype.cc (test01): Likewise.

From-SVN: r220392

varasm.c (do_assemble_alias): Follow transparent alias chain for target.

gcc/

* varasm.c (do_assemble_alias): Follow transparent alias
chain for target.
(default_assemble_visibility): Follow transparent alias
chain for decl name.

gcc/testsuite/

* gcc.target/i386/chkp-hidden-def.c: New.

From-SVN: r220391

re PR middle-end/62103 (Incorrect folding of bitfield in a union on big endian targets)

2015-02-04  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    gcc/
    PR middle-end/62103
    * tree-ssa-sccvn.c (fully_constant_vn_reference_p): Use TYPE_PRECISION
    to compute size of referenced value in the constant case.

    gcc/testsuite/
    PR middle-end/62103
    * gcc.c-torture/execute/bitfld-7.c: New test adapted from bitfld-6.c
    to use 24 bits for bitfield b.

From-SVN: r220390

re PR testsuite/64796 (effective target bswap64 globally caches target-specific use of lp64)

2015-02-04  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    PR testsuite/64796
    * lib/target-supports.exp (check_effective_target_bswap64): Do not
    cache result in a global variable.  Include all 32-bit targets for
    bswap64 tests.

From-SVN: r220388

Daily bump.

From-SVN: r220387

intrinsic.texi (CO_ASSOCIATED): c_prt_1 should be c_ptr_1.

2015-02-03  Steven G. Kargl  <kargl@gcc.gnu.org>

* intrinsic.texi (CO_ASSOCIATED): c_prt_1 should be c_ptr_1.

From-SVN: r220381

go/types: Remove unused testdata files.

The go/types package was removed in July 2013, but the
testdata files were accidentally left behind.

From-SVN: r220378

re PR rtl-optimization/64756 (wrong code at -O3 on x86_64-linux-gnu (in 32-bit mode))

PR rtl-optimization/64756
* cse.c (invalidate_dest): New function.
(cse_insn): Use it.  If dest != SET_DEST (sets[i].rtl) and
HASH (SET_DEST (sets[i].rtl), mode) computation sets do_not_record,
invalidate and do not record it.

* gcc.c-torture/execute/pr64756.c: New test.

From-SVN: r220377

re PR target/64660 ([SH] Convert atomic_fetch_<op> to atomic_<op>_fetch)

gcc/
PR target/64660
* config/sh/sync.md (atomic_<fetchop_name><mode>_hard,
atomic_not<mode>_hard, atomic_<fetchop_name><mode>_soft_tcb,
atomic_not<mode>_soft_tcb, atomic_nand<mode>_hard,
atomic_nand<mode>_soft_tcb): New insns.
(atomic_fetch_<fetchop_name>si_hard): Convert to insn_and_split.
Split into atomic_<fetchop_name>_fetchsi_hard if operands[0] is unused.
(define_insn "atomic_fetch_notsi_hard): Convert to insn_and_split.
Split into atomic_not_fetchsi_hard if operands[0] is unused.
(atomic_fetch_<fetchop_name><mode>_hard): Convert to insn_and_split.
Split into atomic_<fetchop_name><mode>_hard if operands[0] is unused.
(atomic_fetch_not<mode>_hard): Convert to insn_and_split.  Split into
atomic_not<mode>_hard if operands[0] is unused.
(atomic_fetch_<fetchop_name><mode>_soft_gusa): Convert to
insn_and_split.  Split into atomic_<fetchop_name>_fetch<mode>_soft_gusa
if operands[0] is unused.
(atomic_fetch_not<mode>_soft_gusa): Convert to insn_and_split.  Split
into atomic_not_fetch<mode>_soft_gusa if operands[0] is unused.
(atomic_fetch_<fetchop_name><mode>_soft_tcb): Convert to insn_and_split.
Split into atomic_<fetchop_name><mode>_soft_tcb if operands[0] is
unused.
(atomic_fetch_not<mode>_soft_tcb): Convert to insn_and_split.  Split
into atomic_not<mode>_soft_tcb if operands[0] is unused.
(atomic_fetch_<fetchop_name><mode>_soft_imask): Convert to
insn_and_split.  Split into atomic_<fetchop_name>_fetch<mode>_soft_imask
if operands[0] is unused.
(atomic_fetch_not<mode>_soft_imask): Convert to insn_and_split.  Split
into atomic_not_fetch<mode>_soft_imask is operands[0] is unused.
(atomic_fetch_nandsi_hard): Convert to insn_and_split.  Split into
atomic_nand_fetchsi_hard if operands[0] is unused.
(atomic_fetch_nand<mode>_hard): Convert to insn_and_split.  Split into
atomic_nand<mode>_hard if operands[0] is unused.
(atomic_fetch_nand<mode>_soft_gusa): Convert to insn_and_split.  Split
into atomic_nand_fetch<mode>_soft_gusa if operands[0] is unused.
(atomic_fetch_nand<mode>_soft_tcb): Convert to insn_and_split.  Split
into atomic_nand<mode>_soft_tcb if operands[0] is unused.
(atomic_fetch_nand<mode>_soft_imask): Convert to insn_and_split.  Split
into atomic_nand_fetch<mode>_soft_imask if operands[0] is unused.
(atomic_<fetchop_name>_fetch<mode>_hard): Convert to insn_and_split.
Split into atomic_<fetchop_name><mode>_hard if operands[0] is unused.
(atomic_not_fetch<mode>_hard): Convert to insn_and_split.  Split into
atomic_not<mode>_hard if operands[0] is unused.
(atomic_<fetchop_name>_fetch<mode>_soft_tcb): Convert to insn_and_split.
Split into atomic_<fetchop_name><mode>_soft_tcb if operands[0] is
unused.
(atomic_not_fetch<mode>_soft_tcb): Convert to insn_and_split.  Split
into atomic_not<mode>_soft_tcb if operands[0] is unused.
(atomic_nand_fetch<mode>_hard): Convert to insn_and_split.  Split into
atomic_nand<mode>_hard if operands[0] is unused.
(atomic_nand_fetch<mode>_soft_tcb): Convert to insn_and_split.  Split
into atomic_nand<mode>_soft_tcb if operands[0] is unused.

gcc/testsuite/
PR target/64660
* gcc.target/sh/pr64660-0.h: New.
* gcc.target/sh/pr64660-1.c: New.
* gcc.target/sh/pr64660-2.c: New.
* gcc.target/sh/pr64660-3.c: New.
* gcc.target/sh/pr64660-4.c: New.

From-SVN: r220376

jit: option-logging

gcc/jit/ChangeLog:
* jit-logging.h (gcc::jit::log_user::log): Make const.
* jit-recording.c (gcc::jit::recording::context::set_str_option):
Log the new value of the option.
(gcc::jit::recording::context::set_int_option): Likewise.
(gcc::jit::recording::context::set_bool_option): Likewise.
(gcc::jit::recording::context::compile): Log the value of all
options.
(gcc::jit::recording::context::compile_to_file): Likewise.
(gcc::jit::recording::context::log_all_options): New function.
(gcc::jit::recording::context::log_str_option): New function.
(gcc::jit::recording::context::log_int_option): New function.
(gcc::jit::recording::context::log_bool_option): New function.
* jit-recording.h (gcc::jit::recording::context::log_all_options):
New function.
(gcc::jit::recording::context::log_str_option): New function.
(gcc::jit::recording::context::log_int_option): New function.
(gcc::jit::recording::context::log_bool_option): New function.
* docs/internals/test-hello-world.exe.log.txt: Update for above
changes.

From-SVN: r220375

re PR c++/64877 (strange warning message from -Waddress)

/cp
2015-02-03  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/64877
* typeck.c (cp_build_binary_op): Avoid spurious -Waddress warnings
for generated expressions.

/testsuite
2015-02-03  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/64877
* g++.dg/warn/Waddress-2.C: New.

From-SVN: r220374

PR jit/64810: driver, arm, jit: configure-time default options

gcc/ChangeLog:
PR jit/64810
* Makefile.in (GCC_OBJS): Add gcc-main.o.
* gcc-main.c: New file, containing "main" taken from gcc.c.
* gcc.c (do_self_spec): Free decoded_options.
(class driver): Move declaration to gcc.h.
(main): Move declaration and implementation to new file
gcc-main.c.
(driver_get_configure_time_options): New function.
* gcc.h (class driver): Move this declaration here, from
gcc.c.
(driver_get_configure_time_options): New declaration.

gcc/jit/ChangeLog:
PR jit/64810
* Make-lang.in (jit_OBJS): Add jit/jit-spec.o and gcc.o.
(LIBGCCJIT_FILENAME): Add EXTRA_GCC_OBJS.
* jit-playback.c: Include gcc.h.
(gcc::jit::playback::context::compile): Move mutex acquisition
to before the call to make_fake_args.
(append_arg_from_driver): New function.
(gcc::jit::playback::context::make_fake_args): On the first call,
call into driver_get_configure_time_options to get configure-time
default options and cache them.  Add them to the args for
toplev::main.
* jit-spec.c: New source file.
* docs/internals/test-hello-world.exe.log.txt: Update to reflect
above changes.

From-SVN: r220373

ipa-inline-analysis.c (simple_edge_hints): Fix check for cross-module inlining.

* ipa-inline-analysis.c (simple_edge_hints): Fix check for
cross-module inlining.
* cgraph.h (cgraph_node): Add flag merged.
* ipa-icf.c (sem_function::merge): Maintain it.

* lto-symtab.c (lto_cgraph_replace_node): Maintain merged flag.

From-SVN: r220372

arm.c (thumb2_reorg): Test UNARY_P and BINARY_P instead of OBJECT_P.

gcc/
* config/arm/arm.c (thumb2_reorg): Test UNARY_P and BINARY_P
instead of OBJECT_P.

From-SVN: r220371

re PR middle-end/61225 (Several new failures after r210458 on x86_64-*-* with -m32)

PR middle-end/61225
gcc.target/i386/pr49095.c: XFAIL for ia32.

From-SVN: r220370

re PR target/62631 (gcc.dg/tree-ssa/ivopts-lt-2.c FAILs)

PR target/62631
* config/sparc/sparc.h (TARGET_HARD_MUL): Remove TARGET_V8PLUS.
(TARGET_HARD_MUL32): Rewrite based on TARGET_HARD_MUL.
* config/sparc/sparc.c (sparc_rtx_costs) <MULT>: Return costs based on
int_mulX for integers in 64-bit mode if TARGET_HARD_MUL is not set.

From-SVN: r220369

re PR other/63504 (Issues found by --enable-checking=valgrind)

PR other/63504
* combine.c (reg_n_sets_max): New variable.
(can_change_dest_mode, reg_nonzero_bits_for_combine,
reg_num_sign_bit_copies_for_combine, get_last_value_validate,
get_last_value): Use REG_N_SETS only on pseudos < reg_n_sets_max.
(try_combine): Use INC_REG_N_SETS only on pseudos < reg_n_sets_max.
(rest_of_handle_combine): Initialize reg_n_sets_max.

From-SVN: r220368

PR preprocessor/64803 - __LINE__ inside macro is not constant

Consider the example code mentionned in this PR:

 $ cat -n test.c
      1 #define C(a, b) a ## b
      2 #define L(x) C(L, x)
      3 #define M(a) goto L(__LINE__); __LINE__; L(__LINE__):
      4 M(a /* --> this is the line of the expansion point of M.  */
      5   ); /* --> this is the line of the end of the invocation of M.  */
 $

"cc1 -quiet -E test.c" yields:

 goto L5; 5; L4:
    ;

Notice how we have a 'L4' there, where it should be L5.  That is the issue.

My understanding is that during the *second* expansion of __LINE__
(the one between the two L(__LINE__)), builtin_macro() is called by
enter_macro_context() with the location of the expansion point of M
(which is at line 4).  Then _cpp_builtin_macro_text() expands __LINE__
into the line number of the location of the last token that has been
lexed, which is the location of the closing parenthesis of the
invocation of M, at line 5.  So that invocation of __LINE__ is
expanded into 5.

Now let's see why the last invocation of __LINE__ is expanded into 4.

In builtin_macro(), we have this code at some point:

   /* Set pfile->cur_token as required by _cpp_lex_direct.  */
   pfile->cur_token = _cpp_temp_token (pfile);
   cpp_token *token = _cpp_lex_direct (pfile);
   /* We should point to the expansion point of the builtin macro.  */
   token->src_loc = loc;

The first two statements insert a new token in the stream of lexed
token and pfile->cur_token[-1], is the "new" last token that has been
lexed.  But the location of pfile->cur_token[-1] is the same location
as the location of the "previous" pfile->cur_token[-1], by courtesy of
_cpp_temp_token().  So normally, in subsequent invocations of
builtin_macro(), the location of pfile->cur_token[-1] should always be
the location of the closing parenthesis of the invocation of M at line
5.  Except that that code in master now has the statement
"token->src_loc = loc;" on the next line.  That statement actually
sets the location of pfile->cur_token[-1] to 'loc'.  Which is the
location of the expansion point of M, which is on line 4.

So in the subsequent call to builtin_macro() (for the last expansion
of __LINE__ in L(__LINE__)), for _cpp_builtin_macro_text(),
pfile->cur_token[-1].src_loc is going to have a line number of 4.

I think the core issue here is that the location that is passed to
builtin_macro() from enter_macro_context() is not correct when we are
in presence of a top-most function-like macro invocation; in that
case, that location should be the location of the closing parenthesis
of the macro invocation.  Otherwise, if we are in presence of a a
top-most object-like macro invocation then the location passed down
to builtin_macro should be the location of the expansion point of the
macro.

That way, in the particular case of the input code above, the location
received by builtin_macro() will always have line number 5.

Boostrapped and tested on x86_64-unknown-linux-gnu against trunk.

libcpp/ChangeLog:

* internal.h (cpp_reader::top_most_macro_node): New data member.
* macro.c (enter_macro_context): Pass the location of the end of
the top-most invocation of the function-like macro, or the
location of the expansion point of the top-most object-like macro.
(cpp_get_token_1): Store the top-most macro node in the new
pfile->top_most_macro_node data member.
(_cpp_pop_context): Clear the new cpp_reader::top_most_macro_node
data member.

gcc/testsuite/ChangeLog:

* gcc.dg/cpp/builtin-macro-1.c: New test case.

Signed-off-by: Dodji Seketeli <dodji@redhat.com>
From-SVN: r220367

re PR go/64836 (go.test/test/fixedbugs/issue4348.go FAILs)

PR go/64836
PR go/64838

compiler: Use int64_t for backend type size and alignment.

Fixes 32-bit host 64-bit target cross-compilation.

* go-gcc.cc (Gcc_backend::type_size): Change return type to
int64_t.
(Gcc_backend::type_alignment): Likewise.
(Gcc_backend::type_field_alignment): Likewise.
(Gcc_backend::type_field_offset): Likewise.
(Gcc_backend::implicit_variable): Change alignment parameter type
to int64_t.

From-SVN: r220364

re PR c++/64901 (overriding final function defined out of line does not lead to an error)

PR c++/64901
* decl.c (duplicate_decls): Also duplicate DECL_FINAL_P and
DECL_OVERRIDE_P.

From-SVN: r220363

Daily bump.

From-SVN: r220362

ipa-inline.c (early_inliner): Skip inlining only in always_inlined...

* ipa-inline.c (early_inliner): Skip inlining only in always_inlined;
if some always_inline was inlined, apply changes before inlining
heuristically.

* g++.dg/ipa/devirt-37.C: Disable early inlining.

From-SVN: r220359

c-parser.c (c_parser_declspecs): Call invoke_plugin_callbacks after processing enum declaration.

2015-02-02  Bruno Loff <bruno.loff@gmail.com>

* c-parser.c (c_parser_declspecs): Call invoke_plugin_callbacks after
processing enum declaration.

From-SVN: r220358

* tree.c (handle_abi_tag_attribute): Diagnose invalid arguments.

From-SVN: r220356

[ARM/AArch64][testsuite] Add vmax, vmin, vhadd, vhsub and vrhadd tests.

2015-02-02  Christophe Lyon  <christophe.lyon@linaro.org>

* gcc.target/aarch64/advsimd-intrinsics/arm-neon-ref.h
(_ARM_FPSRC): Add DN and AHP fields.
(clean_results): Force DN=1 on AArch64.
* gcc.target/aarch64/advsimd-intrinsics/binary_op_no64.inc: New file.
* gcc.target/aarch64/advsimd-intrinsics/vhadd.c: New file.
* gcc.target/aarch64/advsimd-intrinsics/vhsub.c: New file.
* gcc.target/aarch64/advsimd-intrinsics/vmax.c: New file.
* gcc.target/aarch64/advsimd-intrinsics/vmin.c: New file.
* gcc.target/aarch64/advsimd-intrinsics/vrhadd.c: New file.

From-SVN: r220353

MAINTAINERS (Various Maintainers: testsuite): Remove myself.

* MAINTAINERS (Various Maintainers: testsuite): Remove myself.
(Write After Approval): Update address.

From-SVN: r220352

PR jit/64810: fix for arm_option_override

gcc/ChangeLog:
PR jit/64810
* config/arm/arm.c (arm_option_override): Set
arm_selected_arch/cpu/tune to NULL on entry.

From-SVN: r220351

Regenerate .pot files.

gcc/po:
* gcc.pot: Regenerate.

libcpp/po:
* cpplib.pot: Regenerate.

From-SVN: r220349

re PR target/64231 (SIGSEGV building glibc on aarch64-linux-gnu from r217852)

2015-02-02  Tejas Belagod  <tejas.belagod@arm.com>
    Andrew Pinski  <pinskia@gcc.gnu.org>
    Jakub Jelinek  <jakub@gcc.gnu.org>

PR target/64231
* config/aarch64/aarch64.c (aarch64_classify_symbol): Fix large
integer typing for small model. Use IN_RANGE.

Co-Authored-By: Andrew Pinski <pinskia@gcc.gnu.org>
Co-Authored-By: Jakub Jelinek <jakub@gcc.gnu.org>
From-SVN: r220348

PR jit/64810: support DImode on arm

gcc/jit/ChangeLog:
PR jit/64810
* dummy-frontend.c (jit_langhook_type_for_mode): Support
TYPE_MODE (long_long_integer_type_node).

From-SVN: r220347

tree-ssa-ccp.c (valueize_op_1): Always allow valueizing default-defs.

2015-02-02  Richard Biener  <rguenther@suse.de>

* tree-ssa-ccp.c (valueize_op_1): Always allow valueizing default-defs.
* tree-vrp.c (vrp_valueize_1): Likewise.

From-SVN: r220346

trans-decl.c (gfc_get_symbol_decl): Removed duplicate code.

* trans-decl.c (gfc_get_symbol_decl): Removed duplicate code.
* trans-expr.c (gfc_conv_intrinsic_to_class): Fixed indentation.
Fixed datatype of charlen to be a 32-bit int.

From-SVN: r220345

rs6000.c (rs6000_call_aix): Use unspec rather than mem for toc_restore.

gcc/
* config/rs6000/rs6000.c (rs6000_call_aix): Use unspec rather
than mem for toc_restore.
* config/rs6000/rs6000.md (UNSPEC_TOCSLOT): Define.
(call_indirect_aix, call_value_indirect_aix): Adjust to suit.
(call_indirect_elfv2, call_value_indirect_elfv2): Likewise.
gcc/testsuite/
* gcc.target/powerpc/cprophard.c: New.

From-SVN: r220344

re PR libgomp/64635 (darwin produces libgomp-plugin-host_nonshm.1.dylib but tries to load libgomp-plugin-host_nonshm.so.1)

        PR libgomp/64635
        * configure.tgt (*-*-aix*): Use standard posix plugin-suffix.h.
        Link with -lpthread.
        * config/aix/plugin-suffix.h: Delete.

From-SVN: r220341

re PR target/64047 (ICE: Segmentation fault when compiling gcc.dg/torture/pr52429.c)

        PR target/64047
        * config/rs6000/rs6000.c (rs6000_set_current_function): Handle
        explicit default options.

From-SVN: r220340

Daily bump.

From-SVN: r220339

enum_9.f90: Use arm_eabi instead of arm*-*-linux*.

2015-02-01  Andreas Tobler  <andreast@gcc.gnu.org>

    * gfortran.dg/enum_9.f90: Use arm_eabi instead of arm*-*-linux*.
    * gfortran.dg/enum_10.f90: Likewise. Reorder dg-additional-sources.

From-SVN: r220330

re PR ipa/64872 (ICE: Segmentation fault during Chromium PGO build)

PR ipa/64872
* ipa-utils.c (ipa_merge_profiles): Add release argument.
* ipa-icf.c (sem_function::merge): Do not release body when merging.
* ipa-utils.h (ipa_merge_profiles): Update prototype.

From-SVN: r220329

re PR debug/64817 (compilation hangs at -O3 with -g enabled on x86_64-linux-gnu)

PR debug/64817
* cfgexpand.c (deep_ter_debug_map): New variable.
(avoid_deep_ter_for_debug): New function.
(expand_debug_expr): If TERed SSA_NAME is in
deep_ter_debug_map, use the corresponding DEBUG_EXPR_DECL
instead of trying to expand SSA_NAME's def stmt.
(expand_debug_locations): When expanding debug bind
of a DEBUG_EXPR_DECL to corresponding SSA_NAME,
temporarily remove the DEBUG_EXPR_DECL from deep_ter_debug_map's
value.
(pass_expand::execute): Call avoid_deep_ter_for_debug on
all debug bind stmts.  Delete deep_ter_debug_map after
expand_debug_location if non-NULL and clear it.

* gcc.dg/pr64817-1.c: New test.
* gcc.dg/pr64817-2.c: New test.

From-SVN: r220320

* src/c++11/futex.cc: Do not define for gthr-single.h targets.

From-SVN: r220319

re PR libstdc++/64883 (FAIL: 17_intro/headers/c++*/all_attributes.cc (test for excess errors) on x86_64-apple-darwin10)

PR libstdc++/64883
* include/c_global/cstdio (gets): Use __deprecated__ attribute instead
of deprecated.
* include/c_std/cstdio (gets): Likewise.
* testsuite/17_intro/headers/c++1998/all_attributes.cc: Avoid clashing
with attributes used in darwin headers.

From-SVN: r220318

re PR target/64851 ([SH] Add atomic not)

gcc/
PR target/64851
* config/sh/sync.md (atomic_fetch_notsi_hard,
atomic_fetch_not<mode>_hard, atomic_fetch_not<mode>_soft_gusa,
atomic_fetch_not<mode>_soft_tcb, atomic_fetch_not<mode>_soft_imask,
atomic_not_fetchsi_hard, atomic_not_fetch<mode>_hard,
atomic_not_fetch<mode>_soft_gusa, atomic_not_fetch<mode>_soft_tcb,
atomic_not_fetch<mode>_soft_imask): New insns.

gcc/testsuite/
PR target/64851
* gcc.target/sh/pr64851-0.h: New
* gcc.target/sh/pr64851-1.c: New
* gcc.target/sh/pr64851-2.c: New
* gcc.target/sh/pr64851-3.c: New
* gcc.target/sh/pr64851-4.c: New

From-SVN: r220317

Fix scheduling undeterminism from sorting with DEBUG_INSNs

* haifa-sched.c (INSN_RFS_DEBUG_ORIG_ORDER): New access macro.
(rank_for_schedule_debug): Split from ...
(rank_for_schedule): ... this.
(ready_sort): Sort DEBUG_INSNs separately from normal INSNs.
* sched-int.h (struct _haifa_insn_data): New field rfs_debug_orig_order.

From-SVN: r220316

md.texi (Machine Constraints): Alphabetize table by target.

2015-01-31  Sandra Loosemore  <sandra@codesourcery.com>

gcc/
* doc/md.texi (Machine Constraints): Alphabetize table by target.
* doc/extend.texi (x86 Variable Attributes): Move section to
correct alphabetization after renaming.
(x86 Type Attributes): Likewise.
(Target Builtins): Re-alphabetize menu.
(x86 Built-in Functions): Move section to correct alphabetization
after renaming.
(x86 transactional memory intrinsics): Likewise.
* doc/invoke.texi (Option Summary): Re-alphabetize x86 Options
and x86 Windows Options in table and menu.
(x86 Options): Move section to correct alphabetization after
renaming.
(x86 Windows Options): Likewise.

From-SVN: r220315

Correct Changelog for r220309

From-SVN: r220314

Always pass 0 or option number to gfc_warning*.

Similar to the issue with fatal_error that I fixed in
<https://gcc.gnu.org/ml/gcc-patches/2015-01/msg02690.html>, the
overloads of gfc_warning and gfc_warning_now (with and without a first
argument for an option number) also break gcc.pot regeneration because
xgettext expects the translated string argument to be in a fixed
position for a given function name.  This patch applies the
corresponding fix of always passing a first argument (option number or
0), just like the core diagnostic functions warning and warning_at,
and removing the problem overloads without it.

Bootstrapped with no regressions on x86_64-unknown-linux-gnu.

* error.c (gfc_warning (const char *, ...), gfc_warning_now (const
char *, ...)): Remove functions.
* gfortran.h (gfc_warning (const char *, ...), gfc_warning_now
(const char *, ...)): Remove declarations.
* arith.c, check.c, data.c, decl.c, frontend-passes.c,
interface.c, intrinsic.c, io.c, matchexp.c, module.c, openmp.c,
options.c, parse.c, primary.c, resolve.c, scanner.c, symbol.c,
trans-common.c, trans-const.c, trans-stmt.c: All callers of
gfc_warning and gfc_warning_now changed to pass 0 or option number
as first argument.

From-SVN: r220313

Daily bump.

From-SVN: r220312

extend.texi: Use "x86"...

2015-01-31  Sandra Loosemore  <sandra@codesourcery.com>

gcc/
* doc/extend.texi: Use "x86", "x86-32", and "x86-64" as the
preferred names of the architecture and its 32- and 64-bit
variants.
* doc/invoke.texi: Likewise.
* doc/md.texi: Likewise.

From-SVN: r220309

linux-atomic.c (__kernel_cmpxchg2): Change declaration of oldval and newval to const void *.

* config/pa/linux-atomic.c (__kernel_cmpxchg2): Change declaration of
oldval and newval to const void *.  Fix typo.
(FETCH_AND_OP_2): Use __atomic_load_n to load value.
(FETCH_AND_OP_WORD): Likewise.
(OP_AND_FETCH_WORD): Likewise.
(COMPARE_AND_SWAP_2): Likewise.
(__sync_val_compare_and_swap_4): Likewise.
(__sync_lock_test_and_set_4): Likewise.
(SYNC_LOCK_RELEASE_2): Likewise.
Remove support for long long atomic operations.

From-SVN: r220307

re PR target/64882 (ICE on valid code at -O3 with -g enabled in simplify_subreg, at simplify-rtx.c:5681)

2015-01-31  Uros Bizjak  <ubizjak@gmail.com>

        PR target/64882
* config/i386/predicates.md (address_no_seg_operand): Reject
non-CONST_INT_P operands in invalid mode.

2015-01-31  Uros Bizjak  <ubizjak@gmail.com>

* config/i386/i386.md (*prefetch_prefetchw1): Remove mode of
address_operand 0.  Rename from *prefetch_prefetchwt1_<mode>.
* config/i386/predicates.md (address_no_seg_operand): Call
address_operand with VOIDmode.
(vsib_address_operand): Ditto.
(address_mpx_no_base_operand): Ditto.
(address_mpx_no_index_operand): Ditto.

testsuite/ChangeLog:

2015-01-31  Uros Bizjak  <ubizjak@gmail.com>

        PR target/64882
* gcc.dg/torture/pr64882.c: New test.

From-SVN: r220306

re PR target/64159 (FAIL: gcc.dg/tree-ssa/ssa-dom-cse-2.c scan-tree-dump optimized "return 28;")

        PR target/64159
        * gcc.dg/tree-ssa/ssa-dom-cse-2.c: Add XFAIL for powerpc*-*-* and
        sparc*-*-*.

From-SVN: r220305

* fixincl.c (fix_applies) <case TT_CKSUM:>: Change C++ style comments.

From-SVN: r220304

pr64146.C (dg-final): Cleanup icf ipa dump.

* g++.dg/ipa/pr64146.C (dg-final): Cleanup icf ipa dump.
* gcc.target/i386/chkp-builtins-1.c (dg-final): Cleanup chkp tree dump.
* gcc.target/i386/chkp-builtins-2.c (dg-final): Ditto.
* gcc.target/i386/chkp-builtins-3.c (dg-final): Ditto.
* gcc.target/i386/chkp-builtins-4.c (dg-final): Ditto.
* gcc.target/i386/chkp-const-check-1.c (dg-final): Cleanup chkopt
tree dump.
* gcc.target/i386/chkp-lifetime-1.c (dg-final): Ditto.
* gcc.target/i386/chkp-remove-bndint-1.c (dg-final): Cleanup optimized
tree dump.
* gcc.target/i386/chkp-remove-bndint-2.c (dg-final): Ditto.
* gfortran.dg/goacc/private-1.f95 (dg-final): Cleanup omplower
tree dump.

From-SVN: r220303

compiler: Treat map index expressions as lvalues in range expressions.

Fixes golang/go#9695.

From-SVN: r220301

Daily bump.

From-SVN: r220300

re PR rtl-optimization/64688 (internal compiler error: Max. number of generated reload insns per insn is achieved (90))

2015-01-30  Vladimir Makarov  <vmakarov@redhat.com>

PR target/64688
* lra-constraints.c (original_subreg_reg_mode): New.
(simplify_operand_subreg): Try to simplify subreg of const.  Use
original_subreg_reg_mode for it.
(swap_operands): Update original_subreg_reg_mode.
(curr_insn_transform): Set up original_subreg_reg_mode.

2015-01-30  Vladimir Makarov  <vmakarov@redhat.com>

PR target/64688
* g++.dg/pr64688-2.C: New.

From-SVN: r220297

fuse-caller-save-rec.c: Require nonpic target.

* gcc.target/i386/fuse-caller-save-rec.c: Require nonpic target.
* gcc.target/i386/fuse-caller-save-xmm.c: Ditto.
* gcc.target/i386/fuse-caller-save.c: Ditto.
* gcc.target/i386/headmerge-1.c: Ditto.
* gcc.target/i386/headmerge-2.c: Ditto.
* gcc.target/i386/sibcall-6.c: Ditto.

From-SVN: r220296

re PR target/64617 (ICE: Max. number of generated reload insns per insn is achieved (90) with -ftree-vectorize -mavx512bw -march=slm)

2015-01-30  Vladimir Makarov  <vmakarov@redhat.com>

PR target/64617
* lra-constraints.c (prohibited_class_reg_set_mode_p): New
function.
(process_alt_operands): Use it.
(curr_insn_transform): Check the optional reload pseudo class is
ok for the mode.

2015-01-30  Vladimir Makarov  <vmakarov@redhat.com>

PR target/64617
* gcc.target/i386/pr64617.c: New test.

From-SVN: r220294

Always pass explicit location to fatal_error.

The patch <https://gcc.gnu.org/ml/gcc-patches/2014-11/msg00698.html>
adding an overload for fatal_error that passes an explicit location
broke gcc.pot regeneration because xgettext cannot handle function
overloads with the diagnostic string argument in different positions.

As the desired direction is for all diagnostics to have explicit
locations, this patch addresses the regression by removing the version
of fatal_error that does not pass a location, passing explicit
input_location everywhere (in the hope that those will incrementally
be changed to other locations, much as with the addition of a location
argument to pedwarn some time ago - a lot of cases aren't meaningfully
associated with a source file so UNKNOWN_LOCATION may be better).

Note that this patch does not attempt to fix any existing issues with
these diagnostics (such as wrongly starting with capital letters or
ending with '.' or '\n'); it just adds the input_location argument.

Bootstrapped with no regressions for x86_64-unknown-linux-gnu (Go
excluded).

gcc:
* diagnostic.c (fatal_error (const char *, ...)): Remove function.
* diagnostic-core.h (fatal_error (const char *, ...)): Remove
prototype.
* toplev.h (init_asm_output): Update comment on use of
UNKNOWN_LOCATION with fatal_error.
* cgraph.c, collect-utils.c, collect2.c, config/arc/arc.c,
config/arc/arc.md, config/avr/avr.c, config/c6x/c6x.h,
config/darwin.c, config/host-darwin.c, config/i386/host-cygwin.c,
config/i386/intelmic-mkoffload.c, config/nios2/nios2.c,
config/nvptx/mkoffload.c, config/nvptx/nvptx.h,
config/rs6000/host-darwin.c, config/rs6000/rs6000.c,
config/s390/s390.c, gcc.c, gcov-io.h, gcov-tool.c, ggc-common.c,
ggc-page.c, graph.c, ipa-inline-analysis.c, ipa-reference.c,
lto-cgraph.c, lto-section-in.c, lto-streamer-in.c, lto-streamer.c,
lto-wrapper.c, objc/objc-act.c, opts.c, passes.c, plugin.c,
tlink.c, toplev.c, tree-streamer-in.c, varpool.c: All callers of
fatal_error changed to pass input_location as first argument.

gcc/c-family:
* c-opts.c, c-pch.c, cppspec.c: All callers of fatal_error changed
to pass input_location as first argument.

gcc/cp:
* class.c, except.c, parser.c, pt.c: All callers of fatal_error
changed to pass input_location as first argument.

gcc/fortran:
* f95-lang.c, gfortranspec.c, trans-const.c, trans-expr.c: All
callers of fatal_error changed to pass input_location as first
argument.

gcc/java:
* class.c, expr.c, jcf-parse.c, jvspec.c: All callers of
fatal_error changed to pass input_location as first argument.

gcc/lto:
* lto-object.c, lto-symtab.c, lto.c: All callers of fatal_error
changed to pass input_location as first argument.

libcc1:
* plugin.cc: All callers of fatal_error changed to pass
input_location as first argument.

From-SVN: r220293

compiler: Pass location to fatal_error.

The GCC function is changing.

The error functions need to move to a compiler-indepent
approach so that changes to the GCC functions are not a
concern here.

From-SVN: r220292