[Patch] Teach genrecog/genoutput that scratch registers require write  constraint modifiers

gcc/

* doc/md.texi (Modifiers): Consistently use "read/write"
nomenclature rather than "input/output".
* genrecog.c (constraints_supported_in_insn_p): New.
(validate_pattern): If needed, also check constraints on
MATCH_SCRATCH operands.
* genoutput.c (validate_insn_alternatives): Catch earlyclobber
operands with no '=' or '+' modifier.

From-SVN: r215388

re PR fortran/63152 (needless initialization of local pointer arrays.)

2014-09-19  Joost VandeVondele  <vondele@gcc.gnu.org>

PR fortran/63152
* trans-array.c (gfc_trans_deferred_array): Only nullify allocatables.

2014-09-19  Joost VandeVondele  <vondele@gcc.gnu.org>

PR fortran/63152
* gfortran.dg/auto_char_dummy_array_1.f90: Fix undefined behavior.
* gfortran.dg/pr63152.f90: New test.

From-SVN: r215387

re PR c++/58328 ([C++11] bogus: error: constructor required before non-static data member for)

2014-09-19  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/58328
* g++.dg/cpp0x/nsdmi10.C: New.

From-SVN: r215386

[PATCH AArch64]: Add constraint letter for stack_protect_test  pattern

gcc/

* config/aarch64/aarch64.md (stack_protect_test_<mode>): Mark
scratch register as an output to placate register renaming.

gcc/testsuite/

* gcc.dg/ssp-3.c: New.
* gcc.dg/ssp-4.c: Likewise.

From-SVN: r215385

s390.c (s390_emit_epilogue): Remove bogus assignment.

2014-09-19  Andreas Krebbel  <Andreas.Krebbel@de.ibm.com>

* config/s390/s390.c (s390_emit_epilogue): Remove bogus
assignment.

From-SVN: r215384

s390.md ("trunctdsd2", [...]): New expanders.

2014-09-19  Andreas Krebbel  <Andreas.Krebbel@de.ibm.com>

* config/s390/s390.md ("trunctdsd2", "extendsdtd2"): New
expanders.

2014-09-19  Andreas Krebbel  <Andreas.Krebbel@de.ibm.com>

* gcc.target/s390/dfp-conv1.c: New testcase.

From-SVN: r215382

re PR target/62662 (Miscompilation of Qt on s390x)

2014-09-19  Andreas Krebbel  <Andreas.Krebbel@de.ibm.com>

PR target/62662
* config/s390/s390.c (s390_emit_epilogue): When doing the return
address load optimization force s390_optimize_prologue to leave it
that way.  Only do the optimization if we already decided to push
r14 into a stack slot.

From-SVN: r215381

asan.c (build_check_stmt): Alignment arg was added.

gcc/ChangeLog:

2014-09-19  Marat Zakirov  <m.zakirov@samsung.com>

* asan.c (build_check_stmt): Alignment arg was added.
(asan_expand_check_ifn): Optimization for alignment >= 8.

gcc/testsuite/ChangeLog:

2014-09-19  Marat Zakirov  <m.zakirov@samsung.com>

* c-c++-common/asan/red-align-1.c: New test.
* c-c++-common/asan/red-align-2.c: New test.

From-SVN: r215380

vxworksae.h: Remove obsolete definitions.

2014-09-19  Olivier Hainque  <hainque@adacore.com>

        * config/i386/vxworksae.h: Remove obsolete definitions.
        (STACK_CHECK_PROTECT): Define.
        * config/i386/vx-common.h: Remove.  Merge contents within
        config/i386/vxworks.h.
        * config.gcc (i?86-vxworks*): Use i386/vxworks.h instead of
        i386/vx-common.h.

From-SVN: r215378

config.gcc (powerpc-wrs-vxworksmils): New configuration.

2014-09-18  Olivier Hainque  <hainque@adacore.com>

gcc/
* config.gcc (powerpc-wrs-vxworksmils): New configuration.
* config/rs6000/t-vxworksmils: New file.
* config/rs6000/vxworksmils.h: New file.

libgcc/
* config.host (powerpc-wrs-vxworksmils): New configuration,
same as vxworksae.

contrib/
* config-list.mk (LIST): Add powerpc-wrs-vxworksmils.

From-SVN: r215377

varasm.c (default_section_type_flags): Flag .persistent.bss sections as SECTION_BSS.

2014-09-19  Olivier Hainque  <hainque@adacore.com>

        * varasm.c (default_section_type_flags): Flag .persistent.bss
        sections as SECTION_BSS.

From-SVN: r215376

rl78.c (rl78_expand_epilogue): Generate a USE of the pop'ed registers so that DCE does not eliminate them.

* config/rl78/rl78.c (rl78_expand_epilogue): Generate a USE of the
pop'ed registers so that DCE does not eliminate them.

From-SVN: r215375

re PR ipa/63298 (internal compiler error: in types_same_for_odr, at ipa-devirt.c:449 with LTO)

PR lto/63298
* ipa-devirt.c (odr_subtypes_equivalent_p): Fix thinko in a condition.

From-SVN: r215374

Daily bump.

From-SVN: r215373

Update gcc.dg/pr61053.c for x32

* gcc.dg/pr61053.c: Updated for x32.

From-SVN: r215369

Remove LIBGCC2_TF_CEXT target macro.

This patch removes the (undocumented) LIBGCC2_TF_CEXT target macro,
replacing it by -fbuilding-libgcc predefines (and thereby gets rid of
another LIBGCC2_LONG_DOUBLE_TYPE_SIZE conditional, though some more
patches are needed before that target macro can be eliminated).  This
macro indicated the suffix used on __builtin_huge_val,
__builtin_copysign, __builtin_fabs built-in function names to produce
the names for a given floating-point mode.

Predefines are added for all floating-point modes supported for
libgcc, not just TFmode.  These are fully accurate for modes
corresponding to float, double and long double.  For other modes, the
suffix for *constants* is determined by the targetm.c.mode_for_suffix
hook (the limit to two possible suffixes 'w' and 'q' being hardcoded
in various places).  This is in fact the suffix for built-in functions
as well where such functions exist.

* For i386, the *q functions always exist (whether or not TFmode is
  used for long double).  The *w functions never exist (but this
  doesn't matter for libgcc, since no i386 configuration treats XFmode
  as a supported scalar mode if long double is TFmode; if __float80
  were to be supported for 64-bit Android, properly such functions
  ought to be added).

* For ia64, the *q functions exist for non-HP-UX (under HP-UX, long
  double is TFmode, so they aren't needed).  The *w functions never
  exist.  This is an issue for this libgcc code for the XFmode complex
  functions in libgcc on HP-UX; as I understand it, right now those
  will accidentally be using TFmode versions of those three functions,
  so involving unnecessary conversions, while the sanity check on CEXT
  accidentally passes because all it tests is the sizes of the types.

Because of the lack of 'w' functions, the patch uses 'l' when the
constant suffix is 'w', matching what the existing libgcc code would
do for IA64 HP-UX in that case.

Ideally there would be generic code to create such built-in functions
for all supported floating-point types.  That may be something to
consider if support for TS 18661-3 (standard bindings for IEEE
754-2008, defining names such as _Float128, and function names such as
copysignf128) is added in future.

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

gcc:
* system.h (LIBGCC2_TF_CEXT): Poison.
* config/i386/cygming.h (LIBGCC2_TF_CEXT): Remove.
* config/i386/darwin.h (LIBGCC2_TF_CEXT): Likewise.
* config/i386/dragonfly.h (LIBGCC2_TF_CEXT): Likewise.
* config/i386/freebsd.h (LIBGCC2_TF_CEXT): Likewise.
* config/i386/gnu-user-common.h (LIBGCC2_TF_CEXT): Likewise.
* config/i386/openbsdelf.h (LIBGCC2_TF_CEXT): Likewise.
* config/i386/sol2.h (LIBGCC2_TF_CEXT): Likewise.
* config/ia64/ia64.h (LIBGCC2_TF_CEXT): Likewise.
* config/ia64/linux.h (LIBGCC2_TF_CEXT): Likewise.

gcc/c-family:
* c-cppbuiltin.c (c_cpp_builtins): Define __LIBGCC_*_FUNC_EXT__
for supported floating-point modes.

libgcc:
* libgcc2.c (CEXT): Define using __LIBGCC_*_FUNC_EXT__.

From-SVN: r215368

Add header guard to several header files.

2014-09-19  Kito Cheng  <kito@0xlab.org>

* except.h: Fix header guard.
* addresses.h: Add missing header guard.
* cfghooks.h: Likewise.
* collect-utils.h: Likewise.
* collect2-aix.h: Likewise.
* conditions.h: Likewise.
* cselib.h: Likewise.
* dwarf2asm.h: Likewise.
* graphds.h: Likewise.
* graphite-scop-detection.h: Likewise.
* gsyms.h: Likewise.
* hw-doloop.h: Likewise.
* incpath.h: Likewise.
* ipa-inline.h: Likewise.
* ipa-ref.h: Likewise.
* ira-int.h: Likewise.
* ira.h: Likewise.
* lra-int.h: Likewise.
* lra.h: Likewise.
* lto-section-names.h: Likewise.
* read-md.h: Likewise.
* reload.h: Likewise.
* rtl-error.h: Likewise.
* sdbout.h: Likewise.
* targhooks.h: Likewise.
* tree-affine.h: Likewise.
* xcoff.h: Likewise.
* xcoffout.h: Likewise.

From-SVN: r215365

re PR debug/63285 (-fcompare-debug scheduler related failure)

2014-09-18  Vladimir Makarov  <vmakarov@redhat.com>

PR debug/63285
* haifa-sched.c (schedule_block): Advance cycle at the end of BB
if advance != 0.

2014-09-18  Vladimir Makarov  <vmakarov@redhat.com>

PR debug/63285
* gcc.target/i386/pr63285.c: New test.

From-SVN: r215364

config-list.mk (LIST): Add v850-rtems.

2014-09-18  Joel Sherrill <joel.sherrill@oarcorp.com>

* config-list.mk (LIST): Add v850-rtems.

From-SVN: r215362

re PR c++/63248 (Crash when OpenMP target's array section handling is used with templates)

PR c++/63248
* semantics.c (finish_omp_clauses): Don't call cp_omp_mappable_type
on type of type dependent expressions, and don't call it if
handle_omp_array_sections has kept TREE_LIST because something
was type dependent.
* pt.c (tsubst_expr) <case OMP_TARGET, case OMP_TARGET_DATA>:
Use keep_next_level, begin_omp_structured_block and
finish_omp_structured_block instead of push_stmt_list and
pop_stmt_list.
libgomp/
* testsuite/libgomp.c++/pr63248.C: New test.

From-SVN: r215359

re PR target/61360 (ICE: in lra_update_insn_recog_data, at lra.c:1363 with -mtune=bdver4)

2014-09-18  Vladimir Makarov  <vmakarov@redhat.com>

PR target/61360
* lra.c (lra): Call recog_init.

2014-09-18  Vladimir Makarov  <vmakarov@redhat.com>

PR target/61360
* gcc.target/i386/pr61360.c: New.

From-SVN: r215358

config-list.mk (LIST): Add arm-rtems.

2014-09-18  Sebastian Huber  <sebastian.huber@embedded-brains.de>

* config-list.mk (LIST): Add arm-rtems.
Add nios2-rtems.  Remove extra option from powerpc-rtems.

From-SVN: r215357

re PR c++/62017 (AddressSanitizer reports *-buffer-overflow in destructor when multiple virtual inheritance is used)

PR c++/62017
* asan.c (transform_statements): Don't instrument clobber statements.

* g++.dg/asan/pr62017.C: New test.

From-SVN: r215354

re PR testsuite/63292 (ASan reports global-buffer-overflow in gcc.dg/vect/pr59594.c.)

PR testsuite/63292
* gcc.dg/vect/pr59594.c (b): Increase size to N + 2 elements.

From-SVN: r215352

re PR c++/62232 (-Wnon-virtual-dtor shouldn't warn on final classes)

/cp
2014-09-18  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/62232
* class.c (finish_struct_1): Do not -Wnon-virtual-dtor warn
for final class types.

/testsuite
2014-09-18  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/62232
* g++.dg/cpp0x/Wdtor1.C: New.

From-SVN: r215351

dyncast.cc (__dynamic_cast): Handle mid-destruction dynamic_cast more gracefully.

* libsupc++/dyncast.cc (__dynamic_cast): Handle mid-destruction
dynamic_cast more gracefully.

From-SVN: r215350

Fix i386 FP_TRAPPING_EXCEPTIONS.

The i386 sfp-machine.h defines FP_TRAPPING_EXCEPTIONS in a way that is
always wrong: it treats a set bit as indicating the exception is
trapping, when actually a set bit (both for 387 and SSE floating
point) indicates it is masked, and a clear bit indicates it is
trapping.  This patch fixes this bug.

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

libgcc:
* config/i386/sfp-machine.h (FP_TRAPPING_EXCEPTIONS): Treat clear
bits not set bits as indicating trapping exceptions.

gcc/testsuite:
* gcc.dg/torture/float128-exact-underflow.c: New test.

From-SVN: r215347

[ARM] Fix insn type of movmisalign neon load pattern.

* config/arm/neon.md (*movmisalign<mode>_neon_load): Change type
to neon_load1_1reg<q>.

From-SVN: r215346

re PR c++/61745 (template friend for dyadic operator- is only accepted if the monadic operator- follows it)

2014-09-18  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/61745
* g++.dg/template/pr61745.C: New.

From-SVN: r215345

Daily bump.

From-SVN: r215344

Fix indentation

From-SVN: r215340

re PR c/61854 (Warning single-line comment for -std=c89?)

PR c/61854
libcpp/
* init.c (struct lang_flags): Remove cplusplus_comments.
(cpp_set_lang): Likewise.
(post_options): Likewise.
* lex.c (_cpp_lex_direct): Disallow C++ style comments in C90/C94.
testsuite/
* gcc.dg/cpp/pr61854-1.c: New test.
* gcc.dg/cpp/pr61854-2.c: New test.
* gcc.dg/cpp/pr61854-3.c: New test.
* gcc.dg/cpp/pr61854-3.h: New test.
* gcc.dg/cpp/pr61854-4.c: New test.
* gcc.dg/cpp/pr61854-5.c: New test.
* gcc.dg/cpp/pr61854-6.c: New test.
* gcc.dg/cpp/pr61854-7.c: New test.
* gcc.dg/cpp/pr61854-c90.c: New test.
* gcc.dg/cpp/pr61854-c94.c: New test.

From-SVN: r215339

PR libfortran/62768 Use gfc_unit.filename also when HAVE_TTYNAME{_R} is not defined.

2014-09-18  Janne Blomqvist  <jb@gcc.gnu.org>

PR libfortran/62768
* io/inquire.c (inquire_via_unit): Use gfc_unit.filename also when
HAVE_TTYNAME{_R} is not defined.

From-SVN: r215338

Makefile.in (LIBGCOV_INTERFACE): Add _gcov_dump from ...

* Makefile.in (LIBGCOV_INTERFACE): Add _gcov_dump from ...
(LIBGCOV_DRIVER): ... here.
* libgcov-driver.c (gcov_master): New.
(gcov_exit): Remove from master chain.
(__gcov_init): Add to master chain if version compatible.  Don't
clear the version.
* libgcov_interface (__gcov_flust): Call gcov_dump_int.
(gcov_reset_int): Clear master chain, if compatible.
(gcov_dump_int): New internal interface.  Dump master chain, if
compatible.
(gcov_dump): Alias for gcov_dump_int.
* libgcov.h (struct gcov_root): Add next and prev fields.
(struct gcov_master): New struct.
(__gcov_master): New.
(gcov_dump_int): Declare.

From-SVN: r215337

re PR debug/63284 (-fcompare-debug issue due to redirection to __builtin_unreachable ())

PR debug/63284
* tree-cfgcleanup.c (fixup_noreturn_call): Don't split block
if there are only debug stmts after the noreturn call, instead
remove the debug stmts.

* gcc.dg/pr63284.c: New test.

From-SVN: r215331

* ada/acats/run_all.sh: Redirect mkdir stderr to /dev/null.

From-SVN: r215330

Minor cleanup, don't run tests for which compilation is enough.

* g++.dg/g++.dg/cpp1y/var-templ1.C: Change return to
static_assert, compile instead of running.
* g++.dg/g++.dg/cpp1y/var-templ3.C: Likewise.
* g++.dg/g++.dg/cpp1y/var-templ4.C: Likewise.
* g++.dg/g++.dg/cpp1y/var-templ5.C: Likewise.

From-SVN: r215329

ipa-devirt.c (type_pair, [...]): New types.

* ipa-devirt.c (type_pair, default_hashset_traits): New types.
(odr_types_equivalent_p): Use pair hash.
(odr_subtypes_equivalent_p): Likewise, do structural compare
on ODR types that may be mismatched.
(warn_odr): Support warning when only one field is given.
(odr_types_equivalent_p): Strenghten comparsions made;
support VOIDtype.
(add_type_duplicate): Update VISITED hash set.

From-SVN: r215328

re PR c++/63241 (Internal error in gimplify_init_constructor when using constexr and multidimensional arrays)

2014-09-17  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/63241
* g++.dg/cpp0x/constexpr-63241.C: New.

From-SVN: r215326

config.gcc (*-*-rtems*): Default to 'rtems' thread model.

2014-09-17  Sebastian Huber  <sebastian.huber@embedded-brains.de>

* config.gcc (*-*-rtems*): Default to 'rtems' thread model.
Enable selection of 'posix' or no thread model.

From-SVN: r215324

config.host (x86_64-*-mingw*): Add i386/t-cygming to tmake_file and crtbegin.o + crtend.o to extra_parts.

2014-09-17  Olivier Hainque  <hainque@adacore.com>

        * config.host (x86_64-*-mingw*): Add i386/t-cygming to tmake_file
        and crtbegin.o + crtend.o to extra_parts.

From-SVN: r215323

Disallow -mfpu=neon for unsuitable architectures.

2014-09-17  Andrew Stubbs  <ams@codesourcery.com>

* config/arm/arm.c (arm_option_override): Reject -mfpu=neon
when architecture is older than ARMv7.

From-SVN: r215321

Daily bump.

From-SVN: r215313

re PR target/61853 (ICE: SIGSEGV in store_field)

PR target/61853
* config/pa/pa.c (pa_function_value): Directly handle aggregates
that fit exactly in a word or double word.

From-SVN: r215309

PR libfortran/62768 Handle filenames with embedded null characters.

testsuite ChangeLog:

2014-09-17  Janne Blomqvist  <jb@gcc.gnu.org>

PR libfortran/62768
* gfortran.dg/filename_null.f90: New test.

libgfortran ChangeLog:

2014-09-17  Janne Blomqvist  <jb@gcc.gnu.org>

PR libfortran/62768
* io/io.h (gfc_unit): Store C string for the filename.
* io/close.c (st_close): Use gfc_unit.filename.
* io/inquire.c (inquire_via_unit): Likewise.
* io/open.c (new_unit): Likewise.
(already_open): Likewise, unlink file before freeing filename.
* io/unit.c (init_units): Likewise.
(close_unit_1): Likewise.
(filename_from_unit): Likewise.
* io/unix.c (compare_file_filename): Likewise.
(find_file0): Likewise.
(delete_file): Likewise.

From-SVN: r215307

Fix a typo in gcc.dg/pg-override.c

* gcc.dg/pg-override.c: Fix a typo.

From-SVN: r215304

Fix a typo in gcc.dg/pg.c

* gcc.dg/pg.c: Fix a typo.

From-SVN: r215301

re PR testsuite/63258 (FAIL: gcc.dg/vect/costmodel/ppc/costmodel-slp-33.c scan-tree-dump-times vect "vectorization not profitable" 1)

2014-09-16  Richard Biener  <rguenther@suse.de>

PR testsuite/63258
* gcc.dg/vect/costmodel/ppc/costmodel-slp-33.c: Remove scan
for "vectorization not profitable".

From-SVN: r215299

driver-i386.c (host_detect_local_cpu): Detect lack of zmm/k regs support.

gcc/
2014-09-16  Ilya Tocar  <ilya.tocar@intel.com>

        * config/i386/driver-i386.c (host_detect_local_cpu): Detect lack of
        zmm/k regs support.

testsuite/
2014-09-16  Ilya Tocar  <ilya.tocar@intel.com>

        * gcc.target/i386/avx512f-os-support.h: Remove magic number.

From-SVN: r215297

AVX-512. Extend extract insn patterns.

gcc/
* config/i386/i386.c
(ix86_expand_vector_extract): Handle V32HI and V64QI modes.
* config/i386/sse.md
(define_mode_iterator VI48F_256): New.
(define_mode_attr extract_type): Ditto.
(define_mode_attr extract_suf): Ditto.
(define_mode_iterator AVX512_VEC): Ditto.
(define_expand
"<extract_type>_vextract<shuffletype><extract_suf>_mask"): Use
AVX512_VEC.
(define_insn "avx512dq_vextract<shuffletype>64x2_1_maskm"): New.
(define_insn
"<mask_codefor>avx512dq_vextract<shuffletype>64x2_1<mask_name>"):
Ditto.
(define_mode_attr extract_type_2): Ditto.
(define_mode_attr extract_suf_2): Ditto.
(define_mode_iterator AVX512_VEC_2): Ditto.
(define_expand
"<extract_type_2>_vextract<shuffletype><extract_suf_2>_mask"): Use
AVX512_VEC_2 mode iterator.
(define_insn "vec_extract_hi_<mode>_maskm"): Ditto.
(define_expand "avx512vl_vextractf128<mode>"): Ditto.
(define_insn_and_split "vec_extract_lo_<mode>"): Delete.
(define_insn "vec_extract_lo_<mode><mask_name>"): New.
(define_split for V16FI mode): Ditto.
(define_insn_and_split "vec_extract_lo_<mode>"): Delete.
(define_insn "vec_extract_lo_<mode><mask_name>"): New.
(define_split for VI8F_256 mode): Ditto.
(define_insn "vec_extract_hi_<mode><mask_name>"): Add masking.
(define_insn_and_split "vec_extract_lo_<mode>"): Delete.
(define_insn "vec_extract_lo_<mode><mask_name>"): New.
(define_split for VI4F_256 mode): Ditto.
(define_insn "vec_extract_lo_<mode>_maskm"): Ditto.
(define_insn "vec_extract_hi_<mode>_maskm"): Ditto.
(define_insn "vec_extract_hi_<mode><mask_name>"): Add masking.
(define_mode_iterator VEC_EXTRACT_MODE): Add V64QI and V32HI modes.
(define_insn "vcvtph2ps<mask_name>"): Fix pattern condition.
(define_insn "avx512f_vextract<shuffletype>32x4_1_maskm"): Ditto.
(define_insn "<mask_codefor>avx512f_vextract<shuffletype>32x4_1<mask_name>"):
Update `type' attribute, remove explicit `memory' attribute calculation.

Co-Authored-By: Andrey Turetskiy <andrey.turetskiy@intel.com>
Co-Authored-By: Anna Tikhonova <anna.tikhonova@intel.com>
Co-Authored-By: Ilya Tocar <ilya.tocar@intel.com>
Co-Authored-By: Ilya Verbin <ilya.verbin@intel.com>
Co-Authored-By: Kirill Yukhin <kirill.yukhin@intel.com>
Co-Authored-By: Maxim Kuznetsov <maxim.kuznetsov@intel.com>
Co-Authored-By: Michael Zolotukhin <michael.v.zolotukhin@intel.com>
From-SVN: r215296

re PR fortran/56408 (Fix dependency handling of testsuite/gfortran.dg)

PR fortran/56408
* gfortran.dg/dg.exp (dg-compile-aux-modules): Workaround
missing nexted dg-test call support in dejaGNU 1.4.4.

From-SVN: r215293

Clean up useless initialization for IRA if using LRA.

gcc/
* ira.c (ira): Don't initialize ira_spilled_reg_stack_slots and
ira_spilled_reg_stack_slots_num if using lra.
(do_reload): Remove release ira_spilled_reg_stack_slots part.
* ira-color.c (ira_sort_regnos_for_alter_reg): Add assertion to
make sure not using lra.
(ira_reuse_stack_slot): Likewise.
(ira_mark_new_stack_slot): Likewise.

From-SVN: r215286

Add some more test cases for fentry and pg

Test fentry and no_instrument_function overriding.

gcc/testsuite/:

2014-09-15  Andi Kleen  <ak@linux.intel.com>

* gcc.dg/pg-override.c: New test.
* gcc.dg/pg.c: New test.
* gcc.target/i386/fentry-override.c: New test.
* gcc.target/i386/fentry.c: New test.

From-SVN: r215285

Always set DECL_NO_INSTRUMENT_FUNCTION_ENTRY_EXIT correctly

When profiling is disabled force DECL_NO_INSTRUMENT_FUNCTION_ENTRY_EXIT
for each function to one. This information is then preserved
through LTO.

With this patch for LTO builds -pg needs to be set on both the
LTO final link and the original source build, to allow -pg
(or -pg -fentry) to be active for that source file. This allows
to build large projects mostly with -pg, except for a few files,
and still use LTO.

gcc/:

2014-09-15  Andi Kleen  <ak@linux.intel.com>

* function.c (allocate_struct_function): Force
DECL_NO_INSTRUMENT_FUNCTION_ENTRY_EXIT to one when
profiling is disabled.

From-SVN: r215284

pt.c (lookup_template_class_1): Splice out abi_tag attribute if necessary.

* pt.c (lookup_template_class_1): Splice out abi_tag attribute if
necessary.  Call inherit_targ_abi_tags here.
* class.c (check_bases_and_members): Not here.
(inherit_targ_abi_tags): Check CLASS_TYPE_P.
* cp-tree.h: Declare inherit_targ_abi_tags.

From-SVN: r215283

Replace INSN_DELETED_P with rtx_insn member functions

gcc/

* cfgrtl.c, combine.c, config/arc/arc.c, config/mcore/mcore.c,
config/rs6000/rs6000.c, config/sh/sh.c, cprop.c, dwarf2out.c,
emit-rtl.c, final.c, function.c, gcse.c, jump.c, reg-stack.c,
reload1.c, reorg.c, resource.c, sel-sched-ir.c: Replace INSN_DELETED_P
macro with statically checked member functions.
* rtl.h (rtx_insn::deleted): New method.
(rtx_insn::set_deleted): Likewise.
(rtx_insn::set_undeleted): Likewise.
(INSN_DELETED_P): Remove.

From-SVN: r215282

use rtx_insn * more

gcc/ChangeLog:

2014-09-15  Trevor Saunders  <tsaunders@mozilla.com>

* config/mn10300/mn10300.c (mn10300_insert_setlb_lcc): Assign the
result of emit_jump_insn_before to a new variable.
* jump.c (mark_jump_label): Change the type of insn to rtx_insn *.
(mark_jump_label_1): Likewise.
(mark_jump_label_asm): Likewise.
* reload1.c (gen_reload): Change type of tem to rtx_insn *.
* rtl.h (mark_jump_label): Adjust.

From-SVN: r215281

Daily bump.

From-SVN: r215280

re PR fortran/56408 (Fix dependency handling of testsuite/gfortran.dg)

PR fortran/56408
* gfortran.dg/dg.exp (gfortran_test_path, gfortran_aux_module_flags):
New global vars.
(dg-compile-aux-modules): New procedure.
* gfortran.dg/binding_label_tests_10.f03: Remove comment
and keep-modules.
* gfortran.dg/binding_label_tests_10_main.f03: Remove comment,
use dg-compile-aux-modules.
* gfortran.dg/binding_label_tests_11.f03: Remove comment
and keep-modules.
* gfortran.dg/binding_label_tests_11_main.f03: Remove comment,
use dg-compile-aux-modules.
* gfortran.dg/binding_label_tests_13.f03: Remove comment
and keep-modules.
* gfortran.dg/binding_label_tests_13_main.f03: Remove comment,
use dg-compile-aux-modules.
* gfortran.dg/binding_label_tests_26a.f90: Remove comment
and keep-modules.
* gfortran.dg/binding_label_tests_26b.f90: Remove comment,
use dg-compile-aux-modules.
* gfortran.dg/class_45a.f03: Remove keep-modules.
* gfortran.dg/class_45b.f03: Use dg-compile-aux-modules
and cleanup-modules.
* gfortran.dg/class_4a.f03: Use dg-do link, use
dg-additional-sources and cleanup-modules, remove keep-modules
and update comment.
* gfortran.dg/class_4b.f03: Don't compile anywhere, remove
keep-modules, adjust comment.
* gfortran.dg/class_4c.f03: Don't compile anywhere, remove
dg-additional-sources and cleanup-modules.
* gfortran.dg/coarray_29_1.f90: Remove keep-modules and
associated comment.
* gfortran.dg/coarray_29_2.f90: Use dg-compile-aux-modules,
remove comment.
* gfortran.dg/test_common_binding_labels_2.f03: Remove keep-modules.
* gfortran.dg/test_common_binding_labels_2_main.f03: Remove comment,
use dg-compile-aux-modules.
* gfortran.dg/test_common_binding_labels_3.f03: Remove keep-modules.
* gfortran.dg/test_common_binding_labels_3_main.f03: Remove comment,
use dg-compile-aux-modules.
* gfortran.dg/whole_file_28.f90: Remove comment and keep-modules.
* gfortran.dg/whole_file_29.f90: Remove comment, use
dg-compile-aux-modules.
* gfortran.dg/whole_file_30.f90: Remove comment and keep-modules.
* gfortran.dg/whole_file_31.f90: Remove comment, use
dg-compile-aux-modules.

From-SVN: r215275

Do not diagnose lambda default arguments in c++14 modes.

* parser.c (cp_parser_lambda_declarator_opt): Make the pedwarn
conditional.

From-SVN: r215274

Makefile.in (dg_target_exps): Remove.

gcc/
* Makefile.in (dg_target_exps): Remove.
(check_gcc_parallelize): Change to just an upper bound number.
(check-%-subtargets): Always print the non-parallelized goals.
(check_p_vars, check_p_comma, check_p_subwork): Remove.
(check_p_count, check_p_numbers0, check_p_numbers1, check_p_numbers2,
check_p_numbers3, check_p_numbers4, check_p_numbers5,
check_p_numbers6): New variables.
(check_p_numbers): Set to sequence from 1 to 9999.
(check_p_subdirs): Set to sequence from 1 to minimum of
$(check_p_count) and either GCC_TEST_PARALLEL_SLOTS env var if set,
or 128.
(check-%, check-parallel-%): Rewritten so that for parallelized
testing each job runs all the *.exp files, with
GCC_RUNTEST_PARALLELIZE_DIR set in environment.
gcc/go/
* Make-lang.in (check_go_parallelize): Change to just an upper bound
number.
gcc/fortran/
* Make-lang.in (check_gfortran_parallelize): Change to just an upper
bound number.
gcc/cp/
* Make-lang.in (check_g++_parallelize): Change to just an upper bound
number.
gcc/objc/
* Make-lang.in (check_objc_parallelize): Change to just an upper
bound number.
gcc/ada/
* gcc-interface/Make-lang.in (check_acats_numbers0,
check_acats_numbers1, check_acats_numbers2, check_acats_numbers3,
check_acats_numbers4, check_acats_numbers5, check_acats_numbers6,
check_acats_numbers, check_acats_subdirs): New variables.
(check_acats_targets): Use $(check_acats_subdirs).
(check-acats, check-acats%): Rewritten so that for parallelized
testing each job runs all the chapters files, with
GCC_RUNTEST_PARALLELIZE_DIR set in environment.  Prepare the support
directory sequentially and share it.
(check-acats-subtargets): Always print just check-acats.
gcc/testsuite/
* lib/gcc-defs.exp (gcc_parallel_test_run_p,
gcc_parallel_test_enable): New procedures.  If
GCC_RUNTEST_PARALLELIZE_DIR is set in environment, override
runtest_file_p to invoke also gcc_parallel_test_run_p.
* g++.dg/guality/guality.exp (check_guality): Save/restore
test_counts array around the body of the procedure.
* gcc.dg/guality/guality.exp (check_guality): Likewise.
* g++.dg/plugin/plugin.exp: Run all the tests serially
by the first parallel runtest encountering it.
* gcc.dg/plugin/plugin.exp: Likewise.
* gcc.misc-tests/matrix1.exp: Likewise.
* gcc.misc-tests/dhry.exp: Likewise.
* gcc.misc-tests/acker1.exp: Likewise.
* gcc.misc-tests/linkage.exp: Likewise.
* gcc.misc-tests/mg.exp: Likewise.
* gcc.misc-tests/mg-2.exp: Likewise.
* gcc.misc-tests/sort2.exp: Likewise.
* gcc.misc-tests/sieve.exp: Likewise.
* gcc.misc-tests/options.exp: Likewise.
* gcc.misc-tests/help.exp: Likewise.
* go.test/go-test.exp (go-gc-tests): Use
gcc_parallel_test_enable {0, 1} around all handling of
each test.
* objc.dg/gnu-encoding/gnu-encoding.exp: Likewise.
* ada/acats/run_all.sh: Parallelize using mkdir inside of
a shared directory.  Avoid forking and expr uses if
shell supports $(()).
libstdc++-v3/
* testsuite/Makefile.am (check_p_numbers0, check_p_numbers1,
check_p_numbers2, check_p_numbers3, check_p_numbers4,
check_p_numbers5, check_p_numbers6, check_p_numbers,
check_p_subdirs): New variables.
(check_DEJAGNU_normal_targets): Use check_p_subdirs.
(check-DEJAGNU): Rewritten so that for parallelized
testing each job runs all the *.exp files, with
GCC_RUNTEST_PARALLELIZE_DIR set in environment.
* testsuite/Makefile.in: Regenerated.
* testsuite/lib/libstdc++.exp (gcc_parallel_test_run_p,
gcc_parallel_test_enable): New procedures.  If
GCC_RUNTEST_PARALLELIZE_DIR is set in environment, override
runtest_file_p to invoke also gcc_parallel_test_run_p.
* testsuite/libstdc++-abi/abi.exp: Run all the tests serially
by the first parallel runtest encountering it.  Fix up path
of the extract_symvers script.
* testsuite/libstdc++-xmethods/xmethods.exp: Run all the tests
serially by the first parallel runtest encountering it.  Run
dg-finish even in case of error.

From-SVN: r215273

c.opt (Wpsabi): Use LangEnabledBy.

gcc/c-family/ChangeLog:

2014-09-15  Manuel López-Ibáñez  <manu@gcc.gnu.org>

* c.opt (Wpsabi): Use LangEnabledBy.
* c-opts.c (c_common_handle_option): Do not handle here.

From-SVN: r215272

Instruction attributes take an rtx_insn *

gcc/ChangeLog:
* config/arc/arc-protos.h (arc_attr_type): Strengthen param from
rtx to rtx_insn *.
(arc_sets_cc_p): Likewise.
* config/arc/arc.c (arc_print_operand): Use methods of
"final_sequence" for clarity, and to enable strengthening of
locals "jump" and "delay" from rtx to rtx_insn *.
(arc_adjust_insn_length): Strengthen local "prev" from rtx to
rtx_insn *; use method of rtx_sequence for typesafety.
(arc_get_insn_variants): Use insn method of rtx_sequence for
typesafety.
(arc_pad_return): Likewise.
(arc_attr_type): Strengthen param from rtx to rtx_insn *.
(arc_sets_cc_p): Likewise.  Also, convert a GET_CODE check to a
dyn_cast to rtx_sequence *, using insn method for typesafety.
* config/arc/arc.h (ADJUST_INSN_LENGTH): Add checked casts to
rtx_sequence * and use insn method when invoking get_attr_length.
* config/bfin/bfin.c (type_for_anomaly): Strengthen param from rtx
to rtx_insn *.  Replace a GET_CODE check with a dyn_cast to
rtx_sequence *, introducing a local "seq", using its insn method
from typesafety and clarity.
(add_sched_insns_for_speculation): Strengthen local "next" from
rtx to rtx_insn *.
* config/c6x/c6x.c (get_insn_side): Likewise for param "insn".
(predicate_insn): Likewise.
* config/cris/cris-protos.h (cris_notice_update_cc): Likewise for
second param.
* config/cris/cris.c (cris_notice_update_cc): Likewise.
* config/epiphany/epiphany-protos.h
(extern void epiphany_insert_mode_switch_use): Likewise for param
"insn".
(get_attr_sched_use_fpu): Likewise for param.
* config/epiphany/epiphany.c (epiphany_insert_mode_switch_use):
Likewise for param "insn".
* config/epiphany/mode-switch-use.c (insert_uses): Likewise for
param "insn" of "target_insert_mode_switch_use" callback.
* config/frv/frv.c (frv_insn_unit): Likewise for param "insn".
(frv_issues_to_branch_unit_p): Likewise.
(frv_pack_insn_p): Likewise.
(frv_compare_insns): Strengthen locals "insn1" and "insn2" from
const rtx * (i.e. mutable rtx_def * const *) to
rtx_insn * const *.
* config/i386/i386-protos.h (standard_sse_constant_opcode):
Strengthen first param from rtx to rtx_insn *.
(output_fix_trunc): Likewise.
* config/i386/i386.c (standard_sse_constant_opcode): Likewise.
(output_fix_trunc): Likewise.
(core2i7_first_cycle_multipass_filter_ready_try): Likewise for
local "insn".
(min_insn_size): Likewise for param "insn".
(get_mem_group): Likewise.
(is_cmp): Likewise.
(get_insn_path): Likewise.
(get_insn_group): Likewise.
(count_num_restricted): Likewise.
(fits_dispatch_window): Likewise.
(add_insn_window): Likewise.
(add_to_dispatch_window): Likewise.
(debug_insn_dispatch_info_file): Likewise.
* config/m32c/m32c-protos.h (m32c_output_compare): Likewise for
first param.
* config/m32c/m32c.c (m32c_compare_redundant): Likewise for param
"cmp" and local "prev".
(m32c_output_compare): Likewise for param "insn".
* config/m32r/predicates.md (define_predicate "small_insn_p"): Add
a checked cast to rtx_insn * on "op" after we know it's an INSN_P.
(define_predicate "large_insn_p"): Likewise.
* config/m68k/m68k-protos.h (m68k_sched_attr_size): Strengthen
param from rtx to rtx_insn *.
(attr_op_mem m68k_sched_attr_op_mem): Likewise.
* config/m68k/m68k.c (sched_get_attr_size_int): Likewise.
(m68k_sched_attr_size): Likewise.
(sched_get_opxy_mem_type): Likewise for param "insn".
(m68k_sched_attr_op_mem): Likewise.
(sched_mem_operand_p): Likewise.
* config/mep/mep-protos.h (mep_multi_slot): Likewise for param.
* config/mep/mep.c (mep_multi_slot): Likewise.
* config/mips/mips-protos.h (mips_output_sync_loop): Likewise for
first param.
(mips_sync_loop_insns): Likewise.
* config/mips/mips.c (mips_print_operand_punctuation): Use insn
method of "final_sequence" for typesafety.
(mips_process_sync_loop): Strengthen param "insn" from rtx to
rtx_insn *.
(mips_output_sync_loop): Likewise.
(mips_sync_loop_insns): Likewise.
(mips_74k_agen_init): Likewise.
(mips_sched_init): Use NULL rather than NULL_RTX when working with
insns.
* config/nds32/nds32-fp-as-gp.c (nds32_symbol_load_store_p):
Strengthen param "insn" from rtx to rtx_insn *.
* config/nds32/nds32.c (nds32_target_alignment): Likewise for
local "insn".
* config/pa/pa-protos.h (pa_insn_refs_are_delayed): Likewise for
param.
* config/pa/pa.c (pa_output_function_epilogue): Likewise for local
"insn".  Use method of rtx_sequence for typesafety.
(branch_to_delay_slot_p): Strengthen param "insn" from rtx to
rtx_insn *.
(branch_needs_nop_p): Likewise.
(use_skip_p): Likewise.
(pa_insn_refs_are_delayed): Likewise.
* config/rl78/rl78.c (rl78_propogate_register_origins): Likewise
for locals "insn", "ninsn".
* config/rs6000/rs6000.c (is_microcoded_insn): Likewise for param
"insn".
(is_cracked_insn): Likewise.
(is_branch_slot_insn): Likewise.
(is_nonpipeline_insn): Likewise.
(insn_terminates_group_p): Likewise.
(insn_must_be_first_in_group): Likewise.
(insn_must_be_last_in_group): Likewise.
(force_new_group): Likewise for param "next_insn".
* config/s390/s390.c (s390_get_sched_attrmask): Likewise for param
"insn".
(s390_sched_score): Likewise.
* config/sh/sh-protos.h (output_branch): Likewise for param 2.
(rtx sfunc_uses_reg): Likewise for sole param.
* config/sh/sh.c (sh_print_operand): Use insn method of
final_sequence for typesafety.
(output_branch): Strengthen param "insn" from rtx to rtx_insn *.
Use insn method of final_sequence for typesafety.
(sfunc_uses_reg): Strengthen param "insn" from rtx to rtx_insn *.
* config/sparc/sparc-protos.h (eligible_for_call_delay): Likewise
for param.
(eligible_for_return_delay): Likewise.
(eligible_for_sibcall_delay): Likewise.
* config/sparc/sparc.c (eligible_for_call_delay): Likewise.
(eligible_for_return_delay): Likewise.
(eligible_for_sibcall_delay): Likewise.
* config/stormy16/stormy16-protos.h
(xstormy16_output_cbranch_hi): Likewise for final param.
(xstormy16_output_cbranch_si): Likewise.
* config/stormy16/stormy16.c (xstormy16_output_cbranch_hi): LIkewise.
(xstormy16_output_cbranch_si): Likewise.
* config/v850/v850-protos.h (notice_update_cc): Likewise.
* config/v850/v850.c (notice_update_cc): Likewise.

* final.c (get_attr_length_1): Strengthen param "insn" and param
of "fallback_fn" from rtx to rtx_insn *, eliminating a checked cast.
(get_attr_length): Strengthen param "insn" from rtx to rtx_insn *.
(get_attr_min_length): Likewise.
(shorten_branches): Likewise for signature of locals "length_fun"
and "inner_length_fun".  Introduce local rtx_sequence * "seqn"
from a checked cast and use its methods for clarity and to enable
strengthening local "inner_insn" from rtx to rtx_insn *.
* genattr.c (gen_attr): When writing out the prototypes of the
various generated "get_attr_" functions, strengthen the params of
the non-const functions from rtx to rtx_insn *.
Similarly, strengthen the params of insn_default_length,
insn_min_length, insn_variable_length_p, insn_current_length.
(main): Similarly, strengthen the param of num_delay_slots,
internal_dfa_insn_code, insn_default_latency, bypass_p,
insn_latency, min_issue_delay, print_reservation,
insn_has_dfa_reservation_p and of the "internal_dfa_insn_code" and
"insn_default_latency" callbacks.  Rename hook_int_rtx_unreachable
to hook_int_rtx_insn_unreachable.
* genattrtab.c (write_attr_get): When writing out the generated
"get_attr_" functions, strengthen the param "insn" from rtx to
rtx_insn *, eliminating a checked cast.
(make_automaton_attrs): When writing out prototypes of
"internal_dfa_insn_code_", "insn_default_latency_" functions
and the "internal_dfa_insn_code" and "insn_default_latency"
callbacks, strengthen their params from rtx to rtx_insn *
* genautomata.c (output_internal_insn_code_evaluation): When
writing out code, add a checked cast from rtx to rtx_insn * when
invoking DFA_INSN_CODE_FUNC_NAME aka dfa_insn_code.
(output_dfa_insn_code_func): Strengthen param of generated
function "dfa_insn_code_enlarge" from rtx to rtx_insn *.
(output_trans_func): Likewise for generated function
"state_transition".
(output_internal_insn_latency_func): When writing out generated
function "internal_insn_latency", rename params from "insn" and
"insn2" to "insn_or_const0" and "insn2_or_const0".  Reintroduce
locals "insn" and "insn2" as rtx_insn * with checked casts once
we've proven that we're not dealing with const0_rtx.
(output_insn_latency_func):  Strengthen param of generated
function "insn_latency" from rtx to rtx_insn *.
(output_print_reservation_func): Likewise for generated function
"print_reservation".
(output_insn_has_dfa_reservation_p): Likewise for generated
function "insn_has_dfa_reservation_p".
* hooks.c (hook_int_rtx_unreachable): Rename to...
(hook_int_rtx_insn_unreachable): ...this, and strengthen param
from rtx to rtx_insn *.
* hooks.h (hook_int_rtx_unreachable): Likewise.
(extern int hook_int_rtx_insn_unreachable): Likewise.
* output.h (get_attr_length): Strengthen param from rtx to rtx_insn *.
(get_attr_min_length): Likewise.
* recog.c (get_enabled_alternatives): Likewise.
* recog.h (alternative_mask get_enabled_alternatives): Likewise.
* reorg.c (find_end_label): Introduce local rtx "pat" and
strengthen local "insn" from rtx to rtx_insn *.
(redundant_insn): Use insn method of "seq" rather than element for
typesafety; strengthen local "control" from rtx to rtx_insn *.
* resource.c (mark_referenced_resources): Add checked cast to
rtx_insn * within INSN/JUMP_INSN case.
(mark_set_resources): Likewise.
* sel-sched.c (estimate_insn_cost): Strengthen param "insn" from
rtx to rtx_insn *.

From-SVN: r215271

The various TARGET_ASM_..._MAX_SKIP hooks take an insn

gcc/ChangeLog:
* config/rs6000/rs6000.c (rs6000_loop_align_max_skip): Strengthen
param "label" from rtx to rtx_insn *.
* config/rx/rx.c (rx_max_skip_for_label): Likewise for param "lab"
and local "op".
* doc/tm.texi (TARGET_ASM_JUMP_ALIGN_MAX_SKIP): Autogenerated changes.
(TARGET_ASM_LABEL_ALIGN_AFTER_BARRIER_MAX_SKIP): Likewise.
(TARGET_ASM_LOOP_ALIGN_MAX_SKIP): Likewise.
(TARGET_ASM_LABEL_ALIGN_MAX_SKIP): Likewise.
* final.c (default_label_align_after_barrier_max_skip): Strengthen
param from rtx to rtx_insn *.
(default_loop_align_max_skip): Likewise.
(default_label_align_max_skip): Likewise.
(default_jump_align_max_skip): Likewise.
* target.def (label_align_after_barrier_max_skip): Likewise.
(loop_align_max_skip): Likewise.
(label_align_max_skip): Likewise.
(jump_align_max_skip): Likewise.
* targhooks.h (default_label_align_after_barrier_max_skip):
Likewise.
(default_loop_align_max_skip): Likewise.
(default_label_align_max_skip): Likewise.
(default_jump_align_max_skip): Likewise.

From-SVN: r215270

The TARGET_CAN_FOLLOW_JUMP hook takes insns

gcc/ChangeLog:
* config/arc/arc.c (arc_can_follow_jump): Strengthen both params
from const_rtx to const rtx_insn *.  Update union members from rtx
to rtx_insn *.
* doc/tm.texi (TARGET_CAN_FOLLOW_JUMP): Autogenerated change.
* hooks.c (hook_bool_const_rtx_const_rtx_true): Rename to...
(hook_bool_const_rtx_insn_const_rtx_insn_true): ...this, and
strengthen both params from const_rtx to const rtx_insn *.
* hooks.h (hook_bool_const_rtx_const_rtx_true): Likewise.
(hook_bool_const_rtx_insn_const_rtx_insn_true): Likewise.
* reorg.c (follow_jumps): Strengthen param "jump" from rtx to
rtx_insn *.
* target.def (can_follow_jump): Strengthen both params from
const_rtx to const rtx_insn *, and update default implementation
from hook_bool_const_rtx_const_rtx_true to
hook_bool_const_rtx_insn_const_rtx_insn_true.

From-SVN: r215269

deps_start_bb takes an insn

gcc/ChangeLog:
* sched-deps.c (deps_start_bb): Strengthen param "head" and local
"insn" from rtx to rtx_insn *.
* sched-int.h (deps_start_bb): Likewise for 2nd param.

From-SVN: r215268

longlong.h: Add __udiv_w_sdiv prototype.

2014-09-15  Andreas Krebbel  <Andreas.Krebbel@de.ibm.com>

* longlong.h: Add __udiv_w_sdiv prototype.

From-SVN: r215266

AVX-512. Extend vcvtps2ph insn patterns.

gcc/
* config/i386/sse.md
(define_insn "vcvtph2ps<mask_name>"): Add masking.
(define_insn "*vcvtph2ps_load<mask_name>"): Ditto.
(define_insn "vcvtph2ps256<mask_name>"): Ditto.
(define_expand "vcvtps2ph_mask"): New.
(define_insn "*vcvtps2ph<mask_name>"): Add masking.
(define_insn "*vcvtps2ph_store<mask_name>"): Ditto.
(define_insn "vcvtps2ph256<mask_name>"): Ditto.

Co-Authored-By: Andrey Turetskiy <andrey.turetskiy@intel.com>
Co-Authored-By: Anna Tikhonova <anna.tikhonova@intel.com>
Co-Authored-By: Ilya Tocar <ilya.tocar@intel.com>
Co-Authored-By: Ilya Verbin <ilya.verbin@intel.com>
Co-Authored-By: Kirill Yukhin <kirill.yukhin@intel.com>
Co-Authored-By: Maxim Kuznetsov <maxim.kuznetsov@intel.com>
Co-Authored-By: Michael Zolotukhin <michael.v.zolotukhin@intel.com>
From-SVN: r215263

AVX-512. Extend ashrv insn patterns.

gcc/
* config/i386/sse.md (define_mode_iterator VI248_AVX512BW_AVX512VL):
New.
(define_mode_iterator VI24_AVX512BW_1): Ditto.
(define_insn "<mask_codefor>ashr<mode>3<mask_name>"): Ditto.
(define_insn "<mask_codefor>ashrv2di3<mask_name>"): Ditto.
(define_insn "ashr<VI248_AVX512BW_AVX512VL:mode>3<mask_name>"): Enable
also for TARGET_AVX512VL.
(define_expand "ashrv2di3"): Update to enable TARGET_AVX512VL.

Co-Authored-By: Andrey Turetskiy <andrey.turetskiy@intel.com>
Co-Authored-By: Anna Tikhonova <anna.tikhonova@intel.com>
Co-Authored-By: Ilya Tocar <ilya.tocar@intel.com>
Co-Authored-By: Ilya Verbin <ilya.verbin@intel.com>
Co-Authored-By: Kirill Yukhin <kirill.yukhin@intel.com>
Co-Authored-By: Maxim Kuznetsov <maxim.kuznetsov@intel.com>
Co-Authored-By: Michael Zolotukhin <michael.v.zolotukhin@intel.com>
From-SVN: r215262

regex.h (basic_regex): Rename _Ch_typeraits template parameters to _Ch_traits.

* include/bits/regex.h (basic_regex): Rename _Ch_typeraits template
parameters to _Ch_traits.

From-SVN: r215261

[AArch64] Fix force_simd macro in vdup_lane_2

gcc/

* gcc.target/aarch64/vdup_lane_2.c (force_simd): Emit simd mov.

From-SVN: r215260

doc/install.texi: add documentation for --disable-libsanitizer

This patch adds an item for --disable-libsanitizer to the configuration
documentation.
The option is especially useful during bisection, because it reduces
build time a lot.

2014-09-15  Markus Trippelsdorf  <markus@trippelsdorf.de>

* doc/install.texi (Options specification): add
--disable-libsanitizer item.

From-SVN: r215259

Daily bump.

From-SVN: r215257

re PR target/61407 (Build errors on latest OS X 10.10 Yosemite with Xcode 6 on GCC 4.8.3)

PR target/61407

* config/darwin-c.c (version_as_macro): Added extra 0 for OS X 10.10
and above.
* config/darwin-driver.c (darwin_find_version_from_kernel): Removed
kernel version check to avoid incrementing it after every major OS X
release.
(darwin_default_min_version): Avoid static memory buffer.

* gcc.dg/darwin-minversion-1.c: Fixed formatting
* gcc.dg/darwin-minversion-2.c: Fixed formatting
* gcc.dg/darwin-minversion-3.c: Fixed formatting
* gcc.dg/darwin-minversion-4.c: Added test for OS X 10.10

Co-Authored-By: Francois-Xavier Coudert <fxcoudert@gcc.gnu.org>
From-SVN: r215251

Daily bump.

From-SVN: r215250

Fix mistake in commit 215237

From-SVN: r215242

tree.c (need_assembler_name_p): Store C++ type mangling only for aggregates.

* tree.c (need_assembler_name_p): Store C++ type mangling only
for aggregates.

From-SVN: r215238

* MAINTAINERS: Move myself to reviewers (Fortran).

From-SVN: r215237

re PR fortran/63252 (tree_class_check_failed)

2014-09-13  Tobias Burnus  <burnus@net-b.de>

        PR fortran/63252
        * trans-decl.c (gfc_build_builtin_function_decls): Fix
        caf_unlock declaration.

From-SVN: r215236

re PR c++/60862 (bad location in invalid conversion error)

PR c++/60862
* parser.c (cp_parser_postfix_expression) <case CPP_OPEN_PAREN>: Set
location of a call expression.

* g++.dg/diagnostic/pr60862.C: New test.

From-SVN: r215235

tree.c (protected_set_expr_location): Don't check whether T is non-null here.

* tree.c (protected_set_expr_location): Don't check whether T is
non-null here.

From-SVN: r215234

Daily bump.

From-SVN: r215233

msp430.md (extendhipsi2): Use 20-bit form of RLAM/RRAM.

* config/msp430/msp430.md (extendhipsi2): Use 20-bit form of RLAM/RRAM.
(extend_and_shift1_hipsi2): Likewise.
(extend_and_shift2_hipsi2): Likewise.

From-SVN: r215229

re PR c++/63201 (Full specialization of a member variable template of a class template does not work)

PR c++/63201
* decl.c (start_decl): Handle specialization of member variable
template.
* pt.c (check_explicit_specialization): Adjust error.

From-SVN: r215226

deque (__gnu_debug::deque): Make base class C++11 allocator aware.

* include/debug/deque (__gnu_debug::deque): Make base class C++11
allocator aware.

From-SVN: r215223

params 2 and 3 of reg_set_between_p

gcc/ChangeLog:
* config/alpha/alpha.c (alpha_ra_ever_killed): Replace NULL_RTX
with NULL when dealing with an insn.
* config/sh/sh.c (sh_reorg): Strengthen local "last_float_move"
from rtx to rtx_insn *.
* rtl.h (reg_set_between_p): Strengthen params 2 and 3 from
const_rtx to const rtx_insn *.
* rtlanal.c (reg_set_between_p): Likewise, removing a checked cast.

From-SVN: r215222

re PR libstdc++/59603 (std::random_shuffle tries to swap element with itself)

PR libstdc++/59603
* include/bits/stl_algo.h (random_shuffle): Prevent self-swapping.
* testsuite/25_algorithms/random_shuffle/59603.cc: New.

From-SVN: r215219

fix ChangeLog typo

From-SVN: r215218

fix assert in hash_table pch routines

gcc/ChangeLog:

2014-09-12  Trevor Saunders  <tsaunders@mozilla.com>

* hash-table.h (gt_pch_nx): don't call gt_pch_note_object within an
assert.

From-SVN: r215216

Remove LIBGCC2_HAS_?F_MODE target macros.

This patch removes the LIBGCC2_HAS_{SF,DF,XF,TF}_MODE target macros,
replacing them by predefines with -fbuilding-libgcc, together with a
target hook that can influence those predefines when needed.

The new default is that a floating-point mode is supported in libgcc
if (a) it passes the scalar_mode_supported_p hook (otherwise it's not
plausible for it to be supported in libgcc) and (b) it's one of those
four modes (since those are the modes for which libgcc hardcodes the
possibility of support).  The target hook can override the default
choice (in either direction) for modes that pass
scalar_mode_supported_p (although overriding in the direction of
returning true when the default would return false only makes sense if
all relevant functions are specially defined in libgcc for that
particular target).

The previous default settings depended on various settings such as
LIBGCC2_LONG_DOUBLE_TYPE_SIZE, as well as targets defining the above
target macros if the default wasn't correct.

The default scalar_mode_supported_p only declares a floating-point
mode to be supported if it matches one of float / double / long
double.  This means that in most cases where a mode is only supported
conditionally in libgcc (TFmode only supported if it's the mode of
long double, most commonly), the default gets things right.  Overrides
were needed in the following cases:

* SFmode would always have been supported in libgcc (the condition was
  BITS_PER_UNIT == 8, true for all current targets), but pdp11
  defaults to 64-bit float, and in that case SFmode would fail
  scalar_mode_supported_p.  I don't know if libgcc actually built for
  pdp11 (and the port may well no longer be being used), but this
  patch adds a scalar_mode_supported_p hook to it to ensure SFmode is
  treated as supported.

* Certain i386 and ia64 targets need the new hook to match the
  existing cases for when XFmode or TFmode support is present in
  libgcc.  For i386, the hook can always declare XFmode to be
  supported - the cases where it's not are the cases where long double
  is TFmode, in which case XFmode fails scalar_mode_supported_p[*] -
  but TFmode support needs to be conditional.  (And of the targets not
  defining LIBGCC2_HAS_TF_MODE before this patch, some defined
  LONG_DOUBLE_TYPE_SIZE to 64, so ensuring LIBGCC2_HAS_TF_MODE would
  always be false, while others did not define it, so allowing it to
  be true in the -mlong-double-128 case.  This patch matches that
  logic, although I suspect all the latter targets would have been
  broken if you tried to enable -mlong-double-128 by default, for lack
  of the soft-fp TFmode support in libgcc, which is separately
  configured.)

  [*] I don't know if it's deliberate not to support __float80 at all
  with -mlong-double-128.

In order to implement the default version of the new hook,
insn-modes.h was made to contain macros such as HAVE_TFmode for each
machine mode, so the default hook can contain conditionals on whether
XFmode and TFmode exist (to match the hardcoding of a list of modes in
libgcc).  This is also used in fortran/trans-types.c; previously it
had a conditional on defined(LIBGCC2_HAS_TF_MODE) (a bit dubious,
since it ignored the value of the macro), which is replaced by testing
defined(HAVE_TFmode), in conjunction with requiring
targetm.libgcc_floating_mode_supported_p.

(Fortran is testing something stronger than that hook: not only is
libgcc support required, but also libm or equivalent.  Thus, it has a
test for ENABLE_LIBQUADMATH_SUPPORT in the case that the mode is
TFmode and that's not the same as any of the three standard types.
The old and new tests are intended to accept exactly the same set of
modes for all targets.)

Apart from the four target macros eliminated by this patch, it gets us
closer to eliminating LIBGCC2_LONG_DOUBLE_TYPE_SIZE as well, though a
few more places using that macro need changing first.

Bootstrapped with no regressions on x86_64-unknown-linux-gnu; also
built cc1 for crosses to ia64-elf and pdp11-none as a minimal test of
changes for those targets.

gcc:
* target.def (libgcc_floating_mode_supported_p): New hook.
* targhooks.c (default_libgcc_floating_mode_supported_p): New
function.
* targhooks.h (default_libgcc_floating_mode_supported_p): Declare.
* doc/tm.texi.in (LIBGCC2_HAS_DF_MODE, LIBGCC2_HAS_XF_MODE)
(LIBGCC2_HAS_TF_MODE): Remove.
(TARGET_LIBGCC_FLOATING_MODE_SUPPORTED_P): New @hook.
* doc/tm.texi: Regenerate.
* genmodes.c (emit_insn_modes_h): Define HAVE_%smode for each
machine mode.
* system.h (LIBGCC2_HAS_SF_MODE, LIBGCC2_HAS_DF_MODE)
(LIBGCC2_HAS_XF_MODE, LIBGCC2_HAS_TF_MODE): Poison.
* config/i386/cygming.h (LIBGCC2_HAS_TF_MODE): Remove.
* config/i386/darwin.h (LIBGCC2_HAS_TF_MODE): Remove.
* config/i386/djgpp.h (IX86_MAYBE_NO_LIBGCC_TFMODE): Define.
* config/i386/dragonfly.h (LIBGCC2_HAS_TF_MODE): Remove.
* config/i386/freebsd.h (LIBGCC2_HAS_TF_MODE): Remove.
* config/i386/gnu-user-common.h (LIBGCC2_HAS_TF_MODE): Remove.
* config/i386/i386-interix.h (IX86_NO_LIBGCC_TFMODE): Define.
* config/i386/i386.c (ix86_libgcc_floating_mode_supported_p): New
function.
(TARGET_LIBGCC_FLOATING_MODE_SUPPORTED_P): Define.
* config/i386/i386elf.h (IX86_MAYBE_NO_LIBGCC_TFMODE): Define.
* config/i386/lynx.h (IX86_MAYBE_NO_LIBGCC_TFMODE): Define.
* config/i386/netbsd-elf.h (IX86_MAYBE_NO_LIBGCC_TFMODE): Define.
* config/i386/netbsd64.h (IX86_MAYBE_NO_LIBGCC_TFMODE): Define.
* config/i386/nto.h (IX86_MAYBE_NO_LIBGCC_TFMODE): Define.
* config/i386/openbsd.h (IX86_MAYBE_NO_LIBGCC_TFMODE): Define.
* config/i386/openbsdelf.h (LIBGCC2_HAS_TF_MODE): Remove.
* config/i386/rtemself.h (IX86_NO_LIBGCC_TFMODE): Define.
* config/i386/sol2.h (LIBGCC2_HAS_TF_MODE): Remove.
* config/i386/vx-common.h (IX86_MAYBE_NO_LIBGCC_TFMODE): Define.
* config/ia64/elf.h (IA64_NO_LIBGCC_TFMODE): Define.
* config/ia64/freebsd.h (IA64_NO_LIBGCC_TFMODE): Define.
* config/ia64/hpux.h (LIBGCC2_HAS_XF_MODE, LIBGCC2_HAS_TF_MODE):
Remove.
* config/ia64/ia64.c (TARGET_LIBGCC_FLOATING_MODE_SUPPORTED_P):
New macro.
(ia64_libgcc_floating_mode_supported_p): New function.
* config/ia64/linux.h (LIBGCC2_HAS_TF_MODE): Remove.
* config/ia64/vms.h (IA64_NO_LIBGCC_XFMODE)
(IA64_NO_LIBGCC_TFMODE): Define.
* config/msp430/msp430.h (LIBGCC2_HAS_DF_MODE): Remove.
* config/pdp11/pdp11.c (TARGET_SCALAR_MODE_SUPPORTED_P): New
macro.
(pdp11_scalar_mode_supported_p): New function.
* config/rl78/rl78.h (LIBGCC2_HAS_DF_MODE): Remove.
* config/rx/rx.h (LIBGCC2_HAS_DF_MODE): Remove.

gcc/c-family:
* c-cppbuiltin.c (c_cpp_builtins): Define __LIBGCC_HAS_%s_MODE__
macros for floating-point modes.

gcc/fortran:
* trans-types.c (gfc_init_kinds): Check
targetm.libgcc_floating_mode_supported_p for floating-point
modes.  Check HAVE_TFmode instead of LIBGCC2_HAS_TF_MODE.

libgcc:
* libgcc2.h (LIBGCC2_HAS_SF_MODE): Define using
__LIBGCC_HAS_SF_MODE__.
(LIBGCC2_HAS_DF_MODE): Define using __LIBGCC_HAS_DF_MODE__.
(LIBGCC2_HAS_XF_MODE): Define using __LIBGCC_HAS_XF_MODE__.
(LIBGCC2_HAS_TF_MODE): Define using __LIBGCC_HAS_TF_MODE__.
* config/libbid/bid_gcc_intrinsics.h
(LIBGCC2_LONG_DOUBLE_TYPE_SIZE): Do not define.
(LIBGCC2_HAS_XF_MODE): Define using __LIBGCC_HAS_XF_MODE__.
(LIBGCC2_HAS_TF_MODE): Define using __LIBGCC_HAS_TF_MODE__.
* fixed-bit.h (LIBGCC2_LONG_DOUBLE_TYPE_SIZE): Do not define.
(LIBGCC2_HAS_SF_MODE): Define using __LIBGCC_HAS_SF_MODE__.
(LIBGCC2_HAS_DF_MODE): Define using __LIBGCC_HAS_DF_MODE__.

From-SVN: r215215

re PR middle-end/63237 (error: invalid operand in unary operation)

2014-09-12  Richard Biener  <rguenther@suse.de>

PR middle-end/63237
* gimple-fold.c (get_maxval_strlen): Gimplify string length.

* g++.dg/torture/pr63237.C: New testcase.

From-SVN: r215212

tree.c (integer_each_onep): New function.

2014-09-12  Marc Glisse  <marc.glisse@inria.fr>

gcc/
* tree.c (integer_each_onep): New function.
* tree.h (integer_each_onep): Declare it.
* fold-const.c (fold_binary_loc): Use it for ~A + 1 to -A and
-A - 1 to ~A.  Disable (X & 1) ^ 1, (X ^ 1) & 1 and ~X & 1 to
(X & 1) == 0 for vector and complex.
gcc/testsuite/
* gcc.dg/vec-andxor1.c: New file.

From-SVN: r215209

[AArch64] Add regmove_costs for Cortex-A57 and A53

2014-09-12  Wilco Dijkstra  <wdijkstr@arm.com>

  * gcc/config/aarch64/aarch64.c: (cortexa57_regmove_cost): New cost table for
  A57.
  (cortexa53_regmove_cost): New cost table for A53.  Increase GP2FP/FP2GP cost
  to spilling from integer to FP registers.

From-SVN: r215208

[AArch64] Fix cost for Q register moves

2014-09-12  Wilco Dijkstra  <wdijkstr@arm.com>

  * gcc/config/aarch64/aarch64.c (aarch64_register_move_cost): Fix Q register
  move handling.
  (generic_regmove_cost): Undo raised FP2FP move cost as Q register moves are
  now handled correctly.

From-SVN: r215207

[AArch64] Add cost handling of CALLER_SAVE_REGS and POINTER_REGS

2014-09-12  Wilco Dijkstra  <wdijkstr@arm.com>

  * gcc/config/aarch64/aarch64.c (aarch64_register_move_cost): Add cost handling
  of CALLER_SAVE_REGS and POINTER_REGS.

From-SVN: r215206

[Ree] Ensure inserted copy don't change the number of hard registers

2014-09-12  Wilco Dijkstra  <wilco.dijkstra@arm.com>

  gcc/
    * ree.c (combine_reaching_defs): Ensure inserted copy don't change the
    number of hard registers.

From-SVN: r215205

AVX-512. Extend vpternlog, valign, vrotate insns.

gcc/
* config/i386/sse.md
(define_mode_iterator VI48_AVX512VL): New.
(define_expand "<avx512>_vternlog<mode>_maskz"): Rename from
"avx512f_vternlog<mode>_maskz" and update mode iterator.
(define_insn "<avx512>_vternlog<mode><sd_maskz_name>"): Rename
from "avx512f_vternlog<mode><sd_maskz_name>" and update mode iterator.
(define_insn "<avx512>_vternlog<mode>_mask"): Rename from
"avx512f_vternlog<mode>_mask" and update mode iterator.
(define_insn "<mask_codefor><avx512>_align<mode><mask_name>"): Rename
from "<mask_codefor>avx512f_align<mode><mask_name>" and update mode
iterator.
(define_insn "<avx512>_<rotate>v<mode><mask_name>"): Rename from
"avx512f_<rotate>v<mode><mask_name>" and update mode iterator.
(define_insn "<avx512>_<rotate><mode><mask_name>"): Rename from
"avx512f_<rotate><mode><mask_name>" and update mode iterator.
(define_insn "clz<mode>2<mask_name>"): Use VI48_AVX512VL mode iterator.
(define_insn "<mask_codefor>conflict<mode><mask_name>"): Ditto.

Co-Authored-By: Andrey Turetskiy <andrey.turetskiy@intel.com>
Co-Authored-By: Anna Tikhonova <anna.tikhonova@intel.com>
Co-Authored-By: Ilya Tocar <ilya.tocar@intel.com>
Co-Authored-By: Ilya Verbin <ilya.verbin@intel.com>
Co-Authored-By: Kirill Yukhin <kirill.yukhin@intel.com>
Co-Authored-By: Maxim Kuznetsov <maxim.kuznetsov@intel.com>
Co-Authored-By: Michael Zolotukhin <michael.v.zolotukhin@intel.com>
From-SVN: r215203

AVX-512.  Extend max/min insn patterns.

gcc/
* config/i386/sse.md (VI128_256): Delete.
(define_mode_iterator VI124_256): New.
(define_mode_iterator VI124_256_AVX512F_AVX512BW): Ditto.
(define_expand "<code><mode>3<mask_name><round_name>"): Delete.
(define_expand "<code><VI124_256_AVX512F_AVX512BW:mode>3"): New.
(define_insn "*avx2_<code><VI124_256:mode>3"): Rename from
"*avx2_<code><mode>3<mask_name><round_name>" and update mode iterator.
(define_expand "<code><VI48_AVX512VL:mode>3_mask"): New.
(define_insn "*avx512bw_<code><VI48_AVX512VL:mode>3<mask_name>"): Ditto.
(define_insn "<mask_codefor><code><mode>3<mask_name>"): Update mode
iterator.
(define_expand "<code><VI8_AVX2:mode>3"): Update pettern generation
in presence of AVX-512.

Co-Authored-By: Andrey Turetskiy <andrey.turetskiy@intel.com>
Co-Authored-By: Anna Tikhonova <anna.tikhonova@intel.com>
Co-Authored-By: Ilya Tocar <ilya.tocar@intel.com>
Co-Authored-By: Ilya Verbin <ilya.verbin@intel.com>
Co-Authored-By: Kirill Yukhin <kirill.yukhin@intel.com>
Co-Authored-By: Maxim Kuznetsov <maxim.kuznetsov@intel.com>
Co-Authored-By: Michael Zolotukhin <michael.v.zolotukhin@intel.com>
From-SVN: r215202

AVX-512. Extend gather insn patterns.

gcc/
* config/i386/sse.md
(define_expand "<avx512>_gathersi<mode>"): Rename from
"avx512f_gathersi<mode>".
(define_insn "*avx512f_gathersi<mode>"): Use VI48F.
(define_insn "*avx512f_gathersi<mode>_2"): Ditto.
(define_expand "<avx512>_gatherdi<mode>"): Rename from
"avx512f_gatherdi<mode>".
(define_insn "*avx512f_gatherdi<mode>"): Use VI48F.
(define_insn "*avx512f_gatherdi<mode>_2"): Use VI48F, add 128/256-bit
wide versions.
(define_expand "<avx512>_scattersi<mode>"): Rename from
"avx512f_scattersi<mode>".
(define_insn "*avx512f_scattersi<mode>"): Use VI48F.
(define_expand "<avx512>_scatterdi<mode>"): Rename from
"avx512f_scatterdi<mode>".
(define_insn "*avx512f_scatterdi<mode>"): Use VI48F.

Co-Authored-By: Andrey Turetskiy <andrey.turetskiy@intel.com>
Co-Authored-By: Anna Tikhonova <anna.tikhonova@intel.com>
Co-Authored-By: Ilya Tocar <ilya.tocar@intel.com>
Co-Authored-By: Ilya Verbin <ilya.verbin@intel.com>
Co-Authored-By: Kirill Yukhin <kirill.yukhin@intel.com>
Co-Authored-By: Maxim Kuznetsov <maxim.kuznetsov@intel.com>
Co-Authored-By: Michael Zolotukhin <michael.v.zolotukhin@intel.com>
From-SVN: r215201

ira.h (ira_finish_once): Delete.

gcc/
* ira.h (ira_finish_once): Delete.
* ira-int.h (target_ira_int::~target_ira_int): Declare.
(target_ira_int::free_ira_costs): Likewise.
(target_ira_int::free_register_move_costs): Likewise.
(ira_finish_costs_once): Delete.
* ira.c (free_register_move_costs): Replace with...
(target_ira_int::free_register_move_costs): ...this new function.
(target_ira_int::~target_ira_int): Define.
(ira_init): Call free_register_move_costs as a member function rather
than a global function.
(ira_finish_once): Delete.
* ira-costs.c (free_ira_costs): Replace with...
(target_ira_int::free_ira_costs): ...this new function.
(ira_init_costs): Call free_ira_costs as a member function rather
than a global function.
(ira_finish_costs_once): Delete.
* target-globals.c (target_globals::~target_globals): Call the
target_ira_int destructor.
* toplev.c: Include lra.h.
(finalize): Call lra_finish_once rather than ira_finish_once.

From-SVN: r215200