GFNI enabling [4/4]

gcc/
* config/i386/gfniintrin.h (_mm_gf2p8mul_epi8, _mm256_gf2p8mul_epi8,
_mm_mask_gf2p8mul_epi8, _mm_maskz_gf2p8mul_epi8,
_mm256_mask_gf2p8mul_epi8, _mm256_maskz_gf2p8mul_epi8,
_mm512_mask_gf2p8mul_epi8, _mm512_maskz_gf2p8mul_epi8,
_mm512_gf2p8mul_epi8): New intrinsics.
* config/i386/i386-builtin-types.def
(V64QI_FTYPE_V64QI_V64QI): New type.
* config/i386/i386-builtin.def (__builtin_ia32_vgf2p8mulb_v64qi,
__builtin_ia32_vgf2p8mulb_v64qi_mask, __builtin_ia32_vgf2p8mulb_v32qi,
__builtin_ia32_vgf2p8mulb_v32qi_mask, __builtin_ia32_vgf2p8mulb_v16qi,
__builtin_ia32_vgf2p8mulb_v16qi_mask): New builtins.
* config/i386/sse.md (vgf2p8mulb_*): New pattern.
* config/i386/i386.c (ix86_expand_args_builtin): Handle new type.
gcc/testsuite/
* gcc.target/i386/avx512f-gf2p8mulb-2.c: New runtime tests.
* gcc.target/i386/avx512vl-gf2p8mulb-2.c: Ditto.
* gcc.target/i386/gfni-1.c: Add tests for GF2P8MUL.
* gcc.target/i386/gfni-2.c: Ditto.
* gcc.target/i386/gfni-3.c: Ditto.
* gcc.target/i386/gfni-4.c: Ditto.

From-SVN: r254795

* include/std/numeric (gcd): Fix typo.

From-SVN: r254794

Daily bump.

From-SVN: r254793

i386.c (x86_print_call_or_nop): Emit 5 byte nop explicitly as a stream of bytes.

* config/i386/i386.c (x86_print_call_or_nop): Emit 5 byte nop
explicitly as a stream of bytes.

From-SVN: r254788

altivec.h (vec_xst_be): New #define.

[gcc]

2017-11-15  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

* config/rs6000/altivec.h (vec_xst_be): New #define.
* config/rs6000/altivec.md (altivec_vperm_<mode>_direct): Rename
and externalize from *altivec_vperm_<mode>_internal.
* config/rs6000/rs6000-builtin.def (XL_BE_V16QI): Remove macro
instantiation.
(XL_BE_V8HI): Likewise.
(XL_BE_V4SI): Likewise.
(XL_BE_V4SI): Likewise.
(XL_BE_V2DI): Likewise.
(XL_BE_V4SF): Likewise.
(XL_BE_V2DF): Likewise.
(XST_BE): Add BU_VSX_OVERLOAD_X macro instantiation.
* config/rs6000/rs6000-c.c (altivec_overloaded_builtins): Correct
all array entries with these keys: VSX_BUILTIN_VEC_XL,
VSX_BUILTIN_VEC_XL_BE, VSX_BUILTIN_VEC_XST.  Add entries for key
VSX_BUILTIN_VEC_XST_BE.
* config/rs6000/rs6000.c (altivec_expand_xl_be_builtin): Remove.
(altivec_expand_builtin): Remove handling for VSX_BUILTIN_XL_BE_*
built-ins.
(altivec_init_builtins): Replace conditional calls to def_builtin
for __builtin_vsx_ld_elemrev_{v8hi,v16qi} and
__builtin_vsx_st_elemrev_{v8hi,v16qi} based on TARGET_P9_VECTOR
with unconditional calls.  Remove calls to def_builtin for
__builtin_vsx_le_be_<mode>.  Add a call to def_builtin for
__builtin_vec_xst_be.
* config/rs6000/vsx.md (vsx_ld_elemrev_v8hi): Convert define_insn
to define_expand, and add alternate RTL generation for P8.
(*vsx_ld_elemrev_v8hi_internal): New define_insn based on
vsx_ld_elemrev_v8hi.
(vsx_ld_elemrev_v16qi): Convert define_insn to define_expand, and
add alternate RTL generation for P8.
(*vsx_ld_elemrev_v16qi_internal): New define_insn based on
vsx_ld_elemrev_v16qi.
(vsx_st_elemrev_v8hi): Convert define_insn
to define_expand, and add alternate RTL generation for P8.
(*vsx_st_elemrev_v8hi_internal): New define_insn based on
vsx_st_elemrev_v8hi.
(vsx_st_elemrev_v16qi): Convert define_insn to define_expand, and
add alternate RTL generation for P8.
(*vsx_st_elemrev_v16qi_internal): New define_insn based on
vsx_st_elemrev_v16qi.

[gcc/testsuite]

2017-11-15  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

* gcc.target/powerpc/swaps-p8-26.c: Modify expected code
generation.

From-SVN: r254787

PR testsuite/82988 - g++.dg/cpp0x/lambda/lambda-switch.C fail

gcc/testsuite/ChangeLog:

PR testsuite/82988
* g++.dg/cpp0x/lambda/lambda-switch.C: Prune unimportant warning.

From-SVN: r254786

Fix gcd breakage.

* include/std/numeric (gcd): Use remove_cv_t, not remove_cv.

From-SVN: r254785

Move gcc/testsuite entries to gcc/testsuite/ChangeLog

From-SVN: r254784

i386: Add X86_TUNE_EMIT_VZEROUPPER

Add X86_TUNE_EMIT_VZEROUPPER to indicate if vzeroupper instruction should
be inserted before a transfer of control flow out of the function.  It is
turned on by default unless we are tuning for KNL.  Users can always use
-mzeroupper or -mno-zeroupper to override X86_TUNE_EMIT_VZEROUPPER.

gcc/

PR target/82990
* config/i386/i386.c (pass_insert_vzeroupper::gate): Remove
TARGET_AVX512ER check.
(ix86_option_override_internal): Set MASK_VZEROUPPER if
neither -mzeroupper nor -mno-zeroupper is used and
TARGET_EMIT_VZEROUPPER is set.
* config/i386/i386.h (TARGET_EMIT_VZEROUPPER): New.
* config/i386/x86-tune.def: Add X86_TUNE_EMIT_VZEROUPPER.

gcc/testsuite/

PR target/82990
* gcc.target/i386/pr82942-2.c: Add -mtune=knl.
* gcc.target/i386/pr82990-1.c: New test.
* gcc.target/i386/pr82990-2.c: Likewise.
* gcc.target/i386/pr82990-3.c: Likewise.
* gcc.target/i386/pr82990-4.c: Likewise.
* gcc.target/i386/pr82990-5.c: Likewise.
* gcc.target/i386/pr82990-6.c: Likewise.
* gcc.target/i386/pr82990-7.c: Likewise.

From-SVN: r254783

rs6000.c (rs6000_gimple_fold_builtin): Add support for folding of vector compares.

2017-11-15  Will Schmidt  <will_schmidt@vnet.ibm.com>

[gcc]

* config/rs6000/rs6000.c (rs6000_gimple_fold_builtin): Add support for
folding of vector compares.
(fold_build_vec_cmp): New helper function.
(fold_compare_helper): New helper function.
(builtin_function_type): Add compare builtins to the list of functions
having unsigned arguments.  Cosmetic updates to comment indentation.
* config/rs6000/vsx.md (vcmpneb, vcmpneh, vcmpnew): Update to specify
the not+eq combination.

[testsuite]

* gcc.target/powerpc/builtins-3-p9.c: Add -O1, update
expected codegen checks.
* gcc.target/powerpc/vec-cmp-sel.c: Mark vars as volatile.
* gcc.target/powerpc/vsu/vec-cmpne-0.c: Add -O1.
* gcc.target/powerpc/vsu/vec-cmpne-1.c: Add -O1.
* gcc.target/powerpc/vsu/vec-cmpne-2.c: Add -O1.
* gcc.target/powerpc/vsu/vec-cmpne-3.c: Add -O1.
* gcc.target/powerpc/vsu/vec-cmpne-4.c: Add -O1.
* gcc.target/powerpc/vsu/vec-cmpne-5.c: Add -O1.
* gcc.target/powerpc/vsu/vec-cmpne-6.c: Add -O1.

From-SVN: r254782

Minor improvements to Filesystem tests

* testsuite/27_io/filesystem/iterators/directory_iterator.cc: Leave
error_code unset.
* testsuite/27_io/filesystem/iterators/recursive_directory_iterator.cc:
Check for past-the-end before dereferencing.
* testsuite/experimental/filesystem/iterators/
recursive_directory_iterator.cc: Likewise.

From-SVN: r254781

re PR fortran/78240 (ICE in match_clist_expr, at fortran/decl.c:728)

2017-11-15  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/78240
gfortran.dg/pr78240.f90: Prune run-on errors.

From-SVN: r254780

Add noexcept to generic std::size, std::empty and std::data

* include/bits/range_access.h (size, empty, data): Add conditional
noexcept to generic overloads.

From-SVN: r254779

re PR tree-optimization/82726 (ICE in verify_ssa during GIMPLE pass: pcom)

PR tree-optimization/82726
PR tree-optimization/70754
* tree-predcom.c (order_drefs_by_pos): New function.
(combine_chains): Move code setting has_max_use_after to...
(try_combine_chains): ...here.  New parameter.  Sort combined chains
according to position information.
(tree_predictive_commoning_loop): Update call to above function.
(update_pos_for_combined_chains, pcom_stmt_dominates_stmt_p): New.

gcc/testsuite
* gcc.dg/tree-ssa/pr82726.c: New test.

From-SVN: r254778

re PR tree-optimization/82726 (ICE in verify_ssa during GIMPLE pass: pcom)

PR tree-optimization/82726
Revert
2017-01-23  Bin Cheng  <bin.cheng@arm.com>

PR tree-optimization/70754
* tree-predcom.c (stmt_combining_refs): New parameter INSERT_BEFORE.
(reassociate_to_the_same_stmt): New parameter INSERT_BEFORE.  Insert
combined stmt before it if not NULL.
(combine_chains): Process refs reversely and compute dominance point
for root ref.

Revert
2017-02-23  Bin Cheng  <bin.cheng@arm.com>

PR tree-optimization/79663
* tree-predcom.c (combine_chains): Process refs in reverse order
only for ZERO length chains, and add explaining comment.

From-SVN: r254777

arm-cpus.in (armv8_3, [...]): New

2017-11-15  Tamar Christina  <tamar.christina@arm.com>

        * config/arm/arm-cpus.in (armv8_3, ARMv8_3a, armv8.3-a): New
        * config/arm/arm-tables.opt (armv8.3-a): Regenerated.
        * doc/invoke.texi (ARM Options): Add armv8.3-a.

From-SVN: r254776

arm.h (TARGET_DOTPROD): Add arm_arch8_2.

2017-11-15  Tamar Christina  <tamar.christina@arm.com>

* config/arm/arm.h (TARGET_DOTPROD): Add arm_arch8_2.

From-SVN: r254775

RISC-V: Add Jim Wilson as a maintainer

Jim has recently started working at SiFive, where he'll be contributing
to our GCC port.  Andrew, Kito and I would like him to be a mainatiner.
My understand is that this is the right place to ask.

I've also changed my address to my SiFive one, where I also work -- it
seems cleaner to have everyone use our proper email address.

ChangeLog

2017-11-15  Palmer Dabbelt  <palmer@sifive.com>

        * MAINTAINERS (RISC-V): Add Jim Wilson as a maintainer.
        Use my SiFive email address.

From-SVN: r254774

[ARM] Fix more -Wreturn-type fallout

This patch fixes a couple of more tests that are giving out warnings with -Wreturn-type:
- g++.dg/ext/pr57735.C
- gcc.target/arm/pr54300.C

*** gcc/testsuite/ChangeLog ***

2017-11-10  Sudakshina Das  <sudi.das@arm.com>

        * g++.dg/ext/pr57735.C: Add -Wno-return-type for test.
        * gcc.target/arm/pr54300.C (main): Add return type and
        return a value.

From-SVN: r254773

Compile strncpy-fix-1.c with -Wno-stringop-truncation

2017-11-15  Tom de Vries  <tom@codesourcery.com>

* gcc.dg/strncpy-fix-1.c: Add -Wno-stringop-truncation to dg-options.

From-SVN: r254772

pr81706.c: Adjust asm for darwin.

2017-11-15  Dominique d'Humieres  <dominiq@lps.ens.fr>

* gcc.target/i386/pr81706.c: Adjust asm for darwin.
* g++.dg/ext/pr81706.C: Likewise.

From-SVN: r254770

Add libgomp.oacc-c-c++-common/f-asyncwait-{1,2,3}.c

2017-11-15  Tom de Vries  <tom@codesourcery.com>

* testsuite/libgomp.oacc-c-c++-common/f-asyncwait-1.c: New test, copied
from asyncwait-1.f90.  Rewrite into C.  Rewrite from float to int.
* testsuite/libgomp.oacc-c-c++-common/f-asyncwait-2.c: New test, copied
from asyncwait-2.f90.  Rewrite into C.  Rewrite from float to int.
* testsuite/libgomp.oacc-c-c++-common/f-asyncwait-3.c: New test, copied
from asyncwait-3.f90.  Rewrite into C.  Rewrite from float to int.

From-SVN: r254769

[PR c++/81574] lambda capture of function reference

https://gcc.gnu.org/ml/gcc-patches/2017-11/msg01200.html
PR c++/81574
* lambda.c (lambda_capture_field_type): Function references are
always catured by reference.

PR c++/81574
* g++.dg/cpp1y/pr81574.C: New.

From-SVN: r254768

re PR tree-optimization/82985 (ICC: in vn_nary_build_or_lookup_1, at tree-ssa-sccvn.c:1722)

2017-11-15  Richard Biener  <rguenther@suse.de>

PR tree-optimization/82985
* g++.dg/torture/pr82985.C: Likewise.

From-SVN: r254767

Fix fallout of -fsanitize=vptr.

2017-11-15  Martin Liska  <mliska@suse.cz>

* decl.c (begin_destructor_body): Use cp_build_fold_indirect_ref
instead of cp_build_indirect_ref.

From-SVN: r254765

Disable -Wreturn-type by default in all languages other from C++.

2017-11-15  Martin Liska  <mliska@suse.cz>

* tree-cfg.c (pass_warn_function_return::execute):
Compare warn_return_type for greater than zero.
2017-11-15  Martin Liska  <mliska@suse.cz>

* options.c (gfc_post_options):
Do not set default value of warn_return_type.
* trans-decl.c (gfc_trans_deferred_vars):
Compare warn_return_type for greater than zero.
(generate_local_decl): Likewise
(gfc_generate_function_code): Likewise.

From-SVN: r254764

Fix PR82941 and PR82942 by adding proper vzeroupper generation on SKX.

2017-11-15  Sebastian Peryt  <sebastian.peryt@intel.com>

gcc/

PR target/82941
PR target/82942
* config/i386/i386.c (pass_insert_vzeroupper): Modify gate condition
to return true on Xeon and not on Xeon Phi.
(ix86_check_avx256_register): Changed to ...
(ix86_check_avx_upper_register): ... this. Add extra check for
VALID_AVX512F_REG_OR_XI_MODE.
(ix86_avx_u128_mode_needed): Changed
ix86_check_avx256_register to ix86_check_avx_upper_register.
(ix86_check_avx256_stores): Changed to ...
(ix86_check_avx_upper_stores): ... this. Changed
ix86_check_avx256_register to ix86_check_avx_upper_register.
(ix86_avx_u128_mode_after): Changed
avx_reg256_found to avx_upper_reg_found. Changed
ix86_check_avx256_stores to ix86_check_avx_upper_stores.
(ix86_avx_u128_mode_entry): Changed
ix86_check_avx256_register to ix86_check_avx_upper_register.
(ix86_avx_u128_mode_exit): Ditto.
* config/i386/i386.h: (host_detect_local_cpu): New define.

2017-11-15  Sebastian Peryt  <sebastian.peryt@intel.com>

gcc/testsuite/

PR target/82941
PR target/82942
* gcc.target/i386/pr82941-1.c: New test.
* gcc.target/i386/pr82941-2.c: New test.
* gcc.target/i386/pr82942-1.c: New test.
* gcc.target/i386/pr82942-2.c: New test.

From-SVN: r254763

MAINTAINERS (write after approval): Add myself.

* MAINTAINERS (write after approval): Add myself.

From-SVN: r254762

Fix -Wreturn-type fallout in g++.dg/cpp0x/rv-trivial-bug.C on Solaris 10

* g++.dg/cpp0x/rv-trivial-bug.C (test2): Return a value.

From-SVN: r254760

[AArch64] Improve scheduling model for X-Gene

2017-11-15  Dominik Infuehr  <dominik.infuehr@theobroma-systems.com>

* config/arm/xgene1.md (xgene1): Split into automatons
xgene1_main, xgene1_decoder, xgene1_div, xgene1_simd.
(xgene1_f_load): Adjust reservations and/or types.
(xgene1_f_store): Likewise.
(xgene1_load_pair): Likewise.
(xgene1_store_pair): Likewise.
(xgene1_fp_load1): Likewise.
(xgene1_load1): Likewise.
(xgene1_store1): Likewise.
(xgene1_move): Likewise.
(xgene1_alu): Likewise.
(xgene1_simd): Likewise.
(xgene1_bfm): Likewise.
(xgene1_neon_load1): Likewise.
(xgene1_neon_store1): Likewise.
(xgene1_neon_logic): Likewise.
(xgene1_neon_st1): Likewise.
(xgene1_neon_ld1r): Likewise.
(xgene1_alu_cond): Added.
(xgene1_shift_reg): Likwise.
(xgene1_bfx): Likewise.
(xgene1_mul): Split into xgene1_mul32, xgene1_mul64.

From-SVN: r254759

re PR target/82981 (unnecessary __multi3 call for mips64r6 linux kernel)

PR target/82981
* internal-fn.c: Include gimple-ssa.h, tree-phinodes.h and
ssa-iterators.h.
(can_widen_mult_without_libcall): New function.
(expand_mul_overflow): If only checking unsigned mul overflow,
not result, and can do efficiently MULT_HIGHPART_EXPR, emit that.
Don't use WIDEN_MULT_EXPR if it would involve a libcall, unless
no other way works.  Add MULT_HIGHPART_EXPR + MULT_EXPR support.
(expand_DIVMOD): Formatting fix.
* expmed.h (expand_mult): Add NO_LIBCALL argument.
* expmed.c (expand_mult): Likewise.  Use OPTAB_WIDEN rather
than OPTAB_LIB_WIDEN if NO_LIBCALL is true, and allow it to fail.

* gcc.target/mips/pr82981.c: New test.

From-SVN: r254758

re PR tree-optimization/82977 (AddressSanitizer: heap-use-after-free in strlen_optimize_stmt .././../gcc/tree-ssa-strlen.c:2971)

PR tree-optimization/82977
* tree-ssa-strlen.c (strlen_optimize_stmt): Pass a reference to a copy
constructed temporary to strlen_to_stridx.put.

From-SVN: r254757

Coverage: remove -fkeep-inline-functions from coverage_flags.

2017-11-15  Martin Liska  <mliska@suse.cz>

* configure.ac: Remove -fkeep-inline-functions from coverage_flags.
* configure: Regenerate.

From-SVN: r254756

Use proper probability (PR target/82927)

2017-11-15  Martin Liska  <mliska@suse.cz>

PR target/82927
* config/sh/sh-mem.cc: Use proper probability for
REG_BR_PROB_NOTE.

From-SVN: r254755

Zero vptr in dtor for -fsanitize=vptr.

2017-11-15  Martin Liska  <mliska@suse.cz>

* decl.c (begin_destructor_body): In case of VPTR sanitization
(with disabled recovery), zero vptr in order to catch virtual calls
after lifetime of an object.
2017-11-15  Martin Liska  <mliska@suse.cz>

* g++.dg/ubsan/vptr-12.C: New test.

From-SVN: r254754

explow.c (anti_adjust_stack_and_probe_stack_clash): Avoid probing the red zone for stack_clash_protection_final_dynamic_probe...

* explow.c (anti_adjust_stack_and_probe_stack_clash): Avoid probing
the red zone for stack_clash_protection_final_dynamic_probe targets
when the total dynamic stack size is zero bytes.

From-SVN: r254753

tree-ssa-threadupdate.c (thread_through_all_blocks): Thread blocks is post order.

* tree-ssa-threadupdate.c (thread_through_all_blocks): Thread
blocks is post order.

From-SVN: r254752

introduce TDF_compare_debug, omit OBJ_TYPE_REF casts with it

for  gcc/ChangeLog

* dumpfile.h (TDF_COMPARE_DEBUG): New.
* final.c (rest_of_clean_state): Set it for the
-fcompare-debug dump.
* tree-pretty-print.c (dump_generic_node): Omit OBJ_TYPE_REF
class when TDF_COMPARE_DEBUG is set.

From-SVN: r254751

Use compare-debug for bootstrap-debug-lean to ignore != DW_AT_producer

Unlike bootstrap-debug, bootstrap-debug-lean used to pass compare using
the traditional compare command, because it compiled both stage2 and
stage3 with options that used to generate identical output
(-fcompare-debug= in stage2 vs -fcompare-debug in stage3).

Since we started adding relevant command-line flags to DW_AT_producer,
this is no longer the case, and stages 2 and 3 object files that differ
in nothing but the DW_AT_producer strings.

-fcompare-debug is short for -fcompare-debug=-gtoggle, so stage3
compiles twice, once with the normal options, once with toggled -g, to
then compare the temporary final dumps.  When enabled, both compilations
get from the driver an additional -frandom-seed flag (if none is given
explicitly).

-fcompare-debug= is short for -fno-compare-debug, disabling the second
compilation.

The difference between the DW_AT_producer lines are the different
-fcompare-debug flags, and the presence of the -frandom-seed flag in the
stage3 compilation.

It is easy and sensible enough to filter the -fcompare-debug flags out
of the DW_AT_producer string.  This option should never affect the
compilation output, it just determines whether or not to perform an
additional compilation that should produce the same executable output.

However, dropping -frandom-seed is not quite correct, in that it might
have other consequences on the compilation.  So, it makes little sense
to make the effort to drop it when it's implicit; for those comparing
compiler output differences, it might even hint at what causes, and
could fix, difficult to explain differences, namely, explicitly
supplying -frandom-seed options.

They don't seem to matter for compiler bootstraps, though, at least
for now, so we can safely refrain from issuing -frandom-seed (or maybe
we already issue it where needed :-), and then, for
bootstrap-debug-lean, use the compare-debug script, that strips out
debug information before comparing the object files.

for  config/ChangeLog

* bootstrap-debug-lean.mk (do-compare): Use the
contrib/compare-debug script.

for  gcc/ChangeLog

* dwarf2out.c (gen_producer_string): Discard
OPT_fcompare_debug.

From-SVN: r254750

Add __builtin_tgmath for better tgmath.h implementation (bug 81156).

Various implementations of C99/C11 <tgmath.h> have the property that
their macro expansions contain many copies of the macro arguments, so
resulting in exponential blowup of the size of macro expansions where
a call to such a macro contains other such calls in the macro
arguments.

This patch adds a (C-only) language feature __builtin_tgmath designed
to avoid this problem by implementing the <tgmath.h> function
selection rules directly in the compiler.  The effect is that
type-generic macros can be defined simply as

#define pow(a, b) __builtin_tgmath (powf, pow, powl, \
                                    cpowf, cpow, cpowl, a, b)

as in the example added to the manual, with each macro argument
expanded exactly once.  The details of __builtin_tgmath are as
described in the manual.  This is C-only since C++ uses function
overloading and just defines <ctgmath> to include <ccomplex> and
<cmath>.

__builtin_tgmath handles C99/C11 type-generic macros, and _FloatN,
_FloatNx and decimal floating-point types (following the proposed
resolution to the floating-point TS DR#9 that makes the rules for
finding a common type from arguments to a type-generic macro follow
the usual arithmetic conversions after adjustment of integer arguments
to _Decimal64 or double - or to _Complex double in the case of GNU
complex integer arguments).

Type-generic macros for functions from TS 18661 that round their
results to a narrower type are handled, but there are still some
unresolved questions regarding such macros so further changes in that
regard may be needed in future.  The current implementation follows an
older version of the DR#13 resolution (allowing a function for a
wide-enough argument type to be selected if no exactly-matching
function is available), but with appropriate calls to __builtin_tgmath
is still fully compatible with the latest version of the resolution
(not yet in the DR log), and allowing such not-exactly-matching
argument types to be chosen in that case avoids needing another
special case to treat integers as _Float64 instead of double in
certain cases.

Regarding other possible language/library features, not currently
implemented in GCC:

* Imaginary types could be naturally supported by allowing cases where
  the type-generic type is an imaginary type T and arguments or return
  types may be T (as at present), or the corresponding real type to T
  (as at present), or (new) the corresponding real type if T is real
  or imaginary but T if T is complex.  (tgmath.h would need a series
  of functions such as

  static inline _Imaginary double
  __sin_imag (_Imaginary double __x)
  {
    return _Imaginary_I * sinh (__imag__ __x);
  }

  to be used in __builtin_tgmath calls.)

* __builtin_tgmath would use the constant rounding direction in the
  presence of support for the FENV_ROUND / FENV_DEC_ROUND pragmas.
  Support for those would also require a new __builtin_<something> to
  cause a non-type-generic call to use the constant rounding
  direction (it seems cleaner to add a new __builtin_<something> when
  required than to make __builtin_tgmath handle a non-type-generic
  case with only one function argument).

* TS 18661-5 __STDC_TGMATH_OPERATOR_EVALUATION__ would require new
  __builtin_<something> that evaluates with excess range and precision
  like arithmetic operators do.

* The proposed C bindings for IEEE 754-2018 augmented arithmetic
  operations involve struct return types.  As currently implemented
  __builtin_tgmath does not handle those, but support could be added.

There are many error cases that the implementation diagnoses.  I've
tried to ensure reasonable error messages for erroneous uses of
__builtin_tgmath, but the errors for erroneous uses of the resulting
type-generic macros (that is, when the non-function arguments have
inappropriate types) are more important as they are more likely to be
seen by users.

GCC's own tgmath.h, as used for some targets, is updated in this
patch.  I've tested those changes minimally, via adjusting
gcc.dg/c99-tgmath-* locally to use that tgmath.h version.  I've also
run the glibc testsuite (which has much more thorough tests of
correctness of tgmath.h function selection) with a glibc patch to use
__builtin_tgmath in glibc's tgmath.h.

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

PR c/81156

gcc:
* doc/extend.texi (Other Builtins): Document __builtin_tgmath.
* ginclude/tgmath.h (__tg_cplx, __tg_ldbl, __tg_dbl, __tg_choose)
(__tg_choose_2, __tg_choose_3, __TGMATH_REAL_1_2)
(__TGMATH_REAL_2_3): Remove macros.
(__TGMATH_CPLX, __TGMATH_CPLX_2, __TGMATH_REAL, __TGMATH_REAL_2)
(__TGMATH_REAL_3, __TGMATH_CPLX_ONLY): Define using
__builtin_tgmath.
(frexp, ldexp, nexttoward, scalbn, scalbln): Define using
__TGMATH_REAL_2.
(remquo): Define using __TGMATH_REAL_3.

gcc/c:
* c-parser.c (check_tgmath_function): New function.
(enum tgmath_parm_kind): New enum.
(c_parser_postfix_expression): Handle __builtin_tgmath.

gcc/c-family:
* c-common.c (c_common_reswords): Add __builtin_tgmath.
* c-common.h (enum rid): Add RID_BUILTIN_TGMATH.

gcc/testsuite:
* gcc.dg/builtin-tgmath-1.c, gcc.dg/builtin-tgmath-2.c,
gcc.dg/builtin-tgmath-err-1.c, gcc.dg/builtin-tgmath-err-2.c,
gcc.dg/dfp/builtin-tgmath-dfp-err.c,
gcc.dg/dfp/builtin-tgmath-dfp.c: New tests.

From-SVN: r254749

compiler: remove LHS/RHS context determination for var exprs

    Remove the code that determined LHS/RHS context for var expressions
    (was needed previously for some instances of the back end). LHS/RHS
    context is computed by the back end in all cases, so no need to have
    this code in the front end any more.

    Reviewed-on: https://go-review.googlesource.com/77510

* go-gcc.cc (var_expression): Remove Varexpr_context parameter.

From-SVN: r254748

vr-values.c: New file with contents extracted from tree-vrp.c.

* vr-values.c: New file with contents extracted from tree-vrp.c.
* Makefile.in (OBJS): Add vr-values.o
* tree-vrp.h (set_value_range_to_nonnull): Prototype.
(set_value_range, set_and_canonicalize_value_range): Likewise.
(vrp_bitmap_equal_p, range_is_nonnull): Likewise.
(value_range_constant_singleton, symbolic_range_p): Likewise.
(compare_values, compare_values_warnv, vrp_val_is_min): Likewise.
(vrp_val_is_max, copy_value_range, set_value_range_to_value): Likewise.
(extract_range_from_binary_expr_1, vrp_val_min, vrp_val_max): Likewise.
(set_value_range_to_null, range_int_cst_p, opreand_less_p): Likewise.
(find_case_label_range, find_case_label_index): Likewise.
(zero_nonzero_bits_from_vr, overflow_comparison_p): Likewise.
(range_int_cst_singleton_p, value_inside_range): Likewise.
(get_single_symbol): Likewise.
(switch_update): Move structure definition here.
(to_remove_edges, to_update_switch_stmts): Provide externs.
* tree-vrp.c: Move all methods for vr-values class to vr-values.c
(vrp_val_max, vrp_val_min, vrp_val_is_max): Make externally visible.
(vrp_val_is_min, set_value_range): Likewise.
(set_and_canonicalize_value_range, copy_value_range): Likewise.
(set_value_range_to_value, set_value_range_to_nonnull): Likewise.
(set_value_range_to_null, vrp_bitmap_equal_p): Likewise.
(range_is_nonnull, range_int_cst_p): Likewwise.
(range_int_cst_singleton_p, symbolic_range_p): Likewise.
(get_single_symbol, operand_less_p): Likewise
(compare_values_warnv, compare_values): Likewise.
(value_inside_range, value_range_constant_singleton): Likewise.
(zero_nonzero_bitgs_from_vr): Likewise.
(extract_range_from_binary_expr_1): Likewise.
(overflow_comparison_p): Likewise.
(to_remove_edges, to_update_switch_stmts): Likewise.
(find_case_label-index, find_case_label_range): Likewise.
(switch_update, set_value_range_to_nonnegative): Remove.
(set_value_range_to_truthvalue): Likewise.
(symbolic_range_based_on_p, gimple_assign_nonzero_p): Likewise.
(gimple_stmt_nonzero_p, compare_ranges): Likewise.
(compare_range_with_value, vrp_valueize, vrp_valueize_1): Likewise.
(find_case_label_ranges, test_for_singularity): Likewise.
(range_fits_type_p, simplify_conversion_using_ranges): LIkewise.
(x_vr_values): Move to its remaining use site.

From-SVN: r254747

Daily bump.

From-SVN: r254746

rs6000.md (bswapdi2): On 32-bit ISA 3.0, don't generate the XXBRD instruction.

2017-11-14  Michael Meissner  <meissner@linux.vnet.ibm.com>

* config/rs6000/rs6000.md (bswapdi2): On 32-bit ISA 3.0, don't
generate the XXBRD instruction.

From-SVN: r254742

rs6000-c.c (is_float128_p): New helper function.

[gcc]
2017-11-14  Michael Meissner  <meissner@linux.vnet.ibm.com>

* config/rs6000/rs6000-c.c (is_float128_p): New helper function.
(rs6000_builtin_type_compatible): Treat _Float128 and long double
as being compatible if -mabi=ieeelongdouble.
* config/rs6000/rs6000-builtin.def (BU_FLOAT128_HW_1): New macros
to setup float128 built-ins with hardware support.
(BU_FLOAT128_HW_2): Likewise.
(BU_FLOAT128_HW_3): Likewise.
(BU_FLOAT128_HW_VSX_1): Likewise.
(BU_FLOAT128_HW_VSX_2): Likewise.
(scalar_extract_expq): Change float128 built-in functions to
accommodate having both KFmode and TFmode functions.  Use the
KFmode variant as the default.
(scalar_extract_sigq): Likewise.
(scalar_test_neg_qp): Likewise.
(scalar_insert_exp_q): Likewise.
(scalar_insert_exp_qp): Likewise.
(scalar_test_data_class_qp): Likewise.
(sqrtf128_round_to_odd): Delete processing the round to odd
built-in functions as special built-in functions, and define them
as float128 built-ins.  Use the KFmode variant as the default.
(truncf128_round_to_odd): Likewise.
(addf128_round_to_odd): Likewise.
(subf128_round_to_odd): Likewise.
(mulf128_round_to_odd): Likewise.
(divf128_round_to_odd): Likewise.
(fmaf128_round_to_odd): Likewise.
* config/rs6000/rs6000.c (rs6000_expand_binop_builtin): Add
support for KFmode and TFmode xststdcqp calls.
(rs6000_expand_builtin): If long double is IEEE 128-bit floating
point, switch the built-in handlers for the get/set float128
exponent, get float128 mantissa, float128 test built-ins, and the
float128 round to odd built-in functions.  Eliminate creating the
float128 round to odd built-in functions as special built-ins.
(rs6000_init_builtins): Eliminate special creation of the float128
round to odd built-in functions.
* config/rs6000/vsx.md (xsxexpqp_<mode>): Change float128 built-in
function insns to support both TFmode and KFmode variants.
(xsxsigqp_<mode>): Likewise.
(xsiexpqpf_<mode>): Likewise.
(xsiexpqp_<mode>): Likewise.
(xststdcqp_<mode>): Likewise.
(xststdcnegqp_<mode>): Likewise.
(xststdcqp_<mode>): Likewise.

[gcc/testsuite]
2017-11-14  Michael Meissner  <meissner@linux.vnet.ibm.com>

* gcc.target/powerpc/float128-hw4.c: New test.

From-SVN: r254740

tree-ssa-threadupdate.c (compute_path_counts): Remove unused path_in_freq_ptr parameter.

* tree-ssa-threadupdate.c (compute_path_counts): Remove
unused path_in_freq_ptr parameter.
(ssa_fix_duplicate_block_edges): Do not pass around path_in_freq

From-SVN: r254739

ipa-inline.c (edge_badness): Dump sreal frequency.

* ipa-inline.c (edge_badness): Dump sreal frequency.
(compute_inlined_call_time): Match natural implementaiton ...
* ipa-fnsummary.c (estimate_edge_size_and_time): ... here; remove
forgotten division by CGRAPH_FREQ_BASE.

From-SVN: r254738

Adapt Solaris 12 references

libgcc:
* config.host (*-*-solaris2*): Adapt comment for Solaris 12
renaming.
* config/sol2/crtpg.c (__start_crt_compiler): Likewise.
* configure.ac (libgcc_cv_solaris_crts): Likewise.
* configure: Regenerate.

gcc:
* config.gcc (*-*-solaris2*): Enable default_use_cxa_atexit since
Solaris 11.  Update comment.
* configure.ac (gcc_cv_ld_pid): Adapt comment for Solaris 12
renaming.
* config/sol2.h (STARTFILE_SPEC): Likewise.
* configure: Regenerate.

gcc/testsuite:
* lib/target-supports.exp (check_effective_target_pie): Adapt
comment for Solaris 12 renaming.

* gcc.dg/torture/pr60092.c: Remove *-*-solaris2.11* dg-xfail-run-if.

From-SVN: r254737

Implement LWG 2733 and LWG 2759

* include/experimental/numeric (gcd): Reject cv-qualified bool.
(lcm): Likewise.
* include/std/numeric (gcd): Likewise.
(lcm): Likewise.
* testsuite/26_numerics/gcd/gcd_neg.cc: Add tests and adjust.
* testsuite/26_numerics/lcm/lcm_neg.cc: Likewise.

From-SVN: r254736

Fix typo in std::wbuffer_convert

* include/bits/locale_conv.h (wbuffer_convert::_M_conv_get): Fix typo.
* testsuite/22_locale/conversions/buffer/3.cc: New test.

From-SVN: r254735

rs6000.c (swap_endian_selector_for_mode): Remove le_ and be_ prefixes to swap* variables.

gcc/ChangeLog:

2017-11-14  Carl Love  <cel@us.ibm.com>

* config/rs6000/rs6000.c (swap_endian_selector_for_mode): Remove
le_ and be_ prefixes to swap* variables.  Remove
if (VECTOR_ELT_ORDER_BIG) statement. Remove E_V16QImode case
statements.

gcc/testsuite/ChangeLog:

2017-11-14  Carl Love  <cel@us.ibm.com>

* builtins-revb-runnable.c (dg-do run): Add lp64 directive.  Fix
indentation of printf and abort statements.
* p9-xxbr-1.c (dg-do compile): Add lp64 && p9vector_h directives.

From-SVN: r254732

Support GTY((cache)) on hash_map.

gcc/
* hash-traits.h (ggc_remove): Add ggc_maybe_mx member function.
(ggc_cache_remove): Override it instead of ggc_mx.
* hash-table.h (gt_ggc_mx): Call it instead of ggc_mx.
(gt_cleare_cache): Call ggc_mx instead of gt_ggc_mx.
* hash-map-traits.h (simple_hashmap_traits): Add maybe_mx member.
(simple_cache_map_traits): Override maybe_mx.
* hash-map.h (hash_entry): Add ggc_maybe_mx and keep_cache_entry.
(hash_map): Friend gt_cleare_cache.
(gt_cleare_cache): New.
* tree.h (tree_cache_traits): New hash_map traits class.
(tree_cache_map): New typedef.
gcc/cp/
* decl.c (decomp_type_table): Use tree_cache_map.
* init.c (nsdmi_inst): Likewise.
* pt.c (defarg_ints): Likewise.
* cp-objcp-common.c (cp_get_debug_type): Likewise.

From-SVN: r254731

tree-cfgcleanup.c (cleanup_control_expr_graph): Remove first_p paramter and handling.

2017-11-14  Richard Biener  <rguenther@suse.de>

* tree-cfgcleanup.c (cleanup_control_expr_graph): Remove first_p
paramter and handling.
(cleanup_control_flow_bb): Likewise.
(cleanup_control_flow_pre): New helper performing a DFS walk
to call cleanup_control_flow_bb in PRE order.
(cleanup_tree_cfg_1): Do the first phase of cleanup_control_flow_bb
via cleanup_control_flow_pre.

From-SVN: r254730

libgo: adapt Solaris 12 references

    With the change in the Solaris release model (no more major releases
    like Solaris 12 but only minor ones like 11.4), the Solaris 12
    references in GCC need to be adapted.

    Patch by Rainer Orth.

    Reviewed-on: https://go-review.googlesource.com/77490

From-SVN: r254729

[Patch AArch64] Stop generating BSL for simple integer code

Turn aarch64_simd_bsldi_internal in to an insn_and_split that
knows to split back to integer operations if the register allocation
falls that way. Do this to avoid having to move between integer and
Advanced SIMD register files just for a single BSL.

---
gcc/

* config/aarch64/aarch64-simd.md
(aarch64_simd_bsl<mode>_internal): Remove DImode.
(*aarch64_simd_bsl<mode>_alt): Likewise.
(aarch64_simd_bsldi_internal): New.
(aarch64_simd_bsldi_alt): Likewise.

gcc/testsuite/

* gcc.target/aarch64/bsl-idiom.c: New.
* gcc.target/aarch64/copysign-bsl.c: New.

From-SVN: r254727

tracer.c (better_p): Do not compare frequencies.

* tracer.c (better_p): Do not compare frequencies.
* reg-stack.c (better_edge): Likewise.
* shrink-wrap.c (try_shrink_wrapping): Do not convert to gcov counts
and back.

From-SVN: r254726

auto-profile.c (afdo_annotate_cfg): Use update_max_bb_count.

* auto-profile.c (afdo_annotate_cfg): Use update_max_bb_count.
* cgraphunit.c (cgraph_node::expand_thunk): Use update_max_bb_count.
* ipa-utils.c (ipa_merge_profiles): Use update_max_bb_count.
* lto-streamer-in.c (input_function): Use update_max_bb_count.
* omp-expand.c (expand_omp_taskreg): Use update_max_bb_count.
* predict.c (maybe_hot_frequency_p): Inline to ...
(maybe_hot_count_p): ... here; rewrite to counts.
(counts_to_freqs): Rename to ...
(update_max_bb_count): ... this one.
(expensive_function_p): Use counts.
(estimate_bb_frequencies): Update.
(rebuild_frequencies): Update.
* predict.h (counts_to_freqs): Rename to ...
(update_max_bb_count): ... this one.
* profile.c (compute_branch_probabilities): Add debug info
* tree-inline.c (expand_call_inline): Update debug info.
(optimize_inline_calls): Use update_max_bb_count..
(tree_function_versioning): Use update_max_bb_count..
* value-prof.c (gimple_value_profile_transformations):
Do not use update_max_bb_count.

From-SVN: r254725

ipa-inline.c (compute_uninlined_call_time, [...]): always use frequencies.

* ipa-inline.c (compute_uninlined_call_time, compute_inlined_call_time):
always use frequencies.

From-SVN: r254724

Allow asyncwait-1.c to run for non-nvidia devices

2017-11-14  Tom de Vries  <tom@codesourcery.com>

* testsuite/libgomp.oacc-c-c++-common/asyncwait-1.c: Allow to run for
non-nvidia devices.

From-SVN: r254723

Require alloca for c-c++-common/Wstringop-truncation.c

2017-11-14  Tom de Vries  <tom@codesourcery.com>

* c-c++-common/Wstringop-truncation.c: Require effective target alloca.

From-SVN: r254722

bb-reorder.c: Remove frequencies from comments.

* bb-reorder.c: Remove frequencies from comments.
(better_edge_p): Use profile counts.
(find_traces): Dump profile counts.
(rotate_loop): Use profile counts.
(find_traces_1_round): Likewise.
(connect_better_edge_p): Use counts instead of probabilities for
reverse walk.
(copy_bb_p): Drop early check for non-0 frequency.
(sanitize_hot_paths): Update comments.

From-SVN: r254721

ipa-split.c (struct split_point): Add count.

* ipa-split.c (struct split_point): Add count.
(consider_split): Do not compute incoming frequency; compute incoming
count and store it to split_point.
(split_function): Set count of the call to split part correctly.

* testsuite/gcc.dg/tree-ssa/fnsplit-2.c: New testcase.

From-SVN: r254720

re PR fortran/78240 (ICE in match_clist_expr, at fortran/decl.c:728)

2017-11-13  Fritz Reese <fritzoreese@gmail.com>

    PR fortran/78240

    gcc/fortran/ChangeLog:

PR fortran/78240
* decl.c (match_clist_expr): Replace gcc_assert with proper
handling of bad result from spec_size().
* resolve.c (check_data_variable): Avoid NULL dereference when passing
locus to gfc_error.

    gcc/testsuite/ChangeLog:

PR fortran/78240
* gfortran.dg/dec_structure_23.f90: New.
* gfortran.dg/pr78240.f90: New.

From-SVN: r254718

Daily bump.

From-SVN: r254717

altivec.md (altivec_vsumsws_be): Add define_expand.

gcc/ChangeLog:

2017-11-13  Carl Love  <cel@us.ibm.com>

* config/rs6000/altivec.md (altivec_vsumsws_be): Add define_expand.

gcc/testsuite/ChangeLog:

2017-11-13 Carl Love  <cel@us.ibm.com>

* gcc.target/powerpc/builtin-vec-sums-be-int.c: New test file.

From-SVN: r254714

Capture adjustments for P0588R1.

* semantics.c (process_outer_var_ref): Capture variables when
they are named; complain about non-capture uses when odr-used.
* expr.c (mark_use): Rvalue use looks through capture proxy.
* constexpr.c (potential_constant_expression_1): Improve error about
use of captured variable.
* lambda.c (need_generic_capture, dependent_capture_r)
(do_dependent_capture, processing_nonlambda_template): Remove.
* call.c (build_this): Remove uses of the above.
* decl.c (cp_finish_decl): Likewise.
* semantics.c (maybe_cleanup_point_expr)
(maybe_cleanup_point_expr_void, finish_goto_stmt)
(maybe_convert_cond): Likewise.
* typeck.c (check_return_expr): Likewise.

From-SVN: r254713

Defer folding of *&.

* typeck.c (cp_build_fold_indirect_ref): New.
(cp_build_indirect_ref_1): Split out from cp_build_indirect_ref.
Add 'fold' parameter.
* cp-tree.h: Declare cp_build_fold_indirect_ref.
* call.c, class.c, cp-ubsan.c, decl.c, except.c, init.c, lambda.c,
parser.c, rtti.c, tree.c, typeck.c, typeck2.c: Use it.
* parser.c (do_range_for_auto_deduction): Use RO_UNARY_STAR.
(cp_convert_range_for): Likewise.
* typeck2.c (build_x_arrow): Use RO_ARROW.

From-SVN: r254712

Fix cp-ubsan typo.

* cp-ubsan.c (cp_ubsan_check_member_access_r): Fix handling of
INDIRECT_REF of ADDR_EXPR.

From-SVN: r254711

PR c++/82360 - ICE with static_cast in template.

* call.c (perform_direct_initialization_if_possible): Check
processing_template_decl.
* typeck.c (build_static_cast_1): Likewise.

From-SVN: r254710

Implement __VA_OPT__

This implements __VA_OPT__, a new preprocessor feature added in C++2A.
The paper can be found here:

http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2017/p0306r4.html

gcc/ChangeLog

        * doc/cpp.texi (Variadic Macros): Document __VA_OPT__.

gcc/testsuite/ChangeLog

        * c-c++-common/cpp/va-opt-pedantic.c: New file.
        * c-c++-common/cpp/va-opt.c: New file.
        * c-c++-common/cpp/va-opt-error.c: New file.

libcpp/ChangeLog

        * pch.c (cpp_read_state): Set n__VA_OPT__.
        * macro.c (vaopt_state): New class.
        (_cpp_arguments_ok): Check va_opt flag.
        (replace_args, create_iso_definition): Use vaopt_state.
        * lex.c (lex_identifier_intern): Possibly issue errors for
        __VA_OPT__.
        (lex_identifier): Likewise.
        (maybe_va_opt_error): New function.
        * internal.h (struct lexer_state) <va_args_ok>: Update comment.
        (struct spec_nodes) <n__VA_OPT__>: New field.
        * init.c (struct lang_flags) <va_opt>: New field.
        (lang_defaults): Add entries for C++2A.  Update all entries for
        va_opt.
        (cpp_set_lang): Initialize va_opt.
        * include/cpplib.h (struct cpp_options) <va_opt>: New field.
        * identifiers.c (_cpp_init_hashtable): Initialize n__VA_OPT__.

From-SVN: r254707

rs6000-c.c (altivec_overloaded_builtins): Add support for builtins...

gcc/ChangeLog:

2017-11-13  Carl Love  <cel@us.ibm.com>

* config/rs6000/rs6000-c.c (altivec_overloaded_builtins):
Add support for builtins:
unsigned int vec_first_{,miss}_match_{,or_eos}index,
vector {un,}signed {char,int,short},
vector {un,}signed {char,int,short}) arguments.
* config/rs6000/rs6000-builtin.def (VFIRSTMATCHINDEX,
VFIRSTMATCHOREOSINDEX, VFIRSTMISMATCHINDEX, VFIRSTMISMATCHOREOSINDEX):
Add BU_P9V_AV_2 expansions for the builtins.
* config/rs6000/altivec.h (vec_first_match_index,
vec_first_mismatch_index, vec_first_match_or_eos_index,
vec_first_mismatch_or_eos_index): Add #defines for the builtins.
* config/rs6000/rs6000-protos.h (bytes_in_mode): Add
new extern declaration.
* config/rs6000/rs6000.c (bytes_in_mode): Add new function.
* config/rs6000/vsx.md (first_match_index_<mode>,
first_match_or_eos_index_<mode>, first_mismatch_index_<mode>,
first_mismatch_or_eos_index_<mode>): Add define expand.
(vctzlsbb_<mode>): Add mode field to define_insn for vctzlsbb.
* doc/extend.texi: Update the built-in documenation file for the new
built-in functions.

gcc/testsuite/ChangeLog:

2017-11-13  Carl Love  <cel@us.ibm.com>

* gcc.target/powerpc/builtins-6-p9-runnable.c: Add new runnable test.
* gcc.target/powerpc/vsu/vec-cnttz-lsbb-2.c: Update expected error
message.

From-SVN: r254704

libcpp: move line typedef and column-numbering comment to top of file

The description of our 1-based column-numbering convention was in
a non-obvious place withn line-map.h; this patch moves it to the top
of that header.

libcpp/ChangeLog:
* include/line-map.h (linenum_type): Move this typedef and the
comment describing column numbering to near the top of the file.

From-SVN: r254703

match.pd: Convert fminf<N>...

[gcc]
2017-11-13  Michael Meissner  <meissner@linux.vnet.ibm.com>

* match.pd: Convert fminf<N>, fminf<N>x, fmax<N>, and fmax<N>x
into the min/max operations for _Float<N> and _Float<N>X types.

[gcc/testsuite]
2017-11-13  Michael Meissner  <meissner@linux.vnet.ibm.com>

* gcc.target/powerpc/float128-minmax.c: New test.

From-SVN: r254702

re PR lto/81351 (Many LTO testcases FAIL)

PR lto/81351
* dwarf2out.c (do_eh_frame): New static variable.
(dwarf2out_begin_prologue): Set it.
(dwarf2out_frame_finish): Test it instead of dwarf2out_do_eh_frame.

From-SVN: r254700

* tree-ssa-coalesce.c (coalesce_cost): Fix formating.

From-SVN: r254699

* tree-ssa-sink.c (select_best_block): Do not use frequencies.

From-SVN: r254698

re PR lto/81351 (Many LTO testcases FAIL)

PR lto/81351
* debug.h (dwarf2out_do_eh_frame): Declare.
* dwarf2cfi.c (dwarf2out_do_eh_frame): New predicate.
(dwarf2out_do_frame): Use it.
(dwarf2out_do_cfi_asm): Likewise.
* dwarf2out.c (dwarf2out_frame_finish): Likewise.
(dwarf2out_assembly_start): Likewise.
(dwarf2out_begin_prologue): Fix comment.
* toplev.c (compile_file): Always call dwarf2out_frame_finish
if the target needs either debug or unwind DWARF2 info.
* lto-opts.c (lto_write_options): Do not save -fexceptions,
-fnon-call-exceptions, -ffp-contract, -fmath-errno, -fsigned-zeros,
-ftrapping-math, -ftrapv and -fwrapv.

From-SVN: r254697

cgraph.c (cgraph_edge::sreal_frequency): New function.

* cgraph.c (cgraph_edge::sreal_frequency): New function.
* cgraph.h (cgraph_edge::sreal_frequency): Declare.
* ipa-fnsummary.c (dump_ipa_call_summary): Use sreal_frequency.
(estimate_edge_size_and_time): Likewise.
(ipa_merge_fn_summary_after_inlining): Likewise.
* ipa-inline.c (cgraph_freq_base_rec): Remove.
(compute_uninlined_call_time): Use sreal_frequency.
(compute_inlined_call_time): Likewise.
(ipa_inline): Do not initialize cgraph_freq_base_rec.
* profile-count.c: Include sreal.h.
(profile_count::to_sreal_scale): New.
* profile-count.h: Forward declare sreal.
(profile_count::to_sreal_scale): Declare.

From-SVN: r254696

Remove the null check from placement new in all modes

gcc/cp/

Remove the null check from placement new in all modes
* init.c (build_new_1): Don't do a null check for
a namespace-scope non-replaceable placement new
in any mode unless -fcheck-new is provided.

testsuite/

Remove the null check from placement new in all modes
* g++.dg/init/pr35878_1.C: Adjust.
* g++.dg/init/pr35878_4.C: New.
* g++.dg/torture/pr48695.C: Adjust.
* g++.dg/tree-ssa/pr31146-2.C: Likewise.
* g++.dg/tree-ssa/pr41428.C: Adjust.

From-SVN: r254694

[ARM] Fix more -Wreturn-type fallout (gcc.target/arm/pr67989.C).

2017-11-13  Christophe Lyon  <christophe.lyon@linaro.org>

* gcc.target/arm/pr67989.C: Add -Wno-return-type to
dg-additional-options.

From-SVN: r254692

[Diagnostic Patch] don't print column zero

https://gcc.gnu.org/ml/gcc-patches/2017-10/msg01911.html
* diagnostic.c (maybe_line_and_column): New.
(diagnostic_get_location_text): Use it.
(diagnostic_report_current_module): Likewise.
(test_diagnostic_get_location_text): Add tests.

* lib/gcc-dg.exp (process-message): Use -: for no column.
* c-c++-common/cilk-plus/CK/cilk_for_grain_errors.c: Mark elided
column messages.
* c-c++-common/cpp/pr58844-1.c: Likewise.
* c-c++-common/cpp/pr58844-2.c: Likewise.
* c-c++-common/cpp/warning-zero-location.c: Likewise.
* g++.dg/diagnostic/pr77949.C: Likewise.
* g++.dg/gomp/macro-4.C: Likewise.
* gcc.dg/Wunknownprag.c: Likewise.
* gcc.dg/builtin-redefine.c: Likewise.
* gcc.dg/cpp/Wunknown-pragmas-1.c: Likewise.
* gcc.dg/cpp/Wunused.c: Likewise.
* gcc.dg/cpp/misspelled-directive-1.c: Likewise.
* gcc.dg/cpp/redef2.c: Likewise.
* gcc.dg/cpp/redef3.c: Likewise.
* gcc.dg/cpp/redef4.c: Likewise.
* gcc.dg/cpp/trad/Wunused.c: Likewise.
* gcc.dg/cpp/trad/argcount.c: Likewise.
* gcc.dg/cpp/trad/comment-3.c: Likewise.
* gcc.dg/cpp/trad/comment.c: Likewise.
* gcc.dg/cpp/trad/defined.c: Likewise.
* gcc.dg/cpp/trad/directive.c: Likewise.
* gcc.dg/cpp/trad/funlike-3.c: Likewise.
* gcc.dg/cpp/trad/funlike.c: Likewise.
* gcc.dg/cpp/trad/literals-2.c: Likewise.
* gcc.dg/cpp/trad/macro.c: Likewise.
* gcc.dg/cpp/trad/pr65238-4.c: Likewise.
* gcc.dg/cpp/trad/recurse-1.c: Likewise.
* gcc.dg/cpp/trad/recurse-2.c: Likewise.
* gcc.dg/cpp/trad/redef2.c: Likewise.
* gcc.dg/cpp/ucnid-11.c: Likewise.
* gcc.dg/cpp/unc1.c: Likewise.
* gcc.dg/cpp/unc2.c: Likewise.
* gcc.dg/cpp/unc3.c: Likewise.
* gcc.dg/cpp/unc4.c: Likewise.
* gcc.dg/cpp/undef2.c: Likewise.
* gcc.dg/cpp/warn-redefined-2.c: Likewise.
* gcc.dg/cpp/warn-redefined.c: Likewise.
* gcc.dg/cpp/warn-unused-macros-2.c: Likewise.
* gcc.dg/cpp/warn-unused-macros.c: Likewise.
* gcc.dg/empty-source-2.c: Likewise.
* gcc.dg/empty-source-3.c: Likewise.
* gcc.dg/gomp/macro-4.c: Likewise.
* gcc.dg/noncompile/pr35447-1.c: Likewise.
* gcc.dg/plugin/location-overflow-test-1.c: Likewise.
* gcc.dg/pr20245-1.c: Likewise.
* gcc.dg/pr28419.c: Likewise.
* gcc.dg/rtl/truncated-rtl-file.c: Likewise.
* gcc.dg/unclosed-init.c: Likewise.

From-SVN: r254691

ieee_8.f90: xfail for aarch64*-*-gnu*

gcc/testsuite/ChangeLog:

2017-11-13  Charles Baylis  <charles.baylis@linaro.org>

        * gfortran.dg/ieee/ieee_8.f90: xfail for aarch64*-*-gnu*

From-SVN: r254689

[Documentation] Fix latency in pipeline description example

2017-11-09  Luis Machado  <luis.machado@linaro.org>

gcc/
* doc/md.texi (Specifying processor pipeline description): Fix
incorrect latency for the div instruction example.

From-SVN: r254680

re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)

PR tree-optimization/78821
* gimple-ssa-store-merging.c (compatible_load_p): Don't require
that bit_not_p is the same.
(imm_store_chain_info::coalesce_immediate_stores): Likewise.
(split_group): Count precisely bit_not_p bits in each statement.
(invert_op): New function.
(imm_store_chain_info::output_merged_store): Use invert_op to
emit BIT_XOR_EXPR with a xor_mask instead of BIT_NOT_EXPR if some
but not all orig_stores have BIT_NOT_EXPR in the corresponding spots.

* gcc.dg/store_merging_15.c: New test.

From-SVN: r254679

GCOV: remove typedef of coverage_t.

2017-11-13  Martin Liska  <mliska@suse.cz>

* gcov.c (struct coverage_info): Remove typedef of coverage_t.
(struct source_info): Likewise.
(add_branch_counts): Likewise.
(add_line_counts): Likewise.
(function_summary): Likewise.
(output_intermediate_line): Likewise.
(generate_results): Likewise.

From-SVN: r254678

GCOV: remove typedef for block_t

2017-11-13  Martin Liska  <mliska@suse.cz>

* gcov.c (struct block_info): Remove typedef for block_t.
(struct line_info): Likewise.
(line_info::has_block): Likewise.
(EXIT_BLOCK): Likewise.
(unblock): Likewise.
(circuit): Likewise.
(get_cycles_count): Likewise.
(process_file): Likewise.
(read_graph_file): Likewise.
(solve_flow_graph): Likewise.
(find_exception_blocks): Likewise.
(add_line_counts): Likewise.
(accumulate_line_info): Likewise.
(output_line_details): Likewise.

From-SVN: r254677

GCOV: remove typedef for arc_t

2017-11-13  Martin Liska  <mliska@suse.cz>

* gcov.c (struct arc_info): Remove typedef for arc_t.
(struct line_info): Likewise.
(add_branch_counts): Likewise.
(output_branch_count): Likewise.
(function_info::~function_info): Likewise.
(circuit): Likewise.
(output_intermediate_line): Likewise.
(read_graph_file): Likewise.
(solve_flow_graph): Likewise.
(find_exception_blocks): Likewise.
(add_line_counts): Likewise.
(accumulate_line_info): Likewise.
(output_line_details): Likewise.
(output_function_details): Likewise.

From-SVN: r254676

GCOV: remove typedef for function_t

2017-11-13  Martin Liska  <mliska@suse.cz>

* gcov.c (struct function_info): Remove typedef for function_t.
(struct source_info): Likewise.
(source_info::get_functions_at_location): Likewise.
(solve_flow_graph): Likewise.
(find_exception_blocks): Likewise.
(add_line_counts): Likewise.
(output_intermediate_file): Likewise.
(process_file): Likewise.
(generate_results): Likewise.
(release_structures): Likewise.
(read_graph_file): Likewise.
(read_count_file): Likewise.
(accumulate_line_counts): Likewise.
(output_lines): Likewise.

From-SVN: r254675

GCOV: introduce vector for function_info::counts.

2017-11-13  Martin Liska  <mliska@suse.cz>

* gcov.c (function_info::function_info): Remove num_counts
and add vector<gcov_type>.
(function_info::~function_info): Use the vector.
(process_file): Likewise.
(read_graph_file): Likewise.
(read_count_file): Likewise.
(solve_flow_graph): Likewise.

From-SVN: r254674

GCOV: simplify usage of function_info::artificial.

2017-11-13  Martin Liska  <mliska@suse.cz>

* gcov.c (function_info::is_artificial): New function.
(process_file): Erase all artificial early.
(generate_results): Skip as all artificial are already
removed.

From-SVN: r254673

GCOV: introduce global vector of functions

2017-11-13  Martin Liska  <mliska@suse.cz>

* gcov.c (read_graph_file): Store to global vector of functions.
(read_count_file): Iterate the vector.
(process_file): Likewise.
(generate_results): Likewise.
(release_structures): Likewise.

From-SVN: r254672

re PR tree-optimization/82954 (ICE in fold_binary_loc, at fold-const.c:9061)

PR tree-optimization/82954
* gimple-ssa-store-merging.c
(imm_store_chain_info::coalesce_immediate_stores): If
!infof->ops[N].base_addr, split group if info->ops[N].base_addr.

* gcc.c-torture/execute/pr82954.c: New test.

From-SVN: r254671

[AArch64] More aarch64_endian_lane_rtx

r254466 failed to update some uses of ENDIAN_LANE_N that were added after
the patch was initially written, which meant that we were treating the
mode number as an element count.

2017-11-13  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
* config/aarch64/aarch64-simd.md (aarch64_store_lane0<mode>):
Upddate call to ENDIAN_LANE_N.
(aarch64_<sur>dot_lane<vsi2qi>): Use aarch64_endian_lane_rtx.
(aarch64_<sur>dot_laneq<vsi2qi>): Likewise.
(*aarch64_simd_vec_copy_lane<mode>): Update calls to ENDIAN_LANE_N
and use aarch64_endian_lane_rtx.
(*aarch64_simd_vec_copy_lane_<vswap_width_name><mode>): Likewise.

From-SVN: r254670

Daily bump.

From-SVN: r254669

[riscv] Wrap ASM_OUTPUT_LABELREF in do {} while (0)

2017-11-12  Tom de Vries  <tom@codesourcery.com>

* config/riscv/riscv.h (ASM_OUTPUT_LABELREF): Wrap in do {} while (0).

From-SVN: r254666

Remove semicolon after ASM_OUTPUT_ASCII

2017-11-12  Tom de Vries  <tom@codesourcery.com>

* config/elfos.h (ASM_OUTPUT_ASCII): Remove semicolon after macro body.

From-SVN: r254665

[cr16, powerpcspe, rs6000] Remove semicolon after ASM_OUTPUT_LABELREF macro body

2017-11-12  Tom de Vries  <tom@codesourcery.com>

* config/cr16/cr16.h (ASM_OUTPUT_LABELREF): Remove semicolon after macro
body.
* config/powerpcspe/xcoff.h (ASM_OUTPUT_LABELREF): Same.
* config/rs6000/xcoff.h (ASM_OUTPUT_LABELREF): Same.
* defaults.h (ASM_OUTPUT_LABELREF): Same.

From-SVN: r254664

Daily bump.

From-SVN: r254663

re PR fortran/82932 ([OOP] ICE in update_compcall_arglist, at fortran/resolve.c:5837)

2017-11-11  Janus Weil  <janus@gcc.gnu.org>

PR fortran/82932
* resolve.c (update_compcall_arglist): Improve error recovery,
remove a gcc_assert.

2017-11-11  Janus Weil  <janus@gcc.gnu.org>

PR fortran/82932
* gfortran.dg/typebound_call_29.f90: New test.

From-SVN: r254660

re PR c/81117 (Improve buffer overflow checking in strncpy)

gcc/ChangeLog:

PR c/81117
* doc/extend.texi (attribute nonstring): Remove spurious argument.

From-SVN: r254659