tree-vrp.c (vrp_visit_phi_node): Allow a last iteration if the currently executable edges have fixed ranges.

2016-08-12  Richard Biener  <rguenther@suse.de>

* tree-vrp.c (vrp_visit_phi_node): Allow a last iteration if
the currently executable edges have fixed ranges.  Always
go through update_value_range.

From-SVN: r239404

[PR63240] generate debug info for defaulted member functions

This implements <http://dwarfstd.org/ShowIssue.php?issue=141215.3>, a
proposal already accepted for inclusion in DWARF-5, but using
DW_AT_GNU_defaulted instead of DW_AT_defaulted as the attribute name,
because the attribute id for DW_AT_defaulted is not yet publicly
available.

for  include/ChangeLog

PR debug/63240
* dwarf2.def (DW_AT_deleted, DW_AT_defaulted): New.
* dwarf2.h (enu dwarf_defaulted_attribute): New.

for  gcc/ChangeLog

PR debug/63240
* langhooks-def.h
(LANG_HOOKS_FUNCTION_DECL_EXPLICIT_P): Const_tree-ify.
(LANG_HOOKS_FUNCTION_DECL_DELETED_P): Likewise.
(LANG_HOOKS_FUNCTION_DECL_DEFAULTED): Set default.
(LANG_HOOKS_DECLS): Add it.
* langhooks.h (struct lang_hooks_for_decls): Add
function_decl_defaulted.  Const_tree-ify
function_decl_explicit_p and function_decl_deleted_p.
* dwarf2out.c (gen_subprogram_die): Add DW_AT_defaulted
attribute.  Add DW_AT_deleted instead of DW_AT_GNU_deleted,
also at strict DWARF v5.

for  gcc/cp/ChangeLog

PR debug/63240
* cp-objcp-common.c (cp_function_decl_defaulted): New.
(cp_function_decl_explicit_p): Const_tree-ify.
(cp_function_decl_deleted_p): Likewise.
* cp-objcp-common.h (cp_function_decl_defaulted): Declare.
(cp_function_decl_explicit_p): Const_tree-ify.
(cp_function_decl_deleted_p): Likewise.
(LANG_HOOKS_FUNCTION_DECL_DEFAULTED): Redefine.

for  gcc/testsuite/ChangeLog

PR debug/63240
* g++.dg/debug/dwarf2/defaulted-member-function-1.C: New.
* g++.dg/debug/dwarf2/defaulted-member-function-2.C: New.
* g++.dg/debug/dwarf2/defaulted-member-function-3.C: New.
* g++.dg/debug/dwarf2/deleted-member-function.C: Expect
DW_AT_deleted.

From-SVN: r239403

[PR55641] drop spurious const_type from reference_type variables

Although C++ reference types, denoted by DW_TAG_reference_type in
DWARFv2+ debug info, are unchangeable, we output names of reference type
with DW_TAG_const_type, because internally we mark such variables as
TREE_READONLY.  That's an internal implementation detail that shouldn't
leak to debug information.  This patch fixes this.

for  gcc/ChangeLog

PR debug/55641
* dwarf2out.c (decl_quals): Don't map TREE_READONLY to
TYPE_QUAL_CONST in reference-typed decls.

for  gcc/testsuite/ChangeLog

PR debug/55641
* g++.dg/debug/dwarf2/ref-1.C: New.

From-SVN: r239402

[PR49366] emit loc exprs for C++ non-virtual pmf template value parms

We used to emit, in debug information, the values bound to pointer to
member function template parameters only when they were NULL or
virtual member functions, because those can be represented with
DW_AT_const_value.

In order to represent the symbolic pointer to member function
constants for non-virtual member functions, we'd need to be able to
emit relocations for part of DW_AT_const_value, which we don't.  The
more viable alternative is to use DW_AT_location to represent such
values, as slated for inclusion in DWARFv5, according to
<URL:http://www.dwarfstd.org/ShowIssue.php?issue=130412.1>.

With this patch, when we can't emit a DW_AT_const_value, we emit each
"member" of the pointer to member function "record" as a
DW_OP_stack_value DW_OP_piece, as long as the referenced member
function is output in the same translation unit, otherwise we'd get
relocations to external symbols, something to avoid in debug sections.

for  gcc/ChangeLog

PR debug/49366
* dwarf2out.c (loc_list_from_tree_1): Expand some CONSTRUCTORs
in DW_OP_pieces, just enough to handle pointers to member
functions.
(gen_remaining_tmpl_value_param_die_attribute): Use a location
expression on DWARFv5 if a constant value doesn't work.

for  gcc/testsuite/ChangeLog

PR debug/49366
* g++.dg/debug/dwarf2/template-params-12.H: New.
* g++.dg/debug/dwarf2/template-params-12f.C: New.
* g++.dg/debug/dwarf2/template-params-12g.C: New.
* g++.dg/debug/dwarf2/template-params-12n.C: New.
* g++.dg/debug/dwarf2/template-params-12s.C: New.
* g++.dg/debug/dwarf2/template-params-12u.C: New.
* g++.dg/debug/dwarf2/template-params-12v.C: New.
* g++.dg/debug/dwarf2/template-params-12w.C: New.

From-SVN: r239401

Daily bump.

From-SVN: r239400

Add selftests to selftest.c

gcc/ChangeLog:
* selftest-run-tests.c (selftest::run_tests): Call selftest_c_tests.
* selftest.c (selftest::test_assertions): New function.
(selftest::selftest_c_tests): New function.
* selftest.h (selftest::selftest_c_tests): New declaration.

From-SVN: r239397

re PR rtl-optimization/72855 (Long compile time due to integrity checking during dataflow analysis per loop)

2016-08-11  Richard Biener  <rguenther@suse.de>
    Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

PR rtl-optimization/72855
* df-core.c (df_verify): Turn off DF_VERIFY_SCHEDULED at end.

Co-Authored-By: Bill Schmidt <wschmidt@linux.vnet.ibm.com>
From-SVN: r239395

re PR target/72863 (Powerpc64le: redundant swaps when using vec_vsx_ld/st)

[gcc]

2016-08-11  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

PR target/72863
* vsx.md (vsx_load_<mode>): For P8LE, emit swaps at expand time.
(vsx_store_<mode>): Likewise.

[gcc/testsuite]

2016-08-11  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

PR target/72863
* gcc.target/powerpc/pr72863.c: New test.

From-SVN: r239394

os: fix build tags for dir_regfile.go

    We want to build dir_regfile.go if not GNU/linux, and not solaris/386,
    and not solaris/sparc.  The latter two conditions were incorrect.  To
    write ! solaris/386 we have to write !solaris !386.  I forgot De
    Morgan's Law.

    Reviewed-on: https://go-review.googlesource.com/26870

From-SVN: r239393

constexpr-lambda6.C: Remove dg-do run.

* g++.dg/cpp1z/constexpr-lambda6.C: Remove dg-do run.

From-SVN: r239392

* testsuite/ChangeLog: Remove merge marker.

From-SVN: r239385

Support TImode CONST_WIDE_INT store in 64-bit STV

Support TImode CONST_WIDE_INT store generated from piecewise store.
Need to verify performance impact before enabling TImode CONST_INT
store for __int128.

gcc/

* config/i386/i386.c (timode_scalar_to_vector_candidate_p): Allow
TImode CONST_WIDE_INT store.
(timode_scalar_chain::convert_insn): Handle CONST_WIDE_INT store.

gcc/testsuite/

* gcc.target/i386/pieces-strcpy-1.c: New test.
* gcc.target/i386/pieces-strcpy-2.c: Likewise.

From-SVN: r239383

re PR c++/72868 (Constexpr expressions mistreat case ranges)

PR c++/72868
* constexpr.c (label_matches): Handle case range expressions.

* g++.dg/cpp1y/constexpr-switch4.C: New test.

From-SVN: r239379

Use TImode for piecewise move in 64-bit mode

Use TImode for piecewise move in 64-bit mode.  We should use TImode in
32-bit mode and use OImode or XImode if they are available.  But since
by_pieces_ninsns determines the widest mode with MAX_FIXED_MODE_SIZE,
we can only use TImode in 64-bit mode.

gcc/

* config/i386/i386.h (MOVE_MAX_PIECES): Use TImode in 64-bit
mode if unaligned SSE load and store are optimal.

gcc/testsuite/

* gcc.target/i386/pieces-memcpy-1.c: New test.
* gcc.target/i386/pieces-memcpy-2.c: Likewise.
* gcc.target/i386/pieces-memcpy-3.c: Likewise.
* gcc.target/i386/pieces-memcpy-4.c: Likewise.
* gcc.target/i386/pieces-memcpy-5.c: Likewise.
* gcc.target/i386/pieces-memcpy-6.c: Likewise.

From-SVN: r239378

PR c++/73456 - ICE with constrained parameter pack.

* logic.cc (non_atomic_constraint_p): Handle EXPR_PACK_EXPANSION.

From-SVN: r239376

Include <stdlib.h> in intrinsics/random.c

* intrinsics/random.c: Include <stdlib.h>.

From-SVN: r239375

Fix up ChangeLog entry.

From-SVN: r239373

Implement C++17 make_from_tuple.

* include/std/tuple (__make_from_tuple_impl, make_from_tuple): New.
* testsuite/20_util/tuple/make_from_tuple/1.cc: Likewise.

From-SVN: r239372

re PR c/72816 (ICE on x86_64-linux-gnu (tree check: expected tree that contains ‘decl with RTL’ structure, have ‘field_decl’ in set_decl_rtl, at emit-rtl.c:1282))

PR c/72816
* c-decl.c (grokdeclarator): When adding TYPE_DOMAIN for flexible
array member through typedef, for orig_qual_indirect == 0 clear
orig_qual_type.

* gcc.dg/pr72816.c: New test.

From-SVN: r239371

Implement LWG 2758.

* include/bits/basic_string.h
(append(__sv_type, size_type, size_type)): Turn into a template,
change parameter type, constrain, add a conversion to __sv_type
from the dependent parameter type.
(assign(__sv_type, size_type, size_type)): Likewise.
(insert(size_type, __sv_type, size_type, size_type)): Likewise.
(replace(size_type, size_type, __sv_type, size_type, size_type)):
Likewise.
(compare(size_type, size_type,__sv_type, size_type, size_type)):
Likewise.
* testsuite/21_strings/basic_string/lwg2758.cc: New.

From-SVN: r239370

mark_spam.py: mark attachments as obsolete and rename them

* mark_spam.py: Mark attachments as obsolete and rename them.

From-SVN: r239368

re PR tree-optimization/71083 (Unaligned bit-field address when predictive commoning)

2016-08-11  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        PR tree-optimization/71083
        * tree-predcom.c (ref_at_iteration): Correctly align the
        reference type.

testsuite:
2016-08-11  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        PR tree-optimization/71083
        * gcc.c-torture/execute/pr71083.c: New test.
        * gnat.dg/loop_optimization23.adb: New test.
        * gnat.dg/loop_optimization23_pkg.ads: New test.
        * gnat.dg/loop_optimization23_pkg.adb: New test.

From-SVN: r239362

S/390: Provide low-level builtins with __int128 ops.

gcc/ChangeLog:

2016-08-11  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/s390-builtin-types.def: Add INT128 types.
* config/s390/s390-builtins.def: Add INT128 variants for the add
sub low-level builtins dealing with TImode.
* config/s390/s390.c (s390_expand_builtin): Allow mode conversions
via subreg when expanding a builtin.
* config/s390/s390.md: Remove UNSPEC_VEC_ADDC_U128,
UNSPEC_VEC_SUB_U128, and UNSPEC_VEC_SUBC_U128 constants.
Fix comment.
* config/s390/vecintrin.h: Adjust builtin names accordingly.
* config/s390/vx-builtins.md ("vec_add_u128"): Remove expander.
("vec_addc<mode>", "vec_addc_u128"): Merge to
"vacc<bhfgq>_<mode>".
("vec_adde_u128"): Rename to "vacq". Change mode to TImode.
("vec_addec_u128"): Rename to "vacccq". Change mode to TImode.
("vec_subc<mode>", "vec_subc_u128"): Merge to
"vscbi<bhfgq>_<mode>".
("vec_sube_u128"): Rename to "vsbiq". Change mode to TImode.
("vec_subec_u128"): Rename to "vsbcbiq". Change mode to TImode.

From-SVN: r239361

S/390: Fix vlvg vlgv low-level builtins.

gcc/ChangeLog:

2016-08-11  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/s390-builtins.def: Mark last operand of s390_vlvg*
and s390_vlgv* builtins as element selector.

From-SVN: r239360

aarch64-simd.md (vcond<mode><mode>): Delete unused declaration.

* config/aarch64/aarch64-simd.md (vcond<mode><mode>): Delete
unused declaration.
(vcond<v_cmp_mixed><mode>): Ditto.
(vcondu<mode><mode>, vcondu<mode><v_cmp_mixed>): Ditto.

From-SVN: r239359

aarch64-simd.md (vec_cmp<mode><v_cmp_result>): Init variable explicitly, also assert on it before use.

* config/aarch64/aarch64-simd.md (vec_cmp<mode><v_cmp_result>: Init
variable explicitly, also assert on it before use.

From-SVN: r239358

re PR tree-optimization/72772 (Missed SCEV after pass reordering@236440)

2016-08-11  Richard Biener  <rguenther@suse.de>

PR tree-optimization/72772
* cfgloopmanip.c (create_preheader): Use split_edge if there
is a single loop entry, avoiding degenerate PHIs.

* gcc.dg/graphite/pr35356-1.c: Adjust.
* gcc.dg/tree-ssa/pr59597.c: Likewise.

From-SVN: r239357

Replace KISS PRNG with xorshift1024* using per-thread state.

frontend:

2016-08-11  Janne Blomqvist  <jb@gcc.gnu.org>

* check.c (gfc_check_random_seed): Use new seed size in check.
* intrinsic.texi (RANDOM_NUMBER): Updated documentation.
(RANDOM_SEED): Likewise.

testsuite:

2016-08-11  Janne Blomqvist  <jb@gcc.gnu.org>

* gfortran.dg/random_7.f90: Take into account that the last seed
value is the special p value.
* gfortran.dg/random_seed_1.f90: Seed size is now constant.

libgfortran:
2016-08-11  Janne Blomqvist  <jb@gcc.gnu.org>

* intrinsics/random.c: Replace KISS with xorshift1024* using
per-thread state.
* runtime/main.c (init): Don't call random_seed_i4.

From-SVN: r239356

tree-ssa-threadbackward.c (pass_data_thread_jumps): Remove unconditional TODO_cleanup_cfg.

2016-08-11  Richard Biener  <rguenther@suse.de>

* tree-ssa-threadbackward.c (pass_data_thread_jumps): Remove
unconditional TODO_cleanup_cfg.
(pass_thread_jumps::execute): Initialize loops, perform a CFG
cleanup only if we threaded a jump.

* gcc.dg/tree-ssa/ssa-dom-thread-7.c: Adjust.

From-SVN: r239355

Daily bump.

From-SVN: r239351

missed from last commit

From-SVN: r239343

[LRA] Reload of slow mems

pr71680.c -m64 -O1 -mlra, ira output showing two problem insns.
(insn 7 5 26 3 (set (reg:SI 159 [ a ])
        (mem/c:SI (reg/f:DI 158) [1 a+0 S4 A8])) pr71680.c:13 464 {*movsi_internal1}
     (expr_list:REG_EQUIV (mem/c:SI (reg/f:DI 158) [1 a+0 S4 A8])
        (nil)))
(insn 26 7 27 3 (set (reg:DI 162)
        (unspec:DI [
                (fix:SI (subreg:SF (reg:SI 159 [ a ]) 0))
            ] UNSPEC_FCTIWZ)) pr71680.c:13 372 {fctiwz_sf}
     (expr_list:REG_DEAD (reg:SI 159 [ a ])
        (nil)))
Insn 26 requires that reg 159 be of class FLOAT_REGS.

first lra action:
deleting insn with uid = 7.
Changing pseudo 159 in operand 1 of insn 26 on equiv [r158:DI]
      Creating newreg=164, assigning class ALL_REGS to subreg reg r164
   26: r162:DI=unspec[fix(r164:SI#0)] 7
      REG_DEAD r159:SI
    Inserting subreg reload before:
   30: r164:SI=[r158:DI]
[snip]
      Change to class FLOAT_REGS for r164

Well, that didn't do much.  lra tried the equiv mem, found that didn't
work, and had to reload.  Effectively getting back to the two original
insns but r159 replaced with r164.  simplify_operand_subreg did not do
anything in this case because SLOW_UNALIGNED_ACCESS was true (wrongly
for power8, but that's beside the point).  So now we have, using
abbreviated rtl notation:
r164:SI=[r158:DI]
r162:DI=unspec[fix(r164:SI)]
The problem here is that the first insn isn't valid, due to the rs6000
backend not supporting SImode in fprs, and r164 must be an fpr to make
the second insn valid.

next lra action:
      Creating newreg=165 from oldreg=164, assigning class GENERAL_REGS to r165
   30: r165:SI=[r158:DI]
    Inserting insn reload after:
   31: r164:SI=r165:SI
so now we have
r165:SI=[r158:DI]
r164:SI=r165:SI
r162:DI=unspec[fix(r164:SI)]

This ought to be good on power8, except for one little thing.
r165 is GENERAL_REGS so the first insn is good, a gpr load from mem.
r164 is FLOAT_REGS, making the last insn good, a fctiwz.
The second insn ought to be a sldi, mtvsrd, xscvspdpn combination, but
that is only supported for SFmode.  So lra continue on reloading the
second insn, but in vain because it never tries anything other than
SImode and as noted above, SImode is not valid in fprs.

What this patch does is arrange to emit the two reloads needed for the
SLOW_UNALIGNED_ACCESS case at once, moving the subreg to the second
insn in order to switch modes, producing:

r164:SI=[r158:DI]
r165:SF=r164:SI#0
r162:DI=unspec[fix(r165:SF)]

I've also tidied a couple of other things:
1) "old" is unnecessary as it duplicated "operand".
2) Rejecting mem subregs due to SLOW_UNALIGNED_ACCESS only makes sense
if the original mode was not slow.

PR target/71680
* lra-constraints.c (simplify_operand_subreg): Allow subreg
mode for mem when SLOW_UNALIGNED_ACCESS if inner mode is also
slow.  Emit two reloads for slow mem case, first loading in
fast innermode, then converting to required mode.
testsuite/
* gcc.target/powerpc/pr71680.c: New.

From-SVN: r239342

* c-cppbuiltin.c (c_cpp_builtins): Define __cpp_if_constexpr.

From-SVN: r239340

Implement C++17 constexpr if.

* cp-tree.h (IF_STMT_CONSTEXPR_P): New.
* name-lookup.c (push_to_top_level, pop_from_top_level_1): Handle it.
* parser.h (struct cp_parser): Add in_discarded_stmt field.
* parser.c (cp_parser_selection_statement): Handle 'if constexpr'.
(cp_parser_jump_statement): Avoid deducing from a discarded return.
* pt.c (tsubst_expr): Only instantiate taken branch of constexpr if.
* semantics.c (begin_if_stmt): Set the binding level this_entity.
(finish_if_stmt_cond): Require the condition of a
constexpr if to be constant.
* decl.c (level_for_constexpr_if): New.
(named_label_entry): Add in_constexpr_if field.
(poplevel_named_label_1): Set it.
(check_goto): Check it.
(check_previous_goto_1): Check level_for_constexpr_if.

From-SVN: r239338

altivec.h (vec_extract_exp): New macro.

gcc/ChangeLog:

2016-08-10  Kelvin Nilsen  <kelvin@gcc.gnu.org>

* config/rs6000/altivec.h (vec_extract_exp): New macro.
(vec_extract_sig): New macro.
(vec_insert_exp): New macro.
(vec_test_data_class): New macro.
(scalar_extract_exp): New macro.
(scalar_extract_sig): New macro.
(scalar_insert_exp): New macro.
(scalar_test_data_class): New macro.
(scalar_test_neg): New macro.
(scalar_cmp_exp_gt): New macro.
(scalar_cmp_exp_lt): New macro.
(scalar_cmp_exp_eq): New macro.
(scalar_cmp_exp_unordered): New macro.
* config/rs6000/predicates.md (u7bit_cint_operand): New predicate
to enforce constraint that operand is a 7-bit unsigned literal.
* config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_1): New macro
for power9 built-ins.
(BU_P9V_VSX_2): Likewise.
(BU_P9V_64BIT_VSX_2): Likewise.
(VSEEDP): Add scalar extract exponent support.
(VSEESP): Add scalar extract signature support.
(VSTDCNDP): Add scalar test negative support.
(VSTDCNSP): Likewise.
(VSIEDP): Add scalar insert exponent support.
(VSCEDPGT): Add scalar compare exponent greater than support.
(VSCEDPLT): Add scalar compare exponent less than support.
(VSCEDPEQ): Add scalar compare exponent test-for-equality support.
(VSCEDPUO): Add scalar compare exponent test-for-unordered support.
(VSTDCDP): Add scalar test data class support.
(VSTDCSP): Likewise.
(VSEEDP): Add overload support for scalar extract exponent
operation.
(VSESDP): Add overload support for scalar extract signature
operation.
(VSTDCN): Add overload support for scalar test negative
operation.
(VSTDCNDP): Add overload support for scalar test negative
operation.
(VSTDCNSP): Add overload support for scalar test negative
operation.
(VSIEDP): Add overload support for scalar insert exponent
operation.
(VSTDC): Add overload support for scalar test data class
operation.
(VSTDCDP): Add overload support for scalar test data class
operation.
(VSTDCSP): Add overload support for scalar test data class
opreation.
(VSCEDPGT): Add overload support for scalar compare exponent
greater than operation.
(VSCEDPLT): Add overload support for scalar compare exponent
less than operation.
(VSCEDPEQ): Add overload support for scalar compare exponent
test-for-equality operation.
(VSCEDPUO): Add overload support for scalar compare exponent
test-for-unordered operation.
(VEEDP): Add vector extract exponent support.
(VEESP): Likewise.
(VESDP): Add vector extract significand support.
(VESSP): Likewise.
(VIEDP): Add vector insert exponent support.
(VIESP): Likewise.
(VTDCDP): Add vector test data class support.
(VTDCSP): Likewise.
(VES): Add overload support for vector extract significand operation.
(VESDP): Likewise.
(VESSP): Likewise
(VEE): Add overload support for vector extract exponent operation.
(VEEDP): Likewise.
(VEESP): Likewise.
(VTDC): Add overload support for vector test data class operation.
(VTDCDP): Likewise.
(VTDCSP): Likewise.
(VIE): Add overload support for vector insert exponent operation.
(VIEDP): Likewise.
(VIESP): Likewise.
* config/rs6000/rs6000-c.c (altivec_overloaded_builtins): Add
overloaded binary floating point functions.
(altivec_resolve_overloaded_builtin): Improve error messages to
distinguish between functions not supported in the current
compiler configuration and functions that were invoked with an
invalid parameter combination, and include the built-in function
name in both error messages.
* config/rs6000/rs6000-protos.h (rs6000_overloaded_builtin_name):
New prototype.
* config/rs6000/rs6000.c (rs6000_overloaded_builtin_name): New
function.
(rs6000_expand_binop_builtin): Add check to enforce that argument
2 of the test data class operations is a 7-bit unsigned literal.
(rs6000_invalid_builtin): Add code to issue an error message if a
built-in function that requires the power9_vector and -m32
command-line options is compiled without these options.
* config/rs6000/vsx.md (UNSPEC_VSX_SXEXPDP): New value.
(UNSPEC_VSX_SXSIGDP): New value.
(UNSPEC_VSX_SXSIGPDP): New value.
(UNSPEC_VSX_SIEXPDP): New value.
(UNSPEC_VSX_SCMPEXPDP): New value.
(UNSPEC_VSX_STSTDC): New value.
(UNSPEC_VSX_VXEXP): New value.
(UNSPEC_VSX_VXSIG): New value.
(UNSPEC_VSX_VIEXP): New value.
(UNSPEC_VSX_VTSTDC): New value.
(xsxexpdp): New insn for scalar extract exponent.
(xsxsigdp): New insn for scalar extract significand.
(xsiexpdp): New insn for scalar insert exponent.
(xscmpexpdp_<code>): New expansion for scalar compare exponents.
(*xscmpexpdp): New insn for scalar compare exponents.
(xststdc<Fvsx): New expansion for both single- and
double-precision scalar test data class operations.
(xststdcneg<Fvsx>): New expansion for both single- and
double-precision scalar test for negative value operations.
(*xststdc<Fvsx>): New insn for scalar test data class
operation.
(xvxexp<VSs>): New insn for single- and double-precision
vector extract exponent operation.
(xvxsig<VSs>): New insn for single- and double-precision
vector extract significand operation.
(xviexp<VSs>): New insn for single- and double-precision
vector insert exponent operation.
(xvtstdc<VSs>): New insn for single- and double-precision
vector test data class operation.
* doc/extend.texi (PowerPC AltiVec Built-in Functions): Document
built-in functions to represent the Power9 binary floating-point
support instructions.

gcc/testsuite/ChangeLog:

2016-08-10  Kelvin Nilsen  <kelvin@gcc.gnu.org>

* gcc.target/powerpc/bfp/bfp.exp: New file.
* gcc.target/powerpc/bfp/scalar-cmp-exp-eq-0.c: New test.
* gcc.target/powerpc/bfp/scalar-cmp-exp-eq-1.c: New test.
* gcc.target/powerpc/bfp/scalar-cmp-exp-eq-2.c: New test.
* gcc.target/powerpc/bfp/scalar-cmp-exp-gt-0.c: New test.
* gcc.target/powerpc/bfp/scalar-cmp-exp-gt-1.c: New test.
* gcc.target/powerpc/bfp/scalar-cmp-exp-gt-2.c: New test.
* gcc.target/powerpc/bfp/scalar-cmp-exp-lt-0.c: New test.
* gcc.target/powerpc/bfp/scalar-cmp-exp-lt-1.c: New test.
* gcc.target/powerpc/bfp/scalar-cmp-exp-lt-2.c: New test.
* gcc.target/powerpc/bfp/scalar-cmp-exp-unordered-0.c: New test.
* gcc.target/powerpc/bfp/scalar-cmp-exp-unordered-1.c: New test.
* gcc.target/powerpc/bfp/scalar-cmp-exp-unordered-2.c: New test.
* gcc.target/powerpc/bfp/scalar-extract-exp-0.c: New test.
* gcc.target/powerpc/bfp/scalar-extract-exp-1.c: New test.
* gcc.target/powerpc/bfp/scalar-extract-exp-2.c: New test.
* gcc.target/powerpc/bfp/scalar-extract-sig-0.c: New test.
* gcc.target/powerpc/bfp/scalar-extract-sig-1.c: New test.
* gcc.target/powerpc/bfp/scalar-extract-sig-2.c: New test.
* gcc.target/powerpc/bfp/scalar-insert-exp-0.c: New test.
* gcc.target/powerpc/bfp/scalar-insert-exp-1.c: New test.
* gcc.target/powerpc/bfp/scalar-insert-exp-2.c: New test.
* gcc.target/powerpc/bfp/scalar-test-data-class-0.c: New test.
* gcc.target/powerpc/bfp/scalar-test-data-class-1.c: New test.
* gcc.target/powerpc/bfp/scalar-test-data-class-2.c: New test.
* gcc.target/powerpc/bfp/scalar-test-data-class-3.c: New test.
* gcc.target/powerpc/bfp/scalar-test-data-class-4.c: New test.
* gcc.target/powerpc/bfp/scalar-test-data-class-5.c: New test.
* gcc.target/powerpc/bfp/scalar-test-data-class-6.c: New test.
* gcc.target/powerpc/bfp/scalar-test-data-class-7.c: New test.
* gcc.target/powerpc/bfp/scalar-test-neg-0.c: New test.
* gcc.target/powerpc/bfp/scalar-test-neg-1.c: New test.
* gcc.target/powerpc/bfp/scalar-test-neg-2.c: New test.
* gcc.target/powerpc/bfp/scalar-test-neg-3.c: New test.
* gcc.target/powerpc/bfp/vec-extract-exp-0.c: New test.
* gcc.target/powerpc/bfp/vec-extract-exp-1.c: New test.
* gcc.target/powerpc/bfp/vec-extract-exp-2.c: New test.
* gcc.target/powerpc/bfp/vec-extract-exp-3.c: New test.
* gcc.target/powerpc/bfp/vec-extract-sig-0.c: New test.
* gcc.target/powerpc/bfp/vec-extract-sig-1.c: New test.
* gcc.target/powerpc/bfp/vec-extract-sig-2.c: New test.
* gcc.target/powerpc/bfp/vec-extract-sig-3.c: New test.
* gcc.target/powerpc/bfp/vec-insert-exp-0.c: New test.
* gcc.target/powerpc/bfp/vec-insert-exp-1.c: New test.
* gcc.target/powerpc/bfp/vec-insert-exp-2.c: New test.
* gcc.target/powerpc/bfp/vec-insert-exp-3.c: New test.
* gcc.target/powerpc/bfp/vec-test-data-class-0.c: New test.
* gcc.target/powerpc/bfp/vec-test-data-class-1.c: New test.
* gcc.target/powerpc/bfp/vec-test-data-class-2.c: New test.
* gcc.target/powerpc/bfp/vec-test-data-class-3.c: New test.
* gcc.target/powerpc/bfp/vec-test-data-class-4.c: New test.
* gcc.target/powerpc/bfp/vec-test-data-class-5.c: New test.
* gcc.target/powerpc/bfp/vec-test-data-class-6.c: New test.
* gcc.target/powerpc/bfp/vec-test-data-class-7.c: New test.

From-SVN: r239334

compiler: annotate or fix fall through cases in switch statements

    A new GCC warning, -Wimplicit-fallthrough, detected these three cases
    where a switch case falls through to another.  One of them looks like
    a bug, so I fixed it; the two other look like intended fall throughs
    so I just added a "fall through" comment to suppress the GCC warning.

    Bootstrapped and ran Go testsuite on x86_64-pc-linux-gnu.

    Reviewed-on: https://go-review.googlesource.com/26695

From-SVN: r239332

* doc/sourcebuild.texi (@item vect_cond_mixed): New item.

From-SVN: r239330

target-supports.exp (check_effective_target_vect_cond_mixed): Add aarch64*-*-*.

gcc/testsuite
* lib/target-supports.exp (check_effective_target_vect_cond_mixed):
Add aarch64*-*-*.

From-SVN: r239329

iterators.md (V_cmp_mixed, [...]): New.

* config/aarch64/iterators.md (V_cmp_mixed, v_cmp_mixed): New.
* config/aarch64/aarch64-simd.md (<su><maxmin>v2di3): Call
gen_vcondv2div2di instead of gen_aarch64_vcond_internalv2div2di.
(aarch64_vcond_internal<mode><mode>): Delete pattern.
(aarch64_vcond_internal<VDQF_COND:mode><VDQF:mode>): Ditto.
(vcond<v_cmp_result><mode>): Ditto.
(vcond<mode><mode>): Re-implement using vec_cmp and vcond_mask.
(vcondu<mode><mode>): Ditto.
(vcond<v_cmp_mixed><mode>): New pattern.
(vcondu<mode><v_cmp_mixed>): New pattern.
(aarch64_cmtst<mode>): Revise comment using aarch64_vcond instead
of aarch64_vcond_internal.

gcc/testsuite
* gcc.target/aarch64/simd/vcond-ne.c: New test.

From-SVN: r239328

aarch64-simd.md (vec_cmp<mode><mode>): New pattern.

* config/aarch64/aarch64-simd.md (vec_cmp<mode><mode>): New pattern.
(vec_cmp<mode><v_cmp_result>): New pattern.
(vec_cmpu<mode><mode>): New pattern.
(vcond_mask_<mode><v_cmp_result>): New pattern.

Co-Authored-By: Bin Cheng <bin.cheng@arm.com>
Co-Authored-By: Renlin Li <renlin.li@arm.com>
From-SVN: r239327

re PR middle-end/71734 (FAIL: libgomp.fortran/simd4.f90   -O3 -g  execution test)

Fix PR tree-optimization/71734

2016-08-10  Yuri Rumyantsev  <ysrumyan@gmail.com>

PR tree-optimization/71734
* tree-ssa-loop-im.c (ref_indep_loop_p): Add new argument
REF_LOOP, invoke ref_indep_loop_p_1.
(outermost_indep_loop): Pass LOOP argumnet where REF was defined
to ref_indep_loop_p.
(ref_indep_loop_p_1): Fix commentary, add argument REF_LOOP,
combine it with ref_indep_lopp_p_2, update SAFELEN if only REF
is inside LOOP, do not cache dpendence value for loops with
non-zero SAFELEN.
(ref_indep_loop_p_2): Delete function.
(can_sm_ref_p): Pass LOOP as additional argument to
ref_indep_loop_p.

From-SVN: r239326

re PR target/72853 (gcc/testsuite/gcc.c-torture/execute/20021120-1.c generates incorrect stxssp op with -mcpu=power9)

[gcc]
2016-08-10  Michael Meissner  <meissner@linux.vnet.ibm.com>

PR target/72853
* config/rs6000/rs6000.c (mem_operand_ds_form): Add check for op
being an offsettable address.

[gcc/testsuite]
2016-08-10  Michael Meissner  <meissner@linux.vnet.ibm.com>

PR target/72853
* gcc.target/powerpc/pr72853.c: New test.

From-SVN: r239325

Add new *_atomic counter update function

PR gcov-profile/58306
* Makefile.in: New functions (modules) are added.
* libgcov-profiler.c (__gcov_interval_profiler_atomic): New
function.
(__gcov_pow2_profiler_atomic): New function.
(__gcov_one_value_profiler_body): New argument is instroduced.
(__gcov_one_value_profiler): Call with the new argument.
(__gcov_one_value_profiler_atomic): Likewise.
(__gcov_indirect_call_profiler_v2): Likewise.
(__gcov_time_profiler_atomic): New function.
(__gcov_average_profiler_atomic): Likewise.
(__gcov_ior_profiler_atomic): Likewise.
* libgcov.h: Declare the aforementioned functions.
PR gcov-profile/58306
* gcc.dg/tree-prof/val-profiler-threads-1.c: New test.
PR gcov-profile/58306
* tree-profile.c (gimple_init_edge_profiler): Create conditionally
atomic variants of profile update functions.

From-SVN: r239324

Cherry-pick fprofile-generate-atomic from google/gcc-4_9

Cherry picked (and modified) from google-4_7 branch
2012-12-26  Rong Xu  <xur@google.com>
* common.opt (fprofile-update): Add new flag.
* coretypes.h: Define enum profile_update.
* doc/invoke.texi: Document -fprofile-update.
* gcov-io.h: Declare GCOV_TYPE_ATOMIC_FETCH_ADD and
GCOV_TYPE_ATOMIC_FETCH_ADD_FN.
* tree-profile.c (gimple_init_edge_profiler): Generate
also atomic profiler update.
(gimple_gen_edge_profiler): Likewise.
* g++.dg/gcov/gcov-threads-1.C: New test.

From-SVN: r239323

toplev.c: set FILE * globals to NULL after fclose

gcc/ChangeLog:
* toplev.c (finalize): Set aux_info_file, asm_out_file, and
stack_usage_file to NULL after fclose calls.

From-SVN: r239322

Fix PR 71873 - ICE in push_reload

Extend computation of subreg_in_class to constants and plus expressions
inside SUBREGs, before recursively calling push_reload. SYMBOL_REFs are
also CONSTANT_P, so remove explicit handling of SYMBOL_REFs.

gcc/ChangeLog

PR target/71873
* reload.c (push_reload): Compute subreg_in_class for
subregs of constants and plus expressions. Remove special
handling of SYMBOL_REFs.

gcc/testsuite/ChangeLog

PR target/71873
* gcc.target/avr/pr71873.c: New test.

From-SVN: r239321

[RS6000] e500 part of pr71680

The fallback part of HARD_REGNO_CALLER_SAVE_MODE, choose_hard_reg_mode,
returns DFmode for SImode when TARGET_E500_DOUBLE.  This confuses
lra when attempting to save ctr around a call.

PR target/71680
* config/rs6000/rs6000.h (HARD_REGNO_CALLER_SAVE_MODE): Return
SImode for TARGET_E500_DOUBLE when given SImode.

From-SVN: r239317

gccgo.texi (Invoking gccgo): Document -fgo-optimize-allocs and -fgo-debug-escae.

* gccgo.texi (Invoking gccgo): Document -fgo-optimize-allocs and
-fgo-debug-escae.
(Compiler Directives): New chapter.
(Function Names): Describe using //go:linkname.  Suggest using
-fgo-pkgpath rather than -fgo-prefix.

From-SVN: r239316

compiler: implement go:noinline and go:nosplit directives

    Reviewed-on: https://go-review.googlesource.com/26652

From-SVN: r239315

Daily bump.

From-SVN: r239314

compiler: implement go:linkname compiler directive

    Implement the go:linkname compiler directive for functions (both
    function definitions and function declarations).  At least for now, give
    an error for code that tries to use go:linkname with a non-function.

    Reviewed-on: https://go-review.googlesource.com/26651

From-SVN: r239311

avx512fintrin.h (_mm512_cvtsepi64_epi32): Remove unused variable __O.

* config/i3836/avx512fintrin.h (_mm512_cvtsepi64_epi32): Remove
unused variable __O.

From-SVN: r239310

PR c++/68703 - bogus error with dependent vector length

gcc/c-family/
* c-common.c (c_common_attribute_table): vector_size affects type
identity.
gcc/cp/
* decl2.c (any_dependent_type_attributes_p): New.
* pt.c (dependent_type_p_r, type_dependent_expression_p): Check it.
* semantics.c (finish_id_expression): Check it.
* typeck.c (finish_class_member_access_expr): Check it.

From-SVN: r239309

Add tests for __gcov_dump and __gcov_reset

* g++.dg/gcov/gcov-dump-1.C: New test.
* g++.dg/gcov/gcov-dump-2.C: New test.

From-SVN: r239308

Fix typo in gcov.texi

* doc/gcov.texi: Change _gcov_dump to __gcov_dump and
_gcov_reset to __gcov_reset.
* doc/gcov-tool.texi: Fix typo.
* libgcov-util.c: Fix typo and GNU coding style.

From-SVN: r239307

Remove __gcov_indirect_call_profiler

* Makefile.in: Remove __gcov_indirect_call_profiler.
* libgcov-profiler.c (__gcov_indirect_call_profiler): Remove
function.
* libgcov.h: And the declaration of the function.

From-SVN: r239306

Fix usage of POW2 histogram

* value-prof.c (gimple_divmod_values_to_profile): Do not
instrument MOD histogram if a value is not a SSA name.
* gcc.dg/tree-prof/val-prof-9.c: New test.

From-SVN: r239305

Fix POW2 histogram

* gcc.dg/tree-prof/val-prof-8.c: New test.
* value-prof.c (dump_histogram_value): Swap pow2 and non-pow2
values.
* libgcov-profiler.c (__gcov_pow2_profiler): Consider 0 as not
power of two.

From-SVN: r239304

Set +x for two python scripts.

From-SVN: r239303

re PR bootstrap/72833 (error in fortran/parse.c (unexpected_eof) on Mac OS X 10.7)

2016-08-09  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        PR bootstrap/72833
        * fixincl.tpl (version-compare): Fix generation with autogen 5.18.
        * inclhack.def (darwin_longjmp_noreturn): New fix.
        * fixincl.x: Regenerated.
        * tests/base/i386/setjmp.h [DARWIN_LONGJMP_NORETURN_CHECK]: new test.

From-SVN: r239301

[PATCH][PR64971]Convert function pointer to Pmode when emit call.

gcc/

2016-08-04  Renlin Li  <renlin.li@arm.com>

PR middle-end/64971
* calls.c (prepare_call_address): Convert funexp to Pmode when
necessary.
* config/aarch64/aarch64.md (sibcall): Remove fix for PR 64971.
(sibcall_value): Likewise.

From-SVN: r239300

PR c++/71712 - ABI tags on conversion ops.

* class.c (check_abi_tags): Don't duplicate tags for conversion ops.

From-SVN: r239299

Adjust mangling of ABI tags on class template member functions.

* class.c (missing_abi_tags): New.
(check_abi_tags): Don't check template. Add just_checking mode.
* mangle.c (abi_flag_at_least, any_abi_below, equal_abi_tags): New.
(sorted_abi_tags): Split out from write_abi_tags.
(struct releasing_vec): New.
(write_unqualified_name): Only look for the primary
template for types.  Implement backward compatibility.

From-SVN: r239298

re PR c/7652 (-Wswitch-break : Warn if a switch case falls through)

PR c/7652
gcc/c-family/
* c-ada-spec.c (dump_generic_ada_node): Add return.
gcc/
* cselib.c (cselib_expand_value_rtx_1): Add return.
* gengtype.c (dbgprint_count_type_at): Likewise.
* hsa-gen.c (gen_hsa_insn_for_internal_fn_call): Likewise.
* reg-stack.c (get_true_reg): Restructure to avoid fallthrough warning.

From-SVN: r239297

mksysinfo.sh: always define CLONE_NEWNET

    CLONE_NEWNET is needed to compile the syscall tests on GNU/Linux.
    The symbol is not defined in the CentOS 5.11 header files.

    Patch from Uros Bizjak.

    Reviewed-on: https://go-review.googlesource.com/26630

From-SVN: r239296

[PR ipa/71981] Make get_dynamic_type grok MEM_REF

2016-08-09  Martin Jambor  <mjambor@suse.cz>

        PR ipa/71981
        * ipa-polymorphic-call.c (get_dynamic_type): Bail out gracefully
        if instance is a MEM_REF.

testsuite/
        PR ipa/71981
        * gcc.dg/ipa/pr71981.c: New test.

From-SVN: r239294

re PR rtl-optimization/72843 (internal compiler error: in lra_set_insn_recog_data, at lra.c:964)

PR target/72843
* config/i386/i386.md (*movtf_internal): Use
lra_in_progress || reload_completed instead of !can_create_pseudo_p
in the insn constraint.
(*movxf_internal): Ditto.
(*movdf_internal): Ditto.
(*movsf_internal): Ditto.

From-SVN: r239293

re PR tree-optimization/33707 (scev not handling unsigned conversion)

gcc/testsuite
PR tree-optimization/33707
* gcc.dg/vect/pr33707.c: New test.

From-SVN: r239292

re PR tree-optimization/72772 (Missed SCEV after pass reordering@236440)

PR tree-optimization/72772
* tree-ssa-loop-niter.c (loop_exits_before_overflow): Check equality
for expanded base.

gcc/testsuite
PR tree-optimization/pr72772
* gcc.dg/tree-ssa/pr72772.c: New test.

From-SVN: r239291

re PR tree-optimization/72772 (Missed SCEV after pass reordering@236440)

PR tree-optimization/72772
* tree-ssa-loop-niter.h (simplify_using_initial_conditions): Delete
parameter STOP.
* tree-ssa-loop-niter.c (tree_simplify_using_condition_1): Delete
parameter STOP and update calls.  Move expand_simple_operations
function call from here...
(simplify_using_initial_conditions): ...to here.  Delete parameter
STOP.
(tree_simplify_using_condition): Delete parameter STOP.
* tree-scalar-evolution.c (simple_iv_with_niters): Update call to
simplify_using_initial_conditions.

From-SVN: r239290

PR c++/72849 - ICE with incomplete class.

* constexpr.c (cxx_eval_constant_expression): Check
COMPLETE_TYPE_P before calling is_really_empty_class.
* class.c (is_really_empty_class): Don't call complete_type.

From-SVN: r239289

MIPS: Skip gcc.dg/loop-8.c due to additional invariants

gcc/
PR rtl-optimization/66669
* gcc.dg/loop-8.c: Skip for MIPS due to extra invariants.

From-SVN: r239288

PR c++/56701 - wrong type of &*this

* typeck.c (cp_build_addr_expr_1): Remove special *this handling.

From-SVN: r239285

compiler: rewrite compiler directive support

    Rewrite the compiler directive support to recognize all the compiler
    directives implemented by the current gc compiler.  The directives other
    than go:linkname are now turned into GOPRAGMA flags attached to a
    function or function declaration.  The go:linkname directive is turned
    into a map attached to the Lex object.  No new directives are actually
    implemented yet, they are just recognized.

    Reviewed-on: https://go-review.googlesource.com/26610

From-SVN: r239282

MIPS: Use create_tmp_var_raw in mips_atomic_assign_expand_fenv

gcc/
PR c/65345
* config/mips/mips.c (mips_atomic_assign_expand_fenv):
Use create_tmp_var_raw instead of create_tmp_var.

From-SVN: r239278

tree-ssa-threadbackward.c (profitable_jump_thread_path): Treat same SSA names related.

2016-08-09  Richard Biener  <rguenther@suse.de>

* tree-ssa-threadbackward.c (profitable_jump_thread_path):
        Treat same SSA names related.

From-SVN: r239276

re PR tree-optimization/72824 (Signed floating point zero semantics broken at optimization level -O3 (tree-loop-distribute-patterns))

PR tree-optimization/72824
* tree-loop-distribution.c (const_with_all_bytes_same): Verify
real_zerop is not negative.

* gcc.c-torture/execute/ieee/pr72824.c: New test.

From-SVN: r239275

re PR tree-optimization/71802 (gcc ICE at -O3 on valid code on x86_64-linux-gnu in expand_LOOP_VECTORIZED)

2016-08-09  Richard Biener  <rguenther@suse.de>

PR tree-optimization/71802
* tree-cfgcleanup.c (cleanup_tree_cfg_bb): Make sure to catch
all merge opportunities with the predecessor.

* gcc.dg/torture/pr71802.c: New testcase.

From-SVN: r239274

re PR target/68273 (Wrong code on mips/mipsel due to (invalid?) peeking at alignments in function_arg.)

2016-08-09  Richard Biener  <rguenther@suse.de>

PR ipa/68273
* ipa-prop.c (ipa_modify_formal_parameters): Build
parameter types with natural alignment also for the
over-aligned case.

From-SVN: r239273

re PR c++/72809 (ICE on x86_64-linux-gnu (Segmentation fault, tree_check))

PR c++/72809
* rtti.c (get_pseudo_ti_index): Return TK_CLASS_TYPE for
builtin aggregate types without TYPE_BINFO.

* g++.dg/eh/stdarg1.C: New test.

From-SVN: r239272

Implement C++17 constexpr lambda.

gcc/c-family/
* c-cppbuiltin.c (c_cpp_builtins): Update __cpp_constexpr for
C++17 constexpr lambdas.
gcc/cp/
* class.c (finalize_literal_type_property): Handle lambdas.
* constexpr.c (is_valid_constexpr_fn): Likewise.  No longer static.
(explain_invalid_constexpr_fn, cxx_eval_call_expression): Handle
lambdas.
(cxx_eval_constant_expression): Handle capture proxy.
(var_in_constexpr_fn): Don't check for C++14.
(var_in_maybe_constexpr_fn): New.
(potential_constant_expression_1): Use it.  Check DECL_EXPR for
declarations not allowed in constexpr function.
* decl.c (make_rtl_for_nonlocal_decl): Use var_in_maybe_constexpr_fn.
(finish_function): Set DECL_DECLARED_CONSTEXPR_P on lambda members.
* lambda.c (begin_lambda_type): Set CLASSTYPE_LITERAL_P.
(maybe_add_lambda_conv_op): Clear thunk CALL_EXPR location.
(lambda_static_thunk_p): New.
* parser.c (cp_keyword_starts_decl_specifier_p): Add RID_CONSTEXPR.
(CP_PARSER_FLAGS_ONLY_MUTABLE_OR_CONSTEXPR): New enumerator.
(cp_parser_decl_specifier_seq): Handle it.
(cp_parser_lambda_declarator_opt): Use cp_parser_decl_specifier_seq.
* pt.c (instantiate_class_template_1): Set CLASSTYPE_LITERAL_P.
(tsubst_copy_and_build) [CALL_EXPR]: Propagate CALL_FROM_THUNK_P.
* error.c (dump_function_decl): Check TFF_NO_TEMPLATE_BINDINGS.
(dump_expr) [FUNCTION_DECL]: Pass it.

From-SVN: r239268

Fix empty class parameters with constexpr.

PR c++/67131
* class.c (is_really_empty_class): Call complete_type.
* constexpr.c (cxx_eval_constant_expression): Check
is_really_empty_class.
(potential_constant_expression_1): Likewise.  Check for error type.

From-SVN: r239267

Fix warning breaking profiled bootstrap

This patch fixes an bootstrap error with autoprofiledbootstrap
due to uninitiliazed variables, because the compiler cannot
figure out they don't need to be initialized in an error path.
Just always initialize them.

gcc/:

2016-08-08  Andi Kleen  <ak@linux.intel.com>

* tree-vrp.c (get_single_symbol): Always initialize inv and neg.

From-SVN: r239266

Make autofdo tests safer for parallel builds

There were some reports that the autofdo tests are non deterministic
with parallel builds. I wasn't able to reproduce this, but here are
two changes that may help:

- Always use unique file names for temporary files.
- Don't print file names in the test log because the directories
can vary (suggested by Jeff Law)

gcc/testsuite/:

2016-08-08  Andi Kleen  <ak@linux.intel.com>

* lib/profopt.exp: (auto-profopt-execute): Don't include full
test command line in failure log.
(profopt-execute): dito. Make autofdo file names unique.

From-SVN: r239265

Daily bump.

From-SVN: r239264

text/template: reduce maxExecDepth for gccgo further

    We already lowered the limit of recursive template invocations from
    100,000 to 10,000, but the tests still fail occasionally on
    x86_64-pc-linux-gnu when using GNU ld (so that split stacks are not
    fully functional).  Reduce the limit further, to 1000, enough so that
    the test passes consistently.

    Permitting 1000 recursive template invocations still seems capacious
    enough for real world use.

    Reviewed-on: https://go-review.googlesource.com/25590

From-SVN: r239261

c-format.c: suggest the correct format string to use (PR c/64955)

This adds fix-it hints to c-format.c so that it can (sometimes) suggest
the format string the user should have used.

The patch adds selftests for the new code in c-format.c.  These
selftests are thus lang-specific.  This is the first time we've had
lang-specific selftests, and hence the patch also adds a langhook for
running them.  (Note that currently the Makefile only invokes the
selftests for cc1).

gcc/c-family/ChangeLog:
PR c/64955
* c-common.h (selftest::c_format_c_tests): New declaration.
(selftest::run_c_tests): New declaration.
* c-format.c: Include "selftest.h.
(format_warning_va): Add param "corrected_substring" and use
it to add a replacement fix-it hint.
(format_warning_at_substring): Likewise.
(format_warning_at_char): Update for new param of
format_warning_va.
(argument_parser::check_argument_type): Pass "fki" to
check_format_types.
(check_format_types): Add param "fki" and pass it to
format_type_warning.
(deref_n_times): New function.
(get_modifier_for_format_len): New function.
(selftest::test_get_modifier_for_format_len): New function.
(get_format_for_type): New function.
(format_type_warning): Add param "fki" and use it to attempt
to provide hints for argument types when calling
format_warning_at_substring.
(selftest::get_info): New function.
(selftest::assert_format_for_type_streq): New function.
(ASSERT_FORMAT_FOR_TYPE_STREQ): New macro.
(selftest::test_get_format_for_type_printf): New function.
(selftest::test_get_format_for_type_scanf): New function.
(selftest::c_format_c_tests): New function.

gcc/c/ChangeLog:
PR c/64955
* c-lang.c (LANG_HOOKS_RUN_LANG_SELFTESTS): If CHECKING_P, wire
this up to selftest::run_c_tests.
(selftest::run_c_tests): New function.

gcc/ChangeLog:
PR c/64955
* langhooks-def.h (LANG_HOOKS_RUN_LANG_SELFTESTS): New default
do-nothing langhook.
(LANG_HOOKS_INITIALIZER): Add LANG_HOOKS_RUN_LANG_SELFTESTS.
* langhooks.h (struct lang_hooks): Add run_lang_selftests.
* selftest-run-tests.c: Include "tree.h" and "langhooks.h".
(selftest::run_tests): Call lang_hooks.run_lang_selftests.

gcc/testsuite/ChangeLog:
PR c/64955
* gcc.dg/format/diagnostic-ranges.c: Add fix-it hints to expected
output.

From-SVN: r239260

libgo: don't have .lo depend on .lo.dep

    Having each .lo depend on the corresponding .lo.dep caused too many
    rebuilds, because the .lo.dep files are rebuilt when Makefile changes.
    Instead, if the .lo.dep file changes, remove the .lo file.

    Reviewed-on: https://go-review.googlesource.com/25588

From-SVN: r239258

Fix selftest::test_lexer_string_locations_ebcdic for systems without iconv (PR bootstrap/72844)

selftest::test_lexer_string_locations_ebcdic has this clause:

  /* EBCDIC support requires iconv.  */
  if (!HAVE_ICONV)
    return;

leading to a build failure on systems without iconv.  This conditional
works in libcpp due to this in libcpp/internal.h:

  #if HAVE_ICONV
  #include <iconv.h>
  #else
  #define HAVE_ICONV 0
  typedef int iconv_t;  /* dummy */
  #endif

Fix the problem by ensuring that HAVE_ICONV is always defined within
gcc/input.c.

gcc/ChangeLog:
PR bootstrap/72844
* input.c: Ensure that HAVE_ICONV is defined.

From-SVN: r239257

testsuite: fix gotest for absolute srcdir

    The recent changes to Makefile.am mean that if you configure with an
    absolute path as srcdir then gotest will be invoked with absolute paths
    for the files.  That case never worked.  This patch fixes it.

    Reviewed-on: https://go-review.googlesource.com/25587

From-SVN: r239256

pr71078-1.c: Add require-effective-target c99_runtime.

2016-08-08  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>

testsuite/
* gcc.dg/tree-ssa/pr71078-1.c: Add require-effective-target
c99_runtime.
* gcc.dg/tree-ssa/pr71078-1-double.c: Likewise.
* gcc.dg/tree-ssa/pr71078-1-long-double.c: Likewise.
* gcc.dg/tree-ssa/pr71078-2.c: Likewise.
* gcc.dg/tree-ssa/pr71078-2-double.c: Likewise.
* gcc.dg/tree-ssa/pr71078-2-long-double.c: Likewise.
* gcc.dg/tree-ssa/pr71078-3.c: Likewise.

From-SVN: r239255

Use class substring_loc in c-format.c (PR c/52952)

gcc/c-family/ChangeLog:
PR c/52952
* c-format.c: Include "diagnostic.h".
(location_column_from_byte_offset): Delete.
(location_from_offset): Delete.
(format_warning_va): New function.
(format_warning_at_substring): New function.
(format_warning_at_char): New function.
(check_format_arg): Capture location of format_tree and pass to
check_format_info_main.
(argument_parser): Add fields "start_of_this_format" and
"format_string_cst".
(flag_chars_t::validate): Add param "format_string_cst".  Convert
warning_at call using location_from_offset to call to
format_warning_at_char.
(argument_parser::argument_parser): Add param "format_string_cst_"
and use use it to initialize field "format_string_cst".
Initialize new field "start_of_this_format".
(argument_parser::read_format_flags): Convert warning_at call
using location_from_offset to a call to format_warning_at_char.
(argument_parser::read_any_format_left_precision): Likewise.
(argument_parser::read_any_format_precision): Likewise.
(argument_parser::read_any_other_modifier): Likewise.
(argument_parser::find_format_char_info): Likewise, in three places.
(argument_parser::parse_any_scan_set): Likewise, in one place.
(argument_parser::handle_conversions): Likewise, in two places.
(argument_parser::check_argument_type): Add param "fmt_param_loc"
and use it to make a substring_loc.  Pass the latter to
check_format_types.
(check_format_info_main): Add params "fmt_param_loc" and
"format_string_cst".  Convert warning_at calls using
location_from_offset to calls to format_warning_at_char.  Pass the
new params to the arg_parser ctor.  Pass "format_string_cst" to
flag_chars.validate.  Pass "fmt_param_loc" to
arg_parser.check_argument_type.
(check_format_types): Convert first param from a location_t
to a const substring_loc & and rename to "fmt_loc".  Attempt
to extract the range of the relevant parameter and pass it
to format_type_warning.
(format_type_warning): Convert first param from a location_t
to a const substring_loc & and rename to "fmt_loc".  Add
params "param_range" and "type".  Replace calls to warning_at
with calls to format_warning_at_substring.

gcc/testsuite/ChangeLog:
PR c/52952
* gcc.dg/cpp/pr66415-1.c: Likewise.
* gcc.dg/format/asm_fprintf-1.c: Update column numbers.
* gcc.dg/format/c90-printf-1.c: Likewise.
* gcc.dg/format/diagnostic-ranges.c: New test case.

From-SVN: r239253

re PR go/72814 (reflect FAILs on 32-bit Solaris/SPARC: SIGILL)

PR go/72814

    runtime: treat zero-sized result value as void

    Change the FFI interface to treat a call to a function that returns a
    zero-sized result as a call to a function that returns void.

    This is part of the fix for https://gcc.gnu.org/PR72814.  On 32-bit
    SPARC systems, a call to a function that returns a non-zero-sized struct
    is followed by an unimp instruction that describes the size of the
    struct.  The function returns to the address after the unimp
    instruction.  The libffi library can not represent a zero-sized struct,
    so we wind up treating it as a 1-byte struct.  Thus in that case libffi
    calls the function with an unimp instruction, but the function does not
    adjust the return address.  The result is that the program attempts to
    execute the unimp instruction, causing a crash.

    This is part of a change that fixes the crash by treating all functions
    that return zero bytes as functions that return void.

    Reviewed-on: https://go-review.googlesource.com/25585

* go-gcc.cc (Gcc_backend::function_type): If the return type is
zero bytes, treat the function as returning void.
(return_statement): If the return type is zero bytes, don't
actually return any values.

From-SVN: r239252

re PR c++/58706 (ICE with lambda in OpenMP for-loop)

PR c++/58706
* parser.c: Include tree-iterator.h.
(cp_parser_omp_for_loop_init): Move lambda DECL_EXPRs from init
to FOR_BLOCK.
(cp_parser_omp_for_loop): Handle non-STATEMENT_LIST FOR_BLOCK
entries.

* testsuite/libgomp.c++/pr58706.C: New test.

From-SVN: r239251

re PR fortran/72716 (ICE in gfc_resolve_omp_declare_simd, at fortran/openmp.c:5156)

PR fortran/72716
* openmp.c (gfc_match_omp_declare_simd): Don't stick anything into
BLOCK DATA ns, it will be rejected later.

* gfortran.dg/gomp/pr72716.f90: New test.

From-SVN: r239250

re PR middle-end/72781 (-Wuninitialized false positives in OpenMP code)

PR middle-end/72781
* omp-low.c (lower_lastprivate_clauses): Set TREE_NO_WARNING on the
private vars for lastprivate and for linear iterator.

* gcc.dg/gomp/pr72781.c: New test.

From-SVN: r239249

re PR middle-end/68762 (link error for inline function decorated with OpenMP declare simd)

PR middle-end/68762
* omp-simd-clone.c: Include varasm.h.
(simd_clone_create): Copy over DECL_COMDAT, DECL_WEAK, DECL_EXTERNAL,
DECL_VISIBILITY, DECL_VISIBILITY_SPECIFIED, DECL_DLLIMPORT_P and for
DECL_ONE_ONLY call make_decl_one_only.  Fix up spelling in comment and
update function name.

* g++.dg/vect/pr68762-1.cc: New test.
* g++.dg/vect/pr68762-2.cc: New test.
* g++.dg/vect/pr68762.h: New file.

From-SVN: r239248

c-format.c: cleanup of check_format_info_main

gcc/c-family/ChangeLog:
* c-format.c (class flag_chars_t): New class.
(struct length_modifier): New struct.
(class argument_parser): New class.
(flag_chars_t::flag_chars_t): New ctor.
(flag_chars_t::has_char_p): New method.
(flag_chars_t::add_char): New method.
(flag_chars_t::validate): New method.
(flag_chars_t::get_alloc_flag): New method.
(flag_chars_t::assignment_suppression_p): New method.
(argument_parser::argument_parser): New ctor.
(argument_parser::read_any_dollar): New method.
(argument_parser::read_format_flags): New method.
(argument_parser::read_any_format_width): New method.
(argument_parser::read_any_format_left_precision): New method.
(argument_parser::read_any_format_precision): New method.
(argument_parser::handle_alloc_chars): New method.
(argument_parser::read_any_length_modifier): New method.
(argument_parser::read_any_other_modifier): New method.
(argument_parser::find_format_char_info): New method.
(argument_parser::validate_flag_pairs): New method.
(argument_parser::give_y2k_warnings): New method.
(argument_parser::parse_any_scan_set): New method.
(argument_parser::handle_conversions): New method.
(argument_parser::check_argument_type): New method.
(check_format_info_main): Introduce classes argument_parser
and flag_chars_t, moving the code within the loop into methods
of these classes.  Make various locals "const".

From-SVN: r239247

driver-avr.c (specfiles_doc_url): Remove.

* config/avr/driver-avr.c (specfiles_doc_url): Remove.
(avr_diagnose_devicespecs_error): Remove.
(avr_devicespecs_file): Remove composing absolute path for specfile
and its verbose info. Remove conditions to check specs-file,

From-SVN: r239246

PR testsuite/72838 - FAIL: g++.dg/cpp0x/constexpr-cast.C

gcc/testsuite/ChangeLog:
* gcc/testsuite/g++.dg/cpp0x/constexpr-cast.C: Correct target selector.
* gcc/testsuite/g++.dg/warn/overflow-warn-3.C: Same.
* gcc/testsuite/g++.dg/warn/overflow-warn-4.C: Same.

From-SVN: r239242