Daily bump.

From-SVN: r274224

decl.c (grokdeclarator): Use id_loc and EXPR_LOCATION in a few error messages.

/cp
2019-08-08  Paolo Carlini  <paolo.carlini@oracle.com>

* decl.c (grokdeclarator): Use id_loc and EXPR_LOCATION in
a few error messages.

/testsuite
2019-08-08  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/cpp0x/enum20.C: Test location(s) too.
* g++.dg/other/friend3.C: Likewise.
* g++.dg/parse/dtor5.C: Likewise.
* g++.dg/parse/friend7.C: Likewise.
* g++.dg/template/error22.C: Likewise.
* g++.old-deja/g++.brendan/err-msg5.C: Likewise.

From-SVN: r274220

* doc/extend.texi: Add const qualifier to ld intrinsics.

From-SVN: r274219

rs6000: Rename DFP iterator and attr to DDTD and q

This is more in line with the other iterators we have, and a bit easier
to read and write.

* config/rs6000/dfp.md (D64_D128): Rename to ...
(DDTD): ... this, throughout.
(dfp_suffix): Rename to ...
(q): ... this, throughout.

From-SVN: r274218

rs6000: Use iterators in more DFP patterns

I noticed some patterns in dfp.md could use the D64_D128 iterator but
don't yet.  This converts all remaining simple cases.

* config/rs6000/dfp.md (D64_D128): Move earlier in the file.
(dfp_suffix): Ditto.
(adddd3, addtd3): Merge to ...
(add<mode>3 for D64_D128): ... this.
(subdd3, subtd3): Merge to ...
(sub<mode>3 for D64_D128): ... this.
(muldd3, multd3): Merge to ...
(mul<mode>3 for D64_D128): ... this.
(divdd3, divtd3): Merge to ...
(div<mode>3 for D64_D128): ... this.
(*cmpdd_internal1, *cmptd_internal1): Merge to ...
(*cmp<mode>_internal1 for D64_D128): ... this.
(ftruncdd2, ftrunctd2): Merge to ...
(ftrunc<mode>2 for D64_D128): ... this.
(fixdddi2, fixtddi2): Merge to ...
(fix<mode>di2 for D64_D128): ... this.

From-SVN: r274217

RISC-V: Fix C ABI for flattened struct with 0-length bitfield.

gcc/
PR target/91229
* config/riscv/riscv.c (riscv_flatten_aggregate_field): New arg
ignore_zero_width_bit_field_p.  Skip zero size bitfields when true.
Pass into recursive call.
(riscv_flatten_aggregate_argument): New arg.  Pass to
riscv_flatten_aggregate_field.
(riscv_pass_aggregate_in_fpr_pair_p): New local warned.  Call
riscv_flatten_aggregate_argument twice, with false and true as last
arg.  Process result twice.  Compare results and warn if different.
(riscv_pass_aggregate_in_fpr_and_gpr_p): Likewise.

gcc/testsuite/
* gcc.target/riscv/flattened-struct-abi-1.c: New test.
* gcc.target/riscv/flattened-struct-abi-2.c: New test.

From-SVN: r274215

re PR c++/79520 (Spurious caching for constexpr arguments)

PR c++/79520
* g++.dg/cpp1y/constexpr-79520.C: New test.

From-SVN: r274214

[C] Fix bogus nested enum error message

For:

    enum a { A };
    enum a { B };

we emit a bogus error about nested definitions before the real error:

foo.c:2:6: error: nested redefinition of ‘enum a’
    2 | enum a { B };
      |      ^
foo.c:2:6: error: redeclaration of ‘enum a’
foo.c:1:6: note: originally defined here
    1 | enum a { A };
      |      ^

This is because we weren't clearing C_TYPE_BEING_DEFINED once the
definition was over.

I think it's OK to clear C_TYPE_BEING_DEFINED even for a definition
that actually is nested (and so whose outer definition is still open),
since we'll already have given an error by then.  It means that second
and subsequent attempts to define a nested enum will usually get the
redeclaration error instead of the nested error, but that seems just
as accurate (nested_first and nested_second in the test).  The only
exception is if the first nested enum was also invalid by being empty,
but then the enum as a whole has already produced two errors
(nested_empty in the test).

2019-08-08  Richard Sandiford  <richard.sandiford@arm.com>

gcc/c/
* c-decl.c (finish_enum): Clear C_TYPE_BEING_DEFINED.

gcc/testsuite/
* gcc.dg/pr79983.c (enum E): Don't allow an error about nested
definitions.
* gcc.dg/enum-redef-1.c: New test.

From-SVN: r274213

PR c++/87519 - bogus warning with -Wsign-conversion.

* typeck.c (cp_build_binary_op): Use same_type_p instead of comparing
the types directly.

* g++.dg/warn/Wsign-conversion-5.C: New test.

From-SVN: r274211

constexpr.c (inline_asm_in_constexpr_error): New.

* constexpr.c (inline_asm_in_constexpr_error): New.
(cxx_eval_constant_expression) <case ASM_EXPR>: Call it.
(potential_constant_expression_1) <case ASM_EXPR>: Likewise.

* g++.dg/cpp2a/inline-asm3.C: New test.

From-SVN: r274210

P0325R4 to_array from LFTS with updates

As an extension to what the standard requires, this also adds
conditional noexcept-specifiers to the std::to_array functions.

P0325R4 to_array from LFTS with updates
* include/experimental/array (to_array): Qualify call to __to_array.
* include/std/array (__cpp_lib_to_array, to_array): Define for C++20.
* include/std/version (__cpp_lib_to_array): Likewise.
* testsuite/23_containers/array/creation/1.cc: New test.
* testsuite/23_containers/array/creation/2.cc: New test.
* testsuite/23_containers/array/creation/3_neg.cc: New test.
* testsuite/23_containers/array/tuple_interface/tuple_element_neg.cc:
Use zero for dg-error line number.

From-SVN: r274209

Fix file descriptor existence of MinGW.

2019-08-08  Martin Liska  <mliska@suse.cz>

PR bootstrap/91352
* gcc.c (driver::detect_jobserver): Use is_valid_fd.
* lto-wrapper.c (jobserver_active_p): Likewise.
2019-08-08  Martin Liska  <mliska@suse.cz>

PR bootstrap/91352
* libiberty.h (is_valid_fd): New function.
2019-08-08  Martin Liska  <mliska@suse.cz>

PR bootstrap/91352
* lrealpath.c (is_valid_fd): New function.

From-SVN: r274208

When cloning set operator new/delete to false.

2019-08-08  Martin Liska  <mliska@suse.cz>

* cgraphclones.c (set_new_clone_decl_and_node_flags): Drop
IS_OPERATOR_NEW and IS_OPERATOR_DELETE.
(create_version_clone_with_body): Likewise.

From-SVN: r274207

gimplify.c (omp_add_variable): Use GOVD_PRIVATE | GOVD_EXPLICIT for VLA helper variables on target data even if...

* gimplify.c (omp_add_variable): Use GOVD_PRIVATE | GOVD_EXPLICIT
for VLA helper variables on target data even if not GOVD_FIRSTPRIVATE.
(gimplify_scan_omp_clauses): For OMP_CLAUSE_USE_DEVICE_* use just
GOVD_EXPLICIT flags.
(gimplify_omp_workshare): For OMP_TARGET_DATA move all
OMP_CLAUSE_USE_DEVICE_* clauses to the end of clauses chain.
* omp-low.c (scan_sharing_clauses): For OMP_CLAUSE_USE_DEVICE_*
call install_var_field with mask 11 instead of 3.
(lower_omp_target): For OMP_CLAUSE_USE_DEVICE_* use pass
(splay_tree_key) &DECL_UID (var) to build_sender_ref instead of var.
gcc/c/
* c-typeck.c (c_finish_omp_clauses): For C_ORT_OMP
OMP_CLAUSE_USE_DEVICE_* clauses use oacc_reduction_head bitmap
instead of generic_head to track duplicates.
gcc/cp/
* semantics.c (finish_omp_clauses): For C_ORT_OMP
OMP_CLAUSE_USE_DEVICE_* clauses use oacc_reduction_head bitmap
instead of generic_head to track duplicates.
libgomp/
* target.c (gomp_map_vars_internal): For GOMP_MAP_USE_DEVICE_PTR
perform the lookup in the first loop only if !not_found_cnt, otherwise
perform lookups for it in the second loop guarded with
if (not_found_cnt || has_firstprivate).
* testsuite/libgomp.c/target-37.c: New test.
* testsuite/libgomp.c++/target-22.C: New test.

From-SVN: r274206

Daily bump.

From-SVN: r274205

re PR fortran/91359 (logical function X returns .TRUE. - Warning:  spaghetti code)

2019-08-07  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/91359
* pr91359_2.f:  Fix missing hyphen in dg-do
* pr91359_1.f:  Ditto.  Remove RESULT variable to test actual fix!

From-SVN: r274201

re PR c++/67533 (internal compiler error: in build_call_a, at cp/call.c:372)

PR c++/67533
* g++.dg/tls/thread_local-ice5.C: New test.

From-SVN: r274200

[AArch64] Fix INSR for zero floats

We used INSR to handle zero integers but not zero floats.

2019-08-07  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* config/aarch64/constraints.md (Z): Handle floating-point zeros too.
* config/aarch64/predicates.md (aarch64_reg_or_zero): Likewise.

gcc/testsuite/
* gcc.target/aarch64/sve/init_13.c: New test.

From-SVN: r274193

[AArch64] Prefer FPRs over GPRs for INSR

INSR of GPRs involves a cross-file move while INSR of FPRs doesn't.
We should therefore disparage the GPR version relative to the FPR
version.

The patch also adds MOVPRFX handling, but this is only tested
properly by the ACLE.

2019-08-07  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): Add
MOVPRFX alternatives.  Make the GPR alternatives more expensive
than the FPR ones.

gcc/testsuite/
* gcc.target/aarch64/sve/init_12.c: Expect w1 to be moved into
a temporary FPR.

From-SVN: r274192

[AArch64] Prefer FPRs over GPRs for CLASTB

This patch makes the SVE CLASTB GPR alternative more expensive than the
FPR alternative in order to avoid unnecessary cross-file moves.  It also
fixes the prefix used to print the FPR; <vw> only handles 32-bit and
64-bit elements.

2019-08-07  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* config/aarch64/aarch64-sve.md (fold_extract_last_<mode>):
Disparage the GPR alternative relative to the FPR one.
Fix handling of 8-bit and 16-bit FPR values.

gcc/testsuite/
* gcc.target/aarch64/sve/clastb_8.c: New test.

From-SVN: r274191

[AArch64] Merge SVE reduction patterns

The reorg showed that we had an unnecessary separation between
the bitwise and max/min reductions for integers, and the
addition and max/min reductions for fp.

2019-08-07  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* config/aarch64/iterators.md (BITWISEV): Delete.
(SVE_INT_REDUCTION, SVE_FP_REDUCTION): New int iterators.
(optab): Handle UNSPEC_UMAXV, UNSPEC_UMINV, UNSPEC_SMAXV,
UNSPEC_SMINV, UNSPEC_FADDV, UNSPEC_FMAXNMV, UNSPEC_FMAXV,
UNSPEC_FMINNMV, UNSPEC_FMINV.
(bit_reduc_op): Delete.
(sve_int_op): New int attribute.
(sve_fp_op): Handle UNSPEC_FADDV, UNSPEC_FMAXNMV, UNSPEC_FMAXV,
UNSPEC_FMINNMV, UNSPEC_FMINV.
* config/aarch64/aarch64-sve.md
(reduc_<MAXMINV:maxmin_uns>_scal_<SVE_I:mode>)
(*reduc_<MAXMINV:maxmin_uns>_scal_<SVE_I:mode>)
(reduc_<BITWISEV:optab>_scal_<SVE_I:mode>)
(*reduc_<BITWISEV:optab>_scal_<SVE_I:mode>): Merge into...
(reduc_<SVE_INT_REDUCTION:optab>_scal_<SVE_I:mode>)
(*reduc_<SVE_INT_REDUCTION:optab>_scal_<SVE_I:mode>): ...these
new patterns.
(reduc_plus_scal_<SVE_F:mode>, *reduc_plus_scal_<SVE_I:mode>)
(reduc_<FMAXMINV:optab>_scal_<SVE_F:mode>)
(*reduc_<FMAXMINV:optab>_scal_<SVE_F:mode>): Merge into...
(reduc_<SVE_FP_REDUCTION:optab>_scal_<SVE_F:mode>)
(*reduc_<SVE_FP_REDUCTION:optab>_scal_<SVE_F:mode>): ...these
new patterns.

From-SVN: r274190

[AArch64] Merge SVE ternary FP operations

This patch combines the four individual fused multiply-add optabs
into one pattern and uses unspecs instead of rtx codes.  This is
part of a series of patches that change the SVE FP patterns so that
they can describe cases in which the predicate isn't all-true.

2019-08-07  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* config/aarch64/aarch64-sve.md (fma<mode>4, *fma<mode>4)
(fnma<mode>4, *fnma<mode>4, fnms<mode>4, *fnms<mode>4)
(fms<mode>4, *fms<mode>4): Replace with...
(<SVE_COND_FP_TERNARY:optab><SVE_F:mode>4)
(*<SVE_COND_FP_TERNARY:optab><SVE_F:mode>4): ...these new patterns.
Use unspecs instead of rtx codes.
(cond_<optab><mode>, *cond_<optab><mode>_2, *cond_<optab><mode>_4)
(*cond_<optab><mode>_any): Add the predicate to SVE_COND_FP_TERNARY.

From-SVN: r274189

[AArch64] Merge SVE FMAXNM/FMINNM patterns

This patch makes us use the same define_insn for both the smax/smin
and fmax/fmin optabs.  It also continues the process started by
the earlier FP unary patch of moving predicated FP patterns from
rtx codes to unspecs.

There's no need to handle the FMAX and FMIN instructions until
the ACLE patch, since we only use FMAXNM and FMINNM at present.

2019-08-07  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* config/aarch64/iterators.md (SVE_COND_FP_MAXMIN_PUBLIC): New
int iterator.
(maxmin_uns_op): Handle UNSPEC_COND_FMAXNM and UNSPEC_COND_FMINNM.
* config/aarch64/aarch64-sve.md
(<FMAXMIN:su><FMAXMIN:maxmin><SVE_F:mode>3): Rename to...
(<SVE_COND_FP_MAXMIN_PUBLIC:optab><SVE_F:mode>3): ...this and
use a single unspec for the rhs.
(*<su><maxmin><mode>3): Delete.
(<maxmin_uns><SVE_F:mode>3): Use a single unspec for the rhs.

From-SVN: r274188

[AArch64] Merge SVE FP unary patterns

This patch merges the SVE FP rounding patterns with the other SVE
FP unary patterns.

At the moment, we only generate unary FP operations for full vectors,
so we can use (sqrt:VNx4SF ...) etc. in the rtl pattern.  With the ACLE,
it's also possible to generate predicated operations on partial vectors
without specifying a value for inactive lanes.  (sqrt:VNx4SF ...) would
then have different faulting behaviour from the instruction that the
pattern generates.

This patch therefore uses unspecs to represent the operations instead.
Later patches make this change for other patterns.

2019-08-07  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* config/aarch64/iterators.md (UNSPEC_COND_FABS, UNSPEC_COND_FNEG)
(UNSPEC_COND_FRINTA, UNSPEC_COND_FRINTI, UNSPEC_COND_FRINTM)
(UNSPEC_COND_FRINTN, UNSPEC_COND_FRINTP, UNSPEC_COND_FRINTX)
(UNSPEC_COND_FRINTZ, UNSPEC_COND_FSQRT): New unspecs.
(optab, sve_fp_op): Handle them.
(SVE_FP_UNARY): Delete.
(optab): Remove sqrt entry.
(sve_fp_op): Remove neg, abs and sqrt entries.
(SVE_COND_FP_UNARY): New int iterator.
* config/aarch64/aarch64-sve.md (<frint_pattern><mode>2)
(*<frint_pattern><mode>2): Delete.
(<SVE_FP_UNARY:optab><SVE_F:mode>2): Replace with...
(<SVE_COND_FP_UNARY:optab><SVE_F:mode>2): ...this.
(*<SVE_FP_UNARY:optab><SVE_F:mode>2): Replace with...
(*<SVE_COND_FP_UNARY:optab><SVE_F:mode>2): ...this.

From-SVN: r274187

[AArch64] Remove redundant SVE FADDA pattern

*pred_fold_left_plus_<mode> could no longer match anything, since
UNSPEC_FADDA now takes three operands.  Predicated FADDAs should
now go through mask_fold_left_plus_<mode> instead.

2019-08-07  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* config/aarch64/aarch64-sve.md (*pred_fold_left_plus_<mode>): Delete.

From-SVN: r274186

[AArch64] Make SVE UNSPEC_COND_*s match the insn mnemonic

This patch makes the UNSPEC_COND* names match the instruction mnemonics,
rather than having the previous mixture in which some used instructions
while others used operator names.

2019-08-07  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* config/aarch64/iterators.md (UNSPEC_COND_ADD): Rename to...
(UNSPEC_COND_FADD): ...this.
(UNSPEC_COND_SUB): Rename to...
(UNSPEC_COND_FSUB): ...this.
(UNSPEC_COND_MUL): Rename to...
(UNSPEC_COND_FMUL): ...this.
(UNSPEC_COND_DIV): Rename to...
(UNSPEC_COND_FDIV): ...this.
(UNSPEC_COND_MAX): Rename to...
(UNSPEC_COND_FMAXNM): ...this.
(UNSPEC_COND_MIN): Rename to...
(UNSPEC_COND_FMINNM): ...this.
(UNSPEC_COND_LT): Rename to...
(UNSPEC_COND_FCMLT): ...this.
(UNSPEC_COND_LE): Rename to...
(UNSPEC_COND_FCMLE): ...this.
(UNSPEC_COND_EQ): Rename to...
(UNSPEC_COND_FCMEQ): ...this.
(UNSPEC_COND_NE): Rename to...
(UNSPEC_COND_FCMNE): ...this.
(UNSPEC_COND_GE): Rename to...
(UNSPEC_COND_FCMGE): ...this.
(UNSPEC_COND_GT): Rename to...
(UNSPEC_COND_FCMGT): ...this.
(SVE_COND_FP_BINARY, SVE_COND_FP_CMP, optab, cmp_op, sve_fp_op)
(sve_fp_op_rev): Update accordingly.
* config/aarch64/aarch64.c (aarch64_unspec_cond_code): Likewise.

From-SVN: r274185

[AArch64] Reorganise aarch64-sve.md

aarch64-sve.md was getting a bit jumbled, with related operations
separated by unrelated operations.  Also, many SVE instructions need
to have several patterns due to the various ways in which predication
can be used, and it was getting hard to tell which instructions had a
complete set of patterns.

This patch therefore tries to use an explicit grouping scheme.
Banner comments are usually frowned on, but md files have been
a traditional exception (e.g. when using Mike's template for
new targets).

The patch also lists the instructions implemented by each section,
so that it's possible to search the file by instruction mnemonic.

I wouldn't be surprised if I end up having to rip out the contents
section, but I found it useful for the month or so that that I've
been using it locally.  The patch checks that the contents remain
up-to-date by running a checking script during an early stage of
the build.

No functional change intended.

2019-08-07  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* config/aarch64/aarch64-sve.md: Reorganize contents and add
banner comments.
* config/aarch64/check-sve-md.awk: New file.
* config/aarch64/t-aarch64 (s-check-sve-md): New rule.
(insn-conditions.md): Depend on it.

From-SVN: r274184

re PR target/91385 (Zero-extended negation (*negsi2_1_zext) is not generated)

PR target/91385
* config/i386/sse.md (*negsi2_1_zext): Simplify insn pattern.
(*negsi2_cmpz_zext): Ditto.

testsuite/ChangeLog:

PR target/91385
* gcc.target/i386/pr91385.c: New test.

From-SVN: r274183

[AArch64] Remove unused commutative attribute

The commutative attribute was once used by the SVE conditional binary
expanders, but it's now dead code.

2019-08-07  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* config/aarch64/iterators.md (commutative): Remove.

From-SVN: r274182

PR c++/81429 - wrong parsing of constructor with C++11 attribute.

* parser.c (cp_parser_constructor_declarator_p): Handle the scenario
when a parameter declaration begins with [[attribute]].

* g++.dg/cpp0x/gen-attrs-68.C: New test.
* g++.dg/cpp0x/gen-attrs-69.C: New test.

From-SVN: r274181

PR driver/91130 Use CL_DRIVER when handling of COLLECT_GCC_OPTIONS in lto-wrapper.c

Some options are handled differently by the main driver (gcc, g++,
etc) from the back-end compiler programs (cc1, cc1plus, etc) in that
in the driver they do not take an additional argument, while in the
compiler programs they do.  The processing option option CL_DRIVER
controls this alternative interpretation of the options.

The environment variable COLLECT_GCC_OPTIONS is the list of options to
add to a compile if the compiler re-invokes itself at some point.  As
such, the options are driver options, so CL_DRIVER should be used when
processing this list.  Currently lto-wrapper is doing this
incorrectly.

PR driver/91130
* lto-wrapper.c (find_and_merge_options): Use CL_DRIVER when
processing COLLECT_GCC_OPTIONS.
(run_gcc): Likewise.

From-SVN: r274176

PR c++/91346 - Implement P1668R1, allow unevaluated asm in constexpr.

* constexpr.c (cxx_eval_constant_expression): Handle ASM_EXPR.
(potential_constant_expression_1) <case ASM_EXPR>: Allow.
* cp-tree.h (finish_asm_stmt): Adjust.
* parser.c (cp_parser_asm_definition): Grab the locaion of "asm" and
use it.  Change an error to a pedwarn.  Allow asm in C++2a, warn
otherwise.
* pt.c (tsubst_expr): Pass a location down to finish_asm_stmt.
* semantics.c (finish_asm_stmt): New location_t parameter.  Use it.

* g++.dg/cpp2a/inline-asm1.C: New test.
* g++.dg/cpp2a/inline-asm2.C: New test.
* g++.dg/cpp1y/constexpr-neg1.C: Adjust dg-error.

From-SVN: r274169

re PR tree-optimization/91109 ([arm] gcc.c-torture/execute/20040709-1.c fails since r273135)

2019-08-07  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        PR tree-optimization/91109
        * lra-remat.c (update_scratch_ops): Remove assignment of the
        hard register.

From-SVN: r274163

Make IPA predicate::size a poly_int64

This patch changes the IPA predicate::size field from a HOST_WIDE_INT
to a poly_int64.

2019-08-07  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* data-streamer.h (streamer_write_poly_uint64): Declare.
(streamer_read_poly_uint64): Likewise.
* data-streamer-in.c (streamer_read_poly_uint64): New function.
* data-streamer-out.c (streamer_write_poly_uint64): Likewise.
* ipa-predicate.h (condition::size): Turn into a poly_int64.
(add_condition): Take a poly_int64 size.
* ipa-predicate.c (add_condition): Likewise.
* ipa-prop.h (ipa_load_from_parm_agg): Take a poly_int64 size pointer.
* ipa-prop.c (ipa_load_from_parm_agg): Likewise.
(ipcp_modif_dom_walker::before_dom_children): Update accordingly.
* ipa-fnsummary.c (evaluate_conditions_for_known_args): Handle
condition::size as a poly_int64.
(unmodified_parm_1): Take a poly_int64 size pointer.
(unmodified_parm): Likewise.
(unmodified_parm_or_parm_agg_item): Likewise.
(set_cond_stmt_execution_predicate): Update accordingly.
(set_switch_stmt_execution_predicate): Likewise.
(will_be_nonconstant_expr_predicate): Likewise.
(will_be_nonconstant_predicate): Likewise.
(inline_read_section): Stream condition::size as a poly_int.
(ipa_fn_summary_write): Likewise.

From-SVN: r274162

Replace int with boolean in predicate functions.

2019-08-07  Martin Liska  <mliska@suse.cz>

* fold-const.c (twoval_comparison_p): Replace int
with bool as a return type.
(simple_operand_p): Likewise.
(operand_equal_p): Replace int with bool as a return type.
* fold-const.h (operand_equal_p): Likewise.

From-SVN: r274161

PR 53796 Make inquire(file=, recl=) conform to F2018

In my original patch to fix PR 53796 I forgot to fix the behavior for
unconnected units when inquiring via filename. This patch fixes that.

Regtested on x86_64-pc-linux-gnu, committed as obvious.

libgfortran/ChangeLog:

2019-08-07  Janne Blomqvist  <jb@gcc.gnu.org>

PR fortran/53796
* io/inquire.c (inquire_via_filename): Set recl to -1 for
unconnected units.

gcc/testsuite/ChangeLog:

2019-08-07  Janne Blomqvist  <jb@gcc.gnu.org>

PR fortran/53796
* gfortran.dg/inquire_recl_f2018.f90: Test for unconnected unit
with inquire via filename.

From-SVN: r274160

tree-core.h (enum omp_clause_code): Adjust OMP_CLAUSE_USE_DEVICE_PTR OpenMP description.

* tree-core.h (enum omp_clause_code): Adjust OMP_CLAUSE_USE_DEVICE_PTR
OpenMP description.  Add OMP_CLAUSE_USE_DEVICE_ADDR clause.
* tree.c (omp_clause_num_ops, omp_clause_code_name): Add entries
for OMP_CLAUSE_USE_DEVICE_ADDR clause.
(walk_tree_1): Handle OMP_CLAUSE_USE_DEVICE_ADDR.
* tree-pretty-print.c (dump_omp_clause): Likewise.
* tree-nested.c (convert_nonlocal_omp_clauses,
convert_local_omp_clauses): Likewise.
* gimplify.c (gimplify_scan_omp_clauses, gimplify_adjust_omp_clauses):
Likewise.
* omp-low.c (scan_sharing_clauses, lower_omp_target): Likewise.
Treat OMP_CLAUSE_USE_DEVICE_ADDR like OMP_CLAUSE_USE_DEVICE_PTR
clause with array or reference to array types, no matter what type
except for reference it has.
gcc/c-family/
* c-pragma.h (enum pragma_omp_clause): Add
PRAGMA_OMP_CLAUSE_USE_DEVICE_ADDR.  Set PRAGMA_OACC_CLAUSE_USE_DEVICE
equal to PRAGMA_OMP_CLAUSE_USE_DEVICE_PTR instead of being a separate
enumeration value.
gcc/c/
* c-parser.c (c_parser_omp_clause_name): Parse use_device_addr clause.
(c_parser_omp_clause_use_device_addr): New function.
(c_parser_omp_all_clauses): Handle PRAGMA_OMP_CLAUSE_USE_DEVICE_ADDR.
(OMP_TARGET_DATA_CLAUSE_MASK): Add PRAGMA_OMP_CLAUSE_USE_DEVICE_ADDR.
(c_parser_omp_target_data): Handle PRAGMA_OMP_CLAUSE_USE_DEVICE_ADDR
like PRAGMA_OMP_CLAUSE_USE_DEVICE_PTR, adjust diagnostics about no
map or use_device_* clauses.
* c-typeck.c (c_finish_omp_clauses): For OMP_CLAUSE_USE_DEVICE_PTR
in OpenMP, require pointer type rather than pointer or array type.
Handle OMP_CLAUSE_USE_DEVICE_ADDR.
gcc/cp/
* parser.c (cp_parser_omp_clause_name): Parse use_device_addr clause.
(cp_parser_omp_all_clauses): Handle PRAGMA_OMP_CLAUSE_USE_DEVICE_ADDR.
(OMP_TARGET_DATA_CLAUSE_MASK): Add PRAGMA_OMP_CLAUSE_USE_DEVICE_ADDR.
(cp_parser_omp_target_data): Handle PRAGMA_OMP_CLAUSE_USE_DEVICE_ADDR
like PRAGMA_OMP_CLAUSE_USE_DEVICE_PTR, adjust diagnostics about no
map or use_device_* clauses.
* semantics.c (finish_omp_clauses): For OMP_CLAUSE_USE_DEVICE_PTR
in OpenMP, require pointer or reference to pointer type rather than
pointer or array or reference to pointer or array type. Handle
OMP_CLAUSE_USE_DEVICE_ADDR.
* pt.c (tsubst_omp_clauses): Handle OMP_CLAUSE_USE_DEVICE_ADDR.
gcc/testsuite/
* c-c++-common/gomp/target-data-1.c (foo): Use use_device_addr clause
instead of use_device_ptr clause where required by OpenMP 5.0, add
further tests for both use_device_ptr and use_device_addr clauses.
libgomp/
* testsuite/libgomp.c/target-18.c (struct S): New type.
(foo): Use use_device_addr clause instead of use_device_ptr clause
where required by OpenMP 5.0, add further tests for both use_device_ptr
and use_device_addr clauses.
* testsuite/libgomp.c++/target-9.C (struct S): New type.
(foo): Use use_device_addr clause instead of use_device_ptr clause
where required by OpenMP 5.0, add further tests for both use_device_ptr
and use_device_addr clauses.  Add t and u arguments.
(main): Adjust caller.

From-SVN: r274159

rs6000: support vector int type rotatert

gcc/ChangeLog

2019-08-07  Kewen Lin  <linkw@gcc.gnu.org>

    * config/rs6000/vector.md (vrotr<mode>3): New define_expand.

gcc/testsuite/ChangeLog

2019-08-07  Kewen Lin  <linkw@gcc.gnu.org>

    * gcc.target/powerpc/vec_rotate-1.c: New test.
    * gcc.target/powerpc/vec_rotate-2.c: New test.
    * gcc.target/powerpc/vec_rotate-3.c: New test.
    * gcc.target/powerpc/vec_rotate-4.c: New test.

From-SVN: r274158

wrap math.h for M_PI et al in target/i386 tests

Most but not all of the tests that expect M_PI, M_PI_2 and/or M_PI_4
to be defined in math.h explicitly exclude one target system that does
not satisfy this non-standard assumption.

This patch introduces a wrapper header that includes math.h and then
conditionally supplies the missing non-standard macro definitions.
With that, we can drop the dg-skip-if "no M_PI" exclusions.

for  gcc/testsuite/ChangeLog

* gcc.target/i386/math_m_pi.h: New.
* gcc.target/i386/sse4_1-round.h: Use it.
* gcc.target/i386/pr73350.c: Likewise.
* gcc.target/i386/avx512f-vfixupimmpd-2.c: Likewise.
* gcc.target/i386/avx512f-vfixupimmps-2.c: Likewise.
* gcc.target/i386/avx512f-vfixupimmsd-2.c: Likewise.
* gcc.target/i386/avx512f-vfixupimmss-2.c: Likewise.
* gcc.target/i386/avx512f-vfixupimmss-2.c: Likewise.
* gcc.target/i386/avx-ceil-sfix-2-vec.c: Likewise.  Drop
dg-skip-if "no M_PI".
* gcc.target/i386/avx-cvt-2-vec.c: Likewise.
* gcc.target/i386/avx-floor-sfix-2-vec.c: Likewise.
* gcc.target/i386/avx-rint-sfix-2-vec.c: Likewise.
* gcc.target/i386/avx-round-sfix-2-vec.c: Likewise.
* gcc.target/i386/avx512f-ceil-sfix-vec-1.c: Likewise.
* gcc.target/i386/avx512f-ceil-vec-1.c: Likewise.
* gcc.target/i386/avx512f-ceilf-sfix-vec-1.c: Likewise.
* gcc.target/i386/avx512f-ceilf-vec-1.c: Likewise.
* gcc.target/i386/avx512f-floor-sfix-vec-1.c: Likewise.
* gcc.target/i386/avx512f-floor-vec-1.c: Likewise.
* gcc.target/i386/avx512f-floorf-sfix-vec-1.c: Likewise.
* gcc.target/i386/avx512f-floorf-vec-1.c: Likewise.
* gcc.target/i386/avx512f-rint-sfix-vec-1.c: Likewise.
* gcc.target/i386/avx512f-rintf-sfix-vec-1.c: Likewise.
* gcc.target/i386/avx512f-round-sfix-vec-1.c: Likewise.
* gcc.target/i386/avx512f-roundf-sfix-vec-1.c: Likewise.
* gcc.target/i386/avx512f-trunc-vec-1.c: Likewise.
* gcc.target/i386/avx512f-truncf-vec-1.c: Likewise.
* gcc.target/i386/sse2-cvt-vec.c: Likewise.
* gcc.target/i386/sse4_1-ceil-sfix-vec.c: Likewise.
* gcc.target/i386/sse4_1-ceil-vec.c: Likewise.
* gcc.target/i386/sse4_1-ceilf-sfix-vec.c: Likewise.
* gcc.target/i386/sse4_1-ceilf-vec.c: Likewise.
* gcc.target/i386/sse4_1-floor-sfix-vec.c: Likewise.
* gcc.target/i386/sse4_1-floor-vec.c: Likewise.
* gcc.target/i386/sse4_1-floorf-sfix-vec.c: Likewise.
* gcc.target/i386/sse4_1-floorf-vec.c: Likewise.
* gcc.target/i386/sse4_1-rint-sfix-vec.c: Likewise.
* gcc.target/i386/sse4_1-rint-vec.c: Likewise.
* gcc.target/i386/sse4_1-rintf-sfix-vec.c: Likewise.
* gcc.target/i386/sse4_1-rintf-vec.c: Likewise.
* gcc.target/i386/sse4_1-round-sfix-vec.c: Likewise.
* gcc.target/i386/sse4_1-round-vec.c: Likewise.
* gcc.target/i386/sse4_1-roundf-sfix-vec.c: Likewise.
* gcc.target/i386/sse4_1-roundf-vec.c: Likewise.
* gcc.target/i386/sse4_1-roundsd-4.c: Likewise.
* gcc.target/i386/sse4_1-roundss-4.c: Likewise.
* gcc.target/i386/sse4_1-trunc-vec.c: Likewise.
* gcc.target/i386/sse4_1-truncf-vec.c: Likewise.

From-SVN: r274157

RISC-V: Handle g extension in multilib-generator

gcc/ChangeLog

* gcc/config/riscv/multilib-generator: (canonical_order): Add 'g'.
(arch_canonicalize): Support rv32g and rv64g and fix error
handling.

From-SVN: r274156

Daily bump.

From-SVN: r274155

re PR fortran/91359 (logical function X returns .TRUE. - Warning:  spaghetti code)

2019-08-06  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/91359
* trans-decl.c (gfc_generate_return): Ensure something is returned
from a function.

2019-08-06  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/91359
* gfortran.dg/pr91359_1.f: New test.
* gfortran.dg/pr91359_2.f: Ditto.

From-SVN: r274149

re PR fortran/42546 (ALLOCATED statement typo in the docs and for scalar variables)

2019-08-01  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/42546
* check.c(gfc_check_allocated): Add comment pointing to ...
* intrinsic.c(sort_actual): ... the checking done here.

2019-08-01  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/42546
* gfortran.dg/allocated_1.f90: New test.
* gfortran.dg/allocated_2.f90: Ditto.

From-SVN: r274147

P1651R0 bind_front should not unwrap reference_wrapper

P1651R0 bind_front should not unwrap reference_wrapper
* include/std/functional (bind_front): Don't unwrap reference_wrapper.
* include/std/version (__cpp_lib_bind_front): Update value.
* testsuite/20_util/function_objects/bind_front/1.cc: Fix test for
feature test macro.
* testsuite/20_util/function_objects/bind_front/2.cc: New test.

From-SVN: r274146

Specialize std::numbers constants for __float128

* include/std/numbers [!__STRICT_ANSI__ && _GLIBCXX_USE_FLOAT128]
(e_v, log2e_v, log10e_v, pi_v, inv_pi_v, inv_sqrtpi_v, ln2_v, ln10_v)
(sqrt2_v, sqrt3_v, inv_sqrt3, egamma_v, phi_v): Add explicit
specializations for __float128.
* testsuite/26_numerics/numbers/float128.cc: New test.

From-SVN: r274145

Fix gcc.target/i386/avx512vp2intersect-2intersect-1b.c etc. execution tests

gcc/testsuite:
* gcc.target/i386/avx512vp2intersect-2intersect-1b.c (AVX512F):
Remove.
(AVX512VP2INTERSECT): Define.
* gcc.target/i386/avx512vp2intersect-2intersectvl-1b.c (AVX512F):
Remove.
(AVX512VP2INTERSECT): Define.

From-SVN: r274144

PR c++/91378 - ICE with noexcept and auto return type.

Here, since the call to g is not type-dependent, we call mark_used on it to
determine its return type.  This also wants to instantiate the
noexcept-expression.  But since nothing in maybe_instantiate_noexcept was
calling push_to_top_level, we substituted b.i with processing_template_decl
set, so we left it unresolved for later access checking.  As a result, the
type of C::g<int> remained instantiation-dependent, leading to an ICE in
type_dependent_expression_p on the assert that the type of a function
template with no dependent template arguments must be non-dependent.

* pt.c (maybe_instantiate_noexcept): push_to_top_level.

From-SVN: r274143

decl.c (check_array_designated_initializer): Use cp_expr_loc_or_input_loc in one place.

/cp
2019-08-06  Paolo Carlini  <paolo.carlini@oracle.com>

* decl.c (check_array_designated_initializer): Use
cp_expr_loc_or_input_loc in one place.

/testsuite
2019-08-06  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/cpp0x/desig1.C: Check location too.

From-SVN: r274141

Add operator new/delete to cgraph_node::dump.

2019-08-06  Martin Liska  <mliska@suse.cz>

* cgraph.c (cgraph_node::dump): Dump DECL_IS_OPERATOR_NEW_P
and DECL_IS_OPERATOR_DELETE_P.

From-SVN: r274139

tree.h (OMP_CLAUSE_LASTPRIVATE_TASKLOOP_IV): Rename to ...

* tree.h (OMP_CLAUSE_LASTPRIVATE_TASKLOOP_IV): Rename to ...
(OMP_CLAUSE_LASTPRIVATE_LOOP_IV): ... this.  Adjust comment.
* gimplify.c (gimple_add_tmp_var): In SIMD contexts, turn addressable
new vars into GOVD_PRIVATE rather than GOVD_LOCAL.
(gimplify_omp_for): Don't do C++ random access iterator clause
adjustments on combined constructs from OMP_LOOP.  For OMP_LOOP,
don't predetermine the artificial iterator in case of C++ random
access iterators as lastprivate, but private.  For OMP_LOOP, force
bind expr around simd body and force for_pre_body before the
construct.  Use OMP_CLAUSE_LASTPRIVATE_LOOP_IV instead of
OMP_CLAUSE_LASTPRIVATE_TASKLOOP_IV.
(gimplify_omp_loop): Add firstprivate clauses on OMP_PARALLEL for
diff var of C++ random access iterators.  Handle
OMP_CLAUSE_FIRSTPRIVATE.  For OMP_CLAUSE_LASTPRIVATE_LOOP_IV, if
not outermost also add OMP_CLAUSE_FIRSTPRIVATE, and in both cases
clear OMP_CLAUSE_LASTPRIVATE_LOOP_IV on the lastprivate clause
on the OMP_FOR and OMP_DISTRIBUTE constructs if any.
* omp-low.c (lower_rec_input_clauses): For
OMP_CLAUSE_LASTPRIVATE_LOOP_IV on simd copy construct the private
variables instead of default constructing them.
(lower_lastprivate_clauses): Use OMP_CLAUSE_LASTPRIVATE_LOOP_IV
instead of OMP_CLAUSE_LASTPRIVATE_TASKLOOP_IV and move the
is_taskloop_ctx check from the assert to the guarding condition.
gcc/cp/
* parser.c (cp_parser_omp_for_loop): For OMP_LOOP, ignore parallel
clauses and predetermine iterator as lastprivate.
* semantics.c (handle_omp_for_class_iterator): Use
OMP_CLAUSE_LASTPRIVATE_LOOP_IV instead of
OMP_CLAUSE_LASTPRIVATE_TASKLOOP_IV, set it for lastprivate also
on OMP_LOOP construct.  If a clause is missing for class iterator
on OMP_LOOP, add firstprivate clause, and if there is private
clause, turn it into firstprivate too.
(finish_omp_for): Formatting fix.  For OMP_LOOP, adjust
OMP_CLAUSE_LASTPRIVATE_LOOP_IV clause CP_CLAUSE_INFO, so that it
uses copy ctor instead of default ctor.
* cp-gimplify.c (cp_gimplify_expr): Handle OMP_LOOP like
OMP_DISTRIBUTE etc.
(cp_fold_r): Likewise.
(cp_genericize_r): Likewise.
(cxx_omp_finish_clause): Also finish lastprivate clause with
OMP_CLAUSE_LASTPRIVATE_LOOP_IV flag.
* pt.c (tsubst_omp_clauses): Handle OMP_CLAUSE_BIND.
(tsubst_omp_for_iterator): For OMP_LOOP, ignore parallel
clauses and predetermine iterator as lastprivate.
* constexpr.c (potential_constant_expression_1): Handle OMP_LOOP
like OMP_DISTRIBUTE etc.
libgomp/
* testsuite/libgomp.c++/loop-13.C: New test.
* testsuite/libgomp.c++/loop-14.C: New test.
* testsuite/libgomp.c++/loop-15.C: New test.

From-SVN: r274138

RISC-V: Handle extensions combination correctly in multilib-generator.

Input string consist of four parts:
  <primary arch>-<abi>-<additional arches>-<extensions>

and generator doing combination with <extensions> part, but it just
append the extension at the end of arch string, it might generate
invalid arch.

For example, without this patch `./multilib-generator rv32imafc-ilp32--d`
will generate:

MULTILIB_OPTIONS = march=rv32imafc/march=rv32imafcd mabi=ilp32
                                         ^^^^^^^^^^

and rv32imafcd is not in canonical order.

Tested with python 2.7 and python 3.6/3.7.

gcc/ChangeLog

* gcc/config/riscv/multilib-generator: (canonical_order): New.
(arch_canonicalize): Dito.
Apply arch_canonicalize for alts.

From-SVN: r274137

DR 2413 - typename in conversion-function-ids.

* parser.c (cp_parser_conversion_type_id): Call
cp_parser_type_specifier_seq with CP_PARSER_FLAGS_TYPENAME_OPTIONAL
instead of CP_PARSER_FLAGS_NONE.

* g++.dg/cpp2a/typename17.C: New test.

From-SVN: r274136

PR middle-end/50476 - Warn of pointer set to object whose lifetime is limited

gcc/testsuite/ChangeLog:
* gcc.dg/uninit-pr50476.c: New test.

From-SVN: r274135

Daily bump.

From-SVN: r274134

PR c++/60517 - warning/error for taking address of member of a temporary object

testsuite/ChangeLog:
* g++.dg/pr60517.C: New test.

From-SVN: r274130

Add missing ChangeLog entry.

2019-08-02  Michael Meissner  <meissner@linux.ibm.com>

* config/rs6000/future.md: New file.
* config/rs6000/rs6000.md: Include future.md.
* config/rs6000/t-rs6000 (MD_INCLUDES): Add future.md.

From-SVN: r274129

extend.texi (Common Variable Attributes): Document alias attribute.

gcc/ChangeLog:

* doc/extend.texi (Common Variable Attributes): Document alias
attribute.

From-SVN: r274127

cp-tree.h (cp_expr_loc_or_input_loc): New.

2019-08-05  Paolo Carlini  <paolo.carlini@oracle.com>

* cp-tree.h (cp_expr_loc_or_input_loc): New.
(cxx_incomplete_type_diagnostic): Use it.
* call.c (build_converted_constant_expr_internal, convert_like_real,
convert_arg_to_ellipsis, convert_for_arg_passing, build_over_call,
build_cxx_call, perform_implicit_conversion_flags,
initialize_reference): Likewise.
* constexpr.c (cxx_eval_internal_function, cxx_eval_call_expression,
eval_and_check_array_index, cxx_eval_store_expression,
cxx_eval_statement_list, cxx_eval_loop_expr,
cxx_eval_constant_expression, potential_constant_expression_1):
Likewise.
* constraint.cc (check_for_logical_overloads,
satisfy_predicate_constraint): Likewise.
* cp-gimplify.c (cp_gimplify_expr): Likewise.
* cvt.c (cp_convert_to_pointer, convert_to_reference,
cp_convert_and_check, ocp_convert, maybe_warn_nodiscard): Likewise.
* decl.c (pop_switch): Likewise.
* decl2.c (delete_sanity): Likewise.
* error.c (location_of): Likewise.
* init.c (maybe_warn_list_ctor, build_aggr_init,
warn_placement_new_too_small, build_new_1, build_vec_init): Likewise.
* lex.c (unqualified_name_lookup_error): Likewise.
* parser.c (cp_parser_initializer_list, cp_parser_omp_for_cond):
Likewise.
* pt.c (check_for_bare_parameter_packs, check_valid_ptrmem_cst_expr,
unify_arg_conversion, convert_nontype_argument,
tsubst_copy_and_build, resolve_typename_type): Likewise.
* semantics.c (maybe_convert_cond, finish_call_expr,
cp_build_vec_convert): Likewise.
* typeck.c (decay_conversion, rationalize_conditional_expr,
cp_build_unary_op, build_x_compound_expr_from_list,
maybe_warn_about_returning_address_of_local,
maybe_warn_pessimizing_move): Likewise.
* typeck2.c (check_narrowing, digest_init_r,
process_init_constructor_array): Likewise.

From-SVN: r274124

PR c++/88095, CTAD for literal operator templates per P0732

This patch fixes PR c++/88095: class nontype template parameter UDL string
literals doesn't accepts deduction placeholder

It also addresses a latent issue; literal operator templates with template
parameter packs of literal class type were previously accepted.  The patch
corrects this and adds a test (udlit-class-nttp-neg.C).

This fix is needed for one of the char8_t remediation approaches documented
in P1423, and may be helpful for existing code bases impacted by the char8_t
changes adopted via P0482 for C++20.

gcc/cp/ChangeLog:

2019-08-02  Tom Honermann  <tom@honermann.net>

        * parser.c (cp_parser_template_declaration_after_parameters): Enable
        class template argument deduction for non-type template parameters
        in literal operator templates.

gcc/testsuite/ChangeLog:

2019-08-02  Tom Honermann  <tom@honermann.net>

        PR c++/88095
        * g++.dg/cpp2a/udlit-class-nttp-ctad.C: New test.
        * g++.dg/cpp2a/udlit-class-nttp-ctad-neg.C: New test.
        * g++.dg/cpp2a/udlit-class-nttp-ctad-neg2.C: New test.
        * g++.dg/cpp2a/udlit-class-nttp.C: New test.
        * g++.dg/cpp2a/udlit-class-nttp-neg.C: New test.
        * g++.dg/cpp2a/udlit-class-nttp-neg2.C: New test.

From-SVN: r274123

re PR fortran/91372 (Error: Unclassifiable statement)

2019-08-05  Steven g. Kargl  <kargl@gcc.gnu.org>

PR fortran/91372
* decl.c (gfc_match_data): Allow an implied do-loop to nestle against
DATA.

2019-08-05  Steven g. Kargl  <kargl@gcc.gnu.org>

PR fortran/91372
* gfortran.dg/pr91372.f90: New test.

From-SVN: r274122

PR c++/91338 - Implement P1161R3: Deprecate a[b,c].

* c-opts.c (c_common_post_options): Enable -Wcomma-subscript by
default for C++2a, unless -Wno-deprecated.
* c.opt (Wcomma-subscript): New warning.

* parser.c (cp_parser_postfix_open_square_expression): Warn about uses
of a comma operator within a subscripting expression.
(cp_parser_skip_to_closing_square_bracket_1): New function, made out
of...
(cp_parser_skip_to_closing_square_bracket): ...this.

* doc/invoke.texi: Document -Wcomma-subscript.

* g++.dg/cpp2a/comma1.C: New test.
* g++.dg/cpp2a/comma2.C: New test.
* g++.dg/cpp2a/comma3.C: New test.
* g++.dg/cpp2a/comma4.C: New test.

From-SVN: r274121

* semantics.c (force_paren_expr): Preserve location.

From-SVN: r274120

Make function_code a 32-bit field

Adding SVE intrinsics on top of the existing AArch64 intrinsics blows
the 12-bit function_code in tree_function_decl.  That bitfield has no
spare bits, but it comes at the end of the structure and is preceded
by a pointer, so on LP64 hosts there's currently a 32-bit hole at end.

This patch therefore makes function_code an independent field and
moves the bitfield to the 32-bit hole.

I wondered about instead making function_code 16 bits, so that the
patch leaves 28 spare bits instead of just 12.  That seemed a bit
short-term though; I can't guarantee that we won't blow 16 bits once
the SVE2 functions are added...

If we run out of bits again, we can start chomping from the top
of the enum.  E.g. 24 bits should surely be enough, but there's
no point paying the overhead of the masking until we need it.

2019-08-05  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* tree-core.h (tree_function_decl): Make function_code an
independent field.  Group the remaining bitfields into bytes
and move decl_type so that it contines to be at a byte boundary.
Leave 12 bits for future expansion.

From-SVN: r274119

Fold MASK_LOAD/STORE with an all-true mask

This patch folds IFN_MASK_LOAD and IFN_MASK_STOREs to normal accesses
if the mask is all-true.  This can happen for fully-masked loops that
didn't actually need to be (which could be handled by the vectoriser
instead), or for unrolled fully-masked loops whose first iteration is
guaranteed to operate on a full vector.  It's also useful when the
accesses are generated directly by intrinsics (to follow for SVE).

2019-08-05  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* gimple-fold.c (gimple_fold_mask_load_store_mem_ref)
(gimple_fold_mask_load, gimple_fold_mask_store): New functions.
(gimple_fold_call): Use them to fold IFN_MASK_LOAD and
IFN_MASK_STORE.

gcc/testsuite/
* gcc.target/aarch64/sve/mask_load_1.c: New test.

From-SVN: r274118

Add a gimple_move_vops helper function

I needed to add another instance of this idiom, so thought it'd
be worth having a helper function.

2019-08-05  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* gimple.h (gimple_move_vops): Declare.
* gimple.c (gimple_move_vops): New function
* gimple-fold.c (replace_call_with_call_and_fold)
(gimple_fold_builtin_memory_op, gimple_fold_builtin_memset)
(gimple_fold_builtin_stpcpy, fold_builtin_atomic_compare_exchange)
(gimple_fold_call): Use it.
* ipa-param-manipulation.c (ipa_modify_call_arguments): Likewise.
* tree-call-cdce.c (use_internal_fn): Likewise.
* tree-if-conv.c (predicate_load_or_store): Likewise.
* tree-ssa-ccp.c (optimize_atomic_bit_test_and): Likewise.
* tree-ssa-math-opts.c (pass_cse_reciprocals::execute): Likewise.
* tree-ssa-propagate.c (finish_update_gimple_call): Likewise.
(update_call_from_tree): Likewise.
* tree-vect-stmts.c (vectorizable_load): Likewise.
* tree-vectorizer.c (adjust_simduid_builtins): Likewise.

From-SVN: r274117

pr80993.c: Add cleanup-saved-temps to final actions.

2019-08-05  Jozef Lawrynowicz  <jozef.l@mittosystems.com>

* gcc.target/msp430/pr80993.c: Add cleanup-saved-temps to final
actions.

From-SVN: r274116

Handle new operators with no arguments in DCE.

2019-08-05  Martin Liska  <mliska@suse.cz>

PR c++/91334
* tree-ssa-dce.c (propagate_necessity): Handle new operators
with not arguments.
(eliminate_unnecessary_stmts): Likewise.
2019-08-05  Martin Liska  <mliska@suse.cz>

PR c++/91334
* g++.dg/torture/pr91334.C: New test.

From-SVN: r274115

re PR tree-optimization/91169 (cd2a31a FAILs)

2019-08-05  Richard Biener  <rguenther@suse.de>

PR middle-end/91169
* fold-const.c (get_array_ctor_element_at_index): Create
offset_ints according to the sign of the index type and treat
that as signed if it is obviously so.

* gnat.dg/array37.adb: New testcase.

From-SVN: r274114

re PR target/91341 (Missing AVX Intrinsics: load/store u2)

PR target/91341
* config/i386/avxintrin.h (_mm256_loadu2_m128, _mm256_storeu2_m128,
_mm256_loadu2_m128d, _mm256_storeu2_m128d, _mm256_loadu2_m128i,
_mm256_storeu2_m128i): New function.

* gcc.target/i386/avx-loadu2-m128-1.c: New test.
* gcc.target/i386/avx-loadu2-m128-2.c: New test.
* gcc.target/i386/avx-loadu2-m128d-1.c: New test.
* gcc.target/i386/avx-loadu2-m128d-2.c: New test.
* gcc.target/i386/avx-loadu2-m128i-1.c: New test.
* gcc.target/i386/avx-loadu2-m128i-2.c: New test.
* gcc.target/i386/avx-storeu2-m128-1.c: New test.
* gcc.target/i386/avx-storeu2-m128-2.c: New test.
* gcc.target/i386/avx-storeu2-m128d-1.c: New test.
* gcc.target/i386/avx-storeu2-m128d-2.c: New test.
* gcc.target/i386/avx-storeu2-m128i-1.c: New test.
* gcc.target/i386/avx-storeu2-m128i-2.c: New test.

From-SVN: r274109

RISC-V: Promote type correctly for libcalls

 - argument and return value for libcall won't promote at
   default_promote_function_mode_always_promote, however we expect it
   should sign-extend as normal function.

 - Witout this patch, this test case will fail at -march=rv64i -mabi=lp64.

 - The implementation of riscv_promote_function_mode is borrowed from MIPS.

gcc/ChangeLog

* config/riscv/riscv.c (riscv_promote_function_mode): New.
(TARGET_PROMOTE_FUNCTION_MODE): Use riscv_promote_function_mode.

gcc/testsuite/ChangeLog

* gcc.target/riscv/promote-type-for-libcall.c: New.

From-SVN: r274107

PR91349, powerpc64*-*-freebsd* defines _GNU_SOURCE

rev 266496 (git ab6b1bb456) undefined some macros in rs6000/freebsd.h
but missed doing the same in rs6000/freebsd64.h.

PR target/91349
* config/rs6000/freebsd64.h (CPLUSPLUS_CPP_SPEC),
(LINK_GCC_C_SEQUENCE_SPEC): Undef.

From-SVN: r274105

Daily bump.

From-SVN: r274103

update_web_docs_svn: Proceed even if the invocation of sphinx fails.

* update_web_docs_svn: Proceed even if the invocation of
sphinx fails.

From-SVN: r274099

re PR fortran/88227 (ICE in gfc_convert_boz, at fortran/target-memory.c:788)

2019-08-04  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/88227
* check.c (oct2bin):  New function.  Convert octal string to binary.
(hex2bin): New function.  Convert hexidecimal string to binary.
(bin2real): New function.  Convert binary string to REAL.  Use
oct2bin and hex2bin.
(gfc_boz2real):  Use fallback conversion bin2real.

From-SVN: r274096

documentation_hacking.xml: doxygen.org is now doxygen.nl.

* doc/xml/manual/documentation_hacking.xml: doxygen.org is now
doxygen.nl.

From-SVN: r274094

install.texi (Prerequisites): Remove reference to Tcl 8.6 bug that was fixed in Tcl 8.6.1.

* doc/install.texi (Prerequisites): Remove reference to Tcl 8.6
bug that was fixed in Tcl 8.6.1.

From-SVN: r274089

Daily bump.

From-SVN: r274054

Daily bump.

From-SVN: r274037

re PR fortran/90985 (Wrong error message with variables named "DATA*")

2019-08-02  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/90985
* decl.c (gfc_match_data): In free-form code, DATA be followed by
whitespace.

2019-08-02  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/90985
* gfortran.dg/pr90985.f90: New test.

From-SVN: r274033

Correct the PR number for 90297 to 90986.

From-SVN: r274032

re PR fortran/90297 (gcc/fortran/resolve.c: 2 * possibly redundant code ?)

2019-08-02  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/90297
* match.c (gfc_match_equivalence): Check that EQUIVALENCE is followed
by '('.

2019-08-02  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/90297
* gfortran.dg/equiv_10.f90: New test.

From-SVN: r274031

Add future.md.

2019-08-02  Michael Meissner  <meissner@linux.ibm.com>

* config/rs6000/future.md: New file.
* config/rs6000/rs6000.md: Include future.md.
* config/rs6000/t-rs6000 (MD_INCLUDES): Add future.md.

From-SVN: r274030

rework Ada EH Machine_Occurrence deallocation

Introduce exception handler ABI #1 to ensure single release, no access
after release of reraised Machine_Occurrences, and no failure to
re-reraise a Machine_Occurrence.

Unlike Ada exceptions, foreign exceptions do not get a new
Machine_Occurrence upon reraise, but each handler would delete the
exception upon completion, normal or exceptional, save for the case of
a 'raise;' statement within the handler, that avoided the delete by
clearing the exception pointer that the cleanup would use to release
it.  The cleared exception pointer might then be used by a subsequent
reraise within the same handler.  Get_Current_Excep.all would also
expose the Machine_Occurrence to reuse by Reraise_Occurrence, even for
native exceptions.

Under ABI #1, Begin_Handler_v1 claims responsibility for releasing an
exception by saving its cleanup and setting it to Claimed_Cleanup.
End_Handler_v1 restores the cleanup and runs it, as long as it isn't
still Claimed_Cleanup (which indicates an enclosing handler has
already claimed responsibility for releasing it), and as long as the
same exception is not being propagated up (the next handler of the
propagating exception will then claim responsibility for releasing
it), so reraise no longer needs to clear the exception pointer, and it
can just propagate the exception, just like Reraise_Occurrence.

ABI #1 is fully interoperable with ABI #0, i.e., exception handlers
that call the #0 primitives can be linked together with ones that call
the #1 primitives, and they will not misbehave.  When a #1 handler
claims responsibility for releasing an exception, even #0 reraises
dynamically nested within it will refrain from releasing it.  However,
when a #0 handler is a handler of a foreign exception that would have
been responsible for releasing it with #1, a Reraise_Occurrence of
that foreign or other Machine_Occurrence-carrying exception may still
cause the exception to be released multiple times, and to be used
after it is first released, even if other handlers of the foreign
exception use #1.

for  gcc/ada/ChangeLog

* libgnat/a-exexpr.adb (Begin_Handler_v1, End_Handler_v1): New.
(Claimed_Cleanup): New.
(Begin_Handler, End_Handler): Document.
* gcc-interface/trans.c (gigi): Switch to exception handler
ABI #1.
(Exception_Handler_to_gnu_gcc): Save the original cleanup
returned by begin handler, pass it to end handler, and use
EH_ELSE_EXPR to pass a propagating exception to end handler.
(gnat_to_gnu): Leave the exception pointer alone for reraise.
(add_cleanup): Handle EH_ELSE_EXPR, require it by itself.

From-SVN: r274029

re PR c++/56428 ([C++11] "is not a constant expression" when comparing non-type template argument to nullptr)

PR c++/56428
* g++.dg/cpp0x/nontype4.C: New test.

From-SVN: r274028

re PR c++/53009 (pointer to static member function of template class is “invalid” as a template argument of another template class)

PR c++/53009
* g++.dg/cpp0x/nontype3.C: New test.

From-SVN: r274027

function.c (assign_parm_adjust_stack_rtl): Revise STRICT_ALIGNMENT check to use targetm.slow_unaligned_access instead.

2019-08-02  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        * function.c (assign_parm_adjust_stack_rtl): Revise STRICT_ALIGNMENT
        check to use targetm.slow_unaligned_access instead.

From-SVN: r274025

re PR c++/77575 (Bogus error when alias template yielding a reference type used as template template argument)

PR c++/77575
* g++.dg/cpp0x/nontype2.C: New test.

From-SVN: r274024

function.c (assign_param_data_one): Remove unused data members.

2019-08-02  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        * function.c (assign_param_data_one): Remove unused data members.

From-SVN: r274023

simd_pcs_attribute.c: New test.

2019-08-02  Steve Ellcey  <sellcey@marvell.com>

* gcc.target/aarch64/simd_pcs_attribute.c: New test.
* gcc.target/aarch64/simd_pcs_attribute-2.c: Ditto.
* gcc.target/aarch64/simd_pcs_attribute-3.c: Ditto.

From-SVN: r274020

omp-simd-clone.c (simd_clone_adjust_return_type): Remove call to build_distinct_type_copy.

2019-08-02  Steve Ellcey  <sellcey@marvell.com>

* omp-simd-clone.c (simd_clone_adjust_return_type): Remove call to
build_distinct_type_copy.
(simd_clone_adjust_argument_types): Ditto.
(simd_clone_adjust): Call build_distinct_type_copy here.
(expand_simd_clones): Ditto.

From-SVN: r274019

re PR tree-optimization/91201 (SIMD not generated for horizontal sum of bytes in array)

PR target/91201
* config/i386/sse.md (*vec_extractv16qi_zext): New insn pattern.

testsuite/ChangeLog:

PR target/91201
* gcc.target/i386/sse4_1-pr91201.c: New test.

From-SVN: r274018

tree-ssa-loop-im: simplify casts in comparators

* tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Simplify casts
from 'const void *'.
(sort_locs_in_loop_postorder_cmp): Likewise.

From-SVN: r274011

PR c++/91230 - wrong error with __PRETTY_FUNCTION__ and generic lambda.

* pt.c (value_dependent_expression_p): Consider __PRETTY_FUNCTION__
inside a template function value-dependent.

* g++.dg/cpp1y/lambda-generic-pretty1.C: New test.

From-SVN: r274009

* gcc.dg/torture/pr91323.c: Also check comparison results.

From-SVN: r274008

invoke.texi (hot-bb-count-fraction): Rework description.

* doc/invoke.texi (hot-bb-count-fraction): Rework description.
(hot-bb-count-ws-permille): Likewise.
(hot-bb-frequency-fraction): Likewise.
(unlikely-bb-count-fraction): Likewise.
* params.def (hot-bb-count-fraction): Rework description.
(hot-bb-count-ws-permille): Likewise.
(hot-bb-frequency-fraction): Likewise.
(unlikely-bb-count-fraction): Likewise.  Remove min and max values.
* predict.c (get_hot_bb_threshold): Deal with 0 HOT_BB_COUNT_FRACTION.

From-SVN: r274006

re PR target/91323 (LTGT rtx produces UCOMISS instead of COMISS)

PR target/91323
* config/i386/i386-expand.c (ix86_unordered_fp_compare) <case LTGT>:
Return false.

testsuite/ChangeLog:

PR target/91323
* gcc.dg/torture/pr91323.c: New test.

From-SVN: r274005

vec.h (vec::sort): Add gcc_qsort_r support.

2019-08-02  Richard Biener  <rguenther@suse.de>

* vec.h (vec::sort): Add gcc_qsort_r support.
(vec::bsearch): Add an overload with gcc_qsort_r style callbacks.
* tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Adjust
to gcc_qsort_r style callback.
(sort_locs_in_loop_postorder_cmp): Likewise.
(analyze_memory_references): Use gcc_sort_r interfaces.
(find_ref_loc_in_loop_cmp): Use new bsearch overload.

From-SVN: r274004

Properly detect working jobserver in gcc driver.

2019-08-02  Martin Liska  <mliska@suse.cz>

PR lto/91313
* gcc.c (driver::maybe_run_linker): Call detect_jobserver
to detect working job server.
(driver::detect_jobserver): Test whether jobserver
is active from GCC driver. That will prevent situation where
GCC is invoked from a LD plugin and the linker already uses
file descriptors suggested by make.  That leads to a wrong
detection.
* gcc.h (driver): Add detect_jobserver.
* lto-wrapper.c (jobserver_active_p): Simplify sscanf by
not scanning for --jobserver-auth prefix.

From-SVN: r274003

tree.c (handle_nodiscard_attribute): Do not warn about nodiscard applied to a constructor.

/cp
2019-08-02  Paolo Carlini  <paolo.carlini@oracle.com>

* tree.c (handle_nodiscard_attribute): Do not warn about nodiscard
applied to a constructor.

/testsuite
2019-08-02  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/cpp1z/nodiscard6.C: New.

From-SVN: r274002

Fix gcc.dg/torture/ssa-fre-{5,7}.c failure for avr

The test assumes 32 bit integers (and pointers), and therefore
fails for the avr target, which has 16 bit integers.

Fix failure for avr by adding dg-require-effective-target int32.

gcc/testsuite/ChangeLog:

2019-08-02  Senthil Kumar Selvaraj  <senthilkumar.selvaraj@microchip.com>

* gcc.dg/torture/ssa-fre-5.c: Add dg-require-effective-target int32.
* gcc.dg/torture/ssa-fre-7.c: Likewise.

From-SVN: r274001