alpha.c (direct_return): Move down after struct machine_function definition...

* config/alpha/alpha.c (direct_return): Move down after
struct machine_function definition; use saved frame_size;
return bool.
(struct machine_function): Add sa_mask, sa_size, frame_size.
(alpha_sa_mask, alpha_sa_size, compute_frame_size): Merge into ...
(alpha_compute_frame_layout): ... new function.
(TARGET_COMPUTE_FRAME_LAYOUT): New.
(alpha_initial_elimination_offset): Use saved sa_size.
(alpha_vms_initial_elimination_offset): Likewise.
(alpha_vms_can_eliminate): Remove alpha_sa_size call.
(alpha_expand_prologue): Use saved frame data.  Merge integer
and fp register save loops.
(alpha_expand_epilogue): Likewise.
(alpha_start_function): Use saved frame data.
* config/alpha/alpha-protos.h (direct_return): Update.
(alpha_sa_size): Remove.

From-SVN: r271970

ChangeLog for PR c++/63149

From-SVN: r271969

PR c++/63149 - wrong auto deduction from braced-init-list

2019-06-04  Nina Dinka Ranns  <dinka.ranns@gmail.com>

    gcc/cp/
    * pt.c (listify_autos): Use non cv qualified auto_node in
    std::initializer_list<auto>.

    testsuite/
    * g++.dg/cpp0x/initlist-deduce2.C: New test.

From-SVN: r271968

aix-unwind.h (LR_REGNO): Rename to R_LR.

* config/rs6000/aix-unwind.h (LR_REGNO): Rename to R_LR.
(CR2_REGNO): Rename to R_CR2.
(XER_REGNO): Rename to R_XER.
(FIRST_ALTIVEC_REGNO): Rename to R_FIRST_ALTIVEC.
(VRSAVE_REGNO): Rename to R_VRSAVE.
(VSCR_REGNO): R_VSCR.

From-SVN: r271967

rs6000: Fix new testcase

At least with -m32 you need -maltivec if you #include <altivec.h>.

gcc/testsuite/
* g++.target/powerpc/undef-bool-3.C: Add -maltivec to dg-options.

From-SVN: r271966

c-pragma.c (handle_pragma_weak): Adjust quoting in a diagnostic.

gcc/c-family/ChangeLog:

* c-pragma.c (handle_pragma_weak): Adjust quoting in a diagnostic.
* c.opt (-Wformat-diag): Remove a spurious period.

gcc/testsuite/ChangeLog:

* gcc.dg/weak/weak-19.c: New test.

From-SVN: r271965

fold-const.c (extract_muldiv_1): Do not distribute a multiplication by a power-of-two value.

* fold-const.c (extract_muldiv_1) <PLUS_EXPR>: Do not distribute a
multiplication by a power-of-two value.
(fold_plusminus_mult_expr): Use pow2p_hwi to spot a power-of-two value
and turn the modulo operation into a masking operation.

From-SVN: r271963

[PATCH][GCC][AARCH64] Add tests for pointer authentication B-key

gcc/testsuite/ChangeLog
* gcc.target/aarch64/return_address_sign_b_1.c: New file.
* gcc.target/aarch64/return_address_sign_b_2.c: New file.
* gcc.target/aarch64/return_address_sign_b_3.c: New file.
* gcc.target/aarch64/return_address_sign_builtin.c: New file.
* g++.target/aarch64/return_address_sign_ab_exception.C: New file.
* g++.target/aarch64/return_address_sign_b_exception.C: New file.

From-SVN: r271954

re PR debug/90733 (ICE in simplify_subreg, at simplify-rtx.c:6440)

PR debug/90733
* var-tracking.c (vt_expand_loc_callback): Don't create raw subregs
with VOIDmode inner operands.

* gcc.dg/pr90733.c: New test.

From-SVN: r271952

re PR middle-end/90726 (exponential behavior on SCEV results everywhere)

2019-06-05  Richard Biener  <rguenther@suse.de>

PR middle-end/90726
* tree-ssa-loop-niter.c (expand_simple_operations): Do not
turn an expression graph into a tree.

* gcc.dg/pr90726.c: Enable IVOPTs.

From-SVN: r271950

omp-expand.c (struct omp_region): Add has_lastprivate_conditional member.

* omp-expand.c (struct omp_region): Add has_lastprivate_conditional
member.
(expand_parallel_call): If region->inner->has_lastprivate_conditional,
treat it like explicit monotonic schedule modifier.
(expand_omp_for): Initialize has_lastprivate_conditional.
If fd.lastprivate_conditional != 0, treat it like explicit monotonic
schedule modifier.

From-SVN: r271949

omp-low.c (lower_rec_input_clauses): For lastprivate conditional references...

* omp-low.c (lower_rec_input_clauses): For lastprivate conditional
references, lookup in in hash map MEM_REF operand instead of the
MEM_REF itself.
(lower_omp_1): When looking for lastprivate conditional assignments,
handle MEM_REFs with REFERENCE_TYPE operands.

* testsuite/libgomp.c++/lastprivate-conditional-1.C: New test.
* testsuite/libgomp.c++/lastprivate-conditional-2.C: New test.

From-SVN: r271948

omp-low.c (lower_rec_input_clauses): Force max_vf if is_simd and on privatization clauses OMP_CLAUSE_DECL is...

* omp-low.c (lower_rec_input_clauses): Force max_vf if is_simd and
on privatization clauses OMP_CLAUSE_DECL is privatized by reference
and references a VLA.  Handle references to non-VLAs if is_simd
all privatization clauses like reductions.
(lower_rec_input_clauses) <case do_private, case do_firstprivate>:
If omp_is_reference, use always omp simd arrays and set
DECL_VALUE_EXPR in that case, if lower_rec_simd_input_clauses
fails, emit reference initialization.

* g++.dg/vect/simd-1.cc: New test.

From-SVN: r271947

re PR target/89803 (Missing AVX512 intrinsics)

gcc/
2019-06-05  Hongtao Liu  <hongtao.liu@intel.com>

PR target/89803
* config/i386/avx512dqintrin.h (_mm_mask_fpclass_ss_mask,
_mm_mask_fpclass_sd_mask): New intrinsics.
(_mm_fpclass_ss_mask, _mm_fpclass_sd_mask): Modified, use new builtins.
* config/i386/i386-builtin.def
(__builtin_ia32_fpclassss_mask, __builtin_ia32_fpclasssd_mask):
New builtins.
(__builtin_ia32_fpclassss, __builtin_ia32_fpclasssd): Deleted.
* config/i386/i386-builtin-types.def (DEF_FUNCTION_TYPE (QI, V2DF, INT),
DEF_FUNCTION_TYPE (QI, V4SF, INT)): Deleted.
* config/i386/i386-expand.c (case QI_FTYPE_V4SF_INT,
case QI_FTYPE_V2SF_INT): Ditto.
* config/i386/sse.md
(define_insn "avx512dq_vmfpclass<mode><mask_scalar_merge_name>):
Extended to insnstructions with mask operands.

gcc/testsuite
2019-06-05  Hongtao Liu  <hongtao.liu@intel.com>

PR target/89803
* gcc.target/i386/avx-1.c (__builtin_ia32_fpclasssss,
__builtin_ia32_fpclasssd): Removed.
(__builtin_ia32_fpclassss_mask, __builtin_ia32_fpclasssd_mask): Define.
* gcc.target/i386/sse-13.c (__builtin_ia32_fpclasssss,
__builtin_ia32_fpclasssd): Removed.
(__builtin_ia32_fpclassss_mask, __builtin_ia32_fpclasssd_mask): Define.
* gcc.target/i386/sse-23.c (__builtin_ia32_fpclasssss,
__builtin_ia32_fpclasssd): Removed.
(__builtin_ia32_fpclassss_mask, __builtin_ia32_fpclasssd_mask): Define.
* gcc.target/i386/avx512dq-vfpclassss-2.c: New.
* gcc.target/i386/avx512dq-vfpclasssd-2.c: New.
* gcc.target/i386/avx512dq-vfpclassss-1.c (avx512f_test):
Add test for _mm_mask_fpclass_ss_mask.
* gcc.target/i386/avx512dq-vfpclasssd-1.c (avx512f_test):
Add test for _mm_mask_fpclass_sd_mask.

From-SVN: r271946

compiler: statically allocate constant interface data

    When converting a constant to interface, such as interface{}(42)
    or interface{}("hello"), if the interface escapes, we currently
    generate a heap allocation to hold the constant value.

    This CL changes it to generate a static allocation instead, as
    the gc compiler does. This reduces allocations in such cases.

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/180277

From-SVN: r271945

Daily bump.

From-SVN: r271944

rs6000: Update direct-move* testcases

This fixes some testcases that the last fifteen or so patches broke.
In all these cases we no longer need to set VSX_REG_ATTR: the default
value of "wa" is correct.

gcc/testsuite/
* gcc.target/powerpc/direct-move-double1.c (VSX_REG_ATTR): Delete.
* gcc.target/powerpc/direct-move-double2.c: Ditto.
* gcc.target/powerpc/direct-move-float1.c: Ditto.
* gcc.target/powerpc/direct-move-float2.c: Ditto.
* gcc.target/powerpc/direct-move-vint1.c: Ditto.
* gcc.target/powerpc/direct-move-vint2.c: Ditto.

From-SVN: r271940

rs6000: Remove wp and wq

wp becomes wa with isa p9tf, and wq is replaced by wa with isa p9kf.
To manage to do that, there is the new mode attribute VSisa.

* config/rs6000/constraints.md (define_register_constraint "wp"):
Delete.
(define_register_constraint "wq"): Delete.
* config/rs6000/rs6000.c (rs6000_debug_reg_global): Adjust.
(rs6000_init_hard_regno_mode_ok): Adjust.
* config/rs6000/rs6000.h (enum r6000_reg_class_enum): Delete
RS6000_CONSTRAINT_wp and RS6000_CONSTRAINT_wq.
* config/rs6000/vsx.md (define_mode_attr VSr3): Delete.
(define_mode_attr VSa): Delete.
(define_mode_attr VSisa): New.
(rest of file): Adjust.
* doc/md.texi (Machine Constraints): Adjust.

From-SVN: r271939

rs6000: Add p9kf and p9tf isa values

This adds "p9kf" and "p9tf" isa values, to be used for instruction
alternatives where KFmode resp. TFmode is used.

* config/rs6000/rs6000.md (define_attr "isa"): Add p9kf and p9tf.
(define_attr "enabled"): Handle those new isa values.

From-SVN: r271938

rs6000: More simplification

A whole bunch of mode attributes are used only once.  Things are
easier to read if we just expand those patterns.  It's shorter, too.

* config/rs6000/vsx.md (define_mode_attr VSr4): Delete.
(define_mode_attr VSr5): Delete.
(define_mode_attr VStype_sqrt): Delete.
(define_mode_iterator VSX_SPDP): Delete.
(define_mode_attr VS_spdp_res): Delete.
(define_mode_attr VS_spdp_insn): Delete.
(define_mode_attr VS_spdp_type): Delete.
(*vsx_sqrt<mode>2): Adjust.
(vsx_<VS_spdp_insn>): Delete, split to...
(vsx_xscvdpsp): ... this.  New.  And...
(vsx_xvcvspdp): ... this.  New.  And...
(vsx_xvcvdpsp): ... this.  New.

From-SVN: r271937

rs6000: <VSs> -> <sd>p

We don't need the <VSs> mode attribute, if we make <sd> work for V4SF
and V2DF just like for SF and DF.

* config/rs6000/rs6000.md (define_mode_attr sd): Add values for V4SF
and V2DF.
* config/rs6000/vsx.md (define_mode_attr VSs): Delete.
(rest of file): Adjust.

From-SVN: r271936

rs6000: ww->wa in testsuite

I should have factored this series better.  Oh well.  Near the end,
let's call it loose ends.

gcc/testsuite/
* gcc.target/powerpc/direct-move-float1.c: Use "wa" instead of "ww"
constraint.

From-SVN: r271935

rs6000: VSa->wa for some more cases

* config/rs6000/vsx.md (vsx_<VS_spdp_insn>): Use wa instead of <VSa>.
(vsx_extract_<mode>_var): Ditto.

From-SVN: r271934

rs6000: Simplify <VSa> for VSX_TI

When used in VSX_TI, <VSa> is always just "wa".

* config/rs6000/vsx.md: Replace all <VSa> that are used with VSX_TI
with just "wa".

From-SVN: r271933

rs6000: ww -> wa

"ww" can always be "wa".

* config/rs6000/constraints.md (define_register_constraint "ww"):
Delete.
* config/rs6000/rs6000.c (rs6000_debug_reg_global): Adjust.
(rs6000_init_hard_regno_mode_ok): Adjust.
* config/rs6000/rs6000.h (enum r6000_reg_class_enum): Delete
RS6000_CONSTRAINT_ww.
* config/rs6000/rs6000.md: Adjust.
* config/rs6000/vsx.md: Adjust.
* doc/md.texi (Machine Constraints): Adjust.

From-SVN: r271932

rs6000: Remove Ftrad, Fvsx, Fs; add s and sd

This removes the <Ftrad>, <Fvsx>, and <Fs> mode attributes, and creates
new <sd> and <s> mode attributes instead.  <sd> is either "s" or "d",
depending on whether the mode is single-precision or double-precision
floating point; and <s> is either "s" or nothing.

* config/rs6000/rs6000.md (SFDF, SFDF2): Adjust comments.
(define_mode_attr sd): New.
(define_mode_attr s): New.
(define_mode_attr Ftrad): Delete.
(define_mode_attr Fvsx): Delete.
(define_mode_attr Fs): Delete.
(rest of file): Use the new mode attributes.
* config.rs6000/vsx.md: Use the new mode attributes.

From-SVN: r271931

rs6000: Simplify <VSa> for VSX_W

When used in VSX_W, <VSa> is always just "wa".

* config/rs6000/vsx.md: Replace all <VSa> that are used with VSX_W
with just "wa".

From-SVN: r271930

rs6000: Simplify VS[ra]* for VSX_[BDF]

When used in VSX_B, VSX_D, or VSX_F, both <VSr> and <VSa> are always
just "wa" now.  Similarly <VSr2> and <VSr3>.  The former of those is
always "wa", so we can remove the mode attribute completely.

* config/rs6000/vsx.md (define_mode_attr VSr2): Delete.
(rest of file): Replace all <VSa>, <VSr>, <VSr2>, and <VSr3> that are
used with VSX_B, VSX_D, or VSX_F, with just "wa".

From-SVN: r271929

decl.c (grokdeclarator): Use declarator->id_loc in two additional places.

/cp
2019-06-04  Paolo Carlini  <paolo.carlini@oracle.com>

* decl.c (grokdeclarator): Use declarator->id_loc in two
additional places.

/testsuite
2019-06-04  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/concepts/pr60573.C: Test locations too.
* g++.dg/cpp0x/deleted13.C: Likewise.
* g++.dg/parse/error29.C: Likewise.
* g++.dg/parse/qualified4.C: Likewise.
* g++.dg/template/crash96.C Likewise.
* g++.old-deja/g++.brendan/crash22.C Likewise.
* g++.old-deja/g++.brendan/crash23.C Likewise.
* g++.old-deja/g++.law/visibility10.C Likewise.
* g++.old-deja/g++.other/decl5.C: Likewise.

From-SVN: r271928

re PR target/78263 (Compile failure with AltiVec library on PPC64le and -std=c++11 flag)

[gcc]

2019-06-04  Bill Schmidt  <wschmidt@linux.ibm.com>

PR target/78263
* config/rs6000/altivec.h: Don't #define vector, pixel, bool for
C++ with strict ANSI requirements.

[gcc/testsuite]

2019-06-04  Bill Schmidt  <wschmidt@linux.ibm.com>

PR target/78263
* g++.target/powerpc: New directory.
* g++.target/powerpc/powerpc.exp: New test driver.
* g++.target/powerpc/undef-bool-3.C: New.

From-SVN: r271927

Simplify loop size when step=1

2019-06-04  Marc Glisse  <marc.glisse@inria.fr>

* tree-ssa-loop-niter.c (number_of_iterations_ne): Skip
computations when step is 1.

From-SVN: r271926

rs6000: wf -> wa

"wf" is just "wa".

* config/rs6000/constraints.md (define_register_constraint "wf"):
Delete.
* config/rs6000/rs6000.c (rs6000_debug_reg_global): Adjust.
(rs6000_init_hard_regno_mode_ok): Adjust.
* config/rs6000/rs6000.h (enum r6000_reg_class_enum): Delete
RS6000_CONSTRAINT_wf.
* config/rs6000/rs6000.md: Adjust.
* config/rs6000/vsx.md: Adjust.
* doc/md.texi (Machine Constraints): Adjust.

From-SVN: r271921

AARCH64: ILP32: Fix aarch64_asan_shadow_offset

aarch64_asan_shadow_offset is using the wrong
offset for ILP32.  Change it to be a decent one.

ChangeLog:
* config/aarch64/aarch64.c (aarch64_asan_shadow_offset):
Fix ILP32 value.

From-SVN: r271920

rs6000: wd -> wa

"wd" is just "wa".

* config/rs6000/constraints.md (define_register_constraint "wd"):
Delete.
* config/rs6000/rs6000.c (rs6000_debug_reg_global): Adjust.
(rs6000_init_hard_regno_mode_ok): Adjust.
* config/rs6000/rs6000.h (enum r6000_reg_class_enum): Delete
RS6000_CONSTRAINT_wd.
* config/rs6000/rs6000.md: Adjust.
* config/rs6000/vsx.md: Adjust.
* doc/md.texi (Machine Constraints): Adjust.

From-SVN: r271919

rs6000: Delete Fv2

<Fv2> always is "wa".

* config/rs6000/rs6000.md (define_mode_attr Fv2): Delete.
(rest of file): Adjust.

From-SVN: r271918

rs6000: Delete VS_64reg

<VS_64reg> now always is "wa".  Make that simplification.

* config/rs6000/vsx.md (define_mode_attr VS_64reg): Delete.
(*vsx_extract_<P:mode>_<VSX_D:mode>_load): Adjust.
(vsx_splat_<mode>_reg): Adjust.

From-SVN: r271917

rs6000: ws -> wa

"ws" is just "wa".

* config/rs6000/constraints.md (define_register_constraint "ws"):
Delete.
* config/rs6000/rs6000.c (rs6000_debug_reg_global): Adjust.
(rs6000_init_hard_regno_mode_ok): Adjust.
* config/rs6000/rs6000.h (enum r6000_reg_class_enum): Delete
RS6000_CONSTRAINT_ws.
* config/rs6000/rs6000.md: Adjust.
* config/rs6000/vsx.md: Adjust.
* doc/md.texi (Machine Constraints): Adjust.

From-SVN: r271916

rs6000: wv -> v+p7v

"wv" is "v", but only if VSX is enabled (otherwise it's NO_REGS).  So
this patch sets "isa" "p7v" to all alternatives that used "wv" before
(and that do not already need a later ISA), and changes the constraint.

* config/rs6000/constraints.md (define_register_constraint "wv"):
Delete.
* config/rs6000/rs6000.c (rs6000_debug_reg_global): Adjust.
(rs6000_init_hard_regno_mode_ok): Adjust.
* config/rs6000/rs6000.h (enum r6000_reg_class_enum): Delete
RS6000_CONSTRAINT_wv.
* config/rs6000/rs6000.md: Adjust.
* config/rs6000/vsx.md: Adjust.
* doc/md.texi (Machine Constraints): Adjust.

From-SVN: r271915

rs6000: wi->wa, wt->wa

"wi" and "wt" mean just the same as "wa" these days.  Change them to
the simpler name.

* config/rs6000/constraints.md (define_register_constraint "wi"):
Delete.
(define_register_constraint "wt"): Delete.
* config/rs6000/rs6000.c (rs6000_debug_reg_global): Adjust.
(rs6000_init_hard_regno_mode_ok): Adjust.
* config/rs6000/rs6000.h (enum r6000_reg_class_enum): Delete
RS6000_CONSTRAINT_wi and RS6000_CONSTRAINT_wt.
* config/rs6000/rs6000.md: Adjust.
* config/rs6000/vsx.md: Adjust.
* doc/md.texi (Machine Constraints): Adjust.

From-SVN: r271914

aarch64: fix asm visibility for extern symbols

Commit r271869 broke visibility declarations in asm for extern symbols, because
the new ASM_OUTPUT_EXTERNAL hook failed to call the default hook for elf.

gcc/ChangeLog:

* config/aarch64/aarch64-protos.h (aarch64_asm_output_external): Remove
const.
* config/aarch64/aarch64.c (aarch64_asm_output_external): Call
default_elf_asm_output_external.

From-SVN: r271913

[C++ PATCH] structure tag lookup

https://gcc.gnu.org/ml/gcc-patches/2019-06/msg00179.html
* name-lookup.c (lookup_type_scope_1): Reimplement, handle local
and namespace scopes separately.

From-SVN: r271912

PR c++/60531 - Wrong error about unresolved overloaded function

For PR60531, GCC wrongly rejects function templates with explicitly
specified template arguments as overloaded. They are resolved by
resolve_nondeduced_context, which is normally called by
cp_default_conversion through decay_conversion, but the latter have
extra effects making them unusable here. Calling the former directly
does work.

* typeck.c (cp_build_binary_op): See if overload can be resolved.
(cp_build_unary_op): Ditto.

* g++.dg/template/operator15.C: New test.

From-SVN: r271910

Reduce accumulated garbage in constexpr evaluation.

We want to evaluate the arguments to a call before looking into the cache so
that we have constant values, but if we then find the call in the cache we
end up with a TREE_LIST that we don't end up using; in highly recursive
constexpr evaluation this ends up being a large proportion of the garbage
generated.

The cxx_eval_increment_expression hunk is less important, but it's an easy
tweak; we only use the MODIFY_EXPR to evaluate it, so after that it's
garbage.

* constexpr.c (cxx_eval_call_expression): ggc_free any bindings we
don't save.
(cxx_eval_increment_expression): ggc_free the MODIFY_EXPR after
evaluating it.

From-SVN: r271909

Remove dead code in IPA ICF.

2019-06-04  Martin Liska  <mliska@suse.cz>

* ipa-icf.c (INCLUDE_LIST): Remove.
(sem_item_optimizer::execute): Remove call to init_wpa.
* ipa-icf.h (init_wpa): Remove.

From-SVN: r271908

gimplify.c (gimplify_scan_omp_clauses): Don't sorry_at on lastprivate conditional on combined for simd.

* gimplify.c (gimplify_scan_omp_clauses): Don't sorry_at on lastprivate
conditional on combined for simd.
* omp-low.c (struct omp_context): Add combined_into_simd_safelen0
member.
(lower_rec_input_clauses): For gimple_omp_for_combined_into_p max_vf 1
constructs, don't remove lastprivate_conditional_map, but instead set
ctx->combined_into_simd_safelen0 and adjust hash_map, so that it points
to parent construct temporaries.
(lower_lastprivate_clauses): Handle ctx->combined_into_simd_safelen0
like !ctx->lastprivate_conditional_map.
(lower_omp_1) <case GIMPLE_ASSIGN>: If up->combined_into_simd_safelen0,
use up->outer context instead of up.
* omp-expand.c (expand_omp_for_generic): Perform cond_var bump even if
gimple_omp_for_combined_p.
(expand_omp_for_static_nochunk): Likewise.
(expand_omp_for_static_chunk): Add forgotten cond_var bump that was
probably moved over into expand_omp_for_generic rather than being copied
there.
gcc/cp/
* cp-tree.h (CP_OMP_CLAUSE_INFO): Allow for any clauses up to _condvar_
instead of only up to linear.
gcc/testsuite/
* c-c++-common/gomp/lastprivate-conditional-2.c (foo): Don't expect
a sorry_at on any of the clauses.
libgomp/
* testsuite/libgomp.c-c++-common/lastprivate-conditional-7.c: New test.
* testsuite/libgomp.c-c++-common/lastprivate-conditional-8.c: New test.
* testsuite/libgomp.c-c++-common/lastprivate-conditional-9.c: New test.
* testsuite/libgomp.c-c++-common/lastprivate-conditional-10.c: New test.

From-SVN: r271907

Fix typo in tests.

2019-06-04  Martin Liska  <mliska@suse.cz>

* value-prof.c (dump_histogram_value): Fix typo.
(gimple_mod_subtract_transform): Likewise.

From-SVN: r271904

re PR middle-end/90726 (exponential behavior on SCEV results everywhere)

2019-06-04  Richard Biener  <rguenther@suse.de>

PR middle-end/90726
* tree-chrec.c (chrec_contains_symbols): Add to visited.
(tree_contains_chrecs): Likewise.
(chrec_contains_symbols_defined_in_loop): Move here and avoid
exponential behaivor from ...
* tree-scalar-evolution.c (chrec_contains_symbols_defined_in_loop):
... here.
(expression_expensive_p): Avoid exponential behavior and compute
expanded size, rejecting any expansion.
* tree-ssa-loop-ivopts.c (abnormal_ssa_name_p): Remove.
(idx_contains_abnormal_ssa_name_p): Likewise.
(contains_abnormal_ssa_name_p_1): New helper for walk_tree.
(contains_abnormal_ssa_name_p): Simplify and use
walk_tree_without_duplicates.

* gcc.dg/pr90726.c: New testcase.

From-SVN: r271903

re PR fortran/90738 (gfortran.dg/pointer_array_10.f90 etc. FAIL)

2019-06-04  Richard Biener  <rguenther@suse.de>

PR tree-optimization/90738
Revert
2019-06-03  Richard Biener  <rguenther@suse.de>

* tree-ssa-sccvn.c (ao_ref_init_from_vn_reference): Get original
full reference tree and record in ref->ref.
(vn_reference_lookup_3): Pass in original ref to
ao_ref_init_from_vn_reference.
(vn_reference_lookup): Likewise.
* tree-ssa-sccvn.h (ao_ref_init_from_vn_reference): Adjust prototype.
* tree-ssa-alias.c (nonoverlapping_component_refs_of_decl_p):
Handle non-decl bases in the original reference.

* gcc.dg/tree-ssa/alias-access-path-1.c: Scan fre1.

* gcc.dg/torture/pr90738.c: New testcase.

From-SVN: r271902

IPA ICF: use fibonacci heap instead of list as a worklist.

2019-06-04  Martin Liska  <mliska@suse.cz>

* ipa-icf.c (sem_item_optimizer::add_item_to_class): Count
number of references.
(sem_item_optimizer::do_congruence_step):
(sem_item_optimizer::worklist_push): Dump how references
a class has.
(sem_item_optimizer::worklist_pop): Use heap.
(sem_item_optimizer::process_cong_reduction): Likewise.
* ipa-icf.h: Use fibonacci_heap insteam of std::list.

From-SVN: r271901

IPA ICF: rewrite references into a hash_map.

2019-06-04  Martin Liska  <mliska@suse.cz>

* ipa-icf.h (struct sem_usage_pair_hash): New.
(sem_usage_pair_hash::hash): Likewise.
(sem_usage_pair_hash::equal): Likewise.
(struct sem_usage_hash): Likewise.
* ipa-icf.c (sem_item::sem_item): Initialize
referenced_by_count.
(sem_item::add_reference): Register a reference
in ref_map and not in target->usages.
(sem_item::setup): Remove initialization of
dead vectors.
(sem_item::~sem_item): Remove usage of dead vectors.
(sem_item::dump): Remove dump of references.
(sem_item_optimizer::sem_item_optimizer): Initialize
m_references.
(sem_item_optimizer::read_section): Remove useless
dump.
(sem_item_optimizer::parse_funcs_and_vars): Likewise here.
(sem_item_optimizer::build_graph): Pass m_references
to ::add_reference.
(sem_item_optimizer::verify_classes): Remove usage of dead
vectors.
(sem_item_optimizer::traverse_congruence_split): Return true
when a class is split.
(sem_item_optimizer::do_congruence_step_for_index): Use
hash_map for look up of (sem_item *, index). That brings
significant speed up.
(sem_item_optimizer::do_congruence_step): Return true
when a split is done.
(congruence_class::is_class_used): Use referenced_by_count.
2019-06-04  Martin Liska  <mliska@suse.cz>

* c-c++-common/goacc/acc-icf.c: Change scanned pattern.
* gfortran.dg/goacc/pr78027.f90: Likewise.

From-SVN: r271900

Daily bump.

From-SVN: r271899

PR90689, ICE in extract_insn on ppc64le

PR target/90689
* config/rs6000/rs6000.c (rs6000_call_aix): Correct r271753 merge
error.

From-SVN: r271895

compiler, runtime, reflect: generate unique type descriptors

    Currently, the compiler already generates common symbols for type
    descriptors, so the type descriptors are unique. However, when a
    type is created through reflection, it is not deduplicated with
    compiler-generated types. As a consequence, we cannot assume type
    descriptors are unique, and cannot use pointer equality to
    compare them. Also, when constructing a reflect.Type, it has to
    go through a canonicalization map, which introduces overhead to
    reflect.TypeOf, and lock contentions in concurrent programs.

    In order for the reflect package to deduplicate types with
    compiler-created types, we register all the compiler-created type
    descriptors at startup time. The reflect package, when it needs
    to create a type, looks up the registry of compiler-created types
    before creates a new one. There is no lock contention since the
    registry is read-only after initialization.

    This lets us get rid of the canonicalization map, and also makes
    it possible to compare type descriptors with pointer equality.

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/179598

From-SVN: r271894

parser.c (cp_parser_unqualified_id): Use build_min_nt_loc in five places.

2019-06-03  Paolo Carlini  <paolo.carlini@oracle.com>

* parser.c (cp_parser_unqualified_id): Use build_min_nt_loc in
       five places.

From-SVN: r271893

libgo: delay applying profile stack-frame skip until fixup

    When the runtime collects a stack trace to associate it with some
    profiling event (mem alloc, mutex, etc) there is a skip count passed
    to runtime.Callers (or equivalent) to skip some known count of frames
    in order to get to the "interesting" frame corresponding to the
    profile event. Now that the profiling mechanism uses lazy fixup (when
    removing compiler artifacts like thunks, morestack calls etc), we also
    need to move the frame skipping logic after the fixup, so as to insure
    that the skip count isn't thrown off by these artifacts.

    Fixes golang/go#32290.

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/179740

From-SVN: r271892

compiler: permit inlining references to global variables

    This requires tracking all references to unexported variables, so that
    we can make them global symbols in the object file, and can export
    them so that other compilations can see the right definition for their
    own inline bodies.

    This introduces a syntax for referencing names defined in other
    packages: a <pNN> prefix, where NN is the package index.  This will
    need to be added to gccgoimporter, but I didn't do it yet since it
    isn't yet possible to create an object for which gccgoimporter will
    see a <pNN> prefix.

    This increases the number of inlinable functions in the standard
    library from 181 to 215, adding functions like context.Background.

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/177920

From-SVN: r271891

runtime: remove unnecessary functions calling between C and Go

    These functions were needed during the transition of the runtime from
    C to Go, but are no longer necessary.

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/179879

From-SVN: r271890

rs6000: Delete -mmfpgpr

This patch makes the -mmfpgpr option not do anything except warn that
the option is deprecated.

* config/rs6000/rs6000.h (MASK_MFPGPR): Delete.
* config/rs6000/rs6000.c (direct_move_p): Adjust.
(rs6000_secondary_reload_simple_move): Adjust.
(rs6000_opt_masks): Neuter the "mfpgpr" option.
* config/rs6000/rs6000-c.c (rs6000_target_modify_macros): Adjust.
* config/rs6000/rs6000-cpus.def (ISA_2_5_MASKS_EMBEDDED): Adjust
comment.
(power6x): Adjust.
* config/rs6000/rs6000.md (floatsi<mode>2_lfiwax): Adjust.
(floatunssi<mode>2_lfiwzx): Adjust.
(fix_trunc<mode>si2_stfiwx): Adjust.
(fixuns_trunc<mode>si2_stfiwx): Adjust.
* config/rs6000/rs6000.opt (mno-mfpgpr): New.
(mfpgpr): Mark as deprecated.
* doc/extend.texi (PowerPC Function Attributes): Delete mfpgpr.
(Basic PowerPC Built-in Functions Available on ISA 2.05): Adjust.
* doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mmfpgpr.

gcc/testsuite/
* gcc.target/powerpc/mmfpgpr.c: Delete.

From-SVN: r271889

rs6000: Delete wg

The "wg" constraint is used for the floating point side on mfpgpr
instructions.  Those instructions do not exist on any relevant
hardware.  This patch deletes the constraint and the insns using it.

* config/rs6000/constraints.md (define_register_constraint "wg"):
Delete.
* config/rs6000/rs6000.h (enum r6000_reg_class_enum): Delete
RS6000_CONSTRAINT_wg.
* config/rs6000/rs6000.c (rs6000_debug_reg_global): Adjust.
(rs6000_init_hard_regno_mode_ok): Adjust.
* config/rs6000/rs6000.md (*mov<mode>_softfloat32, *movdi_internal64):
Delete "wg" alternatives.
* doc/md.texi (Machine Constraints): Adjust.

From-SVN: r271888

* sv.po: Update.

From-SVN: r271885

Fix uses of static_assert not guarded by C++11 check

* include/bits/stl_map.h (map): Disable static assert for C++98 mode.
* include/bits/stl_multimap.h (multimap): Likewise.

From-SVN: r271884

runtime: fix assembly syntax

    Some assembler doesn't accept ULL suffix. In fact the suffix
    is not really necessary. Drop it.

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/180217

From-SVN: r271883

bb-reorder.c (copy_bb_p): Don't overflow size calculation.

* bb-reorder.c (copy_bb_p): Don't overflow size calculation.
(get_uncond_jump_length): Assert length less than INT_MAX and
non-negative.

From-SVN: r271877

Rename variables and cleanup comments.

2019-06-03  François Dumont  <fdumont@gcc.gnu.org>

Rename variables and cleanup comments.
* include/bits/hashtable_policy.h
* include/bits/hashtable.h

From-SVN: r271876

enum-1.c: Add -fno-eliminate-unused-debug-symbols on AIX.

        * gcc.dg/debug/enum-1.c: Add -fno-eliminate-unused-debug-symbols
        on AIX.
        * g++.dg/debug/enum-1.C: Same.

From-SVN: r271873

Fix PR64242 - Longjmp expansion incorrect

Improve the fix for PR64242.  Various optimizations can change a memory
reference into a frame access.  Given there are multiple virtual frame pointers
which may be replaced by multiple hard frame pointers, there are no checks for
writes to the various frame pointers.  So updates to a frame pointer tends to
generate incorrect code.  Improve the previous fix to also add clobbers of
several frame pointers and add a scheduling barrier.  This should work in most
cases until GCC supports a generic "don't optimize across this instruction"
feature.

Bootstrap OK. Testcase passes on AArch64 and x86-64.  Inspected x86, Arm,
Thumb-1 and Thumb-2 assembler which looks correct.

    gcc/
PR middle-end/64242
* builtins.c (expand_builtin_longjmp): Add frame clobbers and schedule
block.
(expand_builtin_nonlocal_goto): Likewise.

    testsuite/
PR middle-end/64242
* gcc.c-torture/execute/pr64242.c: Update test.

From-SVN: r271870

aarch64: emit .variant_pcs for aarch64_vector_pcs symbol references

A dynamic linker with lazy binding support may need to handle vector PCS
function symbols specially, so an ELF symbol table marking was
introduced for such symbols.

Function symbol references and definitions that follow the vector PCS
are marked in the generated assembly with .variant_pcs and then the
STO_AARCH64_VARIANT_PCS st_other flag is set on the symbol in the object
file.  The marking is propagated to the dynamic symbol table by the
static linker so a dynamic linker can handle such symbols specially.

For this to work, the assembler, the static linker and the dynamic
linker has to be updated on a system.  Old assembler does not support
the new .variant_pcs directive, so a toolchain with old binutils won't
be able to compile code that references vector PCS symbols.

gcc/ChangeLog:

* config/aarch64/aarch64-protos.h (aarch64_asm_output_alias): Declare.
(aarch64_asm_output_external): Declare.
* config/aarch64/aarch64.c (aarch64_asm_output_variant_pcs): New.
(aarch64_declare_function_name): Call aarch64_asm_output_variant_pcs.
(aarch64_asm_output_alias): New.
(aarch64_asm_output_external): New.
* config/aarch64/aarch64.h (ASM_OUTPUT_DEF_FROM_DECLS): Define.
(ASM_OUTPUT_EXTERNAL): Define.

gcc/testsuite/ChangeLog:

* gcc.target/aarch64/pcs_attribute-2.c: New test.
* gcc.target/aarch64/torture/simd-abi-4.c: Check .variant_pcs support.
* lib/target-supports.exp (check_effective_target_aarch64_variant_pcs):
New.

From-SVN: r271869

Corrections for C++2a library status table

* doc/xml/manual/status_cxx2020.xml: Add missing row for P0920R2.
Fix bgcolor for P0340R3.
* doc/html/*: Regenerate.

From-SVN: r271868

PR libstdc++/90686 update C++2a library status docs

PR libstdc++/90686
* doc/xml/manual/status_cxx2014.xml: Document what's missing from
<experimental/memory_resource>.
* doc/xml/manual/status_cxx2020.xml: Document status of P1285R0,
P0339R6, P0340R3, P1164R1 and P1357R1.
* doc/html/*: Regenerate.

From-SVN: r271867

Enforce allocator::value_type consistency for containers in C++2a

In previous standards it is undefined for a container and its allocator
to have a different value_type. Libstdc++ has traditionally allowed it
as an extension, automatically rebinding the allocator to the
container's value_type. Since GCC 8.1 that extension has been disabled
for C++11 and later when __STRICT_ANSI__ is defined (i.e. for
-std=c++11, -std=c++14, -std=c++17 and -std=c++2a).

Since the acceptance of P1463R1 into the C++2a draft an incorrect
allocator::value_type now requires a diagnostic. This patch implements
that by enabling the static_assert for -std=gnu++2a as well.

* doc/xml/manual/status_cxx2020.xml: Document P1463R1 status.
* include/bits/forward_list.h [__cplusplus > 201703]: Enable
allocator::value_type assertion for C++2a.
* include/bits/hashtable.h: Likewise.
* include/bits/stl_deque.h: Likewise.
* include/bits/stl_list.h: Likewise.
* include/bits/stl_map.h: Likewise.
* include/bits/stl_multimap.h: Likewise.
* include/bits/stl_multiset.h: Likewise.
* include/bits/stl_set.h: Likewise.
* include/bits/stl_vector.h: Likewise.
* testsuite/23_containers/deque/48101-3_neg.cc: New test.
* testsuite/23_containers/forward_list/48101-3_neg.cc: New test.
* testsuite/23_containers/list/48101-3_neg.cc: New test.
* testsuite/23_containers/map/48101-3_neg.cc: New test.
* testsuite/23_containers/multimap/48101-3_neg.cc: New test.
* testsuite/23_containers/multiset/48101-3_neg.cc: New test.
* testsuite/23_containers/set/48101-3_neg.cc: New test.
* testsuite/23_containers/unordered_map/48101-3_neg.cc: New test.
* testsuite/23_containers/unordered_multimap/48101-3_neg.cc: New test.
* testsuite/23_containers/unordered_multiset/48101-3_neg.cc: New test.
* testsuite/23_containers/unordered_set/48101-3_neg.cc: New test.
* testsuite/23_containers/vector/48101-3_neg.cc: New test.

From-SVN: r271866

tree-vrp.h (value_range_base::nonzero_p): New.

* tree-vrp.h (value_range_base::nonzero_p): New.
(value_range_base::set_nonnull): Rename to...
(value_range_base::set_nonzero): ...this.
(value_range_base::set_null): Rename to...
(value_range_base::set_zero): ...this.
(value_range::set_nonnull): Remove.
(value_range::set_null): Remove.
* tree-vrp.c (range_is_null): Remove.
(range_is_nonnull): Remove.
(extract_range_from_binary_expr): Use value_range_base::*zero_p
instead of range_is_*null.
(extract_range_from_unary_expr): Same.
(value_range_base::set_nonnull): Rename to...
(value_range_base::set_nonzero): ...this.
(value_range::set_nonnull): Remove.
(value_range_base::set_null): Rename to...
(value_range_base::set_zero): ...this.
(value_range::set_null): Remove.
(extract_range_from_binary_expr): Rename set_*null uses to
set_*zero.
(extract_range_from_unary_expr): Same.
(union_helper): Same.
* vr-values.c (get_value_range): Use set_*zero instead of
set_*null.
(vr_values::extract_range_from_binary_expr): Same.
(vr_values::extract_range_basic): Same.

From-SVN: r271865

Fix alignment option parser (PR90684)

Fix the alignment option parser to always allow up to 4 alignments.
Now -falign-functions=16:8:8:8 no longer reports an error.

    gcc/
PR driver/90684
* opts.c (parse_and_check_align_values): Allow 4 alignment values.
M    gcc/ChangeLog
M    gcc/opts.c

From-SVN: r271864

[AArch64] Emit TARGET_DOTPROD-specific sequence for <us>sadv16qi

Wilco pointed out that when the Dot Product instructions are available we can use them
to generate an even more efficient expansion for the [us]sadv16qi optab.
Instead of the current:
        uabdl2  v0.8h, v1.16b, v2.16b
        uabal   v0.8h, v1.8b, v2.8b
        uadalp  v3.4s, v0.8h

we can generate:
      (1)  mov    v4.16b, 1
      (2)  uabd    v0.16b, v1.16b, v2.16b
      (3)  udot    v3.4s, v0.16b, v4.16b

Instruction (1) can be CSEd across multiple such expansions and even hoisted outside of loops,
so when this sequence appears frequently back-to-back (like in x264_r) we essentially only have 2 instructions
per sum. Also, the UDOT instruction does the byte-to-word accumulation in one step, which allows us to use
the much simpler UABD instruction before it.

This makes it a shorter and lower-latency sequence overall for targets that support it.

* config/aarch64/iterators.md (MAX_OPP): New code attr.
* config/aarch64/aarch64-simd.md (*aarch64_<su>abd<mode>_3): Rename to...
(aarch64_<su>abd<mode>_3): ... This.
(<sur>sadv16qi): Add TARGET_DOTPROD expansion.

* gcc.target/aarch64/ssadv16qi.c: Add +nodotprod to pragma.
* gcc.target/aarch64/usadv16qi.c: Likewise.
* gcc.target/aarch64/ssadv16qi-dotprod.c: New test.
* gcc.target/aarch64/usadv16qi-dotprod.c: Likewise.

From-SVN: r271863

target-supports.exp (add_options_for_aarch64_sve): New procedure.

2019-06-03  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>

* lib/target-supports.exp (add_options_for_aarch64_sve): New procedure.
(aarch64_sve_hw_bits): Call add_options_for_aarch64_sve.
(check_effective_target_aarch64_sve_hw): Likewise.

From-SVN: r271862

Remove Java Trees from GENERIC manual.

2019-06-03  Martin Liska  <mliska@suse.cz>

* doc/generic.texi: Remove Java Trees.

From-SVN: r271861

tree-ssa-sccvn.c (ao_ref_init_from_vn_reference): Get original full reference tree and record in ref->ref.

2019-06-03  Richard Biener  <rguenther@suse.de>

* tree-ssa-sccvn.c (ao_ref_init_from_vn_reference): Get original
full reference tree and record in ref->ref.
(vn_reference_lookup_3): Pass in original ref to
ao_ref_init_from_vn_reference.
(vn_reference_lookup): Likewise.
* tree-ssa-sccvn.h (ao_ref_init_from_vn_reference): Adjust prototype.
* tree-ssa-alias.c (nonoverlapping_component_refs_of_decl_p):
Handle non-decl bases in the original reference.

* gcc.dg/tree-ssa/alias-access-path-1.c: Scan fre1.

From-SVN: r271860

Fix typo in index comparison of CONSTRUCTOR.

2019-06-03  Martin Liska  <mliska@suse.cz>

* fold-const.c (operand_equal_p): Fix typo as compare_tree_int
returns 0 when operands are equal.

From-SVN: r271859

re PR tree-optimization/90716 (gcc generates wrong debug information at -O2)

2019-06-03  Richard Biener  <rguenther@suse.de>

PR tree-optimization/90716
* tree-loop-distribution.c (destroy_loop): Process blocks in
correct order.

* gcc.dg/guality/pr90716.c: New testcase.

From-SVN: r271858

re PR target/88837 ([SVE] Poor vector construction code in VL-specific mode)

2019-06-03  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>

PR target/88837
* vector-builder.h (vector_builder::count_dups): New method.
* config/aarch64/aarch64-protos.h (aarch64_expand_sve_vector_init):
Declare prototype.
* config/aarch64/aarch64/sve.md (aarch64_sve_rev64<mode>): Use @.
(vec_init<mode><Vel>): New pattern.
* config/aarch64/aarch64.c (emit_insr): New function.
(aarch64_sve_expand_vector_init_handle_trailing_constants): Likewise.
(aarch64_sve_expand_vector_init_insert_elems): Likewise.
(aarch64_sve_expand_vector_init_handle_trailing_same_elem): Likewise.
(aarch64_sve_expand_vector_init): Define two overloaded functions.

testsuite/
* gcc.target/aarch64/sve/init_1.c: New test.
* gcc.target/aarch64/sve/init_1_run.c: Likewise.
* gcc.target/aarch64/sve/init_2.c: Likewise.
* gcc.target/aarch64/sve/init_2_run.c: Likewise.
* gcc.target/aarch64/sve/init_3.c: Likewise.
* gcc.target/aarch64/sve/init_3_run.c: Likewise.
* gcc.target/aarch64/sve/init_4.c: Likewise.
* gcc.target/aarch64/sve/init_4_run.c: Likewise.
* gcc.target/aarch64/sve/init_5.c: Likewise.
* gcc.target/aarch64/sve/init_5_run.c: Likewise.
* gcc.target/aarch64/sve/init_6.c: Likewise.
* gcc.target/aarch64/sve/init_6_run.c: Likewise.
* gcc.target/aarch64/sve/init_7.c: Likewise.
* gcc.target/aarch64/sve/init_7_run.c: Likewise.
* gcc.target/aarch64/sve/init_8.c: Likewise.
* gcc.target/aarch64/sve/init_8_run.c: Likewise.
* gcc.target/aarch64/sve/init_9.c: Likewise.
* gcc.target/aarch64/sve/init_9_run.c: Likewise.
* gcc.target/aarch64/sve/init_10.c: Likewise.
* gcc.target/aarch64/sve/init_10_run.c: Likewise.
* gcc.target/aarch64/sve/init_11.c: Likewise.
* gcc.target/aarch64/sve/init_11_run.c: Likewise.
* gcc.target/aarch64/sve/init_12.c: Likewise.
* gcc.target/aarch64/sve/init_12_run.c: Likewise.

From-SVN: r271857

Fix ICE in vect_slp_analyze_node_operations_1

This patch fixes bug 90681.  It was caused by trying to SLP vectorize a non
groupped load.  We've fixed it by tweaking a bit the implementation: mark
masked loads as not vectorizable, but support them as an special case.  Then
the detect them in the test for normal non-groupped loads that was already
there.

From-SVN: r271856

re PR testsuite/90713 (FAIL: gcc.dg/gimplefe-40.c (internal compiler error))

2019-06-03  Richard Biener  <rguenther@suse.de>

PR testsuite/90713
* gcc.dg/gimplefe-40.c: Add -maltivec for powerpc.

From-SVN: r271855

Make debug(edge) more verbose.

2019-06-03  Martin Liska  <mliska@suse.cz>

* cfg.c (debug): Use TDF_DETAILS for debug and
print edge info only once.

From-SVN: r271854

re PR target/89750 (Wrong code for _mm_comi_round_ss)

2019-05-06  H.J. Lu  <hongjiu.lu@intel.com>
    Hongtao Liu  <hongtao.liu@intel.com>

PR target/89750
PR target/86444
* config/i386/i386-expand.c (ix86_expand_sse_comi_round):
Modified, original implementation isn't correct.

2019-05-06  H.J. Lu  <hongjiu.lu@intel.com>
    Hongtao Liu  <hongtao.liu@intel.com>

PR target/89750
PR target/86444
* gcc.target/i386/avx512f-vcomisd-2.c: New.
* gcc.target/i386/avx512f-vcomisd-2.c: Likewise.

Co-Authored-By: Hongtao Liu <hongtao.liu@intel.com>
From-SVN: r271853

Daily bump.

From-SVN: r271852

re PR fortran/90539 (481.wrf slowdown by 25% on Intel Kaby with -Ofast -march=native starting with r271377)

2019-06-02  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/90539
* trans-expr.c (gfc_conv_subref_array_arg): If the size of the
expression can be determined to be one, treat it as contiguous.
Set likelyhood of presence of an actual argument according to
PRED_FORTRAN_ABSENT_DUMMY and likelyhood of being contiguous
according to PRED_FORTRAN_CONTIGUOUS.

2019-06-02  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/90539
* predict.def (PRED_FORTRAN_CONTIGUOUS): New predictor.

2019-06-02  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/90539
* gfortran.dg/internal_pack_24.f90: New test.

From-SVN: r271844

Daily bump.

From-SVN: r271843

Darwin, x86, testsuite - adjust tests for Darwin PR90698.

We don't have support for -mcmodel={medium, large, kernel} so don't
expect tests for those things to work.

For now mark them as xfail where possible and skip where that isn't.
These changes will be logged onto the PR and therefore can be backed
out when the facility is implemented.

gcc/testsuite/ChangeLog:

2019-06-01  Iain Sandoe  <iain@sandoe.co.uk>

PR target/90698
* gcc.target/i386/pr49866.c: XFAIL for Darwin.
* gcc.target/i386/pr63538.c: Likewise.
* gcc.target/i386/pr61599-1.c: Skip for Darwin.

From-SVN: r271839

PR middle-end/90694 - incorrect representation of ADDR_EXPR involving a pointer to array

gcc/ChangeLog:

PR middle-end/90694
* tree-pretty-print.c (dump_generic_node): Add parentheses.

gcc/testsuite/ChangeLog:

PR middle-end/90694
* gcc.dg/tree-ssa/dump-5.c: New test.

From-SVN: r271838

alias.c: Include ipa-utils.h.

* alias.c: Include ipa-utils.h.
(get_alias_set): Try to complete ODR type via ODR type hash lookup.
* ipa-devirt.c (prevailing_odr_type): New.
* ipa-utils.h (previaling_odr_type): Declare.

* g++.dg/lto/alias-1_0.C: New testcase.
* g++.dg/lto/alias-1_1.C: New testcase.

From-SVN: r271837

Fix changelog

From-SVN: r271836

re PR c++/85254 (boost::is_final does not work for template types)

PR c++/85254

gcc/cp

PR c++/85254
* class.c (fixup_type_variants): Handle CLASSTYPE_FINAL.

testsuite/

PR c++/85254
* g++.dg/ext/is_final.C: Amend.

From-SVN: r271835

Daily bump.

From-SVN: r271832

i386: Don't insert ENDBR after NOTE_INSN_DELETED_LABEL

NOTE_INSN_DELETED_LABEL is used to mark what used to be a 'code_label',
but was not used for other purposes than taking its address which cannot
be used as target for indirect jumps.

Tested on Linux/x86-64 with -fcf-protection.

For x86-64 libc.so on glibc master branch (commit f43b8dd55588c3),

Before: 2961 endbr64
After:  2943 endbr64

gcc/

PR target/89355
* config/i386/i386-features.c (rest_of_insert_endbranch): Remove
NOTE_INSN_DELETED_LABEL check.

gcc/testsuite/

PR target/89355
* gcc.target/i386/cet-label-3.c: New test.
* gcc.target/i386/cet-label-4.c: Likewise.
* gcc.target/i386/cet-label-5.c: Likewise.

Co-Authored-By: Hongtao Liu <hongtao.liu@intel.com>
From-SVN: r271828

* doc/xml/manual/allocator.xml: Move hoard.org back to http.

From-SVN: r271827

mips.c (mips_expand_builtin_insn): Swap the 1st and 3rd operands of the fmadd/fmsub/maddv builtin.

* config/mips/mips.c (mips_expand_builtin_insn): Swap the 1st
and 3rd operands of the fmadd/fmsub/maddv builtin.

* gcc.target/mips/msa-fmadd.c: New.

From-SVN: r271826

tree.h (OMP_CLAUSE__CONDTEMP__ITER): Define.

* tree.h (OMP_CLAUSE__CONDTEMP__ITER): Define.
* gimplify.c (gimplify_scan_omp_clauses): Allow lastprivate conditional
on OMP_SIMD if not nested inside of worksharing loop that also has
lastprivate conditional clause for the same decl.
(gimplify_omp_for): Add _condtemp_ clauses to OMP_SIMD if needed.
* omp-low.c (scan_sharing_clauses): Handle OMP_CLAUSE__CONDTEMP_ also
on simd.
(lower_rec_input_clauses): Likewise.  Handle lastprivate conditional
on simd construct.
(lower_lastprivate_conditional_clauses): Handle lastprivate conditional
on simd construct.
(lower_lastprivate_clauses): Likewise.
(lower_omp_sections): Call lower_lastprivate_conditional_clauses before
calling lower_rec_input_clauses.
(lower_omp_for): Likewise.
(lower_omp_1): Use first rather than second OMP_CLAUSE__CONDTEMP_
clause on simd construct.
* omp-expand.c (expand_omp_simd): Initialize cond_var if
OMP_CLAUSE__CONDTEMP_ clause is present.

* c-c++-common/gomp/lastprivate-conditional-2.c (foo): Don't expect
a sorry on lastprivate conditional on simd construct.
* gcc.dg/vect/vect-simd-6.c: New test.
* gcc.dg/vect/vect-simd-7.c: New test.

From-SVN: r271825

omp-low.c (lower_rec_simd_input_clauses): Set TREE_THIS_NOTRAP on ivar and lvar.

* omp-low.c (lower_rec_simd_input_clauses): Set TREE_THIS_NOTRAP on
ivar and lvar.

* gcc.dg/vect/vect-simd-5.c: New test.

From-SVN: r271824

runtime: drop unused C type reflection code

    In particular, drop __go_type_descriptors_equal, which is no longer
    used, and will be made obsolete by CL 179598.

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/179858

From-SVN: r271823

compiler: optimize append of make

    The gc compiler recognizes append(s, make([]T, n)...), and
    generates code to directly zero the tail instead of allocating a
    new slice and copying. This CL lets the Go frontend do basically
    the same.

    The difficulty is that at the point we handle append, there may
    already be temporaries introduced (e.g. in order_evaluations),
    which makes it hard to find the append-of-make pattern. The
    compiler could "see through" the value of a temporary, but it is
    only safe to do if the temporary is not assigned multiple times.
    For this, we add tracking of assignments and uses for temporaries.

    This also helps in optimizing non-escape slice make. We already
    optimize non-escape slice make with constant len/cap to stack
    allocation. But it failed to handle things like f(make([]T, n))
    (where the slice doesn't escape and n is constant), because of
    the temporary. With tracking of temporary assignments and uses,
    it can handle this now as well.

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/179597

From-SVN: r271822

compiler: handle int-to-string conversion with large integer constant

    Currently, Type_conversion_expression::do_is_constant thinks the
    int-to-string conversion is constant if the integer operand is
    constant, but Type_conversion_expression::do_get_backend actually
    generates a call to runtime.intstring if the integer does not fit
    in a "ushort", which makes it not suitable in constant context,
    such as static initializer.

    This CL makes it handle all constant integer input as constant,
    generating constant string.

    Fixes golang/go#32347.

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/179777

From-SVN: r271821