Use conditional internal functions in if-conversion

This patch uses IFN_COND_* to vectorise conditionally-executed,
potentially-trapping arithmetic, such as most floating-point
ops with -ftrapping-math.  E.g.:

    if (cond) { ... x = a + b; ... }

becomes:

    ...
    x = .COND_ADD (cond, a, b, else_value);
    ...

When this transformation is done on its own, the value of x for
!cond isn't important, so else_value is simply the target's
preferred_else_value (i.e. the value it can handle the most
efficiently).

However, the patch also looks for the equivalent of:

    y = cond ? x : c;

in which the "then" value is the result of the conditionally-executed
operation and the "else" value "c" is some value that is available at x.
In that case we can instead use:

    x = .COND_ADD (cond, a, b, c);

and replace uses of y with uses of x.

The patch also looks for:

    y = !cond ? c : x;

which can be transformed in the same way.  This involved adding a new
utility function inverse_conditions_p, which was already open-coded
in a more limited way in match.pd.

2018-07-12  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
* fold-const.h (inverse_conditions_p): Declare.
* fold-const.c (inverse_conditions_p): New function.
* match.pd: Use inverse_conditions_p.  Add folds of view_converts
that test the inverse condition of a conditional internal function.
* internal-fn.h (vectorized_internal_fn_supported_p): Declare.
* internal-fn.c (internal_fn_mask_index): Handle conditional
internal functions.
(vectorized_internal_fn_supported_p): New function.
* tree-if-conv.c: Include internal-fn.h and fold-const.h.
(any_pred_load_store): Replace with...
(need_to_predicate): ...this new variable.
(redundant_ssa_names): New variable.
(ifcvt_can_use_mask_load_store): Move initial checks to...
(ifcvt_can_predicate): ...this new function.  Handle tree codes
for which a conditional internal function exists.
(if_convertible_gimple_assign_stmt_p): Use ifcvt_can_predicate
instead of ifcvt_can_use_mask_load_store.  Update after variable
name change.
(predicate_load_or_store): New function, split out from
predicate_mem_writes.
(check_redundant_cond_expr): New function.
(value_available_p): Likewise.
(predicate_rhs_code): Likewise.
(predicate_mem_writes): Rename to...
(predicate_statements): ...this.  Use predicate_load_or_store
and predicate_rhs_code.
(combine_blocks, tree_if_conversion): Update after above name changes.
(ifcvt_local_dce): Handle redundant_ssa_names.
* tree-vect-patterns.c (vect_recog_mask_conversion_pattern): Handle
general conditional functions.
* tree-vect-stmts.c (vectorizable_call): Likewise.

gcc/testsuite/
* gcc.dg/vect/vect-cond-arith-4.c: New test.
* gcc.dg/vect/vect-cond-arith-5.c: Likewise.
* gcc.target/aarch64/sve/cond_arith_1.c: Likewise.
* gcc.target/aarch64/sve/cond_arith_1_run.c: Likewise.
* gcc.target/aarch64/sve/cond_arith_2.c: Likewise.
* gcc.target/aarch64/sve/cond_arith_2_run.c: Likewise.
* gcc.target/aarch64/sve/cond_arith_3.c: Likewise.
* gcc.target/aarch64/sve/cond_arith_3_run.c: Likewise.

From-SVN: r262589

Support fused multiply-adds in fully-masked reductions

This patch adds support for fusing a conditional add or subtract
with a multiplication, so that we can use fused multiply-add and
multiply-subtract operations for fully-masked reductions.  E.g.
for SVE we vectorise:

  double res = 0.0;
  for (int i = 0; i < n; ++i)
    res += x[i] * y[i];

using a fully-masked loop in which the loop body has the form:

  res_1 = PHI<0(preheader), res_2(latch)>;
  avec = .MASK_LOAD (loop_mask, a)
  bvec = .MASK_LOAD (loop_mask, b)
  prod = avec * bvec;
  res_2 = .COND_ADD (loop_mask, res_1, prod, res_1);

where the last statement does the equivalent of:

  res_2 = loop_mask ? res_1 + prod : res_1;

(operating elementwise).  The point of the patch is to convert the last
two statements into:

  res_s = .COND_FMA (loop_mask, avec, bvec, res_1, res_1);

which is equivalent to:

  res_2 = loop_mask ? fma (avec, bvec, res_1) : res_1;

(again operating elementwise).

2018-07-12  Richard Sandiford  <richard.sandiford@linaro.org>
    Alan Hayward  <alan.hayward@arm.com>
    David Sherwood  <david.sherwood@arm.com>

gcc/
* internal-fn.h (can_interpret_as_conditional_op_p): Declare.
* internal-fn.c (can_interpret_as_conditional_op_p): New function.
* tree-ssa-math-opts.c (convert_mult_to_fma_1): Handle conditional
plus and minus and convert them into IFN_COND_FMA-based sequences.
(convert_mult_to_fma): Handle conditional plus and minus.

gcc/testsuite/
* gcc.dg/vect/vect-fma-2.c: New test.
* gcc.target/aarch64/sve/reduc_4.c: Likewise.
* gcc.target/aarch64/sve/reduc_6.c: Likewise.
* gcc.target/aarch64/sve/reduc_7.c: Likewise.

Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r262588

Add IFN_COND_FMA functions

This patch adds conditional equivalents of the IFN_FMA built-in functions.
Most of it is just a mechanical extension of the binary stuff.

2018-07-12  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
* doc/md.texi (cond_fma, cond_fms, cond_fnma, cond_fnms): Document.
* optabs.def (cond_fma_optab, cond_fms_optab, cond_fnma_optab)
(cond_fnms_optab): New optabs.
* internal-fn.def (COND_FMA, COND_FMS, COND_FNMA, COND_FNMS): New
internal functions.
(FMA): Use DEF_INTERNAL_FLT_FN rather than DEF_INTERNAL_FLT_FLOATN_FN.
* internal-fn.h (get_conditional_internal_fn): Declare.
(get_unconditional_internal_fn): Likewise.
* internal-fn.c (cond_ternary_direct): New macro.
(expand_cond_ternary_optab_fn): Likewise.
(direct_cond_ternary_optab_supported_p): Likewise.
(FOR_EACH_COND_FN_PAIR): Likewise.
(get_conditional_internal_fn): New function.
(get_unconditional_internal_fn): Likewise.
* gimple-match.h (gimple_match_op::MAX_NUM_OPS): Bump to 5.
(gimple_match_op::gimple_match_op): Add a new overload for 5
operands.
(gimple_match_op::set_op): Likewise.
(gimple_resimplify5): Declare.
* genmatch.c (decision_tree::gen): Generate simplifications for
5 operands.
* gimple-match-head.c (gimple_simplify): Define an overload for
5 operands.  Handle calls with 5 arguments in the top-level overload.
(convert_conditional_op): Handle conversions from unconditional
internal functions to conditional ones.
(gimple_resimplify5): New function.
(build_call_internal): Pass a fifth operand.
(maybe_push_res_to_seq): Likewise.
(try_conditional_simplification): Try converting conditional
internal functions to unconditional internal functions.
Handle 3-operand unconditional forms.
* match.pd (UNCOND_TERNARY, COND_TERNARY): Operator lists.
Define ternary equivalents of the current rules for binary conditional
internal functions.
* config/aarch64/aarch64.c (aarch64_preferred_else_value): Handle
ternary operations.
* config/aarch64/iterators.md (UNSPEC_COND_FMLA, UNSPEC_COND_FMLS)
(UNSPEC_COND_FNMLA, UNSPEC_COND_FNMLS): New unspecs.
(optab): Handle them.
(SVE_COND_FP_TERNARY): New int iterator.
(sve_fmla_op, sve_fmad_op): New int attributes.
* config/aarch64/aarch64-sve.md (cond_<optab><mode>)
(*cond_<optab><mode>_2, *cond_<optab><mode_4)
(*cond_<optab><mode>_any): New SVE_COND_FP_TERNARY patterns.

gcc/testsuite/
* gcc.dg/vect/vect-cond-arith-3.c: New test.
* gcc.target/aarch64/sve/vcond_13.c: Likewise.
* gcc.target/aarch64/sve/vcond_13_run.c: Likewise.
* gcc.target/aarch64/sve/vcond_14.c: Likewise.
* gcc.target/aarch64/sve/vcond_14_run.c: Likewise.
* gcc.target/aarch64/sve/vcond_15.c: Likewise.
* gcc.target/aarch64/sve/vcond_15_run.c: Likewise.
* gcc.target/aarch64/sve/vcond_16.c: Likewise.
* gcc.target/aarch64/sve/vcond_16_run.c: Likewise.

From-SVN: r262587

Extend tree code folds to IFN_COND_*

This patch adds match.pd support for applying normal folds to their
IFN_COND_* forms.  E.g. the rule:

  (plus @0 (negate @1)) -> (minus @0 @1)

also allows the fold:

  (IFN_COND_ADD @0 @1 (negate @2) @3) -> (IFN_COND_SUB @0 @1 @2 @3)

Actually doing this by direct matches in gimple-match.c would
probably lead to combinatorial explosion, so instead, the patch
makes gimple_match_op carry a condition under which the operation
happens ("cond"), and the value to use when the condition is false
("else_value").  Thus in the example above we'd do the following

(a) convert:

      cond:NULL_TREE (IFN_COND_ADD @0 @1 @4 @3) else_value:NULL_TREE

    to:

      cond:@0 (plus @1 @4) else_value:@3

(b) apply gimple_resimplify to (plus @1 @4)

(c) reintroduce cond and else_value when constructing the result.

Nested operations inherit the condition of the outer operation
(so that we don't introduce extra faults) but have a null else_value.
If we try to build such an operation, the target gets to choose what
else_value it can handle efficiently: obvious choices include one of
the operands or a zero constant.  (The alternative would be to have some
representation for an undefined value, but that seems a bit invasive,
and isn't likely to be useful here.)

I've made the condition a mandatory part of the gimple_match_op
constructor so that it doesn't accidentally get dropped.

2018-07-12  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
* target.def (preferred_else_value): New target hook.
* doc/tm.texi.in (TARGET_PREFERRED_ELSE_VALUE): New hook.
* doc/tm.texi: Regenerate.
* targhooks.h (default_preferred_else_value): Declare.
* targhooks.c (default_preferred_else_value): New function.
* internal-fn.h (conditional_internal_fn_code): Declare.
* internal-fn.c (FOR_EACH_CODE_MAPPING): New macro.
(get_conditional_internal_fn): Use it.
(conditional_internal_fn_code): New function.
* gimple-match.h (gimple_match_cond): New struct.
(gimple_match_op): Add a cond member function.
(gimple_match_op::gimple_match_op): Update all forms to take a
gimple_match_cond.
* genmatch.c (expr::gen_transform): Use the same condition as res_op
for the suboperation, but don't specify a particular else_value.
* tree-ssa-sccvn.c (vn_nary_simplify, vn_reference_lookup_3)
(visit_nary_op, visit_reference_op_load): Pass
gimple_match_cond::UNCOND to the gimple_match_op constructor.
* gimple-match-head.c: Include tree-eh.h
(convert_conditional_op): New function.
(maybe_resimplify_conditional_op): Likewise.
(gimple_resimplify1): Call maybe_resimplify_conditional_op.
(gimple_resimplify2): Likewise.
(gimple_resimplify3): Likewise.
(gimple_resimplify4): Likewise.
(maybe_push_res_to_seq): Return null for conditional operations.
(try_conditional_simplification): New function.
(gimple_simplify): Call it.  Pass conditions to the gimple_match_op
constructor.
* match.pd: Fold VEC_COND_EXPRs of an IFN_COND_* call to a new
IFN_COND_* call.
* config/aarch64/aarch64.c (aarch64_preferred_else_value): New
function.
(TARGET_PREFERRED_ELSE_VALUE): Redefine.

gcc/testsuite/
* gcc.dg/vect/vect-cond-arith-2.c: New test.
* gcc.target/aarch64/sve/loop_add_6.c: Likewise.

From-SVN: r262586

lto-streamer-out.c (DFS::DFS_write_tree_body): Do not stream DECL_FCONTEXT

* lto-streamer-out.c (DFS::DFS_write_tree_body): Do not stream
DECL_FCONTEXT
(hash_tree): Do not hash DECL_FCONTEXT
* tree-streamer-in.c (lto_input_ts_field_decl_tree_pointers):
Do not stream DECL_FCONTEXT.
* tree-streamer-out.c (write_ts_field_decl_tree_pointers): Likewise.
* tree.c (free_lang_data_in_decl): Free DECL_FCONTEXT.

From-SVN: r262585

re PR debug/86462 (Quite huge debug info size increase introduced in r262511)

2018-07-12  Richard Biener  <rguenther@suse.de>

PR debug/86462
* dwarf2out.c (gen_block_die): Only output blocks when they have
at least one !DECL_IGNORED_P variable.

From-SVN: r262583

re PR target/84829 (-mieee-fp causes to link with -lieee but that is no longer available)

2018-07-12  Richard Biener  <rguenther@suse.de>

PR target/84829
* config/gnu-user.h (GNU_USER_TARGET_NO_PTHREADS_LIB_SPEC):
Remove -mieee-fp handling.

* gcc.target/i386/pr84829.c: New testcase.

From-SVN: r262582

* gcc.dg/torture/20180712-1.c: New test.

From-SVN: r262579

tree-ssa-sccvn.c (vn_lookup_simplify_result): Remove bogus left-over from last patch.

2018-07-12  Richard Biener  <rguenther@suse.de>

* tree-ssa-sccvn.c (vn_lookup_simplify_result): Remove bogus
left-over from last patch.

From-SVN: r262578

re PR tree-optimization/86492 (store-merging wrong-code)

PR tree-optimization/86492
* gimple-ssa-store-merging.c
(imm_store_chain_info::coalesce_immediate_stores): Call
check_no_overlap even for the merge_overlapping case.  Formatting fix.

* gcc.c-torture/execute/pr86492.c: New test.

From-SVN: r262576

re PR c/86453 (error: type variant differs by TYPE_PACKED in free_lang_data since r255469)

2018-07-12  Richard Biener  <rguenther@suse.de>

PR c/86453
* c-attribs.c (handle_packed_attribute): Do not build a variant
type with TYPE_PACKED, instead ignore the attribute if we may
not apply to the original type.

* g++.dg/warn/pr86453.C: New testcase.

From-SVN: r262575

re PR tree-optimization/86479 ([graphite] ICE in gimplify_modify_expr, at gimplify.c:5756)

2018-07-12  Richard Biener  <rguenther@suse.de>

PR middle-end/86479
* fold-const.c (fold_binary_op_with_conditional_arg): Do not
move possibly trapping operations into the conditional.

* gcc.dg/graphite/pr86479.c: New testcase.

From-SVN: r262574

tree-ssa-sccvn.c (mprts_hook_cnt): Remove.

2018-07-12  Richard Biener  <rguenther@suse.de>

* tree-ssa-sccvn.c (mprts_hook_cnt): Remove.
(vn_lookup_simplify_result): Remove recursion limit applied
here.
(vn_nary_build_or_lookup_1): Adjust.
(try_to_simplify): Likewise.
* gimple-match-head.c (gimple_resimplify1): Instead apply one
here.
(gimple_resimplify2): Likewise.
(gimple_resimplify3): Likewise.
(gimple_resimplify4): Likewise.

From-SVN: r262573

compiler: build a single backend type for type alias

    A type alias and its underlying type are identical. Build a
    single backend type for them. Previously we build two backend
    types, which sometimes confuse the backend's type system.

    Also don't include type aliases into the list of named type
    declarations, since they are not named types.

    Reviewed-on: https://go-review.googlesource.com/123362

From-SVN: r262572

Daily bump.

From-SVN: r262571

avx512bitalgintrin.h (_mm512_mask_bitshuffle_epi64_mask): Use __mmask64 type instead of __mmask8 for __M argument.

* config/i386/avx512bitalgintrin.h (_mm512_mask_bitshuffle_epi64_mask):
Use __mmask64 type instead of __mmask8 for __M argument.
* config/i386/avx512fintrin.h (_mm512_mask_xor_epi64,
_mm512_maskz_xor_epi64): Use __mmask8 type instead of __mmask16 for
__U argument.
(_mm512_mask_cmpneq_epi64_mask): Use __mmask8 type instead of
__mmask16 for __M argument.
(_mm512_maskz_insertf32x4, _mm512_maskz_inserti32x4,
_mm512_mask_insertf32x4, _mm512_mask_inserti32x4): Cast last argument
to __mmask16 instead of __mmask8.
* config/i386/avx512vlintrin.h (_mm_mask_add_ps, _mm_maskz_add_ps,
_mm256_mask_add_ps, _mm256_maskz_add_ps, _mm_mask_sub_ps,
_mm_maskz_sub_ps, _mm256_mask_sub_ps, _mm256_maskz_sub_ps,
_mm256_maskz_cvtepi32_ps, _mm_maskz_cvtepi32_ps): Use __mmask8 type
instead of __mmask16 for __U argument.
* config/i386/avx512vlbwintrin.h (_mm_mask_cmp_epi8_mask): Use
__mmask16 instead of __mmask8 for __U argument.
(_mm256_mask_cmp_epi8_mask): Use __mmask32 instead of __mmask16 for
__U argument.
(_mm256_cmp_epi8_mask): Use __mmask32 return type instead of
__mmask16.
(_mm_mask_cmp_epu8_mask): Use __mmask16 instead of __mmask8 for __U
argument.
(_mm256_mask_cmp_epu8_mask): Use __mmask32 instead of __mmask16 for
__U argument.
(_mm256_cmp_epu8_mask): Use __mmask32 return type instead of
__mmask16.
(_mm_mask_cmp_epi16_mask): Cast last argument to __mmask8 instead
of __mmask16.
(_mm256_mask_cvtepi8_epi16): Use __mmask16 instead of __mmask32 for
__U argument.
(_mm_mask_cvtepi8_epi16): Use __mmask8 instead of __mmask32 for
__U argument.
(_mm256_mask_cvtepu8_epi16): Use __mmask16 instead of __mmask32 for
__U argument.
(_mm_mask_cvtepu8_epi16): Use __mmask8 instead of __mmask32 for
__U argument.
(_mm256_mask_cmpneq_epu8_mask, _mm256_mask_cmplt_epu8_mask,
_mm256_mask_cmpge_epu8_mask, _mm256_mask_cmple_epu8_mask): Change
return type as well as __M argument type and all casts from __mmask8
to __mmask32.
(_mm256_mask_cmpneq_epu16_mask, _mm256_mask_cmplt_epu16_mask,
_mm256_mask_cmpge_epu16_mask, _mm256_mask_cmple_epu16_mask): Change
return type as well as __M argument type and all casts from __mmask8
to __mmask16.
(_mm256_mask_cmpneq_epi8_mask, _mm256_mask_cmplt_epi8_mask,
_mm256_mask_cmpge_epi8_mask, _mm256_mask_cmple_epi8_mask): Change
return type as well as __M argument type and all casts from __mmask8
to __mmask32.
(_mm256_mask_cmpneq_epi16_mask, _mm256_mask_cmplt_epi16_mask,
_mm256_mask_cmpge_epi16_mask, _mm256_mask_cmple_epi16_mask): Change
return type as well as __M argument type and all casts from __mmask8
to __mmask16.
* config/i386/avx512vbmi2vlintrin.h (_mm_mask_shrdi_epi32,
_mm_mask_shldi_epi32): Cast last argument to __mmask8 instead of
__mmask16.

* gcc.target/i386/avx512bw-vpcmpb-2.c (CMP): Use SIZE macro instead
of hardcoding size.  Cast (rel) to MASK_TYPE.
* gcc.target/i386/avx512bw-vpcmpub-2.c (CMP): Likewise.
* gcc.target/i386/avx512f-vinserti32x4-3.c: New test.
* gcc.target/i386/avx512f-vinsertf32x4-3.c: New test.
* gcc.target/i386/avx512vl-vpcmpnequb-2.c: New test.
* gcc.target/i386/avx512vl-vpcmpgeub-2.c: New test.
* gcc.target/i386/avx512vl-vpcmpleb-2.c: New test.
* gcc.target/i386/avx512vl-vpcmpgeb-2.c: New test.
* gcc.target/i386/avx512vl-vpcmpltb-2.c: New test.
* gcc.target/i386/avx512vl-vpcmpltub-2.c: New test.
* gcc.target/i386/avx512vl-vpcmpleub-2.c: New test.
* gcc.target/i386/avx512vl-vpcmpneqb-2.c: New test.
* gcc.target/i386/avx512vl-vpcmpnequw-2.c: New test.
* gcc.target/i386/avx512vl-vpcmpgeuw-2.c: New test.
* gcc.target/i386/avx512vl-vpcmplew-2.c: New test.
* gcc.target/i386/avx512vl-vpcmpgew-2.c: New test.
* gcc.target/i386/avx512vl-vpcmpltw-2.c: New test.
* gcc.target/i386/avx512vl-vpcmpltuw-2.c: New test.
* gcc.target/i386/avx512vl-vpcmpleuw-2.c: New test.
* gcc.target/i386/avx512vl-vpcmpneqw-2.c: New test.

2018-07-11  Grazvydas Ignotas  <notasas@gmail.com>

* config/i386/avx512bwintrin.h: (_mm512_mask_cmp_epi8_mask,
_mm512_mask_cmp_epu8_mask): Use __mmask64 type instead of __mmask32
for __U argument.

* gcc.target/i386/avx512bw-vpcmpb-2.c (SIZE): Define to
(AVX512F_LEN / 8) instead of (AVX512F_LEN / 16).
* gcc.target/i386/avx512bw-vpcmpub-2.c (SIZE): Likewise.

From-SVN: r262566

Correct date on ChangeLog entry.

From-SVN: r262563

md.texi (define_subst): Document how multiple occurrences of the same argument in the replacement...

* doc/md.texi (define_subst): Document how multiple occurrences of
the same argument in the replacement pattern are handled.

From-SVN: r262562

Add documentation for "mode" attribute for types.

* doc/extend.texi (Common Variable Attributes): Move "mode" into
alphabetical order.
(Common Type Attributes): Add "mode" attribute.

From-SVN: r262561

lto-streamer-out.c (DFS::DFS_write_tree_body): Do not stream DECL_ORIGINAL_TYPE.

* lto-streamer-out.c (DFS::DFS_write_tree_body): Do not
stream DECL_ORIGINAL_TYPE.
(DFS::DFS_write_tree_body): Drop hack handling local external decls.
(hash_tree): Do not walk DECL_ORIGINAL_TYPE.
* tree-streamer-in.c (lto_input_ts_decl_non_common_tree_pointers):
Do not walk original type.
* tree-streamer-out.c (streamer_write_chain): Drop hack handling
external decls.
(write_ts_decl_non_common_tree_pointers): Do not stream
DECL_ORIGINAL_TYPE
* tree.c (free_lang_data_in_decl): Clear DECL_ORIGINAL_TYPE.
(find_decls_types_r): Do not walk DEC_ORIGINAL_TYPE.

From-SVN: r262560

tree-ssa-threadupdate.c (thread_through_all_blocks): Do not jump thread twice from the same starting edge.

        * tree-ssa-threadupdate.c (thread_through_all_blocks): Do not jump
thread twice from the same starting edge.

From-SVN: r262559

vr-values.c (gimple_stmt_nonzero_p): Abstract common code to...

        * vr-values.c (gimple_stmt_nonzero_p): Abstract common code to...
        * gimple.c (gimple_call_nonnull_result_p): ...here...
        (gimple_call_nonnull_arg): ...and here.
        * gimple.h (gimple_call_nonnull_result_p): New.
        (gimple_call_nonnull_arg): New.

From-SVN: r262557

[arm] Put CPU's FPU capabilities directly in the ISA specification

As part of the transition from the original support for named FPUs to
general FPU properties I defined an entry in the CPU definitions in
arm-cpus.in to use a named FPU.  However, that has now outlived its
usefulness and increasingly we are likely to find that newer cores do
not fit the legacy FPU names very well.  Furthermore it is now possible
to encode all the FPU capatilities directly in the ISA definitions, often
as simply as using +fp or +simd.

So this patch removes the fpu field from the "define cpu" entries and
instead encodes the same information in the isa field.  This also alows us
to remove a bit of now-dead code from parsecpu.awk.

* config/arm/arm-cpus.in: Move information from fpu field of each
cpu definition to the isa field.
* config/arm/parsecpu.awk (fpu): Delete match rule.
(gen_comm_data): Don't add bits from the CPU's FPU entry.

From-SVN: r262556

compiler: fix evaluation order of LHS index expressions

    The spec says that when an index expression appears on the left hand
    side of an assignment, the operands should be evaluated. The
    gofrontend code was assuming that that only referred to the index
    operand. But discussion of https://golang.org/issue/23188 has
    clarified that this means both the slice/map/string operand and the
    index operand. Adjust the gofrontend code accordingly.

    Fixes golang/go#23188

    Reviewed-on: https://go-review.googlesource.com/123155

From-SVN: r262554

re PR c++/86443 (ICEs on #pragma omp distribute parallel for with class iterators)

PR c++/86443
* testsuite/libgomp.c++/for-15.C (a): Remove unused variable.
(results): Make sure the variable is not inside declare target region.
(qux): Remove unused function.

From-SVN: r262552

re PR debug/86457 (ICE in add_dwarf_attr, at dwarf2out.c:4405 since r251448)

2018-07-11  Richard Biener  <rguenther@suse.de>

PR debug/86457
* dwarf2out.c (init_sections_and_labels): Use
output_asm_line_debug_info consistently.
(dwarf2out_early_finish): Likewise.
(dwarf2out_finish): Remove DW_AT_stmt_list from early generated
type units.

From-SVN: r262551

tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Rework father_bb setting in a way to avoid propagating constants multiple times...

2018-07-10  Richard Biener  <rguenther@suse.de>

* tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1):
Rework father_bb setting in a way to avoid propagating constants
multiple times on a loop body.

From-SVN: r262550

Daily bump.

From-SVN: r262549

PR debug/86459 - Fix -gsplit-dwarf -g3 gcc_assert

There was a typo in the output_macinfo_op gcc_assert.
The function is called dwarf_FORM, not dwarf_form.
Add the provided testcase from the bug to test -gsplit-dwarf -g3.

gcc/ChangeLog:

PR debug/86459
* dwarf2out.c (output_macinfo_op): Fix dwarf_FORM typo in gcc_assert.

gcc/testsuite/ChangeLog:

PR debug/86459
* gcc.dg/pr86459.c: New test.

From-SVN: r262545

PR testsuite/86461 - test case gcc.c-torture/execute/builtins/strlen-3.c fails

gcc/testsuite/ChangeLog:
* gcc.c-torture/execute/builtins/strlen-3.c: Adjust.

From-SVN: r262543

Change dg-warning to db-bogus in a test-case (PR testsuite/86445).

2018-07-10  Martin Liska  <mliska@suse.cz>

        PR testsuite/86445
* gcc.dg/pr84100.c: Change it back to dg-bogus.

From-SVN: r262540

hash-map.h (hash_map::iterator::operator*): Return references to key and value.

2018-07-10  Richard Biener  <rguenther@suse.de>

* hash-map.h (hash_map::iterator::operator*): Return
references to key and value.

From-SVN: r262539

re PR sanitizer/86406 ([UBSAN] -fcompare-debug failure with -fsanitize=undefined)

PR sanitizer/86406
* cp-gimplify.c (cp_maybe_instrument_return): Skip trailing
DEBUG_BEGIN_STMTs.

* g++.dg/ubsan/pr86406.C: New test.

From-SVN: r262536

re PR fortran/86421 (OpenMP declare simd linear ref in module causes gfortran to bail out)

PR fortran/86421
* module.c (omp_declare_simd_clauses): Add LINEAR with _REF, _VAL and
_UVAL suffixes.
(mio_omp_declare_simd): Save and restore ref, val and uval modifiers
on linear clauses.  Initialize n->where to gfc_current_locus.

* gfortran.dg/vect/pr86421.f90: New test.

From-SVN: r262535

re PR c++/86443 (ICEs on #pragma omp distribute parallel for with class iterators)

PR c++/86443
* gimplify.c (find_combined_omp_for): Add DATA argument, in addition
to finding the inner OMP_FOR/OMP_SIMD stmt find non-trivial wrappers,
BLOCKs with BLOCK_VARs, OMP_PARALLEL in between, OMP_FOR in between.
(gimplify_omp_for): For composite loops, move outer
OMP_{DISTRIBUTE,TASKLOOP,FOR,PARALLEL} right around innermost
OMP_FOR/OMP_SIMD if there are any non-trivial wrappers.  For class
iterators add any needed clauses.  Allow OMP_FOR_ORIG_DECLS to contain
TREE_LIST for both the original class iterator and the "last" helper
var.  Gimplify OMP_FOR_PRE_BODY before the outermost composite
loop, remember has_decl_expr from outer composite loops for the
innermost OMP_SIMD in TREE_PRIVATE bit on OMP_FOR_INIT.
gcc/c-family/
* c-omp.c (c_omp_check_loop_iv_r, c_omp_check_loop_iv): Allow declv
to contain TREE_LIST for both the original class iterator and the
"last" helper var.
gcc/cp/
* semantics.c (handle_omp_for_class_iterator): Remove lastp argument,
instead of setting *lastp turn orig_declv elt into a TREE_LIST.
(finish_omp_for): Adjust handle_omp_for_class_iterator caller.
* pt.c (tsubst_omp_for_iterator): Allow OMP_FOR_ORIG_DECLS to contain
TREE_LIST for both the original class iterator and the "last" helper
var.
libgomp/
* testsuite/libgomp.c++/for-15.C: New test.

From-SVN: r262534

compiler: fix double evaluation with interface field expression

    In Interface_field_reference_expression, the interface expression
    is used in two places, so a temporary variable is used. Previously,
    we used a Set_and_use_temporary_expression, which, when evaluated
    twice, causes double evaluation of the underlying expression. Fix
    by setting the temporary once and use Temporary_reference_expression
    instead.

    Fixes golang/go#26248.

    Reviewed-on: https://go-review.googlesource.com/122756

From-SVN: r262533

Daily bump.

From-SVN: r262532

PR tree-optimization/86415 - strlen() not folded for substrings within constant arrays

gcc/testsuite/ChangeLog:

* gcc.dg/strlenopt-53.c: New test.

From-SVN: r262528

PR middle-end/77357 - strlen of constant strings not folded

gcc/ChangeLog:

PR middle-end/77357
PR middle-end/86428
* builtins.c (c_strlen): Avoid out-of-bounds warnings when
accessing implicitly initialized array elements.
* expr.c (string_constant): Handle string initializers of
character arrays within aggregates.
* gimple-fold.c (fold_array_ctor_reference): Add argument.
Store element offset.  As a special case, handle zero size.
(fold_nonarray_ctor_reference): Same.
(fold_ctor_reference): Add argument.  Store subobject offset.
* gimple-fold.h (fold_ctor_reference): Add argument.

gcc/testsuite/ChangeLog:

PR middle-end/77357
* gcc.dg/strlenopt-49.c: New test.
* gcc.dg/strlenopt-50.c: New test.
* gcc.dg/strlenopt-51.c: New test.
* gcc.dg/strlenopt-52.c: New test.

From-SVN: r262522

Improve code generation for pdp11 target.

* config/pdp11/pdp11.c (pdp11_addr_cost): New function.
(pdp11_insn_cost): New function.
(pdp11_md_asm_adjust): New function.
(TARGET_INVALID_WITHIN_DOLOOP): Define.
(pdp11_rtx_costs): Update to match machine better.
(output_addr_const_pdp11): Correct format mismatch warnings.
* config/pdp11/pdp11.h (SLOW_BYTE_ACCESS): Correct definition.
* config/pdp11/pdp11.md: General change to add base_cost and/or
length attributes for use by new pdp11_insn_cost function.
(MIN_BRANCH): Correct definition.
(MIN_SOB): Ditto.
(doloop_end): Use standard pattern name for looping pattern.
(doloop_end_nocc): New.
(movsf): Add another constraint alternative.
(zero_extendqihi2): Add constraint alternatives for not in place
extend.
(zero_extendhisi2): Remove.
(shift patterns): Add CC handling variants.
(bswaphi2): New.
(bswapsi2): New.
(rothi3): New.
(define_peephole2): New peephole to recognize mov that sets CC for
subsequent test.

From-SVN: r262519

Fix overflow handling in sparc.c:sparc_fold_builtin

* config/sparc/sparc.c (sparc_fold_builtin) <SPARC_BUILTIN_PDIST,
SPARC_BUILTIN_PDISTN>: Adapt for signature change of wi::neg,
wi::add.

From-SVN: r262518

re PR c/86420 (nextafter(0x1p-1022,0) is constant folded)

PR c/86420
* real.c (real_nextafter): Return true if result is denormal.

* gcc.dg/nextafter-1.c (TEST): Adjust the tests that expect denormals
to be returned and when first argument is not 0, so that they don't do
anything for NEED_EXC or NEED_ERRNO.

From-SVN: r262517

[testsuite] Use relative line numbers in gcc.dg/guality

Use relative line number where obvious: either the relative line number is
'.', '.-1' or '.+1', or adjacent to another obvious case.

2018-07-09  Tom de Vries  <tdevries@suse.de>

* gcc.dg/guality/asm-1.c: Use relative line numbers where obvious.
* gcc.dg/guality/bswaptest.c: Same.
* gcc.dg/guality/clztest.c: Same.
* gcc.dg/guality/csttest.c: Same.
* gcc.dg/guality/ctztest.c: Same.
* gcc.dg/guality/drap.c: Same.
* gcc.dg/guality/nrv-1.c: Same.
* gcc.dg/guality/pr41353-1.c: Same.
* gcc.dg/guality/pr41353-2.c: Same.
* gcc.dg/guality/pr41404-1.c: Same.
* gcc.dg/guality/pr43051-1.c: Same.
* gcc.dg/guality/pr43077-1.c: Same.
* gcc.dg/guality/pr43177.c: Same.
* gcc.dg/guality/pr43329-1.c: Same.
* gcc.dg/guality/pr43479.c: Same.
* gcc.dg/guality/pr43593.c: Same.
* gcc.dg/guality/pr45003-1.c: Same.
* gcc.dg/guality/pr45003-2.c: Same.
* gcc.dg/guality/pr45003-3.c: Same.
* gcc.dg/guality/pr48437.c: Same.
* gcc.dg/guality/pr48466.c: Same.
* gcc.dg/guality/pr49888.c: Same.
* gcc.dg/guality/pr54200.c: Same.
* gcc.dg/guality/pr54519-1.c: Same.
* gcc.dg/guality/pr54519-2.c: Same.
* gcc.dg/guality/pr54519-3.c: Same.
* gcc.dg/guality/pr54519-4.c: Same.
* gcc.dg/guality/pr54519-5.c: Same.
* gcc.dg/guality/pr54519-6.c: Same.
* gcc.dg/guality/pr54551.c: Same.
* gcc.dg/guality/pr54693-2.c: Same.
* gcc.dg/guality/pr54693.c: Same.
* gcc.dg/guality/pr54796.c: Same.
* gcc.dg/guality/pr54970.c: Same.
* gcc.dg/guality/pr67192.c: Same.
* gcc.dg/guality/pr69947.c: Same.
* gcc.dg/guality/pr78726.c: Same.
* gcc.dg/guality/rotatetest.c: Same.
* gcc.dg/guality/sra-1.c: Same.
* gcc.dg/guality/vla-2.c: Same.

From-SVN: r262516

Add missing Optimization attribute.

2018-07-09  Martin Liska  <mliska@suse.cz>

* common.opt: Add back wrongly removed attribute.

From-SVN: r262513

decl.c (grokdeclarator): Use rich_location::add_range in three more places; include gcc-rich-location.h.

/cp
2018-07-09  Paolo Carlini  <paolo.carlini@oracle.com>

* decl.c (grokdeclarator): Use rich_location::add_range in three
more places; include gcc-rich-location.h.

/testsuite
2018-07-09  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/diagnostic/long-short.C: New.
* g++.dg/diagnostic/signed-unsigned.C: Likewise.
* g++.dg/diagnostic/virtual-friend.C: Likewise.
* g++.old-deja/g++.brendan/crash11.C: Adjust.

From-SVN: r262512

re PR debug/86413 (gcc.dg/guality/pr48437.c fail)

2018-07-09  Richard Biener  <rguenther@suse.de>

PR debug/86413
* dwarf2out.c (gen_block_die): For an early generated DIE
always output high/low PC attributes.

From-SVN: r262511

[debug] Handle debug references to skipped params

When compiling guality/vla-1.c with -O3 -g, vla a in f1 is optimized away, but
f1 still contains a debug expression describing the upper bound of the vla
(D.1914):
...
     __attribute__((noinline))
     f1 (intD.6 iD.1900)
     {
       <bb 2>
       saved_stack.1_2 = __builtin_stack_save ();
       # DEBUG BEGIN_STMT
       # DEBUG D#3 => i_1(D) + 1
       # DEBUG D#2 => (long intD.8) D#3
       # DEBUG D#1 => D#2 + -1
       # DEBUG D.1914 => (sizetype) D#1
...

Then f1 is cloned to a version f1.constprop with no parameters, eliminating
parameter i, and 'DEBUG D#3 => i_1(D) + 1' turns into 'D#3 => NULL'.

This patch fixes that by defining debug expressions for default defs of
eliminated parameters in remap_ssa_name:
...
     __attribute__((noinline))
     f1.constprop ()
     {
       intD.6 iD.1949;

       <bb 3>
       # DEBUG D#8 s=> iD.1900
       # DEBUG iD.1949 => D#8

       <bb 2>
+      # DEBUG D#6 s=> iD.1949
       saved_stack.1_1 = __builtin_stack_save ();
       # DEBUG BEGIN_STMT
-      # DEBUG D#3 => NULL
+      # DEBUG D#3 => D#6 + 1
       # DEBUG D#2 => (long intD.8) D#3
       # DEBUG D#1 => D#2 + -1
       # DEBUG D.1951 => (sizetype) D#1
...

Bootstrapped and reg-tested on x86_64.

2018-07-09  Tom de Vries  <tdevries@suse.de>

* cfgexpand.c (expand_debug_source_expr): Handle VAR_DECL.
* tree-inline.c (remap_ssa_name): Handle default def ssa_name mapping
onto VAR_DECL with abstract origin.

* gcc.dg/vla-1.c: New test.

From-SVN: r262510

Daily bump.

From-SVN: r262509

Daily bump.

From-SVN: r262505

RISC-V: Finish Ada port.

gcc/
* config/riscv/riscv.c (TARGET_CUSTOM_FUNCTION_DESCRIPTORS): New.

From-SVN: r262501

Fix sigsegv on -fdump-tree-all-enumerate_locals

2018-07-07  Tom de Vries  <tdevries@suse.de>

* tree-dfa.c (dump_enumerated_decls): Handle cfun->cfg == NULL.

* gcc.misc-tests/options.exp (check_for_all_options): Clean up dump
files.
(get_dump_flags): New proc.
(toplevel): Test all dump flags.

From-SVN: r262500

c-ada-spec.c (to_ada_name): Remove index parameter.

* c-ada-spec.c (to_ada_name): Remove index parameter.
(pp_ada_tree_identifier): Likewise.
(dump_ada_macros): Adjust call to to_ada_name.
(struct overloaded_name_hash): Delete.
(struct overloaded_name_hasher): Likewise.
(overloaded_names): Likewise.
(compute_overloading_index): Likewise.
(dump_ada_decl_name): Do not call compute_overloading_index and
adjust calls to pp_ada_tree_identifier.
(dump_ada_double_name): Adjust calls to pp_ada_tree_identifier.
(dump_ada_import): Add spc parameter and switch to aspect syntax.
(dump_ada_function_declaration): Adjust call to pp_ada_tree_identifier.
(dump_ada_enum_type): Remove type and display_convention parameters.
Adjust calls to pp_ada_tree_identifier.
(dump_ada_node): Likewise and for dump_ada_structure.
(dump_nested_type) <ENUMERAL_TYPE>: Adjust call to dump_ada_enum_type
and tidy up.
<RECORD_TYPE>: Adjust call to dump_ada_structure and switch to aspect
syntax.
(print_constructor): Adjust call to pp_ada_tree_identifier.
(print_destructor): Likewise.
(dump_ada_declaration): Switch to aspect syntax.
(dump_ada_structure): Likewise and tidy up.  Replace display_convention
parameter with nested parameter.
(dump_ads): Emit pragma Ada_2012 in lieu of pragma Ada_2005.
(dump_ada_specs): Do not delete overloaded_names table.

From-SVN: r262499

decl.c (gnat_to_gnu_entity): Add GNAT_DECL local variable and use it throughout.

* gcc-interface/decl.c (gnat_to_gnu_entity): Add GNAT_DECL local
variable and use it throughout.
<E_Variable>: If the nominal subtype of the object is unconstrained,
compute the Ada size separately and put in on the padding type if the
size is not fixed.
<E_Record_Type>: Minor tweak.
* gcc-interface/misc.c (gnat_type_max_size): Rename max_size_unit
into max_size_unit throughout.

From-SVN: r262498

gigi.h (add_decl_expr): Adjust prototype.

* gcc-interface/gigi.h (add_decl_expr): Adjust prototype.
* gcc-interface/decl.c (gnat_to_gnu_entity): Remove useless test.
* gcc-interface/trans.c (add_stmt_with_node): Remove exceptions.
(add_decl_expr): Change type of second parameter and rename it.
(renaming_from_instantiation_p): New function moved from...
(set_expr_location_from_node): Test for exceptions here and add one
for actual subtypes built for unconstrained composite actuals.
* gcc-interface/utils.c (renaming_from_instantiation_p): ...here.

From-SVN: r262497

trans.c (lvalue_required_p): Remove ALIASED parameter and adjust recursive calls.

* gcc-interface/trans.c (lvalue_required_p): Remove ALIASED parameter
and adjust recursive calls.
(Identifier_to_gnu): Adjust calls to lvalue_required_p.
(gnat_to_gnu): Likewise.

From-SVN: r262496

decl.c (gnat_to_gnu_param): Minor tweak.

* gcc-interface/decl.c (gnat_to_gnu_param): Minor tweak.
(gnat_to_gnu_subprog_type): New pure_flag local variable.  Set it for
a pure Ada function with a by-ref In parameter.  Propagate it onto the
function type by means of the TYPE_QUAL_RESTRICT flag.
* gcc-interface/utils.c (finish_subprog_decl): Set DECL_PURE_P if the
function type has the TYPE_QUAL_RESTRICT flag set.

From-SVN: r262495

tree-vrp.c (vrp_int_const_binop): Change overflow type to overflow_type.

* tree-vrp.c (vrp_int_const_binop): Change overflow type to
overflow_type.
(combine_bound): Use wide-int overflow calculation instead of
rolling our own.
* calls.c (maybe_warn_alloc_args_overflow): Change overflow type to
overflow_type.
* fold-const.c (int_const_binop_2): Same.
(extract_muldiv_1): Same.
(fold_div_compare): Same.
(fold_abs_const): Same.
* match.pd: Same.
* poly-int.h (add): Same.
(sub): Same.
(neg): Same.
(mul): Same.
* predict.c (predict_iv_comparison): Same.
* profile-count.c (slow_safe_scale_64bit): Same.
* simplify-rtx.c (simplify_const_binary_operation): Same.
* tree-chrec.c (tree_fold_binomial): Same.
* tree-data-ref.c (split_constant_offset_1): Same.
* tree-if-conv.c (idx_within_array_bound): Same.
* tree-scalar-evolution.c (iv_can_overflow_p): Same.
* tree-ssa-phiopt.c (minmax_replacement): Same.
* tree-vect-loop.c (is_nonwrapping_integer_induction): Same.
* tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Same.
* vr-values.c (vr_values::adjust_range_with_scev): Same.
* wide-int.cc (wi::add_large): Same.
(wi::mul_internal): Same.
(wi::sub_large): Same.
(wi::divmod_internal): Same.
* wide-int.h: Change overflow type to overflow_type for neg, add,
mul, smul, umul, div_trunc, div_floor, div_ceil, div_round,
mod_trunc, mod_ceil, mod_round, add_large, sub_large,
mul_internal, divmod_internal.
(overflow_type): New enum.
(accumulate_overflow): New.

cp/
* decl.c (build_enumerator): Change overflow type to overflow_type.
* init.c (build_new_1): Same.

From-SVN: r262494

re PR target/84711 (AArch32 big-endian fails when taking subreg of a vector mode to a scalar mode.)

PR target/84711
* gcc.dg/vect/pr84711.c: Remove unnecessary sse
dg-require-effective-target.  Add -msse not just on i386-*, but
on all i?86-* and x86_64-*.

From-SVN: r262493

Daily bump.

From-SVN: r262492

tree-ssa-phiopt.c (cond_removal_in_popcount_pattern): New.

gcc/ChangeLog:

2018-07-06  Kugan Vivekanandarajah  <kugan.vivekanandarajah@linaro.org>

* tree-ssa-phiopt.c (cond_removal_in_popcount_pattern): New.
(tree_ssa_phiopt_worker): Call cond_removal_in_popcount_pattern.

gcc/testsuite/ChangeLog:

2018-07-06  Kugan Vivekanandarajah  <kugan.vivekanandarajah@linaro.org>

* gcc.dg/tree-ssa/phi-opt-16.c: New test.
* gcc.dg/tree-ssa/phi-opt-17.c: New test.
* gcc.dg/tree-ssa/phi-opt-18.c: New test.
* gcc.dg/tree-ssa/phi-opt-19.c: New test.
* gcc.dg/tree-ssa/popcount3.c: New test.

From-SVN: r262488

tree-ssa-loop-niter.c (number_of_iterations_popcount): If popcount argument is checked for zero before entering loop, avoid checking again.

gcc/ChangeLog:

2018-07-06  Kugan Vivekanandarajah  <kuganv@linaro.org>

* tree-ssa-loop-niter.c (number_of_iterations_popcount): If popcount
argument is checked for zero before entering loop, avoid checking again.

From-SVN: r262487

gimplify.h (generic_expr_could_trap_p): Set as global function.

gcc/ChangeLog:

2018-07-06  Kugan Vivekanandarajah  <kuganv@linaro.org>

* gimplify.h (generic_expr_could_trap_p): Set as global function.
* gimplify.h (generic_expr_could_trap_p): Likwise.
* tree-scalar-evolution.c (expression_expensive_p): Handle COND_EXPR.

gcc/testsuite/ChangeLog:

2018-07-06  Kugan Vivekanandarajah  <kuganv@linaro.org>

* gcc.dg/tree-ssa/pr64183.c: Disable final value replacement
to preserve the test.
* gcc.target/i386/pr85073.c: Likewise.

From-SVN: r262486

re PR tree-optimization/86401 (The "For constants M and N, if M == (1LL << cst) - 1 && (N & M) == M,..." opts are only in fold-const.c and in RTL)

PR tree-optimization/86401
* fold-const.c (fold_binary_loc) <case BIT_AND_EXPR>: Move the
((A & N) + B) & M -> (A + B) & M etc. optimization into ...
(fold_bit_and_mask): ... here.  New helper function for match.pd.
* fold-const.h (fold_bit_and_mask): Declare.
* match.pd (((A & N) + B) & M -> (A + B) & M): New optimization.

* gcc.dg/tree-ssa/pr86401-1.c: New test.
* gcc.dg/tree-ssa/pr86401-2.c: New test.
* c-c++-common/rotate-9.c: New test.

From-SVN: r262485

re PR target/86324 (testsuite test divkc3-1.c FAILs when compiling with -mabi=ieeelongdouble)

gcc/
PR target/86324
* target.def (translate_mode_attribute): New hook.
* targhooks.h (default_translate_mode_attribute): Declare.
* targhooks.c (default_translate_mode_attribute): New function.
* doc/tm.texi.in (TARGET_TRANSLATE_MODE_ATTRIBUTE): New hook.
* doc/tm.texi: Regenerate.
* config/rs6000/rs6000.c (TARGET_TRANSLATE_MODE_ATTRIBUTE): Define.
(rs6000_translate_mode_attribute): New function.

gcc/c-family/
PR target/86324
* c-attribs.c (handle_mode_attribute): Call new translate_mode_attribute
target hook.

gcc/testsuite/
PR target/86324
gcc.target/powerpc/pr86324-1.c: New test.
gcc.target/powerpc/pr86324-2.c: Likewise.

From-SVN: r262484

RISC-V: Initial riscv linux Ada port.

* Makefile.rtl: Add riscv*-linux* support.
* libgnarl/s-linux__riscv.ads: New.
* libgnat/system-linux-riscv.ads: New.

From-SVN: r262482

Makefile patches from initial RISC-V cross/native build.

* Make-generated.in (treeprs.ads): Use $(GNATMAKE) instead of gnatmake.
(einfo.h, sinfo.h, stamp-snames, stamp-nmake): Likewise.
* gcc-interface/Makefile.in (xoscons): Likewise.

From-SVN: r262481

functions.h (__gnu_debug::__check_string): Move...

2018-07-06  François Dumont  <fdumont@gcc.gnu.org>

* include/debug/functions.h (__gnu_debug::__check_string): Move...
* include/debug/string (__gnu_debug::__check_string): ... here.
(_GLIBCXX_DEBUG_VERIFY_STR_COND_AT): New.
(__glibcxx_check_string_n_constructor): New.
(__gnu_debug::basic_string<>(const _CharT*, size_type, const _Alloc&)):
Use latter.
(__glibcxx_check_string_constructor): New.
(__gnu_debug::basic_string<>(const _CharT*, const _Alloc&)):
Use latter.
* testsuite/21_strings/basic_string/debug/1_neg.cc: New.
* testsuite/21_strings/basic_string/debug/2_neg.cc: New.

From-SVN: r262480

Update internal documentation to describe that DONE and FAIL
also apply to define_split and define_peephole2.

* doc/md.texi (define_split): Document DONE and FAIL.
(define_peephole2): Ditto.

From-SVN: r262479

PR libstdc++/84928 use std::move in <numeric> algorithms

P0616R0 altered the effects of the <numeric> algorithms to use std::move
on the accumulator values (resolving LWG 2055). This implements the
change for C++2a, but retains the previous behaviour for older
standards.

* include/bits/stl_numeric.h (_GLIBCXX_MOVE_IF_20): Define macro to
conditionally move, according to __cplusplus value.
(accumulate, inner_product, partial_sum, adjacent_difference): Use
_GLIBCXX_MOVE_IF_20.
* testsuite/26_numerics/accumulate/lwg2055.cc: New test.
* testsuite/26_numerics/adjacent_difference/lwg2055.cc: New test.
* testsuite/26_numerics/inner_product/lwg2055.cc: New test.
* testsuite/26_numerics/partial_sum/lwg2055.cc: New test.

From-SVN: r262477

Simplify linker script patterns for std::exception_ptr

* config/abi/pre/gnu.ver: Use wildcards to combine related patterns.

From-SVN: r262475

P0935R0 Eradicating unnecessarily explicit default constructors

This is the last remaining piece of P0935R0. This adds a default
constructor to each of the streambuf and stream types in <sstream> so
that default construction does not use the 'explicit' constructor that
has a single, defaulted argument.

P0935R0 Eradicating unnecessarily explicit default constructors
* config/abi/pre/gnu.ver: Tighten existing patterns and export new
default constructor symbols.
* include/std/sstream (basic_stringbuf, basic_istringstream)
(basic_ostringstream, basic_stringstream): Remove default arguments
from explicit constructors taking ios_base::openmode and add separate
non-explicit default constructors.
* testsuite/27_io/basic_istringstream/cons/default.cc: New.
* testsuite/27_io/basic_ostringstream/cons/default.cc: New.
* testsuite/27_io/basic_stringstream/cons/default.cc: New.
* testsuite/27_io/basic_stringbuf/cons/char/default.cc: New.
* testsuite/27_io/basic_stringbuf/cons/wchar_t/default.cc: New.

From-SVN: r262474

Require sse for testcase on i686.

PR target/84711
* gcc.dg/vect/pr84711.c: Add -msse for i686 targets.

From-SVN: r262472

* include/std/variant (__accepted_index): Use void_t.

From-SVN: r262471

Fix -Wpedantic issue with testcase on AArch64.

2018-07-06  Tamar Christina  <tamar.christina@arm.com>

* gcc.target/aarch64/struct_cpy.c: Remove ;.

From-SVN: r262470

Update config.sub

* config.sub: Sync with upstream version 2018-07-03.

From-SVN: r262469

RTEMS/Ada: Change system configuration

The previous system configuration led to undefined references to
__gnat_raise_nodefer_with_msg().

gcc/ada
* libgnat/system-rtems.ads (Frontend_Exceptions): Set to False.
(ZCX_By_Default): Set to True.

From-SVN: r262467

Update config.guess and config.sub

* config.guess: Sync with upstream version 2018-06-26.
* config.sub: Sync with upstream version 2018-07-02.

From-SVN: r262466

asm-2.c (REGISTER): Override for v850 too.

* gcc.dg/tree-ssa/asm-2.c (REGISTER): Override for v850 too.

2018-07-05  Fritz Reese  <fritzoreese@gmail.com>

From-SVN: r262465

re PR tree-optimization/86010 (redundant memset with smaller size not eliminated)

PR tree-optimization/86010
* tree-ssa-dse.c (compute_trims): More aggressively trim at
both the head and tail of mem* and str* calls.

From-SVN: r262464

RISC-V: Add support for riscv-*-*.

* config.gcc (riscv*-*-*): When setting xlen, handle riscv-*.

From-SVN: r262463

Daily bump.

From-SVN: r262462

aarch64-simd.md: correct flags text for MIN_EXPR replacement.

2018-07-05  Indu Bhagat  <indu.bhagat@oracle.com>

        * config/aarch64/aarch64-simd.md: correct flags text for
        MIN_EXPR replacement.

From-SVN: r262458

PR libstdc++/85831 define move constructors and operators for exceptions

PR libstdc++/85831
* config/abi/pre/gnu.ver: Export move constructors and move
assignment operators for std::logic_error and std::runtime_error.
* include/std/stdexcept: Use _GLIBCXX_NOTHROW instead of
_GLIBCXX_USE_NOEXCEPT.
(logic_error, runtime_error): Declare move constructors and move
assignment operators. When not declared already, define copy
constructors and copy assignment operators as explicit-defaulted.
(domain_error, invalid_argument, length_error, out_of_range)
(overflow_error, underflow_error): Define move constructors and move
assignment operators as explicitly-defaulted.
* libsupc++/exception.h (exception): Likewise.
* src/c++11/cow-stdexcept.cc (logic_error, runtime_error): Define
move constructors and move assignment operators as defaulted.
* testsuite/19_diagnostics/stdexcept.cc: Check that constructors and
assignment operators are defined.

From-SVN: r262456

Add missing '|' as wrong patch was applied.

From-SVN: r262453

dfp.m4 (enable_decimal_float): Enable for x86_64*-*-gnu* to catch x86_64 kFreeBSD and Hurd.

config/
* dfp.m4 (enable_decimal_float): Enable for x86_64*-*-gnu* to
catch x86_64 kFreeBSD and Hurd.

gcc/
* configure: Regenerated.

libdecnumber/
* configure: Regenerated.

libgcc/
* configure: Regenerated.

From-SVN: r262452

libtool.m4: Sort output of 'find' to enable deterministic builds.

        * libtool.m4: Sort output of 'find' to enable deterministic builds.
* ltmain.sh: Likewise.

From-SVN: r262451

Add xfail-if to some tests that fail with COW strings

These tests fail when run with -D_GLIBCXX_USE_CXX11_ABI=0

* testsuite/21_strings/basic_string/cons/char/deduction.cc: XFAIL for
COW strings.
* testsuite/21_strings/basic_string/cons/wchar_t/deduction.cc:
Likewise.
* testsuite/21_strings/basic_string/requirements/
explicit_instantiation/debug.cc: Likewise.

From-SVN: r262448

PR libstdc++/58265 add noexcept to basic_string::assign(basic_string&&)

PR libstdc++/58265
* include/bits/basic_string.h [!_GLIBCXX_USE_CXX11_ABI]
(basic_string::assign(basic_string&&)): Add conditional noexcept
depending on the allocator's is_always_equal property (LWG 2063).
* testsuite/21_strings/basic_string/modifiers/assign/char/
move_assign.cc: Check for non-throwing exception specification.
* testsuite/21_strings/basic_string/modifiers/assign/wchar_t/
move_assign.cc: Likewise.

From-SVN: r262447

re PR fortran/86408 (bogus error: ABSTRACT INTERFACE must not have an assumed character length result (F2003: C418))

2018-07-05  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/86408
* resolve.c.c (resolve_contained_fntype): Reference to C418 is
in F2008 and not F2003.
(resolve_function): Ditto in error message. Also, exclude
deferred character length results from the error.

2018-07-05  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/86408
* gfortran.dg/deferred_character_20.f90: New test.

From-SVN: r262445

PR libstdc++/58265 implement LWG 2063 for COW strings

For COW strings the default constructor does not allocate when
_GLIBCXX_FULLY_DYNAMIC_STRING == 0, so can be noexcept. The move
constructor and swap do not allocate when the allocators are equal, so
add conditional noexcept using allocator_traits::is_always_equal.

PR libstdc++/58265
* include/bits/basic_string.h [!_GLIBCXX_USE_CXX11_ABI]
[_GLIBCXX_FULLY_DYNAMIC_STRING==0] (basic_string::basic_string()):
Add GLIBCXX_NOEXCEPT.
(basic_string::operator=(basic_string&&)): Add _GLIBCXX_NOEXCEPT_IF
to depend on the allocator's is_always_equal property (LWG 2063).
(basic_string::swap(basic_string&)): Likewise.
* include/bits/basic_string.tcc [!_GLIBCXX_USE_CXX11_ABI]
(basic_string::swap(basic_string&)): Likewise.
* testsuite/21_strings/basic_string/allocator/char/move_assign.cc:
Check is_nothrow_move_assignable.
* testsuite/21_strings/basic_string/allocator/wchar_t/move_assign.cc:
Check is_nothrow_move_assignable.
* testsuite/21_strings/basic_string/cons/char/
noexcept_move_construct.cc: Likewise.
* testsuite/21_strings/basic_string/cons/wchar_t/
noexcept_move_construct.cc: Likewise.

From-SVN: r262443

re PR fortran/83183 (Out of memory with option -finit-derived)

2018-07-05  Fritz Reese  <fritzoreese@gmail.com>

    gcc/fortran/ChangeLog:

PR fortran/83183
PR fortran/86325
* expr.c (class_allocatable, class_pointer, comp_allocatable,
comp_pointer): New helpers.
(component_initializer): Generate EXPR_NULL for allocatable or pointer
components. Do not generate initializers for components within BT_CLASS.
Do not assign to comp->initializer.
(gfc_generate_initializer): Use new helpers; move code to generate
EXPR_NULL for class allocatable components into component_initializer().

    gcc/testsuite/ChangeLog:

PR fortran/83183
PR fortran/86325
* gfortran.dg/init_flag_18.f90: New testcase.
* gfortran.dg/init_flag_19.f03: New testcase.

From-SVN: r262442

rs6000-c.c: Map ALTIVEC_BUILTIN_VEC_UNPACKH for float argument to VSX_BUILTIN_DOUBLEH_V4SF.

gcc/ChangeLog:

2018-07-05  Carl Love  <cel@us.ibm.com>

* config/rs6000/rs6000-c.c: Map ALTIVEC_BUILTIN_VEC_UNPACKH for
float argument to VSX_BUILTIN_DOUBLEH_V4SF.
Map ALTIVEC_BUILTIN_VEC_UNPACKL for float argument to
VSX_BUILTIN_DOUBLEL_V4SF.

gcc/testsuite/ChangeLog:

2018-07-05  Carl Love  <cel@us.ibm.com>
* gcc.target/altivec-1-runnable.c: New test file.
* gcc.target/altivec-2-runnable.c: New test file.
* gcc.target/vsx-7.c (main2): Change expected expected instruction
for tests.

From-SVN: r262439

PR tree-optimization/86400 - set<string>::set<char (*)[2]) constructor does not work with array argument

gcc/ChangeLog:
* tree-ssa-strlen.c (maybe_set_strlen_range): Use type size rather
than its domain to compute its the upper bound of a char array.

gcc/testsuite/ChangeLog:
* gcc.dg/strlenopt-47.c: New test.
* gcc.dg/strlenopt-48.c: New test.

From-SVN: r262438

Replace NO_IMPLICIT_EXTERN_C with SYSTEM_IMPLICIT_EXTERN_C.

gcc/
Replace NO_IMPLICIT_EXTERN_C with SYSTEM_IMPLICIT_EXTERN_C.
* doc/cpp.texi: Update comment.
* doc/tm.texi: Rebuilt.
* doc/tm.texi.in (NO_IMPLICIT_EXTERN_C): Replace with ...
(SYSTEM_IMPLICIT_EXTERN_C): ... this, opposite sense.
* doc/extend.texi (Backwards Compatibility): Clarify it is system
headers affected by extern "C".
* system.h: Poison NO_IMPLICIT_EXTERN_C.
* config/alpha/alpha.h, config/arm/uclinux-elf.h,
config/bfin/elf.h, config/cris/cris.h, config/darwin.h,
config/dragonfly.h, config/freebsd.h, config/gnu-user.h,
config/i386/cygming.h, config/i386/djgpp.h, config/i386/nto.h,
config/ia64/hpux.h, config/lm32/lm32.h, config/lm32/uclinux-elf.h,
config/lynx.h, config/mips/elf.h, config/mmix/mmix.h,
config/netbsd.h, config/pa/pa-hpux.h, config/powerpcspe/sysv4.h,
config/riscv/elf.h, config/rs6000/sysv4.h, config/rtems.h,
config/s390/tpf.h, config/sh/newlib.h, config/sol2.h,
config/sparc/openbsd64.h, config/sparc/sp-elf.h,
config/sparc/sp64-elf.h, config/spu/spu.h,
config/stormy16/stormy16.h, config/v850/v850.h,
config/visium/visium.h, config/vx-common.h, config/xtensa/elf.h: Don't
define NO_IMPLICIT_EXTERN_C.
* config/rs6000/aix.h: Set SYSTEM_IMPLICIT_EXTERN_C.

gcc/c-family/
* c-lex.c (fe_file_change): Check SYSTEM_IMPLICIT_EXTERN_C not
NO_IMPLICIT_EXTERN_C.

gcc/cp/
* cp/decl.c (decls_match): Check SYSTEM_IMPLICIT_EXTERN_C not
NO_IMPLICIT_EXTERN_C.
* cp/parser.c (cp_parser_parameter_declaration_clause): Likewise.

From-SVN: r262437

Fix can_change_mode_class for big-endian on Arm

gcc/

PR target/84711
* config/arm/arm.c (arm_can_change_mode_class): Use GET_MODE_UNIT_SIZE
instead of GET_MODE_SIZE when comparing Units.

gcc/testsuite/

PR target/84711
* gcc.target/arm/big-endian-subreg.c: New.

From-SVN: r262436

Correct subreg no-op handling for big-endian vec_select.

gcc/

PR target/84711
* rtlanal.c (set_noop_p): Constrain on mode change,
include hard-reg-set.h

gcc/testuite/

PR target/84711
* gcc.dg/vect/pr84711.c: New.

From-SVN: r262435

Simplify movmem code by always doing overlapping copies when larger than 8 bytes on AArch64.

This changes the movmem code in AArch64 that does copy for data between 4 and 7
bytes to use the smallest possible mode capable of copying the remaining bytes in one
go and then overlapping the reads if needed.

This means that if we're copying 5 bytes we would issue an SImode and QImode
load instead of two SImode loads.

This does smaller memory accesses but also gives the mid-end a chance to realise
that it can CSE the loads in certain circumstances. e.g. when you have something
like

return foo;

where foo is a struct. This would be transformed by the mid-end into SSA form as

D.XXXX = foo;

return D.XXXX;

This movmem routine will handle the first copy, but it's usually not needed,
the mid-end would do SImode and QImode stores into X0 for the 5 bytes example
but without the first copies being in the same mode, it doesn't know it doesn't
need the stores at all.

From-SVN: r262434

Revert
2018-07-04  Maxim Ostapenko  <m.ostapenko@samsung.com>

PR sanitizer/84250
* config/gnu-user.h (LIBASAN_EARLY_SPEC): Pass -lstdc++ for static
libasan.
* gcc.c: Do not pass LIBUBSAN_SPEC if ASan is enabled with UBSan.

* Makefile.am: Reorder libs.
* Makefile.in: Regenerate.
* asan/Makefile.am: Define DCAN_SANITIZE_UB=1, add dependancy from
libsanitizer_ubsan.la.
* asan/Makefile.in: Regenerate.
* ubsan/Makefile.am: Define new libsanitizer_ubsan.la library.
* ubsan/Makefile.in: Regenerate.

From-SVN: r262433

[testsuite] Remove 'note:' from scan-tree-dump directives

2018-07-05  Christophe Lyon  <christophe.lyon@linaro.org>

* c-c++-common/unroll-1.c: Remove 'note:' in matching string.
* c-c++-common/unroll-2.c: Likewise.
* g++.dg/cdce3.C: Likewise.
* g++.dg/ext/unroll-1.C: Likewise.
* g++.dg/ext/unroll-2.C: Likewise.
* g++.dg/ext/unroll-3.C: Likewise.
* gcc.dg/cdce1.c: Likewise.
* gcc.dg/cdce2.c: Likewise.
* gcc.dg/gomp/pr68128-1.c: Likewise.
* gcc.dg/vect/pr46032.c: Likewise.
* gcc.dg/vect/vect-cond-10.c: Likewise.
* gcc.dg/vect/vect-cond-8.c: Likewise.
* gcc.dg/vect/vect-cond-9.c: Likewise.
* gcc.dg/vect/vect-mask-load-1.c: Likewise.
* gcc.dg/vect/vect-mask-loadstore-1.c: Likewise.
* gcc.target/i386/avx-cvt-2.c: Likewise.
* gcc.target/i386/avx-cvt-3.c: Likewise.
* gcc.target/i386/avx2-cvt-2.c: Likewise.
* gcc.target/i386/avx2-gather-2.c: Likewise.
* gcc.target/i386/avx2-gather-6.c: Likewise.
* gcc.target/i386/avx512f-gather-2.c: Likewise.
* gcc.target/i386/sse2-cvt-2.c: Likewise.
* gfortran.dg/directive_unroll_1.f90: Likewise.
* gfortran.dg/directive_unroll_2.f90: Likewise.
* gnat.dg/unroll2.adb: Likewise.
* gnat.dg/unroll3.adb: Likewise.

From-SVN: r262432

[testsuite] Simplify dg-final

2018-07-05  Tom de Vries  <tdevries@suse.de>

* lib/gcc-dg.exp (dg-final): Simplify tcl code.

From-SVN: r262431