re PR target/81593 (Optimize PowerPC vector set from vector extract)

[gcc]
2017-08-07  Michael Meissner  <meissner@linux.vnet.ibm.com>

PR target/81593
* config/rs6000/vsx.md (vsx_concat_<mode>, VSX_D): Cleanup
constraints since the -mupper-regs-* switches have been
eliminated.
(vsx_concat_<mode>_1): New combiner insns to recognize inserting
into a vector from a double word element that was extracted from
another vector, and eliminate extra XXPERMDI instructions.
(vsx_concat_<mode>_2): Likewise.
(vsx_concat_<mode>_3): Likewise.
(vsx_set_<mode>, VSX_D): Rewrite vector set in terms of vector
concat to allow optimizing inserts from previous extracts.

[gcc/testsuite]
2017-08-07  Michael Meissner  <meissner@linux.vnet.ibm.com>

PR target/81593
* gcc.target/powerpc/vec-setup.h: New tests to test various
combinations of setting up vectors of 2 double word elements.
* gcc.target/powerpc/vec-setup-long.c: Likewise.
* gcc.target/powerpc/vec-setup-double.c: Likewise.
* gcc.target/powerpc/vec-setup-be-long.c: Likewise.
* gcc.target/powerpc/vec-setup-be-double.c: Likewise.
* gcc.target/powerpc/vsx-extract-6.c: New tests for optimzing
vector inserts from vector extracts.
* gcc.target/powerpc/vsx-extract-7.c: Likewise.

From-SVN: r250936

i386.c (ix86_stack_protect_guard): Generate memory reference to a SSP offset in TLS address space.

* config/i386/i386.c (ix86_stack_protect_guard): Generate
memory reference to a SSP offset in TLS address space.
(ix86_print_operand) <case '@'>: Remove.
(ix86_print_operand_punct_valid_p): Remove '@' code.
* config/i386/i386.md (unspec): Remove UNSPEC_SP_TLS_SET and
UNSPEC_SP_TLS_TEST.
(stack_tls_protect_set_<mode>): Remove.
(stack_protect_set): Do not call gen_stack_tls_protect_set_<mode>.
(stack_tls_protect_test_<mode>): Remove.
(stack_protect_test): Do not call gen_stack_tls_protect_test_<mode>.

From-SVN: r250932

Olivier Hainque  <hainque@adacore.com>

PR target/81755
* config/vxworksae.h (VXWORKS_HAVE_TLS): Define.

From-SVN: r250931

Makefile.in (install-mkheaders): Fix typo, where the multi_dir variable was referenced as multidir in command.

2017-08-07  Douglas Rupp  <rupp@adacore.com>

* Makefile.in (install-mkheaders): Fix typo, where the multi_dir
variable was referenced as multidir in command.

From-SVN: r250930

re PR c/69389 (bit field incompatible with OpenMP atomic update)

PR c/69389
* gimplify.c (goa_stabilize_expr): Handle BIT_INSERT_EXPR and
BIT_FIELD_REF.

* c-omp.c (c_finish_omp_atomic): Handle atomics on bitfields.

* testsuite/libgomp.c/pr69389.c: New test.
* testsuite/libgomp.c++/pr69389.C: New test.

From-SVN: r250929

Add missing header file attribs.h to couple of targets.

2017-08-07  Martin Liska  <mliska@suse.cz>

* config/m32c/m32c.c: Add include of stringpool.h and attribs.h.
* config/rl78/rl78.c: Add include of attribs.h.
* config/sh/sh.c: Likewise.
* config/v850/v850.c: Likewise.

From-SVN: r250926

Fix diff_type in expand_oacc_for char iter_type

2017-08-07  Tom de Vries  <tom@codesourcery.com>

PR middle-end/78266
* omp-expand.c (expand_oacc_for): Ensure diff_type is large enough.

* testsuite/libgomp.oacc-c-c++-common/vprop-2.c: New test.
* testsuite/libgomp.oacc-c-c++-common/vprop.c: Remove xfail.

From-SVN: r250925

Fix missing include of header file in mips.c.

2017-08-07  Martin Liska  <mliska@suse.cz>

* config/mips/mips.c: Include attribs.h.

From-SVN: r250924

re PR fortran/68829 (Segfaults with -Ofast due to large array on stack)

2017-08-07  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/68829
* doc/invoke.texi: Document change in behvaior for -Ofast for
Fortran.

2017-08-07  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/68829
PR fortran/81701
* options.c: Make -Ofast honor -fmax-stack-var-size.
* invoke.texi: Document change.

2017-08-07  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/68829
PR fortran/81701
* gfortran.dg/o_fast_stacksize.90:  New test.

From-SVN: r250923

* es.po: Update.

From-SVN: r250921

c-ada-spec.c (has_nontrivial_methods): Test for FUNCTION_DECL.

* c-ada-spec.c (has_nontrivial_methods): Test for FUNCTION_DECL.
(print_ada_methods): Likewise.
(print_ada_declaration): Likewise.

From-SVN: r250920

[AArch64] Use gen_frame_mem for callee-saves

The frame code uses a mixture of gen_rtx_MEM and gen_frame_mem for
callee-saves.  Callee-saves never alias with local variables, so using
gen_frame_mem is best.

    gcc/
* config/aarch64/aarch64.c (aarch64_pushwb_single_reg):
Use gen_frame_mem.
(aarch64_pop_regs): Likewise.
(aarch64_gen_load_pair): Likewise.
(aarch64_save_callee_saves): Likewise.
(aarch64_restore_callee_saves): Likewise.

From-SVN: r250919

Fix unresolved in gcc.dg/pr46932.c

Build only if pre-increment is supported.  Given there is no config test,
add a list of targets which have HAVE_PRE_INCREMENT set.

    testsuite/
PR middle-end/46932
* gcc.dg/pr46932.c: Compile on targets with pre-increment.

From-SVN: r250918

Revert r250916

From-SVN: r250917

i386: Don't use frame pointer without stack access

When there is no stack access, there is no need to use frame pointer
even if -fno-omit-frame-pointer is used.

gcc/

PR target/81736
* config/i386/i386.c (ix86_finalize_stack_realign_flags): Renamed
to ...
(ix86_finalize_stack_frame_flags): This.  Also clear
frame_pointer_needed if -fno-omit-frame-pointer is used without
stack access.
(ix86_expand_prologue): Replace ix86_finalize_stack_realign_flags
with ix86_finalize_stack_frame_flags.
(ix86_expand_epilogue): Likewise.
(ix86_expand_split_stack_prologue): Likewise.

gcc/testsuite/

PR target/81736
* gcc.target/i386/pr81736-1.c: New test.
* gcc.target/i386/pr81736-2.c: Likewise.
* gcc.target/i386/pr81736-3.c: Likewise.
* gcc.target/i386/pr81736-4.c: Likewise.

From-SVN: r250916

i386: Set priority to P_AES for Westmere

The difference between Nehalem and Westmere is AES.  We should set
priority to P_AES for Westmere, not P_PROC_SSE4_2 which is for Nehalem.
Otherwise, we will pick Nehalem implementation on Westmere.  Tested on
Westmere.

PR target/81743
* config/i386/i386.c (get_builtin_code_for_version): Set priority
to P_AES for Westmere.

From-SVN: r250915

Share mingw fset-stack-executable with cygwin

This patch is in use by Cygwin for years, upstream to GCC.

* gcc/config/i386/mingw.opt (fset-stack-executable): Removed.
* gcc/config/i386/cygming.opt (fset-stack-executable): Moved
from mingw.opt.
* gcc/config/i386/cygwin.h: Define CHECK_EXECUTE_STACK_ENABLED.
* ligcc/config.host (*-cygwin): Include file from mingw
config/i386/enable-execute-stack-mingw32.c

From-SVN: r250914

print-rtl: NOT is ~, not !

Slim RTL dumps print the RTL code NOT as !, but that is misleading:
its semantics are like the C operator ~.  This fixes it.

* print-rtl.c (print_exp): Print NOT as "~" instead of as "!".

From-SVN: r250913

re PR middle-end/81737 (164.gzip in SPEC CPU 2000 failed to build)

PR middle-end/81737
* fold-const.c (fold_indirect_ref_1): Check type_domain.

* gcc.dg/pr81737.c: New test.

From-SVN: r250912

Canonicalize names of attributes.

2017-08-07  Martin Liska  <mliska@suse.cz>

* attribs.h (canonicalize_attr_name): New function.
(cmp_attribs): Move from c-format.c and adjusted.
(is_attribute_p): Moved from tree.h.
* tree-inline.c: Add new includes.
* tree.c (cmp_attrib_identifiers): Use cmp_attribs.
(private_is_attribute_p): Remove.
(private_lookup_attribute): Likewise.
(private_lookup_attribute_by_prefix): Simplify.
(remove_attribute): Use is_attribute_p.
* tree.h: Remove removed declarations.
2017-08-07  Martin Liska  <mliska@suse.cz>

* array-notation-common.c: Add new includes.
* c-format.c( handle_format_attribute): Canonicalize a format
function name.
* c-lex.c (c_common_has_attribute): Canonicalize name of an
attribute.
* c-pretty-print.c: Add new include.
2017-08-07  Martin Liska  <mliska@suse.cz>

* parser.c (cp_parser_gnu_attribute_list): Canonicalize name of an
attribute.
(cp_parser_std_attribute): Likewise.
* tree.c: Add new include.
2017-08-07  Martin Liska  <mliska@suse.cz>

* c-parser.c (c_parser_attributes): Canonicalize name of an
attribute.
2017-08-07  Martin Liska  <mliska@suse.cz>

* go-gcc.cc (Gcc_backend::function): Look up for no_split_stack
and not __no_split_stack__.
2017-08-07  Martin Liska  <mliska@suse.cz>

* g++.dg/cpp0x/pr65558.C: Update scanned pattern.
* gcc.dg/parm-impl-decl-1.c: Likewise.
* gcc.dg/parm-impl-decl-3.c: Likewise.
* gcc.dg/Wattributes-5.c: New test.

From-SVN: r250911

* include/system/sys/ptrace.h: New file.

From-SVN: r250910

re PR middle-end/81698 (expand_case uses wrong edge as default edge)

PR middle-end/81698
* stmt.c (emit_case_dispatch_table): Add DEFAULT_EDGE argument,
instead of computing it in the function.  Formatting fix.
(expand_case): Don't rely on default_edge being the first edge,
clear it if removing it, pass default_edge to
emit_case_dispatch_table.
(expand_sjlj_dispatch_table): Pass NULL as DEFAULT_EDGE, formatting
fix.

From-SVN: r250909

Daily bump.

From-SVN: r250908

target_attr_10.c: Add -mcpu=generic.

2017-08-06  Andrew Pinski  <apinski@cavium.com>

        * gcc.target/aarch64/target_attr_10.c: Add -mcpu=generic.
        * gcc.target/aarch64/target_attr_13.c: LIkewise.
        * gcc.target/aarch64/target_attr_15.c: LIkewise.
        * gcc.target/aarch64/target_attr_4.c: Likewise.
        * gcc.target/aarch64/target_attr_1.c: Add -march=armv8-a.
        * gcc.target/aarch64/target_attr_2.c: Likewise.
        * gcc.target/aarch64/target_attr_7.c: Likewise.
        * gcc.target/aarch64/target_attr_crypto_ice_1.c: Likewise.
        * gcc.target/aarch64/target_attr_crypto_ice_2.c: Likewise.
        * gcc.target/aarch64/target_attr_3.c: Add -mcpu=generic -march=armv8-a.

From-SVN: r250904

atomic_cmp_exchange_zero_reg_1.c: Pass -march=armv8-a+nolse, skip if -mcpu= is passed.

2017-08-06  Andrew Pinski  <apinski@cavium.com>

        * gcc.target/aarch64/atomic_cmp_exchange_zero_reg_1.c: Pass
        -march=armv8-a+nolse, skip if -mcpu= is passed.
        * gcc.target/aarch64/atomic_cmp_exchange_zero_strong_1.c: Likewise.

From-SVN: r250903

_Float16_1.c: Skip if supplied a -mcpu= option.

2017-08-06  Andrew Pinski  <apinski@cavium.com>

        * gcc.target/aarch64/_Float16_1.c: Skip if supplied a -mcpu= option.
        * gcc.target/aarch64/_Float16_2.c: Likewise.
        * gcc.target/aarch64/_Float16_3.c: Likewise.

From-SVN: r250902

alpha.c (alpha_reorg): If trap is the last active insn in the function, emit NOP after the insn.

* config/alpha/alpha.c (alpha_reorg): If trap is the last active
insn in the function, emit NOP after the insn.

From-SVN: r250901

Add missing edge probabilities in expand_oacc_for, tile case

 ;;   basic block 11, loop depth 2, freq 0, maybe hot
 ;;    prev block 22, next block 23, flags: (NEW, REACHABLE)
 ;;    pred:       22 (TRUE_VALUE)
 ;;                24 (TRUE_VALUE)
   ...
   if (.e_offset.15D.1971 < .e_bound.14D.1972)
-    goto <bb 23>; [100.00%] [count: INV]
+    goto <bb 23>; [80.01%] [count: INV]
   else
-    goto <bb 24>; [INV] [count: INV]
+    goto <bb 24>; [19.99%] [count: INV]

 ;;   basic block 23, loop depth 3, freq 0, maybe hot
 ;;    prev block 11, next block 24, flags: (NEW)
 ;;    pred:       11 [always]  (TRUE_VALUE)
 ;;                23 (TRUE_VALUE)
   ixD.1974 = .outer.4D.1967 + .e_offset.15D.1971;
   .e_offset.15D.1971 = .e_offset.15D.1971 + .e_step.16D.1973;
   if (.e_offset.15D.1971 < .e_bound.14D.1972)
-    goto <bb 23>; [INV] [count: INV]
+    goto <bb 23>; [80.01%] [count: INV]
   else
-    goto <bb 24>; [100.00%] [count: INV]
+    goto <bb 24>; [19.99%] [count: INV]

2017-08-06  Tom de Vries  <tom@codesourcery.com>

* omp-expand.c (expand_oacc_for): Add missing edge probability for tile
and element loops.

From-SVN: r250900

Add missing edge probability in expand_oacc_for

 ;;   basic block 10, loop depth 1, freq 0, maybe hot
 ;;    prev block 9, next block 19, flags: (NEW, REACHABLE)
 ;;    pred:       18 (FALSE_VALUE)
 ;;                9 (FALSE_VALUE)
   GIMPLE_NOP
   .chunk_no.7D.1948 = .chunk_no.7D.1948 + 1;
   if (.chunk_no.7D.1948 < .chunk_max.6D.1949)
-    goto <bb 18>; [INV] [count: INV]
+    goto <bb 18>; [80.01%] [count: INV]
   else
-    goto <bb 19>; [100.00%] [count: INV]
+    goto <bb 19>; [19.99%] [count: INV]

2017-08-06  Tom de Vries  <tom@codesourcery.com>

* omp-expand.c (expand_oacc_for): Add missing edge probability for chunk
loop.

From-SVN: r250899

Require nonlocal_goto for gcc.dg/pr78582.c

2017-08-06  Tom de Vries  <tom@codesourcery.com>

* gcc.dg/pr78582.c: Require effective target nonlocal_goto.

From-SVN: r250898

Fix my_memmove in gcc.c-torture/execute/builtins/memops-asm-lib.c

2017-08-06  Tom de Vries  <tom@codesourcery.com>

* gcc.c-torture/execute/builtins/memops-asm-lib.c (my_memmove):  Fix return
type.  Add missing return.

From-SVN: r250897

Fix call arguments mismatch in gcc.dg/torture/pr78218.c

2017-08-06  Tom de Vries  <tom@codesourcery.com>

PR testsuite/81731
* gcc.dg/torture/pr78218.c (struct v): New type, factored type out of
variable 'a' definition.
(b): Remove unused variable.
(gp): New variable.
(check): Add parameter corresponding to call in main, and handle.

From-SVN: r250896

Daily bump.

From-SVN: r250894

c-ada-spec.c (has_static_fields): Look only into variables.

* c-ada-spec.c (has_static_fields): Look only into variables.
(print_constructor): Add TYPE parameter and use it for the name.
(print_destructor): Likewise.
(print_ada_declaration): Adjust to new constructor/destructor names.
Adjust calls to print_constructor and print_destructor.
(print_ada_struct_decl): Do not test TREE_STATIC on FIELD_DECL.
Look only into variables in the final loop.

From-SVN: r250890

Daily bump.

From-SVN: r250889

Regenerate .pot files.

gcc/po:
* gcc.pot: Regenerate.

libcpp/po:
* cpplib.pot: Regenerate.

From-SVN: r250885

re PR c++/79790 ([C++17] ICE class template argument deduction failed)

/cp
2017-08-04  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/79790
* pt.c (do_class_deduction): Handle the case of no viable implicit
deduction guides; simplify the code generating implicit deduction
guides.

/testsuite
2017-08-04  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/79790
* g++.dg/cpp1z/class-deduction42.C: New.

From-SVN: r250883

re PR c++/79790 ([C++17] ICE class template argument deduction failed)

/cp
2017-08-04  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/79790
* pt.c (do_class_deduction): Handle the case of no viable implicit
deduction guides; simplify the code generating implicit deduction
guides.

/testsuite
2017-08-04  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/79790
* g++.dg/cpp1z/class-deduction42.C: New.

From-SVN: r250882

Remove useless floating point casts in comparisons.

2017-08-04  Yury Gribov  <tetra2005@gmail.com>

PR tree-optimization/57371

gcc/
* match.pd: New pattern.

gcc/testsuite/
* c-c++-common/pr57371-1.c: New test.
* c-c++-common/pr57371-2.c: New test.
* c-c++-common/pr57371-3.c: New test.
* c-c++-common/pr57371-4.c: New test.
* gcc.dg/pr57371-5.c: New test.

From-SVN: r250877

builtins-3.c: Remove ISA 3.0 word variant builtin test cases for vec_mule, and vec_mulo.

gcc/testsuite/ChangeLog:

2017-08-04  Carl Love  <cel@us.ibm.com>

* gcc.target/powerpc/builtins-3.c: Remove ISA 3.0 word variant
builtin test cases for vec_mule, and vec_mulo.
* gcc.target/powerpc/builtins-3-p8.c: Add  ISA 3.0 word variant
builtin test cases for vec_mule, and vec_mulo.

From-SVN: r250876

i386: Rewrite check for AVX512 features

Add a new file, avx512-check.h, to check all AVX512 features.  The test
is skipped if any requested AVX512 features are unavailable.

PR target/81590
* gcc.target/i386/avx512-check.h: New file.
* gcc.target/i386/avx5124fmaps-check.h: Removed.
* gcc.target/i386/avx5124vnniw-check.h: Likewise.
* gcc.target/i386/avx512cd-check.h: Likewise.
* gcc.target/i386/avx512ifma-check.h: Likewise.
* gcc.target/i386/avx512vbmi-check.h: Likewise.
* gcc.target/i386/avx512vpopcntdq-check.h: Likewise.
* gcc.target/i386/avx512bw-check.h: Rewrite.
* gcc.target/i386/avx512dq-check.h: Likewise.
* gcc.target/i386/avx512er-check.h: Likewise.
* gcc.target/i386/avx512f-check.h: Likewise.
* gcc.target/i386/avx512vl-check.h: Likewise.
* gcc.target/i386/avx512f-helper.h: Include "avx512-check.h"
only.
(test_512): Removed.
(avx512*_test): Likewise.
* gcc.target/i386/avx512f-pr71559.c (TEST): Undef.

From-SVN: r250875

re PR hsa/81713 (BIT_FIELD_REF produced by Brig FE do not pass new verification)

Fix PR 81713
 * brigfrontend/brig-basic-inst-handler.cc: replace build_int_cst with
   bitsize_int in building BIT_FIELD_REF.
 * brigfrontend/brig-code-entry-handler.cc: likewise.

From-SVN: r250874

runtime: dump registers for Alpha

    Patch from Uros Bizjak.

    Reviewed-on: https://go-review.googlesource.com/53350

From-SVN: r250873

re PR middle-end/81695 (internal compiler error: in size_binop_loc, at fold-const.c:1768)

PR middle-end/81695
* fold-const.c (fold_indirect_ref_1): For ((int *)&a + 4 -> a[1],
perform the computation in offset_int.

* gcc.dg/pr81695.c: New test.

From-SVN: r250871

Pool alignment information for common bases

This patch is a follow-on to the fix for PR81136.  The testcase for that
PR shows that we can (correctly) calculate different base alignments
for two data_references but still tell that their misalignments wrt the
vector size are equal.  This is because we calculate the base alignments
for each dr individually, without looking at the other drs, and in
general the alignment we calculate is only guaranteed if the dr's DR_REF
actually occurs.

This is working as designed, but it does expose a missed opportunity.
We know that if a vectorised loop is reached, all statements in that
loop execute at least once, so it should be safe to pool the alignment
information for all the statements we're vectorising.  The only catch is
that DR_REFs for masked loads and stores only occur if the mask value is
nonzero.  For example, in:

    struct s __attribute__((aligned(32))) {
      int misaligner;
      int array[N];
    };

    int *ptr;
    for (int i = 0; i < n; ++i)
      ptr[i] = c[i] ? ((struct s *) (ptr - 1))->array[i] : 0;

we can only guarantee that ptr points to a "struct s" if at least
one c[i] is true.

This patch adds a DR_IS_CONDITIONAL_IN_STMT flag to record whether
the DR_REF is guaranteed to occur every time that the statement
executes to completion.  It then pools the alignment information
for references that aren't conditional in this sense.

2017-08-04  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
PR tree-optimization/81136
* tree-vectorizer.h: Include tree-hash-traits.h.
(vec_base_alignments): New typedef.
(vec_info): Add a base_alignments field.
(vect_record_base_alignments): Declare.
* tree-data-ref.h (data_reference): Add an is_conditional_in_stmt
field.
(DR_IS_CONDITIONAL_IN_STMT): New macro.
(create_data_ref): Add an is_conditional_in_stmt argument.
* tree-data-ref.c (create_data_ref): Likewise.  Use it to initialize
the is_conditional_in_stmt field.
(data_ref_loc): Add an is_conditional_in_stmt field.
(get_references_in_stmt): Set the is_conditional_in_stmt field.
(find_data_references_in_stmt): Update call to create_data_ref.
(graphite_find_data_references_in_stmt): Likewise.
* tree-ssa-loop-prefetch.c (determine_loop_nest_reuse): Likewise.
* tree-vect-data-refs.c (vect_analyze_data_refs): Likewise.
(vect_record_base_alignment): New function.
(vect_record_base_alignments): Likewise.
(vect_compute_data_ref_alignment): Adjust base_addr and aligned_to
for nested statements even if we fail to compute a misalignment.
Use pooled base alignments for unconditional references.
(vect_find_same_alignment_drs): Compare base addresses instead
of base objects.
(vect_analyze_data_refs_alignment): Call vect_record_base_alignments.
* tree-vect-slp.c (vect_slp_analyze_bb_1): Likewise.

gcc/testsuite/
PR tree-optimization/81136
* gcc.dg/vect/pr81136.c: Add scan test.

From-SVN: r250870

C++-ify vec_info structures

This patch uses new, delete, constructors and desctructors to manage
vec_info.  This includes making ~vec_info free all the data shared
by bb_vec_info and loop_vec_info, whereas previously the code was
duplicated in destroy_bb_vec_info and destroy_loop_vec_info.  This
in turn meant changing the order of:

  FOR_EACH_VEC_ELT (slp_instances, i, instance)
    vect_free_slp_instance (instance);

and:

  gimple_set_uid (stmt, -1);

in destroy_bb_vec_info/~_bb_vec_info, so that now vect_free_slp_instance
could see a uid of -1 as well as 0.  The patch updates vinfo_for_stmt
so that it returns NULL for a uid of -1.

2017-08-04  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
* tree-vectorizer.h (vec_info): Add a constructor and destructor.
Add an explicit name for the enum.  Use auto_vec for slp_instances
and grouped_stores.
(_loop_vec_info): Add a constructor and destructor.  Use auto_vec
for all vectors.
(_bb_vec_info): Add a constructor and destructor.
(vinfo_for_stmt): Return NULL for uids of -1 as well.
(destroy_loop_vec_info): Delete.
(vect_destroy_datarefs): Likewise.
* tree-vectorizer.c (vect_destroy_datarefs): Delete.
(vec_info::vec_info): New function.
(vec_info::~vec_info): Likewise.
(vectorize_loops): Use delete instead of destroy_loop_vec_info.
* tree-parloops.c (gather_scalar_reductions): Use delete instead of
destroy_loop_vec_info.
* tree-vect-loop.c (new_loop_vec_info): Replace with...
(_loop_vec_info::_loop_vec_info): ...this.
(destroy_loop_vec_info): Replace with...
(_loop_vec_info::~_loop_vec_info): ...this.  Unconditionally delete
the stmt_vec_infos.  Leave handling of vec_info information to its
destructor.  Remove explicit vector releases.
(vect_analyze_loop_form): Use new instead of new_loop_vec_info.
(vect_analyze_loop): Use delete instead of destroy_loop_vec_info.
* tree-vect-slp.c (new_bb_vec_info): Replace with...
(_bb_vec_info::_bb_vec_info): ...this.  Don't reserve space in
BB_VINFO_GROUPED_STORES or BB_VINFO_SLP_INSTANCES.
(destroy_bb_vec_info): Replace with...
(_bb_vec_info::~_bb_vec_info): ...this.  Leave handling of vec_info
information to its destructor.
(vect_slp_analyze_bb_1): Use new and delete instead of
new_bb_vec_info and destroy_bb_vec_info.
(vect_slp_bb): Replace 2 calls to destroy_bb_vec_info with a
single delete.

From-SVN: r250869

Use base inequality for some vector alias checks

This patch checks whether two data references x and y cannot
partially overlap and so are independent whenever &x != &y.
We can then use this in the vectoriser to optimise alias checks.

gcc/
2016-08-04  Richard Sandiford  <richard.sandiford@linaro.org>

* hash-traits.h (pair_hash): New struct.
* tree-data-ref.h (data_dependence_relation): Add object_a and
object_b fields.
(DDR_OBJECT_A, DDR_OBJECT_B): New macros.
* tree-data-ref.c (initialize_data_dependence_relation): Initialize
DDR_OBJECT_A and DDR_OBJECT_B.
* tree-vectorizer.h (vec_object_pair): New type.
(_loop_vec_info): Add a check_unequal_addrs field.
(LOOP_VINFO_CHECK_UNEQUAL_ADDRS): New macro.
(LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Return true if there is an
entry in check_unequal_addrs.  Check comp_alias_ddrs instead of
may_alias_ddrs.
* tree-vect-loop.c (destroy_loop_vec_info): Release
LOOP_VINFO_CHECK_UNEQUAL_ADDRS.
(vect_analyze_loop_2): Likewise, when restarting.
(vect_estimate_min_profitable_iters): Estimate the cost of
LOOP_VINFO_CHECK_UNEQUAL_ADDRS.
* tree-vect-data-refs.c: Include tree-hash-traits.h.
(vect_prune_runtime_alias_test_list): Try to handle conflicts
using LOOP_VINFO_CHECK_UNEQUAL_ADDRS, if the data dependence allows.
Count such tests in the final summary.
* tree-vect-loop-manip.c (chain_cond_expr): New function.
(vect_create_cond_for_align_checks): Use it.
(vect_create_cond_for_unequal_addrs): New function.
(vect_loop_versioning): Call it.

gcc/testsuite/
* gcc.dg/vect/vect-alias-check-6.c: New test.

From-SVN: r250868

Handle data dependence relations with different bases

This patch tries to calculate conservatively-correct distance
vectors for two references whose base addresses are not the same.
It sets a new flag DDR_COULD_BE_INDEPENDENT_P if the dependence
isn't guaranteed to occur.

The motivating example is:

  struct s { int x[8]; };
  void
  f (struct s *a, struct s *b)
  {
    for (int i = 0; i < 8; ++i)
      a->x[i] += b->x[i];
  }

in which the "a" and "b" accesses are either independent or have a
dependence distance of 0 (assuming -fstrict-aliasing).  Neither case
prevents vectorisation, so we can vectorise without an alias check.

I'd originally wanted to do the same thing for arrays as well, e.g.:

  void
  f (int a[][8], struct b[][8])
  {
    for (int i = 0; i < 8; ++i)
      a[0][i] += b[0][i];
  }

I think this is valid because C11 6.7.6.2/6 says:

  For two array types to be compatible, both shall have compatible
  element types, and if both size specifiers are present, and are
  integer constant expressions, then both size specifiers shall have
  the same constant value.

So if we access an array through an int (*)[8], it must have type X[8]
or X[], where X is compatible with int.  It doesn't seem possible in
either case for "a[0]" and "b[0]" to overlap when "a != b".

However, as the comment above "if (same_base_p)" explains, GCC is more
forgiving: it supports arbitrary overlap of arrays and allows arrays to
be accessed with different dimensionality.  There are examples of this
in PR50067.  The patch therefore only handles references that end in a
structure field access.

There are two ways of handling these dependences in the vectoriser:
use them to limit VF, or check at runtime as before.  I've gone for
the approach of checking at runtime if we can, to avoid limiting VF
unnecessarily, but falling back to a VF cap when runtime checks aren't
allowed.

The patch tests whether we queued an alias check with a dependence
distance of X and then picked a VF <= X, in which case it's safe to
drop the alias check.  Since vect_prune_runtime_alias_check_list
can be called twice with different VF for the same loop, it's no
longer safe to clear may_alias_ddrs on exit.  Instead we should use
comp_alias_ddrs to check whether versioning is necessary.

2017-08-04  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
* tree-data-ref.h (subscript): Add access_fn field.
(data_dependence_relation): Add could_be_independent_p.
(SUB_ACCESS_FN, DDR_COULD_BE_INDEPENDENT_P): New macros.
(same_access_functions): Move to tree-data-ref.c.
* tree-data-ref.c (ref_contains_union_access_p): New function.
(access_fn_component_p): Likewise.
(access_fn_components_comparable_p): Likewise.
(dr_analyze_indices): Add a reference to access_fn_component_p.
(dump_data_dependence_relation): Use SUB_ACCESS_FN instead of
DR_ACCESS_FN.
(constant_access_functions): Likewise.
(add_other_self_distances): Likewise.
(same_access_functions): Likewise.  (Moved from tree-data-ref.h.)
(initialize_data_dependence_relation): Use XCNEW and remove
explicit zeroing of DDR_REVERSED_P.  Look for a subsequence
of access functions that have the same type.  Allow the
subsequence to end with different bases in some circumstances.
Record the chosen access functions in SUB_ACCESS_FN.
(build_classic_dist_vector_1): Replace ddr_a and ddr_b with
a_index and b_index.  Use SUB_ACCESS_FN instead of DR_ACCESS_FN.
(subscript_dependence_tester_1): Likewise dra and drb.
(build_classic_dist_vector): Update calls accordingly.
(subscript_dependence_tester): Likewise.
* tree-ssa-loop-prefetch.c (determine_loop_nest_reuse): Check
DDR_COULD_BE_INDEPENDENT_P.
* tree-vectorizer.h (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Test
comp_alias_ddrs instead of may_alias_ddrs.
* tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr):
New function.
(vect_analyze_data_ref_dependence): Use it if
DDR_COULD_BE_INDEPENDENT_P, but fall back to using the recorded
distance vectors if that fails.
(dependence_distance_ge_vf): New function.
(vect_prune_runtime_alias_test_list): Use it.  Don't clear
LOOP_VINFO_MAY_ALIAS_DDRS.

gcc/testsuite/
* gcc.dg/vect/vect-alias-check-3.c: New test.
* gcc.dg/vect/vect-alias-check-4.c: Likewise.
* gcc.dg/vect/vect-alias-check-5.c: Likewise.

From-SVN: r250867

re PR middle-end/81705 (UBSAN: yet another false positive)

2017-08-04  Richard Biener  <rguenther@suse.de>

PR middle-end/81705
* fold-const.c (fold_binary_loc): Properly restrict
minus_var0 && minus_var1 case when associating undefined overflow
entities.

* c-c++-common/ubsan/pr81705.c: New testcase.

From-SVN: r250866

Add missing edge probability in simd_clone_adjust

Currently we generate an if with probability set on only one of the two edges:
  <bb 5> [0.00%] [count: INV]:
  _5 = mask.3[iter.6_3];
  if (_5 == 0)
    goto <bb 6>; [INV] [count: INV]
  else
    goto <bb 2>; [100.00%] [count: INV]

Add the missing edge probability, and set the split to unlikely/likely:
  if (_5 == 0)
    goto <bb 6>; [19.99%] [count: INV]
  else
    goto <bb 2>; [80.01%] [count: INV]

2017-08-04  Tom de Vries  <tom@codesourcery.com>

* omp-simd-clone.c (simd_clone_adjust): Add missing edge probability.

From-SVN: r250865

Daily bump.

From-SVN: r250864

rs6000-c.c (altivec_resolve_overloaded_builtin): Don't start diagnostic messages with a capital letter.

[gcc]

2017-08-03  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

* config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin):
Don't start diagnostic messages with a capital letter.
* config/rs6000/rs6000.c (rs6000_option_override_internal):
Likewise.
(rs6000_invalid_builtin): Likewise.
(rs6000_trampoline_init): Likewise.

[gcc/testsuite]

2017-08-03  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

* gcc.target/powerpc/bfp/scalar-cmp-exp-eq-2.c: Adjust for error
messages that used to start with a capital letter.
* gcc.target/powerpc/bfp/scalar-cmp-exp-gt-2.c: Likewise.
* gcc.target/powerpc/bfp/scalar-cmp-exp-lt-2.c: Likewise.
* gcc.target/powerpc/bfp/scalar-cmp-unordered-2.c: Likewise.
* gcc.target/powerpc/bfp/scalar-extract-exp-1.c: Likewise.
* gcc.target/powerpc/bfp/scalar-extract-exp-4.c: Likewise.
* gcc.target/powerpc/bfp/scalar-extract-sig-1.c: Likewise.
* gcc.target/powerpc/bfp/scalar-extract-sig-4.c: Likewise.
* gcc.target/powerpc/bfp/scalar-insert-exp-1.c: Likewise.
* gcc.target/powerpc/bfp/scalar-insert-exp-10.c: Likewise.
* gcc.target/powerpc/bfp/scalar-insert-exp-4.c: Likewise.
* gcc.target/powerpc/bfp/scalar-insert-exp-7.c: Likewise.
* gcc.target/powerpc/bfp/scalar-test-data-class-11.c: Likewise.
* gcc.target/powerpc/bfp/scalar-test-data-class-6.c: Likewise.
* gcc.target/powerpc/bfp/scalar-test-data-class-7.c: Likewise.
* gcc.target/powerpc/bfp/scalar-test-neg-2.c: Likewise.
* gcc.target/powerpc/bfp/scalar-test-neg-3.c: Likewise.
* gcc.target/powerpc/bfp/scalar-test-neg-5.c: Likewise.
* gcc.target/powerpc/bfp/vec-extract-exp-2.c: Likewise.
* gcc.target/powerpc/bfp/vec-extract-exp-3.c: Likewise.
* gcc.target/powerpc/bfp/vec-extract-sig-2.c: Likewise.
* gcc.target/powerpc/bfp/vec-extract-sig-3.c: Likewise.
* gcc.target/powerpc/bfp/vec-insert-exp-2.c: Likewise.
* gcc.target/powerpc/bfp/vec-insert-exp-3.c: Likewise.
* gcc.target/powerpc/bfp/vec-insert-exp-6.c: Likewise.
* gcc.target/powerpc/bfp/vec-insert-exp-7.c: Likewise.
* gcc.target/powerpc/bfp/vec-test-data-class-2.c: Likewise.
* gcc.target/powerpc/bfp/vec-test-data-class-3.c: Likewise.
* gcc.target/powerpc/byte-in-either-range-1.c: Likewise.
* gcc.target/powerpc/byte-in-range-1.c: Likewise.
* gcc.target/powerpc/byte-in-set-1.c: Likewise.
* gcc.target/powerpc/crypto-builtin-2.c: Likewise.
* gcc.target/powerpc/dfp/dtstsfi-1.c: Likewise.
* gcc.target/powerpc/dfp/dtstsfi-11.c: Likewise.
* gcc.target/powerpc/dfp/dtstsfi-16.c: Likewise.
* gcc.target/powerpc/dfp/dtstsfi-21.c: Likewise.
* gcc.target/powerpc/dfp/dtstsfi-26.c: Likewise.
* gcc.target/powerpc/dfp/dtstsfi-31.c: Likewise.
* gcc.target/powerpc/dfp/dtstsfi-36.c: Likewise.
* gcc.target/powerpc/dfp/dtstsfi-41.c: Likewise.
* gcc.target/powerpc/dfp/dtstsfi-46.c: Likewise.
* gcc.target/powerpc/dfp/dtstsfi-51.c: Likewise.
* gcc.target/powerpc/dfp/dtstsfi-56.c: Likewise.
* gcc.target/powerpc/dfp/dtstsfi-6.c: Likewise.
* gcc.target/powerpc/dfp/dtstsfi-61.c: Likewise.
* gcc.target/powerpc/dfp/dtstsfi-66.c: Likewise.
* gcc.target/powerpc/dfp/dtstsfi-71.c: Likewise.
* gcc.target/powerpc/dfp/dtstsfi-76.c: Likewise.
* gcc.target/powerpc/vsu/vec-all-nez-7.c: Likewise.
* gcc.target/powerpc/vsu/vec-any-eqz-7.c: Likewise.
* gcc.target/powerpc/vsu/vec-cmpnez-7.c: Likewise.
* gcc.target/powerpc/vsu/vec-cntlz-lsbb-2.c: Likewise.
* gcc.target/powerpc/vsu/vec-cnttz-lsbb-2.c: Likewise.
* gcc.target/powerpc/vsu/vec-xl-len-12.c: Likewise.
* gcc.target/powerpc/vsu/vec-xlx-7.c: Likewise.
* gcc.target/powerpc/vsu/vec-xrx-7.c: Likewise.
* gcc.target/powerpc/vsu/vec-xst-len-12.c: Likewise.

From-SVN: r250860

re PR go/81617 (mksigtab.sh fails to resolve NSIG with glibc 2.26)

PR go/81617
    libgo: change mksigtab to recognize glibc 2.26 NSIG expression

    Fixes golang/go#21147
    Fixes GCC PR 81617

    Reviewed-on: https://go-review.googlesource.com/52611

From-SVN: r250858

re PR target/81621 (ICE in delete_insn, at cfgrtl.c:167 with s390x cross compiler)

PR target/81621
* bb-reorder.c (pass_partition_blocks::execute): Return TODO_df_finish
after setting changeable df flags.

* gcc.dg/pr81621.c: New test.

From-SVN: r250857

lto-symtab.h (lto_symtab_prevail_decl): Do not use DECL_ABSTRACT_ORIGIN as flag we can end up using that.

2017-08-03  Richard Biener  <rguenther@suse.de>

* lto-symtab.h (lto_symtab_prevail_decl): Do not use
DECL_ABSTRACT_ORIGIN as flag we can end up using that.  Instead
use DECL_LANG_FLAG_0.
(lto_symtab_prevail_decl): Likewise.

From-SVN: r250856

tree-ssa-reassoc.c (should_break_up_subtract): Also break up if the use is in USE - X.

2017-08-03  Richard Biener  <rguenther@suse.de>

* tree-ssa-reassoc.c (should_break_up_subtract): Also break
up if the use is in USE - X.

* gcc.dg/tree-ssa/reassoc-23.c: Adjust to fool early folding
and CSE.

From-SVN: r250855

toplev: avoid recursive emergency_dump_function

* toplev.c (dumpfile.h): New include.
(internal_error_reentered): New static function.  Use it...
(internal_error_function): ...here to handle reentered internal_error.

From-SVN: r250854

re PR sanitizer/81148 (UBSAN: two more false positives)

2017-08-03 Richard Biener  <rguenther@suse.de>

PR middle-end/81148
* fold-const.c (split_tree): Add minus_var and minus_con
arguments, remove unused loc arg.  Never generate NEGATE_EXPRs
here but always use minus_*.
(associate_trees): Assert we never associate with MINUS_EXPR
and NULL first operand.  Do not recurse for PLUS_EXPR operands
when associating as MINUS_EXPR either.
(fold_binary_loc): Track minus_var and minus_con.

* c-c++-common/ubsan/pr81148.c: New testcase.

From-SVN: r250853

Apply finish_options on DECL_FUNCTION_SPECIFIC_OPTIMIZATION for ACCEL_COMPILER

2017-08-03  Tom de Vries  <tom@codesourcery.com>

PR lto/81430
* tree-streamer-in.c (lto_input_ts_function_decl_tree_pointers): If
ACCEL_COMPILER, apply finish_options on
DECL_FUNCTION_SPECIFIC_OPTIMIZATION.

From-SVN: r250852

Error out on nvptx for fpatchable-function-entry

2017-08-03  Tom de Vries  <tom@codesourcery.com>

PR target/81662
* config/nvptx/nvptx.c (nvptx_option_override): Emit sorry if
function_entry_patch_area_size > 0.

* gcc.target/nvptx/patchable_function_entry-default.c: New test.

From-SVN: r250851

re PR driver/81650 (gcc -m32 mishandles -Walloc-size-larger-than=9223372036854775807)

PR driver/81650
* calls.c (alloc_max_size): Use HOST_WIDE_INT_UC (10??)
instead of 10??LU, perform unit multiplication in wide_int,
don't change alloc_object_size_limit if the limit is larger
than SSIZE_MAX.

* gcc.dg/pr81650.c: New test.

From-SVN: r250850

re PR tree-optimization/81655 (new test case gcc.dg/tree-ssa/pr81588.c fails on powerpc64)

PR tree-optimization/81655
PR tree-optimization/81588
* tree-ssa-reassoc.c (optimize_range_tests_var_bound): Handle also
the case when ranges[i].low and high are 1 for unsigned type with
precision 1.

From-SVN: r250849

re PR c++/71440 (ICE on invalid C++ code in instantiate_type, at cp/class.c:8247)

/cp
2017-08-03  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/71440
* typeck.c (build_x_unary_op): Avoid pretty-printing constructor /
destructor as expressions.

/testsuite
2017-08-03  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/71440
* g++.dg/template/crash127.C: New.

From-SVN: r250848

re PR middle-end/81052 (ICE in verify_dominators, at dominance.c:1184)

PR middle-end/81052
* omp-low.c (diagnose_sb_0): Handle flag_openmp_simd like flag_openmp.
(pass_diagnose_omp_blocks::gate): Enable also for flag_openmp_simd.

* c-c++-common/pr81052.c: New test.

From-SVN: r250847

tree-vrp.h: Add include guard.

2017-08-03  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>

* tree-vrp.h: Add include guard.

From-SVN: r250846

Require alias for gcc.dg/pr56727-2.c

2017-08-03  Tom de Vries  <tom@codesourcery.com>

* gcc.dg/pr56727-2.c: Require alias.

From-SVN: r250845

Require alloca for gcc.dg/attr-noipa.c

2017-08-03  Tom de Vries  <tom@codesourcery.com>

* gcc.dg/attr-noipa.c: Require alloca.

From-SVN: r250844

Require label_values for gcc.dg/torture/pr80163.c

2017-08-03  Tom de Vries  <tom@codesourcery.com>

* gcc.dg/torture/pr80163.c: Require label_values.

From-SVN: r250843

Skip fpatchable-function-entry tests for nvptx

2017-08-03  Tom de Vries  <tom@codesourcery.com>

PR target/81662
* c-c++-common/patchable_function_entry-decl.c: Skip for nvptx.
* c-c++-common/patchable_function_entry-default.c: Same.
* c-c++-common/patchable_function_entry-definition.c: Same.

From-SVN: r250842

Daily bump.

From-SVN: r250841

Added Python implementation of mklog.

From-SVN: r250836

Renamed original mklog to mklog.pl.

From-SVN: r250835

compiler: only finalize embedded fields before finalizing methods

    When finalizing the methods of a named struct type, we used to
    finalize all the field types first.  That can fail if the field types
    refer indirectly to the named type.  Change it to just finalize the
    embedded field types first, and the rest of the fields later.

    Fixes golang/go#21253

    Reviewed-on: https://go-review.googlesource.com/52570

From-SVN: r250832

* ChangeLog: Fix my last entry.

From-SVN: r250831

re PR target/81644 (ICE in rtl_verify_bb_insn, BBRO pass duplicates BB that ends with flow control insn)

PR target/81644
* config/i386/i386.md (unspecv): Add UNSPECV_UD2.
(ud2): New insn pattern.
* config/i386/i386.c (ix86_expand_epilogue):
Generate ud2 instead of trap insn.

testsuite/ChangeLog:

PR target/81644
* gcc.target/i386/pr81644.c: New test.

From-SVN: r250830

re PR other/81667 (trunk/gcc/alloc-pool.h:239: possible missing initialiser ?)

PR other/81667
* alloc-pool.h (base_pool_allocator): Initialize m_elt_size.

From-SVN: r250829

Use relative line number in gcc.dg/Walloca-14.c

2017-08-02  Tom de Vries  <tom@codesourcery.com>

* gcc.dg/Walloca-14.c: Use relative line number.

From-SVN: r250828

funcspec-56.inc (no_3dnowa): Properly test "no-3dnowa" target attribute.

* gcc.target/i386/funcspec-56.inc (no_3dnowa): Properly
test "no-3dnowa" target attribute.

From-SVN: r250826

Add PR number.

From-SVN: r250825

* xcoff.c (xcoff_process_linenos): Initialize incl to NULL.

From-SVN: r250824

Add missing probabilities in nvptx_lock{less,full}_update

2017-08-02  Tom de Vries  <tom@codesourcery.com>
    Cesar Philippidis  <cesar@codesourcery.com>

* config/nvptx/nvptx.c (nvptx_lockless_update, nvptx_lockfull_update):
Add missing edge probabilities.

Co-Authored-By: Cesar Philippidis <cesar@codesourcery.com>
From-SVN: r250823

re PR c/81448 (False positive -Werror=multistatement-macros in openssl)

PR c/81448
PR c/81306
* c-warn.c (warn_for_multistatement_macros): Prevent bogus
warnings.  Avoid walking MACRO_MAP_LOCATIONS.

* c-c++-common/Wmultistatement-macros-13.c: New test.

From-SVN: r250822

re PR c/81289 (ICE in libcpp/line-map.c)

PR c/81289
* c-parser.c (c_parser_unary_expression): Use set_error.

* gcc.dg/noncompile/pr81289.c: New test.

From-SVN: r250821

pr25967-1.c: Delete

* gcc.dg/guality/pr25967-1.c: Delete
        * gcc.dg/guality/pr25967-2.c: Delete
        * gcc.dg/guality/pr25967-3.c: Rename to pr25967-1.c.
        * gcc.dg/guality/pr25967-4.c: Rename to pr25967-2.c.

        * gcc.dg/torture/pr25967-1.c: Delete
        * gcc.dg/torture/pr25967-2.c: Delete
        * gcc.dg/torture/pr25967-3.c: Rename to pr25967-1.c.
        * gcc.dg/torture/pr25967-4.c: Rename to pr25967-2.c.

From-SVN: r250820

aarch64.c (aarch64_reinterpret_float_as_int): Correct endianness.

2017-08-02  Tamar Christina  <tamar.christina@arm.com>

* config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
Correct endianness.

From-SVN: r250818

re PR c++/81640 (ICE in lookup_fnfields_slot_nolazy w/ -Wshadow=compatible-local)

PR c++/81640
* call.c (build_user_type_conversion_1): Only call
lookup_fnfields_slot if totype is CLASS_TYPE_P.

* g++.dg/warn/Wshadow-compatible-local-2.C: New test.

From-SVN: r250816

re PR middle-end/79499 (ICE in rtl_verify_bb_insns, at cfgrtl.c:2661)

PR middle-end/79499
* function.c (thread_prologue_and_epilogue_insns): Determine blocks
for find_many_sub_basic_blocks bitmap by looking up BLOCK_FOR_INSN
of first NONDEBUG_INSN_P in each of the split_prologue_seq and
prologue_seq sequences - if any.

* gcc.dg/pr79499.c: New test.

From-SVN: r250814

tree-vect-stmts.c (vectorizable_store): Perform vector extracts via vectors if supported...

2017-08-02  Richard Biener  <rguenther@suse.de>

* tree-vect-stmts.c (vectorizable_store): Perform vector extracts
via vectors if supported, integer extracts via punning if supported
or otherwise vector extracts.

From-SVN: r250813

tree-ssa-pre.c (bitmap_insert_into_set_1): Remove and inline into ...

2017-08-02  Richard Biener  <rguenther@suse.de>

* tree-ssa-pre.c (bitmap_insert_into_set_1): Remove and inline
into ...
(bitmap_insert_into_set): ... this.

From-SVN: r250812

re PR tree-optimization/81633 (Incorrect floating point result with tree vectoriser)

2017-08-02  Richard Biener  <rguenther@suse.de>

PR tree-optimization/81633
Revert
2015-08-17  Alan Hayward  <alan.hayward@arm.com>

PR tree-optimization/71752
* tree-vect-slp.c (vect_get_slp_defs): Handle null operands.

* gcc.dg/vect/pr81633.c: New testcase.

From-SVN: r250811

Daily bump.

From-SVN: r250810

* fr.po: Update.

From-SVN: r250805

[i386] Remove ix86_frame::outlined_save_offset and machine_function::call_ms2sysv_pad_out

ix86_frame::outlined_save_offset isn't used and
machine_function::call_ms2sysv_pad_out is an ineffective strategy.

2017-08-01  Daniel Santos  <daniel.santos@pobox.com>

        * config/i386/i386.h (ix86_frame::outlined_save_offset): Remove field.
        (machine_function::call_ms2sysv_pad_out): Remove field.
        * config/i386/i386.c (xlogue_layout::get_stack_space_used): Modify.
        (ix86_compute_frame_layout): Likewise.

From-SVN: r250803

c-ada-spec.c (has_static_fields): Look only into fields.

* c-ada-spec.c (has_static_fields): Look only into fields.
(dump_generic_ada_node): Small tweak.
(dump_nested_types): Look only into fields.
(print_ada_declaration): Look only into methods.  Small tweak.
(print_ada_struct_decl): Look only into fields.  Use DECL_VIRTUAL_P.

From-SVN: r250802

i386: Add more naked attribute tests

Add some tests for implementing interrupt handlers with naked attribute
and without asm statements.

* gcc.dg/guality/pr25967-3.c: New test.
* gcc.dg/guality/pr25967-4.c: Likewise.
* gcc.dg/torture/pr25967-3.c: Likewise.
* gcc.dg/torture/pr25967-4.c: Likewise.

From-SVN: r250800

i386: Add some naked attribute tests

Add some tests for implementing interrupt handlers with naked attribute.

* gcc.dg/guality/pr25967-1.c: New test.
* gcc.dg/guality/pr25967-2.c: Likewise.
* gcc.dg/torture/pr25967-1.c: Likewise.
* gcc.dg/torture/pr25967-2.c: Likewise.

From-SVN: r250799

c-ada-spec.c (print_generic_ada_decl): Pass correctly-typed constant.

* c-ada-spec.c (print_generic_ada_decl): Pass correctly-typed constant.
(dump_ada_function_declaration): Likewise.
(dump_generic_ada_node): Likewise.
(print_ada_declaration): Add support for const-qualified variables.

From-SVN: r250797

386: Disallow naked attribute with interrupt attribute

gcc/

PR target/81654
* config/i386/i386.c (ix86_set_func_type): Disallow naked
attribute with interrupt attribute.

gcc/testsuite/

PR target/81654
* gcc.target/i386/pr81654.c: New test.

From-SVN: r250793

re PR fortran/79312 (Empty array in assignment not correctly type-checked)

2017-08-01  Thomas König  <tkoenig@gcc.gnu.org>

PR fortran/79312
* intrisic.c (gfc_convert_type_warn):  Only set typespec for
empty array constructors which don't have it already.

2017-08-01  Thomas König  <tkoenig@gcc.gnu.org>

PR fortran/79312
* gfortran.dg/logical_assignment_1.f90:  New test.

From-SVN: r250792

re PR fortran/45435 (Automatically generate C interop interface blocks from C code)

2017-08-01  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/45435
* lang.opt (fc-prototypes): Add option.
* gfortran.h (gfc_typespec): Add interop_kind to struct.
(gfc_dump_c_prototypes): Add prototype.
* decl.c (gfc_match_kind_spec): Copy symbol used for kind to typespec.
* parse.c (gfc_parse_file): Call gfc_dump_prototypes.
* dump-parse-tree.c (gfc_dump_c_prototypes): New function.
(type_return): New enum.
(get_c_type_name): New function.
(write_decl): New function.
(write_type): New function.
(write_variable): New function.
(write_proc): New function.
(write_interop_decl): New function.
* invoke.texi: Document -fc-prototypes.

From-SVN: r250791