platform/upstream/gcc.git
8 years agore PR fortran/71783 (ICE on valid code in gimplify_var_or_parm_decl at gimplify.c...
Thomas Koenig [Sat, 9 Jul 2016 22:18:29 +0000 (22:18 +0000)]
re PR fortran/71783 (ICE on valid code in gimplify_var_or_parm_decl at gimplify.c:1801)

2016-07-09  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/71783
* frontend-passes.c (create_var):  Always allocate a charlen
for character variables.

2016-07-09  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/71783
* gfortran.dg/dependency_46.f90:  New test.

From-SVN: r238192

8 years agoDaily bump.
GCC Administrator [Sat, 9 Jul 2016 00:16:18 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r238191

8 years agore PR fortran/68426 (Simplification of SPREAD with a derived type element is unimplem...
Steven G. Kargl [Fri, 8 Jul 2016 22:24:44 +0000 (22:24 +0000)]
re PR fortran/68426 (Simplification of SPREAD with a derived type element is unimplemented)

2016-07-08  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/68426
* simplify (gfc_simplify_spread): Adjust locus.

From-SVN: r238186

8 years agoFT32: apply unbias to references to RAM symbols.
James Bowman [Fri, 8 Jul 2016 22:11:30 +0000 (22:11 +0000)]
FT32: apply unbias to references to RAM symbols.

The FT32 binutils use a bias to distinguish between RAM and flash
addresses.

This fix adds an ASM_OUTPUT_SYMBOL_REF() that unbiases references to
RAM symbols.

Only references to RAM objects have the bias applied. Flash objects
(that is, objects in ADDR SPACE 1) are not biased, so for these no bias
should be applied. Likewise references in the gdb section need to use
the biased address, so references in debug sections are not unbiased.

gcc/ChangeLog:

2016-07-08  James Bowman  <james.bowman@ftdichip.com>

* config/ft32/ft32.c (ft32_elf_encode_section_info): New function.
* config/ft32/ft32.h (ASM_OUTPUT_SYMBOL_REF): New function.

From-SVN: r238185

8 years agoUse lvalue_p instead of real_lvalue_p.
Jason Merrill [Fri, 8 Jul 2016 21:32:18 +0000 (17:32 -0400)]
Use lvalue_p instead of real_lvalue_p.

* cp-tree.h: Unpoison lvalue_p.
* call.c, class.c, constexpr.c, cvt.c, init.c, lambda.c, pt.c,
tree.c, typeck.c, typeck2.c: Use lvalue_p instead of
real_lvalue_p.

From-SVN: r238183

8 years agoRename lvalue_p to obvalue_p.
Jason Merrill [Fri, 8 Jul 2016 21:32:12 +0000 (17:32 -0400)]
Rename lvalue_p to obvalue_p.

* tree.c (obvalue_p): Rename from lvalue_p.
(lvalue_p): Define for c-common.
* call.c, cp-tree.h, cvt.c, init.c: Adjust.
* typeck.c: Adjust.
(cp_build_addr_expr_1): Remove obsolete code.

From-SVN: r238182

8 years agoRename lvalue_or_rvalue_with_address_p to glvalue_p.
Jason Merrill [Fri, 8 Jul 2016 21:32:06 +0000 (17:32 -0400)]
Rename lvalue_or_rvalue_with_address_p to glvalue_p.

* tree.c (glvalue_p): Rename from lvalue_or_rvalue_with_address_p.
* call.c, cp-tree.h, typeck.c: Adjust.

From-SVN: r238181

8 years ago* lambda.c (maybe_add_lambda_conv_op): Fix null object argument.
Jason Merrill [Fri, 8 Jul 2016 21:31:59 +0000 (17:31 -0400)]
* lambda.c (maybe_add_lambda_conv_op): Fix null object argument.

From-SVN: r238180

8 years agore PR rtl-optimization/71621 (ICE in assign_by_spills, at lra-assigns.c:1417 (error...
Vladimir Makarov [Fri, 8 Jul 2016 20:29:12 +0000 (20:29 +0000)]
re PR rtl-optimization/71621 (ICE in assign_by_spills, at lra-assigns.c:1417 (error: unable to find a register to spill) w/ -O2 -mavx2 -ftree-vectorize)

2016-07-08  Vladimir Makarov  <vmakarov@redhat.com>

PR rtl-optimization/71621
* lra-constraints.c (process_alt_operands): Check combination of
reg class and mode.

2016-07-08  Vladimir Makarov  <vmakarov@redhat.com>

PR rtl-optimization/71621
* gcc.target/i386/pr71621-1.c: New.
* gcc.target/i386/pr71621-2.c: New.

From-SVN: r238178

8 years agoP0145R2: Refining Expression Order for C++ (assignment 2).
Jason Merrill [Fri, 8 Jul 2016 20:25:38 +0000 (16:25 -0400)]
P0145R2: Refining Expression Order for C++ (assignment 2).

* cp-gimplify.c (lvalue_has_side_effects): New.
(cp_gimplify_expr): Implement assignment ordering.

From-SVN: r238177

8 years agoP0145: Refining Expression Order for C++ (-fstrong-eval-order).
Jason Merrill [Fri, 8 Jul 2016 20:25:31 +0000 (16:25 -0400)]
P0145: Refining Expression Order for C++ (-fstrong-eval-order).

gcc/c-family/
* c.opts (-fargs-in-order): Rename to -fstrong-eval-order.
* c-opts.c: Adjust.
gcc/cp/
* call.c (op_is_ordered, build_over_call): Adjust for
-fargs-in-order renaming to -fstrong-eval-order.
* cp-gimplify.c (cp_gimplify_expr): Likewise.

From-SVN: r238176

8 years agoP0145: Refining Expression Order for C++ (assignment)
Jason Merrill [Fri, 8 Jul 2016 20:23:40 +0000 (16:23 -0400)]
P0145: Refining Expression Order for C++ (assignment)

* gimplify.c (initial_rhs_predicate_for): New.
(gimplfy_modify_expr): Gimplify RHS before LHS.

Co-Authored-By: Richard Biener <rguenther@suse.de>
From-SVN: r238175

8 years agoparse.c (matcha): Define.
Cesar Philippidis [Fri, 8 Jul 2016 18:34:55 +0000 (11:34 -0700)]
parse.c (matcha): Define.

gcc/fortran/
* parse.c (matcha): Define.
(decode_oacc_directive): Add spec_only local var and set it.  Use
matcha to parse acc directives except for routine and declare.  Return
ST_GET_FCN_CHARACTERISTICS if a non-declarative directive could be
matched.

gcc/testsuite/
* gfortran.dg/goacc/pr71704.f90: New test.

From-SVN: r238171

8 years ago* testsuite/23_containers/vector/modifiers/insert/aliasing.cc: New.
Jonathan Wakely [Fri, 8 Jul 2016 16:35:10 +0000 (17:35 +0100)]
* testsuite/23_containers/vector/modifiers/insert/aliasing.cc: New.

From-SVN: r238169

8 years agore PR target/71297 (ICE on invalid code in altivec_resolve_overloaded_builtin (rs6000...
Bill Schmidt [Fri, 8 Jul 2016 15:42:47 +0000 (15:42 +0000)]
re PR target/71297 (ICE on invalid code in altivec_resolve_overloaded_builtin (rs6000-c.c:5106) on powerpc64le-linux)

[gcc]

2016-07-08  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

PR target/71297
* config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin):
Allow standard error handling to take over when a wrong number
of arguments is presented to __builtin_vec_ld () or
__builtin_vec_st ().

[gcc/testsuite]

2016-07-08  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

PR target/71297
* gcc.target/powerpc/pr71297.c: New.

From-SVN: r238168

8 years ago[AArch64] Use fmin/fmax for v[min|max]nm{q} intrinsics
Jiong Wang [Fri, 8 Jul 2016 15:26:51 +0000 (15:26 +0000)]
[AArch64] Use fmin/fmax for v[min|max]nm{q} intrinsics

  smin/smax will actually honor quiet NaN.

gcc/
* config/aarch64/aarch64-simd-builtins.def (smax): Remove float
variants.
(smin): Likewise.
(fmax): New entry.
(fmin): Likewise.
* config/aarch64/arm_neon.h (vmaxnm_f32): Use
__builtin_aarch64_fmaxv2sf.
(vmaxnmq_f32): Likewise.
(vmaxnmq_f64): Likewise.
(vminnm_f32): Likewise.
(vminnmq_f32): Likewise.
(vminnmq_f64): Likewise.

gcc/testsuite/

* gcc.target/aarch64/simd/vminmaxnm_1.c: New.

From-SVN: r238166

8 years agore PR target/71806 (PowerPC -mcpu=power9 enables __float128 without an explicit ...
Michael Meissner [Fri, 8 Jul 2016 14:49:37 +0000 (14:49 +0000)]
re PR target/71806 (PowerPC -mcpu=power9 enables __float128 without an explicit -mfloat128)

[gcc]
2016-07-08  Michael Meissner  <meissner@linux.vnet.ibm.com>

PR target/71806
* config/rs6000/rs6000-cpus.def (ISA_3_0_MASKS_SERVER): Do not
enable -mfloat128-hardware by default.
(ISA_3_0_MASKS_IEEE): New macro to give all of the VSX options
that IEEE 128-bit hardware support needs.
* config/rs6000/rs6000.c (rs6000_option_override_internal): If
-mcpu=power9 -mfloat128, enable -mfloat128-hardware by default.
Use ISA_3_0_MASKS_IEEE as the set of options that IEEE 128-bit
floating point requires.
* doc/invoke.texi (RS/6000 and PowerPC Options): Document
-mfloat128 and -mfloat128-hardware changes.

[gcc/testsuite]
2016-07-08  Michael Meissner  <meissner@linux.vnet.ibm.com>

PR target/71806
* gcc.target/powerpc/p9-lxvx-stxvx-3.c: Add -mfloat128 option.

From-SVN: r238164

8 years agore PR debug/71667 (ICE in as_a, at is-a.h:192 w/ -g -O2 -ftree-vectorize)
Alan Hayward [Fri, 8 Jul 2016 08:50:24 +0000 (08:50 +0000)]
re PR debug/71667 (ICE in as_a, at is-a.h:192 w/ -g -O2 -ftree-vectorize)

2016-07-08  Alan Hayward  <alan.hayward@arm.com>

gcc/
PR tree-optimization/71667
* tree-vect-loop.c (vectorizable_live_operation): ignore DEBUG stmts

testsuite/
PR tree-optimization/71667
* gcc.dg/vect/pr71667.c: New

From-SVN: r238160

8 years agoEnhance documentation of Wundefined-do-loop
Martin Liska [Fri, 8 Jul 2016 08:39:56 +0000 (10:39 +0200)]
Enhance documentation of Wundefined-do-loop

* invoke.texi (Wundefined-do-loop): Enhance documentation.

From-SVN: r238159

8 years agoDo not consider COMPLEX_TYPE as fold_convertible_p
Martin Liska [Fri, 8 Jul 2016 07:52:03 +0000 (09:52 +0200)]
Do not consider COMPLEX_TYPE as fold_convertible_p

PR middle-end/71606
* fold-const.c (fold_convertible_p): As COMPLEX_TYPE
folding produces SAVE_EXPRs, thus return false for the type.
* gcc.dg/torture/pr71606.c: New test.

From-SVN: r238157

8 years agore PR fortran/71764 (ICE in gfc_trans_structure_assign)
Jerry DeLisle [Fri, 8 Jul 2016 04:36:16 +0000 (04:36 +0000)]
re PR fortran/71764 (ICE in gfc_trans_structure_assign)

2016-07-07  Jerry DeLisle  <jvdelisle@gcc.gnu.org>

PR fortran/71764
* trans-expr.c (gfc_trans_structure_assign): Remove assert.

* gfortran.dg/pr71764.f90: New test.

From-SVN: r238156

8 years agoDaily bump.
GCC Administrator [Fri, 8 Jul 2016 00:16:21 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r238155

8 years agore PR c++/70869 (internal compiler error: Segmentation fault on array of pointer...
Jakub Jelinek [Thu, 7 Jul 2016 18:45:43 +0000 (20:45 +0200)]
re PR c++/70869 (internal compiler error: Segmentation fault on array of pointer to function members)

PR c++/70869
PR c++/71054
* cp-gimplify.c (cp_genericize_r): For DECL_EXPR for non-static
artificial vars, genericize their initializers.

* g++.dg/cpp0x/pr70869.C: New test.
* g++.dg/cpp0x/pr71054.C: New test.

Co-Authored-By: Kai Tietz <ktietz70@googlemail.com>
From-SVN: r238124

8 years ago* g++.dg/debug/pr71432.C: Fail on AIX.
David Edelsohn [Thu, 7 Jul 2016 17:59:54 +0000 (17:59 +0000)]
* g++.dg/debug/pr71432.C: Fail on AIX.

From-SVN: r238122

8 years agoUpdate libstdc++ status docs
Jonathan Wakely [Thu, 7 Jul 2016 16:35:43 +0000 (17:35 +0100)]
Update libstdc++ status docs

* doc/xml/manual/status_cxx2014.xml: Update LFTS status table.
* doc/html/*: Regenerate.

From-SVN: r238120

8 years ago[multiple changes]
Arnaud Charlet [Thu, 7 Jul 2016 13:20:30 +0000 (15:20 +0200)]
[multiple changes]

2016-07-07  Ed Schonberg  <schonberg@adacore.com>

* exp_ch6.adb (Expand_Internal_Init_Call): Subsidiary procedure
to Expand_Protected_ Subprogram_Call, to handle properly a
call to a protected function that provides the initialization
expression for a private component of the same protected type.
* sem_ch9.adb (Analyze_Protected_Definition): Layout must be
applied to itypes generated for a private operation of a protected
type that has a formal of an anonymous access to subprogram,
because these itypes have no freeze nodes and are frozen in place.
* sem_ch4.adb (Analyze_Selected_Component): If prefix is a
protected type and it is not a current instance, do not examine
the first private component of the type.

2016-07-07  Arnaud Charlet  <charlet@adacore.com>

* exp_imgv.adb, g-dynhta.adb, s-regexp.adb, s-fatgen.adb, s-poosiz.adb:
Minor removal of extra whitespace.
* einfo.ads: minor removal of repeated "as" in comment

2016-07-07  Vadim Godunko  <godunko@adacore.com>

* adaint.c: Complete previous change.

From-SVN: r238117

8 years ago[multiple changes]
Arnaud Charlet [Thu, 7 Jul 2016 13:17:51 +0000 (15:17 +0200)]
[multiple changes]

2016-07-07  Vadim Godunko  <godunko@adacore.com>

* adainit.h, adainit.c (__gnat_is_read_accessible_file): New
subprogram.
(__gnat_is_write_accessible_file): New subprogram.
* s-os_lib.ads, s-os_lib.adb (Is_Read_Accessible_File): New subprogram.
(Is_Write_Accessible_File): New subprogram.

2016-07-07  Justin Squirek  <squirek@adacore.com>

* sem_ch12.adb (Install_Body): Minor refactoring in the order
of local functions.
(In_Same_Scope): Change loop condition to be more expressive.

From-SVN: r238116

8 years ago[multiple changes]
Arnaud Charlet [Thu, 7 Jul 2016 13:16:05 +0000 (15:16 +0200)]
[multiple changes]

2016-07-07  Gary Dismukes  <dismukes@adacore.com>

* sem_ch3.adb, sem_prag.adb, sem_prag.ads, prj-ext.adb, freeze.adb,
sem_attr.adb: Minor reformatting, fix typos.

2016-07-07  Justin Squirek  <squirek@adacore.com>

* sem_ch12.adb (In_Same_Scope): Created this function to check
a generic package definition against an instantiation for scope
dependancies.
(Install_Body): Add function In_Same_Scope and
amend conditional in charge of delaying the package instance.
(Is_In_Main_Unit): Add guard to check if parent is present in
assignment of Current_Unit.

From-SVN: r238115

8 years agoOptimize fortran loops with +-1 step.
Martin Liska [Thu, 7 Jul 2016 13:15:39 +0000 (15:15 +0200)]
Optimize fortran loops with +-1 step.

* gfortran.dg/do_1.f90: Remove a corner case that triggers
an undefined behavior.
* gfortran.dg/do_3.F90: Likewise.
* gfortran.dg/do_check_11.f90: New test.
* gfortran.dg/do_check_12.f90: New test.
* gfortran.dg/do_corner_warn.f90: New test.
* lang.opt (Wundefined-do-loop): New option.
        * resolve.c (gfc_resolve_iterator): Warn for Wundefined-do-loop.
(gfc_trans_simple_do): Generate a c-style loop.
(gfc_trans_do): Fix GNU coding style.
* invoke.texi: Mention the new warning.

From-SVN: r238114

8 years agosem_ch6.adb (Analyze_Subprogram_Body_Helper): Remove redundant test, adjust comments...
Eric Botcazou [Thu, 7 Jul 2016 13:12:55 +0000 (13:12 +0000)]
sem_ch6.adb (Analyze_Subprogram_Body_Helper): Remove redundant test, adjust comments and formatting.

2016-07-07  Eric Botcazou  <ebotcazou@adacore.com>

* sem_ch6.adb (Analyze_Subprogram_Body_Helper): Remove redundant test,
adjust comments and formatting.
* sem_prag.adb (Inlining_Not_Possible): Do not test Front_End_Inlining
here but...
(Make_Inline): ...here before calling Inlining_Not_Possible instead.
(Set_Inline_Flags): Remove useless test.
(Analyze_Pragma) <Pragma_Inline>: Add comment about -gnatn switch.

From-SVN: r238113

8 years agoAdd PRED_FORTRAN_LOOP_PREHEADER to DO loops with step
Martin Liska [Thu, 7 Jul 2016 13:11:05 +0000 (15:11 +0200)]
Add PRED_FORTRAN_LOOP_PREHEADER to DO loops with step

* trans-stmt.c (gfc_trans_do): Add expect builtin for DO
loops with step bigger than +-1.
* gfortran.dg/predict-1.f90: Ammend the test.
* gfortran.dg/predict-2.f90: Likewise.

From-SVN: r238112

8 years ago[multiple changes]
Arnaud Charlet [Thu, 7 Jul 2016 13:05:08 +0000 (15:05 +0200)]
[multiple changes]

2016-07-07  Ed Schonberg  <schonberg@adacore.com>

* sem_prag.ads, sem_prag.adb (Build_Classwide_Expression): Include
overridden operation as parameter, in order to map formals of
the overridden and overring operation properly prior to rewriting
the inherited condition.
* freeze.adb (Check_Inherited_Cnonditions): Change call to
Build_Class_Wide_Expression accordingly.  In Spark_Mode, add
call to analyze the contract of the parent operation, prior to
mapping formals between operations.

2016-07-07  Arnaud Charlet  <charlet@adacore.com>

* adabkend.adb (Scan_Back_End_Switches): Ignore -o/-G switches
as done in back_end.adb.
(Scan_Compiler_Args): Remove special case for CodePeer/SPARK, no longer
needed, and prevents proper handling of multi-unit sources.

2016-07-07  Thomas Quinot  <quinot@adacore.com>

* g-sechas.adb, g-sechas.ads (GNAT.Secure_Hashes.H): Add Hash_Stream
type with Write primitive calling Update on the underlying context
(and dummy Read primitive raising P_E).

2016-07-07  Thomas Quinot  <quinot@adacore.com>

* sem_ch13.adb: Minor reformatting.

From-SVN: r238111

8 years ago[multiple changes]
Arnaud Charlet [Thu, 7 Jul 2016 13:02:31 +0000 (15:02 +0200)]
[multiple changes]

2016-07-07  Thomas Quinot  <quinot@adacore.com>

* g-socket.ads: Document performance consideration for stream
wrapper.

2016-07-07  Arnaud Charlet  <charlet@adacore.com>

* osint-c.ads (Set_File_Name): Clarify spec.

From-SVN: r238110

8 years agofreeze.adb: Reenable code.
Eric Botcazou [Thu, 7 Jul 2016 13:00:54 +0000 (13:00 +0000)]
freeze.adb: Reenable code.

2016-07-07  Eric Botcazou  <ebotcazou@adacore.com>

* freeze.adb: Reenable code.

From-SVN: r238109

8 years agominor reformatting.
Arnaud Charlet [Thu, 7 Jul 2016 12:59:19 +0000 (14:59 +0200)]
minor reformatting.

From-SVN: r238107

8 years agosem_ch6.adb (Process_Formals): Set ghost flag on formal entities of ghost subprograms.
Yannick Moy [Thu, 7 Jul 2016 12:59:06 +0000 (12:59 +0000)]
sem_ch6.adb (Process_Formals): Set ghost flag on formal entities of ghost subprograms.

2016-07-07  Yannick Moy  <moy@adacore.com>

* sem_ch6.adb (Process_Formals): Set ghost flag
on formal entities of ghost subprograms.
* ghost.adb (Check_Ghost_Context.Is_OK_Ghost_Context): Accept ghost
entities in use type clauses.

From-SVN: r238106

8 years agoPrevent LTO wrappers to process a recursive execution
Martin Liska [Thu, 7 Jul 2016 12:03:39 +0000 (14:03 +0200)]
Prevent LTO wrappers to process a recursive execution

* file-find.c (remove_prefix): New function.
* file-find.h (remove_prefix): Declare the function.
* gcc-ar.c (main): Skip a folder of the wrapper if
a wrapped binary would point to the same file.

From-SVN: r238089

8 years agotree-scalar-evolution.c (iv_can_overflow_p): export.
Jan Hubicka [Thu, 7 Jul 2016 11:50:55 +0000 (13:50 +0200)]
tree-scalar-evolution.c (iv_can_overflow_p): export.

* tree-scalar-evolution.c (iv_can_overflow_p): export.
* tree-scalar-evolution.h (iv_can_overflow_p): Declare.
* tree-ssa-loop-ivopts.c (alloc_iv): Use it.

From-SVN: r238088

8 years agore PR ipa/71624 ([CHKP] internal compiler error: in duplicate_thunk_for_node)
Ilya Enkovich [Thu, 7 Jul 2016 11:45:11 +0000 (11:45 +0000)]
re PR ipa/71624 ([CHKP] internal compiler error: in duplicate_thunk_for_node)

gcc/

PR ipa/71624
* ipa-inline-analysis.c (compute_inline_parameters): Set
local.can_change_signature to false for intrumentation
thunk callees.

gcc/testsuite/

PR ipa/71624
* g++.dg/pr71624.C: New test.

From-SVN: r238086

8 years agoarm.h (TARGET_USE_MOVT): Check MOVT/MOVW availability with TARGET_HAVE_MOVT.
Thomas Preud'homme [Thu, 7 Jul 2016 08:54:59 +0000 (08:54 +0000)]
arm.h (TARGET_USE_MOVT): Check MOVT/MOVW availability with TARGET_HAVE_MOVT.

2016-07-07  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    gcc/
    * config/arm/arm.h (TARGET_USE_MOVT): Check MOVT/MOVW availability
    with TARGET_HAVE_MOVT.
    (TARGET_HAVE_MOVT): Define.
    * config/arm/arm.c (const_ok_for_op): Check MOVT/MOVW
    availability with TARGET_HAVE_MOVT.
    * config/arm/arm.md (arm_movt): Use TARGET_HAVE_MOVT to check MOVT
    availability.
    (addsi splitter): Use TARGET_THUMB && TARGET_HAVE_MOVT rather than
    TARGET_THUMB2.
    (symbol_refs movsi splitter): Remove TARGET_32BIT check.
    (arm_movtas_ze): Use TARGET_HAVE_MOVT to check MOVT availability.
    * config/arm/constraints.md (define_constraint "j"): Use
    TARGET_HAVE_MOVT to check MOVT availability.

From-SVN: r238083

8 years agoarm-protos.h: Reindent FL_FOR_* macro definitions.
Thomas Preud'homme [Thu, 7 Jul 2016 08:54:50 +0000 (08:54 +0000)]
arm-protos.h: Reindent FL_FOR_* macro definitions.

2016-07-07  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    gcc/
    * config/arm/arm-protos.h: Reindent FL_FOR_* macro definitions.

From-SVN: r238082

8 years agoarm-arches.def (armv8-m.base): Define new architecture.
Thomas Preud'homme [Thu, 7 Jul 2016 08:54:40 +0000 (08:54 +0000)]
arm-arches.def (armv8-m.base): Define new architecture.

2016-07-07  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    gcc/
    * config/arm/arm-arches.def (armv8-m.base): Define new architecture.
    (armv8-m.main): Likewise.
    (armv8-m.main+dsp): Likewise.
    * config/arm/arm-protos.h (FL_FOR_ARCH8M_BASE): Define.
    (FL_FOR_ARCH8M_MAIN): Likewise.
    * config/arm/arm-tables.opt: Regenerate.
    * config/arm/bpabi.h: Add armv8-m.base, armv8-m.main and
    armv8-m.main+dsp to BE8_LINK_SPEC.
    * config/arm/arm.h (TARGET_HAVE_LDACQ): Exclude ARMv8-M.
    (enum base_architecture): Add BASE_ARCH_8M_BASE and BASE_ARCH_8M_MAIN.
    * config/arm/arm.c (arm_arch_name): Increase size to work with ARMv8-M
    Baseline and Mainline.
    (arm_option_override_internal): Also disable arm_restrict_it when
    !arm_arch_notm.  Update comment for -munaligned-access to also cover
    ARMv8-M Baseline.
    (arm_file_start): Increase buffer size for printing architecture name.
    * doc/invoke.texi: Document architectures armv8-m.base, armv8-m.main
    and armv8-m.main+dsp.
    (mno-unaligned-access): Clarify that this is disabled by default for
    ARMv8-M Baseline architectures as well.

    gcc/testsuite/
    * lib/target-supports.exp: Generate add_options_for_arm_arch_FUNC and
    check_effective_target_arm_arch_FUNC_multilib for ARMv8-M Baseline and
    ARMv8-M Mainline architectures.

    libgcc/
    * config/arm/lib1funcs.S (__ARM_ARCH__): Define to 8 for ARMv8-M.

From-SVN: r238081

8 years agolib1funcs.S (HAVE_ARM_CLZ): Define for ARMv6* or later and ARMv5t* rather than for...
Thomas Preud'homme [Thu, 7 Jul 2016 08:54:28 +0000 (08:54 +0000)]
lib1funcs.S (HAVE_ARM_CLZ): Define for ARMv6* or later and ARMv5t* rather than for a fixed list of...

2016-07-07  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    libgcc/
    * config/arm/lib1funcs.S (HAVE_ARM_CLZ): Define for ARMv6* or later
    and ARMv5t* rather than for a fixed list of architectures.

From-SVN: r238080

8 years agoelf.h: Use __ARM_ARCH_ISA_THUMB and __ARM_ARCH_ISA_ARM to decide whether to prevent...
Thomas Preud'homme [Thu, 7 Jul 2016 08:54:18 +0000 (08:54 +0000)]
elf.h: Use __ARM_ARCH_ISA_THUMB and __ARM_ARCH_ISA_ARM to decide whether to prevent...

2016-07-07  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    gcc/
    * config/arm/elf.h: Use __ARM_ARCH_ISA_THUMB and __ARM_ARCH_ISA_ARM to
    decide whether to prevent some libgcc routines being included for some
    multilibs rather than __ARM_ARCH_6M__ and add comment to indicate the
    link between this condition and the one in
    libgcc/config/arm/lib1func.S.

    gcc/testsuite/
    * lib/target-supports.exp (check_effective_target_arm_cortex_m): Use
    __ARM_ARCH_ISA_ARM to test for Cortex-M devices.

    libgcc/
    * config/arm/bpabi-v6m.S: Clarify what architectures is the
    implementation suitable for.
    * config/arm/lib1funcs.S (__prefer_thumb__): Define among other cases
    for all Thumb-1 only targets.
    (NOT_ISA_TARGET_32BIT): Define for Thumb-1 only targets.
    (THUMB_LDIV0): Test for NOT_ISA_TARGET_32BIT rather than
    __ARM_ARCH_6M__.
    (EQUIV): Likewise.
    (ARM_FUNC_ALIAS): Likewise.
    (umodsi3): Add check to __ARM_ARCH_ISA_THUMB != 1 to guard the idiv
    version.
    (modsi3): Likewise.
    (clzsi2): Test for NOT_ISA_TARGET_32BIT rather than __ARM_ARCH_6M__.
    (clzdi2): Likewise.
    (ctzsi2): Likewise.
    (L_interwork_call_via_rX): Test for __ARM_ARCH_ISA_ARM rather than
    __ARM_ARCH_6M__ in guard for checking whether it is defined.
    (final includes): Test for NOT_ISA_TARGET_32BIT rather than
    __ARM_ARCH_6M__ and add comment to indicate the connection between
    this condition and the one in gcc/config/arm/elf.h.
    * config/arm/libunwind.S: Test for __ARM_ARCH_ISA_THUMB and
    __ARM_ARCH_ISA_ARM rather than __ARM_ARCH_6M__.
    * config/arm/t-softfp: Likewise.

From-SVN: r238079

8 years agotree-ssa-pre.c: Include alias.h.
Richard Biener [Thu, 7 Jul 2016 07:43:35 +0000 (07:43 +0000)]
tree-ssa-pre.c: Include alias.h.

2016-07-07  Richard Biener  <rguenther@suse.de>

* tree-ssa-pre.c: Include alias.h.
(compute_avail): If we have multiple VN_REFERENCEs with the
same hashtable entry adjust that to make it a valid replacement
for all of them with respect to alignment and aliasing
when doing insertion.
* tree-ssa-sccvn.h (vn_reference_operands_for_lookup): Declare.
* tree-ssa-sccvn.c (vn_reference_operands_for_lookup): New function.

From-SVN: r238078

8 years agors6000: Make the ctr* patterns allow ints in vector regs (PR71763)
Segher Boessenkool [Thu, 7 Jul 2016 03:09:03 +0000 (05:09 +0200)]
rs6000: Make the ctr* patterns allow ints in vector regs (PR71763)

Similar to PR70098, which is about integers in floating point registers,
we can have the completely analogous problem with vector registers as well
now that we allow integers in vector registers.  So, this patch solves it
in the same way.  This only works for targets with direct move.

To recap: register allocation can decide to put an integer mode value in
a floating point or vector register.  If that register is used in a bd*z
instruction, which is a jump instruction, reload can not do an output
reload on it (it does not do output reloads on any jump insns), so the
float or vector register will remain, and we have to allow it here or
recog will ICE.  Later on we will split this to valid instructions,
including a move from that fp/vec register to an int register; it is this
move that will still fail (PR70098) if we do not have direct move enabled.

PR target/70098
PR target/71763
* config/rs6000/rs6000.md (*ctr<mode>_internal1, *ctr<mode>_internal2,
*ctr<mode>_internal5, *ctr<mode>_internal6): Add *wi to the output
constraint.

gcc/testsuite/
PR target/70098
PR target/71763
* gcc.target/powerpc/pr71763.c: New file.

From-SVN: r238076

8 years agoDaily bump.
GCC Administrator [Thu, 7 Jul 2016 00:16:21 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r238073

8 years agoremove unused CTOR_LISTS_DEFINED_EXTERNALLY macro
Trevor Saunders [Wed, 6 Jul 2016 23:55:52 +0000 (23:55 +0000)]
remove unused CTOR_LISTS_DEFINED_EXTERNALLY macro

The last target to use this was i386-interix, so since that is gone we
don't need this anymore.

libgcc/ChangeLog:

2016-07-06  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* libgcc2.c (SYMBOL__MAIN): Remove checks for
CTOR_LISTS_DEFINED_EXTERNALLY.

From-SVN: r238067

8 years agomake side_effects a vec<rtx>
Trevor Saunders [Wed, 6 Jul 2016 23:53:20 +0000 (23:53 +0000)]
make side_effects a vec<rtx>

gcc/ChangeLog:

2016-07-06  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* var-tracking.c (struct adjust_mem_data): Make side_effects a vector.
(adjust_mems): Adjust.
(adjust_insn): Likewise.
(prepare_call_arguments): Likewise.

From-SVN: r238066

8 years agomake stores rtx_insn_list a vec
Trevor Saunders [Wed, 6 Jul 2016 23:53:15 +0000 (23:53 +0000)]
make stores rtx_insn_list a vec

gcc/ChangeLog:

2016-07-06  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* gcse.c (struct ls_expr): Make stores field a vector.
(ldst_entry): Adjust.
(free_ldst_entry): Likewise.
(print_ldst_list): Likewise.
(compute_ld_motion_mems): Likewise.
(update_ld_motion_stores): Likewise.

From-SVN: r238065

8 years agoremove unused loads rtx_insn_list
Trevor Saunders [Wed, 6 Jul 2016 23:53:10 +0000 (23:53 +0000)]
remove unused loads rtx_insn_list

gcc/ChangeLog:

2016-07-06  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* gcse.c (struct ls_expr): Remove loads field.
(ldst_entry): Adjust.
(free_ldst_entry): Likewise.
(print_ldst_list): Likewise.
(compute_ld_motion_mems): Likewise.

From-SVN: r238064

8 years agomake antic_stores a vec<rtx_insn *>
Trevor Saunders [Wed, 6 Jul 2016 23:53:04 +0000 (23:53 +0000)]
make antic_stores a vec<rtx_insn *>

gcc/ChangeLog:

2016-07-06  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* store-motion.c (struct st_expr): Make antic_stores a vector.
(st_expr_entry): Adjust.
(free_st_expr_entry): Likewise.
(print_store_motion_mems): Likewise.
(find_moveable_store): Likewise.
(compute_store_table): Likewise.
(remove_reachable_equiv_notes): Likewise.
(replace_store_insn): Likewise.
(build_store_vectors): Likewise.

From-SVN: r238063

8 years agoImplement std::any.
Ville Voutilainen [Wed, 6 Jul 2016 22:33:20 +0000 (01:33 +0300)]
Implement std::any.

* include/Makefile.am: Add any and c++17_warning.h to exported headers.
* include/Makefile.in: Likewise.
* include/std/any: New.
* testsuite/20_util/any/assign/1.cc: Likewise.
* testsuite/20_util/any/assign/2.cc: Likewise.
* testsuite/20_util/any/assign/self.cc: Likewise.
* testsuite/20_util/any/cons/1.cc: Likewise.
* testsuite/20_util/any/cons/2.cc: Likewise.
* testsuite/20_util/any/cons/aligned.cc: Likewise.
* testsuite/20_util/any/cons/nontrivial.cc: Likewise.
* testsuite/20_util/any/misc/any_cast.cc: Likewise.
* testsuite/20_util/any/misc/any_cast_neg.cc: Likewise.
* testsuite/20_util/any/misc/any_cast_no_rtti.cc: Likewise.
* testsuite/20_util/any/misc/swap.cc: Likewise.
* testsuite/20_util/any/modifiers/1.cc: Likewise.
* testsuite/20_util/any/observers/type.cc: Likewise.
* testsuite/20_util/any/typedefs.cc: Likewise.

From-SVN: r238061

8 years agoAdd a new header for diagnosing the use of C++17 facilities in pre-C++17 modes.
Ville Voutilainen [Wed, 6 Jul 2016 18:07:57 +0000 (21:07 +0300)]
Add a new header for diagnosing the use of C++17 facilities in pre-C++17 modes.

From-SVN: r238058

8 years agoescape: Implement tag phase.
Ian Lance Taylor [Wed, 6 Jul 2016 17:34:03 +0000 (17:34 +0000)]
escape: Implement tag phase.

    Adds notes to function parameters which summarize the escape of that
    parameter with respect to the function's scope.

    Reviewed-on: https://go-review.googlesource.com/18443

From-SVN: r238057

8 years ago[ARM] Add support for some ARMv8-A cores to driver-arm.c
Kyrylo Tkachov [Wed, 6 Jul 2016 16:22:47 +0000 (16:22 +0000)]
[ARM] Add support for some ARMv8-A cores to driver-arm.c

* config/arm/driver-arm.c (arm_cpu_table): Add entries for cortex-a32,
cortex-a35, cortex-a53, cortex-a57, cortex-a72, cortex-a73.

From-SVN: r238056

8 years agore PR tree-optimization/71518 (wrong code at -O3 on x86_64-linux-gnu in 64-bit mode...
Yuri Rumyantsev [Wed, 6 Jul 2016 14:37:26 +0000 (14:37 +0000)]
re PR tree-optimization/71518 (wrong code at -O3 on x86_64-linux-gnu in 64-bit mode (not in 32-bit mode))

gcc/

2016-07-06  Yuri Rumyantsev  <ysrumyan@gmail.com>

PR tree-optimization/71518
* tree-vect-data-refs.c (vect_compute_data_ref_alignment): Adjust
misalign also for outer loops with negative step.

gcc/testsuite/

2016-07-06  Yuri Rumyantsev  <ysrumyan@gmail.com>

PR tree-optimization/71518
        * gcc.dg/pr71518.c: New test.

From-SVN: r238055

8 years agosem_ch6.adb (Check_Inline_Pragma): if the subprogram has no spec then move its aspect...
Javier Miranda [Wed, 6 Jul 2016 13:45:55 +0000 (13:45 +0000)]
sem_ch6.adb (Check_Inline_Pragma): if the subprogram has no spec then move its aspects to the internally built...

2016-07-06  Javier Miranda  <miranda@adacore.com>

* sem_ch6.adb (Check_Inline_Pragma): if the subprogram has no spec
then move its aspects to the internally built subprogram spec.

From-SVN: r238052

8 years agosem_ch6.adb (Analyze_Expression_Function): Mark body of expression function as ghost...
Yannick Moy [Wed, 6 Jul 2016 13:41:20 +0000 (13:41 +0000)]
sem_ch6.adb (Analyze_Expression_Function): Mark body of expression function as ghost if needed when created.

2016-07-06  Yannick Moy  <moy@adacore.com>

* sem_ch6.adb (Analyze_Expression_Function): Mark body of
expression function as ghost if needed when created.
* sem_prag.adb (Analyze_Pragma.Process_Inline.Set_Inline_Flags):
Remove special case.

From-SVN: r238051

8 years ago[multiple changes]
Arnaud Charlet [Wed, 6 Jul 2016 13:38:37 +0000 (15:38 +0200)]
[multiple changes]

2016-07-06  Arnaud Charlet  <charlet@adacore.com>

* lib.adb (Check_Same_Extended_Unit): Complete previous change.
* sem_intr.adb (Errint): New parameter Relaxed. Refine previous
change to only disable errors selectively.
* sem_util.adb: minor style fix in object declaration

2016-07-06  Yannick Moy  <moy@adacore.com>

* sem_warn.adb (Check_Infinite_Loop_Warning.Find_Var): Special case a
call to a volatile function, so that it does not lead to a warning in
that case.

2016-07-06  Hristian Kirtchev  <kirtchev@adacore.com>

* sem_ch12.adb, sem_ch4.adb, sem_ch6.adb: Minor reformatting.

2016-07-06  Hristian Kirtchev  <kirtchev@adacore.com>

* gnat1drv.adb: Code clean up. Do not emit any
code generation errors when the unit is ignored Ghost.

2016-07-06  Ed Schonberg  <schonberg@adacore.com>

* sem_eval.adb (Check_Non_Static_Context): If the expression
is a real literal of a floating point type that is part of a
larger expression and is not a static expression, transform it
into a machine number now so that the rest of the computation,
even if other components are static, is not evaluated with
extra precision.

2016-07-06  Javier Miranda  <miranda@adacore.com>

* sem_ch13.adb (Freeze_Entity_Checks): Undo previous patch and move the
needed functionality to Analyze_Freeze_Generic_Entity.
(Analyze_Freeze_Generic_Entity): If the entity is not already frozen
and has delayed aspects then analyze them.

2016-07-06  Yannick Moy  <moy@adacore.com>

* sem_prag.adb (Analyze_Pragma.Process_Inline.Set_Inline_Flags):
Special case for unanalyzed body entity of ghost expression function.

From-SVN: r238050

8 years agoImplement LWG 2451, optional<T> should 'forward' T's implicit conversions.
Ville Voutilainen [Wed, 6 Jul 2016 13:26:10 +0000 (16:26 +0300)]
Implement LWG 2451, optional<T> should 'forward' T's implicit conversions.

Implement LWG 2451, optional<T> should 'forward' T's
implicit conversions.
* include/experimental/optional (__is_optional_impl, __is_optional):
New.
(optional()): Make constexpr and default.
(optional(_Up&&), optional(const optional<_Up>&),
optional(optional<_Up>&& __t): New.
(operator=(_Up&&)): Constrain.
(operator=(const optional<_Up>&), operator=(optional<_Up>&&)): New.
* testsuite/experimental/optional/cons/value.cc:
Add tests for the functionality added by LWG 2451.
* testsuite/experimental/optional/cons/value_neg.cc: New.

From-SVN: r238049

8 years agoThis patch improves the accuracy of the Cortex-A53 integer scheduler...
Wilco Dijkstra [Wed, 6 Jul 2016 13:17:23 +0000 (13:17 +0000)]
This patch improves the accuracy of the Cortex-A53 integer scheduler...

This patch improves the accuracy of the Cortex-A53 integer scheduler,
resulting in performance gains across a wide range of benchmarks.

    gcc/
* config/arm/cortex-a53.md: Use final_presence_set for in-order.
(cortex_a53_shift): Add mov_shift.
(cortex_a53_shift_reg): Add new reservation for register shifts.
(cortex_a53_alu): Remove bfm.
(cortex_a53_alu_shift): Add bfm, remove mov_shift.
(cortex_a53_alu_extr): Add new reservation for EXTR.
(bypasses): Improve bypass modelling.

From-SVN: r238048

8 years agosem_ch7.adb (Analyze_Package_Specification): Insert its freezing nodes after the...
Javier Miranda [Wed, 6 Jul 2016 13:14:25 +0000 (13:14 +0000)]
sem_ch7.adb (Analyze_Package_Specification): Insert its freezing nodes after the last declaration.

2016-07-06  Javier Miranda  <miranda@adacore.com>

* sem_ch7.adb (Analyze_Package_Specification): Insert its
freezing nodes after the last declaration. Needed to ensure
that global entities referenced in aspects of frozen types are
properly handled.
* freeze.adb (Freeze_Entity): Minor code reorganization to ensure
that freezing nodes of generic packages are handled.
* sem_ch13.adb (Freeze_Entity_Checks): Handle N_Freeze_Generic nodes.
* sem_ch12.adb (Save_References_In_Identifier): Handle selected
components which denote a named number that is constant folded
in the analyzed copy of the tree.

From-SVN: r238047

8 years agoFix prototype in vst1Q_laneu64-1.c to unsigned char*.
Wilco Dijkstra [Wed, 6 Jul 2016 13:02:53 +0000 (13:02 +0000)]
Fix prototype in vst1Q_laneu64-1.c to unsigned char*.

    gcc/testsuite/
* gcc.target/arm/vst1Q_laneu64-1.c (foo): Use unsigned char*.

From-SVN: r238046

8 years ago[multiple changes]
Arnaud Charlet [Wed, 6 Jul 2016 12:40:07 +0000 (14:40 +0200)]
[multiple changes]

2016-07-06  Hristian Kirtchev  <kirtchev@adacore.com>

* exp_aggr.adb Remove with and use clauses for Exp_Ch11 and Inline.
(Initialize_Array_Component): Protect the initialization
statements in an abort defer / undefer block when the associated
component is controlled.
(Initialize_Record_Component): Protect the initialization statements
in an abort defer / undefer block when the associated component is
controlled.
(Process_Transient_Component_Completion): Use Build_Abort_Undefer_Block
to create an abort defer / undefer block.
* exp_ch3.adb Remove with and use clauses for Exp_ch11 and Inline.
(Default_Initialize_Object): Use Build_Abort_Undefer_Block to
create an abort defer / undefer block.
* exp_ch5.adb (Expand_N_Assignment_Statement): Mark an abort
defer / undefer block as such.
* exp_ch9.adb (Find_Enclosing_Context): Do not consider an abort
defer / undefer block as a suitable context for an activation
chain or a master.
* exp_util.adb Add with and use clauses for Exp_Ch11.
(Build_Abort_Undefer_Block): New routine.
* exp_util.ads (Build_Abort_Undefer_Block): New routine.
* sinfo.adb (Is_Abort_Block): New routine.
(Set_Is_Abort_Block): New routine.
* sinfo.ads New attribute Is_Abort_Block along with occurrences
in nodes.
(Is_Abort_Block): New routine along with pragma Inline.
(Set_Is_Abort_Block): New routine along with pragma Inline.

2016-07-06  Justin Squirek  <squirek@adacore.com>

* sem_ch4.adb (Analyze_One_Call): Add a conditional to handle
disambiguation.

From-SVN: r238045

8 years ago[multiple changes]
Arnaud Charlet [Wed, 6 Jul 2016 12:37:54 +0000 (14:37 +0200)]
[multiple changes]

2016-07-06  Hristian Kirtchev  <kirtchev@adacore.com>

* einfo.adb Flag252 is now used as Is_Finalized_Transient. Flag295
is now used as Is_Ignored_Transient.
(Is_Finalized_Transient): New routine.
(Is_Ignored_Transient): New routine.
(Is_Processed_Transient): Removed.
(Set_Is_Finalized_Transient): New routine.
(Set_Is_Ignored_Transient): New routine.
(Set_Is_Processed_Transient): Removed.
(Write_Entity_Flags): Output Flag252 and Flag295.
* einfo.ads: New attributes Is_Finalized_Transient
and Is_Ignored_Transient along with occurrences in
entities. Remove attribute Is_Processed_Transient.
(Is_Finalized_Transient): New routine along with pragma Inline.
(Is_Ignored_Transient): New routine along with pragma Inline.
(Is_Processed_Transient): Removed along with pragma Inline.
(Set_Is_Finalized_Transient): New routine along with pragma Inline.
(Set_Is_Ignored_Transient): New routine along with pragma Inline.
(Set_Is_Processed_Transient): Removed along with pragma Inline.
* exp_aggr.adb Add with and use clauses for Exp_Ch11 and Inline.
(Build_Record_Aggr_Code): Change the handling
of controlled record components.
(Ctrl_Init_Expression): Removed.
(Gen_Assign): Add new formal parameter In_Loop
along with comment on usage.  Remove local variables Stmt and
Stmt_Expr. Change the handling of controlled array components.
(Gen_Loop): Update the call to Gen_Assign.
(Gen_While): Update the call to Gen_Assign.
(Initialize_Array_Component): New routine.
(Initialize_Ctrl_Array_Component): New routine.
(Initialize_Ctrl_Record_Component): New routine.
(Initialize_Record_Component): New routine.
(Process_Transient_Component): New routine.
(Process_Transient_Component_Completion): New routine.
* exp_ch4.adb (Process_Transient_In_Expression): New routine.
(Process_Transient_Object): Removed. Replace all existing calls
to this routine with calls to Process_Transient_In_Expression.
* exp_ch6.adb (Expand_Ctrl_Function_Call): Remove local constant
Is_Elem_Ref. Update the comment on ignoring transients.
* exp_ch7.adb (Process_Declarations): Do not process ignored
or finalized transient objects.
(Process_Transient_In_Scope): New routine.
(Process_Transients_In_Scope): New routine.
(Process_Transient_Objects): Removed. Replace all existing calls
to this routine with calls to Process_Transients_In_Scope.
* exp_util.adb (Build_Transient_Object_Statements): New routine.
(Is_Finalizable_Transient): Do not consider a transient object
which has been finalized.
(Requires_Cleanup_Actions): Do not consider ignored or finalized
transient objects.
* exp_util.ads (Build_Transient_Object_Statements): New routine.
* sem_aggr.adb: Major code clean up.
* sem_res.adb: Update documentation.

2016-07-06  Ed Schonberg  <schonberg@adacore.com>

* sem_ch3.adb (Analyze_Subtype_Declaration): For generated
subtypes, such as actual subtypes of unconstrained formals,
inherit predicate functions, if any, from the parent type rather
than creating redundant new ones.

From-SVN: r238044

8 years ago[multiple changes]
Arnaud Charlet [Wed, 6 Jul 2016 12:34:52 +0000 (14:34 +0200)]
[multiple changes]

2016-07-06  Hristian Kirtchev  <kirtchev@adacore.com>

* exp_attr.adb, sem_attr.adb, sem_ch13.adb: Minor reformatting.

2016-07-06  Arnaud Charlet  <charlet@adacore.com>

* lib.adb (Check_Same_Extended_Unit): Prevent looping forever.
* gnatbind.adb: Disable some consistency checks in codepeer mode,
which are not needed.

2016-07-06  Ed Schonberg  <schonberg@adacore.com>

* sem_ch12.adb (Check_Fixed_Point_Actual): Add a warning when
a formal fixed point type is instantiated with a type that has
a user-defined arithmetic operations, but the generic has no
corresponding formal functions. This is worth a warning because
of the special semantics of fixed-point operators.

From-SVN: r238043

8 years agosem_attr.adb (Analyze_Attribute): Allow any expression of discrete type.
Bob Duff [Wed, 6 Jul 2016 12:32:35 +0000 (12:32 +0000)]
sem_attr.adb (Analyze_Attribute): Allow any expression of discrete type.

2016-07-06  Bob Duff  <duff@adacore.com>

* sem_attr.adb (Analyze_Attribute): Allow any expression of
discrete type.
* exp_attr.adb (Expand_N_Attribute_Reference): Change the
constant-folding code to correctly handle cases newly allowed
by Analyze_Attribute.

From-SVN: r238042

8 years agore PR target/50739 ([avr] nameless error with -fmerge-all-constants)
Senthil Kumar Selvaraj [Wed, 6 Jul 2016 11:40:17 +0000 (11:40 +0000)]
re PR target/50739 ([avr] nameless error with -fmerge-all-constants)

Fix PR target/50739

This patch fixes a problem with fmerge-all-constants and the progmem
attribute.

gcc/

PR target/50739
* config/avr/avr.c (avr_asm_select_section): Strip off
SECTION_DECLARED from flags when calling get_section.

testsuite/

PR target/50739
* gcc.target/avr/pr50739.c: New test

From-SVN: r238041

8 years ago* gcc.dg/tree-ssa/scev-14.c: update template.
Jan Hubicka [Wed, 6 Jul 2016 10:03:08 +0000 (12:03 +0200)]
* gcc.dg/tree-ssa/scev-14.c: update template.

From-SVN: r238040

8 years ago[7/7] Add negative and zero strides to vect_memory_access_type
Richard Sandiford [Wed, 6 Jul 2016 08:16:53 +0000 (08:16 +0000)]
[7/7] Add negative and zero strides to vect_memory_access_type

This patch uses the vect_memory_access_type from patch 6 to represent
the effect of a negative contiguous stride or a zero stride.  The latter
is valid only for loads.

Tested on aarch64-linux-gnu and x86_64-linux-gnu.

gcc/
* tree-vectorizer.h (vect_memory_access_type): Add
VMAT_INVARIANT, VMAT_CONTIGUOUS_DOWN and VMAT_CONTIGUOUS_REVERSED.
* tree-vect-stmts.c (compare_step_with_zero): New function.
(perm_mask_for_reverse): Move further up file.
(get_group_load_store_type): Stick to VMAT_ELEMENTWISE if the
step is negative.
(get_negative_load_store_type): New function.
(get_load_store_type): Call it.  Add an ncopies argument.
(vectorizable_mask_load_store): Update call accordingly and
remove tests for negative steps.
(vectorizable_store, vectorizable_load): Likewise.  Handle new
memory_access_types.

From-SVN: r238039

8 years ago[6/7] Explicitly classify vector loads and stores
Richard Sandiford [Wed, 6 Jul 2016 08:15:28 +0000 (08:15 +0000)]
[6/7] Explicitly classify vector loads and stores

This is the main patch in the series.  It adds a new enum and routines
for classifying a vector load or store implementation.

Originally there were three motivations:

      (1) Reduce cut-&-paste

      (2) Make the chosen vectorisation strategy more obvious.  At the
          moment this is derived implicitly from various other bits of
          state (GROUPED, STRIDED, SLP, etc.)

      (3) Decouple the vectorisation strategy from those other bits of state,
          so that there can be a choice of implementation for a given scalar
          statement.  The specific problem here is that we class:

              for (...)
                {
                  ... = a[i * x];
                  ... = a[i * x + 1];
                }

          as "strided and grouped" but:

              for (...)
                {
                  ... = a[i * 7];
                  ... = a[i * 7 + 1];
                }

          as "non-strided and grouped".  Before the patch, "strided and
          grouped" loads would always try to use separate scalar loads
          while "non-strided and grouped" loads would always try to use
          load-and-permute.  But load-and-permute is never supported for
          a group size of 7, so the effect was that the first loop was
          vectorisable and the second wasn't.  It seemed odd that not
          knowing x (but accepting it could be 7) would allow more
          optimisation opportunities than knowing x is 7.

Unfortunately, it looks like we underestimate the cost of separate
scalar accesses on at least aarch64, so I've disabled (3) for now;
see the "if" statement at the end of get_load_store_type.  I think
the patch still does (1) and (2), so that's the justification for
it in its current form.  It also means that (3) is now simply a
case of removing the FIXME code, once the cost model problems have
been sorted out.  (I did wonder about adding a --param, but that
seems overkill.  I hope to get back to this during GCC 7 stage 1.)

Tested on aarch64-linux-gnu and x86_64-linux-gnu.

gcc/
* tree-vectorizer.h (vect_memory_access_type): New enum.
(_stmt_vec_info): Add a memory_access_type field.
(STMT_VINFO_MEMORY_ACCESS_TYPE): New macro.
(vect_model_store_cost): Take an access type instead of a boolean.
(vect_model_load_cost): Likewise.
* tree-vect-slp.c (vect_analyze_slp_cost_1): Update calls to
vect_model_store_cost and vect_model_load_cost.
* tree-vect-stmts.c (vec_load_store_type): New enum.
(vect_model_store_cost): Take an access type instead of a
store_lanes_p boolean.  Simplify tests.
(vect_model_load_cost): Likewise, but for load_lanes_p.
(get_group_load_store_type, get_load_store_type): New functions.
(vectorizable_store): Use get_load_store_type.  Record the access
type in STMT_VINFO_MEMORY_ACCESS_TYPE.
(vectorizable_load): Likewise.
(vectorizable_mask_load_store): Likewise.  Replace is_store
variable with vls_type.

From-SVN: r238038

8 years ago[5/7] Move the fix for PR65518
Richard Sandiford [Wed, 6 Jul 2016 08:14:41 +0000 (08:14 +0000)]
[5/7] Move the fix for PR65518

This patch moves the fix for PR65518 to the code that checks whether
load-and-permute operations are supported.   If the group size is
greater than the vectorisation factor, it would still be possible
to fall back to elementwise loads (as for strided groups) rather
than fail vectorisation entirely.

Tested on aarch64-linux-gnu and x86_64-linux-gnu.

gcc/
* tree-vectorizer.h (vect_grouped_load_supported): Add a
single_element_p parameter.
* tree-vect-data-refs.c (vect_grouped_load_supported): Likewise.
Check the PR65518 case here rather than in vectorizable_load.
* tree-vect-loop.c (vect_analyze_loop_2): Update call accordignly.
* tree-vect-stmts.c (vectorizable_load): Likewise.

From-SVN: r238037

8 years ago[4/7] Add a gather_scatter_info structure
Richard Sandiford [Wed, 6 Jul 2016 08:13:49 +0000 (08:13 +0000)]
[4/7] Add a gather_scatter_info structure

This patch just refactors the gather/scatter support so that all
information is in a single structure, rather than separate variables.
This reduces the number of arguments to a function added in patch 6.

Tested on aarch64-linux-gnu and x86_64-linux-gnu.

gcc/
* tree-vectorizer.h (gather_scatter_info): New structure.
(vect_check_gather_scatter): Return a bool rather than a decl.
Replace return-by-pointer arguments with a single
gather_scatter_info *.
* tree-vect-data-refs.c (vect_check_gather_scatter): Likewise.
(vect_analyze_data_refs): Update call accordingly.
* tree-vect-stmts.c (vect_mark_stmts_to_be_vectorized): Likewise.
(vectorizable_mask_load_store): Likewise.  Also record the
offset dt and vectype in the gather_scatter_info.
(vectorizable_store): Likewise.
(vectorizable_load): Likewise.

From-SVN: r238036

8 years ago[3/7] Fix load/store costs for strided groups
Richard Sandiford [Wed, 6 Jul 2016 08:12:36 +0000 (08:12 +0000)]
[3/7] Fix load/store costs for strided groups

vect_model_store_cost had:

      /* Costs of the stores.  */
      if (STMT_VINFO_STRIDED_P (stmt_info)
          && !STMT_VINFO_GROUPED_ACCESS (stmt_info))
        {
          /* N scalar stores plus extracting the elements.  */
          inside_cost += record_stmt_cost (body_cost_vec,
       ncopies * TYPE_VECTOR_SUBPARTS (vectype),
       scalar_store, stmt_info, 0, vect_body);

But non-SLP strided groups also use individual scalar stores rather than
vector stores, so I think we should skip this only for SLP groups.

The same applies to vect_model_load_cost.

Tested on aarch64-linux-gnu and x86_64-linux-gnu.

gcc/
* tree-vect-stmts.c (vect_model_store_cost): For non-SLP
strided groups, use the cost of N scalar accesses instead
of ncopies vector accesses.
(vect_model_load_cost): Likewise.

From-SVN: r238035

8 years ago[2/7] Clean up vectorizer load/store costs
Richard Sandiford [Wed, 6 Jul 2016 08:11:31 +0000 (08:11 +0000)]
[2/7] Clean up vectorizer load/store costs

Add a bit more commentary and try to make the structure more obvious.
The horrendous:

      if (grouped_access_p
          && represents_group_p
          && !store_lanes_p
          && !STMT_VINFO_STRIDED_P (stmt_info)
          && !slp_node)

checks go away in patch 6.

Tested on aarch64-linux-gnu and x86_64-linux-gnu.

gcc/
* tree-vect-stmts.c (vect_cost_group_size): Delete.
(vect_model_store_cost): Avoid calling it.  Use first_stmt_p
variable to indicate when once-per-group costs are being used.
(vect_model_load_cost): Likewise.  Fix comment and misindented code.

From-SVN: r238034

8 years ago[1/7] Remove unnecessary peeling for gaps check
Richard Sandiford [Wed, 6 Jul 2016 08:10:29 +0000 (08:10 +0000)]
[1/7] Remove unnecessary peeling for gaps check

I recently relaxed the peeling-for-gaps conditions for LD3 but
kept them as-is for load-and-permute.  I don't think the conditions
are needed for load-and-permute either though.  No current load-and-
permute should load outside the group, so if there is no gap at the end,
the final vector element loaded will correspond to an element loaded
by the original scalar loop.

The patch for PR68559 (a missed optimisation PR) increased the peeled
cases from "exact_log2 (groupsize) == -1" to "vf % group_size == 0", so
before that fix, we didn't peel for gaps if there was no gap at the end
of the group and if the group size was a power of 2.

The only current non-power-of-2 load-and-permute size is 3, which
doesn't require loading more than 3 vectors.

The testcase is based on gcc.dg/vect/pr49038.c.

Tested on aarch64-linux-gnu and x86_64-linux-gnu.

gcc/
* tree-vect-stmts.c (vectorizable_load): Remove unnecessary
peeling-for-gaps condition.

gcc/testsuite/
* gcc.dg/vect/group-no-gaps-1.c: New test.

From-SVN: r238033

8 years agoS/390: Fix vecinit expansion.
Andreas Krebbel [Wed, 6 Jul 2016 07:05:11 +0000 (07:05 +0000)]
S/390: Fix vecinit expansion.

The fallback routine in the S/390 vecinit expander did not check
whether each of the initializer elements is a proper general_operand.
Since revision r236582 the expander is invoked also with e.g. symbol
refs with an odd addend resulting in invalid insns.

Fixed by forcing the element into a register in such cases.

gcc/ChangeLog:

2016-07-06  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/s390.c (s390_expand_vec_init): Force initializer
element to register if it doesn't match general_operand.

From-SVN: r238032

8 years agoFix MPX tests on systems with MPX disabled
Andi Kleen [Wed, 6 Jul 2016 02:38:49 +0000 (02:38 +0000)]
Fix MPX tests on systems with MPX disabled

I have a Skylake system with MPX in the CPU, but MPX is disabled
in the kernel configuration.

This makes all the MPX tests fail because they assume if MPX
is in CPUID it works

Check the output of XGETBV too to detect non MPX kernels.

gcc/testsuite/:

2016-07-05  Andi Kleen  <ak@linux.intel.com>

* gcc.target/i386/mpx/mpx-check.h: Check XGETBV output
if kernel supports MPX.

From-SVN: r238031

8 years agoDaily bump.
GCC Administrator [Wed, 6 Jul 2016 00:16:22 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r238029

8 years agopr69102.c: Require fpic support.
Kito Cheng [Tue, 5 Jul 2016 18:51:38 +0000 (18:51 +0000)]
pr69102.c: Require fpic support.

2016-07-05  Kito Cheng <kito.cheng@gmail.com>

* gcc.c-torture/compile/pr69102.c: Require fpic support.

From-SVN: r238023

8 years agoImplement LWG 2509,
Ville Voutilainen [Tue, 5 Jul 2016 18:33:18 +0000 (21:33 +0300)]
Implement LWG 2509,

any_cast doesn't work with rvalue reference targets and cannot
move with a value target.
* include/experimental/any (any(_ValueType&&)): Constrain and
add an overload that doesn't forward.
(any_cast(any&&)): Constrain and add an overload that moves.
* testsuite/experimental/any/misc/any_cast.cc: Add tests for
the functionality added by LWG 2509.

From-SVN: r238022

8 years agore PR c++/71214 (Typo in feature test macro for rvalue references)
Markus Trippelsdorf [Tue, 5 Jul 2016 17:50:41 +0000 (17:50 +0000)]
re PR c++/71214 (Typo in feature test macro for rvalue references)

Fix PR c++/71214

       PR c++/71214
       * c-cppbuiltin.c (c_cpp_builtins): Define __cpp_rvalue_references.

From-SVN: r238017

8 years agors6000-protos.h (rs6000_split_signbit): New prototype.
Michael Meissner [Tue, 5 Jul 2016 17:50:22 +0000 (17:50 +0000)]
rs6000-protos.h (rs6000_split_signbit): New prototype.

[gcc]

2016-07-05  Michael Meissner  <meissner@linux.vnet.ibm.com>
    Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

* config/rs6000/rs6000-protos.h (rs6000_split_signbit): New
prototype.
* config/rs6000/rs6000.c (rs6000_split_signbit): New function.
* config/rs6000/rs6000.md (UNSPEC_SIGNBIT): New constant.
(SIGNBIT): New mode iterator.
(Fsignbit): New mode attribute.
(signbit<mode>2): Change operand1 to match FLOAT128 instead of
IBM128; dispatch to gen_signbit{kf,tf}2_dm for __float128
when direct moves are available.
(signbit<mode>2_dm): New define_insn_and_split).
(signbit<mode>2_dm2): New define_insn.

[gcc/testsuite]

2016-07-05  Michael Meissner  <meissner@linux.vnet.ibm.com>
    Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

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

Co-Authored-By: Bill Schmidt <wschmidt@linux.vnet.ibm.com>
From-SVN: r238016

8 years ago[RTL ifcvt] PR rtl-optimization/71594: ICE in noce_emit_cmove due to mismatched sourc...
Kyrylo Tkachov [Tue, 5 Jul 2016 16:37:40 +0000 (16:37 +0000)]
[RTL ifcvt] PR rtl-optimization/71594: ICE in noce_emit_cmove due to mismatched source modes

PR rtl-optimization/71594
* ifcvt.c (noce_convert_multiple_sets): Wrap new_val or old_val
into subregs of appropriate mode before trying to emit a conditional
move.

* gcc.dg/torture/pr71594.c: New test.

From-SVN: r238013

8 years agotree-scalar-evolution.c (iv_can_overflow_p): New function.
Jan Hubicka [Tue, 5 Jul 2016 16:17:12 +0000 (18:17 +0200)]
tree-scalar-evolution.c (iv_can_overflow_p): New function.

* tree-scalar-evolution.c (iv_can_overflow_p): New function.
(simple_iv): Use it.

* gcc.dg/tree-ssa/scev-14.c: new testcase.

From-SVN: r238012

8 years ago* tree-ssa-loop-niter.c (nowrap_type_p): Use ANY_INTEGRAL_TYPE_P.
Jan Hubicka [Tue, 5 Jul 2016 16:13:46 +0000 (18:13 +0200)]
* tree-ssa-loop-niter.c (nowrap_type_p): Use ANY_INTEGRAL_TYPE_P.

From-SVN: r238011

8 years ago[LRA] Don't count spilling cost for it offmemok
Jiong Wang [Tue, 5 Jul 2016 16:05:27 +0000 (16:05 +0000)]
[LRA] Don't count spilling cost for it offmemok

* lra-constraints.c (process_alt_operands): Don't add spilling cost for
"offmemok".

From-SVN: r238010

8 years agotree-scalar-evoluiton.c (simple_iv): Use nowrap_type to check if IV can overflow.
Jan Hubicka [Tue, 5 Jul 2016 16:02:52 +0000 (18:02 +0200)]
tree-scalar-evoluiton.c (simple_iv): Use nowrap_type to check if IV can overflow.

* tree-scalar-evoluiton.c (simple_iv): Use nowrap_type to check if
IV can overflow.

From-SVN: r238009

8 years agoPR c++/62314: add fixit hint for "expected ';' after class definition"
David Malcolm [Tue, 5 Jul 2016 15:50:54 +0000 (15:50 +0000)]
PR c++/62314: add fixit hint for "expected ';' after class definition"

gcc/cp/ChangeLog:
PR c++/62314
* parser.c (cp_parser_class_specifier_1): When reporting
missing semicolons, use a fixit-hint to suggest insertion
of a semicolon immediately after the closing brace,
offsetting the reported column accordingly.

gcc/testsuite/ChangeLog:
PR c++/62314
* gcc/testsuite/g++.dg/parse/error5.C: Update column
number of missing semicolon error.
* g++.dg/pr62314-2.C: New test case.

From-SVN: r238008

8 years agoSecond review of STAT= patch + tests
Alessandro Fanfarillo [Tue, 5 Jul 2016 15:33:06 +0000 (09:33 -0600)]
Second review of STAT= patch + tests

From-SVN: r238007

8 years agogimple-ssa-split-paths.c (find_block_to_duplicate_for_splitting_pa): Handle empty...
Richard Biener [Tue, 5 Jul 2016 13:25:47 +0000 (13:25 +0000)]
gimple-ssa-split-paths.c (find_block_to_duplicate_for_splitting_pa): Handle empty else block.

2016-07-05  Richard Biener  <rguenther@suse.de>

* gimple-ssa-split-paths.c (find_block_to_duplicate_for_splitting_pa):
Handle empty else block.
(is_feasible_trace): Likewise.
(split_paths): Likewise.

From-SVN: r238005

8 years agotree-loop-distribution.c (distribute_loop): Fix issue with the cost model loop.
Richard Biener [Tue, 5 Jul 2016 13:25:19 +0000 (13:25 +0000)]
tree-loop-distribution.c (distribute_loop): Fix issue with the cost model loop.

2016-07-05  Richard Biener  <rguenther@suse.de>

* tree-loop-distribution.c (distribute_loop): Fix issue with
the cost model loop.

From-SVN: r238004

8 years agoUpdate documentation.
Arnaud Charlet [Tue, 5 Jul 2016 12:18:48 +0000 (14:18 +0200)]
Update documentation.

From-SVN: r238003

8 years agore PR fortran/71623 (Segfault when allocating deferred-length characters to size...
Andre Vehreschild [Tue, 5 Jul 2016 12:06:22 +0000 (14:06 +0200)]
re PR fortran/71623 (Segfault when allocating deferred-length characters to size of a pointer)

gcc/fortran/ChangeLog:

2016-07-05  Andre Vehreschild  <vehre@gcc.gnu.org>

PR fortran/71623
* trans-stmt.c (gfc_trans_allocate): Add code of pre block of typespec
in allocate to parent block.

gcc/testsuite/ChangeLog:

2016-07-05  Andre Vehreschild  <vehre@gcc.gnu.org>

PR fortran/71623
* gfortran.dg/deferred_character_17.f90: New test.

From-SVN: r238002

8 years agodecl.c (gnat_to_gnu_entity): Invoke global_bindings_p last when possible.
Eric Botcazou [Tue, 5 Jul 2016 10:32:43 +0000 (10:32 +0000)]
decl.c (gnat_to_gnu_entity): Invoke global_bindings_p last when possible.

* gcc-interface/decl.c (gnat_to_gnu_entity): Invoke global_bindings_p
last when possible.  Do not call elaborate_expression_2 on offsets in
local record types and avoid useless processing for constant offsets.

From-SVN: r238001

8 years ago[ARM][testsuite] neon-testgen.ml removal
Christophe Lyon [Tue, 5 Jul 2016 09:24:38 +0000 (09:24 +0000)]
[ARM][testsuite] neon-testgen.ml removal

2016-07-05  Christophe Lyon  <christophe.lyon@linaro.org>

gcc/
* config/arm/neon-testgen.ml: Delete.
* config/arm/neon.ml: Delete.

gcc/testsuite/
* gcc.target/arm/neon/polytypes.c: Move to ...
* gcc.target/arm/polytypes.c: ... here.
* gcc.target/arm/neon/pr51534.c: Move to ...
* gcc.target/arm/pr51534.c: ... here.
* gcc.target/arm/neon/vect-vcvt.c: Move to ...
* gcc.target/arm/vect-vcvt.c: ... here.
* gcc.target/arm/neon/vect-vcvtq.c: Move to ...
* gcc.target/arm/vect-vcvtq.c: ... here.
* gcc.target/arm/neon/vfp-shift-a2t2.c: Move to ...
* gcc.target/arm/vfp-shift-a2t2.c: ... here.
* gcc.target/arm/neon/vst1Q_laneu64-1.c: Move to ...
* gcc.target/arm/vst1Q_laneu64-1.c: ... here. Fix foo() prototype.
* gcc.target/arm/neon/neon.exp: Delete.
* gcc.target/arm/neon/*.c: Delete.

From-SVN: r238000

8 years agoDaily bump.
GCC Administrator [Tue, 5 Jul 2016 00:16:18 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r237998

8 years agore PR fortran/66575 (Endless compilation on missing end interface)
Jerry DeLisle [Mon, 4 Jul 2016 21:04:55 +0000 (21:04 +0000)]
re PR fortran/66575 (Endless compilation on missing end interface)

2016-07-04  Jerry DeLisle  <jvdelisle@gcc.gnu.org>

PR fortran/66575
* decl.c (match_procedure_interface): Exit loop if procedure
interface refers to itself.

* gfortran.dg: pr65575.f90: New test.

From-SVN: r237994

8 years agore PR fortran/35849 ("wrong" line shown in error message for parameter)
Jerry DeLisle [Mon, 4 Jul 2016 19:14:54 +0000 (19:14 +0000)]
re PR fortran/35849 ("wrong" line shown in error message for parameter)

2016-07-04  Jerry DeLisle  <jvdelisle@gcc.gnu.org>
    Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/35849
* simplify.c (gfc_simplify_ishftc): Check that absolute value of
SHIFT is less than or equal to SIZE.

* gfortran.dg: pr35849.f90: New test.

Co-Authored-By: Steven G. Kargl <kargl@gcc.gnu.org>
From-SVN: r237993