fold-const.c (fold_comparsion): Pass OEP_ADDRESS_OF when comparing addresses.

* fold-const.c (fold_comparsion): Pass OEP_ADDRESS_OF when comparing
addresses.
(fold_addr_of_array_ref_difference): Likewise.

From-SVN: r228701

[PATCH] Refactoring FSM bits into their own file

[PATCH] Refactoring FSM bits into their own file
* tree-ssa-threadedge.c (fsm_find_thread_path): Moved from here into
tree-ssa-threadbackward.c.
(fsm_find_control_statement_thread_paths): Likewise.
(thread_through_normal_block): Break out FSM bits and move them
into a new function in tree-ssa-threadbackward.c.  Call new function
instead.
Minimize header file usage.
* tree-ssa-threadbackward.h: New file.
* tree-ssa-threadbackward.c: Likewise.
* Makefile.in (OBJS): Add tree-ssa-threadbackward.o

From-SVN: r228700

Daily bump.

From-SVN: r228699

Test for __cxa_thread_atexit_impl when cross-compiling libstdc++ for GNU targets.

I noticed that when testing glibc with a cross compiler I got

UNSUPPORTED: nptl/tst-thread_local1

because the libstdc++-v3 configuration for cross compiling defaulted
to __cxa_thread_atexit_impl not being available.  This patch fixes
GLIBCXX_CROSSCONFIG to run the same test (for the case covering
targets with glibc) for __cxa_thread_atexit_impl as for native
compilation, just as it runs most of the other tests done for native
compilation (for these targets, it's not possible to build libstdc++
without already having built libc, so link tests are OK).

Tested with no regressions for cross to arm-none-linux-gnueabi.

* crossconfig.m4 (GLIBCXX_CROSSCONFIG) <*-linux* | *-uclinux* |
*-gnu* | *-kfreebsd*-gnu | *-knetbsd*-gnu | *-cygwin*>: Check for
__cxa_thread_atexit_impl.
* configure: Regenerate.

From-SVN: r228695

alpha.h (ALPHA_ROUND): Implement using ROUND_UP macro.

* config/alpha/alpha.h (ALPHA_ROUND): Implement using ROUND_UP macro.

From-SVN: r228693

bb-reorder: Improve the simple algorithm for -Os (PR67864)

As the PR points out, the "simple" reorder algorithm makes bigger code
than the STC algorithm did, for -Os, for x86.  I now tested it for many
different targets and it turns out to be worse everywhere.

This simple patch tunes "simple" a bit; this makes it better than STC
almost everywhere.  The only exceptions (for the targets where I have
results) are x86 and mn10300.  For those targets it may be best to switch
the default algorithm for -Os to STC.

2015-10-11  Segher Boessenkool  <segher@kernel.crashing.org>

PR rtl-optimization/67864
* gcc/bb-reorder (reorder_basic_blocks_simple): Prefer existing
fallthrough edges for conditional jumps.  Don't sort candidate
edges if not optimizing for speed.

From-SVN: r228692

Add bdver4 for multiversioning and fix AMD model detection.

2015-10-09  Venkataramanan kumar <venkataramanan.kumar@amd.com>

        * config/i386/cpuinfo.c (get_amd_cpu): Detect bdver4.
        (__cpu_indicator_init): Fix model selection for AMD CPUs.

From-SVN: r228691

always define REVERSE_CONDITION

gcc/ChangeLog:

2015-10-10  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* defaults.h (REVERSE_CONDITION): New default definition.
* jump.c (reversed_comparison_code_parts): Adjust.

From-SVN: r228690

remove unneeded #if for HARD_FRAME_POINTER_IS_ARG_POINTER

gcc/ChangeLog:

2015-10-10  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* builtins.c (expand_builtin_setjmp_receiver): Don't use #if to
check HARD_FRAME_POINTER_IS_ARG_POINTER.

From-SVN: r228689

always define FRAME_ADDR_RTX

gcc/ChangeLog:

2015-10-10  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* defaults.h (FRAME_ADDR_RTX): New default definition.
* builtins.c (expand_builtin_return_addr): Adjust.

From-SVN: r228688

always define DYNAMIC_CHAIN_ADDRESS

gcc/ChangeLog:

2015-10-10  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* defaults.h (DYNAMIC_CHAIN_ADDRESS): New default definition.
* builtins.c (expand_builtin_return_addr): Adjust.

From-SVN: r228687

always define SETUP_FRAME_ADDRESSES

gcc/ChangeLog:

2015-10-10  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* defaults.h (SETUP_FRAME_ADDRESSES): New default definition.
* builtins.c (expand_builtin_return_addr): Adjust.
* doc/tm.texi: Likewise.
* doc/tm.texi.in: Likewise.
* except.c (expand_builtin_unwind_init): Likewise.

From-SVN: r228686

always define INITIAL_FRAME_ADDRESS_RTX

gcc/ChangeLog:

2015-10-10  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* builtins.c (expand_builtin_return_addr): Adjust.
* defaults.h (INITIAL_FRAME_ADDRESS_RTX): New default definition.

From-SVN: r228685

Daily bump.

From-SVN: r228684

tree.c (type_with_interoperable_signedness): New.

* tree.c (type_with_interoperable_signedness): New.
(gimple_canonical_types_compatible_p): Use it.
* tree.h (type_with_interoperable_signedness): Declare

From-SVN: r228681

tree.c (type_with_interoperable_signedness): New.

* tree.c (type_with_interoperable_signedness): New.
(gimple_canonical_types_compatible_p): Use it.
* tree.h (type_with_interoperable_signedness): Declare

* lto.c (hash_canonical_type): Honor
type_with_interoperable_signedness.

* gfortran.dg/lto/bind_c-2_0.f90: New testcase.
* gfortran.dg/lto/bind_c-2_1.c: New testcase.
* gfortran.dg/lto/bind_c-3_0.f90: New testcase.
* gfortran.dg/lto/bind_c-3_1.c: New testcase.
* gfortran.dg/lto/bind_c-4_0.f90: New testcase.
* gfortran.dg/lto/bind_c-4_1.c: New testcase.
* gfortran.dg/lto/bind_c-5_0.f90: New testcase.
* gfortran.dg/lto/bind_c-5_1.c: New testcase.

From-SVN: r228680

fold-const.c (operand_equal_p): Document OEP_ADDRESS_OF and OEP_CONSTANT_ADDRESS_OF...

* fold-const.c (operand_equal_p): Document OEP_ADDRESS_OF
and OEP_CONSTANT_ADDRESS_OF; skip type compatibility checks
when OEP_ADDRESS_OF is se.

From-SVN: r228679

add dump of data dependences

2015-10-07  Aditya Kumar  <aditya.k7@samsung.com>
                Sebastian Pop  <s.pop@samsung.com>

                * graphite-dependences.c (scop_get_dependences): Add dump of the
                data dependence graph.
                * graphite-poly.c (print_isl_union_map): New.
                (debug_isl_union_map): New.
                * graphite-poly.h (print_isl_union_map): Declare.
                (debug_isl_union_map): Declare.

Co-Authored-By: Sebastian Pop <s.pop@samsung.com>
From-SVN: r228677

clean up debug functions

2015-10-07  Aditya Kumar  <aditya.k7@samsung.com>
                Sebastian Pop  <s.pop@samsung.com>

        * graphite-poly.c (print_iteration_domain): Remove verbosity.
        Remove OpenScop formatting.
        (print_iteration_domains): Same.
        (debug_iteration_domain): Same.
        (debug_iteration_domains): Same.
        (print_pdr): Same.
        (debug_pdr): Same.
        (dump_gbb_cases): Same.
        (dump_gbb_conditions): Same.
        (print_pdrs): Same.
        (debug_pdrs): Same.
        (print_pbb_body): Same.
        (print_pbb): Same.
        (print_scop_params): Same.
        (print_scop_context): Same.
        (print_scop): Same.
        (debug_pbb_domain): Same.
        (debug_pbb): Same.
        (debug_scop_context): Same.
        (debug_scop): Same.
        (debug_scop_params): Same.
        * graphite-poly.h: Same.
        * graphite.c (graphite_transform_loops): Same.

Co-Authored-By: Sebastian Pop <s.pop@samsung.com>
From-SVN: r228676

remove an unneeded as_a<rtx_insn *> ()

gcc/ChangeLog:

2015-10-10  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* function.c (stack_protect_epilogue): Remove as_a<rtx_insn *> ()
call that isn't needed.

From-SVN: r228675

Daily bump.

From-SVN: r228672

Re: [PATCH] Update SSA_NAME manager to use two lists

* tree-ssanames.c (flush_ssaname_freelist): Use splice and truncate
rather than moving each name to the freelist individually.

From-SVN: r228668

mips.exp (mips_option_groups): Add -mframe-header-opt and -mno-frame-header-opt options.

2015-10-09  Steve Ellcey  <sellcey@imgtec.com>

* gcc.target/mips/mips.exp (mips_option_groups): Add -mframe-header-opt
and -mno-frame-header-opt options.
* gcc.target/mips/frame-header-1.c: New file.
* gcc.target/mips/frame-header-2.c: New file.
* gcc.target/mips/frame-header-3.c: New file.

From-SVN: r228667

config.gcc (mips*-*-*): Add frame-header-opt.o to extra_objs.

2015-10-05  Steve Ellcey  <sellcey@imgtec.com>

* config.gcc (mips*-*-*): Add frame-header-opt.o to extra_objs.
* frame-header-opt.c: New file.
* config/mips/mips-proto.h (mips_register_frame_header_opt):
Add prototype.
* config/mips/mips.c (mips_compute_frame_info): Check
optimize_call_stack flag.
(mips_option_override): Register new frame_header_opt pass.
  (mips_frame_info, mips_int_mask, mips_shadow_set,
machine_function): Move these types to...
* config/mips/mips.h: here.
(machine_function): Add does_not_use_frame_header and
optimize_call_stack fields.
* config/mips/t-mips (frame-header-opt.o): Add new make rule.
* doc/invoke.texi (-mframe-header-opt, -mno-frame-header-opt):
Document new flags.
* config/mips/mips.opt (mframe-header-opt): Add new option.

From-SVN: r228666

* config/i386/i386.c
(expand_set_or_movmem_prologue_epilogue_by_misaligned_moves): Use
ROUND_DOWN where applicable.

From-SVN: r228665

[PATCH] [6/n] Fix minor SSA_NAME leaks

* tree-vect-stmts.c (free_stmt_vec_info): Free the LHS of the
correct statement.

From-SVN: r228663

[PATCH][ARM]Add earlyclobber modifier for neon_(vtrn, vuzp, vzip)<mode>_insn rtx
pattern.

gcc/

2015-10-09  Renlin Li  <renlin.li@arm.com>

* config/arm/neon.md (neon_vuzp<mode>_insn): Add & modifier for
        operands[0] and operands[2].
        (neon_vtrn<mode>_insn): Likewise.
        (neon_vzip<mode>_insn): Likewise.

From-SVN: r228662

[PATCH V3][GCC] Algorithmic optimization in match and simplify

gcc/ChangeLog

* match.pd: ((X inner_op C0) outer_op C1) New pattern.
((X & C2) << C1): Expand to...
(X {&,^,|} C2 << C1): ...This.
((X & C2) >> C1): Expand to...
(X {&,^,|} C2 >> C1): ...This.

gcc/testsuite/ChangeLog

* gcc.dg/tree-ssa/forwprop-33.c: New.

Co-Authored-By: Hale Wang <hale.wang@arm.com>
From-SVN: r228661

AVX512: Fix embedded rounding/SAE specifier position.

gcc/

PR target/67895
* config/i386/sse.md (define_insn "sse_cvtsi2ss<round_name>"):
Adjust embedded rounding/SAE specifier position.
(define_insn "sse_cvtsi2ssq<round_name>"): Likewise.
(define_insn "cvtusi2<ssescalarmodesuffix>32<round_name>"): Likewise.
(define_insn "cvtusi2<ssescalarmodesuffix>64<round_name>"): Likewise.
(define_insn "sse2_cvtsi2sdq<round_name>"): Likewise.
(define_insn "avx512dq_rangep<mode><mask_name><round_saeonly_name>"):
Likewise.
(define_insn "avx512dq_ranges<mode><round_saeonly_name>"): Likewise.

gcc/testsuite

PR target/67895
* gcc.target/i386/avx512dq-vrangepd-1.c: Adjust assembly regexp.
* gcc.target/i386/avx512dq-vrangeps-1.c: Likewise.
* gcc.target/i386/avx512dq-vrangesd-1.c: Likewise.
* gcc.target/i386/avx512dq-vrangess-1.c: Likewise.
* gcc.target/i386/avx512f-vcvtsi2sd64-1.c: Likewise.
* gcc.target/i386/avx512f-vcvtsi2ss-1.c: Likewise.
* gcc.target/i386/avx512f-vcvtsi2ss64-1.c: Likewise.
* gcc.target/i386/avx512f-vcvtusi2sd64-1.c: Likewise.
* gcc.target/i386/avx512f-vcvtusi2ss-1.c: Likewise.
* gcc.target/i386/avx512f-vcvtusi2ss64-1.c: Likewise.

From-SVN: r228660

Testsuite: add dg-{begin|end}-multiline-output commands

This patch adds an easy way to write tests for expected multiline
output.  For example we can test carets and underlines for
a particular diagnostic with:

/* { dg-begin-multiline-output "" }
 typedef struct _GMutex GMutex;
                ^~~~~~~
   { dg-end-multiline-output "" } */

multiline.exp is used by prune.exp; hence we need to load it before
prune.exp via *load_gcc_lib* for the testsuites of the various
non-"gcc" support libraries (e.g. boehm-gc).

gcc/testsuite/ChangeLog:
* lib/multiline.exp: New file.
* lib/prune.exp: Load multiline.exp.
(prune_gcc_output): Call into multiline.exp to handle any
multiline output directives.
* lib/libgo.exp: Load multiline.exp before prune.exp, using
load_gcc_lib.

boehm-gc/ChangeLog:
* testsuite/lib/boehm-gc.exp: Load multiline.exp before
prune.exp, using load_gcc_lib.

libatomic/ChangeLog:
* testsuite/lib/libatomic.exp: Load multiline.exp before
prune.exp, using load_gcc_lib.

libgomp/ChangeLog:
* testsuite/lib/libgomp.exp: Load multiline.exp before prune.exp,
using load_gcc_lib.

libitm/ChangeLog:
* testsuite/lib/libitm.exp: Load multiline.exp before prune.exp,
using load_gcc_lib.

libvtv/ChangeLog:
* testsuite/lib/libvtv.exp: Load multiline.exp before prune.exp,
using load_gcc_lib.

From-SVN: r228655

[PR 67794] Also remap SSA_NAMEs defined in ASMs in IPA-SRA

2015-10-09  Martin Jambor  <mjambor@suse.cz>

tree-optimization/67794
* tree-sra.c (replace_removed_params_ssa_names): Do not distinguish
between types of state,ents but accept original definitions as a
parameter.
(ipa_sra_modify_function_body): Use FOR_EACH_SSA_DEF_OPERAND to
iterate over definitions.

testsuite/
        * gcc.dg/ipa/ipa-sra-10.c: Nw test.
        * gcc.dg/torture/pr67794.c: Likewise.

From-SVN: r228654

rs6000.c (rs6000_offload_options): New.

* config/rs6000/rs6000.c (rs6000_offload_options): New.
(TARGET_OFFLOAD_OPTIONS): New.

From-SVN: r228653

[PR67891] don't test is_gimple_reg after parm expansion

for  gcc/ChangeLog

PR middle-end/67891
* cfgexpand.c (set_parm_rtl): Drop is_gimple_reg test.

From-SVN: r228652

[PR67766] reorder handling of parallels, concats and promoted values in return

for  gcc/ChangeLog

PR middle-end/67766
* function.c (expand_function_end): Move return value
promotion past the handling of PARALLELs and CONCATs.

From-SVN: r228651

[PR67828] don't unswitch on default defs of non-parms

for  gcc/ChangeLog

PR rtl-optimizatoin/67828
* tree-ssa-loop-unswitch.c: Include tree-ssa.h.
(tree_may_unswitch_on): Don't unswitch on expressions
involving undefined values.

for  gcc/testsuite/ChangeLog

PR rtl-optimization/67828
* gcc.dg/torture/pr67828.c: New.

From-SVN: r228650

* MAINTAINERS (Write After Approval): Add myself.

From-SVN: r228649

genmatch.c (print_operand): Fix formatting.

2015-10-09  Richard Biener  <rguenther@suse.de>

* genmatch.c (print_operand): Fix formatting.
(dt_node::append_simplify): Warn for multiple simplifiers
that match the same pattern.
* match.pd (log (exp @0)): Remove duplicates.

From-SVN: r228648

fix Changelog entry removed at revision 228479

From-SVN: r228647

Missed ChangeLog for previous libstdc++-v3 commit

Add include/experimental/bits dir

* include/experimental/erase_if.h: Move to ...
* include/experimental/bits/erase_if.h: New.
* include/experimental/fs_dir.h: Move to ...
* include/experimental/bits/fs_dir.h: New.
* include/experimental/fs_fwd.h: Move to ...
* include/experimental/bits/fs_fwd.h: New.
* include/experimental/fs_ops.h: Move to ...
* include/experimental/bits/fs_ops.h: New.
* include/experimental/fs_path.h: Move to ...
* include/experimental/bits/fs_path.h: New.
* include/experimental/string_view.tcc: Move to ...
* include/experimental/bits/string_view.tcc: New.
* include/Makefile.am: Add include/experimental/bits sub-directory.
* include/Makefile.in: Regenerate.
* include/experimental/filesystem: Adjust includes.
* include/experimental/map: Likewise.
* include/experimental/set: Likewise.
* include/experimental/string_view: Likewise.
* include/experimental/unordered_map: Likewise.
* include/experimental/unordered_set: Likewise.

From-SVN: r228646

Add include/experimental/bits dir

From-SVN: r228645

[PATCH PR target/67366 2/2] [gimple-fold.c] Support movmisalign optabs in gimple-fold.c

This patch by Richard allows for movmisalign optabs to be supported
in gimple-fold.c. This caused a bit of pain in the testsuite with strlenopt-8.c
in conjunction with the ARM support for movmisalign_optabs as the test
was coded up to do different things depending on whether the target
supported misaligned access or not. However now with unaligned access
being allowed for different levels of the architecture in the arm backend,
the concept of the helper function non_strict_align mapping identically
to the definition of STRICT_ALIGNMENT disappears.

Adjusted thusly for ARM. The testsuite/lib changes were tested with an
arm-none-eabi multilib that included architecture variants that did not
support unaligned access and architecture variants that did.

The testing matrix for this patch was:

1. x86_64 bootstrap and regression test - no regressions.
2. armhf bootstrap and regression test - no regressions.
3. arm-none-eabi cross build and regression test for

{-marm/-march=armv7-a/-mfpu=vfpv3-d16/-mfloat-abi=softfp}
{-mthumb/-march=armv8-a/-mfpu=crypto-neon-fp-armv8/-mfloat-abi=hard}
{-marm/-mcpu=arm7tdmi/-mfloat-abi=soft}
{-mthumb/-mcpu=arm7tdmi/-mfloat-abi=soft}

with no regressions.

Ok to apply ?

2015-10-09  Richard Biener  <rguenth@suse.de>

PR target/67366
* gimple-fold.c (optabs-query.h): Include
(gimple_fold_builtin_memory_op): Allow unaligned stores
when movmisalign_optabs are available.

2015-10-09  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>

PR target/67366
* lib/target-supports.exp (check_effective_target_non_strict_align):
Adjust for arm*-*-*.
* gcc.target/arm/pr67366.c: New test.

From-SVN: r228644

[Patch PR target/67366 1/2] [ARM] - Add movmisalignhi / si patterns

This adds movmisalignhi and movmisalignsi expanders when unaligned
access is allowed by the architecture. This allows the mid-end
to expand to misaligned loads and stored.

Compared code generated for the Linux kernel and
it changes code generation for a handful of files all for the better
basically by reducing the stack usage.

Tested by :

1. armhf bootstrap and regression test - no regressions.
2.. arm-none-eabi cross build and regression test for

    {-marm/-march=armv7-a/-mfpu=vfpv3-d16/-mfloat-abi=softfp}
    {-mthumb/-march=armv8-a/-mfpu=crypto-neon-fp-armv8/-mfloat-abi=hard}
    {-marm/-mcpu=arm7tdmi/-mfloat-abi=soft}
    {-mthumb/-mcpu=arm7tdmi/-mfloat-abi=soft}

Will apply to trunk once 2/2 is approved.

regards
Ramana

2015-10-09  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>

        PR target/67366
        * config/arm/arm.md (movmisalign<mode>): New.
        * config/arm/iterators.md (HSI): New.

From-SVN: r228643

* MAINTAINERS (Write After Approval): Add myself.

From-SVN: r228642

Make-lang.in: Make sure that GNAT1_OBJS and not just GNAT1_ADA_OBJS are compiled only after...

* gcc-interface/Make-lang.in: Make sure that GNAT1_OBJS and not just
GNAT1_ADA_OBJS are compiled only after generated files are created.

From-SVN: r228636

re PR middle-end/67891 (FAIL: gcc.dg/pr43300.c (internal compiler error) on alpha-linux-gnu)

2015-10-09  Richard Biener  <rguenther@suse.de>

PR tree-optimization/67891
* gimple-match.h (gimple_simplified_result_is_gimple_val):
New helper.
(gimple_resimplify1): Declare.
(gimple_resimplify2): Likewise.
(gimple_resimplify3): Likewise.
* gimple-match-head.c (gimple_resimplify1): Export.
(gimple_resimplify2): Likewise.
(gimple_resimplify3): Likewise.
(maybe_push_res_to_seq): Use gimple_simplified_result_is_gimple_val.
* gimple-fold.c (gimple_fold_stmt_to_constant_1): Likewise.
* tree-ssa-sccvn.c (visit_reference_op_load): Use gimple_resimplify1
to avoid creating stmts without VN info.

* gcc.dg/tree-ssa/pr67891.c: New testcase.

From-SVN: r228635

[PATCH] [5/n] Fix minor SSA_NAME leaks

* value-prof.c (gimple_ic): Add missing calls to unlink_stmt_vdef
and release_ssa_name in two places.
(gimple_stringop_fixed_value): Similarly.

From-SVN: r228632

Daily bump.

From-SVN: r228631

ipa-icf.c (sem_item::compare_symbol_references): Fix use of availability.

* ipa-icf.c (sem_item::compare_symbol_references): Fix use
of availability.

From-SVN: r228628

[PATCH] [4/n] Fix minor SSA_NAME leaks

        * tree-ssa-loop-im.c (rewrite_bittest): Add missing call to
        release_defs.

From-SVN: r228627

[PATCH] [4/n] Fix minor SSA_NAME leaks

* tree-ssa-loop-im.c (rewrite_bittest): Add missing call to
release_defs.

From-SVN: r228626

[PATCH] [3/n] Fix minor SSA_NAME leaks

* tree-stdarg.c (expand_ifn_va_arg_1): Add missing call to
unlink_stmt_vdef and release_ssa_name_fn.

From-SVN: r228625

[PATCH] [2/n] Fix minor SSA_NAME leaks

* tree-ssa-dse.c (dse_optimize_stmt): Add missing call to
release_defs.

From-SVN: r228624

liboffloadmic: Add missed checks for malloc and strdup return values

liboffloadmic/
* runtime/offload_engine.cpp (Engine::init_process): Use strdup instead
of sizeof+malloc+sprintf, check for return value.
* runtime/offload_env.cpp (MicEnvVar::get_env_var_kind): Check for
strdup return value.
* runtime/offload_host.cpp (__offload_init_library_once): Check for
strdup return value.  Fix size calculation of COI_HOST_THREAD_AFFINITY.
* runtime/emulator/coi_device.cpp (COIProcessWaitForShutdown): Check for
malloc return value.

From-SVN: r228622

Round up the SSE register save area only if needed

There is is no point to round up the SSE register save area to 16 bytes if
the incoming stack boundary is less than 16 bytes.

* config/i386/i386.c (ix86_compute_frame_layout): Round up the
SSE register save area to 16 bytes only if the incoming stack
boundary is no less than 16 bytes.

From-SVN: r228621

[PATCH] [1/n] Fix minor SSA_NAME leaks

* tree-ssa-phiopt.c (factor_out_conversion): Add missing calls to
release_ssa_name.  Fix typo in comment.

From-SVN: r228619

nvptx.h (struct machine_function): Add comment.

* config/nvptx/nvptx.h (struct machine_function): Add comment.
* config/nvptx/nvptx.c (nvptx_declare_function_name): Functions
may return pointer as well as in memory.
(nvptx_output_return): Likewise.

From-SVN: r228618

Move sqrt and cbrt simplifications to match.pd

This patch moves the sqrt and cbrt simplification rules to match.pd.
builtins.c now only does the constant folding.

Bootstrapped & regression-tested on x86_64-linux-gnu.

gcc/
* builtins.c (fold_builtin_sqrt, fold_builtin_cbrt): Delete.
(fold_builtin_1): Update accordingly.  Handle constant arguments here.
* match.pd: Add rules previously handled by fold_builtin_sqrt
and fold_builtin_cbrt.

gcc/testsuite/
* gcc.dg/builtins-47.c: Test the optimized dump instead.

From-SVN: r228616

Make tree_expr_nonnegative_warnv_p recurse into SSA names

The upcoming patch to move sqrt and cbrt simplifications to match.pd
caused a regression because the (abs @0)->@0 simplification didn't
trigger for:

        (abs (convert (abs X)))

The simplification is based on tree_expr_nonnegative_p, which at
the moment just gives up if it sees an SSA_NAME.

This patch makes tree_expr_nonnegative_p recurse into SSA name
definitions, but limits the depth of recursion to a small number
for the reason mentioned in the comment (adapted from an existing
comment in gimple_val_nonnegative_real_p).  The patch reuses code
in tree-vrp.c, moving it to gimple-fold.c.  It also replaces calls
to gimple_val_nonnegative_real_p with calls to tree_expr_nonnegative_p.

A knock-on effect is that we can now prove _i_589 < 0 is false in
sequences like:

      i_1917 = ASSERT_EXPR <i_1075, i_1075 == 0>;
      _i_589 = (const int) i_1917;
      _i_1507 = ASSERT_EXPR <_i_589, _i_589 < 0>;

This defeats an assert in tree-vrp.c that ASSERT_EXPR conditions
are never known to be false.  Previously the assert only ever used
local knowledge and so would be limited to cases like x != x for
integer x.  Now that we use global knowledge it's possible to prove
the assertion condition is false in blocks that are in practice
unreachable.  The patch therefore removes the assert.

Bootstrapped & regression-tested on x86_64-linux-gnu.  I didn't write
a specific test because this is already covered by the testsuite if
the follow-on patch is also applied.

gcc/
* params.def (PARAM_MAX_SSA_NAME_QUERY_DEPTH): New param.
* doc/invoke.texi (--param max-ssa-name-query-depth): Document.
* fold-const.h (tree_unary_nonnegative_warnv_p)
(tree_single_nonnegative_warnv_p, tree_call_nonnegative_warnv_p)
(tree_expr_nonnegative_warnv_p): Add depth parameters.
* fold-const.c: Include gimple-fold.h and params.h.
(tree_ssa_name_nonnegative_warnv_p): New function.
(tree_unary_nonnegative_warnv_p, tree_binary_nonnegative_warnv_p)
(tree_single_nonnegative_warnv_p, tree_call_nonnegative_warnv_p)
(tree_invalid_nonnegative_warnv_p, tree_expr_nonnegative_warnv_p):
Add a depth parameter and increment it for recursive calls to
tree_expr_nonnegative_warnv_p.  Use tree_ssa_name_nonnegative_warnv_p
to handle SSA names.
* gimple-fold.h (gimple_val_nonnegative_real_p): Delete.
(gimple_stmt_nonnegative_warnv_p): Declare.
* tree-vrp.c (remove_range_assertions): Remove assert that condition
cannot be proven false.
(gimple_assign_nonnegative_warnv_p, gimple_call_nonnegative_warnv_p)
(gimple_stmt_nonnegative_warnv_p): Move to...
* gimple-fold.c: ...here.  Add depth parameters and pass them
down to the tree routines.  Accept statements that aren't
assignments or calls but just return false for them.
(gimple_val_nonnegative_real_p): Delete.
* tree-ssa-math-opts.c (gimple_expand_builtin_pow): Use
tree_expr_nonnegative_p instead of gimple_val_nonnegative_real_p.
Check HONOR_NANs first.

From-SVN: r228614

Fix Solaris bootstrap

2015-10-08  Martin Jambor  <mjambor@suse.cz>

* ipa-cp.c (meet_with_1): Make the argument of abs signed.  Remove
unnecessary MIN.

From-SVN: r228607

cp-tree.h (SIMPLE_TARGET_EXPR_P): New.

* cp-tree.h (SIMPLE_TARGET_EXPR_P): New.
* init.c (get_nsdmi): Use it.
* typeck2.c (massage_init_elt): Use it.

From-SVN: r228606

format

From-SVN: r228605

* zh_TW.po: Update.

From-SVN: r228603

re PR c++/67557 (Calling copy constructor of base class in constructor of derived class produces crashing code)

PR c++/67557

* call.c (is_base_field_ref): New.
(unsafe_copy_elision_p): New.
(build_over_call): Use it.

From-SVN: r228602

tree-vect-loop.c (vect_analyze_loop_operations): Skip virtual phi in the tail of outer-loop.

gcc/

* tree-vect-loop.c (vect_analyze_loop_operations): Skip virtual phi
in the tail of outer-loop.

gcc/testsuite/

* gcc.dg/vect/vect-outer-simd-3.c: New test.

From-SVN: r228601

rs6000.c (rs6000_xcoff_debug_unwind_info): Always return UI_NONE.

* config/rs6000/rs6000.c (rs6000_xcoff_debug_unwind_info): Always
return UI_NONE.

From-SVN: r228600

tree-ssa-loop-unswitch.c: Include "gimple-iterator.h" and "cfghooks.h"...

gcc/

* tree-ssa-loop-unswitch.c: Include "gimple-iterator.h" and
"cfghooks.h", add prototypes for introduced new functions.
(tree_ssa_unswitch_loops): Use from innermost loop iterator, move all
checks on ability of loop unswitching to tree_unswitch_single_loop;
invoke tree_unswitch_single_loop or tree_unswitch_outer_loop depending
on innermost loop check.
(tree_unswitch_single_loop): Add all required checks on ability of
loop unswitching under zero recursive level guard.
(tree_unswitch_outer_loop): New function.
(find_loop_guard): Likewise.
(empty_bb_without_guard_p): Likewise.
(used_outside_loop_p): Likewise.
(get_vop_from_header): Likewise.
(hoist_guard): Likewise.
(check_exit_phi): Likewise.

gcc/testsuite/

* gcc.dg/loop-unswitch-2.c: New test.
* gcc.dg/loop-unswitch-3.c: Likewise.
* gcc.dg/loop-unswitch-4.c: Likewise.

From-SVN: r228599

tree-ssa-reassoc.c (dump_ops_vector): Print newline after each ops element.

* tree-ssa-reassoc.c (dump_ops_vector): Print newline after each
ops element.

From-SVN: r228598

Dump only required dse1 file

2015-10-08  Tom de Vries  <tom@codesourcery.com>

* gcc.dg/dse.c: Only dump in dse1 pass.
* gcc.dg/tree-ssa/pr30375.c: Same.

From-SVN: r228597

S/390: Use create_tmp_var_raw in s390_atomic_assign_expand_fenv.

gcc/ChangeLog:

2015-10-08  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

PR c/65345
* config/s390/s390.c (s390_atomic_assign_expand_fenv): Use
create_tmp_var_raw instead of create_tmp_var.

From-SVN: r228594

Daily bump.

From-SVN: r228590

expr.c (store_expr_with_bounds): Handle aggregate moves from BLKmode.

* expr.c (store_expr_with_bounds): Handle aggregate moves from
BLKmode.
* gimple-expr.c (useless_type_conversion_p): Do not use TYPE_CANONICAL
to define gimple type system; compare aggregates only by size.

From-SVN: r228586

Re: [PATCH] Improve DOM's optimization of control statements

* tree-ssa-dom.c (optimize_stmt): Don't set LOOPS_NEED_FIXUP here.
* tree-ssa-threadupdate.c (remove_ctrl_stmt_and_useless_edges): Do it
here instead.  Tighten test to avoid setting LOOPS_NEED_FIXUP
unnecessarily.

@@ -1848,12 +1848,6 @@ optimize_stmt (basic_block bb, gimple_stmt_iterator si,

From-SVN: r228585

gather bbs and conditions in a single walk through dominators

Clean up the function to build scop's basic blocks and the function
that gathers the conditions under which a basic block is executed.  We remove
one traversal of the dominator tree.

This refactoring was triggered by the need of a vec<bb> of all the basic blocks
in a region.  We will use that vector in a patch that removes the out-of-ssa
translation of scalar dependences: we will iterate through the basic blocks of a
region to record scalar dependences crossing bbs or going out of the region.

The patch passes bootstrap and regtest on x86_64-linux.

2015-10-06  Aditya Kumar  <aditya.k7@samsung.com>
                Sebastian Pop  <s.pop@samsung.com>

                * graphite-dependences.c (scop_get_dependences): Do not use SCOP_BBS.
                * graphite-isl-ast-to-gimple.c (get_max_schedule_dimensions): Same.
                (generate_isl_schedule): Same.
                * graphite-optimize-isl.c (scop_get_domains): Same.
                (apply_schedule_map_to_scop): Same.
                * graphite-poly.c (print_iteration_domains): Same.
                (remove_gbbs_in_scop): Same.
                (new_scop): Same.
                (free_scop): Same.
                (print_scop): Same.
                * graphite-poly.h (struct scop): Rename bbs to pbbs.
                (SCOP_BBS): Remove.
                * graphite-scop-detection.c (compare_bb_depths): Remove.
                (graphite_sort_dominated_info): Remove.
                (try_generate_gimple_bb): Move out of scop_detection.
                (all_non_dominated_preds_marked_p): Remove.
                (build_scop_bbs_1): Remove.
                (build_scop_bbs): Remove.
                (nb_pbbs_in_loops): Do not use SCOP_BBS.
                (find_scop_parameters): Same.
                (sese_dom_walker): Rename gather_bbs.
                (before_dom_children): Call try_generate_gimple_bb and collect gbb
                and pbb.
                (build_scops): Call gather_bbs.
                * graphite-sese-to-poly.c (build_scop_scattering): Do not use SCOP_BBS.
                (add_conditions_to_constraints): Same.
                (build_scop_iteration_domain): Same.
                (build_scop_drs): Same.
                (new_pbb_from_pbb): Same.
                * sese.c (new_sese_info): Create bbs.
                * sese.h (struct sese_info_t): Add bbs.

Co-Authored-By: Sebastian Pop <s.pop@samsung.com>
From-SVN: r228581

xcoff.h (ASM_PREFERRED_EH_DATA_FORMAT): Use 64-bit encoding in 64-bit mode.

* config/rs6000/xcoff.h (ASM_PREFERRED_EH_DATA_FORMAT): Use 64-bit
encoding in 64-bit mode.

From-SVN: r228579

compiler: Don't make temporaries for constant operands.

    Fixes golang/go#12616.

    Reviewed-on: https://go-review.googlesource.com/14645

From-SVN: r228578

re PR target/66697 (Feature request: -mstackrealign and force_align_arg_pointer for x86_64)

PR target/66697
* config/i386/i386.c (ix86_option_override_internal): Always use
8-byte minimum stack boundary in 64-bit mode.
(ix86_compute_frame_layout): Remove assert on INCOMING_STACK_BOUNDARY.
(ix86_emit_save_reg_using_mov): Support unaligned SSE store.
Add a REG_CFA_EXPRESSION note if needed.
(ix86_emit_restore_sse_regs_using_mov): Support unaligned SSE load.
(ix86_handle_force_align_arg_pointer_attribute): New.
(ix86_minimum_incoming_stack_boundary): Remove TARGET_64BIT check.
(ix86_attribute_table): Set ix86_force_align_arg_pointer_string
with ix86_handle_force_align_arg_pointer_attribute.
* config/i386/i386.h (MIN_STACK_BOUNDARY): Set to BITS_PER_WORD.

testsuite/ChangeLog:

PR target/66697
* gcc.target/i386/20060512-1.c: Remove ia32 requirement.
(PUSH, POP): New defines.
(sse2_test): Use PUSH and POP to misalign runtime stack.
* gcc.target/i386/20060512-2.c: Remove ia32 requirement.

From-SVN: r228577

re PR go/67874 (fd_unix.go does not build when there is fcntl64 and no fcntl syscall)

PR go/67874
    net, runtime: Call C library fcntl function rather than syscall.Syscall.

    Not all systems define a fcntl syscall; some only have fcntl64.

    Fixes GCC PR go/67874.

    Reviewed-on: https://go-review.googlesource.com/15497

From-SVN: r228576

atomic-dg.exp (atomic_link_flags): Move flag setting to atomic_init.

* testsuite/lib/atomic-dg.exp (atomic_link_flags): Move flag setting
to atomic_init. (atomic_init): Restrict flags usage.

From-SVN: r228575

remove unused code

2015-10-07  Aditya Kumar  <aditya.k7@samsung.com>
                Sebastian Pop  <s.pop@samsung.com>

                * graphite-scop-detection.c (parameter_index_in_region): Remove
                use of SESE_ADD_PARAMS.
                (find_scop_parameters): Same.
                * sese.c (new_sese_info): Same.
                * sese.h (struct sese_info_t): Remove add_params.
                (SESE_ADD_PARAMS): Remove.

Co-Authored-By: Sebastian Pop <s.pop@samsung.com>
From-SVN: r228574

move graphite bookkeeping from sese to sese_info

2015-10-06  Aditya Kumar  <aditya.k7@samsung.com>
                Sebastian Pop  <s.pop@samsung.com>

                * graphite-isl-ast-to-gimple.c (translate_isl_ast_to_gimple): Use
                an sese_info_p.
                (copy_def): Same.
                (copy_internal_parameters): Same.
                (translate_isl_ast_to_gimple): Use an sese_l.
                (build_iv_mapping): Same.
                * graphite-poly.c (new_sese): Rename new_sese_info.
                (free_sese): Rename free_sese_info.
                * graphite-poly.h (struct scop): Use an sese_info_p.
                (scop_set_region): Same.
                * graphite-scop-detection.c (struct sese_l): Moved...
                (get_entry_bb): Moved...
                (get_exit_bb): Moved...
                (parameter_index_in_region_1): Use an sese_info_p.
                (parameter_index_in_region): Same.
                (scan_tree_for_params): Same.
                (find_params_in_bb): Same.
                (sese_dom_walker): Use an sese_l.
                * graphite-sese-to-poly.c (remove_invariant_phi): Same.
                (reduction_phi_p): Same.
                (parameter_index_in_region_1): Use an sese_info_p.
                (propagate_expr_outside_region): Use an sese_l.
                * graphite.c: Replace uses of SCOP_REGION.
                * sese.c (sese_record_loop): Use an sese_info_p.
                (build_sese_loop_nests): Same.
                (sese_build_liveouts_use): Same.
                (sese_build_liveouts_bb): Same.
                (sese_build_liveouts_bb): Same.
                (sese_bad_liveouts_use): Same.
                (sese_reset_debug_liveouts_bb): Same.
                (sese_build_liveouts): Same.
                (new_sese): Renamed new_sese_info.
                (free_sese): Renamed free_sese_info.
                (set_rename): Use an sese_info_p.
                (graphite_copy_stmts_from_block): Same.
                (copy_bb_and_scalar_dependences): Same.
                (outermost_loop_in_sese_1): Use an sese_l.
                (outermost_loop_in_sese): Same.
                (if_region_set_false_region): Use an sese_info_p.
                (move_sese_in_condition): Same.
                (scalar_evolution_in_region): Use an sese_l.
                * sese.h (struct sese_l): ... here.
                (SESE_ENTRY): Remove.
                (SESE_ENTRY_BB): Remove.
                (SESE_EXIT): Remove.
                (SESE_EXIT_BB): Remove.
                (sese_contains_loop): Use an sese_info_p.
                (sese_nb_params): Same.
                (bb_in_sese_p): Use an sese_l.
                (stmt_in_sese_p): Same.
                (defined_in_sese_p): Same.
                (loop_in_sese_p): Same.
                (sese_loop_depth): Same.
                (struct ifsese_s): Use an sese_info_p.
                (gbb_loop_at_index): Use an sese_l.
                (nb_common_loops): Same.
                (scev_analyzable_p): Same.

Co-Authored-By: Sebastian Pop <s.pop@samsung.com>
From-SVN: r228572

re PR sanitizer/67867 (UBSan bootstrap fails with error: ‘otype’ may be used uninitialized in this function [-Werror=maybe-uninitialized])

PR sanitizer/67867
* search.c (accessible_p): Initialize OTYPE to NULL_TREE.

From-SVN: r228569

Define x86 CALL_USED_REGISTERS_MASK

Define x86 CALL_USED_REGISTERS_MASK used on x86 CALL_USED_REGISTERS.

* config/i386/i386.c (ix86_conditional_register_usage): Use
CALL_USED_REGISTERS_MASK.
* config/i386/i386.h (CALL_USED_REGISTERS_MASK): New macro.

From-SVN: r228568

Check $READELF_FOR_TARGET for readelf in configure

Similar to as, ld, nm and objdump, gcc configure should check
$READELF_FOR_TARGET for readelf.

PR bootstrap/67385
* configure.ac (gcc_cv_readelf): Check $READELF_FOR_TARGET.
* configure: Regenerated.

From-SVN: r228567

re PR fortran/65889 ([6 Regressions] [OOP] ICE with sizeof a polymorphic variable.)

gcc/fortran/ChangeLog:

2015-10-07  Andre Vehreschild  <vehre@gcc.gnu.org>

PR fortran/65889
* trans-intrinsic.c (gfc_conv_intrinsic_sizeof): Handle pointer to and
on stack class objects as sizeof parameter.

gcc/testsuite/ChangeLog:

2015-10-07  Andre Vehreschild  <vehre@gcc.gnu.org>

PR fortran/65889
* gfortran.dg/sizeof_5.f90: New test.

From-SVN: r228566

Merge ix86_maybe_switch_abi with ix86_set_current_function

ix86_maybe_switch_abi is called to late during RTL expansion and we
use the stale information from compilation of the previous function.
aggregate_value_p uses call_used_regs.  aggregate_value_p is used by
IPA and return value optimization, which are called before
ix86_maybe_switch_abi is called.  This patch merges ix86_maybe_switch_abi
with ix86_set_current_function.

PR target/67850
* config/i386/i386.c (ix86_maybe_switch_abi): Merged with ...
(ix86_set_current_function): This.
(TARGET_EXPAND_TO_RTL_HOOK): Removed.

From-SVN: r228565

Check in forgotten test for fortran/65766

From-SVN: r228564

tree-vectorizer.h (stmt_vec_info_vec): Use vec<stmt_vec_info>.

2015-10-07  Richard Biener  <rguenther@suse.de>

* tree-vectorizer.h (stmt_vec_info_vec): Use vec<stmt_vec_info>.
(vinfo_for_stmt): Adjust.
(set_vinfo_for_stmt): Likewise.
* tree-vectorizer.c (stmt_vec_info_vec): Likewise.
* tree-vect-stmts.c (free_stmt_vec_info_vec): Likewise.
* tree-vect-loop.c (new_loop_vec_info): Remove special-casing
of inner loop.
(vect_analyze_loop_1): Remove.
(vect_analyze_loop_form_1): Avoid building a loop_vec_info for
inner loop when vectorizing an outer loop by splitting out from ...
(vect_analyze_loop_form): ... here.

From-SVN: r228563

Fix PR c/65345 for arm

2015-10-07  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>

PR c/65345
* config/arm/arm-builtins.c (arm_atomic_assign_expand_fenv):
Use create_tmp_var_raw instead of create_tmp_var.

From-SVN: r228562

Cache reals for 1/4, 1/6 and 1/9

We have a global 1/2 and a cached 1/3, but recalculate 1/4, 1/6 and 1/9
each time we need them.  That seems a bit arbitrary and makes the folding
code more noisy (especially once it's moved to match.pd).

This patch caches the other three constants too.  Bootstrapped &
regression-tested on x86_64-linux-gnu.

gcc/
* real.h (dconst_quarter, dconst_sixth, dconst_ninth): New macros.
(dconst_quarter_ptr, dconst_sixth_ptr, dconst_ninth_ptr): Declare.
* real.c (CACHED_FRACTION): New helper macro.
(dconst_third_ptr): Use it.
(dconst_quarter_ptr, dconst_sixth_ptr, dconst_ninth_ptr): New.
* builtins.c (fold_builtin_sqrt): Use dconst_quarter and
dconst_sixth.
(fold_builtin_cbrt): Use dconst_sixth and dconst_ninth.

From-SVN: r228561

cp-gimplify.c (genericize_if_stmt): Use protected_set_expr_location.

* cp-gimplify.c (genericize_if_stmt): Use protected_set_expr_location.
(genericize_cp_loop): Likewise.
* decl.c (cxx_maybe_build_cleanup): Likewise.
* parser.c (cp_parser_binary_expression): Likewise.
(cp_parser_omp_for_loop): Likewise.
(cp_parser_omp_construct): Likewise.
* semantics.c (finish_transaction_stmt): Likewise.
(build_transaction_expr): Likewise.

From-SVN: r228560

[PATCH][PR tree-optimization/67816] Fix jump threading when DOM removes conditionals in jump threading path

PR tree-optimization/67816
* tree-ssa-threadupdate.h (remove_jump_threads_including): Renamed
from remove_jump_threads_starting_at.  Accept an edge rather than
a basic block.
* tree-ssa-threadupdate.c (removed_edges): New hash table.
(remove_jump_threads_including): Note edges that get removed from
the CFG for later pruning of jump threading paths including them.
(thread_through_all_blocks): Remove paths which include edges that
have been removed.
* tree-ssa-dom.c (optimize_stmt): Call remove_jump_threads_including
on each outgoing edges when optimizing away a control statement.

* gcc.c-torture/compile/pr67816.c: New test.

From-SVN: r228559

reorg.c: use vec<rtx_insn *> instead of rtx_insn_list for the delay insn list

gcc/ChangeLog:

2015-10-06  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* reorg.c (emit_delay_sequence): Store list of delay slot insns
in a vector instead of rtx_insn_list.
(add_to_delay_list): Likewise.
(delete_from_delay_slot): Likewise.
(optimize_skip): Likewise.
(redirect_with_delay_list_safe_p): Likewise.
(check_annul_list_true_false): Likewise.
(steal_delay_list_from_target): Likewise.
(steal_delay_list_from_fallthrough): Likewise.
(redundant_insn): Likewise.
(fill_simple_delay_slots): Likewise.
(fill_slots_from_thread): Likewise.
(fill_eager_delay_slots): Likewise.
(relax_delay_slots): Likewise.

From-SVN: r228558

re PR c++/67810 (Non-expression recognized as fold expression)

PR c++/67810
* parser.c (cp_parser_fold_expr_p): Remove.
(is_binary_op): New.
(cp_parser_fold_expression): Take LHS as parameter.
(cp_parser_primary_expression): Call it after parsing an expression.
(cp_parser_binary_expression, cp_parser_assignment_operator_opt)
(cp_parser_expression): Ignore an operator followed by '...'.
(is_binary_op): New.
* pt.c (tsubst_unary_left_fold, tsubst_binary_left_fold)
(tsubst_unary_right_fold, tsubst_binary_right_fold): Handle errors.

From-SVN: r228556

Daily bump.

From-SVN: r228555

[multiple changes]

2015-10-06  Louis Krupp <louis.krupp@zoho.com>
PR fortran/65766
* resolve.c (gfc_resolve_substring_charlen): For derived type,
use typespec of string component when resolving substring length.

2015-10-06  Louis Krupp     <louis.krupp@zoho.com>

PR fortran/65766
* gfortran.dg/substr_alloc_string_comp_1.f90: New.

From-SVN: r228551

compiler: Record each import as a distinct alias.

    This patch introduces the Package_alias class which is a
    finer representation of the different between a package and the
    aliases it is imported under.  Each alias keeps track of the location
    of its import statement and how many times that alias has been used.
    This allows the gofrontend to report when a specific import has not
    been used even if a symbol from the package has been used by another
    import.

    Fixes golang/go#12326.

    Reviewed-on: https://go-review.googlesource.com/14259

From-SVN: r228550

nios2.c (nios2_symbol_ref_in_small_data_p): For -mgpopt=local, also exclude unintialized common symbols.

2015-10-06  Sandra Loosemore  <sandra@codesourcery.com>

gcc/
* config/nios2/nios2.c (nios2_symbol_ref_in_small_data_p):
For -mgpopt=local, also exclude unintialized common symbols.
* doc/invoke.texi (Nios II Options): Document the change.

From-SVN: r228549

iterators.md (vwcore): Add missing cases for V4HF/V8HF modes.

gcc/ChangeLog:

2015-10-07  Kugan Vivekanandarajah  <kuganv@linaro.org>

* config/aarch64/iterators.md (vwcore): Add missing cases for
 V4HF/V8HF modes.

gcc/testsuite/ChangeLog:

2015-10-07  Kugan Vivekanandarajah  <kuganv@linaro.org>

* gcc.target/aarch64/get_lane_f16_1.c: New test.

From-SVN: r228547

move dr->alias_set to a helper structure

2015-10-06  Aditya Kumar  <aditya.k7@samsung.com>
                Sebastian Pop  <s.pop@samsung.com>

                * graphite-poly.c (new_scop): Initialize drs.
                * graphite-poly.h (struct dr_info): New.
                (struct scop): Add drs.
                * graphite-sese-to-poly.c (pdr_add_alias_set): Use dr_info.
                (pdr_add_memory_accesses): Same.
                (build_poly_dr): Same.
                (build_alias_set): Same.
                (build_scop_drs): Same.
                (build_pbb_drs): Remove.
                * tree-data-ref.c (create_data_ref): Do not initialize alias_set.
                * tree-data-ref.h (data_reference): Remove alias_set.

Co-Authored-By: Sebastian Pop <s.pop@samsung.com>
From-SVN: r228545

remove unused struct base_alias_pair

2015-10-06  Aditya Kumar  <aditya.k7@samsung.com>
                Sebastian Pop  <s.pop@samsung.com>

                * graphite-poly.c (free_data_refs_aux): Remove.
                (free_gimple_poly_bb): Do not call free_data_refs_aux.
                * graphite-poly.h (struct base_alias_pair): Remove.
                * graphite-sese-to-poly.c (pdr_add_alias_set): Remove all uses of
                base_alias_pair and dr->aux.
                (build_alias_set): Same.
                * tree-data-ref.c (create_data_ref): Initialize alias_set.
                * tree-data-ref.h (data_reference): Add alias_set.

Co-Authored-By: Sebastian Pop <s.pop@samsung.com>
From-SVN: r228544