re PR target/49687 ([avr] Missed optimization for widening MUL)

PR target/49687
* config/avr/avr.md (mulsi3, *mulsi3, mulu<mode>si3,
muls<mode>si3, mulohisi3, mulhisi3, umulhisi3, usmulhisi3,
*<any_extend:extend_prefix><any_extend2:extend_prefix>mul<QIHI:mode><QIHI2:mode>si3):
Add X to register footprint: Clobber r26/r27.

From-SVN: r176923

builtins.c (fold_builtin_signbit): Build the comparison with a proper type.

2011-07-29  Richard Guenther  <rguenther@suse.de>

* builtins.c (fold_builtin_signbit): Build the comparison
with a proper type.

From-SVN: r176922

re PR middle-end/49893 (-O3 generates wrong code for volatile)

2011-07-29  Richard Guenther  <rguenther@suse.de>

PR tree-optimization/49893
* tree-predcom.c (suitable_reference_p): Volatile references
are not suitable.

From-SVN: r176921

re PR target/49313 (Inefficient libgcc implementations for avr)

PR target/49313
* config/avr/libgcc.S (__ffshi2): Don't skip 2-word instruction.
(__ctzsi2): Result for 0 may be undefined.
(__ctzhi2): Result for 0 may be undefined.
(__popcounthi2): Don't clobber r30. Use __popcounthi2_tail.
(__popcountsi2): Ditto. And don't clobber r26.
(__popcountdi2): Ditto. And don't clobber r27.
* config/avr/avr.md (UNSPEC_COPYSIGN): New c_enum.
(parityhi2): New expand.
(paritysi2): New expand.
(popcounthi2): New expand.
(popcountsi2): New expand.
(clzhi2): New expand.
(clzsi2): New expand.
(ctzhi2): New expand.
(ctzsi2): New expand.
(ffshi2): New expand.
(ffssi2): New expand.
(copysignsf3): New insn.
(bswapsi2): New expand.
(*parityhi2.libgcc): New insn.
(*parityqihi2.libgcc): New insn.
(*paritysihi2.libgcc): New insn.
(*popcounthi2.libgcc): New insn.
(*popcountsi2.libgcc): New insn.
(*popcountqi2.libgcc): New insn.
(*popcountqihi2.libgcc): New insn-and-split.
(*clzhi2.libgcc): New insn.
(*clzsihi2.libgcc): New insn.
(*ctzhi2.libgcc): New insn.
(*ctzsihi2.libgcc): New insn.
(*ffshi2.libgcc): New insn.
(*ffssihi2.libgcc): New insn.
(*bswapsi2.libgcc): New insn.

From-SVN: r176920

re PR tree-optimization/47407 (gcc.dg/tree-ssa/pr42585.c FAILs on Tru64 UNIX V5.1B)

PR tree-optimization/47407
* gcc.dg/tree-ssa/pr42585.c: XFAIL scan-tree-dump-times on
alpha*-dec-osf*.
Sort target list.

From-SVN: r176919

tree-vrp.c (get_value_range): Only set parameter default definitions to varying, leave others at undefined.

2011-07-29  Richard Guenther  <rguenther@suse.de>

* tree-vrp.c (get_value_range): Only set parameter default
definitions to varying, leave others at undefined.
(extract_range_from_binary_expr): Fix undefined handling.
(vrp_visit_phi_node): Handle merged undefined state.

* gcc.dg/uninit-suppress.c: Also disable VRP.
* gcc.dg/uninit-suppress_2.c: Likewise.

From-SVN: r176918

re PR c++/49808 (GCC adds an address-of somewhere!)

PR c++/49808
* pt.c (tsubst) [TEMPLATE_PARM_INDEX]: Call convert_from_reference.
(convert_nontype_argument, tsubst_template_arg): Handle its output.

From-SVN: r176916

Add a testcase for PR rtl-optimization/47958.

2011-07-28  H.J. Lu  <hongjiu.lu@intel.com>

PR rtl-optimization/47958
* gcc.dg/torture/pr47958-1.c: New.

From-SVN: r176914

Add x32 support to config/i386/morestack.S.

2011-07-28  H.J. Lu  <hongjiu.lu@intel.com>

* config/i386/morestack.S (X86_64_SAVE_NEW_STACK_BOUNDARY): New.
Use X86_64_SAVE_NEW_STACK_BOUNDARY to save the new stack boundary
for x86-64.  Properly check __x86_64__ and __LP64__.

From-SVN: r176912

re PR rtl-optimization/49799 (gcc arm generates illegal sbfx instruction)

PR rtl-optimization/49799

* combine.c (make_compound_operation): Check if the bit field is valid
before change it to bit field extraction.

* gcc.dg/pr49799.c: New test case.

From-SVN: r176911

Daily bump.

From-SVN: r176910

re PR rtl-optimization/49891 (ICE in redirect_jump_1)

PR rtl-optimization/49891
* cfgrtl.c (force_nonfallthru_and_redirect): Set JUMP_LABEL for
newly created returnjumps.

From-SVN: r176905

expr.c (expand_expr_addr_expr_1): Detect a user request for a local frame in a naked function...

* expr.c (expand_expr_addr_expr_1): Detect a user request for a
local frame in a naked function, and produce a suitable error for
that specific case.

From-SVN: r176904

m32c.c (m32c_secondary_reload_class): Allow PSI registers to be reloaded in HI classes when the target is HI.

* config/m32c/m32c.c (m32c_secondary_reload_class): Allow PSI
registers to be reloaded in HI classes when the target is HI.

From-SVN: r176903

Replace v1, v2, lb, ub with bound_one, bound_two

2011-07-28  Sebastian Pop  <sebastian.pop@amd.com>

* graphite-clast-to-gimple.c: Replace v1, v2, lb, ub with
bound_one, bound_two.

From-SVN: r176902

Fix PR48648: Handle CLAST assignments.

The CLAST produced by CLooG-ISL contains an assignment and GCC chokes
on it.  The exact CLAST contains an assignment followed by an if:

scat_1 = max(0,ceild(T_4-7,8));
if (scat_1 <= min(1,floord(T_4-1,8))) {
  S7(scat_1);
}

This is equivalent to a loop that iterates only once, and so CLooG
generates an assignment followed by an if instead of a loop.  This is
an important optimization that was improved in ISL, that allows
if-conversion: imagine GCC having to figure out that a loop like the
following actually iterates only once, and can be converted to an if:

for (scat_1 = max(0,ceild(T_4-7,8)); scat_1 <= min(1,floord(T_4-1,8)); scat_1++)
  S7(scat_1);

This patch implements the translation of CLAST assignments.
Bootstrapped and tested on amd64-linux.

2011-07-28  Sebastian Pop  <sebastian.pop@amd.com>

PR middle-end/48648
* graphite-clast-to-gimple.c (clast_get_body_of_loop): Handle
CLAST assignments.
(translate_clast): Same.
(translate_clast_assignment): New.

* gcc.dg/graphite/id-pr48648.c: New.

From-SVN: r176901

Fix PR49876: Continue code generation with integer_zero_node on gloog_error

When setting gloog_error, graphite should continue code generation
without early returns, as otherwise the SSA representation would not
be complete.  So set the new expression to integer_zero_node, that
would not require more SSA updates, and continue code generation as
nothing happened.

Regstrapped on amd64-linux.

2011-07-28  Sebastian Pop  <sebastian.pop@amd.com>

PR tree-optimization/49876
* sese.c (rename_uses): Do not return false on gloog_error: set
the new_expr to integer_zero_node and continue code generation.
(graphite_copy_stmts_from_block): Remove early exit on gloog_error.

From-SVN: r176900

re PR c++/49813 ([C++0x] sinh vs asinh vs constexpr)

2011-07-28  Paolo Carlini  <paolo.carlini@oracle.com>

        PR c++/49813
        * semantics.c (potential_constant_expression_1):  Handle FMA_EXPR.

Checking this in for Paolo.

From-SVN: r176899

line-map.h (struct line_maps): Remove unused field last_listed.

* include/line-map.h (struct line_maps):
Remove unused field last_listed. Update all users.

From-SVN: r176898

re PR fortran/31067 (MINLOC should sometimes be inlined (gas_dyn is sooooo sloooow))

PR fortran/31067
* frontend-passes.c (optimize_minmaxloc): New function.
(optimize_expr): Call it.

* gfortran.dg/maxloc_2.f90: New test.
* gfortran.dg/maxloc_3.f90: New test.
* gfortran.dg/minloc_1.f90: New test.
* gfortran.dg/minloc_2.f90: New test.
* gfortran.dg/minloc_3.f90: New test.
* gfortran.dg/minmaxloc_7.f90: New test.

From-SVN: r176897

Use long long for 64bit int.

2010-07-28  H.J. Lu  <hongjiu.lu@intel.com>

* config/i386/64/sfp-machine.h (_FP_W_TYPE): Always use _WIN64
version.
(_FP_WS_TYPE): Likewise.
(_FP_I_TYPE): Likewise.

From-SVN: r176894

Check ia32 instead of ilp32 for ia32 tests.

2011-07-28  H.J. Lu  <hongjiu.lu@intel.com>

* testsuite/lib/libgomp.exp (libgomp_init): Add -march=i486
for ia32 instead of ilp32.

* testsuite/libgomp.c/atomic-1.c: Require ia32 instead of ilp32.
* testsuite/libgomp.c/atomic-6.c: Likewise.

From-SVN: r176893

re PR debug/49846 (entryval: missing DW_TAG_GNU_call_site_parameter for stack-passed `double' parameter)

PR debug/49846
* var-tracking.c (prepare_call_arguments): For non-MODE_INT stack
arguments also check if they aren't initialized with a MODE_INT
mode of the same size.

From-SVN: r176892

expr.c (get_bit_range): Handle *MEM_REF's.

        * expr.c (get_bit_range): Handle *MEM_REF's.

From-SVN: r176891

rtlanal.c (tablejump_p): False for returns.

* rtlanal.c (tablejump_p): False for returns.
* reorg.c (first_active_target_insn): New static function.
(find_end_label): Set JUMP_LABEL for a new returnjump.
(optimize_skip, get_jump_flags, rare_destination,
mostly_true_jump, get_branch_condition,
steal_delay_list_from_target, own_thread_p,
fill_simple_delay_slots, follow_jumps, fill_slots_from_thread,
fill_eager_delay_slots, relax_delay_slots, make_return_insns,
dbr_schedule): Adjust to handle ret_rtx in JUMP_LABELs.
* jump.c (delete_related_insns): Likewise.
(jump_to_label_p): New function.
(redirect_target): New static function.
(redirect_exp_1): Use it.  Adjust to handle ret_rtx in JUMP_LABELS.
(redirect_jump_1): Assert that the new label is nonnull.
(redirect_jump): Likewise.
(redirect_jump_2): Check for ANY_RETURN_P rather than NULL labels.
* ifcvt.c (find_if_case_1): Take care when redirecting jumps to the
exit block.
(dead_or_predicable): Change NEW_DEST arg to DEST_EDGE.  All callers
changed.  Ensure that the right label is passed to redirect_jump.
* function.c (emit_return_into_block,
thread_prologue_and_epilogue_insns): Ensure new returnjumps have
ret_rtx in their JUMP_LABEL.
* print-rtl.c (print_rtx): Handle ret_rtx in a JUMP_LABEL.
* emit-rtl.c (skip_consecutive_labels): Allow the caller to
pass ret_rtx as label.
* cfglayout.c (fixup_reorder_chain): Use
force_nonfallthru_and_redirect rather than force_nonfallthru.
(duplicate_insn_chain): Copy JUMP_LABELs for returns.
* rtl.h (ANY_RETURN_P): New macro.
(jump_to_label_p): Declare.
* resource.c (find_dead_or_set_registers): Handle ret_rtx in
JUMP_LABELs.
(mark_target_live_regs): Likewise.
* basic-block.h (force_nonfallthru_and_redirect): Declare.
* cfgrtl.c (force_nonfallthru_and_redirect): No longer static.
* config/alpha/alpha.c (alpha_tablejump_addr_vec,
alpha_tablejump_best_label): Remove functions.
* config/alpha/alpha-protos.c (alpha_tablejump_addr_vec,
alpha_tablejump_best_label): Remove declarations.
* config/sh/sh.c (barrier_align, split_branches): Adjust for
ret_rtx in JUMP_LABELs.
* config/arm/arm.c (is_jump_table): Likewise.

From-SVN: r176881

predicates.md (pic_32bit_opreand): Do not define as special predicate.

* config/i386/predicates.md (pic_32bit_opreand): Do not define as
special predicate.  Remove explicit mode checks.

From-SVN: r176879

dwarf2out.c (resolve_addr): For -gdwarf-2 don't optimize DW_AT_data_member_location containing just...

* dwarf2out.c (resolve_addr): For -gdwarf-2 don't
optimize DW_AT_data_member_location containing just
DW_OP_plus_uconst.

From-SVN: r176878

re PR debug/49871 (-gdwarf-3 creates invalid DWARF3 with DW_AT_data_member_location attribute)

PR debug/49871
* dwarf2out.c (size_of_die, value_format, output_die): Use
DW_FORM_udata instead of DW_FORM_data[48] for
dw_val_class_unsigned_const DW_AT_data_member_location for DWARF 3.

* gcc.dg/debug/dwarf2/pr49871.c: New test.

From-SVN: r176876

* ChangeLog: Remove mess.

From-SVN: r176874

Update length attribute on *tls_global_dynamic_64.

2011-07-28  H.J. Lu  <hongjiu.lu@intel.com>

* config/i386/i386.md (*tls_global_dynamic_64): Update length
attribute.

From-SVN: r176873

re PR target/47715 ([x32] TLS doesn't work)

PR target/47715
* config/i386/i386.md (*tls_global_dynamic_64): Remove mode from
tls_symbolic_operand check.  Update code sequence for TARGET_X32.
(tls_global_dynamic_64): Remove mode from tls_symbolic_operand check.
(tls_dynamic_gnu2_64): Ditto.
(*tls_dynamic_gnu2_lea_64): Ditto.
(*tls_dynamic_gnu2_call_64): Ditto.
(*tls_dynamic_gnu2_combine_64): Ditto.

From-SVN: r176872

Set need_64bit_hwint to yes for x86 targets.

gcc/

2011-07-28  H.J. Lu  <hongjiu.lu@intel.com>

* config.gcc: Set need_64bit_hwint to yes for x86 targets.

libcpp/

2011-07-28  H.J. Lu  <hongjiu.lu@intel.com>

* configure.ac: Set need_64bit_hwint to yes for x86 targets.
* configure: Regenerated.

From-SVN: r176871

init.c (__gnat_error_handler): Cast reason to int.

* init.c (__gnat_error_handler): Cast reason to int.
(__gnat_install_handler): Explain sa_sigaction use.

From-SVN: r176870

Only expand strlen to Pmode.

2011-07-28  H.J. Lu  <hongjiu.lu@intel.com>

PR target/47364
* config/i386/i386.md (strlen<mode>): Replace SWI48x with P.

From-SVN: r176868

vfp.md ("*movdf_vfp"): Handle the VFP constraints before the core constraints.

2011-07-28  Ramana Radhakrishnan  <ramana.radhakrishnan@linaro.org>

* config/arm/vfp.md ("*movdf_vfp"): Handle the VFP constraints
before the core constraints. Adjust attributes.
(*thumb2_movdf_vfp"): Likewise.

From-SVN: r176867

tree-vrp.c (simplify_stmt_using_ranges): Remove TRUTH_NOT_EXPR case.

2011-07-28  Kai Tietz  <ktietz@redhat.com>

        * tree-vrp.c (simplify_stmt_using_ranges): Remove
        TRUTH_NOT_EXPR case.
        (simplify_truth_ops_using_ranges): Likewise.
        (build_assert_expr_for): Likewise.
        (build_assert_expr_for_1): Remove TRUTH_NOT_EXPR case
        and handle BIT_NOT_EXPR for truth-operation.

From-SVN: r176866

re PR target/49313 (Inefficient libgcc implementations for avr)

PR target/49313
Undo r176835 from trunk
2011-07-27  Georg-Johann Lay

From-SVN: r176865

update copyright

From-SVN: r176863

re PR target/49687 ([avr] Missed optimization for widening MUL)

PR target/49687
* config/avr/t-avr (LIB1ASMFUNCS): Remove _xmulhisi3_exit.
Add _muluhisi3, _mulshisi3, _usmulhisi3.
* config/avr/libgcc.S (__mulsi3): Rewrite.
(__mulhisi3): Rewrite.
(__umulhisi3): Rewrite.
(__usmulhisi3): New.
(__muluhisi3): New.
(__mulshisi3): New.
(__mulohisi3): New.
(__mulqi3, __mulqihi3, __umulqihi3, __mulhi3): Use DEFUN/ENDF to
declare.
* config/avr/predicates.md (pseudo_register_operand): Rewrite.
(pseudo_register_or_const_int_operand): New.
(combine_pseudo_register_operand): New.
(u16_operand): New.
(s16_operand): New.
(o16_operand): New.
* config/avr/avr.c (avr_rtx_costs): Handle costs for mult:SI.
* config/avr/avr.md (QIHI, QIHI2): New mode iterators.
(any_extend, any_extend2): New code iterators.
(extend_prefix): New code attribute.
(mulsi3): Rewrite. Turn insn to expander.
(mulhisi3): Ditto.
(umulhisi3): Ditto.
(usmulhisi3): New expander.
(*mulsi3): New insn-and-split.
(mulu<mode>si3): New insn-and-split.
(muls<mode>si3): New insn-and-split.
(mulohisi3): New insn-and-split.
(*uumulqihisi3, *uumulhiqisi3, *uumulhihisi3, *uumulqiqisi3,
*usmulqihisi3, *usmulhiqisi3, *usmulhihisi3, *usmulqiqisi3,
*sumulqihisi3, *sumulhiqisi3, *sumulhihisi3, *sumulqiqisi3,
*ssmulqihisi3, *ssmulhiqisi3, *ssmulhihisi3, *ssmulqiqisi3): New
insn-and-split.
(*mulsi3_call): Rewrite.
(*mulhisi3_call): Rewrite.
(*umulhisi3_call): Rewrite.
(*usmulhisi3_call): New insn.
(*muluhisi3_call): New insn.
(*mulshisi3_call): New insn.
(*mulohisi3_call): New insn.
(extendqihi2): Use combine_pseudo_register_operand as predicate
for operand 1.
(extendqisi2): Ditto.
(zero_extendqihi2): Ditto.
(zero_extendqisi2): Ditto.
(zero_extendhisi2): Ditto.
(extendhisi2): Ditto. Don't early-clobber operand 0.

From-SVN: r176862

linux-unwind.h (frob_update_context <__powerpc64__>): Leave r2 REG_UNSAVED if stopped on the instruction that saves r2 in a plt call stub.

* config/rs6000/linux-unwind.h (frob_update_context <__powerpc64__>):
Leave r2 REG_UNSAVED if stopped on the instruction that saves r2
in a plt call stub.  Do restore r2 if stopped on bctrl.

From-SVN: r176861

i386.c (add->lea splitter): Add SWI mode to PLUS RTX.

* config/i386/i386.c (add->lea splitter): Add SWI mode to PLUS RTX.

From-SVN: r176860

i386.exp (check_effective_target_bmi): Make sure the builtin isn't optimized away.

* gcc.target/i386/i386.exp (check_effective_target_bmi): Make sure
the builtin isn't optimized away.

From-SVN: r176859

Daily bump.

From-SVN: r176856

re PR fortran/45586 (ICE non-trivial conversion at assignment)

2011-07-27  Tobias Burnus  <burnus@net-b.de>

        PR fortran/45586
        * trans-types.c (gfc_get_derived_type): Ensure that pointer
        component types are marked as nonrestricted.

2011-07-27  Tobias Burnus  <burnus@net-b.de>

        PR fortran/45586
        * gfortran.dg/lto/pr45586-2_0.f90: New.

From-SVN: r176852

re PR middle-end/49875 (FAIL: c-c++-common/cxxbitfields-4.c)

       PR middle-end/49875
        * c-c++-common/cxxbitfields-4.c: Check for smaller than long
        moves.
        * c-c++-common/cxxbitfields-5.c: Same.

From-SVN: r176848

re PR c++/49813 ([C++0x] sinh vs asinh vs constexpr)

2011-07-27  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/49813
* include/c_global/cmath: Use _GLIBCXX_CONSTEXPR and constexpr.

From-SVN: r176847

target-supports.exp (check_effective_target_mmap): Use check_function_available.

* lib/target-supports.exp (check_effective_target_mmap): Use
check_function_available.

From-SVN: r176846

Propagate the source location of a template's function_decl to the template_decl built out of it.

Propagate the source location of a template's function_decl to the
template_decl built out of it.  Without this, specializations
generated before the template was defined get the location of the
closing ')' instead of the location of the template's name, which can
produce inconsistent locations between translation units, which makes
gold's ODR checker unhappy.

* gcc/cp/pt.c (build_template_decl): Copy the function_decl's source
location to the new template_decl.
* gcc/testsuite/g++.old-deja/g++.pt/crash60.C: Updated.
* libstdc++-v3/testsuite/20_util/weak_ptr/comparison/cmp_neg.cc: Updated.

From-SVN: r176845

re PR target/49866 (-mcmodel=large tail call fails to assemble)

PR target/49866
* gcc.target/i386/pr49866.c: New test.

From-SVN: r176842

Add BMI testcases.

2011-07-27  Yukhin Kirill  <kirill.yukhin@intel.com>

* gcc.target/i386/i386.exp (check_effective_target_bmi): New.
* gcc.target/i386/bmi-andn-1.c: New test.
* gcc.target/i386/bmi-andn-1a.c: Likewise.
* gcc.target/i386/bmi-andn-2.c: Likewise.
* gcc.target/i386/bmi-andn-2a.c: Likewise.
* gcc.target/i386/bmi-bextr-1.c: Likewise.
* gcc.target/i386/bmi-bextr-1a.c: Likewise.
* gcc.target/i386/bmi-bextr-2.c: Likewise.
* gcc.target/i386/bmi-bextr-2a.c: Likewise.
* gcc.target/i386/bmi-blsi-1.c: Likewise.
* gcc.target/i386/bmi-blsi-1a.c: Likewise.
* gcc.target/i386/bmi-blsi-2.c: Likewise.
* gcc.target/i386/bmi-blsi-2a.c: Likewise.
* gcc.target/i386/bmi-blsmsk-1.c: Likewise.
* gcc.target/i386/bmi-blsmsk-1a.c: Likewise.
* gcc.target/i386/bmi-blsmsk-2.c: Likewise.
* gcc.target/i386/bmi-blsmsk-2a.c: Likewise.
* gcc.target/i386/bmi-blsr-1.c: Likewise.
* gcc.target/i386/bmi-blsr-1a.c: Likewise.
* gcc.target/i386/bmi-blsr-2.c: Likewise.
* gcc.target/i386/bmi-blsr-2a.c: Likewise.
* gcc.target/i386/bmi-tzcnt-1.c: Likewise.
* gcc.target/i386/bmi-tzcnt-1a.c: Likewise.
* gcc.target/i386/bmi-tzcnt-2.c: Likewise.
* gcc.target/i386/bmi-tzcnt-2a.c: Likewise.

From-SVN: r176840

Fix PR49471: canonicalize_loop_ivs should not generate unsigned types.

2011-07-27  Sebastian Pop  <sebastian.pop@amd.com>

PR tree-optimization/49471
* tree-ssa-loop-manip.c (canonicalize_loop_ivs): Build an unsigned
iv only when the largest type is unsigned.  Do not call
lang_hooks.types.type_for_size.

* testsuite/libgomp.graphite/force-parallel-1.c: Un-xfail.
* testsuite/libgomp.graphite/force-parallel-2.c: Adjust pattern.

From-SVN: r176838

PR45450: disable legality check after an openscop read

2011-07-27  Sebastian Pop  <sebastian.pop@amd.com>

PR middle-end/45450
* graphite-poly.c (apply_poly_transforms): Disable legality check
after an openscop read.

From-SVN: r176837

Fix PR47691: do not abort compilation when code generation fails

2011-07-27  Sebastian Pop  <sebastian.pop@amd.com>

PR middle-end/47691
* graphite-clast-to-gimple.c (translate_clast_user): Update use of
copy_bb_and_scalar_dependences.
* sese.c (rename_uses): Do not call gcc_assert.  Set gloog_error.
(graphite_copy_stmts_from_block): Update call to rename_uses.
(copy_bb_and_scalar_dependences): Update call to
graphite_copy_stmts_from_block.
* sese.h (copy_bb_and_scalar_dependences): Update declaration.

* gfortran.dg/graphite/id-pr47691.f: New.

From-SVN: r176836

re PR target/49313 (Inefficient libgcc implementations for avr)

PR target/49313
* config/avr/libgcc.S (__ffshi2): Don't skip 2-word instruction.
(__ctzsi2): Result for 0 may be undefined.
(__ctzhi2): Result for 0 may be undefined.
(__popcounthi2): Don't clobber r30. Use __popcounthi2_tail.
(__popcountsi2): Ditto. And don't clobber r26.
(__popcountdi2): Ditto. And don't clobber r27.
* config/avr/avr.md (UNSPEC_COPYSIGN): New c_enum.
(parityhi2): New expand.
(paritysi2): New expand.
(popcounthi2): New expand.
(popcountsi2): New expand.
(clzhi2): New expand.
(clzsi2): New expand.
(ctzhi2): New expand.
(ctzsi2): New expand.
(ffshi2): New expand.
(ffssi2): New expand.
(copysignsf3): New insn.
(bswapsi2): New expand.
(*parityhi2.libgcc): New insn.
(*parityqihi2.libgcc): New insn.
(*paritysihi2.libgcc): New insn.
(*popcounthi2.libgcc): New insn.
(*popcountsi2.libgcc): New insn.
(*popcountqi2.libgcc): New insn.
(*popcountqihi2.libgcc): New insn-and-split.
(*clzhi2.libgcc): New insn.
(*clzsihi2.libgcc): New insn.
(*ctzhi2.libgcc): New insn.
(*ctzsihi2.libgcc): New insn.
(*ffshi2.libgcc): New insn.
(*ffssihi2.libgcc): New insn.
(*bswapsi2.libgcc): New insn.

From-SVN: r176835

i386.c (ix86_expand_move): Do not explicitly check the mode of symbolic_opreand RTXes.

* config/i386/i386.c (ix86_expand_move): Do not explicitly check
the mode of symbolic_opreand RTXes.

From-SVN: r176834

predicates.md (x86_64_movabs_operand): Reject pic32_bit_operand operands.

* config/i386/predicates.md (x86_64_movabs_operand): Reject
pic32_bit_operand operands.
* config/i386/i386.c (ix86_expand_move): Check x86_64_movabs_operand
in DImode.

From-SVN: r176833

avx-os-support.h: New.

* gcc.target/i386/avx-os-support.h: New.
* gcc.target/i386/avx-check.h: Include avx-os-support.h
(main): Check avx_os_support before the test is run.
* gcc.target/i386/aes-avx-check.h: Ditto.
* gcc.target/i386/pclmul-avx-check.h: Ditto.

From-SVN: r176829

i386.c (ix86_option_override_internal): Allow -mabi for 32-bit, too.

        * config/i386/i386.c (ix86_option_override_internal): Allow -mabi
        for 32-bit, too.
        (ix86_handle_abi_attribute): Allow function attributes
        ms_abi/sysv_abi in 32-bit mode, too.
        * doc/extend.texi: Adjust attribute documentation.

        * gcc.target/i386/aggregate-ret3.c: New test.
        * gcc.target/i386/aggregate-ret4.c: New test.

From-SVN: r176828

gimple-fold.c (or_comparisons_1): Remove TRUTH_AND/OR expression handling.

2011-07-27  Kai Tietz  <ktietz@redhat.com>

        * gimple-fold.c (or_comparisons_1): Remove TRUTH_AND/OR
        expression handling.
        (and_var_with_comparison_1): Likewise.

From-SVN: r176827

params.h (ALLOW_STORE_DATA_RACES): New.

* params.h (ALLOW_STORE_DATA_RACES): New.
* params.def (PARAM_ALLOW_STORE_DATA_RACES): New.
* Makefile.in (expr.o): Depend on PARAMS_H.
* machmode.h (get_best_mode): Add argument.
* fold-const.c (optimize_bit_field_compare): Add argument to
get_best_mode.
(fold_truthop): Same.
* ifcvt.c (noce_emit_move_insn): Add argument to store_bit_field.
* expr.c (emit_group_store): Same.
(copy_blkmode_from_reg): Same.
(write_complex_part): Same.
(optimize_bitfield_assignment_op): Add argument.
Add argument to get_best_mode.
(get_bit_range): New.
(expand_assignment): Calculate maxbits and pass it down
accordingly.
(store_field): New argument.
(expand_expr_real_2): New argument to store_field.
Include params.h.
* expr.h (store_bit_field): New argument.
* stor-layout.c (get_best_mode): Restrict mode expansion by taking
into account maxbits.
* calls.c (store_unaligned_arguments_into_pseudos): New argument
to store_bit_field.
* expmed.c (store_bit_field_1): New argument.  Use it.
(store_bit_field): Same.
(store_fixed_bit_field): Same.
(store_split_bit_field): Same.
(extract_bit_field_1): Pass new argument to get_best_mode.
(extract_bit_field): Same.
* stmt.c (store_bit_field): Pass new argument to store_bit_field.
* doc/invoke.texi: Document parameter allow-store-data-races.

From-SVN: r176824

re PR fortran/49755 (ALLOCATE with STAT= produces invalid code for already allocated vars)

2011-07-26  Daniel Carrera  <dcarrera@gmail.com>

PR fortran/49755
* trans.c (gfc_allocate_using_malloc): Change function signature.
Return nothing. New parameter "pointer". Eliminate temorary variables.
(gfc_allocate_using_lib): Ditto.
(gfc_allocate_allocatable): Ditto. Update call to gfc_allocate_using_lib
and gfc_allocate_using_malloc. Do not free and then reallocate a
variable that is already allocated.
(gfc_likely): New function. Basedon gfc_unlikely.
* trans-array.c (gfc_array_init_size): New parameter "descriptor_block".
Instructions to modify the array descriptor are stored in this block
while other instructions continue to be stored in "pblock".
(gfc_array_allocate): Update call to gfc_array_init_size. Move the
descriptor_block so that the array descriptor is only updated if
the array was allocated successfully.
Update calls to gfc_allocate_allocatable and gfc_allocate_using_malloc.
* trans.h (gfc_allocate_allocatable): Change function signature.
Function now returns void.
(gfc_allocate_using_lib): Ditto, and new function parameter.
(gfc_allocate_using_malloc): Ditto.
* trans-openmp.c (gfc_omp_clause_default_ctor,
gfc_omp_clause_copy_ctor,gfc_trans_omp_array_reduction): Replace a call
to gfc_allocate_allocatable with gfc_allocate_using_malloc.
* trans-stmt.c (gfc_trans_allocate): Update function calls for
gfc_allocate_allocatable and gfc_allocate_using_malloc.

2011-07-26  Daniel Carrera  <dcarrera@gmail.com>

PR fortran/49755
* gfortran.dg/multiple_allocation_1.f90: Fix test. Allocating an
allocated array should *not* change its size.
* gfortran.dg/multiple_allocation_3.f90: New test.

From-SVN: r176822

dwarf2out.c (output_macinfo_op): Ensure fd->filename points to GC allocated copy of the string.

* dwarf2out.c (output_macinfo_op): Ensure fd->filename points
to GC allocated copy of the string.
(dwarf2out_finish): Emit .debug_macinfo or .debug_macro sections
before .debug_line, not after it.

From-SVN: r176819

tree-vrp.c (extract_range_from_binary_expr): Remove TRUTH-binary cases and add new bitwise-cases.

2011-07-27  Kai Tietz  <ktietz@redhat.com>

        * tree-vrp.c (extract_range_from_binary_expr): Remove
        TRUTH-binary cases and add new bitwise-cases.
        (extract_range_from_assignment): Likewise.
        (register_edge_assert_for_1): Likeiwise.
        (register_edge_assert_for): Likewise.
        (simplify_truth_ops_using_ranges): Likewise.
        (simplify_stmt_using_ranges): Likewise.

From-SVN: r176818

Daily bump.

From-SVN: r176816

Call simplify_gen_subreg for PIC with mode of x.

2011-07-26  H.J. Lu  <hongjiu.lu@intel.com>

PR target/47372
* config/i386/i386.c (ix86_delegitimize_address): Call
simplify_gen_subreg for PIC with mode of x only if modes of
x and orig_x are different.

From-SVN: r176812

dwarf2out.c (output_macinfo_op): Ensure fd->filename points to GC allocated copy of the string.

* dwarf2out.c (output_macinfo_op): Ensure fd->filename points
to GC allocated copy of the string.
(dwarf2out_finish): Emit .debug_macinfo or .debug_macro sections
before .debug_line, not after it.

From-SVN: r176811

re PR c++/49776 ([C++0x]ICE in build_data_member_initialization, at cp/semantics.c:5499)

/cp
2011-07-26  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/49776
* typeck.c (cp_build_modify_expr): Check digest_init return value
for error_mark_node.

/testsuite
2011-07-26  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/49776
* g++.dg/cpp0x/constexpr-49776.C: New.

From-SVN: r176809

Fix PR47046: correct evolution_function_is_affine_p

"Bug 47046 - gcc.target/i386/sse4_1-movntdqa.c ICEs with -fgraphite-identity"

The problem here is that we are left with the following code to be
translated in the new representation following the transform that
Graphite has chosen:

        D.2709_14 = j_33 * i_32;
        D.2710_15 = D.2709_14 * i_32;
        D.2711_16 = D.2710_15 * sign_34;
        *D.2708_13 = D.2711_16;

In this particular case we have a nonlinear expression "i * i" for
which we have to generate code following the new graphite_iv variables.

The patch fixes the function that detects whether we are passing non
linear stuff to graphite: evolution_function_is_affine_p.  It seems
like for the moment evolution_function_is_affine_p is testing whether
an evolution function is affine only in the innermost loop, without
looking recursively at what happens in outer loops.

The chrec for this case is: {0, +, {0, +, {1, +, 2}_1}_1}_2 and we are
testing whether the evolution is affine only for the loop_2, which is
true as we have {0, +, blah}_2 with blah invariant in loop_2.

The patch adds the recursive call to evolution_function_is_affine_p.

Bootstrapped and tested on amd64-linux.

2011-07-26  Sebastian Pop  <sebastian.pop@amd.com>

PR middle-end/47046
* tree-chrec.h (evolution_function_is_affine_p): Recursively call
evolution_function_is_affine_p on CHREC_RIGHT.

* gcc.dg/graphite/id-pr47046.c: New.

From-SVN: r176805

Avoid use of lang_hooks.types.type_for_size.

2011-07-26  Sebastian Pop  <sebastian.pop@amd.com>

* tree-data-ref.c (max_stmt_executions_tree): Do not call
lang_hooks.types.type_for_size.

From-SVN: r176804

Add testcase for PR47593

2011-07-26  Sebastian Pop  <sebastian.pop@amd.com>

PR middle-end/47653
* gcc.dg/graphite/run-id-pr47593.c: New.

From-SVN: r176803

Fix PR47653: do not handle loops using wrapping semantics in graphite

2011-07-26  Sebastian Pop  <sebastian.pop@amd.com>

PR middle-end/47653
* graphite-scop-detection.c (graphite_can_represent_loop): Discard
loops using wrapping semantics.

* gcc.dg/graphite/run-id-pr47653.c: New.
* gcc.dg/graphite/interchange-3.c: Do not use unsigned types for
induction variables.
* gcc.dg/graphite/scop-16.c: Same.
* gcc.dg/graphite/scop-17.c: Same.
* gcc.dg/graphite/scop-21.c: Same.

From-SVN: r176802

Fix PR48805: Do not instantiate ADDR_EXPRs

With this patch we avoid instantiating ADDR_EXPR: it makes no sense
to translate &b[i] into &b[{0, +, 1}_1].

Bootstrapped and tested on amd64-linux.

2011-07-26  Sebastian Pop  <sebastian.pop@amd.com>

PR middle-end/48805
* tree-scalar-evolution.c (instantiate_scev_r): Return
chrec_dont_know for ADDR_EXPR.

* gcc.dg/graphite/id-pr48805.c: New.

From-SVN: r176801

configure.ac: Set have_compiler based on whether gcc directory exists...

* configure.ac: Set have_compiler based on whether gcc directory
exists, rather than on whether gcc is in configdirs.
* configure: Rebuild.

From-SVN: r176800

re PR target/47369 ([x32] internal compiler error: in extract_insn, at recog.c:2109)

PR target/47369
PR target/49853
* config/i386/i386.c (ix86_expand_move): Call convert_to_mode
if legitimize_tls_address returned operand in wrong mode. Allow
SImode and DImode symbolic operand for PIC.  Call convert_to_mode
if legitimize_pic_address returned operand in wrong mode.

Co-Authored-By: H.J. Lu <hongjiu.lu@intel.com>
From-SVN: r176798

tree-sra.c (tree_non_mode_aligned_mem_p): Strip conversions and return false for invariants.

2011-07-26  Martin Jambor  <mjambor@suse.cz>

* tree-sra.c (tree_non_mode_aligned_mem_p): Strip conversions and
return false for invariants.

From-SVN: r176797

Remove an extra line.

2011-07-26  H.J. Lu  <hongjiu.lu@intel.com>

* testsuite/demangle-expected: Remove an extra line.

From-SVN: r176796

i386.md (add->lea splitter): Implement using SWI mode iterator.

* config/i386/i386.md (add->lea splitter): Implement using SWI
mode iterator.  Change operand 2 predicate to <nonmemory_operand>.
(add->lea zext splitter): Change operand 2 predicate to
x86_64_nonmemory_operand.

From-SVN: r176795

predict.c (maybe_hot_frequency_p): Make sure a zero entry-block frequency makes everything hot.

2011-07-26  Richard Guenther  <rguenther@suse.de>

* predict.c (maybe_hot_frequency_p): Make sure a zero entry-block
frequency makes everything hot.

From-SVN: r176794

cp-demangle.c (d_print_init): Initialize pack_index field.

* cp-demangle.c (d_print_init): Initialize pack_index field.
(d_print_comp): Check for NULL template argument.
* testsuite/demangle-expected: Add test case.

From-SVN: r176791

re PR middle-end/49840 (New test failures)

2011-07-26  Richard Guenther  <rguenther@suse.de>

PR tree-optimization/49840
* tree-vrp.c (range_fits_type_p): Properly handle full
double-int precision.

From-SVN: r176790

re PR bootstrap/49786 (bootstrap failed with bootstrap-profiled)

2011-07-26  Martin Jambor  <mjambor@suse.cz>

PR bootstrap/49786
* ipa-cp.c (update_profiling_info): Avoid overflow when updating
counts.
(update_specialized_profile): Likewise.

From-SVN: r176789

re PR target/47381 ([x32] internal compiler error: in gen_lowpart_general, at rtlhooks.c:59)

PR target/47381
PR target/49832
PR target/49833
* config/i386/i386.md (i): Change SImode attribute to "e".
(g): Change SImode attribute to "rme".
(di): Change SImode attribute to "nF".
(general_operand): Change SImode attribute to x86_64_general_operand.
(general_szext_operand): Change SImode attribute to
x86_64_szext_general_operand.
(immediate_operand): Change SImode attribute to
x86_64_immediate_operand.
(nonmemory_operand): Change SImode attribute to
x86_64_nonmemory_operand.
(*movdi_internal_rex64): Remove mode from pic_32bit_operand check.
(*movsi_internal): Ditto.  Use "e" constraint in alternative 2.
(*lea_1): Use SWI48 mode iterator.
(*lea_1_zext): New insn pattern.
(testsi_ccno_1): Use x86_64_nonmemory_operand predicate for operand 2.
(*bt<mode>): Ditto.
(*add<mode>1): Use x86_64_general_operand predicate for operand 2.
Update operand constraints.
(addsi_1_zext): Ditto.
(*add<mode>2): Ditto.
(*addsi_3_zext): Ditto.
(*subsi_1_zext): Ditto.
(*subsi_2_zext): Ditto.
(*subsi_3_zext): Ditto.
(*addsi3_carry_zext): Ditto.
(*<plusminus_insn>si3_zext_cc_overflow): Ditto.
(*mulsi3_1_zext): Ditto.
(*andsi_1): Ditto.
(*andsi_1_zext): Ditto.
(*andsi_2_zext): Ditto.
(*<any_or:code>si_1_zext): Ditto.
(*<any_or:code>si_2_zext): Ditto.
(*test<mode>_1): Use <general_operand> predicate for operand 1.
(*and<mode>_2): Ditto.
(mov<mode>cc): Use  <general_operand> predicate for operands 1 and 2.
(add->lea splitter): Check operand modes in insn constraint.  Extend
operands less than SImode wide to SImode.
(add->lea zext splitter): Do not extend input operands to DImode.
(*lea_general_1): Handle only QImode and HImode operands.
(*lea_general_2): Ditto.
(*lea_general_3): Ditto.
(*lea_general_1_zext): Remove.
(*lea_general_2_zext): Ditto.
(*lea_general_3_zext): Ditto.
(*lea_general_4): Check operand modes in insn constraint.  Extend
operands less than SImode wide to SImode.
(ashift->lea splitter): Ditto.
* config/i386/i386.c (ix86_print_operand_address): Print address
registers with 'q' modifier on 64bit targets.
* config/i386/predicates.md (pic_32bit_opreand): Define as special
predicate.  Reject non-SI and non-DI modes.

From-SVN: r176788

trans-array.c (CAF_TOKEN_FIELD): New macro constant.

2011-07-26  Tobias Burnus  <burnus@net-b.de>

        * trans-array.c (CAF_TOKEN_FIELD): New macro constant.
        (gfc_conv_descriptor_token): New function.
        * trans-array.h (gfc_conv_descriptor_token): New prototype.
        * trans-types.c (gfc_get_array_descriptor_base): For coarrays
        with -fcoarray=lib, append "void *token" to the array descriptor.
        (gfc_array_descriptor_base_caf): New static variable.
        * trans-expr.c (gfc_conv_procedure_call): Handle token and
        * offset
        when passing a descriptor coarray to a nondescriptor dummy.

2011-07-26  Tobias Burnus  <burnus@net-b.de>

        * gfortran.dg/coarray_lib_token_2.f90: New.

From-SVN: r176784

vect-70.c: Reduce the data size to fit SPU local store.

* gcc.dg/vect/vect-70.c: Reduce the data size to fit
SPU local store.

From-SVN: r176783

re PR middle-end/49671 (volatile goes missing after inlining)

2011-07-25  Andrew Pinski  <apinski@cavium.com>

PR tree-opt/49671
* tree-inline.c (remap_gimple_op_r): Copy TREE_THIS_VOLATILE and
TREE_THIS_NOTRAP into the inner most MEM_REF.
Always copy TREE_THIS_VOLATILE.
* tree-sra.c (ptr_parm_has_direct_uses): Check that the lhs, rhs and
arguments are not volatile references.

2011-07-25  Andrew Pinski  <apinski@cavium.com>

PR tree-opt/49671
* gcc.dg/tree-ssa/pr49671-1.c: New testcase.
* gcc.dg/tree-ssa/pr49671-2.c: New testcase.

From-SVN: r176782

array (at): Remove constexpr when -fno-exceptions.

2011-07-25  Benjamin Kosnik  <bkoz@redhat.com>

* include/std/array (at): Remove constexpr when -fno-exceptions.
* testsuite/23_containers/array/at_neg.cc: Test.

From-SVN: r176780

insn-notes.def (NOTE_INSN_CFA_RESTORE_STATE): Remove.

        * insn-notes.def (NOTE_INSN_CFA_RESTORE_STATE): Remove.
        * final.c (final_scan_insn): Don't test for it.
        (rest_of_clean_state): Likewise.

From-SVN: r176779

Daily bump.

From-SVN: r176778

re PR debug/49841 (AIX bootstrap failure in dwarf2cfi.c:maybe_record_trace_start)

PR debug/49841
* config/rs6000/rs6000.c (rs6000_frame_related): Return the insn.
(emit_frame_save): Likewise.
(rs6000_emit_prologue): Move the FRAME_RELATED_EXPR from the save
insn onto a dummy blockage insn after the join label.

From-SVN: r176773

dwarf2cfi: Generate remember/restore_state.

From-SVN: r176772

dwarf2cfi: Add debug_cfi_row.

From-SVN: r176771

genattr.c (write_upcase, [...]): Move to ...

        * genattr.c (write_upcase, gen_attr <enum definition writing>):
        Move to ...
        * genattr-common.c ... here.
        (main): Call gen_attr.
        * optc-gen.awk: Make generated program include insn-attr-common.h .
        * Makefile.in (oprions.o): Depend on insn-attr-common.h

From-SVN: r176769

re PR bootstrap/49845 (bootstrap failure)

2011-07-25  Paolo Carlini  <paolo.carlini@oracle.com>

PR bootstrap/49845
* parser.c (cp_parser_perform_range_for_lookup): Always assign *being
and *end before returning.

From-SVN: r176765

m32c.h (PRINT_OPERAND, [...]): Remove macro.

* config/m32c/m32c.h (PRINT_OPERAND, PRINT_OPERAND_ADDRESS,
PRINT_OPERAND_PUNCT_VALID_P): Remove macro.
* config/m32c/m32c-protos.h (m32c_print_operand_punct_valid_p,
m32c_print_operand, m32c_print_operand_address): Remove.
* config/m32c/m32c.c (m32c_print_operand_punct_valid_p): Make static.
Change return type to bool. Change argument type to bool.
(m32c_print_operand, m32c_print_operand_address): Make static.
(TARGET_PRINT_OPERAND, TARGET_PRINT_OPERAND_ADDRESS,
TARGET_PRINT_OPERAND_PUNCT_VALID_P): Define.

From-SVN: r176763

Remove parm name from declaration

gcc/c-family

* c-common.h (set_underlying_type): Remove parm name from
declaration.

From-SVN: r176762

re PR libstdc++/49836 ([C++0x] vector<T>::push_back() should not require T to be (move-)assignable)

2011-07-25  Paolo Carlini  <paolo.carlini@oracle.com>
    Nathan Ridge  <zeratul976@hotmail.com>

PR libstdc++/49836
* include/bits/stl_vector.h (vector<>::_M_emplace_back_aux):
Declare.
(vector<>::push_back(const value_type&)): Use it.
* include/bits/vector.tcc: Define.
(vector<>::emplace_back(_Args&&...)): Use it.
* testsuite/util/testsuite_tr1.h (CopyConsOnlyType, MoveConsOnlyType):
Add.
* testsuite/23_containers/vector/modifiers/push_back/49836.cc: New.
* testsuite/23_containers/deque/modifiers/push_back/49836.cc:
Likewise.
* testsuite/23_containers/deque/modifiers/push_front/49836.cc:
Likewise.
* testsuite/23_containers/vector/requirements/dr438/assign_neg.cc:
Adjust dg-error line number.
* testsuite/23_containers/vector/requirements/dr438/insert_neg.cc:
Likewise.
* testsuite/23_containers/vector/requirements/dr438/
constructor_1_neg.cc: Likewise.
* testsuite/23_containers/vector/requirements/dr438/
constructor_2_neg.cc: Likewise.

Co-Authored-By: Nathan Ridge <zeratul976@hotmail.com>
From-SVN: r176761

sourcebuild.texi (Effective-Target Keywords, [...]): Document mmap.

gcc:
* doc/sourcebuild.texi (Effective-Target Keywords, Environment
attributes): Document mmap.

gcc/testsuite:
* lib/target-supports.exp (check_effective_target_mmap): New proc.

* gcc.c-torture/execute/loop-2f.c: Remove #ifdef __unix__.
* gcc.c-torture/execute/loop-2g.c: Likewise.
* gcc.c-torture/execute/loop-2f.x: Load target-supports.exp.
Require mmap support.
* gcc.c-torture/execute/loop-2g.x: Likewise.
* gcc.dg/20030711-1.c: Replace dg-do target list by mmap.
(MAP_ANON): Provide default.
* gcc.dg/20050826-1.c: Likewise.
* gcc.target/i386/pr36533.c: Likewise.
* gcc.dg/vect/pr49038.c: Remove dg-do run.
Use dg-require-effective-target mmap.

From-SVN: r176760

mmix.h (PRINT_OPERAND, [...]): Remove macro.

* config/mmix/mmix.h (PRINT_OPERAND, PRINT_OPERAND_ADDRESS,
PRINT_OPERAND_PUNCT_VALID_P): Remove macro.
* config/mmix/mmix-protos.h (mmix_print_operand_punct_valid_p,
mmix_print_operand, mmix_print_operand_address): Remove.
* config/mmix/mmix.c (mmix_print_operand_punct_valid_p): Make static.
Change return type to bool. Change argument type to bool.
(mmix_print_operand, mmix_print_operand_address): Make static.
(mmix_intval, mmix_output_condition): Change 'x' argument type
to const_rtx.
(TARGET_PRINT_OPERAND, TARGET_PRINT_OPERAND_ADDRESS,
TARGET_PRINT_OPERAND_PUNCT_VALID_P): Define.

From-SVN: r176759

re PR c++/49347 (G++ rejects Solaris spawn.h use of __restrict)

PR c++/49347
* inclhack.def (solaris_posix_spawn_restrict): New fix.
* fixincl.x: Regenerate.
* tests/base/spawn.h: New test.

From-SVN: r176758

re PR target/39386 ([avr] different computation results for O1 and O0 executables)

PR target/39386
* config/avr/avr.c (out_shift_with_cnt): Use tmp_reg as
shift counter for x << x and x >> x shifts.

From-SVN: r176756