genrecog.c (validate_pattern): Add VEC_SELECT validation.

* genrecog.c (validate_pattern): Add VEC_SELECT validation.
* genmodes.c (emit_min_insn_modes_c): Call emit_mode_nunits
and emit_mode_inner.

From-SVN: r246460

S/390: arch12: New builtins.

This patch implements a set of low-level builtins for instruction
which would otherwise not be emitted by the compiler plus a set of
high-level builtins as defined by the IBM XL compiler.  The high-level
builtins will be described in a future revision of the z/OS XL C/C++
Programming Guide.

I'll try to come up with a documentation appropriate for the GCC
manual as well (sometimes in the future).

gcc/ChangeLog:

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/s390-builtins.def: Add VXE builtins.  Add a flags
argument to the overloaded builtin variants.  Use the new flag to
deprecate certain builtin variants.
* config/s390/s390-builtin-types.def: Add new builtin types.
* config/s390/s390-builtins.h: Support new flags field for
overloaded builtins.
* config/s390/s390-c.c (OB_DEF_VAR): New flags field.
(s390_macro_to_expand): Enable vector float data type.
(s390_cpu_cpp_builtins_internal): Indicate support of the new
builtins by incrementing the __VEC__ version number.
(s390_expand_overloaded_builtin): Support expansion of vec_xl and
vec_xst.
(s390_resolve_overloaded_builtin): Emit error messages depending
on the builtin flags.
* config/s390/s390.c (s390_expand_builtin): Support additional
flags argument.  Change error message to match the messages
emitted in s390-c.c.
* config/s390/s390.md: New UNSPEC_* constants.
(op_type): Add new instruction types.
* config/s390/vecintrin.h: Add new builtins and test data class
constants.
* config/s390/vx-builtins.md (V_HW_32_64): Add V4SF.
(V_HW_4, VEC_HW, VECF_HW): New mode iterators.
(VEC_INEXACT, VEC_NOINEXACT): New constants.
("vec_splats<mode>", "vec_insert<mode>", "vec_promote<mode>")
("vec_insert_and_zero<mode>", "vec_mergeh<mode>")
("vec_mergel<mode>"): V_HW -> VEC_HW.

("vlrlrv16qi", "vstrlrv16qi", "vbpermv16qi", "vec_msumv2di")
("vmslg", "*vftci<mode>_cconly", "vftci<mode>_intcconly")
("*vftci<mode>", "vftci<mode>_intcc", "vec_double_s64")
("vec_double_u64", "vfmin<mode>", "vfmax<mode>"): New definition.

("and_av2df3", "and_cv2df3", "vec_andc_av2df3")
("vec_andc_cv2df3", "xor_av2df3", "xor_cv2df3", "vec_nor_av2df3")
("vec_nor_cv2df3", "ior_av2df3", "ior_cv2df3", "vec_nabs")
("*vftcidb", "*vftcidb_cconly", "vftcidb"): Remove definition.

("vec_all_<fpcmpcc:code>v2df", "vec_any_<fpcmpcc:code>v2df")
("vec_scatter_elementv4si_DI", "vec_cmp<fpcmp:code>v2df")
("vec_di_to_df_s64", "vec_di_to_df_u64", "vec_df_to_di_u64")
("vfidb", "*vldeb", "*vledb", "*vec_cmp<insn_cmp>v2df_cconly")
("vec_cmpeqv2df_cc", "vec_cmpeqv2df_cc", "vec_cmphv2df_cc")
("vec_cmphev2df_cc", "*vec_cmpeqv2df_cc")
("*vec_cmphv2df_cc", "*vec_cmphev2df_cc"): Enable new modes as ...

("vec_all_<fpcmpcc:code><mode>", "vec_any_<fpcmpcc:code><mode>")
("vec_scatter_element<V_HW_4:mode>_DI")
("vec_cmp<fpcmp:code><mode>", "vcdgb", "vcdlgb", "vclgdb")
("vec_fpint<mode>", "vflls")
("vflrd", "*vec_cmp<insn_cmp><mode>_cconly", "vec_cmpeq<mode>_cc")
("vec_cmpeq<mode>_cc", "vec_cmph<mode>_cc", "vec_cmphe<mode>_cc")
("*vec_cmpeq<mode>_cc", "*vec_cmph<mode>_cc")
("*vec_cmphe<mode>_cc"): ... these.

("vec_ctd_s64", "vec_ctsl", "vec_ctul", "vec_st2f"): Use rounding
mode constant instead of magic value.

gcc/testsuite/ChangeLog:

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* gcc.target/s390/target-attribute/tattr-3.c: Adjust error message
and remove the high-level builtin.  The error message for the
would prevent compilation from reaching the second.
* gcc.target/s390/target-attribute/tattr-4.c: Likewise.

From-SVN: r246459

S/390: arch12: Support new vector floating point modes.

This patch adds support for the new floating point vector elements (SF
and TF) introduced with arch12.

gcc/ChangeLog:

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/s390.c (s390_expand_vec_compare): Support other
vector floating point modes than just V2DF.
(s390_expand_vcond): Likewise.
(s390_hard_regno_mode_ok): Allow SFmode values in VRs.
(s390_cannot_change_mode_class): Prevent mode changes between TF
and V1TF in vector registers.
* config/s390/s390.md (DF, SF): New mode attributes.
("*cmp<mode>_ccs", "add<mode>3", "sub<mode>3", "mul<mode>3")
("fma<mode>4", "fms<mode>4", "div<mode>3", "*neg<mode>2"): Add
SFmode support for VRs.
* config/s390/vector.md (V_HW, V_HW2, VT_HW, ti*, nonvec): Add new
vector fp modes.
(VFT, VF_HW): New mode iterators.
(vw, sdx): New mode attributes.
("addv2df3", "subv2df3", "mulv2df3", "divv2df3", "sqrtv2df2")
("fmav2df4","fmsv2df4", "negv2df2", "absv2df2", "*negabsv2df2")
("smaxv2df3", "sminv2df3", "*vec_cmp<VFCMP_HW_OP:code>v2df_nocc")
("vec_cmpuneqv2df", "vec_cmpltgtv2df", "vec_orderedv2df")
("vec_unorderedv2df"): Adjust the v2df only patterns to support
also the new vector floating point modes.  Renaming to ...

("add<mode>3", "sub<mode>3", "mul<mode>3", "div<mode>3")
("sqrt<mode>2", "fma<mode>4", "fms<mode>4", "neg<mode>2")
("abs<mode>2", "negabs<mode>2", "smax<mode>3")
("smin<mode>3", "*vec_cmp<VFCMP_HW_OP:code><mode>_nocc")
("vec_cmpuneq<mode>", "vec_cmpltgt<mode>", "vec_ordered<mode>")
("vec_unordered<mode>"): ... these.

("neg_fma<mode>4", "neg_fms<mode>4", "*smax<mode>3_vxe")
("*smin<mode>3_vxe", "*sminv2df3_vx", "*vec_extendv4sf")
("*vec_extendv2df"): New insn definitions.

gcc/testsuite/ChangeLog:

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* gcc.target/s390/vxe/negfma-1.c: New test.

From-SVN: r246458

S/390: arch12: Support the mul/add/subtract
 instructions.

gcc/ChangeLog:

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/s390.md ("*adddi3_sign", "*subdi3_sign", "mulditi3")
("mulditi3_2", "*muldi3_sign"): New patterns.
("muldi3", "*muldi3", "mulsi3", "*mulsi3"): Add an expander and
rename the pattern definition.

gcc/testsuite/ChangeLog:

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* gcc.target/s390/arch12/aghsghmgh-1.c: New test.
* gcc.target/s390/arch12/mul-1.c: New test.
* gcc.target/s390/arch12/mul-2.c: New test.

From-SVN: r246457

S/390: arch12: Add indirect branch pattern

This adds support for the branch indirect instruction.

gcc/ChangeLog:

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/s390.md ("indirect_jump"): Turn insn definition into
expander.
("*indirect_jump", "*indirect2_jump"): New pattern definitions.

From-SVN: r246456

S/390: arch12: Add vllezlf instruction.

This adds support for the vector load element and zero instruction and
makes sure it is used when initializing vectors with elements while
setting the rest to 0.

gcc/ChangeLog:

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/s390.c (s390_expand_vec_init): Use vllezl
instruction if possible.
* config/s390/vector.md (vec_halfnumelts): New mode
attribute.
("*vec_vllezlf<mode>"): New pattern.

gcc/testsuite/ChangeLog:

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* gcc.target/s390/vxe/vllezlf-1.c: New test.

From-SVN: r246455

S/390: arch12: New vector popcount variants

arch12 provides pop count vector instructions for bigger elements than
just chars.

gcc/testsuite/ChangeLog:

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* gcc.target/s390/vxe/popcount-1.c: New test.

gcc/ChangeLog:

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/vector.md ("popcountv16qi2", "popcountv8hi2")
("popcountv4si2", "popcountv2di2"): Rename to ...
("popcount<mode>2", "popcountv8hi2_vx", "popcountv4si2_vx")
("popcountv2di2_vx"): ... these and add !TARGET_VXE to the
condition.
("popcount<mode>2_vxe"): New pattern.

From-SVN: r246454

S/390: arch12: Add support for new vector bit
 operations.

This patch adds support for the new bit operations introduced with
arch12.

The patch also renames the one complement pattern to the proper RTL
standard name.

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/s390.c (s390_rtx_costs): Return low costs for the
canonical form of ~AND to make sure the new instruction will be
used.
* config/s390/vector.md ("notand<mode>3", "ior_not<mode>3")
("notxor<mode>3"): Add new pattern definitions.
("*not<mode>"): Rename to ...
("one_cmpl<mode>2"): ... this.

gcc/testsuite/ChangeLog:

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* gcc.target/s390/vxe/bitops-1.c: New test.

From-SVN: r246453

S/390: arch12: Add arch12 option.

This patch covers the mechanical work of making the new architecture
option arch12 available wherever it will be needed later.

gcc/testsuite/ChangeLog:

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* gcc.target/s390/s390.exp: Run tests in arch12 and vxe dirs.
* lib/target-supports.exp: Add effective target check s390_vxe.

gcc/ChangeLog:

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* common/config/s390/s390-common.c (processor_flags_table): Add
arch12.
* config.gcc: Add arch12.
* config/s390/driver-native.c (s390_host_detect_local_cpu):
Default to arch12 for unknown CPU model numbers.
* config/s390/s390-builtins.def: Add B_VXE builtin flag.
* config/s390/s390-c.c (s390_cpu_cpp_builtins_internal): Adjust
PROCESSOR_max sanity check.
* config/s390/s390-opts.h (enum processor_type): Add
PROCESSOR_ARCH12.
* config/s390/s390.c (processor_table): Add arch12.
(s390_expand_builtin): Add check for B_VXE flag.
(s390_issue_rate): Add PROCESSOR_ARCH12.
(s390_get_sched_attrmask): Likewise.
(s390_get_unit_mask): Likewise.
(s390_sched_score): Enable z13 scheduling for arch12.
(s390_sched_reorder): Likewise.
(s390_sched_variable_issue): Likewise.
* config/s390/s390.h (enum processor_flags): Add PF_ARCH12 and
PF_VXE.
(s390_tune_attr): Use z13 scheduling also for arch12.
(TARGET_CPU_ARCH12, TARGET_CPU_ARCH12_P, TARGET_CPU_VXE)
(TARGET_CPU_VXE_P, TARGET_ARCH12, TARGET_ARCH12_P, TARGET_VXE)
(TARGET_VXE_P): New macros.
* config/s390/s390.md: Add arch12 to cpu attribute.  Add arch12
and vxe to cpu_facility.  Add arch12 and vxe to enabled attribute.
* config/s390/s390.opt: Add arch12 as processor_type.

From-SVN: r246452

S/390: Rearrange fixuns_trunc pattern definitions.

This reworks the fixuns_trunc* patterns a bit which got quite confusing
after adding z13 support.  Now we just have a single RTL standard name
expander definition ("fixuns_trunc<FP:mode><GPR:mode>2") which then
multiplexes to either the emulation variants *_emu or the hardware
implementations.

gcc/ChangeLog:

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/s390.md
("fixuns_truncdddi2", "fixuns_trunctddi2")
("fixuns_trunc<BFP:mode><GPR:mode>2"): Merge into ...
("fixuns_trunc<FP:mode><GPR:mode>2"): New expander.

("fixuns_trunc<BFP:mode><GPR:mode>2", "fixuns_trunc<mode>si2"):
Rename expanders to ...

("fixuns_trunc<BFP:mode><GPR:mode>2_emu")
("fixuns_truncdddi2_emu"): ... these.

("fixuns_trunc<mode>si2_emu"): New expander.

("*fixuns_truncdfdi2_z13"): Rename to ...
("*fixuns_truncdfdi2_vx"): ... this.

From-SVN: r246451

S/390: Use wfc for scalar vector compares

The z13 vector support used the vector style comparison instructions
also for the scalar compares in vector registers.  However, it is much
more convenient to just use the compare scalar instruction for that
purpose.  The advantage is that this instruction generates a CC result
as our compares usually do.  So this results in quite some code to be
removed from the backend.

Regression tested on s390x.

gcc/ChangeLog:

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/2964.md: Remove the single element vector compare
instructions which are no longer used.
* config/s390/s390.c (s390_select_ccmode): Remove handling of
vector CCmodes.
(s390_canonicalize_comparison): Remove handling of DFmode
compares.
(s390_expand_vec_compare_scalar): Remove function.
(s390_emit_compare): Don't call s390_expand_vec_compare_scalar.
* config/s390/s390.md ("*vec_cmp<insn_cmp>df_cconly"): Remove
pattern.
("*cmp<mode>_ccs"): Add wfcdb instruction.

gcc/testsuite/ChangeLog:

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* gcc.target/s390/vector/vec-scalar-cmp-1.c: Adjust for the
comparison instructions used from now on.

From-SVN: r246450

S/390: Move and rename vector check.

Move the target support routine for the vector facility to the common
code file.  This is required to enable the generic vectorization tests
on S/390.  While doing this the too generic name for the check (vector)
is changed to s390_vx.  The renaming required to modify all the
testcases currently using that check.

gcc/testsuite/ChangeLog:

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* gcc.target/s390/s390.exp (check_effective_target_vector):
Include target-supports.exp and move target_vector check routine
...
* lib/target-supports.exp (check_effective_target_s390_vx): ... to
here and rename it.
* gcc.target/s390/htm-builtins-z13-1.c: Rename effective target
check from vector to s390_vx.
* gcc.target/s390/target-attribute/tpragma-struct-vx-1.c: Likewise.
* gcc.target/s390/target-attribute/tpragma-struct-vx-2.c: Likewise.
* gcc.target/s390/vector/stpcpy-1.c: Likewise.
* gcc.target/s390/vector/vec-abi-vararg-1.c: Likewise.
* gcc.target/s390/vector/vec-clobber-1.c: Likewise.
* gcc.target/s390/vector/vec-genbytemask-1.c: Likewise.
* gcc.target/s390/vector/vec-genmask-1.c: Likewise.
* gcc.target/s390/vector/vec-nopeel-1.c: Likewise.
* gcc.target/s390/vector/vec-vrepi-1.c: Likewise.

From-SVN: r246449

S/390: movdf improvements

This patch add the vector load element from immediate instruction to the
movdf/dd pattern for loading a FP zero and it removes the vector
instructions from the mov<mode>_64 pattern. These were pointless in
there because z13 support implies DFP support so these instructions will
always be matched in the mov<mode>_64dfp pattern instead.

Regression tested on s390x

gcc/ChangeLog:

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/s390.md ("mov<mode>_64dfp" DD_DF): Use vleig for loading a
FP zero.
("*mov<mode>_64" DD_DF): Remove the vector instructions. These
will anyway by matched by mov<mode>_64dfp.

From-SVN: r246448

S/390: movsf/sd pattern fixes.

The SD/SFmode move pattern used a wrong mnemonic for vector load
element.
On the vector load element instruction was an operand missing.

Regression tested on s390x.

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/s390.md ("mov<mode>" SD_SF): Change vleg/vsteg to
vlef/vstef.  Add missing operand to vleif.

From-SVN: r246447

S/390: vec_init improvements

This enables the vec_init pattern also for V4SF, V1TI, and V1TF.

gcc/testsuite/ChangeLog:

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* gcc.target/s390/vector/vec-init-2.c: New test.

gcc/ChangeLog:

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/s390.c (s390_expand_vec_init): Enable vector load
pair for all vector types with 64 bit elements.
* config/s390/vx-builtins.md (V_HW_64): Move mode iterator to ...
* config/s390/vector.md (V_HW_64): ... here.
(V_128_NOSINGLE): New mode iterator.
("vec_init<V_HW:mode>"): Use V_128 as mode iterator.
("*vec_splat<mode>"): Use V_128_NOSINGLE mode iterator.
("*vec_tf_to_v1tf", "*vec_ti_to_v1ti"): New pattern definitions.
("*vec_load_pairv2di"): Change to ...
("*vec_load_pair<mode>"): ... this one.

From-SVN: r246446

S/390: Improve support of 128 bit vectors in GPRs

This patch improves the handling of 128 bit vectors residing in GPRs
by adding more alternatives to the move pattern.

Regression tested on s390x.

gcc/ChangeLog:

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/constraints.md: Add comments.
(jKK): Reject element sizes > 8 bytes.
* config/s390/s390.c (s390_split_ok_p): Enable splitting also for
s_operands.
* config/s390/s390.md: Add the s_operand checks formerly in
s390_split_ok_p to various splitters where they are still
required.
* config/s390/vector.md ("mov<mode>" V_128): Add GPR alternatives
for 128 bit vectors.  Plus two splitters.

From-SVN: r246445

S/390: Rename cpu facility vec to vx.

gcc/ChangeLog:

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/s390.md: Rename the cpu facilty vec to vx throughout
the file.

From-SVN: r246444

S/390: PR79904: Disallow reg + sym_ref literal pool addresses.

We accept reg + sym_ref as valid address if sym_ref is a literal pool
reference knowing that it will be rewritten as r13 + reg + offset.
However, annotate_constant_pool_refs was never able to handle that.

With the patch only single sym_refs are accepted.

Regression tested on s390x.

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

PR target/79904
* config/s390/s390.c (s390_decompose_address): Reject reg +
sym_ref literal pool references.

gcc/testsuite/ChangeLog:

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* gcc.dg/ubsan/pr79904-2.c: New test.

From-SVN: r246443

S/390: PR79893: Add diagnostics vec_load_bndry builtin.

The boundary argument of the vec_load_bndry builtin needs to be
rewritten.  At that point it must be constant already.  The current
diagnostics in s390_expand_builtins is too late for this.  The patch
adds an additional check for that builtin which will be triggered
already during preprocessing.

Regression tested on s390x.

gcc/testsuite/ChangeLog:

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

PR target/79893
* gcc.target/s390/zvector/pr79893.c: New test.

gcc/ChangeLog:

2017-03-24  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

PR target/79893
* config/s390/s390-c.c (s390_adjust_builtin_arglist): Issue an
error if the boundary argument is not constant.

From-SVN: r246442

re PR rtl-optimization/80112 (ICE in doloop_condition_get at loop-doloop.c:158)

PR rtl-optimization/80112
* loop-doloop.c (doloop_condition_get): Don't check condition
if cmp isn't SET with IF_THEN_ELSE src.

* gcc.dg/pr80112.c: New test.

From-SVN: r246441

Adjust c-c++-common/Wimplicit-fallthrough-7.c etc. line numbers

* c-c++-common/Wimplicit-fallthrough-7.c: Adjust dg-warning
etc. line numbers.
* gcc.dg/Walloca-1.c: Likewise.
* gcc.dg/Walloca-2.c: Likewise.
* gcc.dg/Wvla-larger-than-2.c: Likewise.

From-SVN: r246440

re PR tree-optimization/80158 (ICE in all_phi_incrs_profitable)

2017-03-24  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

PR tree-optimization/80158
* gimple-ssa-strength-reduction.c (replace_mult_candidate): When
replacing a candidate statement, also replace it for the
candidate's alternate interpretation.
(replace_rhs_if_not_dup): Likewise.
(replace_one_candidate): Likewise.

* gfortran.fortran-torture/compile/pr80158.f: New file.

From-SVN: r246439

re PR tree-optimization/80167 (ICE in translate_isl_ast_to_gimple::is_valid_rename at gcc/graphite-isl-ast-to-gimple.c:1139)

2017-03-24  Richard Biener  <rguenther@suse.de>

PR tree-optimization/80167
* graphite-isl-ast-to-gimple.c
(translate_isl_ast_to_gimple::is_valid_rename): Handle default-defs
properly.
(translate_isl_ast_to_gimple::get_rename): Likewise.

* gcc.dg/graphite/pr80167.c: New testcase.

From-SVN: r246438

Require effective target global_constructor for two testcases

2017-03-24  Tom de Vries  <tom@codesourcery.com>

PR testsuite/80092
* gcc.dg/no_profile_instrument_function-attr-1.c: Add
dg-require-effective-target global_constructor.
* gcc.dg/tls/emutls-2.c: Same.

From-SVN: r246436

Require effective target indirect_jump in Wimplicit-fallthrough-34.c

2017-03-24  Tom de Vries  <tom@codesourcery.com>

PR testsuite/80092
* c-c++-common/Wimplicit-fallthrough-34.c: Add
dg-require-effective-target indirect_jumps.

From-SVN: r246435

Require effective target nonlocal_goto in pr79244.c

2017-03-24  Tom de Vries  <tom@codesourcery.com>

PR testsuite/80092
* gcc.dg/torture/pr79244.c: Add dg-require-effective-target
nonlocal_goto.

From-SVN: r246434

Add missing dg-require-effective-target alloca in gcc testsuite

2017-03-24  Tom de Vries  <tom@codesourcery.com>

PR testsuite/80092
* gcc.dg/Walloca-7.c: Add dg-require-effective-target alloca.
* gcc.dg/Walloca-12.c: Same.
* gcc.dg/attr-alloc_size-8.c: Same.
* gcc.dg/Walloca-4.c: Same.
* gcc.dg/Walloca-8.c: Same.
* gcc.dg/Walloca-13.c: Same.
* gcc.dg/Walloca-14.c: Same.
* gcc.dg/attr-alloc_size-9.c: Same.
* gcc.dg/Walloca-1.c: Same.
* gcc.dg/tree-ssa/builtin-sprintf-warn-3.c: Same.
* gcc.dg/Walloca-5.c: Same.
* gcc.dg/Walloca-10.c: Same.
* gcc.dg/Walloca-9.c: Same.
* gcc.dg/attr-alloc_size-6.c: Same.
* gcc.dg/Wvla-larger-than-1.c: Same.
* gcc.dg/torture/pr71881.c: Same.
* gcc.dg/torture/pr71901.c: Same.
* gcc.dg/torture/pr78742.c: Same.
* gcc.dg/builtin-alloc-size.c: Same.
* gcc.dg/Walloca-2.c: Same.
* gcc.dg/Walloca-6.c: Same.
* gcc.dg/Walloca-11.c: Same.
* gcc.dg/attr-alloc_size-7.c: Same.
* gcc.dg/Wvla-larger-than-2.c: Same.
* gcc.dg/Walloca-3.c: Same.
* c-c++-common/Wimplicit-fallthrough-7.c: Same.
* gcc.c-torture/compile/pr79413.c: Same.
* gcc.c-torture/compile/pr78439.c: Same.

From-SVN: r246433

Daily bump.

From-SVN: r246431

p9-options-1.c: New test.

gcc/testsuite/ChangeLog:

2017-03-23  Kelvin Nilsen  <kelvin@gcc.gnu.org>

* gcc.target/powerpc/p9-options-1.c: New test.

gcc/ChangeLog:

2017-03-23  Kelvin Nilsen  <kelvin@gcc.gnu.org>

* config/rs6000/rs6000.c (rs6000_option_override_internal): Change
handling of certain combinations of target options, including the
combinations -mpower8-vector vs. -mno-vsx, -mpower9-vector vs.
-mno-power8-vector, and -mpower9_dform vs. -mno-power9-vector.

From-SVN: r246428

Fix Debug Mode test failures

* testsuite/23_containers/array/tuple_interface/
tuple_element_debug_neg.cc: Adjust dg-error.
* testsuite/23_containers/list/operations/78389.cc: Fix less-than to
define a valid strict weak ordering.
* testsuite/23_containers/priority_queue/67085.cc: Disable test for
Debug Mode, due to debug checks making extra copies of predicate.
* testsuite/ext/pb_ds/regression/priority_queue_binary_heap-62045.cc:
Likewise.

From-SVN: r246426

Fix broken links in manual and remove outdated info

* doc/xml/faq.xml: Add link.
* doc/xml/manual/backwards_compatibility.xml: Remove outdated
information on pre-ISO headers. Replace broken link to C++ FAQ Lite.
* doc/xml/manual/io.xml: Update broken link.
* doc/html/*: Regenerate.

From-SVN: r246425

Implement LWG 2686, std::hash<error_condition>, for C++17

2017-03-23  Daniel Kruegler  <daniel.kruegler@gmail.com>

Implement LWG 2686, Why is std::hash specialized for error_code,
but not error_condition?
* include/std/system_error (hash<error_condition>): Define for C++17.
* testsuite/20_util/hash/operators/size_t.cc (hash<error_condition>):
Instantiate test for error_condition.
* testsuite/20_util/hash/requirements/explicit_instantiation.cc
(hash<error_condition>): Instantiate hash<error_condition>.

From-SVN: r246424

Implement P0607R0 "Inline Variables for Standard Library" for C++17

2017-03-23  Daniel Kruegler  <daniel.kruegler@gmail.com>

* include/bits/c++config (_GLIBCXX17_INLINE): Define.
* include/bits/regex_constants.h (All std::regex_constants constants):
Add _GLIBCXX17_INLINE as per P0607R0.
* include/bits/std_mutex.h (defer_lock, try_to_lock, adopt_lock):
Likewise.
* include/bits/stl_pair.h (piecewise_construct): Likewise.
* include/bits/uses_allocator.h (allocator_arg, uses_allocator_v)
(__is_uses_allocator_constructible_v)
(__is_nothrow_uses_allocator_constructible_v): Likewise.
* include/std/chrono (treat_as_floating_point_v): Likewise.
* include/std/functional (is_bind_expression_v, is_placeholder_v):
Likewise.
* include/std/optional (nullopt): Likewise.
* include/std/ratio (ratio_equal_v, ratio_not_equal_v, ratio_less_v)
ratio_less_equal_v, ratio_greater_v, ratio_greater_equal_v): Likewise.
* include/std/system_error (is_error_code_enum_v)
(is_error_condition_enum_v): Likewise.
* include/std/tuple (tuple_size_v, ignore): Likewise.
(ignore): Declare ignore constexpr as per LWG 2773, declare assignment
constexpr as per LWG 2933.
* include/std/type_traits (All variable templates): Add
_GLIBCXX17_INLINE as per P0607R0.
* include/std/variant (variant_size_v, variant_npos, __index_of_v)
(__tuple_count_v, __exactly_once): Likewise.
* testsuite/18_support/headers/new/synopsis.cc
(hardware_destructive_interference_size)
(hardware_constructive_interference_size): Likewise for commented-out
variables.
* testsuite/20_util/tuple/creation_functions/constexpr.cc: Add new
test function for constexpr std::ignore (LWG 2773).
* testsuite/20_util/tuple/creation_functions/constexpr_cpp14.cc: New
test for LWG 2933.

From-SVN: r246423

PR c++/80150 - ICE with overloaded variadic deduction.

* pt.c (try_one_overload): Remove asserts.

From-SVN: r246422

[ARM] PR target/71436: Restrict *load_multiple pattern till after LRA

PR target/71436
* config/arm/arm.md (*load_multiple): Add reload_completed to
matching condition.

* gcc.c-torture/compile/pr71436.c: New test.

From-SVN: r246419

re PR tree-optimization/79908 (ICE in gimplify_expr (gimplify.c:12155) gimplification failed)

[gcc]

2017-03-23  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
            Richard Biener  <rguenth@suse.com>

PR tree-optimization/79908
PR tree-optimization/80136
* tree-stdarg.c (expand_ifn_va_arg_1): For a VA_ARG whose LHS has
been cast away, gimplify_and_add suffices.

[gcc/testsuite]

2017-03-23  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
            Richard Biener  <rguenther@suse.de>

PR tree-optimization/79908
PR tree-optimization/80136
* gcc.dg/torture/pr79908.c: New file.

From-SVN: r246418

PR c++/77563 - missing ambiguous conversion error.

* call.c (convert_like_real): Use LOOKUP_IMPLICIT.

From-SVN: r246417

c-tree.h: Remove a C_RID_YYCODE reference.

* c-tree.h: Remove a C_RID_YYCODE reference.

* cp-tree.h: Remove a C_RID_YYCODE reference.

From-SVN: r246416

Fix memory leak in identify_jump_threads()

* tree-vrp.c (identify_jump_threads): Delete avail_exprs.

From-SVN: r246415

re PR tree-optimization/80032 (C++ excessive stack usage (no stack reuse))

2017-03-23  Richard Biener  <rguenther@suse.de>

PR tree-optimization/80032
* gimplify.c (gimple_push_cleanup): Forced unconditional
cleanups still have to go to the conditional_cleanups
sequence.

From-SVN: r246414

Daily bump.

From-SVN: r246413

internal/syscall/unix: add randomTrap for m68k

    Patch by Andrwas Schwab.

    Reviewed-on: https://go-review.googlesource.com/38456

From-SVN: r246410

Also fixed by revision r246391.

2017-03-22  Dominique d'Humieres  <dominiq@lps.ens.fr>

PR fortran/79838
* module.c: Remove trailing period.

From-SVN: r246409

re PR tree-optimization/80072 (ICE in gimple_build_assign_1 with -O3 -march=broadwell/skylake-avx512)

PR tree-optimization/80072
* tree-ssa-reassoc.c (struct operand_entry): Change id field type
to unsigned int.
(next_operand_entry_id): Change type to unsigned int.
(sort_by_operand_rank): Make sure to return the right return value
even if unsigned fields are bigger than INT_MAX.
(struct oecount): Change cnt and id type to unsigned int.
(oecount_hasher::equal): Formatting fix.
(oecount_cmp): Make sure to return the right return value
even if unsigned fields are bigger than INT_MAX.
(undistribute_ops_list): Change next_oecount_id type to unsigned int.

From-SVN: r246408

compiler: initialize gogo fields

    A couple of the data members in the Gogo class were not
    being initialized properly. This was causing "uninitialized value"
    errors during Valgrind memcheck runs. This patch insures that
    all of the fields receive an initial value.

    Reviewed-on: https://go-review.googlesource.com/38470

From-SVN: r246407

re PR fortran/39239 (Reject SAVEd variables EQUIVALENCEd to a COMMON)

2017-03-12 Nicolas Koenig <koenigni@student.ethz.ch>

PR fortran/39239
fortran/Changelog:  Add entry for rev. 246284.

From-SVN: r246406

re PR fortran/39239 (Reject SAVEd variables EQUIVALENCEd to a COMMON)

2017-03-22 Nicolas Koenig <koenigni@student.ethz.ch>

PR fortran/39239
* gfortran.dg/equiv_constraint_bind_c.f90: New test.

From-SVN: r246405

ChangeLog: Add forgotten entry for r246322.

2017-03-21  Nicolas Koenig  <koenigni@student.ethz.ch>

* fortran/ChangeLog: Add forgotten entry for r246322.
* testsuite/ChangeLog: Likewise.

From-SVN: r246404

re PR c++/80141 (ICE with pragma omp declare)

PR c++/80141
* semantics.c (finish_omp_clause) <case OMP_CLAUSE_SIMDLEN,
case OMP_CLAUSE_ALIGNED>: Call maybe_constant_value only when not
processing_template_decl.

* g++.dg/gomp/pr80141.C: New test.

From-SVN: r246403

re PR sanitizer/78158 (Strange data race detection with thread sanitizer)

PR sanitizer/78158
* tsan/tsan_interface_atomic.cc: Cherry-pick upstream r298378.

From-SVN: r246402

re PR c++/80129 (wrong code with ternary struct assignment to const)

PR c++/80129
* gimplify.c (gimplify_modify_expr_rhs) <case COND_EXPR>: Clear
TREE_READONLY on result if writing it more than once.

* g++.dg/torture/pr80129.C: New test.

From-SVN: r246401

re PR sanitizer/80110 (error: statement marked for throw, but doesn’t w/ -fsanitize=thread)

PR sanitizer/80110
* doc/invoke.texi (-fsanitize=thread): Document that with
-fnon-call-exceptions atomics are not able to throw
exceptions.

From-SVN: r246400

re PR sanitizer/80110 (error: statement marked for throw, but doesn’t w/ -fsanitize=thread)

PR sanitizer/80110
* tsan.c: Include tree-eh.h.
(instrument_builtin_call): Call maybe_clean_eh_stmt or
maybe_clean_or_replace_eh_stmt where needed.
(instrument_memory_accesses): Add cfg_changed argument.
Call gimple_purge_dead_eh_edges on each block and set *cfg_changed
if it returned true.
(tsan_pass): Adjust caller.  Return TODO_cleanup_cfg if cfg_changed.

* g++.dg/tsan/pr80110.C: New test.

From-SVN: r246399

re PR rtl-optimization/63191 (32-bit gcc uses excessive memory during dead store elimination with -fPIC)

PR rtl-optimization/63191
* config/i386/i386.c (ix86_delegitimize_address): Turn into small
wrapper function, moved the whole old content into ...
(ix86_delegitimize_address_1): ... this.  New inline function.
(ix86_find_base_term): Use ix86_delegitimize_address_1 with
true as last argument instead of ix86_delegitimize_address.

From-SVN: r246398

Recently we've put a lot of effort into improving ifcvt to use CSEL on AArch64.

In  https://gcc.gnu.org/ml/gcc-patches/2015-11/msg01639.html James determined
the best value for AArch64 code generation.  Although this setting is used when
explicitly targeting Cortex cores, it is not otherwise used.  This means by
default GCC will not use (F)CSEL in many common cases.

Change the generic_branch_cost to be the same as cortexa57_branch_cost so that
all supported cores benefit from CSEL.  This is generally faster and smaller.
On one benchmark the new setting fixes a regression since GCC6 and improves
performance by 49%.

    gcc/
* config/aarch64/aarch64.c (generic_branch_cost):
Copy cortexa57_branch_cost.

From-SVN: r246397

configure.ac: Enable LTO by default on darwin >= 9.

* configure.ac: Enable LTO by default on darwin >= 9.
* configure: Regenerate.

From-SVN: r246396

Many supported cores implement fusion of AES instructions.

Many supported cores implement fusion of AES instructions.  When fusion
happens it can give a significant performance gain.  If not, scheduling
fusion candidates next to each other has almost no effect on performance.
Due to the high benefit/low cost it makes sense to enable AES fusion with
-mcpu=generic so that cores that support it always benefit.

    gcc/
* config/aarch64/aarch64.c (generic_tunings): Add AES fusion.

From-SVN: r246395

re PR target/80123 (libgomp tests pr66199-2.c and pr66199-5.c fail with -mcpu=power9)

2017-03-21  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>

PR target/80123
* doc/md.texi (Constraints): Document wA constraint.
* config/rs6000/constraints.md (wA): New.
* config/rs6000/rs6000.c (rs6000_debug_reg_global): Add wA reg_class.
(rs6000_init_hard_regno_mode_ok): Init wA constraint.
* config/rs6000/rs6000.h (RS6000_CONSTRAINT_wA): New.
* config/rs6000/vsx.md (vsx_splat_<mode>): Use wA constraint.

From-SVN: r246394

re PR fortran/79602 (translation: globally replace '%s' with %qs)

2017-03-22  Dominique d'Humieres  <dominiq@lps.ens.fr>

PR fortran/79602
* decl.c: Replace '%s' with %qs.
* expr.c: Likewise.
* interface.c: Likewise.
* match.c: Likewise.
* primary.c: Likewise.
* resolve.c: Likewise.

PR fortran/79844
PR fortran/80011
* io.c: Remove trailing spaces.
* match.c: Likewise.
* openmp.c: Likewise.
* resolve.c: Likewise.
* trans-intrinsic.c: Likewise.

PR fortran/79853
* expr.c: Remove a double spaces.

PR fortran/79859
* primary.c: Remove spurious quotes around %qs.

From-SVN: r246391

Add deduction guides for C++17 (P0433R2, partial)

* include/bits/shared_ptr.h (shared_ptr, weak_ptr): Add deduction
guides for C++17.
* include/bits/std_function.h (function): Likewise.
* include/bits/stl_pair.h (pair): Likewise.
* include/debug/array (__gnu_debug::array): Likewise.
* include/std/array (array): Likewise.
* include/std/functional (make_default_searcher)
(make_boyer_moore_searcher, make_boyer_moore_horspool_searcher):
Remove generator functions.
* include/std/tuple (tuple): Add deduction guides.
* include/std/valarray (valarray): Likewise.
* testsuite/20_util/function_objects/searchers.cc: Adjust to use
class template argument deduction instead of generator functions.
* testsuite/20_util/function/cons/deduction.cc: New test.
* testsuite/20_util/optional/cons/deduction_guide.cc: Rename to ...
* testsuite/20_util/optional/cons/deduction.cc: ... here.
* testsuite/20_util/pair/cons/deduction.cc: New test.
* testsuite/20_util/shared_ptr/cons/deduction.cc: New test.
* testsuite/20_util/tuple/cons/deduction.cc: New test.
* testsuite/20_util/tuple/element_access/get_neg.cc: Adjust dg-error.
* testsuite/20_util/unique_ptr/cons/deduction_neg.cc: New test.
* testsuite/20_util/weak_ptr/cons/deduction.cc: New test.
* testsuite/23_containers/array/cons/deduction.cc: New test.
* testsuite/23_containers/array/cons/deduction_neg.cc: New test.
* testsuite/23_containers/array/tuple_interface/get_debug_neg.cc:
Adjust dg-error.
* testsuite/23_containers/array/tuple_interface/get_neg.cc: Likewise.
* testsuite/23_containers/array/tuple_interface/tuple_element_neg.cc:
Likewise.
* testsuite/26_numerics/valarray/deduction.cc: New test.
* testsuite/30_threads/lock_guard/cons/deduction.cc: New test.
* testsuite/30_threads/scoped_lock/cons/deduction.cc: New test.
* testsuite/30_threads/unique_lock/cons/deduction.cc: New test.

From-SVN: r246389

re PR fortran/80142 (Warning: No location in expression ... with -O / -ffrontend-optimize)

2017-03-22  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/80142
* frontend-passes.c (combine_array_constructor): Take
location of new expression from constructor expression instead
of constructor.

2017-03-22  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/80142
* gfortran.dg/any_loc.f90: New test case.

From-SVN: r246388

re PR go/80128 (go1: internal compiler error: in write_specific_type_functions, at go/gofrontend/types.cc:2002)

PR go/80128
    compiler: check backend alignment for memequalNN functions

    The code was assuming the usual required alignment for the memequalNN
    functions (16 bits for int16, 32 for int32, etc.). However, on m68k
    the required alignment of int32 is only 16 bits. Assuming the
    memequalNN alignment caused the compiler to incorrectly decide that
    int32 required a specially generated function rather than calling
    memequal32. This then crashed if the type descriptor were generated
    after type-specific functions had been written.

    Fixes GCC PR 80128.

    Reviewed-on: https://go-review.googlesource.com/38433

From-SVN: r246382

re PR c++/80029 (valgrind error in new_omp_context(omp_region_type) (gimplify.c:400))

PR c++/80029

gcc/
* gimplify.c (is_oacc_declared): New function.
(oacc_default_clause): Use it to set default flags for acc declared
variables inside parallel regions.
(gimplify_scan_omp_clauses): Strip firstprivate pointers for acc
declared variables.
(gimplify_oacc_declare): Gimplify the declare clauses.  Add the
declare attribute to any decl as necessary.

libgomp/
* testsuite/libgomp.oacc-c-c++-common/declare-vla.c: New test.

From-SVN: r246381

Fix PR80082: LDRD erronously used for 64bit load on ARMv7-R

2017-03-22  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    gcc/
    PR target/80082
    * config/arm/arm-isa.h (isa_bit_lpae): New feature bit.
    (ISA_ARMv7ve): Add isa_bit_lpae to the definition.
    * config/arm/arm-protos.h (arm_arch7ve): Rename into ...
    (arm_arch_lpae): This.
    * config/arm/arm.c (arm_arch7ve): Rename into ...
    (arm_arch_lpae): This.  Define it in term of isa_bit_lpae.
    * config/arm/arm.h (TARGET_HAVE_LPAE): Redefine in term of
    arm_arch_lpae.

    gcc/testsuite/
    PR target/80082
    * gcc.target/arm/atomic_loaddi_10.c: New testcase.
    * gcc.target/arm/atomic_loaddi_11.c: Likewise.

From-SVN: r246365

Error message on target attribute on power target (PR target/79906)

2017-03-22  Martin Liska  <mliska@suse.cz>

PR target/79906
* config/rs6000/rs6000.c (rs6000_inner_target_options): Show
error message instead of an ICE.
2017-03-22  Martin Liska  <mliska@suse.cz>

PR target/79906
* g++.dg/ext/mv8.C: Add power* targets.

From-SVN: r246345

extend.texi (6.11 Additional Floating Types): Revise.

2017-03-21  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

* doc/extend.texi (6.11 Additional Floating Types): Revise.

From-SVN: r246343

Daily bump.

From-SVN: r246342

rs6000-c.c (rs6000_target_modify_macros): Add comments.

gcc/ChangeLog:

2017-03-21  Kelvin Nilsen  <kelvin@gcc.gnu.org>

* config/rs6000/rs6000-c.c (rs6000_target_modify_macros): Add
comments.
* config/rs6000/rs6000.c (rs6000_option_override_internal): Add
comments.

From-SVN: r246339

PR c++/79548 - missing -Wunused-variable on a typedef'd variable in a function template

gcc/c-family/ChangeLog:

PR c++/79548
* c-common.c (set_underlying_type): Mark type used only when
original del is declared unused.

gcc/testsuite/ChangeLog:

PR c++/79548
* g++.dg/warn/Wunused-var-26.C: New test.

From-SVN: r246335

Use the more formal "cannot" instead of the informal "can't."

gcc/ChangeLog:
* doc/extend.texi: Use "cannot" instead of "can't."
* doc/hostconfig.texi: Same.
* doc/install.texi: Same.
* doc/invoke.texi: Same.
* doc/loop.texi: Same.
* doc/md.texi: Same.
* doc/objc.texi: Same.
* doc/rtl.texi: Same.
* doc/tm.texi: Same.
* doc/tm.texi.in: Same.
* doc/trouble.texi: Same.

From-SVN: r246334

[PR63238] include alignment debug information in DIE checksum

Add DW_AT_alignment to the DIE checksum.

for gcc/ChangeLog

PR debug/63238
* dwarf2out.c (struct checksum_attributes): Add at_alignment.
(collect_checksum_attributes): Set it.
(die_checksum_ordered): Use it.

From-SVN: r246331

re PR tree-optimization/79908 (ICE in gimplify_expr (gimplify.c:12155) gimplification failed)

[gcc]

2017-03-21  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

PR tree-optimization/79908
* tree-stdarg.c (expand_ifn_va_arg_1): Revert the following
change: For a VA_ARG whose LHS has been cast away, use
force_gimple_operand to construct the side effects.

[gcc/testsuite]

2017-03-21  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

PR tree-optimization/79908
* gcc.dg/torture/pr79908.c: Revert addition of new file.

From-SVN: r246330

omp-offload.c: translation fixes (PR translation/80001)

gcc/ChangeLog:
PR translation/80001
* omp-offload.c (oacc_loop_fixed_partitions): Make diagnostics
more amenable to translation.
(oacc_loop_auto_partitions): Likewise.

From-SVN: r246326

re PR tree-optimization/80109 (ICE in get_range_info, at tree-ssanames.c:375)

PR tree-optimization/80109
* gimple-ssa-warn-alloca.c (alloca_call_type): Only call get_range_info
on INTEGRAL_TYPE_P.

* gcc.dg/Walloca-14.c: New test.

Co-Authored-By: Martin Sebor <msebor@redhat.com>
From-SVN: r246325

re PR target/80125 (r246297 causes segfault in reg_used_between_p())

PR target/80125
* combine.c (can_combine_p): Revert the 2017-03-20 change, only
check reg_used_between_p between insn and one of succ or succ2
depending on if succ is artificial insn not inserted into insn
stream.

* gcc.target/powerpc/pr80125.c: New test.

Co-Authored-By: Segher Boessenkool <segher@kernel.crashing.org>
From-SVN: r246323

re PR fortran/69498 (ICE on disjunct cases with displaced or incomplete embedded statement)

2017-03-18  Nicolas Koenig  <koenigni@student.ethz.ch>

                PR fortran/69498
                * decl.c (add_hidden_procptr_result): Fixed Refs count of the created "ppr@" symbol.

2017-03-18  Nicolas Koenig  <koenigni@student.ethz.ch>

                PR fortran/69498
                * gfortran.dg/unexp_attribute.f90: New test

From-SVN: r246322

Document gcov-dump and fix installation of gcov-tool (PR gcov-profile/80081).

2017-03-21  Martin Liska  <mliska@suse.cz>

PR gcov-profile/80081
* Makefile.in: Add gcov-dump and fix installation of gcov-tool.
* doc/gcc.texi: Include gcov-dump stuff.
* doc/gcov-dump.texi: New file.

From-SVN: r246321

Apply temporary fix for PR rtl-optimization/79150.

gcc/
PR rtl-optimization/79150
* config/mips/mips.c (mips_block_move_loop): Emit a NOP after the
conditional jump, if the jump is the last insn of the loop.

From-SVN: r246320

re PR tree-optimization/79908 (ICE in gimplify_expr (gimplify.c:12155) gimplification failed)

[gcc]

2017-03-21  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
            Richard Biener  <rguenth@suse.com>

PR tree-optimization/79908
* tree-stdarg.c (expand_ifn_va_arg_1): For a VA_ARG whose LHS has
been cast away, use force_gimple_operand to construct the side
effects.

[gcc/testsuite]

2017-03-21  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
            Richard Biener  <rguenther@suse.de>

PR tree-optimization/79908
* gcc.dg/torture/pr79908.c: New file.

Co-Authored-By: Richard Biener <rguenth@suse.com>
Co-Authored-By: Richard Biener <rguenther@suse.de>
From-SVN: r246319

Fix failing overflow-1.c for avr

The test assumes 32 bit ints, and expects a constant in the
dump that is only valid for 32 bit ints. Fix by explicitly
specifying __UINT32_TYPE__.

gcc/testsuite/

2017-03-21  Senthil Kumar Selvaraj  <senthil_kumar.selvaraj@atmel.com>

* gcc.dg/tree-ssa/overflow-1.c: Use __UINT32_TYPE__ for targets
with sizeof(int) < 4.

From-SVN: r246318

Fix another profiledbootstrap warning (PR libfortran/79956).

2017-03-21  Martin Liska  <mliska@suse.cz>

PR libfortran/79956
* simplify-rtx.c (simplify_immed_subreg): Initialize a variable
to NULL.

From-SVN: r246317

Fix dg-error for a test

2017-03-21  Martin Liska  <mliska@suse.cz>

* gcc.target/i386/pr65044.c: Add '.' in order to catch
apostrophes.

From-SVN: r246316

re PR plugins/80094 (GCC plugin hash table corruption on hash table expansion (>10 plugins) on GCC 4.5+)

2017-03-21  Brad Spengler <spender@grsecurity.net>

PR plugins/80094
* plugin.c (htab_hash_plugin): New function.
(add_new_plugin): Use it and adjust.
(parse_plugin_arg_opt): Adjust.
(init_one_plugin): Likewise.

From-SVN: r246315

re PR tree-optimization/80032 (C++ excessive stack usage (no stack reuse))

2017-03-21  Richard Biener  <rguenther@suse.de>

PR tree-optimization/80032
* gimplify.c (gimple_push_cleanup): Add force_uncond parameter,
if set force the cleanup to happen unconditionally.
(gimplify_target_expr): Push inserted clobbers with force_uncond
to avoid them being removed by control-dependent DCE.

* g++.dg/opt/pr80032.C: New testcase.

From-SVN: r246314

re PR tree-optimization/80122 (__builtin_va_arg_pack() and __builtin_va_arg_pack_len() does not work correctly)

2017-03-21  Richard Biener  <rguenther@suse.de>

PR tree-optimization/80122
* tree-inline.c (copy_bb): Do not expans va-arg packs or
va_arg_pack_len when the inlined call stmt requires pack
expansion itself.
* tree-inline.h (struct copy_body_data): Make call_stmt a gcall *.

* gcc.dg/torture/pr80122.c: New testcase.

From-SVN: r246313

Fix search_line_fast for aarch64/ILP32

* lex.c (search_line_fast) [__ARM_NEON && __ARM_64BIT_STATE]:
Convert 64-bit value to boolean before passing to
__builtin_expect.

From-SVN: r246312

Skip gcc.dg/pic-2.c and gcc.dg/pie-2.c for MIPS.

gcc/testsuite/

* gcc.dg/pic-2.c: Skip for MIPS.
* gcc.dg/pie-2.c: Skip for MIPS.

From-SVN: r246311

re PR c++/77752 (ICE on C++ code on x86_64-linux-gnu (internal compiler error: Segmentation fault, build_list_conv, implicit_conversion))

/cp
2017-03-21  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/77752
* name-lookup.c (pushtag_1): Add check for bogus, non template,
std::initializer_list.

/testsuite
2017-03-21  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/77752
* g++.dg/cpp0x/initlist97.C: New.
* g++.dg/cpp0x/initlist85.C: Update.

From-SVN: r246310

re PR sanitizer/78158 (Strange data race detection with thread sanitizer)

PR sanitizer/78158
* tsan.c (instrument_builtin_call): If the memory model argument
is not a constant, assume it is valid.

From-SVN: r246306

re PR middle-end/67338 (fold-const sanitizer runtime error  in roundup_loc)

PR c/67338
* fold-const.c (round_up_loc): Negate divisor in unsigned type to
avoid UB.

* gcc.dg/pr67338.c: New test.

From-SVN: r246305

Fix CL.

From-SVN: r246304

re PR c++/35878 ([LWG 2302] Useless NULL pointer check when constructing object)

PR c++/35878
* init.c (std_placement_new_fn_p, build_new_1): Formatting fixes.

* g++.dg/init/pr35878_1.C: Rewrite directives to scan optimized
dump instead of assembler.
* g++.dg/init/pr35878_2.C: Likewise.
* g++.dg/init/pr35878_3.C: Likewise.

From-SVN: r246303

re PR c/80097 (internal compiler error in c_fully_fold_internal with std=c89 and -fsanitize=float-divide-by-zero)

PR c/80097
* c-typeck.c (build_binary_op): Add EXCESS_PRECISION_EXPR only around
optional COMPOUND_EXPR with ubsan instrumentation.

* gcc.dg/ubsan/pr80097.c: New test.

From-SVN: r246302

re PR c++/35878 ([LWG 2302] Useless NULL pointer check when constructing object)

gcc/

PR c++/35878
* cp/init.c (std_placement_new_fn_p): New.
(build_new_1): Call it.

testsuite/

PR c++/35878
* g++.dg/init/pr35878_1.C: New.
* g++.dg/init/pr35878_2.C: Likewise.
* g++.dg/init/pr35878_3.C: Likewise.

From-SVN: r246301

Daily bump.

From-SVN: r246300

combine: Fix 79910

If the dest of an I0 or I1 is used in an insn before I2, as can happen
in various uncommon cases, and we manage to do the combination, the set
is moved to I2, which is wrong.  Don't allow combining the insns in this
case.

PR rtl-optimization/79910
* combine.c (can_combine_p): Do not allow combining an I0 or I1
if its dest is used by an insn before I2 (other than the combined
insns themselves, which are properly handled already).

From-SVN: r246297

Revert:

2017-03-17  Bernd Schmidt  <bschmidt@redhat.com>

* combine.c (record_used_regs): New static function.
(try_combine): Handle situations where there is an additional
instruction between I2 and I3 which needs to have a LOG_LINK
updated.

Revert:
2017-03-17  Jim Wilson  <jim.wilson@linaro.org>

* combine.c (try_combine): Delete redundant i1 test.  Call
prev_nonnote_nondebug_insn instead of prev_nonnote_insn.

From-SVN: r246296

* de.po, fr.po: Update.

From-SVN: r246295

stl_deque.h (deque): Access allocator value_type only if concept checks are enabled.

2017-03-20  François Dumont  <fdumont@gcc.gnu.org>

* include/bits/stl_deque.h (deque): Access allocator value_type only if
concept checks are enabled.
* include/bits/stl_stack.h (stack): Likewise.
* include/bits/stl_vector.h (vector): Likewise.
* include/bits/stl_list.h (list): Likewise and check
_SGIAssignableConcept only in C++03.
* include/bits/stl_map.h (map): Likewise.
* include/bits/stl_set.h (set): Likewise.
* include/bits/stl_multimap.h (multimap): Likewise.
* include/bits/stl_multiset.h (multiset): Likewise.
* include/bits/stl_queue.h (queue, priority_queue): Likewise.

From-SVN: r246294

re PR target/80083 (libgomp doacross2.f90 regtest fails with -mcpu=power9 -O1)

2017-03-20  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>

PR target/80083
* config/rs6000/rs6000.md (*movsi_internal1): Fix constraints for
alternatives 13/14.

From-SVN: r246293