[16/77] Add scalar_int_mode_pod

This patch adds a POD class for scalar integers, as an instance
of a new pod_mode template.  Later patches will use pod_mode in
situations that really do need to be POD; this patch is simply
using PODs to remove load-time initialisation.

2017-08-30  Richard Sandiford  <richard.sandiford@linaro.org>
    Alan Hayward  <alan.hayward@arm.com>
    David Sherwood  <david.sherwood@arm.com>

gcc/
* coretypes.h (pod_mode): New type.
(scalar_int_mode_pod): New typedef.
* machmode.h (pod_mode): New class.
(int_n_data_t::m): Change type to scalar_int_mode_pod.
* genmodes.c (emit_mode_int_n): Update accordingly.
* lower-subreg.h (target_lower_subreg): Change type to
scalar_int_mode_pod.
* gdbhooks.py (build_pretty_printer): Handle pod_mode and
scalar_int_mode_pod.

Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r251468

[15/77] Add scalar_int_mode

Similar to the previous scalar_float_mode patch, but for modes that
satisfy SCALAR_INT_MODE_P.  There are very many uses of scalar integers,
so this patch only makes a token change to the types of byte_mode,
word_mode, ptr_mode and rs6000_pmode.  The next patches in the series
gradually replace more uses.

The patch also removes and adds casts to some target-specific code
due to the new types of SImode, DImode and Pmode.

The as_a <scalar_int_mode> goes away in a later patch.

2017-08-30  Richard Sandiford  <richard.sandiford@linaro.org>
    Alan Hayward  <alan.hayward@arm.com>
    David Sherwood  <david.sherwood@arm.com>

gcc/
* config/powerpcspe/powerpcspe.h (rs6000_pmode): Change type from
machine_mode to scalar_int_mode.
* config/powerpcspe/powerpcspe.c (rs6000_pmode): Likewise.
(rs6000_option_override_internal): Remove cast to int.
* config/rs6000/rs6000.h (rs6000_pmode): Change type from
machine_mode to scalar_int_mode.
* config/rs6000/rs6000.c (rs6000_pmode): Likewise.
(rs6000_option_override_internal): Remove cast to int.
* config/s390/s390.h (Pmode): Remove cast to machine_mode.
* config/epiphany/epiphany.h (RTX_OK_FOR_OFFSET_P): Add cast
to machine_mode.
* config/s390/s390.c (s390_expand_builtin): Likewise.
* coretypes.h (scalar_int_mode): New type.
(opt_scalar_int_mode): New typedef.
* machmode.h (scalar_int_mode): New class.
(scalar_int_mode::includes_p): New function.
(byte_mode): Change type to scalar_int_mode.
(word_mode): Likewise.
(ptr_mode): Likewise.
* emit-rtl.c (byte_mode): Likewise.
(word_mode): Likewise.
(ptr_mode): Likewise.
(init_derived_machine_modes): Update accordingly.
* genmodes.c (get_mode_class): Return scalar_int_mode for MODE_INT
and MODE_PARTIAL_INT.
* gdbhooks.py (build_pretty_printer): Handle scalar_int_mode and
opt_scalar_int_mode.

Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r251467

[14/77] Make libgcc_floating_mode_supported_p take a scalar_float_mode

As per subject.

2017-08-30  Richard Sandiford  <richard.sandiford@linaro.org>
    Alan Hayward  <alan.hayward@arm.com>
    David Sherwood  <david.sherwood@arm.com>

gcc/
* target.def (libgcc_floating_mode_supported_p): Take a
scalar_float_mode.
* doc/tm.texi: Regenerate.
* targhooks.h (default_libgcc_floating_mode_supported_p): Take a
scalar_float_mode.
* targhooks.c (default_libgcc_floating_mode_supported_p): Likewise.
* config/aarch64/aarch64.c (aarch64_libgcc_floating_mode_supported_p):
Likewise.

Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r251466

[13/77] Make floatn_mode return an opt_scalar_float_mode

As per subject.

2017-08-30  Richard Sandiford  <richard.sandiford@linaro.org>
    Alan Hayward  <alan.hayward@arm.com>
    David Sherwood  <david.sherwood@arm.com>

gcc/
* target.def (default_floatn_mode): Return an opt_scalar_float_mode.
* doc/tm.texi: Regenerate.
* config/arm/arm.c (arm_floatn_mode): Return an opt_scalar_float_mode.
* config/powerpcspe/powerpcspe.c (rs6000_floatn_mode): Likewise.
* config/rs6000/rs6000.c (rs6000_floatn_mode): Likewise.
* targhooks.h (default_floatn_mode): Likewise.
* targhooks.c (default_floatn_mode): Likewise.
* tree.c (build_common_tree_nodes): Update accordingly.

Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r251465

[12/77] Use opt_scalar_float_mode when iterating over float modes

This means that we know when accessing the modes that the size is
a compile-time constant, even for SVE.  It also enables stricter
type safety in later patches.

2017-08-30  Richard Sandiford  <richard.sandiford@linaro.org>
    Alan Hayward  <alan.hayward@arm.com>
    David Sherwood  <david.sherwood@arm.com>

gcc/
* machmode.h (mode_iterator::start): Provide overload for opt_modes.
(mode_iterator::iterate_p): Likewise.
(mode_iterator::get_wider): Likewise.
* expr.c (init_expr_target): Use opt_scalar_float_mode.

gcc/ada/
* gcc-interface/misc.c (fp_prec_to_size): Use opt_scalar_float_mode.
(fp_size_to_prec): Likewise.

gcc/c-family/
* c-cppbuiltin.c (c_cpp_builtins): Use opt_scalar_float_mode.

gcc/fortran/
* trans-types.c (gfc_init_kinds): Use opt_scalar_float_mode
and FOR_EACH_MODE_IN_CLASS.

Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r251464

[11/77] Add a float_mode_for_size helper function

This provides a type-safe way to ask for a float mode and get it as a
scalar_float_mode.

2017-08-30  Richard Sandiford  <richard.sandiford@linaro.org>
    Alan Hayward  <alan.hayward@arm.com>
    David Sherwood  <david.sherwood@arm.com>

gcc/
* coretypes.h (opt_scalar_float_mode): New typedef.
* machmode.h (float_mode_for_size): New function.
* emit-rtl.c (double_mode): Delete.
(init_emit_once): Use float_mode_for_size.
* stor-layout.c (layout_type): Likewise.
* gdbhooks.py (build_pretty_printer): Handle opt_scalar_float_mode.

Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r251463

Fix test-case vrp104.c.

2017-08-30  Martin Liska  <mliska@suse.cz>

* gcc.dg/tree-ssa/vrp104.c: Change dump file name.

From-SVN: r251462

[10/77] Make assemble_real take a scalar_float_mode

As per subject.

2017-08-30  Richard Sandiford  <richard.sandiford@linaro.org>
    Alan Hayward  <alan.hayward@arm.com>
    David Sherwood  <david.sherwood@arm.com>

gcc/
* output.h (assemble_real): Take a scalar_float_mode.
* config/arm/arm.c (arm_assemble_integer): Update accordingly.
* config/arm/arm.md (consttable_4): Likewise.
(consttable_8): Likewise.
(consttable_16): Likewise.
* config/mips/mips.md (consttable_float): Likewise.
* config/s390/s390.c (s390_output_pool_entry): Likewise.
* varasm.c (assemble_real): Take a scalar_float_mode.
(output_constant_pool_2): Update accordingly.
(output_constant): Likewise.

Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r251461

[9/77] Add SCALAR_FLOAT_TYPE_MODE

This patch adds a macro that extracts the TYPE_MODE and forcibly
converts it to a scalar_float_mode.  The forcible conversion
includes a gcc_checking_assert that the mode is a SCALAR_FLOAT_MODE_P.

This becomes important as more static type checking is added by
later patches.  It has the additional benefit of bypassing the
VECTOR_TYPE_P (...) ? vector_type_mode (...) : ... condition
in TYPE_MODE; in release builds the new macro is a simple
field access.

2017-08-30  Richard Sandiford  <richard.sandiford@linaro.org>
    Alan Hayward  <alan.hayward@arm.com>
    David Sherwood  <david.sherwood@arm.com>

gcc/
* tree.h (SCALAR_FLOAT_TYPE_MODE): New macro.
* builtins.c (expand_builtin_signbit): Use it instead of TYPE_MODE.
* fold-const.c (fold_convert_const_real_from_fixed): Likewise.
(native_encode_real): Likewise.
(native_interpret_real): Likewise.
* hsa-brig.c (emit_immediate_scalar_to_buffer): Likewise.
* tree-vrp.c (simplify_float_conversion_using_ranges): Likewise.

gcc/cp/
* mangle.c (write_real_cst): Use SCALAR_FLOAT_TYPE_MODE
instead of TYPE_MODE.

gcc/fortran/
* target-memory.c (size_float): Use SCALAR_FLOAT_TYPE_MODE
instead of TYPE_MODE.

gcc/objc/
* objc-encoding.c (encode_type): Use SCALAR_FLOAT_TYPE_MODE
instead of TYPE_MODE.

Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r251460

[8/77] Simplify gen_trunc/extend_conv_libfunc

Replace checks of:

  GET_MODE_CLASS (fmode) != MODE_FLOAT && !DECIMAL_FLOAT_MODE_P (fmode)

with !is_a <scalar_float_mode> and use MODE_CLASS equality/inequality
instead of:

  (GET_MODE_CLASS (tmode) == MODE_FLOAT && DECIMAL_FLOAT_MODE_P (fmode))
  || (GET_MODE_CLASS (fmode) == MODE_FLOAT && DECIMAL_FLOAT_MODE_P (tmode))

and:

  (GET_MODE_CLASS (tmode) == MODE_FLOAT
   && GET_MODE_CLASS (fmode) == MODE_FLOAT)
  || (DECIMAL_FLOAT_MODE_P (fmode) && DECIMAL_FLOAT_MODE_P (tmode))

2017-08-30  Richard Sandiford  <richard.sandiford@linaro.org>
    Alan Hayward  <alan.hayward@arm.com>
    David Sherwood  <david.sherwood@arm.com>

gcc/
* optabs-libfuncs.c (gen_trunc_conv_libfunc): Use is_a
<scalar_float_mode>.  Simplify.
(gen_extend_conv_libfunc): Likewise.

Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r251459

[7/77] Add scalar_float_mode

This patch adds a scalar_float_mode class, which wraps a mode enum
that is known to satisfy SCALAR_FLOAT_MODE_P.  Things like "SFmode"
now give a scalar_float_mode object instead of a machine_mode.
This in turn needs a change to the real.h format_helper, so that
it can accept both machine_modes and scalar_float_modes.

2017-08-30  Richard Sandiford  <richard.sandiford@linaro.org>
    Alan Hayward  <alan.hayward@arm.com>
    David Sherwood  <david.sherwood@arm.com>

gcc/
* coretypes.h (scalar_float_mode): New type.
* machmode.h (mode_traits::from_int): Use machine_mode if
USE_ENUM_MODES is defined.
(is_a): New function.
(as_a): Likewise.
(dyn_cast): Likewise.
(scalar_float_mode): New class.
(scalar_float_mode::includes_p): New function.
(is_float_mode): Likewise.
* gdbhooks.py (MachineModePrinter): New class.
(build_pretty_printer): Use it for scalar_float_mode.
* real.h (FLOAT_MODE_FORMAT): Use as_a <scalar_float_mode>.
(format_helper::format_helper): Turn into a template.
* genmodes.c (get_mode_class): New function.
(emit_insn_modes_h): Give modes the class returned by get_mode_class,
or machine_mode if none.
* config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Use
as_a <scalar_float_mode>.
* dwarf2out.c (mem_loc_descriptor): Likewise.
(insert_float): Likewise.
(add_const_value_attribute): Likewise.
* simplify-rtx.c (simplify_immed_subreg): Likewise.
* optabs.c (expand_absneg_bit): Take a scalar_float_mode.
(expand_unop): Update accordingly.
(expand_abs_nojump): Likewise.
(expand_copysign_absneg): Take a scalar_float_mode.
(expand_copysign_bit): Likewise.
(expand_copysign): Update accordingly.

gcc/ada/
* gcc-interface/utils.c (gnat_type_for_mode): Use is_a
<scalar_float_mode> instead of SCALAR_FLOAT_MODE_P.

gcc/go/
* go-lang.c (go_langhook_type_for_mode): Use is_float_mode.

Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r251458

[6/77] Make GET_MODE_WIDER return an opt_mode

GET_MODE_WIDER previously returned VOIDmode if no wider mode existed.
That would cause problems with stricter mode classes, since VOIDmode
isn't for example a valid scalar integer or floating-point mode.
This patch instead makes it return a new opt_mode<T> class, which
holds either a T or nothing.

2017-08-30  Richard Sandiford  <richard.sandiford@linaro.org>
    Alan Hayward  <alan.hayward@arm.com>
    David Sherwood  <david.sherwood@arm.com>

gcc/
* coretypes.h (opt_mode): New class.
* machmode.h (opt_mode): Likewise.
(opt_mode::else_void): New function.
(opt_mode::require): Likewise.
(opt_mode::exists): Likewise.
(GET_MODE_WIDER_MODE): Turn into a function and return an opt_mode.
(GET_MODE_2XWIDER_MODE): Likewise.
(mode_iterator::get_wider): Update accordingly.
(mode_iterator::get_2xwider): Likewise.
(mode_iterator::get_known_wider): Likewise, turning into a template.
* combine.c (make_extraction): Update use of GET_MODE_WIDER_MODE,
forcing a wider mode to exist.
* config/cr16/cr16.h (LONG_REG_P): Likewise.
* rtlanal.c (init_num_sign_bit_copies_in_rep): Likewise.
* config/c6x/c6x.c (c6x_rtx_costs): Update use of
GET_MODE_2XWIDER_MODE, forcing a wider mode to exist.
* lower-subreg.c (init_lower_subreg): Likewise.
* optabs-libfuncs.c (init_sync_libfuncs_1): Likewise, but not
on the final iteration.
* config/i386/i386.c (ix86_expand_set_or_movmem): Check whether
a wider mode exists before asking for a move pattern.
(get_mode_wider_vector): Update use of GET_MODE_WIDER_MODE,
forcing a wider mode to exist.
(expand_vselect_vconcat): Update use of GET_MODE_2XWIDER_MODE,
returning false if no such mode exists.
* config/ia64/ia64.c (expand_vselect_vconcat): Likewise.
* config/mips/mips.c (mips_expand_vselect_vconcat): Likewise.
* expmed.c (init_expmed_one_mode): Update use of GET_MODE_WIDER_MODE.
Avoid checking for a MODE_INT if we already know the mode is not a
SCALAR_INT_MODE_P.
(extract_high_half): Update use of GET_MODE_WIDER_MODE,
forcing a wider mode to exist.
(expmed_mult_highpart_optab): Likewise.
(expmed_mult_highpart): Likewise.
* expr.c (expand_expr_real_2): Update use of GET_MODE_WIDER_MODE,
using else_void.
* lto-streamer-in.c (lto_input_mode_table): Likewise.
* optabs-query.c (find_widening_optab_handler_and_mode): Likewise.
* stor-layout.c (bit_field_mode_iterator::next_mode): Likewise.
* internal-fn.c (expand_mul_overflow): Update use of
GET_MODE_2XWIDER_MODE.
* omp-low.c (omp_clause_aligned_alignment): Likewise.
* tree-ssa-math-opts.c (convert_mult_to_widen): Update use of
GET_MODE_WIDER_MODE.
(convert_plusminus_to_widen): Likewise.
* tree-switch-conversion.c (array_value_type): Likewise.
* var-tracking.c (emit_note_insn_var_location): Likewise.
* tree-vrp.c (simplify_float_conversion_using_ranges): Likewise.
Return false inside rather than outside the loop if no wider mode
exists
* optabs.c (expand_binop): Update use of GET_MODE_WIDER_MODE
and GET_MODE_2XWIDER_MODE
(can_compare_p): Use else_void.
* gdbhooks.py (OptMachineModePrinter): New class.
(build_pretty_printer): Use it for opt_mode.

gcc/ada/
* gcc-interface/decl.c (validate_size): Update use of
GET_MODE_WIDER_MODE, forcing a wider mode to exist.

Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r251457

[5/77] Small tweak to array_value_type

Store the type mode in a variable so that a later,
more mechanical patch can change its type.

2017-08-30  Richard Sandiford  <richard.sandiford@linaro.org>
    Alan Hayward  <alan.hayward@arm.com>
    David Sherwood  <david.sherwood@arm.com>

gcc/
* tree-switch-conversion.c (array_value_type): Only read TYPE_MODE
once.  Use get_narrowest_mode instead of GET_CLASS_NARROWEST_MODE.

Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r251456

[4/77] Add FOR_EACH iterators for modes

The new iterators are:

- FOR_EACH_MODE_IN_CLASS: iterate over all the modes in a mode class.

- FOR_EACH_MODE_FROM: iterate over all the modes in a class,
  starting at a given mode.

- FOR_EACH_WIDER_MODE: iterate over all the modes in a class,
  starting at the next widest mode after a given mode.

- FOR_EACH_2XWIDER_MODE: same, but considering only modes that
  are two times wider than the previous mode.

- FOR_EACH_MODE_UNTIL: iterate over all the modes in a class until
  a given mode is reached.

- FOR_EACH_MODE: iterate over all the modes in a class between
  two given modes, inclusive of the first but not the second.

These help with the stronger type checking added by later patches,
since every new mode will be in the same class as the previous one.

2017-08-30  Richard Sandiford  <richard.sandiford@linaro.org>
            Alan Hayward  <alan.hayward@arm.com>
            David Sherwood  <david.sherwood@arm.com>

gcc/
* machmode.h (mode_traits): New structure.
(get_narrowest_mode): New function.
(mode_iterator::start): Likewise.
(mode_iterator::iterate_p): Likewise.
(mode_iterator::get_wider): Likewise.
(mode_iterator::get_known_wider): Likewise.
(mode_iterator::get_2xwider): Likewise.
(FOR_EACH_MODE_IN_CLASS): New mode iterator.
(FOR_EACH_MODE): Likewise.
(FOR_EACH_MODE_FROM): Likewise.
(FOR_EACH_MODE_UNTIL): Likewise.
(FOR_EACH_WIDER_MODE): Likewise.
(FOR_EACH_2XWIDER_MODE): Likewise.
* builtins.c (expand_builtin_strlen): Use new mode iterators.
* combine.c (simplify_comparison): Likewise
* config/i386/i386.c (type_natural_mode): Likewise.
* cse.c (cse_insn): Likewise.
* dse.c (find_shift_sequence): Likewise.
* emit-rtl.c (init_derived_machine_modes): Likewise.
(init_emit_once): Likewise.
* explow.c (hard_function_value): Likewise.
* expmed.c (extract_fixed_bit_field_1): Likewise.
(extract_bit_field_1): Likewise.
(expand_divmod): Likewise.
(emit_store_flag_1): Likewise.
* expr.c (init_expr_target): Likewise.
(convert_move): Likewise.
(alignment_for_piecewise_move): Likewise.
(widest_int_mode_for_size): Likewise.
(emit_block_move_via_movmem): Likewise.
(copy_blkmode_to_reg): Likewise.
(set_storage_via_setmem): Likewise.
(compress_float_constant): Likewise.
* omp-low.c (omp_clause_aligned_alignment): Likewise.
* optabs-query.c (get_best_extraction_insn): Likewise.
* optabs.c (expand_binop): Likewise.
(expand_twoval_unop): Likewise.
(expand_twoval_binop): Likewise.
(widen_leading): Likewise.
(widen_bswap): Likewise.
(expand_parity): Likewise.
(expand_unop): Likewise.
(prepare_cmp_insn): Likewise.
(prepare_float_lib_cmp): Likewise.
(expand_float): Likewise.
(expand_fix): Likewise.
(expand_sfix_optab): Likewise.
* postreload.c (move2add_use_add2_insn): Likewise.
* reg-stack.c (reg_to_stack): Likewise.
* reginfo.c (choose_hard_reg_mode): Likewise.
* rtlanal.c (init_num_sign_bit_copies_in_rep): Likewise.
* stor-layout.c (mode_for_size): Likewise.
(smallest_mode_for_size): Likewise.
(mode_for_vector): Likewise.
(finish_bitfield_representative): Likewise.
* tree-ssa-math-opts.c (target_supports_divmod_p): Likewise.
* tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
* tree-vect-stmts.c (vectorizable_conversion): Likewise.
* var-tracking.c (prepare_call_arguments): Likewise.

gcc/ada/
* gcc-interface/misc.c (fp_prec_to_size): Use new mode iterators.
(fp_size_to_prec): Likewise.

gcc/c-family/
* c-common.c (c_common_fixed_point_type_for_size): Use new mode
iterators.
* c-cppbuiltin.c (c_cpp_builtins): Likewise.

Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r251455

[3/77] Allow machine modes to be classes

This patch makes various changes that allow modes like SImode to be
classes rather than enums.

Firstly, it adds inline functions for all mode size properties,
with the macros being simple wrappers around them.  This is necessary
for the __builtin_constant_p trick to continue working effectively when
the mode arguments are slightly more complex (but still foldable at
compile time).

These inline functions are trivial and heavily used.  There's not much
point keeping them out-of-line at -O0: if anything it would make
debugging harder rather than easier, and it would also slow things down.
The patch therefore marks them as "always_inline", if that's available.
Later patches use this approach too.

Using inline functions means that it's no longer possible to pass
an int to GET_MODE_PRECISION etc.  The Fortran and powerpcspe-c.c
parts are needed to avoid instances of that.

The patch continues to use enums for gencondmd.c, so that more
mode comparisons are integer constant expressions when checking
for always-true or always-false conditions.  This is the only
intended use of USE_ENUM_MODES.

The patch also enforces the previous replacement of case statements
by defining modes as:

  #define FOOmode ((void) 0, E_FOOmode)

This adds no overhead but makes sure that new uses of "case FOOmode:"
don't accidentally creep in when FOOmode has no specific class associated
with it.

2017-08-30  Richard Sandiford  <richard.sandiford@linaro.org>
    Alan Hayward  <alan.hayward@arm.com>
    David Sherwood  <david.sherwood@arm.com>

gcc/
* genconditions.c (write_header): Add a "#define USE_ENUM_MODES".
* genmodes.c (emit_insn_modes_h): Define FOOmode to E_FOOmode if
USE_ENUM_MODES is defined and to ((void) 0, E_FOOmode) otherwise.
* machmode.h (mode_size): Move earlier in file.
(mode_precision): Likewise.
(mode_inner): Likewise.
(mode_nunits): Likewise.
(mode_unit_size): Likewise.
(unit_unit_precision): Likewise.
(mode_wider): Likewise.
(mode_2xwider): Likewise.
(machine_mode): New class.
(mode_to_bytes): New function.
(mode_to_bits): Likewise.
(mode_to_precision): Likewise.
(mode_to_inner): Likewise.
(mode_to_unit_size): Likewise.
(mode_to_unit_precision): Likewise.
(mode_to_nunits): Likewise.
(GET_MODE_SIZE): Use mode_to_bytes.
(GET_MODE_BITSIZE): Use mode_to_bits.
(GET_MODE_PRECISION): Use mode_to_precision.
(GET_MODE_INNER): Use mode_to_inner.
(GET_MODE_UNIT_SIZE): Use mode_to_unit_size.
(GET_MODE_UNIT_PRECISION): Use mode_to_unit_precision.
(GET_MODE_NUNITS): Use mode_to_nunits.
* system.h (ALWAYS_INLINE): New macro.
* config/powerpcspe/powerpcspe-c.c
(altivec_resolve_overloaded_builtin): Use machine_mode instead of
int for arg1_mode and arg2_mode.

gcc/fortran/
* trans-types.c (gfc_init_kinds): Use machine_mode instead of int
for "mode".

Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r251454

[2/77] Add an E_ prefix to case statements

All case statements need to be updated to use the prefixed names,
since the unprefixed names will eventually not be integer constant
expressions.  This patch does a mechanical substitution over the whole
codebase.

2017-08-30  Richard Sandiford  <richard.sandiford@linaro.org>
    Alan Hayward  <alan.hayward@arm.com>
    David Sherwood  <david.sherwood@arm.com>

gcc/
* config/aarch64/aarch64-builtins.c (aarch64_simd_builtin_std_type):
Prefix mode names with E_ in case statements.
* config/aarch64/aarch64-elf.h (ASM_OUTPUT_ADDR_DIFF_ELT): Likewise.
* config/aarch64/aarch64.c (aarch64_split_simd_combine): Likewise.
(aarch64_split_simd_move): Likewise.
(aarch64_gen_storewb_pair): Likewise.
(aarch64_gen_loadwb_pair): Likewise.
(aarch64_gen_store_pair): Likewise.
(aarch64_gen_load_pair): Likewise.
(aarch64_get_condition_code_1): Likewise.
(aarch64_constant_pool_reload_icode): Likewise.
(get_rsqrte_type): Likewise.
(get_rsqrts_type): Likewise.
(get_recpe_type): Likewise.
(get_recps_type): Likewise.
(aarch64_gimplify_va_arg_expr): Likewise.
(aarch64_simd_container_mode): Likewise.
(aarch64_emit_load_exclusive): Likewise.
(aarch64_emit_store_exclusive): Likewise.
(aarch64_expand_compare_and_swap): Likewise.
(aarch64_gen_atomic_cas): Likewise.
(aarch64_emit_bic): Likewise.
(aarch64_emit_atomic_swap): Likewise.
(aarch64_emit_atomic_load_op): Likewise.
(aarch64_evpc_trn): Likewise.
(aarch64_evpc_uzp): Likewise.
(aarch64_evpc_zip): Likewise.
(aarch64_evpc_ext): Likewise.
(aarch64_evpc_rev): Likewise.
(aarch64_evpc_dup): Likewise.
(aarch64_gen_ccmp_first): Likewise.
(aarch64_gen_ccmp_next): Likewise.
* config/alpha/alpha.c (alpha_scalar_mode_supported_p): Likewise.
(alpha_emit_xfloating_libcall): Likewise.
(emit_insxl): Likewise.
(alpha_arg_type): Likewise.
* config/arc/arc.c (arc_vector_mode_supported_p): Likewise.
(arc_preferred_simd_mode): Likewise.
(arc_secondary_reload): Likewise.
(get_arc_condition_code): Likewise.
(arc_print_operand): Likewise.
(arc_legitimate_constant_p): Likewise.
* config/arc/arc.h (ASM_OUTPUT_ADDR_DIFF_ELT): Likewise.
* config/arc/arc.md (casesi_load): Likewise.
(casesi_compact_jump): Likewise.
* config/arc/predicates.md (proper_comparison_operator): Likewise.
(cc_use_register): Likewise.
* config/arm/aout.h (ASM_OUTPUT_ADDR_DIFF_ELT): Likewise.
* config/arm/arm-builtins.c (arm_simd_builtin_std_type): Likewise.
(arm_init_iwmmxt_builtins): Likewise.
* config/arm/arm.c (thumb1_size_rtx_costs): Likewise.
(neon_expand_vector_init): Likewise.
(arm_attr_length_move_neon): Likewise.
(maybe_get_arm_condition_code): Likewise.
(arm_emit_vector_const): Likewise.
(arm_preferred_simd_mode): Likewise.
(arm_output_iwmmxt_tinsr): Likewise.
(thumb1_output_casesi): Likewise.
(thumb2_output_casesi): Likewise.
(arm_emit_load_exclusive): Likewise.
(arm_emit_store_exclusive): Likewise.
(arm_expand_compare_and_swap): Likewise.
(arm_evpc_neon_vuzp): Likewise.
(arm_evpc_neon_vzip): Likewise.
(arm_evpc_neon_vrev): Likewise.
(arm_evpc_neon_vtrn): Likewise.
(arm_evpc_neon_vext): Likewise.
(arm_validize_comparison): Likewise.
* config/arm/neon.md (neon_vc<cmp_op><mode>): Likewise.
* config/avr/avr-c.c (avr_resolve_overloaded_builtin): Likewise.
* config/avr/avr.c (avr_rtx_costs_1): Likewise.
* config/c6x/c6x.c (c6x_vector_mode_supported_p): Likewise.
(c6x_preferred_simd_mode): Likewise.
* config/epiphany/epiphany.c (get_epiphany_condition_code): Likewise.
(epiphany_rtx_costs): Likewise.
* config/epiphany/predicates.md (proper_comparison_operator):
Likewise.
* config/frv/frv.c (condexec_memory_operand): Likewise.
(frv_emit_move): Likewise.
(output_move_single): Likewise.
(output_condmove_single): Likewise.
(frv_hard_regno_mode_ok): Likewise.
(frv_matching_accg_mode): Likewise.
* config/h8300/h8300.c (split_adds_subs): Likewise.
(h8300_rtx_costs): Likewise.
(h8300_print_operand): Likewise.
(compute_mov_length): Likewise.
(output_logical_op): Likewise.
(compute_logical_op_length): Likewise.
(compute_logical_op_cc): Likewise.
(h8300_shift_needs_scratch_p): Likewise.
(output_a_shift): Likewise.
(compute_a_shift_length): Likewise.
(compute_a_shift_cc): Likewise.
(expand_a_rotate): Likewise.
(output_a_rotate): Likewise.
* config/i386/i386.c (classify_argument): Likewise.
(function_arg_advance_32): Likewise.
(function_arg_32): Likewise.
(function_arg_64): Likewise.
(function_value_64): Likewise.
(ix86_gimplify_va_arg): Likewise.
(ix86_legitimate_constant_p): Likewise.
(put_condition_code): Likewise.
(split_double_mode): Likewise.
(ix86_avx256_split_vector_move_misalign): Likewise.
(ix86_expand_vector_logical_operator): Likewise.
(ix86_split_idivmod): Likewise.
(ix86_expand_adjust_ufix_to_sfix_si): Likewise.
(ix86_build_const_vector): Likewise.
(ix86_build_signbit_mask): Likewise.
(ix86_match_ccmode): Likewise.
(ix86_cc_modes_compatible): Likewise.
(ix86_expand_branch): Likewise.
(ix86_expand_sse_cmp): Likewise.
(ix86_expand_sse_movcc): Likewise.
(ix86_expand_int_sse_cmp): Likewise.
(ix86_expand_vec_perm_vpermi2): Likewise.
(ix86_expand_vec_perm): Likewise.
(ix86_expand_sse_unpack): Likewise.
(ix86_expand_int_addcc): Likewise.
(ix86_split_to_parts): Likewise.
(ix86_vectorize_builtin_gather): Likewise.
(ix86_vectorize_builtin_scatter): Likewise.
(avx_vpermilp_parallel): Likewise.
(inline_memory_move_cost): Likewise.
(ix86_tieable_integer_mode_p): Likewise.
(x86_maybe_negate_const_int): Likewise.
(ix86_expand_vector_init_duplicate): Likewise.
(ix86_expand_vector_init_one_nonzero): Likewise.
(ix86_expand_vector_init_one_var): Likewise.
(ix86_expand_vector_init_concat): Likewise.
(ix86_expand_vector_init_interleave): Likewise.
(ix86_expand_vector_init_general): Likewise.
(ix86_expand_vector_set): Likewise.
(ix86_expand_vector_extract): Likewise.
(emit_reduc_half): Likewise.
(ix86_emit_i387_round): Likewise.
(ix86_mangle_type): Likewise.
(ix86_expand_round_sse4): Likewise.
(expand_vec_perm_blend): Likewise.
(canonicalize_vector_int_perm): Likewise.
(ix86_expand_vec_one_operand_perm_avx512): Likewise.
(expand_vec_perm_1): Likewise.
(expand_vec_perm_interleave3): Likewise.
(expand_vec_perm_even_odd_pack): Likewise.
(expand_vec_perm_even_odd_1): Likewise.
(expand_vec_perm_broadcast_1): Likewise.
(ix86_vectorize_vec_perm_const_ok): Likewise.
(ix86_expand_vecop_qihi): Likewise.
(ix86_expand_mul_widen_hilo): Likewise.
(ix86_expand_sse2_abs): Likewise.
(ix86_expand_pextr): Likewise.
(ix86_expand_pinsr): Likewise.
(ix86_preferred_simd_mode): Likewise.
(ix86_simd_clone_compute_vecsize_and_simdlen): Likewise.
* config/i386/sse.md (*andnot<mode>3): Likewise.
(<mask_codefor><code><mode>3<mask_name>): Likewise.
(*<code><mode>3): Likewise.
* config/ia64/ia64.c (ia64_expand_vecint_compare): Likewise.
(ia64_expand_atomic_op): Likewise.
(ia64_arg_type): Likewise.
(ia64_mode_to_int): Likewise.
(ia64_scalar_mode_supported_p): Likewise.
(ia64_vector_mode_supported_p): Likewise.
(expand_vec_perm_broadcast): Likewise.
* config/iq2000/iq2000.c (iq2000_move_1word): Likewise.
(iq2000_function_arg_advance): Likewise.
(iq2000_function_arg): Likewise.
* config/m32c/m32c.c (m32c_preferred_reload_class): Likewise.
* config/m68k/m68k.c (output_dbcc_and_branch): Likewise.
(m68k_libcall_value): Likewise.
(m68k_function_value): Likewise.
(sched_attr_op_type): Likewise.
* config/mcore/mcore.c (mcore_output_move): Likewise.
* config/microblaze/microblaze.c (microblaze_function_arg_advance):
Likewise.
(microblaze_function_arg): Likewise.
* config/mips/mips.c (mips16_build_call_stub): Likewise.
(mips_print_operand): Likewise.
(mips_mode_ok_for_mov_fmt_p): Likewise.
(mips_vector_mode_supported_p): Likewise.
(mips_preferred_simd_mode): Likewise.
(mips_expand_vpc_loongson_even_odd): Likewise.
(mips_expand_vec_unpack): Likewise.
(mips_expand_vi_broadcast): Likewise.
(mips_expand_vector_init): Likewise.
(mips_expand_vec_reduc): Likewise.
(mips_expand_msa_cmp): Likewise.
* config/mips/mips.md (casesi_internal_mips16_<mode>): Likewise.
* config/mn10300/mn10300.c (mn10300_print_operand): Likewise.
(cc_flags_for_mode): Likewise.
* config/msp430/msp430.c (msp430_print_operand): Likewise.
* config/nds32/nds32-md-auxiliary.c (nds32_mem_format): Likewise.
(nds32_output_casesi_pc_relative): Likewise.
* config/nds32/nds32.h (ASM_OUTPUT_ADDR_DIFF_ELT): Likewise.
* config/nvptx/nvptx.c (nvptx_ptx_type_from_mode): Likewise.
(nvptx_gen_unpack): Likewise.
(nvptx_gen_pack): Likewise.
(nvptx_gen_shuffle): Likewise.
(nvptx_gen_wcast): Likewise.
(nvptx_preferred_simd_mode): Likewise.
* config/pa/pa.c (pa_secondary_reload): Likewise.
* config/pa/predicates.md (base14_operand): Likewise.
* config/powerpcspe/powerpcspe-c.c
(altivec_resolve_overloaded_builtin): Likewise.
* config/powerpcspe/powerpcspe.c (rs6000_setup_reg_addr_masks):
Likewise.
(rs6000_preferred_simd_mode): Likewise.
(output_vec_const_move): Likewise.
(rs6000_expand_vector_extract): Likewise.
(rs6000_split_vec_extract_var): Likewise.
(reg_offset_addressing_ok_p): Likewise.
(rs6000_legitimate_offset_address_p): Likewise.
(rs6000_legitimize_address): Likewise.
(rs6000_emit_set_const): Likewise.
(rs6000_const_vec): Likewise.
(rs6000_emit_move): Likewise.
(spe_build_register_parallel): Likewise.
(rs6000_darwin64_record_arg_recurse): Likewise.
(swap_selector_for_mode): Likewise.
(spe_init_builtins): Likewise.
(paired_init_builtins): Likewise.
(altivec_init_builtins): Likewise.
(do_load_for_compare): Likewise.
(rs6000_generate_compare): Likewise.
(rs6000_expand_float128_convert): Likewise.
(emit_load_locked): Likewise.
(emit_store_conditional): Likewise.
(rs6000_output_function_epilogue): Likewise.
(rs6000_handle_altivec_attribute): Likewise.
(rs6000_function_value): Likewise.
(emit_fusion_gpr_load): Likewise.
(emit_fusion_p9_load): Likewise.
(emit_fusion_p9_store): Likewise.
* config/powerpcspe/predicates.md (easy_fp_constant): Likewise.
(fusion_gpr_mem_load): Likewise.
(fusion_addis_mem_combo_load): Likewise.
(fusion_addis_mem_combo_store): Likewise.
* config/rs6000/predicates.md (easy_fp_constant): Likewise.
(fusion_gpr_mem_load): Likewise.
(fusion_addis_mem_combo_load): Likewise.
(fusion_addis_mem_combo_store): Likewise.
* config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin):
Likewise.
* config/rs6000/rs6000-string.c (do_load_for_compare): Likewise.
* config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): Likewise.
(rs6000_preferred_simd_mode): Likewise.
(output_vec_const_move): Likewise.
(rs6000_expand_vector_extract): Likewise.
(rs6000_split_vec_extract_var): Likewise.
(reg_offset_addressing_ok_p): Likewise.
(rs6000_legitimate_offset_address_p): Likewise.
(rs6000_legitimize_address): Likewise.
(rs6000_emit_set_const): Likewise.
(rs6000_const_vec): Likewise.
(rs6000_emit_move): Likewise.
(rs6000_darwin64_record_arg_recurse): Likewise.
(swap_selector_for_mode): Likewise.
(paired_init_builtins): Likewise.
(altivec_init_builtins): Likewise.
(rs6000_expand_float128_convert): Likewise.
(emit_load_locked): Likewise.
(emit_store_conditional): Likewise.
(rs6000_output_function_epilogue): Likewise.
(rs6000_handle_altivec_attribute): Likewise.
(rs6000_function_value): Likewise.
(emit_fusion_gpr_load): Likewise.
(emit_fusion_p9_load): Likewise.
(emit_fusion_p9_store): Likewise.
* config/rx/rx.c (rx_gen_move_template): Likewise.
(flags_from_mode): Likewise.
* config/s390/predicates.md (s390_alc_comparison): Likewise.
(s390_slb_comparison): Likewise.
* config/s390/s390.c (s390_handle_vectorbool_attribute): Likewise.
(s390_vector_mode_supported_p): Likewise.
(s390_cc_modes_compatible): Likewise.
(s390_match_ccmode_set): Likewise.
(s390_canonicalize_comparison): Likewise.
(s390_emit_compare_and_swap): Likewise.
(s390_branch_condition_mask): Likewise.
(s390_rtx_costs): Likewise.
(s390_secondary_reload): Likewise.
(__SECONDARY_RELOAD_CASE): Likewise.
(s390_expand_cs): Likewise.
(s390_preferred_simd_mode): Likewise.
* config/s390/vx-builtins.md (vec_packsu_u<mode>): Likewise.
* config/sh/sh.c (sh_print_operand): Likewise.
(dump_table): Likewise.
(sh_secondary_reload): Likewise.
* config/sh/sh.h (ASM_OUTPUT_ADDR_DIFF_ELT): Likewise.
* config/sh/sh.md (casesi_worker_1): Likewise.
(casesi_worker_2): Likewise.
* config/sparc/predicates.md (icc_comparison_operator): Likewise.
(fcc_comparison_operator): Likewise.
* config/sparc/sparc.c (sparc_expand_move): Likewise.
(emit_soft_tfmode_cvt): Likewise.
(sparc_preferred_simd_mode): Likewise.
(output_cbranch): Likewise.
(sparc_print_operand): Likewise.
(sparc_expand_vec_perm_bmask): Likewise.
(vector_init_bshuffle): Likewise.
* config/spu/spu.c (spu_scalar_mode_supported_p): Likewise.
(spu_vector_mode_supported_p): Likewise.
(spu_expand_insv): Likewise.
(spu_emit_branch_or_set): Likewise.
(spu_handle_vector_attribute): Likewise.
(spu_builtin_splats): Likewise.
(spu_builtin_extract): Likewise.
(spu_builtin_promote): Likewise.
(spu_expand_sign_extend): Likewise.
* config/tilegx/tilegx.c (tilegx_scalar_mode_supported_p): Likewise.
(tilegx_simd_int): Likewise.
* config/tilepro/tilepro.c (tilepro_scalar_mode_supported_p): Likewise.
(tilepro_simd_int): Likewise.
* config/v850/v850.c (const_double_split): Likewise.
(v850_print_operand): Likewise.
(ep_memory_offset): Likewise.
* config/vax/vax.c (vax_rtx_costs): Likewise.
(vax_output_int_move): Likewise.
(vax_output_int_add): Likewise.
(vax_output_int_subtract): Likewise.
* config/visium/predicates.md (visium_branch_operator): Likewise.
* config/visium/visium.c (rtx_ok_for_offset_p): Likewise.
(visium_print_operand_address): Likewise.
* config/visium/visium.h (ASM_OUTPUT_ADDR_DIFF_ELT): Likewise.
* config/xtensa/xtensa.c (xtensa_mem_offset): Likewise.
(xtensa_expand_conditional_branch): Likewise.
(xtensa_copy_incoming_a7): Likewise.
(xtensa_output_literal): Likewise.
* dfp.c (decimal_real_maxval): Likewise.
* targhooks.c (default_libgcc_floating_mode_supported_p): Likewise.

gcc/c-family/
* c-cppbuiltin.c (mode_has_fma): Prefix mode names with E_ in
case statements.

gcc/objc/
* objc-encoding.c (encode_gnu_bitfield): Prefix mode names with E_ in
case statements.

libobjc/
* encoding.c (_darwin_rs6000_special_round_type_align): Prefix mode
names with E_ in case statements.

Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r251453

[1/77] Add an E_ prefix to mode names

Later patches will add wrapper types for specific classes
of mode.  E.g. SImode will be a scalar_int_mode, SFmode will be a
scalar_float_mode, etc.  This patch prepares for that change by adding
an E_ prefix to the mode enum values.  It also adds #defines that map
the unprefixed names to the prefixed names; e.g:

  #define QImode E_QImode

Later patches will change this to use things like scalar_int_mode
where appropriate.

The patch continues to use enum values to initialise static data.
This isn't necessary for correctness, but it cuts down on the amount
of load-time initialisation and shouldn't have any downsides.

The patch also changes things like:

  cmp_mode == DImode ? DFmode : DImode

to:

  cmp_mode == DImode ? E_DFmode : E_DImode

This is because DImode and DFmode will eventually be different
classes, so the original ?: wouldn't be well-formed.

2017-08-30  Richard Sandiford  <richard.sandiford@linaro.org>
    Alan Hayward  <alan.hayward@arm.com>
    David Sherwood  <david.sherwood@arm.com>

gcc/
* genmodes.c (mode_size_inline): Add an E_ prefix to mode names.
(mode_nunits_inline): Likewise.
(mode_inner_inline): Likewise.
(mode_unit_size_inline): Likewise.
(mode_unit_precision_inline): Likewise.
(emit_insn_modes_h): Likewise.  Also emit a #define of the
unprefixed name.
(emit_mode_wider): Add an E_ prefix to mode names.
(emit_mode_complex): Likewise.
(emit_mode_inner): Likewise.
(emit_mode_adjustments): Likewise.
(emit_mode_int_n): Likewise.
* config/aarch64/aarch64-builtins.c (v8qi_UP, v4hi_UP, v4hf_UP)
(v2si_UP, v2sf_UP, v1df_UP, di_UP, df_UP, v16qi_UP, v8hi_UP, v8hf_UP)
(v4si_UP, v4sf_UP, v2di_UP, v2df_UP, ti_UP, oi_UP, ci_UP, xi_UP)
(si_UP, sf_UP, hi_UP, hf_UP, qi_UP): Likewise.
(CRC32_BUILTIN, ENTRY): Likewise.
* config/aarch64/aarch64.c (aarch64_push_regs): Likewise.
(aarch64_pop_regs): Likewise.
(aarch64_process_components): Likewise.
* config/alpha/alpha.c (alpha_emit_conditional_move): Likewise.
* config/arm/arm-builtins.c (v8qi_UP, v4hi_UP, v4hf_UP, v2si_UP)
(v2sf_UP, di_UP, v16qi_UP, v8hi_UP, v8hf_UP, v4si_UP, v4sf_UP)
(v2di_UP, ti_UP, ei_UP, oi_UP, hf_UP, si_UP, void_UP): Likewise.
* config/arm/arm.c (arm_init_libfuncs): Likewise.
* config/i386/i386-builtin-types.awk (ix86_builtin_type_vect_mode):
Likewise.
* config/i386/i386-builtin.def (pcmpestr): Likewise.
(pcmpistr): Likewise.
* config/microblaze/microblaze.c (double_memory_operand): Likewise.
* config/mmix/mmix.c (mmix_output_condition): Likewise.
* config/powerpcspe/powerpcspe.c (rs6000_init_hard_regno_mode_ok):
Likewise.
* config/rl78/rl78.c (mduc_regs): Likewise.
* config/rs6000/rs6000.c (rs6000_init_hard_regno_mode_ok): Likewise.
(htm_expand_builtin): Likewise.
* config/sh/sh.h (REGISTER_NATURAL_MODE): Likewise.
* config/sparc/sparc.c (emit_save_or_restore_regs): Likewise.
* config/xtensa/xtensa.c (print_operand): Likewise.
* expmed.h (NUM_MODE_PARTIAL_INT): Likewise.
(NUM_MODE_VECTOR_INT): Likewise.
* genoutput.c (null_operand): Likewise.
(output_operand_data): Likewise.
* genrecog.c (print_parameter_value): Likewise.
* lra.c (debug_operand_data): Likewise.

Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r251452

Split out parts of scompare_loc_descriptor and emit_store_flag

This patch splits some cases out of scompare_loc_descriptor and
emit_store_flag, which helps with the upcoming machmode series.

2017-08-30  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
* dwarf2out.c (scompare_loc_descriptor_wide)
(scompare_loc_descriptor_narrow): New functions, split out from...
(scompare_loc_descriptor): ...here.
* expmed.c (emit_store_flag_int): New function, split out from...
(emit_store_flag): ...here.

From-SVN: r251451

dwarf2out.c (dwarf2out_finish): Remove setting AT_pubnames.

2017-08-30  Richard Biener  <rguenther@suse.de>

* dwarf2out.c (dwarf2out_finish): Remove setting AT_pubnames.
(dwarf2out_early_finish): Move setting of AT_pubnames from
early debug output to early finish.

From-SVN: r251450

msp430.h: Pass -mcode/data-region to the linker and -mdata-region to the assembler.

* gcc/config/msp430/msp430.h: Pass -mcode/data-region to the linker
and -mdata-region to the assembler.

From-SVN: r251449

dwarf2out.c (add_dwarf_attr): Check we don't add duplicate attributes.

2017-08-30  Richard Biener  <rguenther@suse.de>

* dwarf2out.c (add_dwarf_attr): Check we don't add duplicate
attributes.
(gen_subprogram_die): Add DW_AT_object_pointer only early.
(dwarf2out_early_global_decl): Only generate a DIE for the
abstract origin if it doesn't already exist or is a declaration DIE.
(resolve_addr): Do not add the linkage name twice when
generating a stub DIE for the DW_TAG_GNU_call_site target.

* g++.dg/pr78112-2.C: Do not expect duplicate DW_AT_object_pointer.

From-SVN: r251448

[rs6000] int->machine_mode in rs6000-c.c

2017-08-30  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
* config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin):
Use machine_mode rather than int for arg1_mode.

From-SVN: r251447

Daily bump.

From-SVN: r251444

net/internal/socktest: build sys_unix.go on AIX

    Reviewed-on: https://go-review.googlesource.com/59913

From-SVN: r251440

libgo: netinet/icmp6.h require netinet/in.h on AIX

    Reviewed-on: https://go-review.googlesource.com/59912

From-SVN: r251439

PR c++/81236 - ICE with template-id in generic lambda

* semantics.c (finish_id_expression): Remove special dependent case.
Avoid some later pieces when dependent.
(finish_qualified_id_expr): Do normal BASELINK handling in a
template.  Always build a SCOPE_REF for a destructor BIT_NOT_EXPR.
(parsing_default_capturing_generic_lambda_in_template): Remove.
* parser.c (cp_parser_postfix_dot_deref_expression): Always give an
error for types that will never be complete.
* mangle.c (write_expression): Add sanity check.
* tree.c (build_qualified_name): Add sanity check.
(cp_walk_subtrees): Walk into the class context of a BASELINK.
* lambda.c (add_capture): Improve diagnostic for generic lambda
capture failure.
* call.c (build_new_method_call_1): Print the right constructor
name.

From-SVN: r251438

libgo: fix Stat_t on AIX

    Reviewed-on: https://go-review.googlesource.com/59911

From-SVN: r251436

runtime: make go-nosys.c compile with !HAVE_SYSCALL

    Reviewed-on: https://go-review.googlesource.com/59910

From-SVN: r251435

tweak ChangeLog

From-SVN: r251434

Reimplement handling of lambdas in templates.

* cp-tree.h (LAMBDA_FUNCTION_P): Check DECL_DECLARES_FUNCTION_P.
* decl.c (start_preparsed_function): Call start_lambda_scope.
(finish_function): Call finish_lambda_scope.
* init.c (get_nsdmi): Call start/finish_lambda_scope.
* lambda.c (start_lambda_scope): Only ignore VAR_DECL in a function.
* parser.c (cp_parser_function_definition_after_declarator): Don't
call start/finish_lambda_scope.
* pt.c (retrieve_specialization): Ignore lambda functions in
templates.
(find_parameter_packs_r): Ignore capture proxies.  Look into
lambdas.
(check_for_bare_parameter_packs): Allow bare packs in lambdas.
(tsubst_default_argument): Call start/finish_lambda_scope.
(tsubst_function_decl): Handle lambda functions differently.
(tsubst_template_decl): Likewise.
(tsubst_expr) [DECL_EXPR]: Skip closure declarations and capture
proxies.
(tsubst_lambda_expr): Create a new closure rather than instantiate
the one from the template.
(tsubst_copy_and_build): Don't register a specialization of a pack.
(regenerate_decl_from_template): Call start/finish_lambda_scope.
(instantiate_decl): Remove special lambda function handling.
* semantics.c (process_outer_var_ref): Remove special generic lambda
handling.  Don't implicitly capture in a lambda in a template.  Look
for an existing proxy.
* class.c (current_nonlambda_class_type): Use decl_type_context.

From-SVN: r251433

re PR target/82015 (PowerPC should check if 2nd argument to __builtin_unpackv1ti and similar functions is 0 or 1)

[gcc]
2017-08-29  Michael Meissner  <meissner@linux.vnet.ibm.com>

PR target/82015
* config/rs6000/rs6000.c (rs6000_expand_binop_builtin): Insure
that the second argument of the built-in functions to unpack
128-bit scalar types to 64-bit values is 0 or 1.  Change to use a
switch statement instead a lot of if statements.
* config/rs6000/rs6000.md (unpack<mode>, FMOVE128_VSX iterator):
Allow 64-bit values to be in Altivec registers as well as
traditional floating point registers.
(pack<mode>, FMOVE128_VSX iterator): Likewise.

[gcc/testsuite]
2017-08-29  Michael Meissner  <meissner@linux.vnet.ibm.com>

PR target/82015
* gcc.target/powerpc/pr82015.c: New test.

From-SVN: r251432

PR c++/78840 - ICE with const and nested generic lambda

* g++.dg/cpp1y/lambda-generic-const5.C: New.

From-SVN: r251431

Remove unnecessary LAMBDA_EXPR fields.

* cp-tree.h (LAMBDA_EXPR_CLOSURE): Use TREE_TYPE.
(LAMBDA_EXPR_RETURN_TYPE): Remove.
(struct tree_lambda_expr): Remove closure and return_type fields.
* lambda.c (build_lambda_expr): Don't set LAMBDA_EXPR_RETURN_TYPE.
* pt.c (tsubst_copy_and_build): Likewise.
* parser.c (cp_parser_lambda_declarator_opt): Track return type.
(cp_parser_lambda_body): Adjust unspecified return type check.
* ptree.c (cxx_print_lambda_node): Don't print closure or
return type.

From-SVN: r251430

PR c++/80935 - wrong C++17 error with lambda

* decl.c (check_for_uninitialized_const_var): Check
is_instantiation_of_constexpr.
* constexpr.c (ensure_literal_type_for_constexpr_object): Check
is_instantiation_of_constexpr.
(potential_constant_expression_1): Check var_in_maybe_constexpr_fn.

From-SVN: r251429

Various small fixes.

* lambda.c (build_lambda_object): Check for error_mark_node.
* pt.c (make_pack_expansion): Set PACK_EXPANSION_LOCAL_P on the type
pack as well.
(tsubst_decl) [FUNCTION_DECL]: Set DECL_CONTEXT on the parameters.
(tsubst) [TEMPLATE_PARM_INDEX]: Check for error_mark_node.

From-SVN: r251428

PR c++/80767 - unnecessary instantiation of generic lambda

* call.c (convert_like_real): Call build_user_type_conversion_1 if
cand is null.
(add_conv_candidate): Build a ck_user conversion with no candidate.

From-SVN: r251427

Fix lambdas in template default argument of inherited ctor.

* method.c (synthesized_method_base_walk): Replace an inherited
template with its specialization.
(synthesized_method_walk): Make inheriting_ctor a pointer.
(maybe_explain_implicit_delete, explain_implicit_non_constexpr)
(deduce_inheriting_ctor, implicitly_declare_fn): Adjust.

From-SVN: r251426

Adjust C++17 class deduction use of DECL_ABSTRACT_ORIGIN.

* pt.c (build_deduction_guide): Set DECL_ABSTRACT_ORIGIN on the
template, not the function.
(template_guide_p): Adjust.

From-SVN: r251425

Support copying local_specializations.

* cp-tree.h (enum lss_policy): New.
(local_specialization_stack): Add policy parameter to default ctor.
* pt.c (local_specialization_stack): Copy local_specializations if
lss_copy.

From-SVN: r251424

Add immediate potential_constant_expression variants.

* constexpr.c (potential_constant_expression_1): Add "now" parm.
(is_constant_expression, require_constant_expression): New.
(is_static_init_expression, is_nondependent_constant_expression)
(is_nondependent_static_init_expression): Drop "potential".
* except.c (build_must_not_throw_expr): Do type conversion on
value-dependent argument.
* pt.c, semantics.c, typeck2.c: Use variants without "potential".

From-SVN: r251423

Instantiate default arguments/member initializers once.

* init.c (get_nsdmi): Remember NSDMI instantiations.
* parser.c (inject_this_parameter): Be more picky about
current_class_ptr.
* pt.c (tsubst_copy): Simplify 'this' handling.
(tsubst_default_argument): Remember default argument
instantiations.  Take parameter number.
(tsubst_default_arguments): Pass it.
* call.c (convert_default_arg): Likewise.

From-SVN: r251422

Fix default argument conversion failure and SFINAE.

* call.c (build_over_call): Check convert_default_arg result for
error_mark_node.
* parser.c (cp_parser_late_parsing_default_args): Remember
error_mark_node.

From-SVN: r251421

runtime: fix lfstack for 64-bit AIX

    Reviewed-on: https://go-review.googlesource.com/57550

From-SVN: r251420

*.c: Remove excess braces from target selectors.

* gcc.target/i386/*.c: Remove excess braces from target selectors.

From-SVN: r251419

ira-costs: avoid missing base registers in record_address_regs

* ira-costs.c (record_address_regs): Handle both operands of PLUS for
MAX_REGS_PER_ADDRESS == 1.

From-SVN: r251418

20030926-1.c: Add dg-additional-options.

* gcc.target/i386/20030926-1.c: Add dg-additional-options.
* gcc.target/i386/abi-2.c: Ditto.
* gcc.target/i386/interrupt-sibcall-2.c: Ditto.
* gcc.target/i386/pr22076.c: Ditto.
* gcc.target/i386/pr37216.c: Ditto.
* gcc.target/i386/pr39431.c: Ditto.
* gcc.target/i386/pr40906-1.c: Ditto.
* gcc.target/i386/pr40906-2.c: Ditto.
* gcc.target/i386/pr43766.c: Ditto.
* gcc.target/i386/pr46226.c: Ditto.
* gcc.target/i386/pr46470.c: Ditto.
* gcc.target/i386/pr59929.c: Ditto.
* gcc.target/i386/sse-10.c: Ditto.
* gcc.target/i386/vararg-1.c: Ditto.
* gcc.target/i386/vararg-1.c: Ditto.
* gcc.target/i386/asm-6.c: Compile for fpic target only.
* gcc.target/i386/pr44223.c: Ditto.
* gcc.target/i386/bitfield1.c (dg-options): Remove target selector.
* gcc.target/i386/bitfield2.c (dg-options): Ditto.
* gcc.target/i386/pr67480.c (dg-options): Ditto.
* gcc.target/i386/vect-cond-1.c (dg-options): Ditto.
* gcc.target/i386/bittest.c (scan-assembler-times): Ditto.
* gcc.target/i386/darwin-fpmath.c (dg-do): Simplify target selector.
* gcc.target/i386/mvc9.c: Compile for lto target only.
* gcc.target/i386/pr45234.c: Compile for ia32 target only.
* gcc.target/i386/pr49866.c: Compile for lp64 target only.
* gcc.target/i386/pr57091.c: Ditto.
* gcc.target/i386/pr61599-2.c: Ditto.

From-SVN: r251417

i386.opt (flag_fentry): Do not init to -1.

* config/i386/i386.opt (flag_fentry): Do not init to -1.
* config/i386/i386.c (ix86_option_override_internal): Simplify
setting of opts->x_flag_entry.

From-SVN: r251416

re PR tree-optimization/81503 (Wrong code at -O2)

[gcc]

2017-08-29  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
    Jakub Jelinek  <jakub@redhat.com>
    Richard Biener  <rguenther@suse.de>

PR tree-optimization/81503
* gimple-ssa-strength-reduction.c (replace_mult_candidate): Ensure
folded constant fits in the target type; reorder tests for clarity.

[gcc/testsuite]

2017-08-29  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

PR tree-optimization/81503
* gcc.c-torture/execute/pr81503.c: New file.

Co-Authored-By: Jakub Jelinek <jakub@redhat.com>
Co-Authored-By: Richard Biener <rguenther@suse.de>
From-SVN: r251414

Make expansion of balanced binary trees of switches on tree level.

2017-08-29  Martin Liska  <mliska@suse.cz>

* passes.def: Include pass_lower_switch.
* stmt.c (dump_case_nodes): Remove and move to
tree-switch-conversion.
(case_values_threshold): Likewise.
(expand_switch_as_decision_tree_p): Likewise.
(emit_case_decision_tree): Likewise.
(expand_case): Likewise.
(balance_case_nodes): Likewise.
(node_has_low_bound): Likewise.
(node_has_high_bound): Likewise.
(node_is_bounded): Likewise.
(emit_case_nodes): Likewise.
(struct simple_case_node): New struct.
(add_case_node): Remove.
(emit_case_dispatch_table): Use vector instead of case_list.
(reset_out_edges_aux): Remove.
(compute_cases_per_edge): Likewise.
(expand_case): Build list of simple_case_node.
(expand_sjlj_dispatch_table): Use it.
* tree-switch-conversion.c (struct case_node): Moved from
stmt.c and adjusted.
(emit_case_nodes): Likewise.
(node_has_low_bound): Likewise.
(node_has_high_bound): Likewise.
(node_is_bounded): Likewise.
(case_values_threshold): Likewise.
(reset_out_edges_aux): Likewise.
(compute_cases_per_edge): Likewise.
(add_case_node): Likewise.
(dump_case_nodes): Likewise.
(balance_case_nodes): Likewise.
(expand_switch_as_decision_tree_p): Likewise.
(emit_jump): Likewise.
(emit_case_decision_tree): Likewise.
(try_switch_expansion): Likewise.
(do_jump_if_equal): Likewise.
(emit_cmp_and_jump_insns): Likewise.
(fix_phi_operands_for_edge): New function.
(record_phi_operand_mapping): Likewise.
(class pass_lower_switch): New pass.
(pass_lower_switch::execute): New function.
(make_pass_lower_switch): Likewise.
(conditional_probability):
* timevar.def: Add TV_TREE_SWITCH_LOWERING.
* tree-pass.h: Add make_pass_lower_switch.
2017-08-29  Martin Liska  <mliska@suse.cz>

* gcc.dg/tree-prof/update-loopch.c: Scan patterns in
switchlower.
* gcc.dg/tree-ssa/vrp104.c: Likewise.

From-SVN: r251412

re PR target/80993 ([msp430] __attribute__((interrupt)) should imply __attribute__((used)))

PR target/80993
* gcc/config/msp430/msp430.c (msp430_attr): Mark interrupt
handlers as used.

From-SVN: r251410

dwarf2out.c (add_dwarf_attr): When checking is enabled verify we do not add a DW_AT_inline attribute twice.

2017-08-29  Richard Biener  <rguenther@suse.de>

* dwarf2out.c (add_dwarf_attr): When checking is enabled verify
we do not add a DW_AT_inline attribute twice.
(gen_subprogram_die): Remove code setting DW_AT_inline on
DECL_ABSTRACT_P nodes.

From-SVN: r251409

Fix --help=target (Ada) (PR other/39851)

2017-08-29  Martin Liska  <mliska@suse.cz>

PR other/39851
* gcc-interface/trans.c (Pragma_to_gnu): Set argument to NULL.

From-SVN: r251406

Set the call nothrow flag more often

This patch sets the nothrow flag for various calls to internal functions
that are not inherently NOTHROW (and so can't be declared that way in
internal-fn.def) but that are used in contexts that can guarantee
NOTHROWness.

2017-08-29  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
* gimplify.c (gimplify_call_expr): Copy the nothrow flag to
calls to internal functions.
(gimplify_modify_expr): Likewise.
* tree-call-cdce.c (use_internal_fn): Likewise.
* tree-ssa-math-opts.c (pass_cse_reciprocals::execute): Likewise.
(convert_to_divmod): Set the nothrow flag.
* tree-if-conv.c (predicate_mem_writes):  Likewise.
* tree-vect-stmts.c (vectorizable_mask_load_store): Likewise.
(vectorizable_call): Likewise.
(vectorizable_store): Likewise.
(vectorizable_load): Likewise.
* tree-vect-patterns.c (vect_recog_pow_pattern): Likewise.
(vect_recog_mask_conversion_pattern): Likewise.

From-SVN: r251401

Fix --help=target (PR other/39851).

2017-08-29  Martin Liska  <mliska@suse.cz>

PR other/39851
* gcc.c (driver_handle_option): Add new argument.
* opts-common.c (handle_option): Pass
target_option_override_hook.
* opts-global.c (lang_handle_option): Add new option.
(set_default_handlers):  Add new argument.
(decode_options): Likewise.
* opts.c (target_handle_option): Likewise.
(common_handle_option): Call target_option_override_hook.
* opts.h (struct cl_option_handler_func): Add hook for
target option override.
(struct cl_option_handlers): Likewise.
(set_default_handlers): Add new argument.
(decode_options): Likewise.
(common_handle_option): Likewise.
(target_handle_option): Likewise.
* toplev.c (toplev::main): Pass targetm.target_option.override
hook.
2017-08-29  Martin Liska  <mliska@suse.cz>

PR other/39851
* c-common.c (parse_optimize_options): Add argument to function
call.
* c-pragma.c (handle_pragma_diagnostic): Likewise.

From-SVN: r251400

tree-vect-slp.c (vect_bb_slp_scalar_cost): Properly confine life to the active subtree.

2017-08-29  Richard Biener  <rguenther@suse.de>
Dominik Infuehr <dominik.infuehr@theobroma-systems.com>

* tree-vect-slp.c (vect_bb_slp_scalar_cost): Properly confine
life to the active subtree.

* gcc.dg/vect/costmodel/x86_64/costmodel-vect-slp.c: New testcase.

Co-Authored-By: Dominik Infuehr <dominik.infuehr@theobroma-systems.com>
From-SVN: r251398

tree-ssa-dom.c (edge_info::record_simple_equiv): Call derive_equivalences.

* tree-ssa-dom.c (edge_info::record_simple_equiv): Call
derive_equivalences.
(derive_equivalences_from_bit_ior, record_temporary_equivalences):
Code moved into....
(edge_info::derive_equivalences): New private member function

* gcc.dg/torture/pr57214.c: Fix type of loop counter.
* gcc.dg/tree-ssa/ssa-sink-16.c: Disable DOM.
* gcc.dg/tree-ssa/ssa-dom-thread-11.c: New test.
* gcc.dg/tree-ssa/ssa-dom-thread-12.c: New test.
* gcc.dg/tree-ssa/ssa-dom-thread-13.c: New test.
* gcc.dg/tree-ssa/ssa-dom-thread-14.c: New test.
* gcc.dg/tree-ssa/ssa-dom-thread-15.c: New test.
* gcc.dg/tree-ssa/ssa-dom-thread-16.c: New test.
* gcc.dg/tree-ssa/ssa-dom-thread-17.c: New test.

From-SVN: r251397

tree-ssa-dom.c (class edge_info): Changed from a struct to a class.

* tree-ssa-dom.c (class edge_info): Changed from a struct
to a class.  Add ctor/dtor, methods and data members.
(edge_info::edge_info): Renamed from allocate_edge_info.
Initialize additional members.
(edge_info::~edge_info): New.
(free_dom_edge_info): Delete the edge info.
(record_edge_info): Use new class & associated member functions.
Tighten forms for testing for edge equivalences.
(record_temporary_equivalences): Iterate over the simple
equivalences rather than assuming there's only one per edge.
(cprop_into_successor_phis): Iterate over the simple
equivalences rather than assuming there's only one per edge.
(optimize_stmt): Use operand_equal_p rather than pointer
equality for mini-DSE code.

From-SVN: r251396

Daily bump.

From-SVN: r251395

re PR fortran/81770 (Bogus warning: Pointer in pointer assignment might outlive the pointer target)

2017-08-28  Janus Weil  <janus@gcc.gnu.org>

PR fortran/81770
* expr.c (gfc_check_pointer_assign): Improve the check whether pointer
may outlive pointer target.

2017-08-28  Janus Weil  <janus@gcc.gnu.org>

PR fortran/81770
* gfortran.dg/warn_target_lifetime_3.f90: Fix a typo.
* gfortran.dg/warn_target_lifetime_4.f90: New testcase.

From-SVN: r251390

dec_init_1.f90 (dummy): Use ichar in print statement.

2017-08-25  Steve Ellcey  <sellcey@cavium.com>

* gfortran.dg/dec_init_1.f90 (dummy): Use ichar in print statement.
* gfortran.dg/dec_init_2.f90 (dummy): Ditto.
* gfortran.dg/dec_init_3.f90 (dummy): Ditto.

From-SVN: r251389

cp-tree.h (lang_type): Replace sorted_fields vector with bindings map.

* cp-tree.h (lang_type): Replace sorted_fields vector with
bindings map.
(CLASSTYPE_SORTED_FIELDS): Delete.
(CLASSTYPE_BINDINGS): New.
* decl.c (finish_enum_value_list): Swap args of
insert_late_enum_def_bindings.
* name-lookup.c (lookup_field_1): Replace binary search of sorted
fields with map->get.
(sorted_fields_type_new, count_fields,
add_fields_to_record_type, add_enum_fields_to_record_type): Delete.
(add_class_member, add_class_members): New.
(set_class_bindings): Create map and insert.
(insert_late_enum_def_binding): Swap parms.  Use add_clasS_member.
* ptree.c (cxx_print_type): Delete sorted fields printing.

From-SVN: r251388

cp-tree.h (insert_late_enum_def_into_classtype_sorted_fields): Delete.

* cp-tree.h (insert_late_enum_def_into_classtype_sorted_fields):
Delete.
* name-lookup.h (set_class_bindings,
insert_late_enum_def_bindings): Declare.
* decl.c (finish_enum_value_list): Adjust for
insert_late_enum_def_bindings name change.
* class.c (finish_struct_1): Call set_class_bindings.
(count_fields, add_fields_to_record_type,
add_enum_fields_to_record_type, sorted_fields_type_new,
insert_into_classtype_sorted_fields,
insert_late_enum_def_into_classtype_sorted_fields): Move to ...
* name-lookup.h (count_fields, add_fields_to_record_type,
add_enum_fields_to_record_type, sorted_fields_type_new,
set_class_bindings, insert_late_enum_def_bindings): ... here.

From-SVN: r251387

gcc.c (execute): Fold SIGPIPE handling into switch statement.

* gcc.c (execute): Fold SIGPIPE handling into switch
statement.  Adjust internal error message.

From-SVN: r251385

re PR lto/81968 (early lto debug objects make Solaris ld SEGV)

2017-08-28  Richard Biener  <rguenther@suse.de>

PR lto/81968
* simple-object-elf.c (simple_object_elf_copy_lto_debug_section):
Adjust field with for sh_type write, set SHF_EXCLUDE only for
removed sections.

From-SVN: r251384

re PR debug/81993 (-gsplit-dwarf removes some symbols, causing some undefined references)

2017-08-28  Richard Biener  <rguenther@suse.de>

PR debug/81993
* dwarf2out.c (gen_remaining_tmpl_value_param_die_attributes):
Do nothing for removed DIEs.

From-SVN: r251382

re PR tree-optimization/81977 (Issue with inline memcpy with optimizations enabled)

2017-08-28  Richard Biener  <rguenther@suse.de>

PR tree-optimization/81977
* tree-ssa-sccvn.c (vn_reference_lookup_3): Fix look through
memcpy.

* g++.dg/torture/pr81977.C: New testcase.

From-SVN: r251378

optabs: ensure mem_thread_fence is a compiler barrier

PR target/80640
* doc/md.texi (mem_thread_fence): Remove mention of mode.  Rewrite.
* optabs.c (expand_mem_thread_fence): Emit a compiler barrier when
using targetm.gen_mem_thread_fence.
testsuite/
* gcc.dg/atomic/pr80640.c: New testcase.

From-SVN: r251377

re PR fortran/78387 (OpenMP segfault/stack size exceeded writing to internal file)

2017-08-27  Jerry DeLisle  <jvdelisle@gcc.gnu.org>

PR libgfortran/78387
* io/list_read.c (nml_read_obj): Remove use of stash.
* io/transfer.c (st_read_done, st_write_done): Likewise.
* io/unit.c (stash_internal_unit): Delete function.
(get_unit): Remove use of stash.
(init_units): Likewise.
(close_units): Likewise.
* io/write.c (nml_write_obj): Likewise:

From-SVN: r251374

Daily bump.

From-SVN: r251373

re PR target/81995 (gcc/reg-stack.c:2073:1: error: unrecognizable insn:)

PR target/81995
* config/i386/i386.md (*<btsc><mode>): Change operand 2
predicate to register_operand.  Reorder operands.
(*btr<mode>): Ditto.
(*<btsc><mode>_mask): Change operand 3 predicate to register_operand.
(*btr<mode>_mask): Ditto.

testsuite/ChangeLog:

PR target/81995
* gcc.target/i386/pr46091-4.c: Add -mregparm=2 for 32bit targets.
* gcc.target/i386/pr46091-4a.c: Ditto.

From-SVN: r251369

re PR fortran/81974 (ICE verify_gimple failed  type mismatch in binary expression)

2017-08-27  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/81974
* frontend-passes (inline_matumul_assign):  Explicity
set typespec for call to CONJG.

2017-08-27  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/81974
* gfortran.dg/inline_matmul_19.f90:  New test.

From-SVN: r251368

Daily bump.

From-SVN: r251367

Daily bump.

From-SVN: r251362

Part 3/3 for contributing PPC64LE support for X86 SSE instrisics.
This patch includes testsuite/gcc.target tests for the intrinsics
in xmmintrin.h.  For these tests I added -Wno-psabi to dg-options
to suppress warnings associated with the vector ABI change in GCC5.

From-SVN: r251357

Part 2/3 for contributing PPC64LE support for X86 SSE instrisics.

Part 2/3 for contributing PPC64LE support for X86 SSE
instrisics. This patch includes the new (for PPC) xmmintrin.h,
changes x86intrin.h to include xmmintrin.h and associated
config.gcc changes.

From-SVN: r251356

re PR target/81504 (gcc-7 regression: vec_st  in loop misoptimized)

2017-08-25  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

PR target/81504
* config/rs6000/rs6000-p8swap.c (find_alignment_op): Add reference
parameter and_insn and return it.
(recombine_lvx_pattern): Insert a copy to ensure availability of
the base register of the copied masking operation at the point of
the instruction replacement.
(recombine_stvx_pattern): Likewise.

From-SVN: r251355

rs6000.opt (-mpower9-dform-scalar): Delete undocumented switches.

[gcc]
2017-08-24  Michael Meissner  <meissner@linux.vnet.ibm.com>

* config/rs6000/rs6000.opt (-mpower9-dform-scalar): Delete
undocumented switches.
(-mpower9-dform-vector): Likewise.
(-mpower9-dform): Likewise.
* config/rs6000/rs6000-c.c (rs6000_target_modify_macros): Update
comments to delete references to -mpower9-dform* switches.
* config/rs6000/predicates.md (vsx_quad_dform_memory_operand):
Delete reference to -mpower9-dform* switches, test for
-mpower9-vector instead.
* config/rs6000/rs6000-cpus.def (ISA_3_0_MASKS_SERVER): Likewise.
(OTHER_P9_VECTOR_MASKS): Likewise.
(POWERPC_MASKS): Likewise.
* config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): Change
tests against -mpower9-dform* to -mpower9-vector.  Delete code
that checked for -mpower9-dform* consistancy with other options.
Add test for -mpower9-misc to enable other power9 switches.
(rs6000_init_hard_regno_mode_ok): Likewise.
(rs6000_option_override_internal): Likewise.
(rs6000_emit_prologue): Likewise.
(rs6000_emit_epilogue): Likewise.
(rs6000_opt_masks): Delete -mpower9-dform-{scalar,vector}.
(rs6000_disable_incompatiable_switches): Delete -mpower9-dform.
(emit_fusion_p9_load): Change tests for -mpower9-dform-scalar
-mpower9-vector.
(emit_fusion_p9_store): Likewise.
* config/rs6000/rs6000.h (TARGET_P9_DFORM_SCALAR): Delete
resetting these macros if the assembler does not support ISA 3.0
instructions.
(TARGET_P9_DFORM_VECTOR): Likewise.
* config/rs6000/rs6000.md (peepholes to optimize altivec memory):
Change to use -mpower9-vector instead of -mpower9-dform-scalar.

[gcc/testsuite]
2017-08-24  Michael Meissner  <meissner@linux.vnet.ibm.com>

* gcc.target/powerpc/dform-1.c: Delete -mpower9-dform*
options.
* gcc.target/powerpc/dform-2.c: Likewise.
* gcc.target/powerpc/dform-3.c: Likewise.
* gcc.target/powerpc/pr71656-1.c: Likewise.
* gcc.target/powerpc/pr71656-2.c: Likewise.
* gcc.target/powerpc/pr80103-1.c: Likewise.
* gcc.target/powerpc/pr80098-1.c: Likewise.

From-SVN: r251352

re PR target/81921 (Fails to always-inline intrinsics with -flto)

Retabify 2017-08-24  Richard Biener  <rguenther@suse.de>
PR target/81921

From-SVN: r251351

class.c (method_name_cmp, [...]): Method names can never be NULL.

* class.c (method_name_cmp, resort_method_name_cmp): Method names
can never be NULL.

From-SVN: r251350

PR81747, ICE in operator[]

PR rtl-optimization/81747
* cse.c (cse_extended_basic_block): Don't attempt to record
equivalences for degenerate conditional branches that branch
to their fall-through.

From-SVN: r251349

Conversion operators have a special name

Conversion operators have a special name
* cp-tree.h (CPTI_CONV_OP_MARKER, CPTI_CONV_OP_IDENTIFIER): New.
(conv_op_marker, conv_op_identifier): New.
(CLASSTYPE_FIRST_CONVERSION_SLOT): Delete.
* decl.c (initialize_predefined_identifiers): Add
conv_op_identifier.
(cxx_init_decl_processing): Create conv_op_marker.
* decl2.c (check_classfn): Lookup conv-ops by name.
* class.c (add_method): Use conv_op_identifier & conv_op_marker.
(resort_type_method_vec): Don't skip conv-ops.
(finish_struct_methods, warn_hidden): Likewise.
* name-lookup.h (lookup_all_conversions): Delete.
* name-lookup.c (lookup_conversion_operator): Replace with ...
(extract_conversion_operator): ... this.
(lookup_fnfields_slot_nolazy): Find conv-ops by name.
(lookup_all_conversions): Delete.
* pt.c (check_explicit_specialization): Find conv-ops by name.
* search.c (lookup_conversions_r): Likewise.

From-SVN: r251348

PR middle-end/81908 - FAIL: gfortran.dg/alloc_comp_auto_array_2.f90 -O3 -g -m32

gcc/ChangeLog:

PR middle-end/81908
* gimple-fold.c (size_must_be_zero_p): New function.
(gimple_fold_builtin_memory_op): Call it.

gcc/testsuite/ChangeLog:

PR middle-end/81908
* gcc.dg/tree-ssa/builtins-folding-gimple-2.c: New test.
* gcc.dg/tree-ssa/builtins-folding-gimple-3.c: New test.
* gcc.dg/tree-ssa/pr81908.c: New test.

From-SVN: r251347

Daily bump.

From-SVN: r251346

stl_list.h [!_GLIBCXX_INLINE_VERSION](_List_base<>::_S_distance): Remove.

2017-08-24  François Dumont  <fdumont@gcc.gnu.org>

* include/bits/stl_list.h
[!_GLIBCXX_INLINE_VERSION](_List_base<>::_S_distance): Remove.
(_List_impl(_Node_alloc_type&&, _List_impl&&)): New.
(_List_base(_Node_alloc_type&&, _List_base&&)): New, use latter.
[!_GLIBCXX_INLINE_VERSION](_List_base(_Node_alloc_type&&,_List_base&&)):
Remove.
(_List_base(_Node_alloc_type&&)): New.
[!_GLIBCXX_INLINE_VERSION](_List_base<>::_M_distance): Remove.
[!_GLIBCXX_INLINE_VERSION](_List_base<>::_M_node_count): Remove.
(list<>::_M_node_count): New.
(list<>::size()): Adapt to call latter.
(list<>::_S_distance(const_iterator, const_iterator)): New.
(list<>::splice(iterator, list&, const_iterator, const_iterator)):
Adapt to call latter.
(list(list&&, const allocator_type&, true_type)): New.
(list(list&&, const allocator_type&, false_type)): New.
(list(list&&, const allocator_type&)): Adapt to call latters.

From-SVN: r251342

rs6000.opt (-mpower9-dform-scalar): Delete undocumented switches.

[gcc]
2017-08-24  Michael Meissner  <meissner@linux.vnet.ibm.com>

* config/rs6000/rs6000.opt (-mpower9-dform-scalar): Delete
undocumented switches.
(-mpower9-dform-vector): Likewise.
(-mpower9-dform): Likewise.
* config/rs6000/rs6000-c.c (rs6000_target_modify_macros): Update
comments to delete references to -mpower9-dform* switches.
* config/rs6000/predicates.md (vsx_quad_dform_memory_operand):
Delete reference to -mpower9-dform* switches, test for
-mpower9-vector instead.
* config/rs6000/rs6000-cpus.def (ISA_3_0_MASKS_SERVER): Likewise.
(OTHER_P9_VECTOR_MASKS): Likewise.
(POWERPC_MASKS): Likewise.
* config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): Change
tests against -mpower9-dform* to -mpower9-vector.  Delete code
that checked for -mpower9-dform* consistancy with other options.
Add test for -mpower9-misc to enable other power9 switches.
(rs6000_init_hard_regno_mode_ok): Likewise.
(rs6000_option_override_internal): Likewise.
(rs6000_emit_prologue): Likewise.
(rs6000_emit_epilogue): Likewise.
(rs6000_opt_masks): Delete -mpower9-dform-{scalar,vector}.
(rs6000_disable_incompatiable_switches): Delete -mpower9-dform.
(emit_fusion_p9_load): Change tests for -mpower9-dform-scalar
-mpower9-vector.
(emit_fusion_p9_store): Likewise.
* config/rs6000/rs6000.h (TARGET_P9_DFORM_SCALAR): Delete
resetting these macros if the assembler does not support ISA 3.0
instructions.
(TARGET_P9_DFORM_VECTOR): Likewise.
* config/rs6000/rs6000.md (peepholes to optimize altivec memory):
Change to use -mpower9-vector instead of -mpower9-dform-scalar.

[gcc/testsuite]
2017-08-24  Michael Meissner  <meissner@linux.vnet.ibm.com>

* gcc.target/powerpc/dform-1.c: Delete -mpower9-dform*
options.
* gcc.target/powerpc/dform-2.c: Likewise.
* gcc.target/powerpc/dform-3.c: Likewise.
* gcc.target/powerpc/pr71656-1.c: Likewise.
* gcc.target/powerpc/pr71656-2.c: Likewise.
* gcc.target/powerpc/pr80103-1.c: Likewise.
* gcc.target/powerpc/pr80098-1.c: Likewise.

From-SVN: r251341

Conversion operators kept on single overload set

Conversion operators kept on single overload set
* class.c (add_method): Keep all conv-ops on one slot.
* name-lookup.c (lookup_conversion_operator): Pull the desired
conv op out of overload set.
* search.c (lookup_conversions_r): Lose template/non-template
distinction.
(lookup_conversions): Likewise.

From-SVN: r251340

Adds the powerpc equivalent mm_malloc.h header which is needed by xmmintrin.h...

Adds the powerpc equivalent mm_malloc.h header which is needed by
xmmintrin.h and cleans up some noisy warnings from the previous MMX
commit.

From-SVN: r251339

target-supports.exp: Group together target-dependant checks.

* lib/target-supports.exp: Group together target-dependant checks.

From-SVN: r251338

re PR tree-optimization/81913 (wrong code at -O1)

PR tree-optimization/81913
* tree-ssa-loop-niter.c (number_of_iterations_cond): Skip niter
analysis when either IVs in condition can wrap.

gcc/testsuite
* gcc.c-torture/execute/pr81913.c: New test.
* gcc.dg/tree-ssa/loop-niter-1.c: New test.
* gcc.dg/tree-ssa/loop-niter-2.c: New test.

From-SVN: r251337

C: fix logic within c_expr::get_location

In r251239 I added a c_expr::get_location method for use by
c_parser_expr_list for building the vec<location_t> for
an expression list, rather than using the location of the first token.

When determining whether to use the location within the tree node,
or fall back to the range in the c_expr, I used EXPR_CAN_HAVE_LOCATION,
rather than EXPR_HAS_LOCATION.  This meant that any tree nodes of kinds
that *can* have a location but which erroneously had
   EXPR_LOCATION (value) == UNKNOWN_LOCATION
had that value added to the vec<location_t>, leading to missing
location information when reporting on the issue
(seen with gcc.dg/Wtraditional-conversion-2.c for m68k).

This patch addresses this in two ways:

(a) it fixes the specific issue in this failing test case, by
    setting up the location properly on the EXCESS_PRECISION_EXPR.

(b) updating c_expr::get_location by only using the EXPR_LOCATION
    if it's sane.  Arguably this is papering over bugs, but they are
    pre-existing ones exposed by r251239, and I'd rather have this
    fix in place than play whack-a-mole on any other such "missing
    location" bugs that are lurking in the codebase.

gcc/c/ChangeLog:
* c-tree.h (c_expr::get_location) Use EXPR_HAS_LOCATION rather
than CAN_HAVE_LOCATION_P when determining whether to use the
location_t value within "value".

gcc/c-family/ChangeLog:
* c-lex.c (interpret_float): Use token location
when building an EXCESS_PRECISION_EXPR.

From-SVN: r251335

dwarf2out.c (MAX_ARTIFICIAL_LABEL_BYTES): Increase to 40.

* dwarf2out.c (MAX_ARTIFICIAL_LABEL_BYTES): Increase to 40.
* dwarf2cfi.c (MAX_ARTIFICIAL_LABEL_BYTES): Remove.

From-SVN: r251334

re PR target/81921 (Fails to always-inline intrinsics with -flto)

2017-08-23  Richard Biener  <rguenther@suse.de>

        PR target/81921
        * targhooks.c (default_target_can_inline_p): Properly
        use target_option_default_node when no DECL_FUNCTION_SPECIFIC_TARGET
        is present and always compare.
        * config/i386/i386.c (ix86_valid_target_attribute_tree): Do not
imply -mfpmath=sse from TARGET_SSE_P.
        (ix86_can_inline_p): Properly use target_option_default_node when
        no DECL_FUNCTION_SPECIFIC_TARGET is present and always compare.

* gcc.target/i386/pr81921.c: New testcase.

From-SVN: r251333

re PR debug/81936 (ICE in dwarf2out_die_ref_for_decl, at dwarf2out.c:5543)

2017-08-24  Richard Biener  <rguenther@suse.de>

PR debug/81936
* dwarf2out.c (output_die): Handle flag_generate_offload like
flag_generate_lto.
(output_comp_unit): Likewise.
(gen_array_type_die): Likewise.
(dwarf2out_early_finish): Likewise.
(note_variable_value_in_expr): Likewise.
(dwarf2out_finish): Likewise.  Adjust assert.
* cgraphunit.c (symbol_table::compile): Move setting of
flag_generate_offload earlier ...
(symbol_table::finalize_compilation_unit): ... here, before
early debug finalization.

From-SVN: r251332

i386.c: Include symbol-summary.h, ipa-prop.h and ipa-fnsummary.h.

2017-08-24  Richard Biener  <rguenther@suse.de>

* config/i386/i386.c: Include symbol-summary.h, ipa-prop.h
and ipa-fnsummary.h.
(ix86_can_inline_p): When ix86_fpmath flags do not match
check whether the callee uses FP math at all.

From-SVN: r251330

re PR middle-end/81931 (r251260 breaks many tests on powerpc64)

PR middle-end/81931
* tree-ssanames.c (get_nonzero_bits): Use element_precision
instead of TYPE_PRECISION.

From-SVN: r251328

Make more use of subreg_offset_from_lsb

This patch makes use of the subreg_offset_from_lsb function
added by an earlier patch in the SVE series.

2017-08-24  Richard Sandiford  <richard.sandiford@linaro.org>
    Alan Hayward  <alan.hayward@arm.com>
    David Sherwood  <david.sherwood@arm.com>

gcc/
* combine.c (make_extraction): Use subreg_offset_from_lsb.

Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r251326

Daily bump.

From-SVN: r251325

PR target/80969 - Fix ICE with -mabi=ms -mavx512f, reduce wasted space when realigning stack.

2017-08-23  Daniel Santos  <daniel.santos@pobox.com>

* config/i386/i386.h (ix86_frame::stack_realign_allocate_offset):
Remove field.
(ix86_frame::stack_realign_allocate): New field.
(struct machine_frame_state): Modify comments.
(machine_frame_state::sp_realigned_fp_end): New field.
* config/i386/i386.c (ix86_compute_frame_layout): Rework stack frame
layout calculation.
(sp_valid_at): Add assertion to assure no attempt to access invalid
offset of a realigned stack.
(fp_valid_at): Likewise.
(choose_baseaddr): Modify comments.
(ix86_emit_outlined_ms2sysv_save): Adjust to changes in
ix86_expand_prologue.
(ix86_expand_prologue): Modify stack realignment and allocation.
(ix86_expand_epilogue): Modify comments.
* doc/sourcebuild.texi: Add documentation for target selectors avx2,
avx2_runtime, avx512f, and avx512f_runtime.

2017-08-23  Daniel Santos  <daniel.santos@pobox.com>

* gcc.target/i386/pr80969-1.c: New testcase.
* gcc.target/i386/pr80969-2a.c: Likewise.
* gcc.target/i386/pr80969-2.c: Likewise.
* gcc.target/i386/pr80969-3.c: Likewise.
* gcc.target/i386/pr80969-4a.c: Likewise.
* gcc.target/i386/pr80969-4b.c: Likewise.
* gcc.target/i386/pr80969-4.c: Likewise.
* gcc.target/i386/pr80969-4.h: New header common to pr80969-4*.c
* lib/target-supports.exp (check_avx512_os_support_available,
check_avx512f_hw_available, check_effective_target_avx512f_runtime):
New proceedures for target avx512f and avx512f_runtime selectors.
(check_avx2_hw_available): Fix breakage due NULL being undefined.

From-SVN: r251321

cp-tree.h (lookup_field_1, [...]): Move declatations to ...

* cp-tree.h (lookup_field_1, lookup_fnfields_slot,
lookup_fnfields_slot_nolazy, lookup_all_conversions): Move
declatations to ...
* name-lookup.h (lookup_field_1, lookup_fnfields_slot,
lookup_fnfields_slot_nolazy, lookup_all_conversions): ... here.
* search.c (lookup_conversion_operator,
lookup_fnfields_slot_nolazy, lookup_field_1, lookup_fnfields_slot,
lookup_all_conversions): Move to ...
* name-lookup.c (lookup_conversion_operator,
lookup_fnfields_slot_nolazy, lookup_field_1, lookup_fnfields_slot,
lookup_all_conversions): ... here.

From-SVN: r251319

semantics.c (finish_member_declaration): Move USING_DECL check earlier.

* semantics.c (finish_member_declaration): Move USING_DECL check
earlier.  Always set C++ linkage.  Commonize TYPE_FIELD and
template decl list insertion.

From-SVN: r251317