Re: [PATCH v3] c++: parser - Support for target address spaces in C++

[Jason Merrill <jason@redhat.com>]

On 10/18/22 13:01, Paul Iannetta wrote:
> Thank you very much for the detailed review.
> 
> On Tue, Oct 18, 2022 at 10:24:23AM -0400, Jason Merrill wrote:
>> On 10/18/22 03:37, Paul Iannetta wrote:
>>> On Fri, Oct 14, 2022 at 11:19:50AM -0400, Jason Merrill wrote:
>>>> On 10/13/22 17:57, Paul Iannetta wrote:
>>>>> On Thu, Oct 13, 2022 at 03:41:16PM -0400, Jason Merrill wrote:
>>>>>> On 10/13/22 12:02, Paul Iannetta wrote:
>>>>>>> On Thu, Oct 13, 2022 at 11:47:42AM -0400, Jason Merrill wrote:
>>>>>>>> On 10/13/22 11:23, Paul Iannetta wrote:
>>>>>>>>> On Thu, Oct 13, 2022 at 11:02:24AM -0400, Jason Merrill wrote:
>>>>>>>>>> On 10/12/22 20:52, Paul Iannetta wrote:
>>>>>>>>>>> There are quite a few things I would like to clarify concerning some
>>>>>>>>>>> implementation details.
>>>>>>>>>>>         - A variable with automatic storage (which is neither a pointer nor
>>>>>>>>>>>           a reference) cannot be qualified with an address space.  I detect
>>>>>>>>>>>           this by the combination of `sc_none' and `! toplevel_bindings_p ()',
>>>>>>>>>>>           but I've also seen the use of `at_function_scope' at other places.
>>>>>>>>>>>           And I'm unsure which one is appropriate here.
>>>>>>>>>>>           This detection happens at the very end of grokdeclarator because I
>>>>>>>>>>>           need to know that the type is a pointer, which is not know until
>>>>>>>>>>>           very late in the function.
>>>>>>>>>>
>>>>>>>>>> At that point you have the decl, and you can ask directly what its storage
>>>>>>>>>> duration is, perhaps using decl_storage_duration.
>>>>>>>>>>
>>>>>>>>>> But why do you need to know whether the type is a pointer?  The attribute
>>>>>>>>>> applies to the target type of the pointer, not the pointer type.  I think
>>>>>>>>>> the problem is that you're looking at declspecs when you ought to be looking
>>>>>>>>>> at type_quals.
>>>>>>>>>
>>>>>>>>> I need to know that the base type is a pointer to reject invalid
>>>>>>>>> declarations such as:
>>>>>>>>>
>>>>>>>>>          int f (__seg_fs int a) { }     or     int f () { __seg_fs int a; }
>>>>>>>>>
>>>>>>>>> because parameters and auto variables can have an address space
>>>>>>>>> qualifier only if they are pointer or reference type, which I can't
>>>>>>>>> tell only from type_quals.
>>>>>>>>
>>>>>>>> But "int *__seg_fs a" is just as invalid as the above; the difference is not
>>>>>>>> whether a is a pointer, but whether the address-space-qualified is the type
>>>>>>>> of a itself or some sub-type.
>>>>>>>
>>>>>>> I agree that "int * __seg_fs a" is invalid but it is accepted by the C
>>>>>>> front-end, and by clang (both C and C++), the behavior is that the
>>>>>>> address-name is silently ignored.
>>>>>>
>>>>>> Hmm, that sounds like a bug; in that case, presumably the user meant to
>>>>>> qualify the pointed-to type, and silently ignoring seems unlikely to give
>>>>>> the effect they want.
>>>>>>
>>>>>
>>>>> Well, actually, I'm re-reading the draft and "int * __seg_fs a" is
>>>>> valid.  It means "pointer in address space __seg_fs pointing to an
>>>>> object in the generic address space", whereas "__seg_fs int * a" means
>>>>> "pointer in the generic address space pointing to an object in the
>>>>> __seg_fs address-space".
>>>>>
>>>>> Oddities such as, "__seg_fs int * __seg_gs a" are also perfectly
>>>>> valid.
>>>>
>>>> If a has static storage duration, sure; I was still thinking about
>>>> declarations with automatic storage duration such as in your example above.
>>>>
>>>
>>> Thanks, I only use type_quals now. I also took into account the style
>>> recommendations from Jakub, and included the other template tests.
>>> I rebased over trunk, bootstrapped the compiler and run the "make
>>> check-gcc" with no regressions on x86.
>>>
>>> Paul
>>>
>>> # ------------------------ >8 ------------------------
>>> Add support for custom address spaces in C++
>>>
>>> gcc/
>>>           * tree.h (ENCODE_QUAL_ADDR_SPACE): Missing parentheses.
>>>
>>> gcc/c/
>>>           * c-decl.cc: Remove c_register_addr_space.
>>>
>>> gcc/c-family/
>>>           * c-common.cc (c_register_addr_space): Imported from c-decl.cc
>>>           (addr_space_superset): Imported from gcc/c/c-typecheck.cc
>>>           * c-common.h: Remove the FIXME.
>>>           (addr_space_superset): New declaration.
>>>
>>> gcc/cp/
>>>           * cp-tree.h (enum cp_decl_spec): Add addr_space support.
>>>           (struct cp_decl_specifier_seq): Likewise.
>>>           * decl.cc (get_type_quals): Likewise.
>>>           (check_tag_decl): Likewise.
>>> 	(grokdeclarator): Likewise.
>>>           * parser.cc (cp_parser_type_specifier): Likewise.
>>>           (cp_parser_cv_qualifier_seq_opt): Likewise.
>>>           (cp_parser_postfix_expression): Likewise.
>>>           (cp_parser_type_specifier): Likewise.
>>>           (set_and_check_decl_spec_loc): Likewise.
>>>           * typeck.cc (composite_pointer_type): Likewise
>>>           (comp_ptr_ttypes_real): Likewise.
>>> 	(same_type_ignoring_top_level_qualifiers_p): Likewise.
>>>           * pt.cc (check_cv_quals_for_unify): Likewise.
>>>           (unify): Likewise.
>>>           * tree.cc: Remove c_register_addr_space stub.
>>>           * mangle.cc (write_CV_qualifiers_for_type): Mangle address spaces
>>>             using the extended qualifier notation.
>>>
>>> gcc/doc
>>>           * extend.texi (Named Address Spaces): add a mention about C++
>>>             support.
>>>
>>> gcc/testsuite/
>>>           * g++.dg/abi/mangle-addr-space1.C: New test.
>>>           * g++.dg/abi/mangle-addr-space2.C: New test.
>>>           * g++.dg/parse/addr-space.C: New test.
>>>           * g++.dg/parse/addr-space1.C: New test.
>>>           * g++.dg/parse/addr-space2.C: New test.
>>>           * g++.dg/parse/template/spec-addr-space.C: New test.
>>>           * g++.dg/ext/addr-space-decl.C: New test.
>>>           * g++.dg/ext/addr-space-ref.C: New test.
>>>           * g++.dg/ext/addr-space-ops.C: New test.
>>>           * g++.dg/template/addr-space-overload.C: New test.
>>>           * g++.dg/template/addr-space-strip1.C: New test.
>>>           * g++.dg/template/addr-space-strip2.C: New test.
>>>
>>> # ------------------------ >8 ------------------------
>>> diff --git a/gcc/c-family/c-common.cc b/gcc/c-family/c-common.cc
>>> index 9ec9100cc90..3b79dc47515 100644
>>> --- a/gcc/c-family/c-common.cc
>>> +++ b/gcc/c-family/c-common.cc
>>> @@ -588,6 +588,33 @@ c_addr_space_name (addr_space_t as)
>>>      return IDENTIFIER_POINTER (ridpointers [rid]);
>>>    }
>>> +/* Return true if between two named address spaces, whether there is a superset
>>> +   named address space that encompasses both address spaces.  If there is a
>>> +   superset, return which address space is the superset.  */
>>> +
>>> +bool
>>> +addr_space_superset (addr_space_t as1, addr_space_t as2,
>>> +		     addr_space_t * common)
>>> +{
>>> +  if (as1 == as2)
>>> +    {
>>> +      *common = as1;
>>> +      return true;
>>> +    }
>>> +  else if (targetm.addr_space.subset_p (as1, as2))
>>> +    {
>>> +      *common = as2;
>>> +      return true;
>>> +    }
>>> +  else if (targetm.addr_space.subset_p (as2, as1))
>>> +    {
>>> +      *common = as1;
>>> +      return true;
>>> +    }
>>> +  else
>>> +    return false;
>>> +}
>>> +
>>>    /* Push current bindings for the function name VAR_DECLS.  */
>>>    void
>>> @@ -2785,6 +2812,25 @@ c_build_bitfield_integer_type (unsigned HOST_WIDE_INT width, int unsignedp)
>>>      return build_nonstandard_integer_type (width, unsignedp);
>>>    }
>>> +/* Register reserved keyword WORD as qualifier for address space AS.  */
>>> +
>>> +void
>>> +c_register_addr_space (const char *word, addr_space_t as)
>>> +{
>>> +  int rid = RID_FIRST_ADDR_SPACE + as;
>>> +  tree id;
>>> +
>>> +  /* Address space qualifiers are only supported
>>> +     in C with GNU extensions enabled.  */
>>> +  if (c_dialect_objc () || flag_no_asm)
>>> +    return;
>>> +
>>> +  id = get_identifier (word);
>>> +  C_SET_RID_CODE (id, rid);
>>> +  TREE_LANG_FLAG_0 (id) = 1;
>>> +  ridpointers[rid] = id;
>>> +}
>>> +
>>>    /* The C version of the register_builtin_type langhook.  */
>>>    void
>>> diff --git a/gcc/c-family/c-common.h b/gcc/c-family/c-common.h
>>> index 62ab4ba437b..a3864d874aa 100644
>>> --- a/gcc/c-family/c-common.h
>>> +++ b/gcc/c-family/c-common.h
>>> @@ -829,12 +829,11 @@ extern const struct attribute_spec c_common_format_attribute_table[];
>>>    extern tree (*make_fname_decl) (location_t, tree, int);
>>> -/* In c-decl.cc and cp/tree.cc.  FIXME.  */
>>> -extern void c_register_addr_space (const char *str, addr_space_t as);
>>> -
>>>    /* In c-common.cc.  */
>>>    extern bool in_late_binary_op;
>>>    extern const char *c_addr_space_name (addr_space_t as);
>>> +extern const char *c_addr_space_name (addr_space_t as);
>>> +extern bool addr_space_superset (addr_space_t, addr_space_t, addr_space_t *);
>>>    extern tree identifier_global_value (tree);
>>>    extern tree identifier_global_tag (tree);
>>>    extern bool names_builtin_p (const char *);
>>> @@ -951,6 +950,7 @@ extern bool c_common_init (void);
>>>    extern void c_common_finish (void);
>>>    extern void c_common_parse_file (void);
>>>    extern alias_set_type c_common_get_alias_set (tree);
>>> +extern void c_register_addr_space (const char *, addr_space_t);
>>>    extern void c_register_builtin_type (tree, const char*);
>>>    extern bool c_promoting_integer_type_p (const_tree);
>>>    extern bool self_promoting_args_p (const_tree);
>>> diff --git a/gcc/c/c-decl.cc b/gcc/c/c-decl.cc
>>> index a7571cc7542..b1f69997ff7 100644
>>> --- a/gcc/c/c-decl.cc
>>> +++ b/gcc/c/c-decl.cc
>>> @@ -12531,25 +12531,6 @@ c_parse_final_cleanups (void)
>>>      ext_block = NULL;
>>>    }
>>> -/* Register reserved keyword WORD as qualifier for address space AS.  */
>>> -
>>> -void
>>> -c_register_addr_space (const char *word, addr_space_t as)
>>> -{
>>> -  int rid = RID_FIRST_ADDR_SPACE + as;
>>> -  tree id;
>>> -
>>> -  /* Address space qualifiers are only supported
>>> -     in C with GNU extensions enabled.  */
>>> -  if (c_dialect_objc () || flag_no_asm)
>>> -    return;
>>> -
>>> -  id = get_identifier (word);
>>> -  C_SET_RID_CODE (id, rid);
>>> -  C_IS_RESERVED_WORD (id) = 1;
>>> -  ridpointers [rid] = id;
>>> -}
>>> -
>>>    /* Return identifier to look up for omp declare reduction.  */
>>>    tree
>>> diff --git a/gcc/c/c-typeck.cc b/gcc/c/c-typeck.cc
>>> index fdb96c28c51..2a700bbaff3 100644
>>> --- a/gcc/c/c-typeck.cc
>>> +++ b/gcc/c/c-typeck.cc
>>> @@ -303,32 +303,6 @@ c_type_promotes_to (tree type)
>>>      return type;
>>>    }
>>> -/* Return true if between two named address spaces, whether there is a superset
>>> -   named address space that encompasses both address spaces.  If there is a
>>> -   superset, return which address space is the superset.  */
>>> -
>>> -static bool
>>> -addr_space_superset (addr_space_t as1, addr_space_t as2, addr_space_t *common)
>>> -{
>>> -  if (as1 == as2)
>>> -    {
>>> -      *common = as1;
>>> -      return true;
>>> -    }
>>> -  else if (targetm.addr_space.subset_p (as1, as2))
>>> -    {
>>> -      *common = as2;
>>> -      return true;
>>> -    }
>>> -  else if (targetm.addr_space.subset_p (as2, as1))
>>> -    {
>>> -      *common = as1;
>>> -      return true;
>>> -    }
>>> -  else
>>> -    return false;
>>> -}
>>> -
>>>    /* Return a variant of TYPE which has all the type qualifiers of LIKE
>>>       as well as those of TYPE.  */
>>> diff --git a/gcc/cp/cp-tree.h b/gcc/cp/cp-tree.h
>>> index e2607f09c19..0248569a95b 100644
>>> --- a/gcc/cp/cp-tree.h
>>> +++ b/gcc/cp/cp-tree.h
>>> @@ -6235,6 +6235,7 @@ enum cp_decl_spec {
>>>      ds_const,
>>>      ds_volatile,
>>>      ds_restrict,
>>> +  ds_addr_space,
>>>      ds_inline,
>>>      ds_virtual,
>>>      ds_explicit,
>>> @@ -6281,6 +6282,8 @@ struct cp_decl_specifier_seq {
>>>      cp_storage_class storage_class;
>>>      /* For the __intN declspec, this stores the index into the int_n_* arrays.  */
>>>      int int_n_idx;
>>> +  /* The address space that the declaration belongs to.  */
>>> +  addr_space_t address_space;
>>>      /* True iff TYPE_SPEC defines a class or enum.  */
>>>      BOOL_BITFIELD type_definition_p : 1;
>>>      /* True iff multiple types were (erroneously) specified for this
>>> diff --git a/gcc/cp/decl.cc b/gcc/cp/decl.cc
>>> index 85b892cddf0..a87fed04529 100644
>>> --- a/gcc/cp/decl.cc
>>> +++ b/gcc/cp/decl.cc
>>> @@ -5290,6 +5290,8 @@ get_type_quals (const cp_decl_specifier_seq *declspecs)
>>>        type_quals |= TYPE_QUAL_VOLATILE;
>>>      if (decl_spec_seq_has_spec_p (declspecs, ds_restrict))
>>>        type_quals |= TYPE_QUAL_RESTRICT;
>>> +  if (decl_spec_seq_has_spec_p (declspecs, ds_addr_space))
>>> +    type_quals |= ENCODE_QUAL_ADDR_SPACE (declspecs->address_space);
>>>      return type_quals;
>>>    }
>>> @@ -5412,6 +5414,10 @@ check_tag_decl (cp_decl_specifier_seq *declspecs,
>>>    	error_at (declspecs->locations[ds_restrict],
>>>    		  "%<__restrict%> can only be specified for objects and "
>>>    		  "functions");
>>> +      else if (decl_spec_seq_has_spec_p (declspecs, ds_addr_space))
>>> +	error_at (declspecs->locations[ds_addr_space],
>>> +		  "address space can only be specified for objects and "
>>> +		  "functions");
>>>          else if (decl_spec_seq_has_spec_p (declspecs, ds_thread))
>>>    	error_at (declspecs->locations[ds_thread],
>>>    		  "%<__thread%> can only be specified for objects "
>>> @@ -14608,6 +14614,59 @@ grokdeclarator (const cp_declarator *declarator,
>>>        if (!processing_template_decl)
>>>          cp_apply_type_quals_to_decl (type_quals, decl);
>>> +  /* Warn about address space used for things other than static memory or
>>> +     pointers.  */
>>> +    addr_space_t address_space = DECODE_QUAL_ADDR_SPACE (type_quals);
>>> +      if (!ADDR_SPACE_GENERIC_P (address_space))
>>> +      {
>>> +	if (decl_context == NORMAL)
>>> +	  {
>>> +	    switch (storage_class)
>>
>> I would still suggest checking decl_storage_duration at this point rather
>> than the storage_class specifier.
> 
> Unless I misunderstand something, I can't weed out register variables
> if I rely on decl_storage_duration.

Yes, but register variables are automatic, so they'll get that error; I 
don't think they need their own specific error.

>>> +	      {
>>> +	      case sc_auto:
>>> +		error ("%qs combined with C++98 %<auto%> qualifier for %qs",
>>> +		       c_addr_space_name (address_space), name);
>>> +		break;
>>> +	      case sc_register:
>>> +		error ("%qs combined with %<register%> qualifier for %qs",
>>> +		       c_addr_space_name (address_space), name);
>>> +		break;
>>> +	      case sc_none:
>>> +		if (! toplevel_bindings_p ())
>>> +		  error ("%qs specified for auto variable %qs",
>>
>> And let's refer to automatic storage duration rather than shorten to 'auto'.
>>
> Right.
>>> +			 c_addr_space_name (address_space), name);
>>> +		break;
>>> +	      case sc_mutable:
>>> +		error ("%qs combined with %<mutable%> qualifier for %qs",
>>> +		       c_addr_space_name (address_space), name);
>>> +		break;
>>> +	      case sc_static:
>>> +	      case sc_extern:
>>> +		break;
>>> +	      default:
>>> +		gcc_unreachable ();
>>> +	      }
>>> +	  }
>>> +	else if (decl_context == PARM && TREE_CODE (type) != ARRAY_TYPE)
>>> +	  {
>>> +	    if (name)
>>> +	      error ("%qs specified for parameter %qs",
>>> +		     c_addr_space_name (address_space), name);
>>> +	    else
>>> +	      error ("%qs specified for unnamed parameter",
>>> +		     c_addr_space_name (address_space));
>>> +	  }
>>> +	else if (decl_context == FIELD)
>>> +	  {
>>> +	    if (name)
>>> +	      error ("%qs specified for structure field %qs",
>>> +		     c_addr_space_name (address_space), name);
>>> +	    else
>>> +	      error ("%qs specified for structure field",
>>> +		     c_addr_space_name (address_space));
>>> +	  }
>>> +      }
>>> +
>>>        return decl;
>>>      }
>>>    }
>>> diff --git a/gcc/cp/mangle.cc b/gcc/cp/mangle.cc
>>> index 1215463089b..aafff98f05a 100644
>>> --- a/gcc/cp/mangle.cc
>>> +++ b/gcc/cp/mangle.cc
>>> @@ -2520,6 +2520,14 @@ write_CV_qualifiers_for_type (const tree type)
>>>         array.  */
>>>      cp_cv_quals quals = TYPE_QUALS (type);
>>> +  if (addr_space_t as = DECODE_QUAL_ADDR_SPACE (quals))
>>> +    {
>>> +      const char *as_name = c_addr_space_name (as);
>>> +      write_char ('U');
>>> +      write_unsigned_number (strlen (as_name));
>>> +      write_string (as_name);
>>> +      ++num_qualifiers;
>>> +    }
>>>      if (quals & TYPE_QUAL_RESTRICT)
>>>        {
>>>          write_char ('r');
>>> diff --git a/gcc/cp/parser.cc b/gcc/cp/parser.cc
>>> index 9ddfb027ff9..c82059d1efd 100644
>>> --- a/gcc/cp/parser.cc
>>> +++ b/gcc/cp/parser.cc
>>> @@ -7703,6 +7703,15 @@ cp_parser_postfix_expression (cp_parser *parser, bool address_p, bool cast_p,
>>>    		    postfix_expression = error_mark_node;
>>>    		    break;
>>>    		  }
>>> +		if (type != error_mark_node
>>> +		    && !ADDR_SPACE_GENERIC_P (TYPE_ADDR_SPACE (type))
>>> +		    && current_function_decl)
>>> +		  {
>>> +		    error
>>> +		      ("compound literal qualified by address-space "
>>> +		       "qualifier");
>>> +		    type = error_mark_node;
>>> +		  }
>>>    		/* Form the representation of the compound-literal.  */
>>>    		postfix_expression
>>>    		  = finish_compound_literal (type, initializer,
>>> @@ -19445,6 +19454,15 @@ cp_parser_type_specifier (cp_parser* parser,
>>>          break;
>>>        }
>>> +
>>> +  if (RID_FIRST_ADDR_SPACE <= keyword && keyword <= RID_LAST_ADDR_SPACE)
>>> +    {
>>> +      ds = ds_addr_space;
>>> +      if (is_cv_qualifier)
>>> +	*is_cv_qualifier = true;
>>> +    }
>>> +
>>> +
>>
>> I don't think we need two blank lines before and after this block, one each
>> should be enough.
>>
> Indeed.
>>>      /* Handle simple keywords.  */
>>>      if (ds != ds_last)
>>>        {
>>> @@ -23837,6 +23855,7 @@ cp_parser_ptr_operator (cp_parser* parser,
>>>       GNU Extension:
>>>       cv-qualifier:
>>> +     address-space-qualifier
>>>         __restrict__
>>>       Returns a bitmask representing the cv-qualifiers.  */
>>> @@ -23873,6 +23892,11 @@ cp_parser_cv_qualifier_seq_opt (cp_parser* parser)
>>>    	  break;
>>>    	}
>>> +      if (RID_FIRST_ADDR_SPACE <= token->keyword
>>> +	  && token->keyword <= RID_LAST_ADDR_SPACE)
>>> +	cv_qualifier
>>> +	  = ENCODE_QUAL_ADDR_SPACE (token->keyword - RID_FIRST_ADDR_SPACE);
>>> +
>>>          if (!cv_qualifier)
>>>    	break;
>>> @@ -32893,6 +32917,8 @@ set_and_check_decl_spec_loc (cp_decl_specifier_seq *decl_specs,
>>>          decl_specs->locations[ds] = location;
>>>          if (ds == ds_thread)
>>>    	decl_specs->gnu_thread_keyword_p = token_is__thread (token);
>>> +      else if (ds == ds_addr_space)
>>> +	decl_specs->address_space = token->keyword - RID_FIRST_ADDR_SPACE;
>>>        }
>>>      else
>>>        {
>>> @@ -32925,6 +32951,25 @@ set_and_check_decl_spec_loc (cp_decl_specifier_seq *decl_specs,
>>>    	      error_at (&richloc, "duplicate %qD", token->u.value);
>>>    	    }
>>>    	}
>>> +      else if (ds == ds_addr_space)
>>> +	{
>>> +	  addr_space_t as1 = decl_specs->address_space;
>>> +	  addr_space_t as2 = token->keyword - RID_FIRST_ADDR_SPACE;
>>> +
>>> +	  gcc_rich_location richloc (location);
>>> +	  richloc.add_fixit_remove ();
>>> +	  if (!ADDR_SPACE_GENERIC_P (as1) && !ADDR_SPACE_GENERIC_P (as2)
>>> +	      && as1 != as2)
>>> +	    error_at (&richloc,
>>> +		      "conflicting named address spaces (%s vs %s)",
>>> +		      c_addr_space_name (as1), c_addr_space_name (as2));
>>> +	  if (as1 == as2 && !ADDR_SPACE_GENERIC_P (as1))
>>> +	    error_at (&richloc,
>>> +		      "duplicate named address space %s",
>>> +		      c_addr_space_name (as1));
>>> +
>>> +	  decl_specs->address_space = as2;
>>> +	}
>>>          else
>>>    	{
>>>    	  static const char *const decl_spec_names[] = {
>>> diff --git a/gcc/cp/pt.cc b/gcc/cp/pt.cc
>>> index e4dca9d4f9d..7b73a57091e 100644
>>> --- a/gcc/cp/pt.cc
>>> +++ b/gcc/cp/pt.cc
>>> @@ -23778,8 +23778,19 @@ template_decl_level (tree decl)
>>>    static int
>>>    check_cv_quals_for_unify (int strict, tree arg, tree parm)
>>>    {
>>> -  int arg_quals = cp_type_quals (arg);
>>> -  int parm_quals = cp_type_quals (parm);
>>> +  int arg_quals = CLEAR_QUAL_ADDR_SPACE (cp_type_quals (arg));
>>> +  int parm_quals = CLEAR_QUAL_ADDR_SPACE (cp_type_quals (parm));
>>> +
>>> +  /*  Try to unify ARG's address space into PARM's address space.
>>> +      If PARM does not have any address space qualifiers (ie., as_parm is 0),
>>> +      there are no constraints on address spaces for this type.  */
>>> +  addr_space_t as_arg = DECODE_QUAL_ADDR_SPACE (cp_type_quals (arg));
>>> +  addr_space_t as_parm = DECODE_QUAL_ADDR_SPACE (cp_type_quals (parm));
>>> +  addr_space_t as_common;
>>> +  addr_space_superset (as_arg, as_parm, &as_common);
>>> +
>>> +  if (!(as_parm == as_common || as_parm == 0))
>>> +    return 0;
>>
>> I'd expect address space qualifiers to follow the 'strict' parameter like
>> the other qualifiers; the above test seems to assume
>> UNIFY_ALLOW_{OUTER_,}LESS_CV_QUAL.
>>
> The reason I ignored strict was to enforce that the deduced address
> space is always at most "as_parm" unless "as_parm" is the generic address
> space, and prevent unifying if the two address spaces are disjoint
> unless "parm" does not have any address space constraints; and avoid the
> addition/deletion of an address space to "arg" during the unifying
> process.
> 
> Since I don't really understand the whole picture behind strict, and when
> check_cv_quals_for_unify gets called with which variant of restrict it
> might be me who tried to be overcareful when unifying the address
> spaces.

How we need to handle differing qualifiers varies between different 
template argument deduction contexts.

The code you wrote above is correct for the function call context, since 
https://eel.is/c++draft/temp.deduct.call#4.2 says the deduced type can 
be convertable by qualification conversion, i.e. parm more qualified 
than arg (and my "LESS" above was backwards).  This is a bit different 
for address space qualifiers given that the qualification conversion 
would be removing the address space qualifier or changing it to a more 
general one, but the principle is the same.

But the allowance for qualifier changes doesn't apply to all deduction 
contexts: for instance,

template <class T> void f(T * const *);
struct A {
   template <class T> operator T**();
};
int main()
{
   f((void**)0); // void** -> void*const* is a valid qualification conv
   (void *const*)A(); // same conversion
   void (*p)(void **) = f; // error, type mismatch
}

so similarly,

template <class T> void f(T **);
struct A {
   template <class T> operator T*__seg_fs*();
};
int main()
{
   f((void* __seg_fs *)0); // void*__seg_fs* -> void** should be OK
   (void **)A(); // same conversion
   void (*p)(void * __seg_fs *) = f; // error
}


>>>      if (TREE_CODE (parm) == TEMPLATE_TYPE_PARM
>>>          && !(strict & UNIFY_ALLOW_OUTER_MORE_CV_QUAL))
>>> @@ -24415,10 +24426,28 @@ unify (tree tparms, tree targs, tree parm, tree arg, int strict,
>>>    					 arg, parm))
>>>    	    return unify_cv_qual_mismatch (explain_p, parm, arg);
>>> +	  int arg_cv_quals = cp_type_quals (arg);
>>> +	  int parm_cv_quals = cp_type_quals (parm);
>>> +
>>> +	  /* If PARM does not contain any address spaces constraints it can
>>> +	     fully match the address space of ARG.  However, if PARM contains an
>>> +	     address space constraints, it becomes the upper bound.  That is,
>>> +	     AS_ARG may be promoted to AS_PARM but not the converse.  If we
>>> +	     ended up here, it means that `check_cv_quals_for_unify' succeeded
>>> +	     and that either AS_PARM is 0 (ie., no constraints) or AS_COMMON ==
>>> +	     AS_PARM.  */
>>> +	  addr_space_t as_arg = DECODE_QUAL_ADDR_SPACE (arg_cv_quals);
>>> +	  addr_space_t as_parm = DECODE_QUAL_ADDR_SPACE (parm_cv_quals);
>>> +	  addr_space_t as_common = as_parm ? 0 : as_arg;
>>
>> Hmm, I'd think we also want as_common = as_arg when it's a subset of
>> as_parm.
> 
> Let's assume that "PARM" is "__as1 T", and since the call to
> check_cv_quals_for_unify succeeded we know that "as_common" is
> "__as1". That is ARG is of the form "__as2 U" with "__as2" a
> subset of "__as1", hence we are trying to unify
>                    __as1 T = __as1 U
> which does not give any constraints over PARM since it alreay contains
> the common address space, hence there is no more constraints on T and
> as_common = 0.

Agreed.

> However, if PARM's address space is 0, we are trying to unify
>                           T = __as1 U
> and we need to add __addr_space1 to the constraints of T.

Agreed.

> If as_parm is not the generic address space (ie, as_parm != 0)

Looks like this comment got cut off?  This is the case I was talking 
about.  When we are trying to unify

   __as1 T = __as2 U

and __as2 is a subset of __as1, I think we want T to be deduced to __as2 
U, and then substitution will need to handle substituting __as2 U for T 
into __as1 T to get __as2 U.

>>>    	  /* Consider the case where ARG is `const volatile int' and
>>>    	     PARM is `const T'.  Then, T should be `volatile int'.  */
>>>    	  arg = cp_build_qualified_type
>>>    	    (arg, cp_type_quals (arg) & ~cp_type_quals (parm), tf_none);
>>> +	  int unified_cv =
>>> +	    (CLEAR_QUAL_ADDR_SPACE (arg_cv_quals & ~parm_cv_quals)
>>> +	    | ENCODE_QUAL_ADDR_SPACE (as_common));
>>> +	  arg = cp_build_qualified_type (arg, unified_cv, tf_none);
>>>    	  if (arg == error_mark_node)
>>>    	    return unify_invalid (explain_p);
>>> diff --git a/gcc/cp/tree.cc b/gcc/cp/tree.cc
>>> index 45348c58bb6..1f330ca93ed 100644
>>> --- a/gcc/cp/tree.cc
>>> +++ b/gcc/cp/tree.cc
>>> @@ -6072,15 +6072,6 @@ cp_free_lang_data (tree t)
>>>        DECL_CHAIN (t) = NULL_TREE;
>>>    }
>>> -/* Stub for c-common.  Please keep in sync with c-decl.cc.
>>> -   FIXME: If address space support is target specific, then this
>>> -   should be a C target hook.  But currently this is not possible,
>>> -   because this function is called via REGISTER_TARGET_PRAGMAS.  */
>>> -void
>>> -c_register_addr_space (const char * /*word*/, addr_space_t /*as*/)
>>> -{
>>> -}
>>> -
>>>    /* Return the number of operands in T that we care about for things like
>>>       mangling.  */
>>> diff --git a/gcc/cp/typeck.cc b/gcc/cp/typeck.cc
>>> index da0e1427b97..93cfdc70e2d 100644
>>> --- a/gcc/cp/typeck.cc
>>> +++ b/gcc/cp/typeck.cc
>>> @@ -803,10 +803,28 @@ composite_pointer_type (const op_location_t &location,
>>>    	  else
>>>    	    return error_mark_node;
>>>            }
>>> +      /* If possible merge the address space into the superset of the address
>>> +	  spaces of t1 and t2, or raise an error. */
>>> +      addr_space_t as_t1 = TYPE_ADDR_SPACE (t1);
>>> +      addr_space_t as_t2 = TYPE_ADDR_SPACE (t2);
>>> +      addr_space_t as_common;
>>> +
>>> +      /* If the two named address spaces are different, determine the common
>>> +	 superset address space.  If there isn't one, raise an error.  */
>>> +      if (!addr_space_superset (as_t1, as_t2, &as_common))
>>> +	{
>>> +	  as_common = as_t1;
>>> +	  error_at (location,
>>> +		    "%qT and %qT are in disjoint named address spaces",
>>> +		    t1, t2);
>>
>> Why not return error_mark_node here?
>>
> That's a mistake. Thanks.
>>> +	}
>>> +      int quals_t1 = cp_type_quals (TREE_TYPE (t1));
>>> +      int quals_t2 = cp_type_quals (TREE_TYPE (t2));
>>>          result_type
>>>    	= cp_build_qualified_type (void_type_node,
>>> -				   (cp_type_quals (TREE_TYPE (t1))
>>> -				    | cp_type_quals (TREE_TYPE (t2))));
>>> +				   (CLEAR_QUAL_ADDR_SPACE (quals_t1)
>>> +				    | CLEAR_QUAL_ADDR_SPACE (quals_t2)
>>> +				    | ENCODE_QUAL_ADDR_SPACE (as_common)));
>>>          result_type = build_pointer_type (result_type);
>>>          /* Merge the attributes.  */
>>>          attributes = (*targetm.merge_type_attributes) (t1, t2);
>>> @@ -1731,7 +1749,9 @@ comptypes (tree t1, tree t2, int strict)
>>>    }
>>>    /* Returns nonzero iff TYPE1 and TYPE2 are the same type, ignoring
>>> -   top-level qualifiers.  */
>>> +   top-level qualifiers, except for named address spaces.  If the pointers point
>>> +   to different named addresses spaces, then we must determine if one address
>>> +   space is a subset of the other.  */
>>>    bool
>>>    same_type_ignoring_top_level_qualifiers_p (tree type1, tree type2)
>>> @@ -1741,6 +1761,14 @@ same_type_ignoring_top_level_qualifiers_p (tree type1, tree type2)
>>>      if (type1 == type2)
>>>        return true;
>>> +  addr_space_t as_type1 = TYPE_ADDR_SPACE (type1);
>>> +  addr_space_t as_type2 = TYPE_ADDR_SPACE (type2);
>>> +  addr_space_t as_common;
>>> +
>>> +  /* Fail if pointers point to incompatible address spaces.  */
>>> +  if (!addr_space_superset (as_type1, as_type2, &as_common))
>>> +    return false;
>>
>> Why do you need this change?  I'd expect this function to ignore top level
>> address space qualifiers like the other qualifiers.
> 
> I am mirroring the C front-end here, which does the same thing in
> "comp_target_types" (gcc/c/c-typeck.cc), which ignores qualifiers but
> not address spaces when checking if two pointer types are equivalent.

This function serves a very different function from comp_target_types, 
which deals with the types that pointers point to; this function is 
ignoring top-level qualifiers that should not affect the type.

...except now I see that cp_build_binary_op is wierdly using this 
function for pointer subtraction.  I'd think it should use 
composite_pointer_type instead, like EQ_EXPR does.

>>>      type1 = cp_build_qualified_type (type1, TYPE_UNQUALIFIED);
>>>      type2 = cp_build_qualified_type (type2, TYPE_UNQUALIFIED);
>>>      return same_type_p (type1, type2);
>>> @@ -6672,10 +6700,32 @@ static tree
>>>    pointer_diff (location_t loc, tree op0, tree op1, tree ptrtype,
>>>    	      tsubst_flags_t complain, tree *instrument_expr)
>>>    {
>>> -  tree result, inttype;
>>>      tree restype = ptrdiff_type_node;
>>> +  tree result, inttype;
>>> +
>>> +  addr_space_t as0 = TYPE_ADDR_SPACE (TREE_TYPE (TREE_TYPE (op0)));
>>> +  addr_space_t as1 = TYPE_ADDR_SPACE (TREE_TYPE (TREE_TYPE (op1)));
>>>      tree target_type = TREE_TYPE (ptrtype);
>>> +  /* If the operands point into different address spaces, we need to
>>> +     explicitly convert them to pointers into the common address space
>>> +     before we can subtract the numerical address values.  */
>>> +  if (as0 != as1)
>>> +    {
>>> +      addr_space_t as_common;
>>> +      tree common_type;
>>> +
>>> +      /* Determine the common superset address space.  This is guaranteed
>>> +	 to exist because the caller verified that comp_target_types
>>> +	 returned non-zero.  */
>>> +      if (!addr_space_superset (as0, as1, &as_common))
>>> +	gcc_unreachable ();
>>> +
>>> +      common_type = common_pointer_type (TREE_TYPE (op0), TREE_TYPE (op1));
>>> +      op0 = convert (common_type, op0);
>>> +      op1 = convert (common_type, op1);
>>> +    }
>>
>> I think you shouldn't need to change pointer_diff if composite_pointer_type
>> returns error_mark_node above.
> 
> I'll have a look, the idea here is to prevent "a - b" with "a" and "b"
> from different address spaces.

As above, I think this should have been handled in cp_build_binary_op.

>>>      if (!complete_type_or_maybe_complain (target_type, NULL_TREE, complain))
>>>        return error_mark_node;
>>> @@ -11286,6 +11336,19 @@ comp_ptr_ttypes_real (tree to, tree from, int constp)
>>>    	      to_more_cv_qualified = true;
>>>    	    }
>>> +      /* Warn about conversions between pointers to disjoint
>>> +	 address spaces.  */
>>> +      if (TREE_CODE (from) == POINTER_TYPE
>>> +	  && TREE_CODE (to) == POINTER_TYPE)
>>> +	{
>>> +	  addr_space_t as_from = TYPE_ADDR_SPACE (TREE_TYPE (from));
>>> +	  addr_space_t as_to = TYPE_ADDR_SPACE (TREE_TYPE (to));
>>> +	  addr_space_t as_common;
>>> +
>>> +	  if (!addr_space_superset (as_to, as_from, &as_common))
>>> +	    return false;
>>
>> I think you also want to check that as_common == as_to here?
>>
> Yes.
>>> +	}
>>> +
>>>    	  if (constp > 0)
>>>    	    constp &= TYPE_READONLY (to);
>>>    	}
>>> diff --git a/gcc/doc/extend.texi b/gcc/doc/extend.texi
>>> index cfbe32afce9..ef75f6b83a2 100644
>>> --- a/gcc/doc/extend.texi
>>> +++ b/gcc/doc/extend.texi
>>> @@ -1448,7 +1448,7 @@ Fixed-point types are supported by the DWARF debug information format.
>>>    @section Named Address Spaces
>>>    @cindex Named Address Spaces
>>> -As an extension, GNU C supports named address spaces as
>>> +As an extension, GNU C and GNU C++ support named address spaces as
>>>    defined in the N1275 draft of ISO/IEC DTR 18037.  Support for named
>>>    address spaces in GCC will evolve as the draft technical report
>>>    changes.  Calling conventions for any target might also change.  At
>>> diff --git a/gcc/testsuite/g++.dg/abi/mangle-addr-space1.C b/gcc/testsuite/g++.dg/abi/mangle-addr-space1.C
>>> new file mode 100644
>>> index 00000000000..c01f8d6054a
>>> --- /dev/null
>>> +++ b/gcc/testsuite/g++.dg/abi/mangle-addr-space1.C
>>> @@ -0,0 +1,10 @@
>>> +// { dg-do run { target { i?86-*-* x86_64-*-* } } }
>>
>> This can be dg-do compile, I don't think you get anything from running an
>> empty main.
>>
> Yes.
>>> +// { dg-options "-fabi-version=8 -Wabi -save-temps" }
>>
>> And then you don't need -save-temps.  What are the other options for?
>>
> I forgot to remove -Wabi and -fabi-version, this was from my first
> attempt when I used AS<number> to mangle which changed the ABI. I'll
> remove them.
>>> +// { dg-final { scan-assembler "_Z1fPU8__seg_fsVi" } }
>>> +
>>> +int f (int volatile __seg_fs *a)
>>> +{
>>> +  return *a;
>>> +}
>>> +
>>> +int main () {}
>>> diff --git a/gcc/testsuite/g++.dg/abi/mangle-addr-space2.C b/gcc/testsuite/g++.dg/abi/mangle-addr-space2.C
>>> new file mode 100644
>>> index 00000000000..862bbbdcdf2
>>> --- /dev/null
>>> +++ b/gcc/testsuite/g++.dg/abi/mangle-addr-space2.C
>>> @@ -0,0 +1,9 @@
>>> +// { dg-do run { target { i?86-*-* x86_64-*-* } } }
>>> +// { dg-options "-fabi-version=8 -Wabi -save-temps" }
>>
>> Also not clear that running is important for this test.
>>
> Noted.
>>> +// { dg-final { scan-assembler "_Z1fIU8__seg_fsiEiPT_" } }
>>> +
>>> +template <class T>
>>> +int f (T *p) { return *p; }
>>> +int g (__seg_fs int *p) { return *p; }
>>> +__seg_fs int *a;
>>> +int main() { f(a); }
>>> diff --git a/gcc/testsuite/g++.dg/ext/addr-space-decl.C b/gcc/testsuite/g++.dg/ext/addr-space-decl.C
>>> new file mode 100644
>>> index 00000000000..c04d2f497da
>>> --- /dev/null
>>> +++ b/gcc/testsuite/g++.dg/ext/addr-space-decl.C
>>> @@ -0,0 +1,5 @@
>>> +// { dg-do compile { target { i?86-*-* x86_64-*-* } } }
>>> +__seg_fs char a, b, c;
>>> +__seg_fs const int *p;
>>> +static /* give internal linkage to the following anonymous struct */
>>
>> Hmm, this 'static' gives internal linkage to the variable q, not the type.
>> What do you want it for?
>>
> Yes, the idea is to give internal linkage to q, otherwise g++
> complains in -std=c++98 mode because q is externally visible but it
> can't be reffered from anywhere else since there is no tag for this
> structure.

Then let's change the comment to /* give internal linkage to q */

>>> +__seg_fs struct { int a; char b; } * __seg_gs q;
>>> diff --git a/gcc/testsuite/g++.dg/ext/addr-space-ops.C b/gcc/testsuite/g++.dg/ext/addr-space-ops.C
>>> new file mode 100644
>>> index 00000000000..86c02d1e7f5
>>> --- /dev/null
>>> +++ b/gcc/testsuite/g++.dg/ext/addr-space-ops.C
>>> @@ -0,0 +1,20 @@
>>> +// { dg-do compile { target { i?86-*-* x86_64-*-* } } }
>>> +int __seg_fs * fs1;
>>> +int __seg_fs * fs2;
>>> +float __seg_gs * gs1;
>>> +float __seg_gs * gs2;
>>> +
>>> +int
>>> +main ()
>>> +{
>>> +  fs1 + fs2; // { dg-error "invalid operands of types .__seg_fs int.. and .__seg_fs int.. to binary .operator.." }
>>> +  fs1 - fs2;
>>> +  fs1 - gs2; // { dg-error "invalid operands of types .__seg_fs int.. and .__seg_gs float.. to binary .operator.." }
>>> +  fs1 == fs2;
>>> +  fs1 != gs2; // { dg-error "comparison between distinct pointer types .__seg_fs int.. and .__seg_gs float.. lacks a cast" }
>>> +  fs1 = fs2;
>>> +  fs1 = gs2; // { dg-error "cannot convert .__seg_gs float.. to .__seg_fs int.. in assignment" }
>>> +  fs1 > fs2;
>>> +  fs1 < gs2; // { dg-error "comparison between distinct pointer types .__seg_fs int.. and .__seg_gs float.. lacks a cast" }
>>> +  return 0;
>>> +}
>>> diff --git a/gcc/testsuite/g++.dg/ext/addr-space-ref.C b/gcc/testsuite/g++.dg/ext/addr-space-ref.C
>>> new file mode 100644
>>> index 00000000000..12d7975e560
>>> --- /dev/null
>>> +++ b/gcc/testsuite/g++.dg/ext/addr-space-ref.C
>>> @@ -0,0 +1,31 @@
>>> +// { dg-do compile { target { i?86-*-* x86_64-*-* } } }
>>> +// { dg-prune-output "does not allow .register. storage class specifier" }
>>> +int __seg_fs * outer_b;
>>> +
>>> +struct s {
>>> +  __seg_fs int * ok;
>>> +  __seg_gs int ko; // { dg-error ".__seg_gs. specified for structure field .ko." }
>>> +};
>>> +
>>> +int register __seg_fs reg_fs; // { dg-error ".__seg_fs. combined with .register. qualifier for .reg_fs." }
>>> +
>>> +namespace ns_a
>>> +{
>>> +  int __seg_fs * inner_b;
>>> +
>>> +  template<typename T>
>>> +  int f (T &a) { return a; }
>>> +  int g (__seg_fs int a) { return a; } // { dg-error ".__seg_fs. specified for parameter .a." }
>>> +  int h (__seg_fs int *a) { return *a; }
>>> +}
>>> +
>>> +int
>>> +main ()
>>> +{
>>> +  int register __seg_gs reg_gs; // { dg-error ".__seg_gs. combined with .register. qualifier for .reg_gs." }
>>> +  static __seg_gs int static_gs;
>>> +  __seg_fs int auto_fs; // { dg-error ".__seg_fs. specified for auto variable .auto_fs." }
>>> +  __seg_fs int *pa = outer_b;
>>> +  __seg_fs int& ra = *ns_a::inner_b;
>>> +  return ns_a::f(ra) + ns_a::f(*pa);
>>> +}
>>> diff --git a/gcc/testsuite/g++.dg/parse/addr-space.C b/gcc/testsuite/g++.dg/parse/addr-space.C
>>> new file mode 100644
>>> index 00000000000..ebb6316054a
>>> --- /dev/null
>>> +++ b/gcc/testsuite/g++.dg/parse/addr-space.C
>>> @@ -0,0 +1,9 @@
>>> +// { dg-do compile { target { i?86-*-* x86_64-*-* } } }
>>> +
>>> +__seg_fs struct foo; // { dg-error "address space can only be specified for objects and functions" }
>>> +
>>> +int
>>> +main ()
>>> +{
>>> +	return 0;
>>> +}
>>> diff --git a/gcc/testsuite/g++.dg/parse/addr-space1.C b/gcc/testsuite/g++.dg/parse/addr-space1.C
>>> new file mode 100644
>>> index 00000000000..2e8ee32a885
>>> --- /dev/null
>>> +++ b/gcc/testsuite/g++.dg/parse/addr-space1.C
>>> @@ -0,0 +1,10 @@
>>> +// { dg-do compile { target { i?86-*-* x86_64-*-* } } }
>>> +// { dg-options "-std=gnu++98" }
>>> +
>>> +int
>>> +main ()
>>> +{
>>> +	struct foo {int a; char b[2];} structure;
>>> +	structure = ((__seg_fs struct foo) {1 + 2, 'a', 0}); // { dg-error "compound literal qualified by address-space qualifier" }
>>> +	return 0;
>>> +}
>>> diff --git a/gcc/testsuite/g++.dg/parse/addr-space2.C b/gcc/testsuite/g++.dg/parse/addr-space2.C
>>> new file mode 100644
>>> index 00000000000..5b2c0f28078
>>> --- /dev/null
>>> +++ b/gcc/testsuite/g++.dg/parse/addr-space2.C
>>> @@ -0,0 +1,9 @@
>>> +// { dg-do compile { target { i?86-*-* x86_64-*-* } } }
>>> +
>>> +__seg_fs __seg_gs int *a; // { dg-error "conflicting named address spaces .__seg_fs vs __seg_gs." }
>>> +
>>> +int
>>> +main ()
>>> +{
>>> +	return 0;
>>> +}
>>> diff --git a/gcc/testsuite/g++.dg/template/addr-space-overload.C b/gcc/testsuite/g++.dg/template/addr-space-overload.C
>>> new file mode 100644
>>> index 00000000000..70dfcce53fa
>>> --- /dev/null
>>> +++ b/gcc/testsuite/g++.dg/template/addr-space-overload.C
>>> @@ -0,0 +1,17 @@
>>> +// { dg-do compile { target { i?86-*-* x86_64-*-* } } }
>>> +
>>> +int __seg_fs * fs1;
>>> +int __seg_gs * gs1;
>>> +
>>> +template<typename T, typename U>
>>> +__seg_fs T* f (T __seg_fs * a, U __seg_gs * b) { return a; }
>>> +template<typename T, typename U>
>>> +__seg_gs T* f (T __seg_gs * a, U __seg_fs * b) { return a; }
>>> +
>>> +int
>>> +main ()
>>> +{
>>> +    f (fs1, gs1);
>>> +    f (gs1, fs1);
>>> +    return 0;
>>> +}
>>> diff --git a/gcc/testsuite/g++.dg/template/addr-space-strip1.C b/gcc/testsuite/g++.dg/template/addr-space-strip1.C
>>> new file mode 100644
>>> index 00000000000..5df115db939
>>> --- /dev/null
>>> +++ b/gcc/testsuite/g++.dg/template/addr-space-strip1.C
>>> @@ -0,0 +1,17 @@
>>> +// { dg-do compile { target { i?86-*-* x86_64-*-* } } }
>>> +// { dg-skip-if "" { *-*-* } { "-std=c++98" "-std=c++03" "-std=gnu++98" "-std=gnu++03" } { "" } }
>>
>> This can be { dg-require-effective-target c++11 }

Or put the x86 requirement in dg-require-effective-target, and put c++11 
in the dg-do target spec, either way.

>>> +// decltype is ony available since c++11
>>
>> "only"
>>
>>> +
>>> +int __seg_fs * fs1;
>>> +int __seg_gs * gs1;
>>> +
>>> +template<typename T> struct strip;
>>> +template<typename T> struct strip<__seg_fs T *> { typedef T type; };
>>> +template<typename T> struct strip<__seg_gs T *> { typedef T type; };
>>> +
>>> +int
>>> +main ()
>>> +{
>>> +    *(strip<decltype(fs1)>::type *) fs1 == *(strip<decltype(gs1)>::type *) gs1;
>>> +    return 0;
>>> +}
>>> diff --git a/gcc/testsuite/g++.dg/template/addr-space-strip2.C b/gcc/testsuite/g++.dg/template/addr-space-strip2.C
>>> new file mode 100644
>>> index 00000000000..526bbaa56b7
>>> --- /dev/null
>>> +++ b/gcc/testsuite/g++.dg/template/addr-space-strip2.C
>>> @@ -0,0 +1,16 @@
>>> +// { dg-do compile { target { i?86-*-* x86_64-*-* } } }
>>> +
>>> +int __seg_fs * fs1;
>>> +int __seg_gs * gs1;
>>> +
>>> +template<typename T, typename U>
>>> +bool f (T __seg_fs * a, U __seg_gs * b)
>>> +{
>>> +    return *(T *) a == *(U *) b;
>>> +}
>>> +
>>> +int
>>> +main ()
>>> +{
>>> +    return f (fs1, gs1);
>>> +}
>>> diff --git a/gcc/testsuite/g++.dg/template/spec-addr-space.C b/gcc/testsuite/g++.dg/template/spec-addr-space.C
>>> new file mode 100644
>>> index 00000000000..ae9f4de0e1f
>>> --- /dev/null
>>> +++ b/gcc/testsuite/g++.dg/template/spec-addr-space.C
>>> @@ -0,0 +1,8 @@
>>> +// { dg-do compile { target { i?86-*-* x86_64-*-* } } }
>>> +
>>> +template <class T>
>>> +int f (T __seg_gs *p) { return *p; } // { dg-note "candidate: 'template<class T> int f.__seg_gs T\*." }
>>> +				     // { dg-note "template argument deduction/substitution failed:" "" { target *-*-* } .-1 }
>>> +__seg_fs int *a;
>>> +int main() { f(a); } // { dg-error "no matching" }
>>> +// { dg-note "types .__seg_gs T. and .__seg_fs int. have incompatible cv-qualifiers" "" { target *-*-* } .-1 }
>>> diff --git a/gcc/tree.h b/gcc/tree.h
>>> index 9af971cf401..4aebfef854b 100644
>>> --- a/gcc/tree.h
>>> +++ b/gcc/tree.h
>>> @@ -2292,7 +2292,7 @@ extern tree vector_element_bits_tree (const_tree);
>>>    /* Encode/decode the named memory support as part of the qualifier.  If more
>>>       than 8 qualifiers are added, these macros need to be adjusted.  */
>>> -#define ENCODE_QUAL_ADDR_SPACE(NUM) ((NUM & 0xFF) << 8)
>>> +#define ENCODE_QUAL_ADDR_SPACE(NUM) (((NUM) & 0xFF) << 8)
>>>    #define DECODE_QUAL_ADDR_SPACE(X) (((X) >> 8) & 0xFF)
>>>    /* Return all qualifiers except for the address space qualifiers.  */
> 
> 
> 
> 
> 
>