Jason Merrill <ja...@redhat.com> writes:
> On 12/18/19 1:24 PM, Richard Sandiford wrote:
>> The SVE port needs to maintain a different type identity for
>> GNU vectors and "SVE vectors" even during LTO, since the types
>> use different ABIs. The easiest way of doing that seemed to be
>> to use type attributes. However, these type attributes shouldn't
>> be user-facing; they're just a convenient way of representing the
>> types internally in GCC.
>>
>> There are already several internal-only attributes, such as "fn spec"
>> and "omp declare simd". They're distinguished from normal user-facing
>> attributes by having a space in their name, which means that it isn't
>> possible to write them directly in C or C++.
>>
>> Taking the same approach mostly works well for SVE. The only snag
>> I've hit so far is that the new attribute needs to (and only exists to)
>> affect type identity. This means that it would normally get included
>> in mangled names, to distinguish it from types without the attribute.
>>
>> However, the SVE ABI specifies a separate mangling for SVE vector types,
>> rather than using an attribute mangling + a normal vector mangling.
>> So we need some way of suppressing the attribute mangling for this case.
>>
>> There are currently no other target-independent or target-specific
>> internal-only attributes that affect type identity, so this patch goes
>> for the simplest fix of skipping mangling for attributes whose names
>> contain a space. Other options I thought about were:
>>
>> (1) Also make sure that targetm.mangled_type returns nonnull.
>>
>> (2) Check directly for the target-specific name.
>>
>> (3) Add a new target hook.
>>
>> (4) Add new information to attribute_spec. This would be very invasive
>> at this stage, but maybe we should consider replacing all the boolean
>> fields with flags? That should make the tables slightly easier to
>> read and would make adding new flags much simpler in future.
>>
>> What do you think? Do any of these sound OK, or is there a better
>> way of doing it?
>>
>> Tested on aarch64-linux-gnu and x86_64-linux-gnu.
>>
>> Thanks,
>> Richard
>>
>>
>> 2019-12-18 Richard Sandiford <richard.sandif...@arm.com>
>>
>> gcc/cp/
>> * mangle.c (write_CV_qualifiers_for_type): Don't mangle attributes
>> that contain a space.
>
> This approach works for me. Though that condition is getting large
> enough that I might break it out into a separate function.
OK, here's a version with that change. I also expanded the comment
slightly.
Tested on aarch64-linux-gnu and x86_64-linux-gnu. OK to install?
Thanks,
Richard
2019-12-27 Richard Sandiford <richard.sandif...@arm.com>
gcc/cp/
* mangle.c (mangle_type_attribute_p): New function, split out from...
(write_CV_qualifiers_for_type): ...here. Don't mangle attributes
that contain a space.
Index: gcc/cp/mangle.c
===================================================================
--- gcc/cp/mangle.c 2019-12-27 15:59:54.000000000 +0000
+++ gcc/cp/mangle.c 2019-12-27 15:59:55.236935396 +0000
@@ -2348,6 +2348,34 @@ attr_strcmp (const void *p1, const void
return strcmp (as1->name, as2->name);
}
+/* Return true if we should mangle a type attribute with name NAME. */
+
+static bool
+mangle_type_attribute_p (tree name)
+{
+ const attribute_spec *as = lookup_attribute_spec (name);
+ if (!as || !as->affects_type_identity)
+ return false;
+
+ /* Skip internal-only attributes, which are distinguished from others
+ by having a space. At present, all internal-only attributes that
+ affect type identity are target-specific and are handled by
+ targetm.mangle_type instead.
+
+ Another reason to do this is that a space isn't a valid identifier
+ character for most file formats. */
+ if (strchr (IDENTIFIER_POINTER (name), ' '))
+ return false;
+
+ if (is_attribute_p ("transaction_safe", name))
+ return false;
+
+ if (is_attribute_p ("abi_tag", name))
+ return false;
+
+ return true;
+}
+
/* Non-terminal <CV-qualifiers> for type nodes. Returns the number of
CV-qualifiers written for TYPE.
@@ -2373,14 +2401,8 @@ write_CV_qualifiers_for_type (const tree
{
auto_vec<tree> vec;
for (tree a = TYPE_ATTRIBUTES (type); a; a = TREE_CHAIN (a))
- {
- tree name = get_attribute_name (a);
- const attribute_spec *as = lookup_attribute_spec (name);
- if (as && as->affects_type_identity
- && !is_attribute_p ("transaction_safe", name)
- && !is_attribute_p ("abi_tag", name))
- vec.safe_push (a);
- }
+ if (mangle_type_attribute_p (get_attribute_name (a)))
+ vec.safe_push (a);
if (abi_warn_or_compat_version_crosses (10) && !vec.is_empty ())
G.need_abi_warning = true;
if (abi_version_at_least (10))
