[PATCH v2] aarch64: Fix SVE ACLE builtins with LTO [PR99216]

[PATCH v2] aarch64: Fix SVE ACLE builtins with LTO [PR99216]

[Alex Coplan]

[-- Attachment #1: Type: text/plain, Size: 1821 bytes --]

Hi all,

Here is a v2 patch which provides a more obviously fake answer to
TARGET_BUILTIN_DECL: this can hopefully go in for GCC 11. For GCC 12, it
seems that we should consider removing the target hook.

Original patch:
https://gcc.gnu.org/pipermail/gcc-patches/2021-March/566405.html

---

As discussed in the PR, we currently have two different numbering
schemes for SVE builtins: one for C, and one for C++. This is
problematic for LTO, where we end up getting confused about which
intrinsic we're talking about. This patch inserts placeholders into the
registered_functions vector to ensure that there is a consistent
numbering scheme for both C and C++.

This patch uses integer_zero_node as a placeholder node instead of
building a function decl. This is safe because the node is only returned
by the TARGET_BUILTIN_DECL hook, which (on AArch64) is only used for
validation when builtin decls are streamed into lto1.

Bootstrapped and regtested on aarch64-linux-gnu, OK for trunk?

Thanks,
Alex

gcc/ChangeLog:

	PR target/99216
	* config/aarch64/aarch64-sve-builtins.cc
	(function_builder::add_function): Add placeholder_p argument, use
	placeholder decls if this is set.
	(function_builder::add_unique_function): Instead of conditionally adding
	direct overloads, unconditionally add either a direct overload or a
	placeholder.
	(function_builder::add_overloaded_function): Set placeholder_p if we're
	using C++ overloads. Use the obstack for string storage instead
	of relying on the tree nodes.
	(function_builder::add_overloaded_functions): Don't return early for
	m_direct_overloads: we need to add placeholders.
	* config/aarch64/aarch64-sve-builtins.h
	(function_builder::add_function): Add placeholder_p argument.

gcc/testsuite/ChangeLog:

	PR target/99216
	* g++.target/aarch64/sve/pr99216.C: New test.

[-- Attachment #2: patch_v2.txt --]
[-- Type: text/plain, Size: 5686 bytes --]

diff --git a/gcc/config/aarch64/aarch64-sve-builtins.cc b/gcc/config/aarch64/aarch64-sve-builtins.cc
index 25612d2ea52..f44f81f1375 100644
--- a/gcc/config/aarch64/aarch64-sve-builtins.cc
+++ b/gcc/config/aarch64/aarch64-sve-builtins.cc
@@ -999,12 +999,29 @@ registered_function &
 function_builder::add_function (const function_instance &instance,
 				const char *name, tree fntype, tree attrs,
 				uint64_t required_extensions,
-				bool overloaded_p)
+				bool overloaded_p,
+				bool placeholder_p)
 {
   unsigned int code = vec_safe_length (registered_functions);
   code = (code << AARCH64_BUILTIN_SHIFT) | AARCH64_BUILTIN_SVE;
-  tree decl = simulate_builtin_function_decl (input_location, name, fntype,
-					      code, NULL, attrs);
+
+  /* We need to be able to generate placeholders to enusre that we have a
+     consistent numbering scheme for function codes between the C and C++
+     frontends, so that everything ties up in LTO.
+
+     Currently, tree-streamer-in.c:unpack_ts_function_decl_value_fields
+     validates that tree nodes returned by TARGET_BUILTIN_DECL are non-NULL and
+     some node other than error_mark_node. This is a holdover from when builtin
+     decls were streamed by code rather than by value.
+
+     Ultimately, we should be able to remove this validation of BUILT_IN_MD
+     nodes and remove the target hook. For now, however, we need to appease the
+     validation and return a non-NULL, non-error_mark_node node, so we
+     arbitrarily choose integer_zero_node.  */
+  tree decl = placeholder_p
+    ? integer_zero_node
+    : simulate_builtin_function_decl (input_location, name, fntype,
+				      code, NULL, attrs);
 
   registered_function &rfn = *ggc_alloc <registered_function> ();
   rfn.instance = instance;
@@ -1036,7 +1053,7 @@ function_builder::add_unique_function (const function_instance &instance,
 					   argument_types.address ());
   tree attrs = get_attributes (instance);
   registered_function &rfn = add_function (instance, name, fntype, attrs,
-					   required_extensions, false);
+					   required_extensions, false, false);
 
   /* Enter the function into the hash table.  */
   hashval_t hash = instance.hash ();
@@ -1047,16 +1064,14 @@ function_builder::add_unique_function (const function_instance &instance,
 
   /* Also add the function under its overloaded alias, if we want
      a separate decl for each instance of an overloaded function.  */
-  if (m_direct_overloads || force_direct_overloads)
+  char *overload_name = get_name (instance, true);
+  if (strcmp (name, overload_name) != 0)
     {
-      char *overload_name = get_name (instance, true);
-      if (strcmp (name, overload_name) != 0)
-	{
-	  /* Attribute lists shouldn't be shared.  */
-	  tree attrs = get_attributes (instance);
-	  add_function (instance, overload_name, fntype, attrs,
-			required_extensions, false);
-	}
+      /* Attribute lists shouldn't be shared.  */
+      tree attrs = get_attributes (instance);
+      bool placeholder_p = !(m_direct_overloads || force_direct_overloads);
+      add_function (instance, overload_name, fntype, attrs,
+		    required_extensions, false, placeholder_p);
     }
 
   obstack_free (&m_string_obstack, name);
@@ -1077,18 +1092,19 @@ function_builder::add_overloaded_function (const function_instance &instance,
 {
   char *name = get_name (instance, true);
   if (registered_function **map_value = m_overload_names.get (name))
-    gcc_assert ((*map_value)->instance == instance
-		&& ((*map_value)->required_extensions
-		    & ~required_extensions) == 0);
+    {
+      gcc_assert ((*map_value)->instance == instance
+		  && ((*map_value)->required_extensions
+		      & ~required_extensions) == 0);
+      obstack_free (&m_string_obstack, name);
+    }
   else
     {
       registered_function &rfn
 	= add_function (instance, name, m_overload_type, NULL_TREE,
-			required_extensions, true);
-      const char *permanent_name = IDENTIFIER_POINTER (DECL_NAME (rfn.decl));
-      m_overload_names.put (permanent_name, &rfn);
+			required_extensions, true, m_direct_overloads);
+      m_overload_names.put (name, &rfn);
     }
-  obstack_free (&m_string_obstack, name);
 }
 
 /* If we are using manual overload resolution, add one function decl
@@ -1098,9 +1114,6 @@ void
 function_builder::add_overloaded_functions (const function_group_info &group,
 					    mode_suffix_index mode)
 {
-  if (m_direct_overloads)
-    return;
-
   unsigned int explicit_type0 = (*group.shape)->explicit_type_suffix_p (0);
   unsigned int explicit_type1 = (*group.shape)->explicit_type_suffix_p (1);
   for (unsigned int pi = 0; group.preds[pi] != NUM_PREDS; ++pi)
diff --git a/gcc/config/aarch64/aarch64-sve-builtins.h b/gcc/config/aarch64/aarch64-sve-builtins.h
index 620e18841b2..b701f90ac1a 100644
--- a/gcc/config/aarch64/aarch64-sve-builtins.h
+++ b/gcc/config/aarch64/aarch64-sve-builtins.h
@@ -337,7 +337,8 @@ private:
   tree get_attributes (const function_instance &);
 
   registered_function &add_function (const function_instance &,
-				     const char *, tree, tree, uint64_t, bool);
+				     const char *, tree, tree,
+				     uint64_t, bool, bool);
 
   /* The function type to use for functions that are resolved by
      function_resolver.  */
diff --git a/gcc/testsuite/g++.target/aarch64/sve/pr99216.C b/gcc/testsuite/g++.target/aarch64/sve/pr99216.C
new file mode 100644
index 00000000000..61a58a7fcf4
--- /dev/null
+++ b/gcc/testsuite/g++.target/aarch64/sve/pr99216.C
@@ -0,0 +1,5 @@
+/* { dg-do link { target aarch64_asm_sve_ok } } */
+/* { dg-additional-options "-flto" } */
+#include <arm_sve.h>
+bool a;
+int main() { a = svaddv(svptrue_b8(), svdup_s8(0)); }

Re: [PATCH v2] aarch64: Fix SVE ACLE builtins with LTO [PR99216]

[Alex Coplan]

On 09/03/2021 15:41, Alex Coplan via Gcc-patches wrote:
> Hi all,
> 
> Here is a v2 patch which provides a more obviously fake answer to
> TARGET_BUILTIN_DECL: this can hopefully go in for GCC 11. For GCC 12, it
> seems that we should consider removing the target hook.
> 
> Original patch:
> https://gcc.gnu.org/pipermail/gcc-patches/2021-March/566405.html

Ping. OK for trunk?

> 
> ---
> 
> As discussed in the PR, we currently have two different numbering
> schemes for SVE builtins: one for C, and one for C++. This is
> problematic for LTO, where we end up getting confused about which
> intrinsic we're talking about. This patch inserts placeholders into the
> registered_functions vector to ensure that there is a consistent
> numbering scheme for both C and C++.
> 
> This patch uses integer_zero_node as a placeholder node instead of
> building a function decl. This is safe because the node is only returned
> by the TARGET_BUILTIN_DECL hook, which (on AArch64) is only used for
> validation when builtin decls are streamed into lto1.
> 
> Bootstrapped and regtested on aarch64-linux-gnu, OK for trunk?
> 
> Thanks,
> Alex
> 
> gcc/ChangeLog:
> 
> 	PR target/99216
> 	* config/aarch64/aarch64-sve-builtins.cc
> 	(function_builder::add_function): Add placeholder_p argument, use
> 	placeholder decls if this is set.
> 	(function_builder::add_unique_function): Instead of conditionally adding
> 	direct overloads, unconditionally add either a direct overload or a
> 	placeholder.
> 	(function_builder::add_overloaded_function): Set placeholder_p if we're
> 	using C++ overloads. Use the obstack for string storage instead
> 	of relying on the tree nodes.
> 	(function_builder::add_overloaded_functions): Don't return early for
> 	m_direct_overloads: we need to add placeholders.
> 	* config/aarch64/aarch64-sve-builtins.h
> 	(function_builder::add_function): Add placeholder_p argument.
> 
> gcc/testsuite/ChangeLog:
> 
> 	PR target/99216
> 	* g++.target/aarch64/sve/pr99216.C: New test.

> diff --git a/gcc/config/aarch64/aarch64-sve-builtins.cc b/gcc/config/aarch64/aarch64-sve-builtins.cc
> index 25612d2ea52..f44f81f1375 100644
> --- a/gcc/config/aarch64/aarch64-sve-builtins.cc
> +++ b/gcc/config/aarch64/aarch64-sve-builtins.cc
> @@ -999,12 +999,29 @@ registered_function &
>  function_builder::add_function (const function_instance &instance,
>  				const char *name, tree fntype, tree attrs,
>  				uint64_t required_extensions,
> -				bool overloaded_p)
> +				bool overloaded_p,
> +				bool placeholder_p)
>  {
>    unsigned int code = vec_safe_length (registered_functions);
>    code = (code << AARCH64_BUILTIN_SHIFT) | AARCH64_BUILTIN_SVE;
> -  tree decl = simulate_builtin_function_decl (input_location, name, fntype,
> -					      code, NULL, attrs);
> +
> +  /* We need to be able to generate placeholders to enusre that we have a
> +     consistent numbering scheme for function codes between the C and C++
> +     frontends, so that everything ties up in LTO.
> +
> +     Currently, tree-streamer-in.c:unpack_ts_function_decl_value_fields
> +     validates that tree nodes returned by TARGET_BUILTIN_DECL are non-NULL and
> +     some node other than error_mark_node. This is a holdover from when builtin
> +     decls were streamed by code rather than by value.
> +
> +     Ultimately, we should be able to remove this validation of BUILT_IN_MD
> +     nodes and remove the target hook. For now, however, we need to appease the
> +     validation and return a non-NULL, non-error_mark_node node, so we
> +     arbitrarily choose integer_zero_node.  */
> +  tree decl = placeholder_p
> +    ? integer_zero_node
> +    : simulate_builtin_function_decl (input_location, name, fntype,
> +				      code, NULL, attrs);
>  
>    registered_function &rfn = *ggc_alloc <registered_function> ();
>    rfn.instance = instance;
> @@ -1036,7 +1053,7 @@ function_builder::add_unique_function (const function_instance &instance,
>  					   argument_types.address ());
>    tree attrs = get_attributes (instance);
>    registered_function &rfn = add_function (instance, name, fntype, attrs,
> -					   required_extensions, false);
> +					   required_extensions, false, false);
>  
>    /* Enter the function into the hash table.  */
>    hashval_t hash = instance.hash ();
> @@ -1047,16 +1064,14 @@ function_builder::add_unique_function (const function_instance &instance,
>  
>    /* Also add the function under its overloaded alias, if we want
>       a separate decl for each instance of an overloaded function.  */
> -  if (m_direct_overloads || force_direct_overloads)
> +  char *overload_name = get_name (instance, true);
> +  if (strcmp (name, overload_name) != 0)
>      {
> -      char *overload_name = get_name (instance, true);
> -      if (strcmp (name, overload_name) != 0)
> -	{
> -	  /* Attribute lists shouldn't be shared.  */
> -	  tree attrs = get_attributes (instance);
> -	  add_function (instance, overload_name, fntype, attrs,
> -			required_extensions, false);
> -	}
> +      /* Attribute lists shouldn't be shared.  */
> +      tree attrs = get_attributes (instance);
> +      bool placeholder_p = !(m_direct_overloads || force_direct_overloads);
> +      add_function (instance, overload_name, fntype, attrs,
> +		    required_extensions, false, placeholder_p);
>      }
>  
>    obstack_free (&m_string_obstack, name);
> @@ -1077,18 +1092,19 @@ function_builder::add_overloaded_function (const function_instance &instance,
>  {
>    char *name = get_name (instance, true);
>    if (registered_function **map_value = m_overload_names.get (name))
> -    gcc_assert ((*map_value)->instance == instance
> -		&& ((*map_value)->required_extensions
> -		    & ~required_extensions) == 0);
> +    {
> +      gcc_assert ((*map_value)->instance == instance
> +		  && ((*map_value)->required_extensions
> +		      & ~required_extensions) == 0);
> +      obstack_free (&m_string_obstack, name);
> +    }
>    else
>      {
>        registered_function &rfn
>  	= add_function (instance, name, m_overload_type, NULL_TREE,
> -			required_extensions, true);
> -      const char *permanent_name = IDENTIFIER_POINTER (DECL_NAME (rfn.decl));
> -      m_overload_names.put (permanent_name, &rfn);
> +			required_extensions, true, m_direct_overloads);
> +      m_overload_names.put (name, &rfn);
>      }
> -  obstack_free (&m_string_obstack, name);
>  }
>  
>  /* If we are using manual overload resolution, add one function decl
> @@ -1098,9 +1114,6 @@ void
>  function_builder::add_overloaded_functions (const function_group_info &group,
>  					    mode_suffix_index mode)
>  {
> -  if (m_direct_overloads)
> -    return;
> -
>    unsigned int explicit_type0 = (*group.shape)->explicit_type_suffix_p (0);
>    unsigned int explicit_type1 = (*group.shape)->explicit_type_suffix_p (1);
>    for (unsigned int pi = 0; group.preds[pi] != NUM_PREDS; ++pi)
> diff --git a/gcc/config/aarch64/aarch64-sve-builtins.h b/gcc/config/aarch64/aarch64-sve-builtins.h
> index 620e18841b2..b701f90ac1a 100644
> --- a/gcc/config/aarch64/aarch64-sve-builtins.h
> +++ b/gcc/config/aarch64/aarch64-sve-builtins.h
> @@ -337,7 +337,8 @@ private:
>    tree get_attributes (const function_instance &);
>  
>    registered_function &add_function (const function_instance &,
> -				     const char *, tree, tree, uint64_t, bool);
> +				     const char *, tree, tree,
> +				     uint64_t, bool, bool);
>  
>    /* The function type to use for functions that are resolved by
>       function_resolver.  */
> diff --git a/gcc/testsuite/g++.target/aarch64/sve/pr99216.C b/gcc/testsuite/g++.target/aarch64/sve/pr99216.C
> new file mode 100644
> index 00000000000..61a58a7fcf4
> --- /dev/null
> +++ b/gcc/testsuite/g++.target/aarch64/sve/pr99216.C
> @@ -0,0 +1,5 @@
> +/* { dg-do link { target aarch64_asm_sve_ok } } */
> +/* { dg-additional-options "-flto" } */
> +#include <arm_sve.h>
> +bool a;
> +int main() { a = svaddv(svptrue_b8(), svdup_s8(0)); }


-- 
Alex

Re: [PATCH v2] aarch64: Fix SVE ACLE builtins with LTO [PR99216]

[Richard Sandiford]

Alex Coplan <alex.coplan@arm.com> writes:
> Hi all,
>
> Here is a v2 patch which provides a more obviously fake answer to
> TARGET_BUILTIN_DECL: this can hopefully go in for GCC 11. For GCC 12, it
> seems that we should consider removing the target hook.
>
> Original patch:
> https://gcc.gnu.org/pipermail/gcc-patches/2021-March/566405.html
>
> ---
>
> As discussed in the PR, we currently have two different numbering
> schemes for SVE builtins: one for C, and one for C++. This is
> problematic for LTO, where we end up getting confused about which
> intrinsic we're talking about. This patch inserts placeholders into the
> registered_functions vector to ensure that there is a consistent
> numbering scheme for both C and C++.
>
> This patch uses integer_zero_node as a placeholder node instead of
> building a function decl. This is safe because the node is only returned
> by the TARGET_BUILTIN_DECL hook, which (on AArch64) is only used for
> validation when builtin decls are streamed into lto1.
>
> Bootstrapped and regtested on aarch64-linux-gnu, OK for trunk?
>
> Thanks,
> Alex
>
> gcc/ChangeLog:
>
> 	PR target/99216
> 	* config/aarch64/aarch64-sve-builtins.cc
> 	(function_builder::add_function): Add placeholder_p argument, use
> 	placeholder decls if this is set.
> 	(function_builder::add_unique_function): Instead of conditionally adding
> 	direct overloads, unconditionally add either a direct overload or a
> 	placeholder.
> 	(function_builder::add_overloaded_function): Set placeholder_p if we're
> 	using C++ overloads. Use the obstack for string storage instead
> 	of relying on the tree nodes.
> 	(function_builder::add_overloaded_functions): Don't return early for
> 	m_direct_overloads: we need to add placeholders.
> 	* config/aarch64/aarch64-sve-builtins.h
> 	(function_builder::add_function): Add placeholder_p argument.
>
> gcc/testsuite/ChangeLog:
>
> 	PR target/99216
> 	* g++.target/aarch64/sve/pr99216.C: New test.

OK, thanks, and sorry for the delay in reviewing.

Richard