[PATCH] cgraphclones: Don't share DECL_ARGUMENTS between thunk and its artificial thunk [PR108854]

[PATCH] cgraphclones: Don't share DECL_ARGUMENTS between thunk and its artificial thunk [PR108854]

[Jakub Jelinek]

Hi!

The following testcase ICEs on x86_64-linux with -m32.  The problem is
we create an artificial thunk and because of -fPIC, ia32 and thunk
destination which doesn't bind locally can't use a mi thunk.
The ICE is because during expansion to RTL we see SSA_NAME for a PARM_DECL,
but the PARM_DECL doesn't have DECL_CONTEXT of the current function.
This is because duplicate_thunk_for_node creates a new DECL_ARGUMENTS chain
only if some arguments need modification.

The following patch fixes it by copying the DECL_ARGUMENTS list even if
the arguments can stay as is, to update DECL_CONTEXT on them.  While for
mi thunks it doesn't really matter because we don't use those arguments
in any way, for other thunks it is important.

Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

2023-02-23  Jakub Jelinek  <jakub@redhat.com>

	PR middle-end/108854
	* cgraphclones.cc (duplicate_thunk_for_node): If no parameter
	changes are needed, copy at least DECL_ARGUMENTS PARM_DECL
	nodes and adjust their DECL_CONTEXT.

	* g++.dg/opt/pr108854.C: New test.

--- gcc/cgraphclones.cc.jj	2023-02-22 20:50:27.417519830 +0100
+++ gcc/cgraphclones.cc	2023-02-23 17:12:59.875133883 +0100
@@ -218,7 +218,17 @@ duplicate_thunk_for_node (cgraph_node *t
       body_adj.modify_formal_parameters ();
     }
   else
-    new_decl = copy_node (thunk->decl);
+    {
+      new_decl = copy_node (thunk->decl);
+      for (tree *arg = &DECL_ARGUMENTS (new_decl);
+	   *arg; arg = &DECL_CHAIN (*arg))
+	{
+	  tree next = DECL_CHAIN (*arg);
+	  *arg = copy_node (*arg);
+	  DECL_CONTEXT (*arg) = new_decl;
+	  DECL_CHAIN (*arg) = next;
+	}
+    }
 
   gcc_checking_assert (!DECL_STRUCT_FUNCTION (new_decl));
   gcc_checking_assert (!DECL_INITIAL (new_decl));
--- gcc/testsuite/g++.dg/opt/pr108854.C.jj	2023-02-23 17:11:19.275583506 +0100
+++ gcc/testsuite/g++.dg/opt/pr108854.C	2023-02-23 17:11:02.723822009 +0100
@@ -0,0 +1,37 @@
+// PR middle-end/108854
+// { dg-do compile { target c++11 } }
+// { dg-options "-O3" }
+// { dg-additional-options "-fPIC" { target fpic } }
+
+struct A { A (int); ~A (); };
+struct B { B (int, bool); ~B (); };
+template <typename T>
+struct C { void m1 (T); void m2 (T &&); };
+class D;
+struct E { virtual void m3 (); };
+template <typename>
+struct F { virtual bool m4 (D &); };
+struct D { virtual D m5 () { return D (); } };
+void foo (void *, void *);
+struct G {
+  int a;
+  C <D *> b;
+  void m4 (D &r) { B l (a, true); r.m5 (); b.m1 (&r); b.m2 (&r); }
+};
+struct H : E, F <int> {
+  template <typename T>
+  H (int, T);
+  bool m4 (D &r) { A l (a); b.m4 (r); if (c) return true; } // { dg-warning "control reaches end of non-void function" }
+  int a;
+  bool c;
+  G b;
+};
+inline void bar (F <int> &p) { D s, t; p.m4 (t); foo (&p, &s); }
+enum I { I1, I2 };
+template <I>
+struct J;
+template <class, class T, class, class, class, class>
+void baz () { int g = 0, h = 0; T i (g, h); bar (i); }
+template <class, int, I T>
+void qux () { baz <int, H, int, int, E, J<T>> (); }
+void corge () { qux <int, I2, I1> (); qux <int, I2, I2> (); }

	Jakub

Re: [PATCH] cgraphclones: Don't share DECL_ARGUMENTS between thunk and its artificial thunk [PR108854]

[Richard Biener]

On Fri, 24 Feb 2023, Jakub Jelinek wrote:

> Hi!
> 
> The following testcase ICEs on x86_64-linux with -m32.  The problem is
> we create an artificial thunk and because of -fPIC, ia32 and thunk
> destination which doesn't bind locally can't use a mi thunk.
> The ICE is because during expansion to RTL we see SSA_NAME for a PARM_DECL,
> but the PARM_DECL doesn't have DECL_CONTEXT of the current function.
> This is because duplicate_thunk_for_node creates a new DECL_ARGUMENTS chain
> only if some arguments need modification.
> 
> The following patch fixes it by copying the DECL_ARGUMENTS list even if
> the arguments can stay as is, to update DECL_CONTEXT on them.  While for
> mi thunks it doesn't really matter because we don't use those arguments
> in any way, for other thunks it is important.
> 
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

OK.

> 2023-02-23  Jakub Jelinek  <jakub@redhat.com>
> 
> 	PR middle-end/108854
> 	* cgraphclones.cc (duplicate_thunk_for_node): If no parameter
> 	changes are needed, copy at least DECL_ARGUMENTS PARM_DECL
> 	nodes and adjust their DECL_CONTEXT.
> 
> 	* g++.dg/opt/pr108854.C: New test.
> 
> --- gcc/cgraphclones.cc.jj	2023-02-22 20:50:27.417519830 +0100
> +++ gcc/cgraphclones.cc	2023-02-23 17:12:59.875133883 +0100
> @@ -218,7 +218,17 @@ duplicate_thunk_for_node (cgraph_node *t
>        body_adj.modify_formal_parameters ();
>      }
>    else
> -    new_decl = copy_node (thunk->decl);
> +    {
> +      new_decl = copy_node (thunk->decl);
> +      for (tree *arg = &DECL_ARGUMENTS (new_decl);
> +	   *arg; arg = &DECL_CHAIN (*arg))
> +	{
> +	  tree next = DECL_CHAIN (*arg);
> +	  *arg = copy_node (*arg);
> +	  DECL_CONTEXT (*arg) = new_decl;
> +	  DECL_CHAIN (*arg) = next;
> +	}
> +    }
>  
>    gcc_checking_assert (!DECL_STRUCT_FUNCTION (new_decl));
>    gcc_checking_assert (!DECL_INITIAL (new_decl));
> --- gcc/testsuite/g++.dg/opt/pr108854.C.jj	2023-02-23 17:11:19.275583506 +0100
> +++ gcc/testsuite/g++.dg/opt/pr108854.C	2023-02-23 17:11:02.723822009 +0100
> @@ -0,0 +1,37 @@
> +// PR middle-end/108854
> +// { dg-do compile { target c++11 } }
> +// { dg-options "-O3" }
> +// { dg-additional-options "-fPIC" { target fpic } }
> +
> +struct A { A (int); ~A (); };
> +struct B { B (int, bool); ~B (); };
> +template <typename T>
> +struct C { void m1 (T); void m2 (T &&); };
> +class D;
> +struct E { virtual void m3 (); };
> +template <typename>
> +struct F { virtual bool m4 (D &); };
> +struct D { virtual D m5 () { return D (); } };
> +void foo (void *, void *);
> +struct G {
> +  int a;
> +  C <D *> b;
> +  void m4 (D &r) { B l (a, true); r.m5 (); b.m1 (&r); b.m2 (&r); }
> +};
> +struct H : E, F <int> {
> +  template <typename T>
> +  H (int, T);
> +  bool m4 (D &r) { A l (a); b.m4 (r); if (c) return true; } // { dg-warning "control reaches end of non-void function" }
> +  int a;
> +  bool c;
> +  G b;
> +};
> +inline void bar (F <int> &p) { D s, t; p.m4 (t); foo (&p, &s); }
> +enum I { I1, I2 };
> +template <I>
> +struct J;
> +template <class, class T, class, class, class, class>
> +void baz () { int g = 0, h = 0; T i (g, h); bar (i); }
> +template <class, int, I T>
> +void qux () { baz <int, H, int, int, E, J<T>> (); }
> +void corge () { qux <int, I2, I1> (); qux <int, I2, I2> (); }
> 
> 	Jakub
> 
> 

-- 
Richard Biener <rguenther@suse.de>
SUSE Software Solutions Germany GmbH, Frankenstrasse 146, 90461 Nuernberg,
Germany; GF: Ivo Totev, Andrew Myers, Andrew McDonald, Boudien Moerman;
HRB 36809 (AG Nuernberg)

Re: [PATCH] cgraphclones: Don't share DECL_ARGUMENTS between thunk and its artificial thunk [PR108854]

[Jan Hubicka]

> Hi!
> 
> The following testcase ICEs on x86_64-linux with -m32.  The problem is
> we create an artificial thunk and because of -fPIC, ia32 and thunk
> destination which doesn't bind locally can't use a mi thunk.
> The ICE is because during expansion to RTL we see SSA_NAME for a PARM_DECL,
> but the PARM_DECL doesn't have DECL_CONTEXT of the current function.
> This is because duplicate_thunk_for_node creates a new DECL_ARGUMENTS chain
> only if some arguments need modification.
> 
> The following patch fixes it by copying the DECL_ARGUMENTS list even if
> the arguments can stay as is, to update DECL_CONTEXT on them.  While for
> mi thunks it doesn't really matter because we don't use those arguments
> in any way, for other thunks it is important.
> 
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
> 
> 2023-02-23  Jakub Jelinek  <jakub@redhat.com>
> 
> 	PR middle-end/108854
> 	* cgraphclones.cc (duplicate_thunk_for_node): If no parameter
> 	changes are needed, copy at least DECL_ARGUMENTS PARM_DECL
> 	nodes and adjust their DECL_CONTEXT.
> 
> 	* g++.dg/opt/pr108854.C: New test.
> 
> --- gcc/cgraphclones.cc.jj	2023-02-22 20:50:27.417519830 +0100
> +++ gcc/cgraphclones.cc	2023-02-23 17:12:59.875133883 +0100
> @@ -218,7 +218,17 @@ duplicate_thunk_for_node (cgraph_node *t
>        body_adj.modify_formal_parameters ();
>      }
>    else
> -    new_decl = copy_node (thunk->decl);
> +    {
> +      new_decl = copy_node (thunk->decl);
> +      for (tree *arg = &DECL_ARGUMENTS (new_decl);
> +	   *arg; arg = &DECL_CHAIN (*arg))
> +	{
> +	  tree next = DECL_CHAIN (*arg);
> +	  *arg = copy_node (*arg);
> +	  DECL_CONTEXT (*arg) = new_decl;
> +	  DECL_CHAIN (*arg) = next;

This makes sense to me. I wonder if we don't want to update abstract
origin too like we do in tree-inline?
Maybe it is unecessary since we don't do debug info for thunks....

Jan