[Bug fortran/118896] The fortran compiler is unable to devirtualize typebound indirect calls

[mikael at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118896

--- Comment #2 from Mikael Morin <mikael at gcc dot gnu.org> ---
This area of the code has an interesting history.

TREE_READONLY was introduced for virtual tables with r0-114543 together with
the fix for PR51809

diff --git a/gcc/fortran/trans-decl.c b/gcc/fortran/trans-decl.c
index
0761ebb26d1d438c1e3b8570fdf5ea00b9744453..e8e54c77b79c5b94c664977a62a7b8500fb6bff0
100644 (file)
--- a/gcc/fortran/trans-decl.c
+++ b/gcc/fortran/trans-decl.c
@@ -1485,7 +1485,10 @@ gfc_get_symbol_decl (gfc_symbol * sym)

   if (sym->attr.vtab
       || (sym->name[0] == '_' && strncmp ("__def_init", sym->name, 10) == 0))
-    GFC_DECL_PUSH_TOPLEVEL (decl) = 1;
+    {
+      TREE_READONLY (decl) = 1;
+      GFC_DECL_PUSH_TOPLEVEL (decl) = 1;
+    }

   return decl;
 }


Then TREE_READONLY was removed in favor of DECL_ARTIFICIAL in r10-3606 to fix
PR84487

diff --git a/gcc/fortran/trans-decl.c b/gcc/fortran/trans-decl.c
index
a113f08e26be87be7783c802c5b1129d64548801..b701f4934401a00341178bba846566ea790c6eac
100644 (file)
--- a/gcc/fortran/trans-decl.c
+++ b/gcc/fortran/trans-decl.c
@@ -1911,9 +1911,13 @@ gfc_get_symbol_decl (gfc_symbol * sym)
   if (sym->attr.associate_var)
     GFC_DECL_ASSOCIATE_VAR_P (decl) = 1;

+  /* We no longer mark __def_init as read-only so it does not take up
+     space in the read-only section and dan go into the BSS instead,
+     see PR 84487.  Marking this as artificial means that OpenMP will
+     treat this as predetermined shared.  */
   if (sym->attr.vtab
       || (sym->name[0] == '_' && gfc_str_startswith (sym->name,
"__def_init")))
-    TREE_READONLY (decl) = 1;
+    DECL_ARTIFICIAL (decl) = 1;

   return decl;
 }


And finally TREE_READONLY was restored for __def_init variables with
initializer with r10-4319 to fix PR92113.

diff --git a/gcc/fortran/trans-decl.c b/gcc/fortran/trans-decl.c
index
630682cbac7210e25b5254a3792f17712baec053..ffa61111316c29d423e40e9f48c2e7e0b3d698a5
100644 (file)
--- a/gcc/fortran/trans-decl.c
+++ b/gcc/fortran/trans-decl.c
@@ -1904,13 +1904,20 @@ gfc_get_symbol_decl (gfc_symbol * sym)
   if (sym->attr.associate_var)
     GFC_DECL_ASSOCIATE_VAR_P (decl) = 1;

-  /* We no longer mark __def_init as read-only so it does not take up
-     space in the read-only section and dan go into the BSS instead,
-     see PR 84487.  Marking this as artificial means that OpenMP will
-     treat this as predetermined shared.  */
-  if (sym->attr.vtab
-      || (sym->name[0] == '_' && gfc_str_startswith (sym->name,
"__def_init")))
-    DECL_ARTIFICIAL (decl) = 1;
+  /* We only longer mark __def_init as read-only if it actually has an
+     initializer, it does not needlessly take up space in the
+     read-only section and can go into the BSS instead, see PR 84487.
+     Marking this as artificial means that OpenMP will treat this as
+     predetermined shared.  */
+
+  bool def_init = gfc_str_startswith (sym->name, "__def_init");
+
+  if (sym->attr.vtab || def_init)
+    {
+      DECL_ARTIFICIAL (decl) = 1;
+      if (def_init && sym->value)
+       TREE_READONLY (decl) = 1;
+    }

   return decl;
 }


But the latter patch has a different condition guarding TREE_READONLY from the
one that was submitted to the list (
https://gcc.gnu.org/pipermail/fortran/2019-November/053368.html ) and used for
backports ( r9-7960 and r8-9817 ).
The condition in the submitted patch is: 
  if (sym->attr.vtab || sym->value)
which is equivalent to comment #1.