Use __builtin_atomic_thread_fence and implement SYNC_MEMORY.

gcc/fortran/ChangeLog:

	* trans.c (gfc_trans_memory_barrier_fence): New
	function.
	* trans.h (gfc_trans_memory_barrier_fence): Prototype it.
	* trans-stmt.c (gfc_trans_sync): For shared coarrays,
	use memory fence. Don't do anything else for SYNC MEMORY.

gcc/testsuite/ChangeLog:

	* gfortran.dg/caf-shared/sync_all_1.f90: New test.
	* gfortran.dg/caf-shared/sync_memory_1.f90: New test.

5 files changed, 50 insertions, 5 deletions

diff --git a/gcc/fortran/trans-stmt.c b/gcc/fortran/trans-stmt.c
index 368165e..8db4333 100644
--- a/gcc/fortran/trans-stmt.c
+++ b/gcc/fortran/trans-stmt.c
@@ -1307,11 +1307,16 @@ gfc_trans_sync (gfc_code *code, gfc_exec_op type)
 
   /* Per F2008, 8.5.1, a SYNC MEMORY is implied by calling the
      image control statements SYNC IMAGES and SYNC ALL.  */
-  if (flag_coarray == GFC_FCOARRAY_LIB || flag_coarray == GFC_FCOARRAY_SHARED)
+  if (flag_coarray == GFC_FCOARRAY_LIB)
     {
       tmp = gfc_trans_memory_barrier ();
       gfc_add_expr_to_block (&se.pre, tmp);
     }
+  else if (flag_coarray == GFC_FCOARRAY_SHARED)
+    {
+      tmp = gfc_trans_memory_barrier_fence ();
+      gfc_add_expr_to_block (&se.pre, tmp);
+    }
 
   if (flag_coarray != GFC_FCOARRAY_LIB && flag_coarray != GFC_FCOARRAY_SHARED)
     {
@@ -1332,8 +1337,16 @@ gfc_trans_sync (gfc_code *code, gfc_exec_op type)
 	    stat = gfc_build_addr_expr (NULL, stat);
 
 	  if(type == EXEC_SYNC_MEMORY)
-	    tmp = build_call_expr_loc (input_location, gfor_fndecl_caf_sync_memory,
-				       3, stat, errmsg, errmsglen);
+	    {
+	      /* For shared coarrays, there is no need for a memory
+		 fence here because that is emitted anyway below.  */
+	      if (flag_coarray != GFC_FCOARRAY_SHARED)
+		tmp = build_call_expr_loc (input_location,
+					   gfor_fndecl_caf_sync_memory,
+					   3, stat, errmsg, errmsglen);
+	      else
+		tmp = NULL_TREE;
+	    }
 	  else
 	    {
 	      if (flag_coarray == GFC_FCOARRAY_LIB)
@@ -1343,8 +1356,8 @@ gfc_trans_sync (gfc_code *code, gfc_exec_op type)
 		tmp = build_call_expr_loc (input_location, gfor_fndecl_cas_sync_all,
 					   1, stat);
 	    }
-
-	  gfc_add_expr_to_block (&se.pre, tmp);
+	  if (tmp != NULL_TREE)
+	    gfc_add_expr_to_block (&se.pre, tmp);
 	}
       else
 	{
diff --git a/gcc/fortran/trans.c b/gcc/fortran/trans.c
index aa30710..fa7fd9d 100644
--- a/gcc/fortran/trans.c
+++ b/gcc/fortran/trans.c
@@ -34,6 +34,7 @@ along with GCC; see the file COPYING3.  If not see
 #include "trans-types.h"
 #include "trans-const.h"
 #include "diagnostic-core.h"
+#include "memmodel.h"   /* For MEMMODEL_ enums.  */
 
 /* Naming convention for backend interface code:
 
@@ -49,6 +50,7 @@ const char gfc_msg_fault[] = N_("Array reference out of bounds");
 const char gfc_msg_wrong_return[] = N_("Incorrect function return value");
 
 /* Insert a memory barrier into the code.  */
+
 tree
 gfc_trans_memory_barrier (void)
 {
@@ -63,6 +65,20 @@ gfc_trans_memory_barrier (void)
   return tmp;
 }
 
+/* Same as above, but do it by calling
+   __builtin_atomic_thread_fence.  */
+
+tree
+gfc_trans_memory_barrier_fence (void)
+{
+  tree call, mode;
+  call = builtin_decl_explicit (BUILT_IN_ATOMIC_THREAD_FENCE);
+  mode = build_int_cst (integer_type_node, MEMMODEL_ACQ_REL);
+  call = build_call_expr_loc (input_location, call, 1, mode);
+  return call;
+}
+
+
 /* Return a location_t suitable for 'tree' for a gfortran locus.  The way the
    parser works in gfortran, loc->lb->location contains only the line number
    and LOCATION_COLUMN is 0; hence, the column has to be added when generating
diff --git a/gcc/fortran/trans.h b/gcc/fortran/trans.h
index eef90f6..95e4741 100644
--- a/gcc/fortran/trans.h
+++ b/gcc/fortran/trans.h
@@ -504,6 +504,7 @@ void gfc_conv_expr_type (gfc_se * se, gfc_expr *, tree);
 /* Insert a memory barrier into the code.  */
 
 tree gfc_trans_memory_barrier (void);
+tree gfc_trans_memory_barrier_fence (void);
 
 /* trans-expr.c */
 tree gfc_conv_scalar_to_descriptor (gfc_se *, tree, symbol_attribute);
diff --git a/gcc/testsuite/gfortran.dg/caf-shared/sync_all_1.f90 b/gcc/testsuite/gfortran.dg/caf-shared/sync_all_1.f90
new file mode 100644
index 0000000..95343f5
--- /dev/null
+++ b/gcc/testsuite/gfortran.dg/caf-shared/sync_all_1.f90
@@ -0,0 +1,9 @@
+! { dg-do compile }
+! { dg-options "-fdump-tree-original" }
+program main
+  sync all
+end program main
+! { dg-final { scan-tree-dump-times "__atomic_thread_fence \\(4\\)" 1 "original" } }
+! { dg-final { scan-tree-dump-times "_gfortran_cas_coarray_sync_all" 1 "original" } }
+
+
diff --git a/gcc/testsuite/gfortran.dg/caf-shared/sync_memory_1.f90 b/gcc/testsuite/gfortran.dg/caf-shared/sync_memory_1.f90
new file mode 100644
index 0000000..25a7dcc
--- /dev/null
+++ b/gcc/testsuite/gfortran.dg/caf-shared/sync_memory_1.f90
@@ -0,0 +1,6 @@
+! { dg-do compile }
+! { dg-options "-fdump-tree-original" }
+program main
+  sync memory
+end program main
+! { dg-final { scan-tree-dump-times "__atomic_thread_fence \\(4\\)" 1 "original" } }