nvptx.h (PARM_BOUNDARY): Set to 32.

	gcc/
	* config/nvptx/nvptx.h (PARM_BOUNDARY): Set to 32.
	* config/nvptx/nvptx.c (PASS_IN_REG_P, RETURN_IN_REG_P): Delete.
	(pass_in_memory, promote_arg, promote_return): New.
	(nvptx_function_arg_boundary): Delete.
	(nvptx_function_value): Use promote_return.
	(nvptx_pass_by_reference): Use pass_in_memory.
	(nvptx_return_in_memory): Use pass_in_memory.
	(nvptx_promote_function_mode): Use promote_arg.
	(write_arg): Adjust arg splitting logic.
	(write_return): Check and clear ret_reg_mode, if needed.
	(write_fn_proto, nvptx_declare_function_name): Adust write_return
	calls.
	(TARGET_RUNCTION_ARG_BOUNDARY,
	TARGET_FUNCTION_ARG_ROUND_BOUNDARY): Don't override.

	gcc/testsuite/
	* g++.dg/abi/nvptx-nrv1.C: New.
	* g++.dg/abi/nvptx-ptrmem1.C: New.
	* gcc.target/nvptx/abi-complex-arg.c: New.
	* gcc.target/nvptx/abi-complex-ret.c: New.
	* gcc.target/nvptx/abi-enum-arg.c: New.
	* gcc.target/nvptx/abi-enum-ret.c: New.
	* gcc.target/nvptx/abi-knr-arg.c: New.
	* gcc.target/nvptx/abi-knr-ret.c: New.
	* gcc.target/nvptx/abi-scalar-arg.c: New.
	* gcc.target/nvptx/abi-scalar-ret.c: New.
	* gcc.target/nvptx/abi-struct-arg.c: New.
	* gcc.target/nvptx/abi-struct-ret.c: New.
	* gcc.target/nvptx/abi-vararg-1.c: New.
	* gcc.target/nvptx/abi-vararg-2.c: New.
	* gcc.target/nvptx/abi-vect-arg.c: New.
	* gcc.target/nvptx/abi-vect-ret.c: New.

From-SVN: r231628

20 files changed, 1718 insertions, 113 deletions

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 31c79d8..e1fcab6 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,20 @@
+2015-12-14  Nathan Sidwell  <nathan@acm.org>
+
+	* config/nvptx/nvptx.h (PARM_BOUNDARY): Set to 32.
+	* config/nvptx/nvptx.c (PASS_IN_REG_P, RETURN_IN_REG_P): Delete.
+	(pass_in_memory, promote_arg, promote_return): New.
+	(nvptx_function_arg_boundary): Delete.
+	(nvptx_function_value): Use promote_return.
+	(nvptx_pass_by_reference): Use pass_in_memory.
+	(nvptx_return_in_memory): Use pass_in_memory.
+	(nvptx_promote_function_mode): Use promote_arg.
+	(write_arg): Adjust arg splitting logic.
+	(write_return): Check and clear ret_reg_mode, if needed.
+	(write_fn_proto, nvptx_declare_function_name): Adust write_return
+	calls.
+	(TARGET_RUNCTION_ARG_BOUNDARY,
+	TARGET_FUNCTION_ARG_ROUND_BOUNDARY): Don't override.
+
 2015-12-14  Jakub Jelinek  <jakub@redhat.com>
 
 	PR c/68833
diff --git a/gcc/config/nvptx/nvptx.c b/gcc/config/nvptx/nvptx.c
index 44b0c30..bb4c384 100644
--- a/gcc/config/nvptx/nvptx.c
+++ b/gcc/config/nvptx/nvptx.c
@@ -365,18 +365,6 @@ nvptx_emit_joining (unsigned mask, bool is_call)
     }
 }
 
-#define PASS_IN_REG_P(MODE, TYPE)				\
-  ((GET_MODE_CLASS (MODE) == MODE_INT				\
-    || GET_MODE_CLASS (MODE) == MODE_FLOAT			\
-    || ((GET_MODE_CLASS (MODE) == MODE_COMPLEX_INT		\
-	 || GET_MODE_CLASS (MODE) == MODE_COMPLEX_FLOAT)	\
-	&& !AGGREGATE_TYPE_P (TYPE)))				\
-   && (MODE) != TImode)
-
-#define RETURN_IN_REG_P(MODE)			\
-  ((GET_MODE_CLASS (MODE) == MODE_INT		\
-    || GET_MODE_CLASS (MODE) == MODE_FLOAT)	\
-   && GET_MODE_SIZE (MODE) <= 8)
 
 /* Perform a mode promotion for a function argument with MODE.  Return
    the promoted mode.  */
@@ -389,6 +377,61 @@ arg_promotion (machine_mode mode)
   return mode;
 }
 
+/* Determine whether MODE and TYPE (possibly NULL) should be passed or
+   returned in memory.  Integer and floating types supported by the
+   machine are passed in registers, everything else is passed in
+   memory.  Complex types are split.  */
+
+static bool
+pass_in_memory (machine_mode mode, const_tree type, bool for_return)
+{
+  if (type)
+    {
+      if (AGGREGATE_TYPE_P (type))
+	return true;
+      if (TREE_CODE (type) == VECTOR_TYPE)
+	return true;
+    }
+
+  if (!for_return && COMPLEX_MODE_P (mode))
+    /* Complex types are passed as two underlying args.  */
+    mode = GET_MODE_INNER (mode);
+
+  if (GET_MODE_CLASS (mode) != MODE_INT
+      && GET_MODE_CLASS (mode) != MODE_FLOAT)
+    return true;
+
+  if (GET_MODE_SIZE (mode) > UNITS_PER_WORD)
+    return true;
+
+  return false;
+}
+
+/* A non-memory argument of mode MODE is being passed, determine the mode it
+   should be promoted to.  This is also used for determining return
+   type promotion.  */
+
+static machine_mode
+promote_arg (machine_mode mode, bool prototyped)
+{
+  if (!prototyped && mode == SFmode)
+    /* K&R float promotion for unprototyped functions.  */
+    mode = DFmode;
+  else if (GET_MODE_SIZE (mode) < GET_MODE_SIZE (SImode))
+    mode = SImode;
+
+  return mode;
+}
+
+/* A non-memory return type of MODE is being returned.  Determine the
+   mode it should be promoted to.  */
+
+static machine_mode
+promote_return (machine_mode mode)
+{
+  return promote_arg (mode, true);
+}
+
 /* Implement TARGET_FUNCTION_ARG.  */
 
 static rtx
@@ -450,40 +493,6 @@ nvptx_strict_argument_naming (cumulative_args_t cum_v)
   return cum->fntype == NULL_TREE || stdarg_p (cum->fntype);
 }
 
-/* Implement TARGET_FUNCTION_ARG_BOUNDARY.  */
-
-static unsigned int
-nvptx_function_arg_boundary (machine_mode mode, const_tree type)
-{
-  unsigned int boundary = type ? TYPE_ALIGN (type) : GET_MODE_BITSIZE (mode);
-
-  if (boundary > UNITS_PER_WORD * BITS_PER_UNIT)
-    boundary = UNITS_PER_WORD * BITS_PER_UNIT;
-  else if (mode == BLKmode)
-    {
-      HOST_WIDE_INT size = int_size_in_bytes (type);
-
-      if (size > UNITS_PER_WORD)
-	boundary = UNITS_PER_WORD;
-      else
-	{
-	  /* Keep rounding up until only 1 bit set.  */
-	  unsigned lsb = (unsigned) size;
-
-	  boundary = 0;
-	  do
-	    {
-	      boundary += lsb;
-	      lsb = boundary & -boundary;
-	    }
-	  while (boundary != lsb);
-	}
-      boundary *= BITS_PER_UNIT;
-    }
-
-  return boundary;
-}
-
 /* Implement TARGET_LIBCALL_VALUE.  */
 
 static rtx
@@ -501,13 +510,11 @@ nvptx_libcall_value (machine_mode mode, const_rtx)
    where function FUNC returns or receives a value of data type TYPE.  */
 
 static rtx
-nvptx_function_value (const_tree type, const_tree func ATTRIBUTE_UNUSED,
+nvptx_function_value (const_tree type, const_tree ARG_UNUSED (func),
 		      bool outgoing)
 {
-  int unsignedp = TYPE_UNSIGNED (type);
-  machine_mode orig_mode = TYPE_MODE (type);
-  machine_mode mode = promote_function_mode (type, orig_mode,
-					     &unsignedp, NULL_TREE, 1);
+  machine_mode mode = promote_return (TYPE_MODE (type));
+
   if (outgoing)
     return gen_rtx_REG (mode, NVPTX_RETURN_REGNUM);
 
@@ -529,7 +536,7 @@ static bool
 nvptx_pass_by_reference (cumulative_args_t ARG_UNUSED (cum), machine_mode mode,
 			 const_tree type, bool ARG_UNUSED (named))
 {
-  return !PASS_IN_REG_P (mode, type);
+  return pass_in_memory (mode, type, false);
 }
 
 /* Implement TARGET_RETURN_IN_MEMORY.  */
@@ -537,35 +544,17 @@ nvptx_pass_by_reference (cumulative_args_t ARG_UNUSED (cum), machine_mode mode,
 static bool
 nvptx_return_in_memory (const_tree type, const_tree)
 {
-  machine_mode mode = TYPE_MODE (type);
-  if (!RETURN_IN_REG_P (mode))
-    return true;
-  return false;
+  return pass_in_memory (TYPE_MODE (type), type, true);
 }
 
 /* Implement TARGET_PROMOTE_FUNCTION_MODE.  */
 
 static machine_mode
 nvptx_promote_function_mode (const_tree type, machine_mode mode,
-			     int *punsignedp,
+			     int *ARG_UNUSED (punsignedp),
 			     const_tree funtype, int for_return)
 {
-  if (type == NULL_TREE)
-    return mode;
-  if (for_return)
-    return promote_mode (type, mode, punsignedp);
-  /* For K&R-style functions, try to match the language promotion rules to
-     minimize type mismatches at assembly time.  */
-  if (TYPE_ARG_TYPES (funtype) == NULL_TREE
-      && type != NULL_TREE
-      && !AGGREGATE_TYPE_P (type))
-    {
-      if (mode == SFmode)
-	mode = DFmode;
-      mode = arg_promotion (mode);
-    }
-
-  return mode;
+  return promote_arg (mode, for_return || !type || TYPE_ARG_TYPES (funtype));
 }
 
 /* Implement TARGET_STATIC_CHAIN.  */
@@ -576,7 +565,6 @@ nvptx_static_chain (const_tree fndecl, bool incoming_p)
   if (!DECL_STATIC_CHAIN (fndecl))
     return NULL;
 
-
   return gen_rtx_REG (Pmode, (incoming_p ? STATIC_CHAIN_REGNUM
 			      : OUTGOING_STATIC_CHAIN_REGNUM));
 }
@@ -620,8 +608,9 @@ write_one_arg (std::stringstream &s, int for_reg, int argno, machine_mode mode)
 }
 
 /* Process function parameter TYPE to emit one or more PTX
-   arguments.  PROTOTYPED is true, if this is a prototyped function,
-   rather than an old-style C declaration.
+   arguments. S, FOR_REG and ARGNO as for write_one_arg.  PROTOTYPED
+   is true, if this is a prototyped function, rather than an old-style
+   C declaration.  Returns the next argument number to use.
 
    The promotion behaviour here must match the regular GCC function
    parameter marshalling machinery.  */
@@ -635,50 +624,71 @@ write_arg (std::stringstream &s, int for_reg, int argno,
   if (mode == VOIDmode)
     return argno;
 
-  if (!PASS_IN_REG_P (mode, type))
+  if (pass_in_memory (mode, type, false))
     mode = Pmode;
+  else
+    {
+      bool split = TREE_CODE (type) == COMPLEX_TYPE;
 
-  machine_mode split = maybe_split_mode (mode);
-  if (split != VOIDmode)
-    mode = split;
+      if (split)
+	{
+	  /* Complex types are sent as two separate args.  */
+	  type = TREE_TYPE (type);
+	  mode  = TYPE_MODE (type);
+	  prototyped = true;
+	}
 
-  if (!prototyped && !AGGREGATE_TYPE_P (type))
-    {
-      if (mode == SFmode)
-	mode = DFmode;
-      mode = arg_promotion (mode);
+      mode = promote_arg (mode, prototyped);
+      if (split)
+	argno = write_one_arg (s, for_reg, argno, mode);
     }
-  else if (for_reg >= 0)
-    mode = arg_promotion (mode);
 
-  if (split != VOIDmode)
-    argno = write_one_arg (s, for_reg, argno, mode);
   return write_one_arg (s, for_reg, argno, mode);
 }
 
+/* Process a function return TYPE to emit a PTX return as a prototype
+   or function prologue declaration.  DECL_RESULT is the decl result
+   of the function and needed for determining named result
+   behaviour. Returns true if return is via an additional pointer
+   parameter.  The promotion behaviour here must match the regular GCC
+   function return mashalling.  */
+
 static bool
-write_return (std::stringstream &s, bool for_proto, tree type,
-	      machine_mode ret_mode)
+write_return (std::stringstream &s, bool for_proto, tree type)
 {
   machine_mode mode = TYPE_MODE (type);
-  bool return_in_mem = mode != VOIDmode && !RETURN_IN_REG_P (mode);
 
-  mode = arg_promotion (mode);
-  if (for_proto)
+  if (mode == VOIDmode)
+    return false;
+
+  bool return_in_mem = pass_in_memory (mode, type, true);
+
+  if (return_in_mem)
     {
-      if (!return_in_mem && mode != VOIDmode)
-	s << "(.param" << nvptx_ptx_type_from_mode (mode, false)
-	  << " %out_retval) ";
+      if (for_proto)
+	return return_in_mem;
+      
+      /* Named return values can cause us to return a pointer as well
+	 as expect an argument for the return location.  This is
+	 optimization-level specific, so no caller can make use of
+	 this data, but more importantly for us, we must ensure it
+	 doesn't change the PTX prototype.  */
+      mode = (machine_mode) cfun->machine->ret_reg_mode;
+      if (mode == VOIDmode)
+	return return_in_mem;
+
+      /* Clear ret_reg_mode to inhibit copy of retval to non-existent
+	 retval parameter.  */
+      cfun->machine->ret_reg_mode = VOIDmode;
     }
   else
-    {
-      /* Prologue.  C++11 ABI causes us to return a reference to the
-	 passed in pointer for return_in_mem.  */
-      ret_mode = arg_promotion (ret_mode);
-      if (ret_mode != VOIDmode)
-	s << "\t.reg" << nvptx_ptx_type_from_mode (ret_mode, false)
-	  << " %retval;\n";
-    }
+    mode = promote_return (mode);
+
+  const char *ptx_type  = nvptx_ptx_type_from_mode (mode, false);
+  if (for_proto)
+    s << "(.param" << ptx_type << " %out_retval) ";
+  else
+    s << "\t.reg" << ptx_type << " %retval;\n";
 
   return return_in_mem;
 }
@@ -751,7 +761,7 @@ write_fn_proto (std::stringstream &s, bool is_defn,
   tree result_type = TREE_TYPE (fntype);
 
   /* Declare the result.  */
-  bool return_in_mem = write_return (s, true, result_type, VOIDmode);
+  bool return_in_mem = write_return (s, true, result_type);
 
   s << name;
 
@@ -943,8 +953,7 @@ nvptx_declare_function_name (FILE *file, const char *name, const_tree decl)
   write_fn_proto (s, true, name, decl);
   s << "{\n";
 
-  bool return_in_mem = write_return (s, false, result_type,
-				     (machine_mode)cfun->machine->ret_reg_mode);
+  bool return_in_mem = write_return (s, false, result_type);
   if (return_in_mem)
     argno = write_arg (s, 0, argno, ptr_type_node, true);
   
@@ -1203,6 +1212,7 @@ nvptx_expand_call (rtx retval, rtx address)
   if (tmp_retval != retval)
     emit_move_insn (retval, tmp_retval);
 }
+
 /* Emit a comparison COMPARE, and return the new test to be used in the
    jump.  */
 
@@ -4840,10 +4850,6 @@ nvptx_goacc_reduction (gcall *call)
 #define TARGET_FUNCTION_INCOMING_ARG nvptx_function_incoming_arg
 #undef TARGET_FUNCTION_ARG_ADVANCE
 #define TARGET_FUNCTION_ARG_ADVANCE nvptx_function_arg_advance
-#undef TARGET_FUNCTION_ARG_BOUNDARY
-#define TARGET_FUNCTION_ARG_BOUNDARY nvptx_function_arg_boundary
-#undef TARGET_FUNCTION_ARG_ROUND_BOUNDARY
-#define TARGET_FUNCTION_ARG_ROUND_BOUNDARY nvptx_function_arg_boundary
 #undef TARGET_PASS_BY_REFERENCE
 #define TARGET_PASS_BY_REFERENCE nvptx_pass_by_reference
 #undef TARGET_FUNCTION_VALUE_REGNO_P
diff --git a/gcc/config/nvptx/nvptx.h b/gcc/config/nvptx/nvptx.h
index 33b2c72..ed0f28e 100644
--- a/gcc/config/nvptx/nvptx.h
+++ b/gcc/config/nvptx/nvptx.h
@@ -46,7 +46,8 @@
 /* Chosen such that we won't have to deal with multi-word subregs.  */
 #define UNITS_PER_WORD 8
 
-#define PARM_BOUNDARY 8
+/* Alignments in bits.  */
+#define PARM_BOUNDARY 32
 #define STACK_BOUNDARY 64
 #define FUNCTION_BOUNDARY 32
 #define BIGGEST_ALIGNMENT 64
diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index 2bbd9dc..ff02575 100644
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,3 +1,22 @@
+2015-12-14  Nathan Sidwell  <nathan@acm.org>
+
+	* g++.dg/abi/nvptx-nrv1.C: New.
+	* g++.dg/abi/nvptx-ptrmem1.C: New.
+	* gcc.target/nvptx/abi-complex-arg.c: New.
+	* gcc.target/nvptx/abi-complex-ret.c: New.
+	* gcc.target/nvptx/abi-enum-arg.c: New.
+	* gcc.target/nvptx/abi-enum-ret.c: New.
+	* gcc.target/nvptx/abi-knr-arg.c: New.
+	* gcc.target/nvptx/abi-knr-ret.c: New.
+	* gcc.target/nvptx/abi-scalar-arg.c: New.
+	* gcc.target/nvptx/abi-scalar-ret.c: New.
+	* gcc.target/nvptx/abi-struct-arg.c: New.
+	* gcc.target/nvptx/abi-struct-ret.c: New.
+	* gcc.target/nvptx/abi-vararg-1.c: New.
+	* gcc.target/nvptx/abi-vararg-2.c: New.
+	* gcc.target/nvptx/abi-vect-arg.c: New.
+	* gcc.target/nvptx/abi-vect-ret.c: New.
+
 2015-12-14  Jakub Jelinek  <jakub@redhat.com>
 
 	PR c/68833
diff --git a/gcc/testsuite/g++.dg/abi/nvptx-nrv1.C b/gcc/testsuite/g++.dg/abi/nvptx-nrv1.C
new file mode 100644
index 0000000..6a05b4f
--- /dev/null
+++ b/gcc/testsuite/g++.dg/abi/nvptx-nrv1.C
@@ -0,0 +1,71 @@
+// { dg-do compile { target nvptx-*-* } }
+// { dg-additional-options "-m64" }
+
+// Check NRV optimization doesn't change the PTX prototypes.
+
+struct A
+{
+  int d;
+
+// { dg-final { scan-assembler-times ".weak .func _ZN1AC1Ev \\(.param.u64 %in_ar0\\)(?:;|\[\r\n\]+\{)" 2 } }
+  A () { d = 123; }
+  A (const A & o) { d = o.d; }
+  void frob ();
+};
+
+
+namespace unopt 
+{
+  // { dg-final { scan-assembler ".extern .func _ZN5unopt3fooEv \\(.param.u64 %in_ar0\\);" } }
+  A __attribute__ ((__optimize__ ("O0"))) foo ();
+
+  // { dg-final { scan-assembler-times ".visible .func _ZN5unopt3barEv \\(.param.u64 %in_ar0\\)(?:;|\[\r\n\]+\{)" 2 } }
+  A __attribute__ ((__optimize__ ("O0"), noinline)) bar()
+  {
+    A l;
+    return l;
+  }
+
+  // { dg-final { scan-assembler-times ".visible .func _ZN5unopt3bazEv \\(.param.u64 %in_ar0\\)(?:;|\[\r\n\]+\{)" 2 } }
+  A __attribute__ ((__optimize__ ("O0"), noinline)) baz ()
+  {
+    return foo ();
+  }
+
+  void __attribute__ ((__optimize__ ("O0"), noinline)) quux ()
+  {
+    bar ().frob ();
+    baz ().frob ();
+    foo ().frob ();
+  }
+  
+}
+
+namespace opt
+{
+  // { dg-final { scan-assembler ".extern .func _ZN3opt3fooEv \\(.param.u64 %in_ar0\\);" } }
+  A __attribute__ ((__optimize__ ("O2"), noinline)) foo ();
+
+  // { dg-final { scan-assembler-times ".visible .func _ZN3opt3barEv \\(.param.u64 %in_ar0\\)(?:;|\[\r\n\]+\{)" 2 } }
+  A __attribute__ ((__optimize__ ("O2"), noinline)) bar()
+  {
+    A l;
+    return l;
+  }
+
+  // { dg-final { scan-assembler-times ".visible .func _ZN3opt3bazEv \\(.param.u64 %in_ar0\\)(?:;|\[\r\n\]+\{)" 2 } }
+  A __attribute__ ((__optimize__ ("O2"))) baz ()
+  {
+    return foo ();
+  }
+
+  void __attribute__ ((__optimize__ ("O2"), noinline)) quux ()
+  {
+    bar ().frob ();
+    baz ().frob ();
+    foo ().frob ();
+  }
+}
+
+// Make sure we're not trying return a return value anywhere.
+// { dg-final { scan-assembler-not "st.param.u64\[\t \]*\\\[%out_retval\\\], %\[_a-z0-9\]*;" } }
diff --git a/gcc/testsuite/g++.dg/abi/nvptx-ptrmem1.C b/gcc/testsuite/g++.dg/abi/nvptx-ptrmem1.C
new file mode 100644
index 0000000..08ffd7e
--- /dev/null
+++ b/gcc/testsuite/g++.dg/abi/nvptx-ptrmem1.C
@@ -0,0 +1,52 @@
+// { dg-do compile { target nvptx-*-* } }
+// { dg-additional-options "-m64" }
+
+// ptr to member data is like sizeof.
+// ptr to member fn is like a struct.
+
+struct A{};
+
+typedef int A::*pmd;
+typedef void (A::*pmf) ();
+
+// { dg-final { scan-assembler ".extern .func \\(.param.u64 %out_retval\\) _Z8dcl_rpmdv;" } }
+pmd dcl_rpmd ();
+
+// { dg-final { scan-assembler ".extern .func _Z8dcl_rpmfv \\(.param.u64 %in_ar0\\);" } }
+pmf dcl_rpmf ();
+
+// { dg-final { scan-assembler ".extern .func _Z8dcl_apmdM1Ai \\(.param.u64 %in_ar0\\);" } }
+void dcl_apmd (pmd);
+
+// { dg-final { scan-assembler ".extern .func _Z8dcl_apmfM1AFvvE \\(.param.u64 %in_ar0\\);" } }
+void dcl_apmf (pmf);
+
+void test_1 ()
+{
+  dcl_rpmd ();
+  dcl_rpmf ();
+  dcl_apmd (0);
+  dcl_apmf (0);
+}
+
+// { dg-final { scan-assembler-times ".visible .func \\(.param.u64 %out_retval\\) _Z8dfn_rpmdv(?:;|\[\r\n\]+\{)" 2 } }
+pmd dfn_rpmd ()
+{
+  return 0;
+}
+
+// { dg-final { scan-assembler-times ".visible .func _Z8dfn_rpmfv \\(.param.u64 %in_ar0\\)(?:;|\[\r\n\]+\{)" 2 } }
+pmf dfn_rpmf ()
+{
+  return 0;
+}
+
+// { dg-final { scan-assembler-times ".visible .func _Z8dfn_apmdM1Ai \\(.param.u64 %in_ar0\\)(?:;|\[\r\n\]+\{)" 2 } }
+void dfn_apmd (pmd)
+{
+}
+
+// { dg-final { scan-assembler-times ".visible .func _Z8dfn_apmfM1AFvvE \\(.param.u64 %in_ar0\\)(?:;|\[\r\n\]+\{)" 2 } }
+void dfn_apmf (pmf)
+{
+}
diff --git a/gcc/testsuite/gcc.target/nvptx/abi-complex-arg.c b/gcc/testsuite/gcc.target/nvptx/abi-complex-arg.c
new file mode 100644
index 0000000..e8f0832
--- /dev/null
+++ b/gcc/testsuite/gcc.target/nvptx/abi-complex-arg.c
@@ -0,0 +1,64 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-Wno-pedantic -Wno-long-long -m64" } */
+
+/* Comples arg types.  All these should be in 2 registers.  */
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_acc \\(.param.u32 %\[_a-z0-9\]*, .param.u32 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_acc (_Complex char);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_acs \\(.param.u32 %\[_a-z0-9\]*, .param.u32 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_acs (_Complex short);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_aci \\(.param.u32 %\[_a-z0-9\]*, .param.u32 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_aci (_Complex int);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_acll \\(.param.u64 %\[_a-z0-9\]*, .param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_acll (_Complex long);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_acf \\(.param.f32 %\[_a-z0-9\]*, .param.f32 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_acf (_Complex float);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_acd \\(.param.f64 %\[_a-z0-9\]*, .param.f64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_acd (_Complex double);
+
+#define M(T,r,i) ({_Complex T x; __real__ (x) = (r), __imag__(x) == (i); x; })
+
+void test_1 (void)
+{
+  dcl_acc (M (char, 1, 2));
+  dcl_acs (M (short, 3, 4));
+  dcl_aci (M (int, 5, 6));
+  dcl_acll (M (long long, 7, 8));
+  dcl_acf (M (float, 9, 10));
+  dcl_acd (M (double, 11, 12));
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_acc \\(.param.u32 %\[_a-z0-9\]*, .param.u32 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_acc (_Complex char c)
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_acs \\(.param.u32 %\[_a-z0-9\]*, .param.u32 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_acs (_Complex short s)
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_aci \\(.param.u32 %\[_a-z0-9\]*, .param.u32 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_aci (_Complex int i)
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_acll \\(.param.u64 %\[_a-z0-9\]*, .param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_acll (_Complex long long ll)
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_acf \\(.param.f32 %\[_a-z0-9\]*, .param.f32 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_acf (_Complex float f)
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_acd \\(.param.f64 %\[_a-z0-9\]*, .param.f64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_acd (_Complex double d)
+{
+}
diff --git a/gcc/testsuite/gcc.target/nvptx/abi-complex-ret.c b/gcc/testsuite/gcc.target/nvptx/abi-complex-ret.c
new file mode 100644
index 0000000..bfaac14
--- /dev/null
+++ b/gcc/testsuite/gcc.target/nvptx/abi-complex-ret.c
@@ -0,0 +1,70 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-Wno-pedantic -Wno-long-long -m64" } */
+
+/* Complex return.  Returned via pointer.  */
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_rcc \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+_Complex char dcl_rcc (void);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_rcs \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+_Complex short dcl_rcs (void);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_rci \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+_Complex int dcl_rci (void);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_rcll \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+_Complex long long dcl_rcll (void);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_rcf \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+_Complex float dcl_rcf (void);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_rcd \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+_Complex double dcl_rcd (void);
+
+void test_1 (void)
+{
+  dcl_rcc ();
+  dcl_rcs ();
+  dcl_rci ();
+  dcl_rcll ();
+  dcl_rcf ();
+  dcl_rcd ();
+}
+
+#define M(T,r,i) ({_Complex T x; __real__ (x) = (r), __imag__(x) == (i); x; })
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_rcc \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+_Complex char dfn_rcc (void)
+{
+  return M (char,1,2);
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_rcs \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+_Complex short dfn_rcs (void)
+{
+  return M (short,3,4);
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_rci \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+_Complex int dfn_rci (void)
+{
+  return M (int,5,6);
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_rcll \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+_Complex long long dfn_rcll (void)
+{
+  return M (long long,7,8);
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_rcf \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+_Complex float dfn_rcf (void)
+{
+  return M (float,9,10);
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_rcd \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+_Complex double dfn_rcd (void)
+{
+  return M (double,11,12);
+}
diff --git a/gcc/testsuite/gcc.target/nvptx/abi-enum-arg.c b/gcc/testsuite/gcc.target/nvptx/abi-enum-arg.c
new file mode 100644
index 0000000..d362676
--- /dev/null
+++ b/gcc/testsuite/gcc.target/nvptx/abi-enum-arg.c
@@ -0,0 +1,49 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-Wno-pedantic -Wno-long-long -m64 -fshort-enums" } */
+
+/* Enum return types.  Passed as the underlying integer.  */
+
+typedef enum { a = 0x1, b } Echar;
+typedef enum { c = 0x100, d } Eshort;
+typedef enum { e = 0x10000, f } Eint;
+typedef enum { g = 0x100000000LL, h } Elonglong;
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_ac \\(.param.u32 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_ac (Echar);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_as \\(.param.u32 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_as (Eshort);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_ai \\(.param.u32 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_ai (Eint);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_all \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_all (Elonglong);
+
+void test_1 (void)
+{
+  dcl_ac (1);
+  dcl_as (2);
+  dcl_ai (3);
+  dcl_all (4);
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_ac \\(.param.u32 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_ac (Echar c)
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_as \\(.param.u32 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_as (Eshort s)
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_ai \\(.param.u32 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_ai (Eint i)
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_all \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_all (Elonglong ll)
+{
+}
diff --git a/gcc/testsuite/gcc.target/nvptx/abi-enum-ret.c b/gcc/testsuite/gcc.target/nvptx/abi-enum-ret.c
new file mode 100644
index 0000000..c562656
--- /dev/null
+++ b/gcc/testsuite/gcc.target/nvptx/abi-enum-ret.c
@@ -0,0 +1,53 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-Wno-pedantic -Wno-long-long -m64 -fshort-enums" } */
+
+/* Enum return types.  Passed as the underlying integer.  */
+
+typedef enum { a = 0x1, b } Echar;
+typedef enum { c = 0x100, d } Eshort;
+typedef enum { e = 0x10000, f } Eint;
+typedef enum { g = 0x100000000LL, h } Elonglong;
+
+/* { dg-final { scan-assembler-times ".extern .func \\(.param.u32 %\[_a-z\]*\\) dcl_rc;" 1 } } */
+Echar dcl_rc (void);
+
+/* { dg-final { scan-assembler-times ".extern .func \\(.param.u32 %\[_a-z\]*\\) dcl_rs;" 1 } } */
+Eshort dcl_rs (void);
+
+/* { dg-final { scan-assembler-times ".extern .func \\(.param.u32 %\[_a-z\]*\\) dcl_ri;" 1 } } */
+Eint dcl_ri (void);
+
+/* { dg-final { scan-assembler-times ".extern .func \\(.param.u64 %\[_a-z\]*\\) dcl_rll;" 1 } } */
+Elonglong dcl_rll (void);
+
+void test_1 (void)
+{
+  dcl_rc ();
+  dcl_rs ();
+  dcl_ri ();
+  dcl_rll ();
+}
+
+/* { dg-final { scan-assembler-times ".visible .func \\(.param.u32 %\[_a-z0-9\]*\\) dfn_rc(?:;|\[\r\n\]+\{)" 2 } } */
+Echar dfn_rc (void)
+{
+  return 1;
+}
+
+/* { dg-final { scan-assembler-times ".visible .func \\(.param.u32 %\[_a-z0-0\]*\\) dfn_rs(?:;|\[\r\n\]+\{)" 2 } } */
+Eshort dfn_rs (void)
+{
+  return 2;
+}
+
+/* { dg-final { scan-assembler-times ".visible .func \\(.param.u32 %\[_a-z0-9\]*\\) dfn_ri(?:;|\[\r\n\]+\{)" 2 } } */
+Eint dfn_ri (void)
+{
+  return 3;
+}
+
+/* { dg-final { scan-assembler-times ".visible .func \\(.param.u64 %\[_a-z0-9\]*\\) dfn_rll(?:;|\[\r\n\]+\{)" 2 } } */
+Elonglong dfn_rll (void)
+{
+  return 4;
+}
diff --git a/gcc/testsuite/gcc.target/nvptx/abi-knr-arg.c b/gcc/testsuite/gcc.target/nvptx/abi-knr-arg.c
new file mode 100644
index 0000000..aca5073
--- /dev/null
+++ b/gcc/testsuite/gcc.target/nvptx/abi-knr-arg.c
@@ -0,0 +1,95 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-Wno-pedantic -Wno-long-long -m64" } */
+
+/* K+R args.  as for ANSI except float promotes to double.  */
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_av;" 1 } } */
+void dcl_av ();
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_ac \\(.param.u32 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_ac ();
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_as \\(.param.u32 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_as ();
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_ai \\(.param.u32 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_ai ();
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_all \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_all ();
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_af \\(.param.f64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_af ();
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_ad \\(.param.f64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_ad ();
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_ap \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_ap ();
+
+void test_1 (void)
+{
+  dcl_av ();
+  dcl_ac ((char)1);
+  dcl_as ((short)2);
+  dcl_ai ((int)3);
+  dcl_all ((long long)4);
+  dcl_af ((float)5);
+  dcl_ad ((double)6);
+  dcl_ap ((void *)0);
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_av(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_av ()
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_ac \\(.param.u32 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_ac (c)
+  char c;
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_as \\(.param.u32 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_as (s)
+  short s;
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_ai \\(.param.u32 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_ai (i)
+  int i;
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_all \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_all (ll)
+  long long ll;
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_af \\(.param.f64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_af (f)
+  float f;
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_ad \\(.param.f64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_ad (d)
+  double d;
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_ap \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_ap (p)
+  void *p;
+{
+}
+
+/*  But complex float is passed as two floats.  (K&R doesn't have
+    complex, so why obey k&r for the components of such an object?)  */
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_acf \\(.param.f32 %\[_a-z0-9\]*, .param.f32 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_acf (_Complex float f)
+{
+}
diff --git a/gcc/testsuite/gcc.target/nvptx/abi-knr-ret.c b/gcc/testsuite/gcc.target/nvptx/abi-knr-ret.c
new file mode 100644
index 0000000..17bfd75
--- /dev/null
+++ b/gcc/testsuite/gcc.target/nvptx/abi-knr-ret.c
@@ -0,0 +1,88 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-Wno-pedantic -Wno-long-long -m64" } */
+
+/* K+R returns.  as for ANSI .  */
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_rv;" 1 } } */
+void dcl_rv ();
+
+/* { dg-final { scan-assembler-times ".extern .func \\(.param.u32 %\[_a-z\]*\\) dcl_rc;" 1 } } */
+char dcl_rc ();
+
+/* { dg-final { scan-assembler-times ".extern .func \\(.param.u32 %\[_a-z\]*\\) dcl_rs;" 1 } } */
+short dcl_rs ();
+
+/* { dg-final { scan-assembler-times ".extern .func \\(.param.u32 %\[_a-z\]*\\) dcl_ri;" 1 } } */
+int dcl_ri ();
+
+/* { dg-final { scan-assembler-times ".extern .func \\(.param.u64 %\[_a-z\]*\\) dcl_rll;" 1 } } */
+long long dcl_rll ();
+
+/* { dg-final { scan-assembler-times ".extern .func \\(.param.f32 %\[_a-z\]*\\) dcl_rf;" 1 } } */
+float dcl_rf ();
+
+/* { dg-final { scan-assembler-times ".extern .func \\(.param.f64 %\[_a-z\]*\\) dcl_rd;" 1 } } */
+double dcl_rd ();
+
+/* { dg-final { scan-assembler-times ".extern .func \\(.param.u64 %\[_a-z\]*\\) dcl_rp;" 1 } } */
+void *dcl_rp ();
+
+void test_1 ()
+{
+  dcl_rv ();
+  dcl_rc ();
+  dcl_rs ();
+  dcl_ri ();
+  dcl_rll ();
+  dcl_rf ();
+  dcl_rd ();
+  dcl_rp ();
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_rv(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_rv ()
+{
+  return;
+}
+
+/* { dg-final { scan-assembler-times ".visible .func \\(.param.u32 %\[_a-z0-9\]*\\) dfn_rc(?:;|\[\r\n\]+\{)" 2 } } */
+char dfn_rc ()
+{
+  return 1;
+}
+
+/* { dg-final { scan-assembler-times ".visible .func \\(.param.u32 %\[_a-z0-0\]*\\) dfn_rs(?:;|\[\r\n\]+\{)" 2 } } */
+short dfn_rs ()
+{
+  return 2;
+}
+
+/* { dg-final { scan-assembler-times ".visible .func \\(.param.u32 %\[_a-z0-9\]*\\) dfn_ri(?:;|\[\r\n\]+\{)" 2 } } */
+int dfn_ri ()
+{
+  return 3;
+}
+
+/* { dg-final { scan-assembler-times ".visible .func \\(.param.u64 %\[_a-z0-9\]*\\) dfn_rll(?:;|\[\r\n\]+\{)" 2 } } */
+long long dfn_rll ()
+{
+  return 4;
+}
+
+/* { dg-final { scan-assembler-times ".visible .func \\(.param.f32 %\[_a-z0-9\]*\\) dfn_rf(?:;|\[\r\n\]+\{)" 2 } } */
+float dfn_rf ()
+{
+  return 5;
+}
+
+/* { dg-final { scan-assembler-times ".visible .func \\(.param.f64 %\[_a-z0-9\]*\\) dfn_rd(?:;|\[\r\n\]+\{)" 2 } } */
+double dfn_rd ()
+{
+  return 6;
+}
+
+/* { dg-final { scan-assembler-times ".visible .func \\(.param.u64 %\[_a-z0-9\]*\\) dfn_rp(?:;|\[\r\n\]+\{)" 2 } } */
+void *dfn_rp ()
+{
+  return 0;
+}
diff --git a/gcc/testsuite/gcc.target/nvptx/abi-scalar-arg.c b/gcc/testsuite/gcc.target/nvptx/abi-scalar-arg.c
new file mode 100644
index 0000000..0fe1e67
--- /dev/null
+++ b/gcc/testsuite/gcc.target/nvptx/abi-scalar-arg.c
@@ -0,0 +1,91 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-Wno-pedantic -Wno-long-long -m64" } */
+
+/* Scalar return types.  In registers when <= 64 bit.  */
+
+typedef int __attribute__((mode(TI))) ti;
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_av;" 1 } } */
+void dcl_av (void);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_ac \\(.param.u32 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_ac (char);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_as \\(.param.u32 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_as (short);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_ai \\(.param.u32 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_ai (int);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_all \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_all (long long);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_ati \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_ati (ti);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_af \\(.param.f32 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_af (float);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_ad \\(.param.f64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_ad (double);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_ap \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_ap (void *);
+
+void test_1 (void)
+{
+  dcl_av ();
+  dcl_ac (1);
+  dcl_as (2);
+  dcl_ai (3);
+  dcl_all (4);
+  dcl_ati (5);
+  dcl_af (6);
+  dcl_ad (7);
+  dcl_ap (0);
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_av(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_av (void)
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_ac \\(.param.u32 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_ac (char c)
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_as \\(.param.u32 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_as (short s)
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_ai \\(.param.u32 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_ai (int i)
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_all \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_all (long long ll)
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_ati \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_ati (ti t)
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_af \\(.param.f32 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_af (float f)
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_ad \\(.param.f64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_ad (double d)
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_ap \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_ap (void *p)
+{
+}
diff --git a/gcc/testsuite/gcc.target/nvptx/abi-scalar-ret.c b/gcc/testsuite/gcc.target/nvptx/abi-scalar-ret.c
new file mode 100644
index 0000000..e9b11ba
--- /dev/null
+++ b/gcc/testsuite/gcc.target/nvptx/abi-scalar-ret.c
@@ -0,0 +1,100 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-Wno-pedantic -Wno-long-long -m64" } */
+
+/* Scalar return types.  In registers when <= 64 bit.  */
+
+typedef int __attribute__((mode(TI))) ti;
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_rv;" 1 } } */
+void dcl_rv (void);
+
+/* { dg-final { scan-assembler-times ".extern .func \\(.param.u32 %\[_a-z\]*\\) dcl_rc;" 1 } } */
+char dcl_rc (void);
+
+/* { dg-final { scan-assembler-times ".extern .func \\(.param.u32 %\[_a-z\]*\\) dcl_rs;" 1 } } */
+short dcl_rs (void);
+
+/* { dg-final { scan-assembler-times ".extern .func \\(.param.u32 %\[_a-z\]*\\) dcl_ri;" 1 } } */
+int dcl_ri (void);
+
+/* { dg-final { scan-assembler-times ".extern .func \\(.param.u64 %\[_a-z\]*\\) dcl_rll;" 1 } } */
+long long dcl_rll (void);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_rti \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+ti dcl_rti (void);
+
+/* { dg-final { scan-assembler-times ".extern .func \\(.param.f32 %\[_a-z\]*\\) dcl_rf;" 1 } } */
+float dcl_rf (void);
+
+/* { dg-final { scan-assembler-times ".extern .func \\(.param.f64 %\[_a-z\]*\\) dcl_rd;" 1 } } */
+double dcl_rd (void);
+
+/* { dg-final { scan-assembler-times ".extern .func \\(.param.u64 %\[_a-z\]*\\) dcl_rp;" 1 } } */
+void *dcl_rp (void);
+
+void test_1 (void)
+{
+  dcl_rv ();
+  dcl_rc ();
+  dcl_rs ();
+  dcl_ri ();
+  dcl_rll ();
+  dcl_rti ();
+  dcl_rf ();
+  dcl_rd ();
+  dcl_rp ();
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_rv(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_rv (void)
+{
+  return;
+}
+
+/* { dg-final { scan-assembler-times ".visible .func \\(.param.u32 %\[_a-z0-9\]*\\) dfn_rc(?:;|\[\r\n\]+\{)" 2 } } */
+char dfn_rc (void)
+{
+  return 1;
+}
+
+/* { dg-final { scan-assembler-times ".visible .func \\(.param.u32 %\[_a-z0-0\]*\\) dfn_rs(?:;|\[\r\n\]+\{)" 2 } } */
+short dfn_rs (void)
+{
+  return 2;
+}
+
+/* { dg-final { scan-assembler-times ".visible .func \\(.param.u32 %\[_a-z0-9\]*\\) dfn_ri(?:;|\[\r\n\]+\{)" 2 } } */
+int dfn_ri (void)
+{
+  return 3;
+}
+
+/* { dg-final { scan-assembler-times ".visible .func \\(.param.u64 %\[_a-z0-9\]*\\) dfn_rll(?:;|\[\r\n\]+\{)" 2 } } */
+long long dfn_rll (void)
+{
+  return 4;
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_rti \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+ti dfn_rti (void)
+{
+  return 5;
+}
+
+/* { dg-final { scan-assembler-times ".visible .func \\(.param.f32 %\[_a-z0-9\]*\\) dfn_rf(?:;|\[\r\n\]+\{)" 2 } } */
+float dfn_rf (void)
+{
+  return 6;
+}
+
+/* { dg-final { scan-assembler-times ".visible .func \\(.param.f64 %\[_a-z0-9\]*\\) dfn_rd(?:;|\[\r\n\]+\{)" 2 } } */
+double dfn_rd (void)
+{
+  return 7;
+}
+
+/* { dg-final { scan-assembler-times ".visible .func \\(.param.u64 %\[_a-z0-9\]*\\) dfn_rp(?:;|\[\r\n\]+\{)" 2 } } */
+void *dfn_rp (void)
+{
+  return 0;
+}
diff --git a/gcc/testsuite/gcc.target/nvptx/abi-struct-arg.c b/gcc/testsuite/gcc.target/nvptx/abi-struct-arg.c
new file mode 100644
index 0000000..54ae651
--- /dev/null
+++ b/gcc/testsuite/gcc.target/nvptx/abi-struct-arg.c
@@ -0,0 +1,62 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-Wno-pedantic -Wno-long-long -m64" } */
+
+/* Struct arg.  Passed via pointer.  */
+
+typedef struct {char a;} one;
+typedef struct {short a;} two;
+typedef struct {int a;} four;
+typedef struct {long long a;} eight;
+typedef struct {int a, b[12];} big;
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_aone \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_aone (one);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_atwo \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_atwo (two);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_afour \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_afour (four);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_aeight \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_aeight (eight);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_abig \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_abig (big);
+
+#define M(T, v) ({T t; t.a = v; t;})
+
+void test_1 (void)
+{
+  dcl_aone (M (one, 1));
+  dcl_atwo (M (two, 2));
+  dcl_afour (M (four, 3));
+  dcl_aeight (M (eight, 4));
+  dcl_abig (M (big, 5));
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_aone \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_aone (one one)
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_atwo \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_atwo (two two)
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_afour \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_afour (four four)
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_aeight \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_aeight (eight eight)
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_abig \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_abig (big big)
+{
+}
+
diff --git a/gcc/testsuite/gcc.target/nvptx/abi-struct-ret.c b/gcc/testsuite/gcc.target/nvptx/abi-struct-ret.c
new file mode 100644
index 0000000..d48a82d
--- /dev/null
+++ b/gcc/testsuite/gcc.target/nvptx/abi-struct-ret.c
@@ -0,0 +1,66 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-Wno-pedantic -Wno-long-long -m64" } */
+
+/* Struct return.  Returned via pointer.  */
+
+typedef struct {char a;} one;
+typedef struct {short a;} two;
+typedef struct {int a;} four;
+typedef struct {long long a;} eight;
+typedef struct {int a, b[12];} big;
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_rone \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+one dcl_rone (void);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_rtwo \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+two dcl_rtwo (void);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_rfour \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+four dcl_rfour (void);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_reight \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+eight dcl_reight (void);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_rbig \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+big dcl_rbig (void);
+
+void test_1 (void)
+{
+  dcl_rone ();
+  dcl_rtwo ();
+  dcl_rfour ();
+  dcl_reight ();
+  dcl_rbig ();
+}
+
+#define M(T, v) ({T t; t.a = v; t;})
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_rone \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+one dfn_rone (void)
+{
+  return M (one, 1);
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_rtwo \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+two dfn_rtwo (void)
+{
+  return M (two, 2);
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_rfour \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+four dfn_rfour (void)
+{
+  return M (four, 3);
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_reight \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+eight dfn_reight (void)
+{
+  return M (eight, 4);
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_rbig \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+big dfn_rbig (void)
+{
+  return M (big, 5);
+}
diff --git a/gcc/testsuite/gcc.target/nvptx/abi-vararg-1.c b/gcc/testsuite/gcc.target/nvptx/abi-vararg-1.c
new file mode 100644
index 0000000..7e56b88
--- /dev/null
+++ b/gcc/testsuite/gcc.target/nvptx/abi-vararg-1.c
@@ -0,0 +1,19 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-Wno-pedantic -Wno-long-long -m64" } */
+
+/* varargs, passed as pointer to array of args.   */
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_av \\(.param.u32 %\[_a-z0-9\]*, .param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_av (int, ...);
+
+void test_1 (void)
+{
+  dcl_av (1, 1);
+  dcl_av (2, 1, 2);
+  dcl_av (2, 1, 2, 3);
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_av \\(.param.u32 %\[_a-z0-9\]*, .param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_av (int a, ...)
+{
+}
diff --git a/gcc/testsuite/gcc.target/nvptx/abi-vararg-2.c b/gcc/testsuite/gcc.target/nvptx/abi-vararg-2.c
new file mode 100644
index 0000000..bcaeb5b
--- /dev/null
+++ b/gcc/testsuite/gcc.target/nvptx/abi-vararg-2.c
@@ -0,0 +1,208 @@
+/* { dg-do run } */
+/* { dg-additional-options "-Wno-pedantic -Wno-long-long -fshort-enums" } */
+
+/* va args, promoted as regular knr args.  */
+
+void __attribute__ ((noinline)) chars (int i, ...)
+{
+  __builtin_va_list args;
+  __builtin_va_start (args, i);
+
+  if (__builtin_va_arg (args, int) != 1) __builtin_abort ();
+  if (__builtin_va_arg (args, int) != 2) __builtin_abort ();
+  __builtin_va_end (args);
+}
+
+void __attribute__ ((noinline)) shorts (int i, ...)
+{
+  __builtin_va_list args;
+  __builtin_va_start (args, i);
+
+  if (__builtin_va_arg (args, int) != 1) __builtin_abort ();
+  if (__builtin_va_arg (args, int) != 2) __builtin_abort ();
+  if (__builtin_va_arg (args, int) != 3) __builtin_abort ();
+  __builtin_va_end (args);
+}
+
+void __attribute__ ((noinline)) ints (int i, ...)
+{
+  __builtin_va_list args;
+  __builtin_va_start (args, i);
+
+  if (__builtin_va_arg (args, int) != 1) __builtin_abort ();
+  if (__builtin_va_arg (args, int) != 2) __builtin_abort ();
+  if (__builtin_va_arg (args, int) != 3) __builtin_abort ();
+  __builtin_va_end (args);
+}
+
+void __attribute__ ((noinline)) longlongs (int i, ...)
+{
+  __builtin_va_list args;
+  __builtin_va_start (args, i);
+
+  if (__builtin_va_arg (args, int) != 1) __builtin_abort ();
+  if (__builtin_va_arg (args, long long) != 2) __builtin_abort ();
+  if (__builtin_va_arg (args, long long) != 3) __builtin_abort ();
+  __builtin_va_end (args);
+}
+
+typedef int __attribute__ ((mode(TI))) ti;
+
+void __attribute__ ((noinline)) tis (int i, ...)
+{
+  __builtin_va_list args;
+  __builtin_va_start (args, i);
+
+  if (__builtin_va_arg (args, int) != 1) __builtin_abort ();
+  if (__builtin_va_arg (args, ti) != 2) __builtin_abort ();
+  if (__builtin_va_arg (args, ti) != 3) __builtin_abort ();
+  __builtin_va_end (args);
+}
+
+void __attribute__ ((noinline)) floats (int i, ...)
+{
+  __builtin_va_list args;
+  __builtin_va_start (args, i);
+
+  if (__builtin_va_arg (args, int) != 1) __builtin_abort ();
+  if (__builtin_va_arg (args, double) != 2) __builtin_abort ();
+  if (__builtin_va_arg (args, double) != 3) __builtin_abort ();
+  __builtin_va_end (args);
+}
+
+void __attribute__ ((noinline)) doubles (int i, ...)
+{
+  __builtin_va_list args;
+  __builtin_va_start (args, i);
+
+  if (__builtin_va_arg (args, int) != 1) __builtin_abort ();
+  if (__builtin_va_arg (args, double) != 2) __builtin_abort ();
+  if (__builtin_va_arg (args, double) != 3) __builtin_abort ();
+  __builtin_va_end (args);
+}
+
+typedef enum {ec1, ec2, ecmax = 0xf} echar;
+typedef enum {es1, es2, esmax = 0xfff} eshort;
+typedef enum {ei1, ei2, eimax = 0xfffff} eint;
+
+void __attribute__ ((noinline)) echars (int i, ...)
+{
+  __builtin_va_list args;
+  __builtin_va_start (args, i);
+
+  if (__builtin_va_arg (args, int) != ec1) __builtin_abort ();
+  if (__builtin_va_arg (args, int) != ec2) __builtin_abort ();
+  if (__builtin_va_arg (args, int) != ecmax) __builtin_abort ();
+  __builtin_va_end (args);
+}
+
+void __attribute__ ((noinline)) eshorts (int i, ...)
+{
+  __builtin_va_list args;
+  __builtin_va_start (args, i);
+
+  if (__builtin_va_arg (args, int) != es1) __builtin_abort ();
+  if (__builtin_va_arg (args, int) != es2) __builtin_abort ();
+  if (__builtin_va_arg (args, int) != esmax) __builtin_abort ();
+  __builtin_va_end (args);
+}
+
+void __attribute__ ((noinline)) eints (int i, ...)
+{
+  __builtin_va_list args;
+  __builtin_va_start (args, i);
+
+  if (__builtin_va_arg (args, int) != ei1) __builtin_abort ();
+  if (__builtin_va_arg (args, int) != ei2) __builtin_abort ();
+  if (__builtin_va_arg (args, int) != eimax) __builtin_abort ();
+  __builtin_va_end (args);
+}
+
+typedef struct {char a;} one;
+typedef struct {short a;} two;
+typedef struct {int a;} four;
+typedef struct {long long a;} eight;
+typedef struct {int a, b[12];} big;
+
+void __attribute__ ((noinline)) structs (int i, ...)
+{
+  __builtin_va_list args;
+  __builtin_va_start (args, i);
+
+  if (__builtin_va_arg (args, int) != 1) __builtin_abort ();
+  if (__builtin_va_arg (args, one).a != 2) __builtin_abort ();
+  if (__builtin_va_arg (args, int) != 3) __builtin_abort ();
+  if (__builtin_va_arg (args, two).a != 4) __builtin_abort ();
+  if (__builtin_va_arg (args, int) != 5) __builtin_abort ();
+  if (__builtin_va_arg (args, four).a != 6) __builtin_abort ();
+  if (__builtin_va_arg (args, int) != 7) __builtin_abort ();
+  if (__builtin_va_arg (args, eight).a != 8) __builtin_abort ();
+  if (__builtin_va_arg (args, int) != 9) __builtin_abort ();
+  if (__builtin_va_arg (args, big).a != 10) __builtin_abort ();
+  __builtin_va_end (args);
+}
+
+int main ()
+{
+  char vc1, vc2, vc3, vc4, vc5;
+  short vs1, vs2, vs3;
+  int vi1, vi2, vi3;
+  long long vll1, vll2;
+  ti vti1, vti2;
+  float vf1, vf2;
+  double vd1, vd2;
+
+  one vone;
+  two vtwo;
+  four vfour;
+  eight veight;
+  big vbig;
+  echar vec1, vec2,  vec3;
+  eshort ves1, ves2,  ves3;
+  eint vei1, vei2,  vei3;
+
+  vc1 = 1, vc2 = 2;
+  chars (1, vc1, vc2);
+
+  vc1 = 1, vs1 = 2, vs2 = 3;
+  shorts (1, vc1, vs1, vs2);
+
+  vc1 = 1, vi1 = 2, vi2 = 3;
+  ints (1, vc1, vi1, vi2);
+
+  vc1 = 1,  vll1 = 2, vll2 = 3;
+  longlongs (1, vc1, vll1, vll2);
+
+  vc1 = 1, vti1 = 2, vti2 = 3;
+  tis (1, vc1,  vti1, vti2);
+
+  vc1 = 1,  vf1 = 2, vf2 = 3;
+  floats (1, vc1, vf1, vf2);
+  doubles (1, vc1, vf1, vf2); /* Floats are promoted, so this should work. */
+
+  vc1 = 1, vd1 = 2, vd2 = 3;
+  floats (1, vc1, vf1, vf2); /* Floats are promoted, so this should work. */
+  doubles (1, vc1, vf1, vf2);
+
+  vec1 = ec1, vec2 = ec2, vec3 = ecmax;
+  echars (1, vec1, vec2, vec3);
+  vc1 = ec1, vc2 = ec2, vc3 = ecmax;
+  echars (1, vc1, vc2, vc3);
+
+  ves1 = ec1, ves2 = ec2, ves3 = esmax;
+  eshorts (1, ves1, ves2, ves3);
+  vs1 = ec1, vs2 = ec2, vs3 = esmax;
+  eshorts (1, vs1, vs2, vs3);
+
+  vei1 = ec1, vei2 = ec2, vei3 = eimax;
+  eints (1, vei1, vei2, vei3);
+  vi1 = ec1, vi2 = ec2, vi3 = eimax;
+  eints (1, vi1, vi2, vi3);
+
+  vc1 = 1, vone.a = 2, vc2 = 3, vtwo.a = 4,
+    vc3 = 5, vfour.a = 6, vc4 = 7, veight.a = 8,
+    vc5 = 9, vbig.a = 10;
+  structs (1, vc1,  vone, vc2, vtwo, vc3, vfour, vc4, veight, vc5, vbig);
+  
+  return 0;
+}
diff --git a/gcc/testsuite/gcc.target/nvptx/abi-vect-arg.c b/gcc/testsuite/gcc.target/nvptx/abi-vect-arg.c
new file mode 100644
index 0000000..1247b29
--- /dev/null
+++ b/gcc/testsuite/gcc.target/nvptx/abi-vect-arg.c
@@ -0,0 +1,225 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-Wno-pedantic -Wno-long-long -m64" } */
+
+/* Vector arg.  Pass via pointer.  */
+
+typedef char __attribute__ ((vector_size (1))) vc1;
+typedef char __attribute__ ((vector_size (2))) vc2;
+typedef char __attribute__ ((vector_size (4))) vc4;
+typedef char __attribute__ ((vector_size (8))) vc8;
+
+typedef short __attribute__ ((vector_size (2))) vs1;
+typedef short __attribute__ ((vector_size (4))) vs2;
+typedef short __attribute__ ((vector_size (8))) vs4;
+typedef short __attribute__ ((vector_size (16))) vs8;
+
+typedef int __attribute__ ((vector_size (4))) vi1;
+typedef int __attribute__ ((vector_size (8))) vi2;
+typedef int __attribute__ ((vector_size (16))) vi4;
+typedef int __attribute__ ((vector_size (32))) vi8;
+
+typedef long long __attribute__ ((vector_size (8))) vll1;
+typedef long long __attribute__ ((vector_size (16))) vll2;
+typedef long long __attribute__ ((vector_size (32))) vll4;
+typedef long long __attribute__ ((vector_size (64))) vll8;
+
+typedef float __attribute__ ((vector_size (4))) vf1;
+typedef float __attribute__ ((vector_size (8))) vf2;
+typedef float __attribute__ ((vector_size (16))) vf4;
+typedef float __attribute__ ((vector_size (32))) vf8;
+
+typedef double __attribute__ ((vector_size (8))) vd1;
+typedef double __attribute__ ((vector_size (16))) vd2;
+typedef double __attribute__ ((vector_size (32))) vd4;
+typedef double __attribute__ ((vector_size (64))) vd8;
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_avc1 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_avc1 (vc1);
+/* { dg-final { scan-assembler-times ".extern .func dcl_avc2 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_avc2 (vc2);
+/* { dg-final { scan-assembler-times ".extern .func dcl_avc4 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_avc4 (vc4);
+/* { dg-final { scan-assembler-times ".extern .func dcl_avc8 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_avc8 (vc8);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_avs1 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_avs1 (vs1);
+/* { dg-final { scan-assembler-times ".extern .func dcl_avs2 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_avs2 (vs2);
+/* { dg-final { scan-assembler-times ".extern .func dcl_avs4 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_avs4 (vs4);
+/* { dg-final { scan-assembler-times ".extern .func dcl_avs8 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_avs8 (vs8);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_avi1 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_avi1 (vi1);
+/* { dg-final { scan-assembler-times ".extern .func dcl_avi2 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_avi2 (vi2);
+/* { dg-final { scan-assembler-times ".extern .func dcl_avi4 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_avi4 (vi4);
+/* { dg-final { scan-assembler-times ".extern .func dcl_avi8 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_avi8 (vi8);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_avll1 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_avll1 (vll1);
+/* { dg-final { scan-assembler-times ".extern .func dcl_avll2 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_avll2 (vll2);
+/* { dg-final { scan-assembler-times ".extern .func dcl_avll4 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_avll4 (vll4);
+/* { dg-final { scan-assembler-times ".extern .func dcl_avll8 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_avll8 (vll8);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_avf1 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_avf1 (vf1);
+/* { dg-final { scan-assembler-times ".extern .func dcl_avf2 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_avf2 (vf2);
+/* { dg-final { scan-assembler-times ".extern .func dcl_avf4 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_avf4 (vf4);
+/* { dg-final { scan-assembler-times ".extern .func dcl_avf8 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_avf8 (vf8);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_avd1 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_avd1 (vd1);
+/* { dg-final { scan-assembler-times ".extern .func dcl_avd2 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_avd2 (vd2);
+/* { dg-final { scan-assembler-times ".extern .func dcl_avd4 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_avd4 (vd4);
+/* { dg-final { scan-assembler-times ".extern .func dcl_avd8 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+void dcl_avd8 (vd8);
+
+#define M(T, V) ({T t;t[0]= V;t;})
+
+void  test_1 (void)
+{
+  dcl_avc1 (M (vc1, 1));
+  dcl_avc2 (M (vc2, 2));
+  dcl_avc4 (M (vc4, 3));
+  dcl_avc8 (M (vc8, 4));
+
+  dcl_avs1 (M (vs1, 5));
+  dcl_avs2 (M (vs2, 6));
+  dcl_avs4 (M (vs4, 7));
+  dcl_avs8 (M (vs8, 8));
+
+  dcl_avi1 (M (vi1, 9));
+  dcl_avi2 (M (vi2, 10));
+  dcl_avi4 (M (vi4, 11));
+  dcl_avi8 (M (vi8, 12));
+
+  dcl_avll1 (M (vll1, 13));
+  dcl_avll2 (M (vll2, 14));
+  dcl_avll4 (M (vll4, 15));
+  dcl_avll8 (M (vll8, 16));
+
+  dcl_avf1 (M (vf1, 17));
+  dcl_avf2 (M (vf2, 18));
+  dcl_avf4 (M (vf4, 19));
+  dcl_avf8 (M (vf8, 20));
+
+  dcl_avd1 (M (vd1, 21));
+  dcl_avd2 (M (vd2, 22));
+  dcl_avd4 (M (vd4, 23));
+  dcl_avd8 (M (vd8, 24));
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_avc1 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_avc1(vc1 a)
+{
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_avc2 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_avc2(vc2 a)
+{
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_avc4 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_avc4(vc4 a)
+{
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_avc8 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_avc8(vc8 a)
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_avs1 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_avs1(vs1 a)
+{
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_avs2 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_avs2(vs2 a)
+{
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_avs4 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_avs4(vs4 a)
+{
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_avs8 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_avs8(vs8 a)
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_avi1 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_avi1(vi1 a)
+{
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_avi2 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_avi2(vi2 a)
+{
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_avi4 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_avi4(vi4 a)
+{
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_avi8 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_avi8(vi8 a)
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_avll1 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_avll1(vll1 a)
+{
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_avll2 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_avll2(vll2 a)
+{
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_avll4 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_avll4(vll4 a)
+{
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_avll8 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_avll8(vll8 a)
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_avf1 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_avf1(vf1 a)
+{
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_avf2 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_avf2(vf2 a)
+{
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_avf4 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_avf4(vf4 a)
+{
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_avf8 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_avf8(vf8 a)
+{
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_avd1 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_avd1(vd1 a)
+{
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_avd2 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_avd2(vd2 a)
+{
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_avd4 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_avd4(vd4 a)
+{
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_avd8 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+void dfn_avd8(vd8 a)
+{
+}
diff --git a/gcc/testsuite/gcc.target/nvptx/abi-vect-ret.c b/gcc/testsuite/gcc.target/nvptx/abi-vect-ret.c
new file mode 100644
index 0000000..78adf50
--- /dev/null
+++ b/gcc/testsuite/gcc.target/nvptx/abi-vect-ret.c
@@ -0,0 +1,249 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-Wno-pedantic -Wno-long-long -m64" } */
+
+/* Vector return.  Return via pointer.  */
+
+typedef char __attribute__ ((vector_size (1))) vc1;
+typedef char __attribute__ ((vector_size (2))) vc2;
+typedef char __attribute__ ((vector_size (4))) vc4;
+typedef char __attribute__ ((vector_size (8))) vc8;
+
+typedef short __attribute__ ((vector_size (2))) vs1;
+typedef short __attribute__ ((vector_size (4))) vs2;
+typedef short __attribute__ ((vector_size (8))) vs4;
+typedef short __attribute__ ((vector_size (16))) vs8;
+
+typedef int __attribute__ ((vector_size (4))) vi1;
+typedef int __attribute__ ((vector_size (8))) vi2;
+typedef int __attribute__ ((vector_size (16))) vi4;
+typedef int __attribute__ ((vector_size (32))) vi8;
+
+typedef long long __attribute__ ((vector_size (8))) vll1;
+typedef long long __attribute__ ((vector_size (16))) vll2;
+typedef long long __attribute__ ((vector_size (32))) vll4;
+typedef long long __attribute__ ((vector_size (64))) vll8;
+
+typedef float __attribute__ ((vector_size (4))) vf1;
+typedef float __attribute__ ((vector_size (8))) vf2;
+typedef float __attribute__ ((vector_size (16))) vf4;
+typedef float __attribute__ ((vector_size (32))) vf8;
+
+typedef double __attribute__ ((vector_size (8))) vd1;
+typedef double __attribute__ ((vector_size (16))) vd2;
+typedef double __attribute__ ((vector_size (32))) vd4;
+typedef double __attribute__ ((vector_size (64))) vd8;
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_rvc1 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+vc1 dcl_rvc1 (void);
+/* { dg-final { scan-assembler-times ".extern .func dcl_rvc2 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+vc2 dcl_rvc2 (void);
+/* { dg-final { scan-assembler-times ".extern .func dcl_rvc4 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+vc4 dcl_rvc4 (void);
+/* { dg-final { scan-assembler-times ".extern .func dcl_rvc8 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+vc8 dcl_rvc8 (void);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_rvs1 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+vs1 dcl_rvs1 (void);
+/* { dg-final { scan-assembler-times ".extern .func dcl_rvs2 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+vs2 dcl_rvs2 (void);
+/* { dg-final { scan-assembler-times ".extern .func dcl_rvs4 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+vs4 dcl_rvs4 (void);
+/* { dg-final { scan-assembler-times ".extern .func dcl_rvs8 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+vs8 dcl_rvs8 (void);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_rvi1 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+vi1 dcl_rvi1 (void);
+/* { dg-final { scan-assembler-times ".extern .func dcl_rvi2 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+vi2 dcl_rvi2 (void);
+/* { dg-final { scan-assembler-times ".extern .func dcl_rvi4 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+vi4 dcl_rvi4 (void);
+/* { dg-final { scan-assembler-times ".extern .func dcl_rvi8 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+vi8 dcl_rvi8 (void);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_rvll1 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+vll1 dcl_rvll1 (void);
+/* { dg-final { scan-assembler-times ".extern .func dcl_rvll2 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+vll2 dcl_rvll2 (void);
+/* { dg-final { scan-assembler-times ".extern .func dcl_rvll4 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+vll4 dcl_rvll4 (void);
+/* { dg-final { scan-assembler-times ".extern .func dcl_rvll8 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+vll8 dcl_rvll8 (void);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_rvf1 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+vf1 dcl_rvf1 (void);
+/* { dg-final { scan-assembler-times ".extern .func dcl_rvf2 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+vf2 dcl_rvf2 (void);
+/* { dg-final { scan-assembler-times ".extern .func dcl_rvf4 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+vf4 dcl_rvf4 (void);
+/* { dg-final { scan-assembler-times ".extern .func dcl_rvf8 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+vf8 dcl_rvf8 (void);
+
+/* { dg-final { scan-assembler-times ".extern .func dcl_rvd1 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+vd1 dcl_rvd1 (void);
+/* { dg-final { scan-assembler-times ".extern .func dcl_rvd2 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+vd2 dcl_rvd2 (void);
+/* { dg-final { scan-assembler-times ".extern .func dcl_rvd4 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+vd4 dcl_rvd4 (void);
+/* { dg-final { scan-assembler-times ".extern .func dcl_rvd8 \\(.param.u64 %\[_a-z0-9\]*\\);" 1 } } */
+vd8 dcl_rvd8 (void);
+
+void  test_1 (void)
+{
+  dcl_rvc1 ();
+  dcl_rvc2 ();
+  dcl_rvc4 ();
+  dcl_rvc8 ();
+  
+  dcl_rvs1 ();
+  dcl_rvs2 ();
+  dcl_rvs4 ();
+  dcl_rvs8 ();
+  
+  dcl_rvi1 ();
+  dcl_rvi2 ();
+  dcl_rvi4 ();
+  dcl_rvi8 ();
+
+  dcl_rvll1 ();
+  dcl_rvll2 ();
+  dcl_rvll4 ();
+  dcl_rvll8 ();
+  
+  dcl_rvf1 ();
+  dcl_rvf2 ();
+  dcl_rvf4 ();
+  dcl_rvf8 ();
+
+  dcl_rvd1 ();
+  dcl_rvd2 ();
+  dcl_rvd4 ();
+  dcl_rvd8 ();
+}
+
+#define M(T, V) ({T t;t[0]= V;t;})
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_rvc1 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+vc1 dfn_rvc1 (void)
+{
+  return M (vc1, 1);
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_rvc2 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+vc2 dfn_rvc2 (void)
+{
+  return M (vc2, 2);
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_rvc4 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+vc4 dfn_rvc4 (void)
+{
+  return M (vc4, 3);
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_rvc8 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+vc8 dfn_rvc8 (void)
+{
+  return M (vc8, 4);
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_rvs1 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+vs1 dfn_rvs1 (void)
+{
+  return M (vs1, 5);
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_rvs2 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+vs2 dfn_rvs2 (void)
+{
+  return M (vs2, 6);
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_rvs4 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+vs4 dfn_rvs4 (void)
+{
+  return M (vs4, 7);
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_rvs8 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+vs8 dfn_rvs8 (void)
+{
+  return M (vs8, 8);
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_rvi1 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+vi1 dfn_rvi1 (void)
+{
+  return M (vi1, 9);
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_rvi2 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+vi2 dfn_rvi2 (void)
+{
+  return M (vi2, 10);
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_rvi4 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+vi4 dfn_rvi4 (void)
+{
+  return M (vi4, 11);
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_rvi8 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+vi8 dfn_rvi8 (void)
+{
+  return M (vi8, 12);
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_rvll1 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+vll1 dfn_rvll1 (void)
+{
+  return M (vll1, 13);
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_rvll2 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+vll2 dfn_rvll2 (void)
+{
+  return M (vll2, 14);
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_rvll4 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+vll4 dfn_rvll4 (void)
+{
+  return M (vll4, 16);
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_rvll8 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+vll8 dfn_rvll8 (void)
+{
+  return M (vll8, 6);
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_rvf1 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+vf1 dfn_rvf1 (void)
+{
+  return M (vf1, 17);
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_rvf2 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+vf2 dfn_rvf2 (void)
+{
+  return M (vf2, 18);
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_rvf4 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+vf4 dfn_rvf4 (void)
+{
+  return M (vf4, 19);
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_rvf8 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+vf8 dfn_rvf8 (void)
+{
+  return M (vf8, 20);
+}
+
+/* { dg-final { scan-assembler-times ".visible .func dfn_rvd1 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+vd1 dfn_rvd1 (void)
+{
+  return M (vd1, 21);
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_rvd2 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+vd2 dfn_rvd2 (void)
+{
+  return M (vd2, 22);
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_rvd4 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+vd4 dfn_rvd4 (void)
+{
+  return M (vd4, 23);
+}
+/* { dg-final { scan-assembler-times ".visible .func dfn_rvd8 \\(.param.u64 %\[_a-z0-9\]*\\)(?:;|\[\r\n\]+\{)" 2 } } */
+vd8 dfn_rvd8 (void)
+{
+  return M (vd8, 24);
+}