public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
* constexpr vector operations
@ 2012-11-29 13:16 Marc Glisse
  2012-11-29 15:18 ` Jason Merrill
  0 siblings, 1 reply; 2+ messages in thread
From: Marc Glisse @ 2012-11-29 13:16 UTC (permalink / raw)
  To: gcc-patches; +Cc: jason

[-- Attachment #1: Type: TEXT/PLAIN, Size: 713 bytes --]

Hello,

this patch fixes several ICEs when using constexpr SIMD vectors. Support 
for subscripting is still missing though, which is why I am not adding 
some static_asserts to the testcase. I don't use build_vector_from_ctor 
because it doesn't check for non-constant elements and doesn't handle 
vectors in a constructor (the middle-end produces those, and I expect the 
front-ends will too eventually).

Bootstrap+testsuite on x86_64-linux-gnu.

2012-11-29  Marc Glisse  <marc.glisse@inria.fr>

 	PR c++/53094
gcc/
 	* fold-const.c (fold): Replace a CONSTRUCTOR with a VECTOR_CST.
gcc/cp/
 	* cvt.c (ocp_convert): Call convert_to_vector.
gcc/testsuite/
 	* g++.dg/ext/vector20.C: New testcase.

-- 
Marc Glisse

[-- Attachment #2: Type: TEXT/PLAIN, Size: 3236 bytes --]

Index: gcc/cp/cvt.c
===================================================================
--- gcc/cp/cvt.c	(revision 193922)
+++ gcc/cp/cvt.c	(working copy)
@@ -683,20 +683,22 @@ ocp_convert (tree type, tree expr, int c
 	   might be expected, since if one of the types is a typedef;
 	   the comparison in fold is just equality of pointers, not a
 	   call to comptypes.  We don't call fold in this case because
 	   that can result in infinite recursion; fold will call
 	   convert, which will call ocp_convert, etc.  */
 	return e;
       /* For complex data types, we need to perform componentwise
 	 conversion.  */
       else if (TREE_CODE (type) == COMPLEX_TYPE)
 	return fold_if_not_in_template (convert_to_complex (type, e));
+      else if (TREE_CODE (type) == VECTOR_TYPE)
+	return fold_if_not_in_template (convert_to_vector (type, e));
       else if (TREE_CODE (e) == TARGET_EXPR)
 	{
 	  /* Don't build a NOP_EXPR of class type.  Instead, change the
 	     type of the temporary.  */
 	  TREE_TYPE (e) = TREE_TYPE (TARGET_EXPR_SLOT (e)) = type;
 	  return e;
 	}
       else
 	{
 	  /* We shouldn't be treating objects of ADDRESSABLE type as
Index: gcc/testsuite/g++.dg/ext/vector20.C
===================================================================
--- gcc/testsuite/g++.dg/ext/vector20.C	(revision 0)
+++ gcc/testsuite/g++.dg/ext/vector20.C	(revision 0)
@@ -0,0 +1,7 @@
+/* { dg-do compile } */
+/* { dg-options "-std=c++11" } */
+
+typedef long vec __attribute__((vector_size (2 * sizeof (long))));
+constexpr vec v = { 3, 4 };
+constexpr vec s = v + v;
+constexpr vec w = __builtin_shuffle (v, v);

Property changes on: gcc/testsuite/g++.dg/ext/vector20.C
___________________________________________________________________
Added: svn:keywords
   + Author Date Id Revision URL
Added: svn:eol-style
   + native

Index: gcc/fold-const.c
===================================================================
--- gcc/fold-const.c	(revision 193922)
+++ gcc/fold-const.c	(working copy)
@@ -14380,20 +14380,49 @@ fold (tree expr)
 			 && tree_int_cst_lt (op1, TREE_OPERAND (index, 0)))
 		  end = middle;
 		else
 		  return (*elts)[middle].value;
 	      }
 	  }
 
 	return t;
       }
 
+      /* Return a VECTOR_CST if possible.  */
+    case CONSTRUCTOR:
+      {
+	tree type = TREE_TYPE (t);
+	if (TREE_CODE (type) != VECTOR_TYPE)
+	  return t;
+
+	tree *vec = XALLOCAVEC (tree, TYPE_VECTOR_SUBPARTS (type));
+	unsigned HOST_WIDE_INT idx, pos = 0;
+	tree value;
+
+	FOR_EACH_CONSTRUCTOR_VALUE (CONSTRUCTOR_ELTS (t), idx, value)
+	  {
+	    if (!CONSTANT_CLASS_P (value))
+	      return t;
+	    if (TREE_CODE (value) == VECTOR_CST)
+	      {
+		for (unsigned i = 0; i < VECTOR_CST_NELTS (value); ++i)
+		  vec[pos++] = VECTOR_CST_ELT (value, i);
+	      }
+	    else
+	      vec[pos++] = value;
+	  }
+	for (; pos < TYPE_VECTOR_SUBPARTS (type); ++pos)
+	  vec[pos] = build_zero_cst (TREE_TYPE (type));
+
+	return build_vector (type, vec);
+      }
+
     case CONST_DECL:
       return fold (DECL_INITIAL (t));
 
     default:
       return t;
     } /* switch (code) */
 }
 
 #ifdef ENABLE_FOLD_CHECKING
 #undef fold

^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: constexpr vector operations
  2012-11-29 13:16 constexpr vector operations Marc Glisse
@ 2012-11-29 15:18 ` Jason Merrill
  0 siblings, 0 replies; 2+ messages in thread
From: Jason Merrill @ 2012-11-29 15:18 UTC (permalink / raw)
  To: Marc Glisse; +Cc: gcc-patches

OK.

Jason

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2012-11-29 15:18 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2012-11-29 13:16 constexpr vector operations Marc Glisse
2012-11-29 15:18 ` Jason Merrill

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).