[rs6000] Add support for gimple folding vec_perm()

Message ID 1531163335.17604.41.camel@brimstone.rchland.ibm.com
State New
Headers show
Series
  • [rs6000] Add support for gimple folding vec_perm()
Related show

Commit Message

Will Schmidt July 9, 2018, 7:08 p.m.
Hi,
   Add support for early gimple folding of vec_perm.   Testcases are already in-tree as
gcc.target/powerpc/fold-vec-perm-*.c

OK for trunk?

Thanks,
-Will

[gcc]
    
2018-07-09  Will Schmidt  <will_schmidt@vnet.ibm.com>
    
	* gcc/config/rs6000/rs6000.c (rs6000_gimple_fold_builtin): Add support
	for folding vec_perm.

Comments

Segher Boessenkool July 11, 2018, 11:08 p.m. | #1
On Mon, Jul 09, 2018 at 02:08:55PM -0500, Will Schmidt wrote:
>    Add support for early gimple folding of vec_perm.   Testcases are already in-tree as
> gcc.target/powerpc/fold-vec-perm-*.c
> 
> OK for trunk?

Looks fine to me.  Okay if no one else complains :-)


Segher


> 	* gcc/config/rs6000/rs6000.c (rs6000_gimple_fold_builtin): Add support
> 	for folding vec_perm.

Patch

diff --git a/gcc/config/rs6000/rs6000.c b/gcc/config/rs6000/rs6000.c
index 1335661..7e4370c9 100644
--- a/gcc/config/rs6000/rs6000.c
+++ b/gcc/config/rs6000/rs6000.c
@@ -16155,10 +16155,41 @@  rs6000_gimple_fold_builtin (gimple_stmt_iterator *gsi)
     case ALTIVEC_BUILTIN_VUPKLPX:
       {
        return false;
       }
 
+    /* vec_perm.  */
+    case ALTIVEC_BUILTIN_VPERM_16QI:
+    case ALTIVEC_BUILTIN_VPERM_8HI:
+    case ALTIVEC_BUILTIN_VPERM_4SI:
+    case ALTIVEC_BUILTIN_VPERM_2DI:
+    case ALTIVEC_BUILTIN_VPERM_4SF:
+    case ALTIVEC_BUILTIN_VPERM_2DF:
+      {
+	arg0 = gimple_call_arg (stmt, 0);
+	arg1 = gimple_call_arg (stmt, 1);
+	tree permute = gimple_call_arg (stmt, 2);
+	lhs = gimple_call_lhs (stmt);
+	location_t loc = gimple_location (stmt);
+	gimple_seq stmts = NULL;
+	// convert arg0 and arg1 to match the type of the permute
+	// for the VEC_PERM_EXPR operation.
+	tree permute_type = (TREE_TYPE (permute));
+	tree arg0_ptype = gimple_convert (&stmts, loc, permute_type, arg0);
+	tree arg1_ptype = gimple_convert (&stmts, loc, permute_type, arg1);
+	tree lhs_ptype = gimple_build (&stmts, loc, VEC_PERM_EXPR,
+				      permute_type, arg0_ptype, arg1_ptype,
+				      permute);
+	// Convert the result back to the desired lhs type upon completion.
+	tree temp = gimple_convert (&stmts, loc, TREE_TYPE (lhs), lhs_ptype);
+	gsi_insert_seq_before (gsi, stmts, GSI_SAME_STMT);
+	g = gimple_build_assign (lhs, temp);
+	gimple_set_location (g, loc);
+	gsi_replace (gsi, g, true);
+	return true;
+      }
+
     default:
       if (TARGET_DEBUG_BUILTIN)
 	fprintf (stderr, "gimple builtin intrinsic not matched:%d %s %s\n",
 		 fn_code, fn_name1, fn_name2);
       break;