[PATCH, rs6000] Early gimple folding of vec_mergeh and vec_mergel for float

[Will Schmidt]

Hi,
   This adds support for gimple folding of vec_mergeh and vec_mergel
for float and double types.   Support for the integral types is already
in-tree.
    
This change includes updates to the fold_mergehl_helper function to handle
creating the (integral) permute vector when the vector type non integer.

Relevant testcases are already in-tree.

OK for trunk?

Thanks,
-Will
    
[gcc]
    
2018-08-07 Will Schmidt  <will_schm...@vnet.ibm.com>
    
        * config/rs6000/rs600.c (rs6000_gimple_fold_builtin): Add entries to
        allow folding of mergeh() and mergel() for the float and double types.
        (fold_mergehl_helper): Rework to handle building a permute tree
        for float vectors.

diff --git a/gcc/config/rs6000/rs6000.c b/gcc/config/rs6000/rs6000.c
index 6cb5c87..35c32be 100644
--- a/gcc/config/rs6000/rs6000.c
+++ b/gcc/config/rs6000/rs6000.c
@@ -15119,24 +15119,39 @@ fold_mergehl_helper (gimple_stmt_iterator *gsi, 
gimple *stmt, int use_high)
 {
   tree arg0 = gimple_call_arg (stmt, 0);
   tree arg1 = gimple_call_arg (stmt, 1);
   tree lhs = gimple_call_lhs (stmt);
   tree lhs_type = TREE_TYPE (lhs);
-  tree lhs_type_type = TREE_TYPE (lhs_type);
   int n_elts = TYPE_VECTOR_SUBPARTS (lhs_type);
   int midpoint = n_elts / 2;
   int offset = 0;
 
   if (use_high == 1)
     offset = midpoint;
 
-  tree_vector_builder elts (lhs_type, VECTOR_CST_NELTS (arg0), 1);
+  /* The permute_type will match the lhs for integral types.  For double and
+     float types, the permute type needs to map to the V2 or V4 type that
+     matches size.  */
+  tree permute_type;
+  if (INTEGRAL_TYPE_P (TREE_TYPE (lhs_type)))
+    permute_type = lhs_type;
+  else
+    if (TREE_TYPE (lhs_type) == TREE_TYPE (V2DF_type_node))
+      permute_type = V2DI_type_node;
+    else if (TREE_TYPE (lhs_type) == TREE_TYPE (V4SF_type_node))
+      permute_type = V4SI_type_node;
+    else
+      gcc_unreachable ();
+
+  tree_vector_builder elts (permute_type, VECTOR_CST_NELTS (arg0), 1);
 
   for (int i = 0; i < midpoint; i++)
     {
-      elts.safe_push (build_int_cst (lhs_type_type, offset + i));
-      elts.safe_push (build_int_cst (lhs_type_type, offset + n_elts + i));
+      elts.safe_push (build_int_cst (TREE_TYPE (permute_type),
+                                    offset + i));
+      elts.safe_push (build_int_cst (TREE_TYPE (permute_type),
+                                    offset + n_elts + i));
     }
 
   tree permute = elts.build ();
 
   gimple *g = gimple_build_assign (lhs, VEC_PERM_EXPR, arg0, arg1, permute);
@@ -15757,18 +15772,22 @@ rs6000_gimple_fold_builtin (gimple_stmt_iterator *gsi)
     case ALTIVEC_BUILTIN_VMRGLH:
     case ALTIVEC_BUILTIN_VMRGLW:
     case VSX_BUILTIN_XXMRGLW_4SI:
     case ALTIVEC_BUILTIN_VMRGLB:
     case VSX_BUILTIN_VEC_MERGEL_V2DI:
+    case VSX_BUILTIN_XXMRGLW_4SF:
+    case VSX_BUILTIN_VEC_MERGEL_V2DF:
        fold_mergehl_helper (gsi, stmt, 1);
        return true;
     /* vec_mergeh (integrals).  */
     case ALTIVEC_BUILTIN_VMRGHH:
     case ALTIVEC_BUILTIN_VMRGHW:
     case VSX_BUILTIN_XXMRGHW_4SI:
     case ALTIVEC_BUILTIN_VMRGHB:
     case VSX_BUILTIN_VEC_MERGEH_V2DI:
+    case VSX_BUILTIN_XXMRGHW_4SF:
+    case VSX_BUILTIN_VEC_MERGEH_V2DF:
        fold_mergehl_helper (gsi, stmt, 0);
        return true;
 
     /* d = vec_pack (a, b) */
     case P8V_BUILTIN_VPKUDUM: