[gcc(refs/vendors/vrull/heads/slp-improvements)] Implement the missing transformation for SLP rearrange_pattern 3

[Philipp Tomsich]

https://gcc.gnu.org/g:89f3268cde0fcd58f45311f4d28b3ea1ee13af8c

commit 89f3268cde0fcd58f45311f4d28b3ea1ee13af8c
Author: Manolis Tsamis <manolis.tsamis@vrull.eu>
Date:   Wed Dec 13 11:31:45 2023 +0100

    Implement the missing transformation for SLP rearrange_pattern 3
    
    Ref #342

Diff:
---
 gcc/tree-vect-slp.cc | 41 ++++++++++++++++++++++++-----------------
 1 file changed, 24 insertions(+), 17 deletions(-)

diff --git a/gcc/tree-vect-slp.cc b/gcc/tree-vect-slp.cc
index f2ee40b25f9..4a9da25558a 100644
--- a/gcc/tree-vect-slp.cc
+++ b/gcc/tree-vect-slp.cc
@@ -1880,7 +1880,7 @@ try_rearrange_oprnd_info (vec<slp_oprnd_info> &oprnds_info, unsigned group_size)
       dump_printf (MSG_NOTE, " pattern detected.\n");
     }
 
-  if (pattern == 1)
+  if (pattern == 1 || pattern == 3)
     for (unsigned int j = 0; j < group_size; j += 4)
       {
 	oprnds_info[0]->def_stmts[j+2] = oprnds_info[1]->def_stmts[j];
@@ -1900,9 +1900,6 @@ try_rearrange_oprnd_info (vec<slp_oprnd_info> &oprnds_info, unsigned group_size)
 	oprnds_info[0]->def_stmts[j+3] = oprnds_info[1]->def_stmts[j+2];
 	oprnds_info[0]->ops[j+3] = oprnds_info[1]->ops[j+2];
       }
-  else if (pattern == 3)
-    /* No need to handle that for now.  */
-    return 0;
 
   if (dump_enabled_p ())
     {
@@ -2763,14 +2760,14 @@ fail:
       SLP_TREE_CHILDREN (one).safe_splice (children);
       SLP_TREE_CHILDREN (two).safe_splice (children);
 
+      SLP_TREE_CODE (one) = VEC_PERM_EXPR;
+      SLP_TREE_CODE (two) = VEC_PERM_EXPR;
+      unsigned int h = group_size / 2;
+      SLP_TREE_REPRESENTATIVE (one) = stmts[0];
+      SLP_TREE_REPRESENTATIVE (two) = stmts[h];
+
       if (rearrange_pattern == 1)
 	{
-	  SLP_TREE_CODE (one) = VEC_PERM_EXPR;
-	  SLP_TREE_CODE (two) = VEC_PERM_EXPR;
-	  unsigned int h = group_size / 2;
-	  SLP_TREE_REPRESENTATIVE (one) = stmts[0];
-	  SLP_TREE_REPRESENTATIVE (two) = stmts[h];
-
 	  for (unsigned int j = 0; j < h; j += 2)
 	    {
 	      SLP_TREE_LANE_PERMUTATION(one).safe_push (std::make_pair (0, j));
@@ -2788,12 +2785,6 @@ fail:
 	}
       else if (rearrange_pattern == 2)
 	{
-	  SLP_TREE_CODE (one) = VEC_PERM_EXPR;
-	  SLP_TREE_CODE (two) = VEC_PERM_EXPR;
-	  unsigned int h = group_size / 2;
-	  SLP_TREE_REPRESENTATIVE (one) = stmts[0];
-	  SLP_TREE_REPRESENTATIVE (two) = stmts[h];
-
 	  for (unsigned int j = 0; j < h; j += 2)
 	    {
 	      SLP_TREE_LANE_PERMUTATION(one).safe_push (std::make_pair (0, j));
@@ -2809,6 +2800,23 @@ fail:
 	      SLP_TREE_LANE_PERMUTATION(two).safe_push (std::make_pair (0, h + j + 1));
 	    }
 	}
+      else if (rearrange_pattern == 3)
+	{
+	  for (unsigned int j = 0; j < h; j += 2)
+	    {
+	      SLP_TREE_LANE_PERMUTATION(one).safe_push (std::make_pair (0, j));
+	      SLP_TREE_LANE_PERMUTATION(one).safe_push (std::make_pair (0, j + 1));
+	      SLP_TREE_LANE_PERMUTATION(one).safe_push (std::make_pair (0, j + 1));
+	      SLP_TREE_LANE_PERMUTATION(one).safe_push (std::make_pair (0, j));
+	    }
+	  for (unsigned int j = 0; j < h; j += 2)
+	    {
+	      SLP_TREE_LANE_PERMUTATION(two).safe_push (std::make_pair (0, h + j));
+	      SLP_TREE_LANE_PERMUTATION(two).safe_push (std::make_pair (0, h + j + 1));
+	      SLP_TREE_LANE_PERMUTATION(two).safe_push (std::make_pair (0, h + j + 1));
+	      SLP_TREE_LANE_PERMUTATION(two).safe_push (std::make_pair (0, h + j));
+	    }
+	}
 
       slp_tree child;
       FOR_EACH_VEC_ELT (SLP_TREE_CHILDREN (two), i, child)
@@ -2826,7 +2834,6 @@ fail:
       children.safe_splice(SLP_TREE_CHILDREN (node));
     }
 
-
   if (two_operators)
     {
       /* ???  We'd likely want to either cache in bst_map sth like