[PATCH] Allow all target supported load permutations in BB SLP

[PATCH] Allow all target supported load permutations in BB SLP

[Richard Biener]

Well - only those with no gaps in the groups with this patch.  More
as followup.

Bootstrapped and tested on x86_64-unknown-linux-gnu, applied.

Richard.

2015-05-06  Richard Biener  <rguenther@suse.de>

	* tree-vect-slp.c (vect_supported_load_permutation_p): Use
	vect_transform_slp_perm_load to check if we support a permutation
	for basic-block vectorization.

	* gcc.dg/vect/bb-slp-34.c: New testcase.

Index: gcc/tree-vect-slp.c
===================================================================
*** gcc/tree-vect-slp.c	(revision 222758)
--- gcc/tree-vect-slp.c	(working copy)
*************** vect_supported_load_permutation_p (slp_i
*** 1313,1329 ****
       FORNOW: not supported in loop SLP because of realignment compications.  */
    if (STMT_VINFO_BB_VINFO (vinfo_for_stmt (stmt)))
      {
!       /* Check that for every node in the instance the loads
! 	 form a subchain.  */
        FOR_EACH_VEC_ELT (SLP_INSTANCE_LOADS (slp_instn), i, node)
          {
            next_load = NULL;
            FOR_EACH_VEC_ELT (SLP_TREE_SCALAR_STMTS (node), j, load)
              {
                if (j != 0 && next_load != load)
! 		return false;
                next_load = GROUP_NEXT_ELEMENT (vinfo_for_stmt (load));
              }
          }
  
        /* Check that the alignment of the first load in every subchain, i.e.,
--- 1313,1348 ----
       FORNOW: not supported in loop SLP because of realignment compications.  */
    if (STMT_VINFO_BB_VINFO (vinfo_for_stmt (stmt)))
      {
!       /* Check whether the loads in an instance form a subchain and thus
!          no permutation is necessary.  */
        FOR_EACH_VEC_ELT (SLP_INSTANCE_LOADS (slp_instn), i, node)
          {
+ 	  bool subchain_p = true;
            next_load = NULL;
            FOR_EACH_VEC_ELT (SLP_TREE_SCALAR_STMTS (node), j, load)
              {
                if (j != 0 && next_load != load)
! 		{
! 		  subchain_p = false;
! 		  break;
! 		}
                next_load = GROUP_NEXT_ELEMENT (vinfo_for_stmt (load));
              }
+ 	  if (subchain_p)
+ 	    SLP_TREE_LOAD_PERMUTATION (node).release ();
+ 	  else
+ 	    {
+ 	      /* Verify the permutation can be generated.  */
+ 	      vec<tree> tem;
+ 	      if (!vect_transform_slp_perm_load (node, tem, NULL,
+ 						 1, slp_instn, true))
+ 		{
+ 		  dump_printf_loc (MSG_MISSED_OPTIMIZATION,
+ 				   vect_location,
+ 				   "unsupported load permutation\n");
+ 		  return false;
+ 		}
+ 	    }
          }
  
        /* Check that the alignment of the first load in every subchain, i.e.,
*************** vect_supported_load_permutation_p (slp_i
*** 1352,1360 ****
  	    }
  	}
  
-       /* We are done, no actual permutations need to be generated.  */
-       FOR_EACH_VEC_ELT (SLP_INSTANCE_LOADS (slp_instn), i, node)
- 	SLP_TREE_LOAD_PERMUTATION (node).release ();
        return true;
      }
  
--- 1371,1376 ----
Index: gcc/testsuite/gcc.dg/vect/bb-slp-34.c
===================================================================
*** gcc/testsuite/gcc.dg/vect/bb-slp-34.c	(revision 0)
--- gcc/testsuite/gcc.dg/vect/bb-slp-34.c	(working copy)
***************
*** 0 ****
--- 1,37 ----
+ /* { dg-require-effective-target vect_int } */
+ 
+ #include "tree-vect.h"
+ 
+ extern void abort (void);
+ 
+ int a[8], b[8];
+ 
+ void __attribute__((noinline,noclone))
+ foo(void)
+ {
+   a[0] = b[3];
+   a[1] = b[2];
+   a[2] = b[1];
+   a[3] = b[0];
+   a[4] = b[2];
+   a[5] = b[3];
+   a[6] = b[4];
+   a[7] = b[5];
+ }
+ 
+ int main()
+ {
+   int i;
+   check_vect ();
+   for (i = 0; i < 8; ++i)
+     b[i] = i;
+   foo ();
+   if (a[0] != 3 || a[1] != 2 || a[2] != 1 || a[3] != 0
+       || a[4] != 2 || a[5] != 3 || a[6] != 4 || a[7] != 5)
+     abort ();
+   return 0;
+ }
+ 
+ /* { dg-final { scan-tree-dump "basic block vectorized" "slp2" { target vect_perm } } } */
+ /* { dg-final { cleanup-tree-dump "slp1" } } */
+ /* { dg-final { cleanup-tree-dump "slp2" } } */