Change vec_perm checking and expansion level.

The can_vec_perm_p interface changed to use a C integer array.  This
allows easy re-use from the rtl level and the gimple level within
the vectorizer.  It allows both to determine if a given permutation
is (un-)supported without having to create tree/rtl garbage.

The expand_vec_perm interface changed to use rtl.  This allows easy
re-use from the rtl level, so that expand_vec_perm can be used in the
fallback implementation of other optabs.

        * target.def (vec_perm_const_ok): Change parameters to mode and
        array of indicies.
        * doc/tm.texi: Rebuild.
        * config/i386/i386.c (ix86_vectorize_vec_perm_const_ok): Change
        parameters to mode and array of indicies.
        * expr.c (expand_expr_real_2) [VEC_PERM_EXPR]: Expand operands here.
        * optabs.c (can_vec_perm_p): Rename from can_vec_perm_expr_p.
        Change parameters to mode and array of indicies.
        (expand_vec_perm_1): Rename from expand_vec_perm_expr_1.
        (expand_vec_perm): Rename from expand_vec_perm_expr.  Change
        parameters to mode and rtx inputs.  Try lowering to QImode
        vec_perm_const before trying fully variable permutation.
        * optabs.h: Update decls.
        * tree-vect-generic.c (lower_vec_perm): Extract array of indices from
        VECTOR_CST to pass to can_vec_perm_p.
        * tree-vect-slp.c (vect_get_mask_element): Change mask parameter type
        from int pointer to unsigned char pointer.
        (vect_transform_slp_perm_load): Update for change to can_vec_perm_p.
        * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@180449 138bc75d-0d04-0410-961f-82ee72b054a4

10 files changed, 198 insertions, 148 deletions

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index b15e4717ea6..af51915eae5 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,5 +1,25 @@
 2011-10-25  Richard Henderson  <rth@redhat.com>
 
+	* target.def (vec_perm_const_ok): Change parameters to mode and
+	array of indicies.
+	* doc/tm.texi: Rebuild.
+	* config/i386/i386.c (ix86_vectorize_vec_perm_const_ok): Change
+	parameters to mode and array of indicies.
+	* expr.c (expand_expr_real_2) [VEC_PERM_EXPR]: Expand operands here.
+	* optabs.c (can_vec_perm_p): Rename from can_vec_perm_expr_p.
+	Change parameters to mode and array of indicies.
+	(expand_vec_perm_1): Rename from expand_vec_perm_expr_1.
+	(expand_vec_perm): Rename from expand_vec_perm_expr.  Change
+	parameters to mode and rtx inputs.  Try lowering to QImode
+	vec_perm_const before trying fully variable permutation.
+	* optabs.h: Update decls.
+	* tree-vect-generic.c (lower_vec_perm): Extract array of indices from
+	VECTOR_CST to pass to can_vec_perm_p.
+	* tree-vect-slp.c (vect_get_mask_element): Change mask parameter type
+	from int pointer to unsigned char pointer.
+	(vect_transform_slp_perm_load): Update for change to can_vec_perm_p.
+	* tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
+
 	* tree.def (VEC_EXTRACT_EVEN_EXPR): Fix typo in text name.
 	(VEC_EXTRACT_ODD_EXPR, VEC_INTERLEAVE_HIGH_EXPR,
 	VEC_INTERLEAVE_LOW_EXPR): Likewise.
diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
index e336d5abcd5..0d5063e0e7e 100644
--- a/gcc/config/i386/i386.c
+++ b/gcc/config/i386/i386.c
@@ -36477,14 +36477,14 @@ ix86_expand_vec_perm_const (rtx operands[4])
 /* Implement targetm.vectorize.vec_perm_const_ok.  */
 
 static bool
-ix86_vectorize_vec_perm_const_ok (tree vec_type, tree mask)
+ix86_vectorize_vec_perm_const_ok (enum machine_mode vmode,
+				  const unsigned char *sel)
 {
   struct expand_vec_perm_d d;
   unsigned int i, nelt, which;
   bool ret, one_vec;
-  tree list;
 
-  d.vmode = TYPE_MODE (vec_type);
+  d.vmode = vmode;
   d.nelt = nelt = GET_MODE_NUNITS (d.vmode);
   d.testing_p = true;
 
@@ -36505,19 +36505,13 @@ ix86_vectorize_vec_perm_const_ok (tree vec_type, tree mask)
 
   /* Extract the values from the vector CST into the permutation
      array in D.  */
-  list = TREE_VECTOR_CST_ELTS (mask);
-  for (i = which = 0; i < nelt; ++i, list = TREE_CHAIN (list))
+  memcpy (d.perm, sel, nelt);
+  for (i = which = 0; i < nelt; ++i)
     {
-      unsigned HOST_WIDE_INT e;
-
-      gcc_checking_assert (host_integerp (TREE_VALUE (list), 1));
-      e = tree_low_cst (TREE_VALUE (list), 1);
+      unsigned char e = d.perm[i];
       gcc_assert (e < 2 * nelt);
-
       which |= (e < nelt ? 1 : 2);
-      d.perm[i] = e;
     }
-  gcc_assert (list == NULL);
 
   /* For all elements from second vector, fold the elements to first.  */
   if (which == 2)
diff --git a/gcc/doc/tm.texi b/gcc/doc/tm.texi
index a43ce3d1099..422f74c5874 100644
--- a/gcc/doc/tm.texi
+++ b/gcc/doc/tm.texi
@@ -5711,7 +5711,7 @@ misalignment value (@var{misalign}).
 Return true if vector alignment is reachable (by peeling N iterations) for the given type.
 @end deftypefn
 
-@deftypefn {Target Hook} bool TARGET_VECTORIZE_VEC_PERM_CONST_OK (tree @var{vec_type}, tree @var{mask})
+@deftypefn {Target Hook} bool TARGET_VECTORIZE_VEC_PERM_CONST_OK (enum @var{machine_mode}, const unsigned char *@var{sel})
 Return true if a vector created for @code{vec_perm_const} is valid.
 @end deftypefn
 
diff --git a/gcc/expr.c b/gcc/expr.c
index a4cfee005d0..121db5eaf21 100644
--- a/gcc/expr.c
+++ b/gcc/expr.c
@@ -8752,9 +8752,11 @@ expand_expr_real_2 (sepops ops, rtx target, enum machine_mode tmode,
       goto binop;
 
     case VEC_PERM_EXPR:
-      target = expand_vec_perm_expr (type, treeop0, treeop1, treeop2, target);
-      gcc_assert (target);
-      return target;
+      expand_operands (treeop0, treeop1, target, &op0, &op1, EXPAND_NORMAL);
+      op2 = expand_normal (treeop2);
+      temp = expand_vec_perm (mode, op0, op1, op2, target);
+      gcc_assert (temp);
+      return temp;
 
     case DOT_PROD_EXPR:
       {
diff --git a/gcc/optabs.c b/gcc/optabs.c
index 5036856524a..26669f404ff 100644
--- a/gcc/optabs.c
+++ b/gcc/optabs.c
@@ -6701,20 +6701,22 @@ vector_compare_rtx (tree cond, bool unsignedp, enum insn_code icode)
    of the CPU.  SEL may be NULL, which stands for an unknown constant.  */
 
 bool
-can_vec_perm_expr_p (tree type, tree sel)
+can_vec_perm_p (enum machine_mode mode, bool variable,
+		const unsigned char *sel)
 {
-  enum machine_mode mode, qimode;
-  mode = TYPE_MODE (type);
+  enum machine_mode qimode;
 
   /* If the target doesn't implement a vector mode for the vector type,
      then no operations are supported.  */
   if (!VECTOR_MODE_P (mode))
     return false;
 
-  if (sel == NULL || TREE_CODE (sel) == VECTOR_CST)
+  if (!variable)
     {
       if (direct_optab_handler (vec_perm_const_optab, mode) != CODE_FOR_nothing
-	  && (sel == NULL || targetm.vectorize.vec_perm_const_ok (type, sel)))
+	  && (sel == NULL
+	      || targetm.vectorize.vec_perm_const_ok == NULL
+	      || targetm.vectorize.vec_perm_const_ok (mode, sel)))
 	return true;
     }
 
@@ -6722,6 +6724,8 @@ can_vec_perm_expr_p (tree type, tree sel)
     return true;
 
   /* We allow fallback to a QI vector mode, and adjust the mask.  */
+  if (GET_MODE_INNER (mode) == QImode)
+    return false;
   qimode = mode_for_vector (QImode, GET_MODE_SIZE (mode));
   if (!VECTOR_MODE_P (qimode))
     return false;
@@ -6732,9 +6736,9 @@ can_vec_perm_expr_p (tree type, tree sel)
   if (direct_optab_handler (vec_perm_optab, qimode) == CODE_FOR_nothing)
     return false;
 
-  /* In order to support the lowering of non-constant permutations,
+  /* In order to support the lowering of variable permutations,
      we need to support shifts and adds.  */
-  if (sel != NULL && TREE_CODE (sel) != VECTOR_CST)
+  if (variable)
     {
       if (GET_MODE_UNIT_SIZE (mode) > 2
 	  && optab_handler (ashl_optab, mode) == CODE_FOR_nothing
@@ -6747,11 +6751,11 @@ can_vec_perm_expr_p (tree type, tree sel)
   return true;
 }
 
-/* A subroutine of expand_vec_perm_expr for expanding one vec_perm insn.  */
+/* A subroutine of expand_vec_perm for expanding one vec_perm insn.  */
 
 static rtx
-expand_vec_perm_expr_1 (enum insn_code icode, rtx target,
-			rtx v0, rtx v1, rtx sel)
+expand_vec_perm_1 (enum insn_code icode, rtx target,
+		   rtx v0, rtx v1, rtx sel)
 {
   enum machine_mode tmode = GET_MODE (target);
   enum machine_mode smode = GET_MODE (sel);
@@ -6783,119 +6787,130 @@ expand_vec_perm_expr_1 (enum insn_code icode, rtx target,
   return NULL_RTX;
 }
 
-/* Generate instructions for VEC_PERM_EXPR given its type and three
-   operands.  */
+/* Generate instructions for vec_perm optab given its mode
+   and three operands.  */
+
 rtx
-expand_vec_perm_expr (tree type, tree v0, tree v1, tree sel, rtx target)
+expand_vec_perm (enum machine_mode mode, rtx v0, rtx v1, rtx sel, rtx target)
 {
   enum insn_code icode;
-  enum machine_mode mode = TYPE_MODE (type);
   enum machine_mode qimode;
-  rtx v0_rtx, v1_rtx, sel_rtx, *vec, vt, tmp;
   unsigned int i, w, e, u;
+  rtx tmp, sel_qi;
+  rtvec vec;
 
-  if (!target)
+  if (!target || GET_MODE (target) != mode)
     target = gen_reg_rtx (mode);
-  v0_rtx = expand_normal (v0);
-  if (operand_equal_p (v0, v1, 0))
-    v1_rtx = v0_rtx;
-  else
-    v1_rtx = expand_normal (v1);
-  sel_rtx = expand_normal (sel);
+
+  w = GET_MODE_SIZE (mode);
+  e = GET_MODE_NUNITS (mode);
+  u = GET_MODE_UNIT_SIZE (mode);
+
+  /* Set QIMODE to a different vector mode with byte elements.
+     If no such mode, or if MODE already has byte elements, use VOIDmode.  */
+  qimode = VOIDmode;
+  if (GET_MODE_INNER (mode) != QImode)
+    {
+      qimode = mode_for_vector (QImode, w);
+      if (!VECTOR_MODE_P (qimode))
+	qimode = VOIDmode;
+    }
 
   /* If the input is a constant, expand it specially.  */
-  if (CONSTANT_P (sel_rtx))
+  if (CONSTANT_P (sel))
     {
       icode = direct_optab_handler (vec_perm_const_optab, mode);
-      if (icode != CODE_FOR_nothing
-	  && targetm.vectorize.vec_perm_const_ok (TREE_TYPE (v0), sel)
-	  && (tmp = expand_vec_perm_expr_1 (icode, target, v0_rtx,
-					    v1_rtx, sel_rtx)) != NULL)
-	return tmp;
+      if (icode != CODE_FOR_nothing)
+	{
+	  tmp = expand_vec_perm_1 (icode, target, v0, v1, sel);
+	  if (tmp)
+	    return tmp;
+	}
+
+      /* Fall back to a constant byte-based permutation.  */
+      if (qimode != VOIDmode)
+	{
+	  icode = direct_optab_handler (vec_perm_const_optab, qimode);
+	  if (icode != CODE_FOR_nothing)
+	    {
+	      vec = rtvec_alloc (w);
+	      for (i = 0; i < e; ++i)
+		{
+		  unsigned int j, this_e;
+
+		  this_e = INTVAL (XVECEXP (sel, 0, i));
+		  this_e &= 2 * e - 1;
+		  this_e *= u;
+
+		  for (j = 0; j < u; ++j)
+		    RTVEC_ELT (vec, i * u + j) = GEN_INT (this_e + j);
+		}
+	      sel_qi = gen_rtx_CONST_VECTOR (qimode, vec);
+
+	      tmp = expand_vec_perm_1 (icode, gen_lowpart (qimode, target),
+				       gen_lowpart (qimode, v0),
+				       gen_lowpart (qimode, v1), sel_qi);
+	      if (tmp)
+		return gen_lowpart (mode, tmp);
+	    }
+	}
     }
 
-  /* Otherwise fall back to a fully variable permuation.  */
+  /* Otherwise expand as a fully variable permuation.  */
   icode = direct_optab_handler (vec_perm_optab, mode);
-  if (icode != CODE_FOR_nothing
-      && (tmp = expand_vec_perm_expr_1 (icode, target, v0_rtx,
-					v1_rtx, sel_rtx)) != NULL)
-    return tmp;
+  if (icode != CODE_FOR_nothing)
+    {
+      tmp = expand_vec_perm_1 (icode, target, v0, v1, sel);
+      if (tmp)
+	return tmp;
+    }
 
   /* As a special case to aid several targets, lower the element-based
      permutation to a byte-based permutation and try again.  */
-  qimode = mode_for_vector (QImode, GET_MODE_SIZE (mode));
-  if (!VECTOR_MODE_P (qimode))
+  if (qimode == VOIDmode)
     return NULL_RTX;
-
-  /* ??? For completeness, we ought to check the QImode version of
-     vec_perm_const_optab.  But all users of this implicit lowering
-     feature implement the variable vec_perm_optab.  */
   icode = direct_optab_handler (vec_perm_optab, qimode);
   if (icode == CODE_FOR_nothing)
     return NULL_RTX;
 
-  w = GET_MODE_SIZE (mode);
-  e = GET_MODE_NUNITS (mode);
-  u = GET_MODE_UNIT_SIZE (mode);
-  vec = XALLOCAVEC (rtx, w);
-
-  if (CONSTANT_P (sel_rtx))
-    {
-      unsigned int j;
-      for (i = 0; i < e; ++i)
-	{
-	  unsigned int this_e = INTVAL (XVECEXP (sel_rtx, 0, i));
-	  this_e &= 2 * e - 1;
-          this_e *= u;
-
-	  for (j = 0; j < u; ++j)
-	    vec[i * u + j] = GEN_INT (this_e + j);
-	}
-      sel_rtx = gen_rtx_CONST_VECTOR (qimode, gen_rtvec_v (w, vec));
-    }
+  /* Multiply each element by its byte size.  */
+  if (u == 2)
+    sel = expand_simple_binop (mode, PLUS, sel, sel, sel, 0, OPTAB_DIRECT);
   else
-    {
-      /* Multiply each element by its byte size.  */
-      if (u == 2)
-	sel_rtx = expand_simple_binop (mode, PLUS, sel_rtx, sel_rtx,
-				       sel_rtx, 0, OPTAB_DIRECT);
-      else
-	sel_rtx = expand_simple_binop (mode, ASHIFT, sel_rtx,
-				       GEN_INT (exact_log2 (u)),
-				       sel_rtx, 0, OPTAB_DIRECT);
-      gcc_assert (sel_rtx);
-
-      /* Broadcast the low byte each element into each of its bytes.  */
-      for (i = 0; i < w; ++i)
-	{
-	  int this_e = i / u * u;
-	  if (BYTES_BIG_ENDIAN)
-	    this_e += u - 1;
-	  vec[i] = GEN_INT (this_e);
-	}
-      vt = gen_rtx_CONST_VECTOR (qimode, gen_rtvec_v (w, vec));
-      sel_rtx = gen_lowpart (qimode, sel_rtx);
-      sel_rtx = expand_vec_perm_expr_1 (icode, gen_reg_rtx (qimode),
-					sel_rtx, sel_rtx, vt);
-      gcc_assert (sel_rtx != NULL);
-
-      /* Add the byte offset to each byte element.  */
-      /* Note that the definition of the indicies here is memory ordering,
-	 so there should be no difference between big and little endian.  */
-      for (i = 0; i < w; ++i)
-	vec[i] = GEN_INT (i % u);
-      vt = gen_rtx_CONST_VECTOR (qimode, gen_rtvec_v (w, vec));
-      sel_rtx = expand_simple_binop (qimode, PLUS, sel_rtx, vt,
-				     NULL_RTX, 0, OPTAB_DIRECT);
-      gcc_assert (sel_rtx);
-    }
-
-  tmp = expand_vec_perm_expr_1 (icode, gen_lowpart (qimode, target),
-			        gen_lowpart (qimode, v0_rtx),
-			        gen_lowpart (qimode, v1_rtx), sel_rtx);
-  gcc_assert (tmp != NULL);
-
-  return gen_lowpart (mode, tmp);
+    sel = expand_simple_binop (mode, ASHIFT, sel, GEN_INT (exact_log2 (u)),
+			       sel, 0, OPTAB_DIRECT);
+  gcc_assert (sel != NULL);
+
+  /* Broadcast the low byte each element into each of its bytes.  */
+  vec = rtvec_alloc (w);
+  for (i = 0; i < w; ++i)
+    {
+      int this_e = i / u * u;
+      if (BYTES_BIG_ENDIAN)
+	this_e += u - 1;
+      RTVEC_ELT (vec, i) = GEN_INT (this_e);
+    }
+  tmp = gen_rtx_CONST_VECTOR (qimode, vec);
+  sel = gen_lowpart (qimode, sel);
+  sel = expand_vec_perm (qimode, sel, sel, tmp, NULL);
+  gcc_assert (sel != NULL);
+
+  /* Add the byte offset to each byte element.  */
+  /* Note that the definition of the indicies here is memory ordering,
+     so there should be no difference between big and little endian.  */
+  vec = rtvec_alloc (w);
+  for (i = 0; i < w; ++i)
+    RTVEC_ELT (vec, i) = GEN_INT (i % u);
+  tmp = gen_rtx_CONST_VECTOR (qimode, vec);
+  sel = expand_simple_binop (qimode, PLUS, sel, tmp, sel, 0, OPTAB_DIRECT);
+  gcc_assert (sel != NULL);
+
+  tmp = expand_vec_perm_1 (icode, gen_lowpart (qimode, target),
+			   gen_lowpart (qimode, v0),
+			   gen_lowpart (qimode, v1), sel);
+  if (tmp)
+    tmp = gen_lowpart (mode, tmp);
+  return tmp;
 }
 
 
diff --git a/gcc/optabs.h b/gcc/optabs.h
index 86e0ec93bc1..9e3c5b06fc4 100644
--- a/gcc/optabs.h
+++ b/gcc/optabs.h
@@ -901,10 +901,10 @@ extern rtx expand_vec_cond_expr (tree, tree, tree, tree, rtx);
 extern rtx expand_vec_shift_expr (sepops, rtx);
 
 /* Return tree if target supports vector operations for VEC_PERM_EXPR.  */
-extern bool can_vec_perm_expr_p (tree, tree);
+extern bool can_vec_perm_p (enum machine_mode, bool, const unsigned char *);
 
 /* Generate code for VEC_PERM_EXPR.  */
-extern rtx expand_vec_perm_expr (tree, tree, tree, tree, rtx);
+extern rtx expand_vec_perm (enum machine_mode, rtx, rtx, rtx, rtx);
 
 /* Return the insn used to implement mode MODE of OP, or CODE_FOR_nothing
    if the target does not have such an insn.  */
diff --git a/gcc/target.def b/gcc/target.def
index 60fad2a813a..f89bb519369 100644
--- a/gcc/target.def
+++ b/gcc/target.def
@@ -985,12 +985,13 @@ DEFHOOK
  bool, (const_tree type, bool is_packed),
  default_builtin_vector_alignment_reachable)
 
-/* Return true if a vector created for vec_perm_const is valid.  */
+/* Return true if a vector created for vec_perm_const is valid.
+   A NULL indicates that all constants are valid permutations.  */
 DEFHOOK
 (vec_perm_const_ok,
  "",
- bool, (tree vec_type, tree mask),
- hook_bool_tree_tree_true)
+ bool, (enum machine_mode, const unsigned char *sel),
+ NULL)
 
 /* Return true if the target supports misaligned store/load of a
    specific factor denoted in the third parameter.  The last parameter
diff --git a/gcc/tree-vect-generic.c b/gcc/tree-vect-generic.c
index 37ff807b1b5..42ce2e3d729 100644
--- a/gcc/tree-vect-generic.c
+++ b/gcc/tree-vect-generic.c
@@ -641,13 +641,23 @@ lower_vec_perm (gimple_stmt_iterator *gsi)
   location_t loc = gimple_location (gsi_stmt (*gsi));
   unsigned i;
 
-  if (can_vec_perm_expr_p (vect_type, mask))
+  if (TREE_CODE (mask) == VECTOR_CST)
+    {
+      unsigned char *sel_int = XALLOCAVEC (unsigned char, elements);
+      tree vals = TREE_VECTOR_CST_ELTS (mask);
+
+      for (i = 0; i < elements; ++i, vals = TREE_CHAIN (vals))
+	sel_int[i] = TREE_INT_CST_LOW (TREE_VALUE (vals));
+
+      if (can_vec_perm_p (TYPE_MODE (vect_type), false, sel_int))
+	return;
+    }
+  else if (can_vec_perm_p (TYPE_MODE (vect_type), true, NULL))
     return;
   
   warning_at (loc, OPT_Wvector_operation_performance,
               "vector shuffling operation will be expanded piecewise");
 
-
   v = VEC_alloc (constructor_elt, gc, elements);
   for (i = 0; i < elements; i++)
     {
diff --git a/gcc/tree-vect-slp.c b/gcc/tree-vect-slp.c
index f75817b985f..790f2dd0a75 100644
--- a/gcc/tree-vect-slp.c
+++ b/gcc/tree-vect-slp.c
@@ -2356,7 +2356,7 @@ vect_create_mask_and_perm (gimple stmt, gimple next_scalar_stmt,
 static bool
 vect_get_mask_element (gimple stmt, int first_mask_element, int m,
                        int mask_nunits, bool only_one_vec, int index,
-                       int *mask, int *current_mask_element,
+		       unsigned char *mask, int *current_mask_element,
                        bool *need_next_vector, int *number_of_mask_fixes,
                        bool *mask_fixed, bool *needs_first_vector)
 {
@@ -2443,14 +2443,18 @@ vect_transform_slp_perm_load (gimple stmt, VEC (tree, heap) *dr_chain,
   gimple next_scalar_stmt;
   int group_size = SLP_INSTANCE_GROUP_SIZE (slp_node_instance);
   int first_mask_element;
-  int index, unroll_factor, *mask, current_mask_element, ncopies;
+  int index, unroll_factor, current_mask_element, ncopies;
+  unsigned char *mask;
   bool only_one_vec = false, need_next_vector = false;
   int first_vec_index, second_vec_index, orig_vec_stmts_num, vect_stmts_counter;
   int number_of_mask_fixes = 1;
   bool mask_fixed = false;
   bool needs_first_vector = false;
+  enum machine_mode mode;
 
-  if (!can_vec_perm_expr_p (vectype, NULL_TREE))
+  mode = TYPE_MODE (vectype);
+
+  if (!can_vec_perm_p (mode, false, NULL))
     {
       if (vect_print_dump_info (REPORT_DETAILS))
         {
@@ -2467,7 +2471,7 @@ vect_transform_slp_perm_load (gimple stmt, VEC (tree, heap) *dr_chain,
     (TREE_INT_CST_LOW (TYPE_SIZE (TREE_TYPE (vectype))), 1);
   mask_type = get_vectype_for_scalar_type (mask_element_type);
   nunits = TYPE_VECTOR_SUBPARTS (vectype);
-  mask = (int *) xmalloc (sizeof (int) * nunits);
+  mask = XALLOCAVEC (unsigned char, nunits);
   unroll_factor = SLP_INSTANCE_UNROLLING_FACTOR (slp_node_instance);
 
   /* The number of vector stmts to generate based only on SLP_NODE_INSTANCE
@@ -2529,6 +2533,18 @@ vect_transform_slp_perm_load (gimple stmt, VEC (tree, heap) *dr_chain,
                 {
 		  tree mask_vec = NULL;
 
+		  if (!can_vec_perm_p (mode, false, mask))
+		    {
+		      if (vect_print_dump_info (REPORT_DETAILS))
+			{
+			  fprintf (vect_dump, "unsupported vect permute { ");
+			  for (i = 0; i < nunits; ++i)
+			    fprintf (vect_dump, "%d ", mask[i]);
+			  fprintf (vect_dump, "}\n");
+			}
+		      return false;
+		    }
+
 		  while (--index >= 0)
 		    {
 		      tree t = build_int_cst (mask_element_type, mask[index]);
@@ -2537,17 +2553,6 @@ vect_transform_slp_perm_load (gimple stmt, VEC (tree, heap) *dr_chain,
 		  mask_vec = build_vector (mask_type, mask_vec);
 		  index = 0;
 
-		  if (!can_vec_perm_expr_p (vectype, mask_vec))
-		    {
-		      if (vect_print_dump_info (REPORT_DETAILS))
-			{
-			  fprintf (vect_dump, "unsupported vect permute ");
-			  print_generic_expr (vect_dump, mask_vec, 0);
-			}
-		      free (mask);
-		      return false;
-		    }
-
                   if (!analyze_only)
                     {
                       if (need_next_vector)
@@ -2569,7 +2574,6 @@ vect_transform_slp_perm_load (gimple stmt, VEC (tree, heap) *dr_chain,
         }
     }
 
-  free (mask);
   return true;
 }
 
diff --git a/gcc/tree-vect-stmts.c b/gcc/tree-vect-stmts.c
index 1aba74601c1..4cd582dad89 100644
--- a/gcc/tree-vect-stmts.c
+++ b/gcc/tree-vect-stmts.c
@@ -4090,25 +4090,29 @@ vectorizable_store (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt,
 static tree
 perm_mask_for_reverse (tree vectype)
 {
-  tree mask_element_type, mask_type, mask_vec = NULL;
+  tree mask_elt_type, mask_type, mask_vec;
   int i, nunits;
+  unsigned char *sel;
 
-  if (!can_vec_perm_expr_p (vectype, NULL_TREE))
+  nunits = TYPE_VECTOR_SUBPARTS (vectype);
+  sel = XALLOCAVEC (unsigned char, nunits);
+
+  for (i = 0; i < nunits; ++i)
+    sel[i] = nunits - 1 - i;
+
+  if (!can_vec_perm_p (TYPE_MODE (vectype), false, sel))
     return NULL;
 
-  mask_element_type
+  mask_elt_type
     = lang_hooks.types.type_for_size
     (TREE_INT_CST_LOW (TYPE_SIZE (TREE_TYPE (vectype))), 1);
-  mask_type = get_vectype_for_scalar_type (mask_element_type);
-  nunits = TYPE_VECTOR_SUBPARTS (vectype);
+  mask_type = get_vectype_for_scalar_type (mask_elt_type);
 
+  mask_vec = NULL;
   for (i = 0; i < nunits; i++)
-    mask_vec = tree_cons (NULL, build_int_cst (mask_element_type, i), mask_vec);
+    mask_vec = tree_cons (NULL, build_int_cst (mask_elt_type, i), mask_vec);
   mask_vec = build_vector (mask_type, mask_vec);
 
-  if (!can_vec_perm_expr_p (vectype, mask_vec))
-    return NULL;
-
   return mask_vec;
 }