* tree-vrp.c (execute_vrp): Do not pass loops structure through

[pf3gnuchains/gcc-fork.git] / gcc / tree-vect-transform.c

/* Transformation Utilities for Loop Vectorization.
   Copyright (C) 2003,2004,2005,2006 Free Software Foundation, Inc.
   Contributed by Dorit Naishlos <dorit@il.ibm.com>

This file is part of GCC.

GCC is free software; you can redistribute it and/or modify it under
the terms of the GNU General Public License as published by the Free
Software Foundation; either version 2, or (at your option) any later
version.

GCC is distributed in the hope that it will be useful, but WITHOUT ANY
WARRANTY; without even the implied warranty of MERCHANTABILITY or
FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
for more details.

You should have received a copy of the GNU General Public License
along with GCC; see the file COPYING.  If not, write to the Free
Software Foundation, 51 Franklin Street, Fifth Floor, Boston, MA
02110-1301, USA.  */

#include "config.h"
#include "system.h"
#include "coretypes.h"
#include "tm.h"
#include "ggc.h"
#include "tree.h"
#include "target.h"
#include "rtl.h"
#include "basic-block.h"
#include "diagnostic.h"
#include "tree-flow.h"
#include "tree-dump.h"
#include "timevar.h"
#include "cfgloop.h"
#include "expr.h"
#include "optabs.h"
#include "recog.h"
#include "tree-data-ref.h"
#include "tree-chrec.h"
#include "tree-scalar-evolution.h"
#include "tree-vectorizer.h"
#include "langhooks.h"
#include "tree-pass.h"
#include "toplev.h"
#include "real.h"

/* Utility functions for the code transformation.  */
static bool vect_transform_stmt (tree, block_stmt_iterator *, bool *);
static tree vect_create_destination_var (tree, tree);
static tree vect_create_data_ref_ptr 
  (tree, block_stmt_iterator *, tree, tree *, tree *, bool); 
static tree vect_create_addr_base_for_vector_ref (tree, tree *, tree);
static tree vect_setup_realignment (tree, block_stmt_iterator *, tree *);
static tree vect_get_new_vect_var (tree, enum vect_var_kind, const char *);
static tree vect_get_vec_def_for_operand (tree, tree, tree *);
static tree vect_init_vector (tree, tree);
static void vect_finish_stmt_generation 
  (tree stmt, tree vec_stmt, block_stmt_iterator *bsi);
static bool vect_is_simple_cond (tree, loop_vec_info); 
static void update_vuses_to_preheader (tree, struct loop*);
static void vect_create_epilog_for_reduction (tree, tree, enum tree_code, tree);
static tree get_initial_def_for_reduction (tree, tree, tree *);

/* Utility function dealing with loop peeling (not peeling itself).  */
static void vect_generate_tmps_on_preheader 
  (loop_vec_info, tree *, tree *, tree *);
static tree vect_build_loop_niters (loop_vec_info);
static void vect_update_ivs_after_vectorizer (loop_vec_info, tree, edge); 
static tree vect_gen_niters_for_prolog_loop (loop_vec_info, tree);
static void vect_update_init_of_dr (struct data_reference *, tree niters);
static void vect_update_inits_of_drs (loop_vec_info, tree);
static int vect_min_worthwhile_factor (enum tree_code);


/* Function vect_get_new_vect_var.

   Returns a name for a new variable. The current naming scheme appends the 
   prefix "vect_" or "vect_p" (depending on the value of VAR_KIND) to 
   the name of vectorizer generated variables, and appends that to NAME if 
   provided.  */

static tree
vect_get_new_vect_var (tree type, enum vect_var_kind var_kind, const char *name)
{
  const char *prefix;
  tree new_vect_var;

  switch (var_kind)
  {
  case vect_simple_var:
    prefix = "vect_";
    break;
  case vect_scalar_var:
    prefix = "stmp_";
    break;
  case vect_pointer_var:
    prefix = "vect_p";
    break;
  default:
    gcc_unreachable ();
  }

  if (name)
    new_vect_var = create_tmp_var (type, concat (prefix, name, NULL));
  else
    new_vect_var = create_tmp_var (type, prefix);

  return new_vect_var;
}


/* Function vect_create_addr_base_for_vector_ref.

   Create an expression that computes the address of the first memory location
   that will be accessed for a data reference.

   Input:
   STMT: The statement containing the data reference.
   NEW_STMT_LIST: Must be initialized to NULL_TREE or a statement list.
   OFFSET: Optional. If supplied, it is be added to the initial address.

   Output:
   1. Return an SSA_NAME whose value is the address of the memory location of 
      the first vector of the data reference.
   2. If new_stmt_list is not NULL_TREE after return then the caller must insert
      these statement(s) which define the returned SSA_NAME.

   FORNOW: We are only handling array accesses with step 1.  */

static tree
vect_create_addr_base_for_vector_ref (tree stmt,
                                      tree *new_stmt_list,
                                      tree offset)
{
  stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
  struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info);
  tree data_ref_base = unshare_expr (DR_BASE_ADDRESS (dr));
  tree base_name = build_fold_indirect_ref (data_ref_base);
  tree ref = DR_REF (dr);
  tree scalar_type = TREE_TYPE (ref);
  tree scalar_ptr_type = build_pointer_type (scalar_type);
  tree vec_stmt;
  tree new_temp;
  tree addr_base, addr_expr;
  tree dest, new_stmt;
  tree base_offset = unshare_expr (DR_OFFSET (dr));
  tree init = unshare_expr (DR_INIT (dr));

  /* Create base_offset */
  base_offset = size_binop (PLUS_EXPR, base_offset, init);
  dest = create_tmp_var (TREE_TYPE (base_offset), "base_off");
  add_referenced_var (dest);
  base_offset = force_gimple_operand (base_offset, &new_stmt, false, dest);  
  append_to_statement_list_force (new_stmt, new_stmt_list);

  if (offset)
    {
      tree tmp = create_tmp_var (TREE_TYPE (base_offset), "offset");
      tree step; 

      /* For interleaved access step we divide STEP by the size of the
        interleaving group.  */
      if (DR_GROUP_SIZE (stmt_info))
        step = fold_build2 (TRUNC_DIV_EXPR, TREE_TYPE (offset), DR_STEP (dr),
                            build_int_cst (TREE_TYPE (offset),
                                           DR_GROUP_SIZE (stmt_info)));
      else
        step = DR_STEP (dr);

      add_referenced_var (tmp);
      offset = fold_build2 (MULT_EXPR, TREE_TYPE (offset), offset, step);
      base_offset = fold_build2 (PLUS_EXPR, TREE_TYPE (base_offset),
                                 base_offset, offset);
      base_offset = force_gimple_operand (base_offset, &new_stmt, false, tmp);  
      append_to_statement_list_force (new_stmt, new_stmt_list);
    }
  
  /* base + base_offset */
  addr_base = fold_build2 (PLUS_EXPR, TREE_TYPE (data_ref_base), data_ref_base,
                           base_offset);

  /* addr_expr = addr_base */
  addr_expr = vect_get_new_vect_var (scalar_ptr_type, vect_pointer_var,
                                     get_name (base_name));
  add_referenced_var (addr_expr);
  vec_stmt = build2 (MODIFY_EXPR, void_type_node, addr_expr, addr_base);
  new_temp = make_ssa_name (addr_expr, vec_stmt);
  TREE_OPERAND (vec_stmt, 0) = new_temp;
  append_to_statement_list_force (vec_stmt, new_stmt_list);

  if (vect_print_dump_info (REPORT_DETAILS))
    {
      fprintf (vect_dump, "created ");
      print_generic_expr (vect_dump, vec_stmt, TDF_SLIM);
    }
  return new_temp;
}


/* Function vect_create_data_ref_ptr.

   Create a new pointer to vector type (vp), that points to the first location
   accessed in the loop by STMT, along with the def-use update chain to 
   appropriately advance the pointer through the loop iterations. Also set
   aliasing information for the pointer.  This vector pointer is used by the
   callers to this function to create a memory reference expression for vector 
   load/store access.

   Input:
   1. STMT: a stmt that references memory. Expected to be of the form
         MODIFY_EXPR <name, data-ref> or MODIFY_EXPR <data-ref, name>.
   2. BSI: block_stmt_iterator where new stmts can be added.
   3. OFFSET (optional): an offset to be added to the initial address accessed
        by the data-ref in STMT.
   4. ONLY_INIT: indicate if vp is to be updated in the loop, or remain
        pointing to the initial address.

   Output:
   1. Declare a new ptr to vector_type, and have it point to the base of the
      data reference (initial addressed accessed by the data reference).
      For example, for vector of type V8HI, the following code is generated:

      v8hi *vp;
      vp = (v8hi *)initial_address;

      if OFFSET is not supplied:
         initial_address = &a[init];
      if OFFSET is supplied:
         initial_address = &a[init + OFFSET];

      Return the initial_address in INITIAL_ADDRESS.

   2. If ONLY_INIT is true, just return the initial pointer.  Otherwise, also
      update the pointer in each iteration of the loop.  

      Return the increment stmt that updates the pointer in PTR_INCR.

   3. Return the pointer.  */

static tree
vect_create_data_ref_ptr (tree stmt,
                          block_stmt_iterator *bsi ATTRIBUTE_UNUSED,
                          tree offset, tree *initial_address, tree *ptr_incr,
                          bool only_init)
{
  tree base_name;
  stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
  loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
  struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
  tree vectype = STMT_VINFO_VECTYPE (stmt_info);
  tree vect_ptr_type;
  tree vect_ptr;
  tree tag;
  tree new_temp;
  tree vec_stmt;
  tree new_stmt_list = NULL_TREE;
  edge pe = loop_preheader_edge (loop);
  basic_block new_bb;
  tree vect_ptr_init;
  struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info);

  base_name =  build_fold_indirect_ref (unshare_expr (DR_BASE_ADDRESS (dr)));

  if (vect_print_dump_info (REPORT_DETAILS))
    {
      tree data_ref_base = base_name;
      fprintf (vect_dump, "create vector-pointer variable to type: ");
      print_generic_expr (vect_dump, vectype, TDF_SLIM);
      if (TREE_CODE (data_ref_base) == VAR_DECL)
        fprintf (vect_dump, "  vectorizing a one dimensional array ref: ");
      else if (TREE_CODE (data_ref_base) == ARRAY_REF)
        fprintf (vect_dump, "  vectorizing a multidimensional array ref: ");
      else if (TREE_CODE (data_ref_base) == COMPONENT_REF)
        fprintf (vect_dump, "  vectorizing a record based array ref: ");
      else if (TREE_CODE (data_ref_base) == SSA_NAME)
        fprintf (vect_dump, "  vectorizing a pointer ref: ");
      print_generic_expr (vect_dump, base_name, TDF_SLIM);
    }

  /** (1) Create the new vector-pointer variable:  **/

  vect_ptr_type = build_pointer_type (vectype);
  vect_ptr = vect_get_new_vect_var (vect_ptr_type, vect_pointer_var,
                                    get_name (base_name));
  add_referenced_var (vect_ptr);
  
  
  /** (2) Add aliasing information to the new vector-pointer:
          (The points-to info (DR_PTR_INFO) may be defined later.)  **/
  
  tag = DR_MEMTAG (dr);
  gcc_assert (tag);

  /* If tag is a variable (and NOT_A_TAG) than a new symbol memory
     tag must be created with tag added to its may alias list.  */
  if (!MTAG_P (tag))
    new_type_alias (vect_ptr, tag, DR_REF (dr));
  else
    var_ann (vect_ptr)->symbol_mem_tag = tag;

  var_ann (vect_ptr)->subvars = DR_SUBVARS (dr);

  /** (3) Calculate the initial address the vector-pointer, and set
          the vector-pointer to point to it before the loop:  **/

  /* Create: (&(base[init_val+offset]) in the loop preheader.  */
  new_temp = vect_create_addr_base_for_vector_ref (stmt, &new_stmt_list,
                                                   offset);
  pe = loop_preheader_edge (loop);
  new_bb = bsi_insert_on_edge_immediate (pe, new_stmt_list);
  gcc_assert (!new_bb);
  *initial_address = new_temp;

  /* Create: p = (vectype *) initial_base  */
  vec_stmt = fold_convert (vect_ptr_type, new_temp);
  vec_stmt = build2 (MODIFY_EXPR, void_type_node, vect_ptr, vec_stmt);
  vect_ptr_init = make_ssa_name (vect_ptr, vec_stmt);
  TREE_OPERAND (vec_stmt, 0) = vect_ptr_init;
  new_bb = bsi_insert_on_edge_immediate (pe, vec_stmt);
  gcc_assert (!new_bb);


  /** (4) Handle the updating of the vector-pointer inside the loop: **/

  if (only_init) /* No update in loop is required.  */
    {
      /* Copy the points-to information if it exists. */
      if (DR_PTR_INFO (dr))
        duplicate_ssa_name_ptr_info (vect_ptr_init, DR_PTR_INFO (dr));
      return vect_ptr_init;
    }
  else
    {
      block_stmt_iterator incr_bsi;
      bool insert_after;
      tree indx_before_incr, indx_after_incr;
      tree incr;

      standard_iv_increment_position (loop, &incr_bsi, &insert_after);
      create_iv (vect_ptr_init,
                 fold_convert (vect_ptr_type, TYPE_SIZE_UNIT (vectype)),
                 NULL_TREE, loop, &incr_bsi, insert_after,
                 &indx_before_incr, &indx_after_incr);
      incr = bsi_stmt (incr_bsi);
      set_stmt_info (stmt_ann (incr),
                     new_stmt_vec_info (incr, loop_vinfo));

      /* Copy the points-to information if it exists. */
      if (DR_PTR_INFO (dr))
        {
          duplicate_ssa_name_ptr_info (indx_before_incr, DR_PTR_INFO (dr));
          duplicate_ssa_name_ptr_info (indx_after_incr, DR_PTR_INFO (dr));
        }
      merge_alias_info (vect_ptr_init, indx_before_incr);
      merge_alias_info (vect_ptr_init, indx_after_incr);
      if (ptr_incr)
        *ptr_incr = incr;

      return indx_before_incr;
    }
}


/* Function bump_vector_ptr

   Increment a pointer (to a vector type) by vector-size. Connect the new 
   increment stmt to the exising def-use update-chain of the pointer.

   The pointer def-use update-chain before this function:
                        DATAREF_PTR = phi (p_0, p_2)
                        ....
        PTR_INCR:       p_2 = DATAREF_PTR + step 

   The pointer def-use update-chain after this function:
                        DATAREF_PTR = phi (p_0, p_2)
                        ....
                        NEW_DATAREF_PTR = DATAREF_PTR + vector_size
                        ....
        PTR_INCR:       p_2 = NEW_DATAREF_PTR + step

   Input:
   DATAREF_PTR - ssa_name of a pointer (to vector type) that is being updated 
                 in the loop.
   PTR_INCR - the stmt that updates the pointer in each iteration of the loop.
              The increment amount across iterations is also expected to be
              vector_size.      
   BSI - location where the new update stmt is to be placed.
   STMT - the original scalar memory-access stmt that is being vectorized.

   Output: Return NEW_DATAREF_PTR as illustrated above.
   
*/

static tree
bump_vector_ptr (tree dataref_ptr, tree ptr_incr, block_stmt_iterator *bsi,
                 tree stmt)
{
  stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
  struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info);
  tree vectype = STMT_VINFO_VECTYPE (stmt_info);
  tree vptr_type = TREE_TYPE (dataref_ptr);
  tree ptr_var = SSA_NAME_VAR (dataref_ptr);
  tree update = fold_convert (vptr_type, TYPE_SIZE_UNIT (vectype));
  tree incr_stmt;
  ssa_op_iter iter;
  use_operand_p use_p;
  tree new_dataref_ptr;

  incr_stmt = build2 (MODIFY_EXPR, void_type_node, ptr_var,
                build2 (PLUS_EXPR, vptr_type, dataref_ptr, update));
  new_dataref_ptr = make_ssa_name (ptr_var, incr_stmt);
  TREE_OPERAND (incr_stmt, 0) = new_dataref_ptr;
  vect_finish_stmt_generation (stmt, incr_stmt, bsi);

  /* Update the vector-pointer's cross-iteration increment.  */
  FOR_EACH_SSA_USE_OPERAND (use_p, ptr_incr, iter, SSA_OP_USE)
    {
      tree use = USE_FROM_PTR (use_p);

      if (use == dataref_ptr)
        SET_USE (use_p, new_dataref_ptr);
      else
        gcc_assert (tree_int_cst_compare (use, update) == 0);
    }

  /* Copy the points-to information if it exists. */
  if (DR_PTR_INFO (dr))
    duplicate_ssa_name_ptr_info (new_dataref_ptr, DR_PTR_INFO (dr));
  merge_alias_info (new_dataref_ptr, dataref_ptr);

  return new_dataref_ptr;
}


/* Function vect_create_destination_var.

   Create a new temporary of type VECTYPE.  */

static tree
vect_create_destination_var (tree scalar_dest, tree vectype)
{
  tree vec_dest;
  const char *new_name;
  tree type;
  enum vect_var_kind kind;

  kind = vectype ? vect_simple_var : vect_scalar_var;
  type = vectype ? vectype : TREE_TYPE (scalar_dest);

  gcc_assert (TREE_CODE (scalar_dest) == SSA_NAME);

  new_name = get_name (scalar_dest);
  if (!new_name)
    new_name = "var_";
  vec_dest = vect_get_new_vect_var (type, vect_simple_var, new_name);
  add_referenced_var (vec_dest);

  return vec_dest;
}


/* Function vect_init_vector.

   Insert a new stmt (INIT_STMT) that initializes a new vector variable with
   the vector elements of VECTOR_VAR. Return the DEF of INIT_STMT. It will be
   used in the vectorization of STMT.  */

static tree
vect_init_vector (tree stmt, tree vector_var)
{
  stmt_vec_info stmt_vinfo = vinfo_for_stmt (stmt);
  loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_vinfo);
  struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
  tree new_var;
  tree init_stmt;
  tree vectype = STMT_VINFO_VECTYPE (stmt_vinfo); 
  tree vec_oprnd;
  edge pe;
  tree new_temp;
  basic_block new_bb;
 
  new_var = vect_get_new_vect_var (vectype, vect_simple_var, "cst_");
  add_referenced_var (new_var); 
 
  init_stmt = build2 (MODIFY_EXPR, vectype, new_var, vector_var);
  new_temp = make_ssa_name (new_var, init_stmt);
  TREE_OPERAND (init_stmt, 0) = new_temp;

  pe = loop_preheader_edge (loop);
  new_bb = bsi_insert_on_edge_immediate (pe, init_stmt);
  gcc_assert (!new_bb);

  if (vect_print_dump_info (REPORT_DETAILS))
    {
      fprintf (vect_dump, "created new init_stmt: ");
      print_generic_expr (vect_dump, init_stmt, TDF_SLIM);
    }

  vec_oprnd = TREE_OPERAND (init_stmt, 0);
  return vec_oprnd;
}


/* Function vect_get_vec_def_for_operand.

   OP is an operand in STMT. This function returns a (vector) def that will be
   used in the vectorized stmt for STMT.

   In the case that OP is an SSA_NAME which is defined in the loop, then
   STMT_VINFO_VEC_STMT of the defining stmt holds the relevant def.

   In case OP is an invariant or constant, a new stmt that creates a vector def
   needs to be introduced.  */

static tree
vect_get_vec_def_for_operand (tree op, tree stmt, tree *scalar_def)
{
  tree vec_oprnd;
  tree vec_stmt;
  tree def_stmt;
  stmt_vec_info def_stmt_info = NULL;
  stmt_vec_info stmt_vinfo = vinfo_for_stmt (stmt);
  tree vectype = STMT_VINFO_VECTYPE (stmt_vinfo);
  int nunits = TYPE_VECTOR_SUBPARTS (vectype);
  loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_vinfo);
  struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
  tree vec_inv;
  tree vec_cst;
  tree t = NULL_TREE;
  tree def;
  int i;
  enum vect_def_type dt;
  bool is_simple_use;

  if (vect_print_dump_info (REPORT_DETAILS))
    {
      fprintf (vect_dump, "vect_get_vec_def_for_operand: ");
      print_generic_expr (vect_dump, op, TDF_SLIM);
    }

  is_simple_use = vect_is_simple_use (op, loop_vinfo, &def_stmt, &def, &dt);
  gcc_assert (is_simple_use);
  if (vect_print_dump_info (REPORT_DETAILS))
    {
      if (def)
        {
          fprintf (vect_dump, "def =  ");
          print_generic_expr (vect_dump, def, TDF_SLIM);
        }
      if (def_stmt)
        {
          fprintf (vect_dump, "  def_stmt =  ");
          print_generic_expr (vect_dump, def_stmt, TDF_SLIM);
        }
    }

  switch (dt)
    {
    /* Case 1: operand is a constant.  */
    case vect_constant_def:
      {
        if (scalar_def) 
          *scalar_def = op;

        /* Create 'vect_cst_ = {cst,cst,...,cst}'  */
        if (vect_print_dump_info (REPORT_DETAILS))
          fprintf (vect_dump, "Create vector_cst. nunits = %d", nunits);

        for (i = nunits - 1; i >= 0; --i)
          {
            t = tree_cons (NULL_TREE, op, t);
          }
        vec_cst = build_vector (vectype, t);
        return vect_init_vector (stmt, vec_cst);
      }

    /* Case 2: operand is defined outside the loop - loop invariant.  */
    case vect_invariant_def:
      {
        if (scalar_def) 
          *scalar_def = def;

        /* Create 'vec_inv = {inv,inv,..,inv}'  */
        if (vect_print_dump_info (REPORT_DETAILS))
          fprintf (vect_dump, "Create vector_inv.");

        for (i = nunits - 1; i >= 0; --i)
          {
            t = tree_cons (NULL_TREE, def, t);
          }

        /* FIXME: use build_constructor directly.  */
        vec_inv = build_constructor_from_list (vectype, t);
        return vect_init_vector (stmt, vec_inv);
      }

    /* Case 3: operand is defined inside the loop.  */
    case vect_loop_def:
      {
        if (scalar_def) 
          *scalar_def = def_stmt;

        /* Get the def from the vectorized stmt.  */
        def_stmt_info = vinfo_for_stmt (def_stmt);
        vec_stmt = STMT_VINFO_VEC_STMT (def_stmt_info);
        gcc_assert (vec_stmt);
        vec_oprnd = TREE_OPERAND (vec_stmt, 0);
        return vec_oprnd;
      }

    /* Case 4: operand is defined by a loop header phi - reduction  */
    case vect_reduction_def:
      {
        gcc_assert (TREE_CODE (def_stmt) == PHI_NODE);

        /* Get the def before the loop  */
        op = PHI_ARG_DEF_FROM_EDGE (def_stmt, loop_preheader_edge (loop));
        return get_initial_def_for_reduction (stmt, op, scalar_def);
     }

    /* Case 5: operand is defined by loop-header phi - induction.  */
    case vect_induction_def:
      {
        if (vect_print_dump_info (REPORT_DETAILS))
          fprintf (vect_dump, "induction - unsupported.");
        internal_error ("no support for induction"); /* FORNOW */
      }

    default:
      gcc_unreachable ();
    }
}


/* Function vect_get_vec_def_for_stmt_copy

   Return a vector-def for an operand. This function is used when the 
   vectorized stmt to be created (by the caller to this function) is a "copy" 
   created in case the vectorized result cannot fit in one vector, and several 
   copies of the vector-stmt are required. In this case the vector-def is 
   retrieved from the vector stmt recorded in the STMT_VINFO_RELATED_STMT field 
   of the stmt that defines VEC_OPRND. 
   DT is the type of the vector def VEC_OPRND.

   Context:
        In case the vectorization factor (VF) is bigger than the number
   of elements that can fit in a vectype (nunits), we have to generate
   more than one vector stmt to vectorize the scalar stmt. This situation
   arises when there are multiple data-types operated upon in the loop; the 
   smallest data-type determines the VF, and as a result, when vectorizing
   stmts operating on wider types we need to create 'VF/nunits' "copies" of the
   vector stmt (each computing a vector of 'nunits' results, and together
   computing 'VF' results in each iteration).  This function is called when 
   vectorizing such a stmt (e.g. vectorizing S2 in the illusration below, in 
   which VF=16 and nuniti=4, so the number of copies required is 4):

   scalar stmt:         vectorized into:        STMT_VINFO_RELATED_STMT
 
   S1: x = load         VS1.0:  vx.0 = memref0      VS1.1
                        VS1.1:  vx.1 = memref1      VS1.2
                        VS1.2:  vx.2 = memref2      VS1.3
                        VS1.3:  vx.3 = memref3 

   S2: z = x + ...      VSnew.0:  vz0 = vx.0 + ...  VSnew.1
                        VSnew.1:  vz1 = vx.1 + ...  VSnew.2
                        VSnew.2:  vz2 = vx.2 + ...  VSnew.3
                        VSnew.3:  vz3 = vx.3 + ...

   The vectorization of S1 is explained in vectorizable_load.
   The vectorization of S2:
        To create the first vector-stmt out of the 4 copies - VSnew.0 - 
   the function 'vect_get_vec_def_for_operand' is called to 
   get the relevant vector-def for each operand of S2. For operand x it
   returns  the vector-def 'vx.0'.

        To create the remaining copies of the vector-stmt (VSnew.j), this 
   function is called to get the relevant vector-def for each operand.  It is 
   obtained from the respective VS1.j stmt, which is recorded in the 
   STMT_VINFO_RELATED_STMT field of the stmt that defines VEC_OPRND.

        For example, to obtain the vector-def 'vx.1' in order to create the 
   vector stmt 'VSnew.1', this function is called with VEC_OPRND='vx.0'. 
   Given 'vx0' we obtain the stmt that defines it ('VS1.0'); from the 
   STMT_VINFO_RELATED_STMT field of 'VS1.0' we obtain the next copy - 'VS1.1',
   and return its def ('vx.1').
   Overall, to create the above sequence this function will be called 3 times:
        vx.1 = vect_get_vec_def_for_stmt_copy (dt, vx.0);
        vx.2 = vect_get_vec_def_for_stmt_copy (dt, vx.1);
        vx.3 = vect_get_vec_def_for_stmt_copy (dt, vx.2);  */

static tree
vect_get_vec_def_for_stmt_copy (enum vect_def_type dt, tree vec_oprnd)
{
  tree vec_stmt_for_operand;
  stmt_vec_info def_stmt_info;

  if (dt == vect_invariant_def || dt == vect_constant_def)
    {
      /* Do nothing; can reuse same def.  */ ;
      return vec_oprnd;
    }

  vec_stmt_for_operand = SSA_NAME_DEF_STMT (vec_oprnd);
  def_stmt_info = vinfo_for_stmt (vec_stmt_for_operand);
  gcc_assert (def_stmt_info);
  vec_stmt_for_operand = STMT_VINFO_RELATED_STMT (def_stmt_info);
  gcc_assert (vec_stmt_for_operand);
  vec_oprnd = TREE_OPERAND (vec_stmt_for_operand, 0);

  return vec_oprnd;
}


/* Function vect_finish_stmt_generation.

   Insert a new stmt.  */

static void
vect_finish_stmt_generation (tree stmt, tree vec_stmt, 
                             block_stmt_iterator *bsi)
{
  stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
  loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);

  bsi_insert_before (bsi, vec_stmt, BSI_SAME_STMT);
  set_stmt_info (get_stmt_ann (vec_stmt), 
                 new_stmt_vec_info (vec_stmt, loop_vinfo)); 

  if (vect_print_dump_info (REPORT_DETAILS))
    {
      fprintf (vect_dump, "add new stmt: ");
      print_generic_expr (vect_dump, vec_stmt, TDF_SLIM);
    }

  /* Make sure bsi points to the stmt that is being vectorized.  */
  gcc_assert (stmt == bsi_stmt (*bsi));

#ifdef USE_MAPPED_LOCATION
  SET_EXPR_LOCATION (vec_stmt, EXPR_LOCATION (stmt));
#else
  SET_EXPR_LOCUS (vec_stmt, EXPR_LOCUS (stmt));
#endif
}


#define ADJUST_IN_EPILOG 1

/* Function get_initial_def_for_reduction

   Input:
   STMT - a stmt that performs a reduction operation in the loop.
   INIT_VAL - the initial value of the reduction variable

   Output:
   SCALAR_DEF - a tree that holds a value to be added to the final result
        of the reduction (used for "ADJUST_IN_EPILOG" - see below).
   Return a vector variable, initialized according to the operation that STMT
        performs. This vector will be used as the initial value of the
        vector of partial results.

   Option1 ("ADJUST_IN_EPILOG"): Initialize the vector as follows:
     add:         [0,0,...,0,0]
     mult:        [1,1,...,1,1]
     min/max:     [init_val,init_val,..,init_val,init_val]
     bit and/or:  [init_val,init_val,..,init_val,init_val]
   and when necessary (e.g. add/mult case) let the caller know 
   that it needs to adjust the result by init_val.

   Option2: Initialize the vector as follows:
     add:         [0,0,...,0,init_val]
     mult:        [1,1,...,1,init_val]
     min/max:     [init_val,init_val,...,init_val]
     bit and/or:  [init_val,init_val,...,init_val]
   and no adjustments are needed.

   For example, for the following code:

   s = init_val;
   for (i=0;i<n;i++)
     s = s + a[i];

   STMT is 's = s + a[i]', and the reduction variable is 's'.
   For a vector of 4 units, we want to return either [0,0,0,init_val],
   or [0,0,0,0] and let the caller know that it needs to adjust
   the result at the end by 'init_val'.

   FORNOW: We use the "ADJUST_IN_EPILOG" scheme.
   TODO: Use some cost-model to estimate which scheme is more profitable.
*/

static tree
get_initial_def_for_reduction (tree stmt, tree init_val, tree *scalar_def)
{
  stmt_vec_info stmt_vinfo = vinfo_for_stmt (stmt);
  tree vectype = STMT_VINFO_VECTYPE (stmt_vinfo);
  int nunits = GET_MODE_NUNITS (TYPE_MODE (vectype));
  int nelements;
  enum tree_code code = TREE_CODE (TREE_OPERAND (stmt, 1));
  tree type = TREE_TYPE (init_val);
  tree def;
  tree vec, t = NULL_TREE;
  bool need_epilog_adjust;
  int i;

  gcc_assert (INTEGRAL_TYPE_P (type) || SCALAR_FLOAT_TYPE_P (type));

  switch (code)
  {
  case WIDEN_SUM_EXPR:
  case DOT_PROD_EXPR:
  case PLUS_EXPR:
    if (INTEGRAL_TYPE_P (type))
      def = build_int_cst (type, 0);
    else
      def = build_real (type, dconst0);

#ifdef ADJUST_IN_EPILOG
    /* All the 'nunits' elements are set to 0. The final result will be
       adjusted by 'init_val' at the loop epilog.  */
    nelements = nunits;
    need_epilog_adjust = true;
#else
    /* 'nunits - 1' elements are set to 0; The last element is set to 
        'init_val'.  No further adjustments at the epilog are needed.  */
    nelements = nunits - 1;
    need_epilog_adjust = false;
#endif
    break;

  case MIN_EXPR:
  case MAX_EXPR:
    def = init_val;
    nelements = nunits;
    need_epilog_adjust = false;
    break;

  default:
    gcc_unreachable ();
  }

  for (i = nelements - 1; i >= 0; --i)
    t = tree_cons (NULL_TREE, def, t);

  if (nelements == nunits - 1)
    {
      /* Set the last element of the vector.  */
      t = tree_cons (NULL_TREE, init_val, t);
      nelements += 1;
    }
  gcc_assert (nelements == nunits);
  
  if (TREE_CODE (init_val) == INTEGER_CST || TREE_CODE (init_val) == REAL_CST)
    vec = build_vector (vectype, t);
  else
    vec = build_constructor_from_list (vectype, t);
    
  if (!need_epilog_adjust)
    *scalar_def = NULL_TREE;
  else
    *scalar_def = init_val;

  return vect_init_vector (stmt, vec);
}


/* Function vect_create_epilog_for_reduction
    
   Create code at the loop-epilog to finalize the result of a reduction
   computation. 
  
   VECT_DEF is a vector of partial results. 
   REDUC_CODE is the tree-code for the epilog reduction.
   STMT is the scalar reduction stmt that is being vectorized.
   REDUCTION_PHI is the phi-node that carries the reduction computation.

   This function:
   1. Creates the reduction def-use cycle: sets the the arguments for 
      REDUCTION_PHI:
      The loop-entry argument is the vectorized initial-value of the reduction.
      The loop-latch argument is VECT_DEF - the vector of partial sums.
   2. "Reduces" the vector of partial results VECT_DEF into a single result,
      by applying the operation specified by REDUC_CODE if available, or by 
      other means (whole-vector shifts or a scalar loop).
      The function also creates a new phi node at the loop exit to preserve 
      loop-closed form, as illustrated below.
  
     The flow at the entry to this function:
    
        loop:
          vec_def = phi <null, null>            # REDUCTION_PHI
          VECT_DEF = vector_stmt                # vectorized form of STMT       
          s_loop = scalar_stmt                  # (scalar) STMT
        loop_exit:
          s_out0 = phi <s_loop>                 # (scalar) EXIT_PHI
          use <s_out0>
          use <s_out0>

     The above is transformed by this function into:

        loop:
          vec_def = phi <vec_init, VECT_DEF>    # REDUCTION_PHI
          VECT_DEF = vector_stmt                # vectorized form of STMT
          s_loop = scalar_stmt                  # (scalar) STMT 
        loop_exit:
          s_out0 = phi <s_loop>                 # (scalar) EXIT_PHI
          v_out1 = phi <VECT_DEF>               # NEW_EXIT_PHI
          v_out2 = reduce <v_out1>
          s_out3 = extract_field <v_out2, 0>
          s_out4 = adjust_result <s_out3>
          use <s_out4>
          use <s_out4>
*/

static void
vect_create_epilog_for_reduction (tree vect_def, tree stmt,
                                  enum tree_code reduc_code, tree reduction_phi)
{
  stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
  tree vectype;
  enum machine_mode mode;
  loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
  struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
  basic_block exit_bb;
  tree scalar_dest;
  tree scalar_type;
  tree new_phi;
  block_stmt_iterator exit_bsi;
  tree vec_dest;
  tree new_temp;
  tree new_name;
  tree epilog_stmt;
  tree new_scalar_dest, exit_phi;
  tree bitsize, bitpos, bytesize; 
  enum tree_code code = TREE_CODE (TREE_OPERAND (stmt, 1));
  tree scalar_initial_def;
  tree vec_initial_def;
  tree orig_name;
  imm_use_iterator imm_iter;
  use_operand_p use_p;
  bool extract_scalar_result;
  tree reduction_op;
  tree orig_stmt;
  tree use_stmt;
  tree operation = TREE_OPERAND (stmt, 1);
  int op_type;
  
  op_type = TREE_CODE_LENGTH (TREE_CODE (operation));
  reduction_op = TREE_OPERAND (operation, op_type-1);
  vectype = get_vectype_for_scalar_type (TREE_TYPE (reduction_op));
  mode = TYPE_MODE (vectype);

  /*** 1. Create the reduction def-use cycle  ***/
  
  /* 1.1 set the loop-entry arg of the reduction-phi:  */
  /* For the case of reduction, vect_get_vec_def_for_operand returns
     the scalar def before the loop, that defines the initial value
     of the reduction variable.  */
  vec_initial_def = vect_get_vec_def_for_operand (reduction_op, stmt,
                                                  &scalar_initial_def);
  add_phi_arg (reduction_phi, vec_initial_def, loop_preheader_edge (loop));

  /* 1.2 set the loop-latch arg for the reduction-phi:  */
  add_phi_arg (reduction_phi, vect_def, loop_latch_edge (loop));

  if (vect_print_dump_info (REPORT_DETAILS))
    {
      fprintf (vect_dump, "transform reduction: created def-use cycle:");
      print_generic_expr (vect_dump, reduction_phi, TDF_SLIM);
      fprintf (vect_dump, "\n");
      print_generic_expr (vect_dump, SSA_NAME_DEF_STMT (vect_def), TDF_SLIM);
    }


  /*** 2. Create epilog code
          The reduction epilog code operates across the elements of the vector
          of partial results computed by the vectorized loop.
          The reduction epilog code consists of:
          step 1: compute the scalar result in a vector (v_out2)
          step 2: extract the scalar result (s_out3) from the vector (v_out2)
          step 3: adjust the scalar result (s_out3) if needed.

          Step 1 can be accomplished using one the following three schemes:
          (scheme 1) using reduc_code, if available.
          (scheme 2) using whole-vector shifts, if available.
          (scheme 3) using a scalar loop. In this case steps 1+2 above are 
                     combined.
                
          The overall epilog code looks like this:

          s_out0 = phi <s_loop>         # original EXIT_PHI
          v_out1 = phi <VECT_DEF>       # NEW_EXIT_PHI
          v_out2 = reduce <v_out1>              # step 1
          s_out3 = extract_field <v_out2, 0>    # step 2
          s_out4 = adjust_result <s_out3>       # step 3

          (step 3 is optional, and step2 1 and 2 may be combined).
          Lastly, the uses of s_out0 are replaced by s_out4.

          ***/

  /* 2.1 Create new loop-exit-phi to preserve loop-closed form:
        v_out1 = phi <v_loop>  */

  exit_bb = single_exit (loop)->dest;
  new_phi = create_phi_node (SSA_NAME_VAR (vect_def), exit_bb);
  SET_PHI_ARG_DEF (new_phi, single_exit (loop)->dest_idx, vect_def);
  exit_bsi = bsi_start (exit_bb);

  /* 2.2 Get the relevant tree-code to use in the epilog for schemes 2,3 
         (i.e. when reduc_code is not available) and in the final adjustment code
         (if needed).  Also get the original scalar reduction variable as
         defined in the loop.  In case STMT is a "pattern-stmt" (i.e. - it 
         represents a reduction pattern), the tree-code and scalar-def are 
         taken from the original stmt that the pattern-stmt (STMT) replaces.  
         Otherwise (it is a regular reduction) - the tree-code and scalar-def
         are taken from STMT.  */ 

  orig_stmt = STMT_VINFO_RELATED_STMT (stmt_info);
  if (!orig_stmt)
    {
      /* Regular reduction  */
      orig_stmt = stmt;
    }
  else
    {
      /* Reduction pattern  */
      stmt_vec_info stmt_vinfo = vinfo_for_stmt (orig_stmt);
      gcc_assert (STMT_VINFO_IN_PATTERN_P (stmt_vinfo));
      gcc_assert (STMT_VINFO_RELATED_STMT (stmt_vinfo) == stmt);
    }
  code = TREE_CODE (TREE_OPERAND (orig_stmt, 1));
  scalar_dest = TREE_OPERAND (orig_stmt, 0);
  scalar_type = TREE_TYPE (scalar_dest);
  new_scalar_dest = vect_create_destination_var (scalar_dest, NULL);
  bitsize = TYPE_SIZE (scalar_type);
  bytesize = TYPE_SIZE_UNIT (scalar_type);

  /* 2.3 Create the reduction code, using one of the three schemes described
         above.  */

  if (reduc_code < NUM_TREE_CODES)
    {
      /*** Case 1:  Create:
           v_out2 = reduc_expr <v_out1>  */

      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "Reduce using direct vector reduction.");

      vec_dest = vect_create_destination_var (scalar_dest, vectype);
      epilog_stmt = build2 (MODIFY_EXPR, vectype, vec_dest,
                        build1 (reduc_code, vectype,  PHI_RESULT (new_phi)));
      new_temp = make_ssa_name (vec_dest, epilog_stmt);
      TREE_OPERAND (epilog_stmt, 0) = new_temp;
      bsi_insert_after (&exit_bsi, epilog_stmt, BSI_NEW_STMT);

      extract_scalar_result = true;
    }
  else
    {
      enum tree_code shift_code = 0;
      bool have_whole_vector_shift = true;
      int bit_offset;
      int element_bitsize = tree_low_cst (bitsize, 1);
      int vec_size_in_bits = tree_low_cst (TYPE_SIZE (vectype), 1);
      tree vec_temp;

      if (vec_shr_optab->handlers[mode].insn_code != CODE_FOR_nothing)
        shift_code = VEC_RSHIFT_EXPR;
      else
        have_whole_vector_shift = false;

      /* Regardless of whether we have a whole vector shift, if we're
         emulating the operation via tree-vect-generic, we don't want
         to use it.  Only the first round of the reduction is likely
         to still be profitable via emulation.  */
      /* ??? It might be better to emit a reduction tree code here, so that
         tree-vect-generic can expand the first round via bit tricks.  */
      if (!VECTOR_MODE_P (mode))
        have_whole_vector_shift = false;
      else
        {
          optab optab = optab_for_tree_code (code, vectype);
          if (optab->handlers[mode].insn_code == CODE_FOR_nothing)
            have_whole_vector_shift = false;
        }

      if (have_whole_vector_shift)
        {
          /*** Case 2: Create:
             for (offset = VS/2; offset >= element_size; offset/=2)
                {
                  Create:  va' = vec_shift <va, offset>
                  Create:  va = vop <va, va'>
                }  */

          if (vect_print_dump_info (REPORT_DETAILS))
            fprintf (vect_dump, "Reduce using vector shifts");

          vec_dest = vect_create_destination_var (scalar_dest, vectype);
          new_temp = PHI_RESULT (new_phi);

          for (bit_offset = vec_size_in_bits/2;
               bit_offset >= element_bitsize;
               bit_offset /= 2)
            {
              tree bitpos = size_int (bit_offset);

              epilog_stmt = build2 (MODIFY_EXPR, vectype, vec_dest,
                                    build2 (shift_code, vectype,
                                            new_temp, bitpos));
              new_name = make_ssa_name (vec_dest, epilog_stmt);
              TREE_OPERAND (epilog_stmt, 0) = new_name;
              bsi_insert_after (&exit_bsi, epilog_stmt, BSI_NEW_STMT);

              epilog_stmt = build2 (MODIFY_EXPR, vectype, vec_dest,
                                    build2 (code, vectype,
                                            new_name, new_temp));
              new_temp = make_ssa_name (vec_dest, epilog_stmt);
              TREE_OPERAND (epilog_stmt, 0) = new_temp;
              bsi_insert_after (&exit_bsi, epilog_stmt, BSI_NEW_STMT);
            }

          extract_scalar_result = true;
        }
      else
        {
          tree rhs;

          /*** Case 3: Create:  
             s = extract_field <v_out2, 0>
             for (offset = element_size; 
                  offset < vector_size; 
                  offset += element_size;)
               {
                 Create:  s' = extract_field <v_out2, offset>
                 Create:  s = op <s, s'>
               }  */

          if (vect_print_dump_info (REPORT_DETAILS))
            fprintf (vect_dump, "Reduce using scalar code. ");

          vec_temp = PHI_RESULT (new_phi);
          vec_size_in_bits = tree_low_cst (TYPE_SIZE (vectype), 1);
          rhs = build3 (BIT_FIELD_REF, scalar_type, vec_temp, bitsize,
                         bitsize_zero_node);
          BIT_FIELD_REF_UNSIGNED (rhs) = TYPE_UNSIGNED (scalar_type);
          epilog_stmt = build2 (MODIFY_EXPR, scalar_type, new_scalar_dest, rhs);
          new_temp = make_ssa_name (new_scalar_dest, epilog_stmt);
          TREE_OPERAND (epilog_stmt, 0) = new_temp;
          bsi_insert_after (&exit_bsi, epilog_stmt, BSI_NEW_STMT);
              
          for (bit_offset = element_bitsize;
               bit_offset < vec_size_in_bits;
               bit_offset += element_bitsize)
            { 
              tree bitpos = bitsize_int (bit_offset);
              tree rhs = build3 (BIT_FIELD_REF, scalar_type, vec_temp, bitsize,
                                 bitpos);
                
              BIT_FIELD_REF_UNSIGNED (rhs) = TYPE_UNSIGNED (scalar_type);
              epilog_stmt = build2 (MODIFY_EXPR, scalar_type, new_scalar_dest,
                                    rhs);       
              new_name = make_ssa_name (new_scalar_dest, epilog_stmt);
              TREE_OPERAND (epilog_stmt, 0) = new_name;
              bsi_insert_after (&exit_bsi, epilog_stmt, BSI_NEW_STMT);

              epilog_stmt = build2 (MODIFY_EXPR, scalar_type, new_scalar_dest,
                                build2 (code, scalar_type, new_name, new_temp));
              new_temp = make_ssa_name (new_scalar_dest, epilog_stmt);
              TREE_OPERAND (epilog_stmt, 0) = new_temp;
              bsi_insert_after (&exit_bsi, epilog_stmt, BSI_NEW_STMT);
            }

          extract_scalar_result = false;
        }
    }

  /* 2.4  Extract the final scalar result.  Create:
         s_out3 = extract_field <v_out2, bitpos>  */
  
  if (extract_scalar_result)
    {
      tree rhs;

      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "extract scalar result");

      if (BYTES_BIG_ENDIAN)
        bitpos = size_binop (MULT_EXPR,
                       bitsize_int (TYPE_VECTOR_SUBPARTS (vectype) - 1),
                       TYPE_SIZE (scalar_type));
      else
        bitpos = bitsize_zero_node;

      rhs = build3 (BIT_FIELD_REF, scalar_type, new_temp, bitsize, bitpos);
      BIT_FIELD_REF_UNSIGNED (rhs) = TYPE_UNSIGNED (scalar_type);
      epilog_stmt = build2 (MODIFY_EXPR, scalar_type, new_scalar_dest, rhs);
      new_temp = make_ssa_name (new_scalar_dest, epilog_stmt);
      TREE_OPERAND (epilog_stmt, 0) = new_temp; 
      bsi_insert_after (&exit_bsi, epilog_stmt, BSI_NEW_STMT);
    }

  /* 2.4 Adjust the final result by the initial value of the reduction
         variable. (When such adjustment is not needed, then
         'scalar_initial_def' is zero).

         Create: 
         s_out4 = scalar_expr <s_out3, scalar_initial_def>  */
  
  if (scalar_initial_def)
    {
      epilog_stmt = build2 (MODIFY_EXPR, scalar_type, new_scalar_dest,
                      build2 (code, scalar_type, new_temp, scalar_initial_def));
      new_temp = make_ssa_name (new_scalar_dest, epilog_stmt);
      TREE_OPERAND (epilog_stmt, 0) = new_temp;
      bsi_insert_after (&exit_bsi, epilog_stmt, BSI_NEW_STMT);
    }

  /* 2.6 Replace uses of s_out0 with uses of s_out3  */

  /* Find the loop-closed-use at the loop exit of the original scalar result.  
     (The reduction result is expected to have two immediate uses - one at the 
     latch block, and one at the loop exit).  */
  exit_phi = NULL;
  FOR_EACH_IMM_USE_FAST (use_p, imm_iter, scalar_dest)
    {
      if (!flow_bb_inside_loop_p (loop, bb_for_stmt (USE_STMT (use_p))))
        {
          exit_phi = USE_STMT (use_p);
          break;
        }
    }
  /* We expect to have found an exit_phi because of loop-closed-ssa form.  */
  gcc_assert (exit_phi);
  /* Replace the uses:  */
  orig_name = PHI_RESULT (exit_phi);
  FOR_EACH_IMM_USE_STMT (use_stmt, imm_iter, orig_name)
    FOR_EACH_IMM_USE_ON_STMT (use_p, imm_iter)
      SET_USE (use_p, new_temp);
} 


/* Function vectorizable_reduction.

   Check if STMT performs a reduction operation that can be vectorized.
   If VEC_STMT is also passed, vectorize the STMT: create a vectorized
   stmt to replace it, put it in VEC_STMT, and insert it at BSI.
   Return FALSE if not a vectorizable STMT, TRUE otherwise.

   This function also handles reduction idioms (patterns) that have been 
   recognized in advance during vect_pattern_recog. In this case, STMT may be
   of this form:
     X = pattern_expr (arg0, arg1, ..., X)
   and it's STMT_VINFO_RELATED_STMT points to the last stmt in the original
   sequence that had been detected and replaced by the pattern-stmt (STMT).
  
   In some cases of reduction patterns, the type of the reduction variable X is 
   different than the type of the other arguments of STMT.
   In such cases, the vectype that is used when transforming STMT into a vector
   stmt is different than the vectype that is used to determine the 
   vectorization factor, because it consists of a different number of elements 
   than the actual number of elements that are being operated upon in parallel.

   For example, consider an accumulation of shorts into an int accumulator. 
   On some targets it's possible to vectorize this pattern operating on 8
   shorts at a time (hence, the vectype for purposes of determining the
   vectorization factor should be V8HI); on the other hand, the vectype that
   is used to create the vector form is actually V4SI (the type of the result). 

   Upon entry to this function, STMT_VINFO_VECTYPE records the vectype that 
   indicates what is the actual level of parallelism (V8HI in the example), so 
   that the right vectorization factor would be derived. This vectype 
   corresponds to the type of arguments to the reduction stmt, and should *NOT* 
   be used to create the vectorized stmt. The right vectype for the vectorized
   stmt is obtained from the type of the result X: 
        get_vectype_for_scalar_type (TREE_TYPE (X))

   This means that, contrary to "regular" reductions (or "regular" stmts in 
   general), the following equation:
      STMT_VINFO_VECTYPE == get_vectype_for_scalar_type (TREE_TYPE (X))
   does *NOT* necessarily hold for reduction patterns.  */

bool
vectorizable_reduction (tree stmt, block_stmt_iterator *bsi, tree *vec_stmt)
{
  tree vec_dest;
  tree scalar_dest;
  tree op;
  tree loop_vec_def0 = NULL_TREE, loop_vec_def1 = NULL_TREE;
  stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
  tree vectype = STMT_VINFO_VECTYPE (stmt_info);
  loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
  struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
  tree operation;
  enum tree_code code, orig_code, epilog_reduc_code = 0;
  enum machine_mode vec_mode;
  int op_type;
  optab optab, reduc_optab;
  tree new_temp = NULL_TREE;
  tree def, def_stmt;
  enum vect_def_type dt;
  tree new_phi;
  tree scalar_type;
  bool is_simple_use;
  tree orig_stmt;
  stmt_vec_info orig_stmt_info;
  tree expr = NULL_TREE;
  int i;
  int nunits = TYPE_VECTOR_SUBPARTS (vectype);
  int ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits;
  stmt_vec_info prev_stmt_info;
  tree reduc_def;
  tree new_stmt = NULL_TREE;
  int j;

  gcc_assert (ncopies >= 1);

  /* 1. Is vectorizable reduction?  */

  /* Not supportable if the reduction variable is used in the loop.  */
  if (STMT_VINFO_RELEVANT_P (stmt_info))
    return false;

  if (!STMT_VINFO_LIVE_P (stmt_info))
    return false;

  /* Make sure it was already recognized as a reduction computation.  */
  if (STMT_VINFO_DEF_TYPE (stmt_info) != vect_reduction_def)
    return false;

  /* 2. Has this been recognized as a reduction pattern? 

     Check if STMT represents a pattern that has been recognized
     in earlier analysis stages.  For stmts that represent a pattern,
     the STMT_VINFO_RELATED_STMT field records the last stmt in
     the original sequence that constitutes the pattern.  */

  orig_stmt = STMT_VINFO_RELATED_STMT (stmt_info);
  if (orig_stmt)
    {
      orig_stmt_info = vinfo_for_stmt (orig_stmt);
      gcc_assert (STMT_VINFO_RELATED_STMT (orig_stmt_info) == stmt);
      gcc_assert (STMT_VINFO_IN_PATTERN_P (orig_stmt_info));
      gcc_assert (!STMT_VINFO_IN_PATTERN_P (stmt_info));
    }
 
  /* 3. Check the operands of the operation. The first operands are defined
        inside the loop body. The last operand is the reduction variable,
        which is defined by the loop-header-phi.  */

  gcc_assert (TREE_CODE (stmt) == MODIFY_EXPR);

  operation = TREE_OPERAND (stmt, 1);
  code = TREE_CODE (operation);
  op_type = TREE_CODE_LENGTH (code);
  if (op_type != binary_op && op_type != ternary_op)
    return false;
  scalar_dest = TREE_OPERAND (stmt, 0);
  scalar_type = TREE_TYPE (scalar_dest);

  /* All uses but the last are expected to be defined in the loop.
     The last use is the reduction variable.  */
  for (i = 0; i < op_type-1; i++)
    {
      op = TREE_OPERAND (operation, i);
      is_simple_use = vect_is_simple_use (op, loop_vinfo, &def_stmt, &def, &dt);
      gcc_assert (is_simple_use);
      gcc_assert (dt == vect_loop_def || dt == vect_invariant_def ||
                  dt == vect_constant_def);
    }

  op = TREE_OPERAND (operation, i);
  is_simple_use = vect_is_simple_use (op, loop_vinfo, &def_stmt, &def, &dt);
  gcc_assert (is_simple_use);
  gcc_assert (dt == vect_reduction_def);
  gcc_assert (TREE_CODE (def_stmt) == PHI_NODE);
  if (orig_stmt) 
    gcc_assert (orig_stmt == vect_is_simple_reduction (loop, def_stmt));
  else
    gcc_assert (stmt == vect_is_simple_reduction (loop, def_stmt));
  
  if (STMT_VINFO_LIVE_P (vinfo_for_stmt (def_stmt)))
    return false;

  /* 4. Supportable by target?  */

  /* 4.1. check support for the operation in the loop  */
  optab = optab_for_tree_code (code, vectype);
  if (!optab)
    {
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "no optab.");
      return false;
    }
  vec_mode = TYPE_MODE (vectype);
  if (optab->handlers[(int) vec_mode].insn_code == CODE_FOR_nothing)
    {
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "op not supported by target.");
      if (GET_MODE_SIZE (vec_mode) != UNITS_PER_WORD
          || LOOP_VINFO_VECT_FACTOR (loop_vinfo)
             < vect_min_worthwhile_factor (code))
        return false;
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "proceeding using word mode.");
    }

  /* Worthwhile without SIMD support?  */
  if (!VECTOR_MODE_P (TYPE_MODE (vectype))
      && LOOP_VINFO_VECT_FACTOR (loop_vinfo)
         < vect_min_worthwhile_factor (code))
    {
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "not worthwhile without SIMD support.");
      return false;
    }

  /* 4.2. Check support for the epilog operation.

          If STMT represents a reduction pattern, then the type of the
          reduction variable may be different than the type of the rest
          of the arguments.  For example, consider the case of accumulation
          of shorts into an int accumulator; The original code:
                        S1: int_a = (int) short_a;
          orig_stmt->   S2: int_acc = plus <int_a ,int_acc>;

          was replaced with:
                        STMT: int_acc = widen_sum <short_a, int_acc>

          This means that:
          1. The tree-code that is used to create the vector operation in the 
             epilog code (that reduces the partial results) is not the 
             tree-code of STMT, but is rather the tree-code of the original 
             stmt from the pattern that STMT is replacing. I.e, in the example 
             above we want to use 'widen_sum' in the loop, but 'plus' in the 
             epilog.
          2. The type (mode) we use to check available target support
             for the vector operation to be created in the *epilog*, is 
             determined by the type of the reduction variable (in the example 
             above we'd check this: plus_optab[vect_int_mode]).
             However the type (mode) we use to check available target support
             for the vector operation to be created *inside the loop*, is
             determined by the type of the other arguments to STMT (in the
             example we'd check this: widen_sum_optab[vect_short_mode]).
  
          This is contrary to "regular" reductions, in which the types of all 
          the arguments are the same as the type of the reduction variable. 
          For "regular" reductions we can therefore use the same vector type 
          (and also the same tree-code) when generating the epilog code and
          when generating the code inside the loop.  */

  if (orig_stmt)
    {
      /* This is a reduction pattern: get the vectype from the type of the
         reduction variable, and get the tree-code from orig_stmt.  */
      orig_code = TREE_CODE (TREE_OPERAND (orig_stmt, 1));
      vectype = get_vectype_for_scalar_type (TREE_TYPE (def));
      vec_mode = TYPE_MODE (vectype);
    }
  else
    {
      /* Regular reduction: use the same vectype and tree-code as used for
         the vector code inside the loop can be used for the epilog code. */
      orig_code = code;
    }

  if (!reduction_code_for_scalar_code (orig_code, &epilog_reduc_code))
    return false;
  reduc_optab = optab_for_tree_code (epilog_reduc_code, vectype);
  if (!reduc_optab)
    {
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "no optab for reduction.");
      epilog_reduc_code = NUM_TREE_CODES;
    }
  if (reduc_optab->handlers[(int) vec_mode].insn_code == CODE_FOR_nothing)
    {
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "reduc op not supported by target.");
      epilog_reduc_code = NUM_TREE_CODES;
    }
 
  if (!vec_stmt) /* transformation not required.  */
    {
      STMT_VINFO_TYPE (stmt_info) = reduc_vec_info_type;
      return true;
    }

  /** Transform.  **/

  if (vect_print_dump_info (REPORT_DETAILS))
    fprintf (vect_dump, "transform reduction.");

  /* Create the destination vector  */
  vec_dest = vect_create_destination_var (scalar_dest, vectype);

  /* Create the reduction-phi that defines the reduction-operand.  */
  new_phi = create_phi_node (vec_dest, loop->header);

  /* In case the vectorization factor (VF) is bigger than the number
     of elements that we can fit in a vectype (nunits), we have to generate
     more than one vector stmt - i.e - we need to "unroll" the
     vector stmt by a factor VF/nunits.  For more details see documentation
     in vectorizable_operation.  */

  prev_stmt_info = NULL;
  for (j = 0; j < ncopies; j++)
    {
      /* Handle uses.  */
      if (j == 0)
        {
          op = TREE_OPERAND (operation, 0);
          loop_vec_def0 = vect_get_vec_def_for_operand (op, stmt, NULL);
          if (op_type == ternary_op)
            {
              op = TREE_OPERAND (operation, 1);
              loop_vec_def1 = vect_get_vec_def_for_operand (op, stmt, NULL);
            }
                                                                                
          /* Get the vector def for the reduction variable from the phi node */
          reduc_def = PHI_RESULT (new_phi);
        }
      else
        {
          enum vect_def_type dt = vect_unknown_def_type; /* Dummy */
          loop_vec_def0 = vect_get_vec_def_for_stmt_copy (dt, loop_vec_def0);
          if (op_type == ternary_op)
            loop_vec_def1 = vect_get_vec_def_for_stmt_copy (dt, loop_vec_def1);
                                                                                
          /* Get the vector def for the reduction variable from the vectorized
             reduction operation generated in the previous iteration (j-1)  */
          reduc_def = TREE_OPERAND (new_stmt ,0);
        }
                                                                                
      /* Arguments are ready. create the new vector stmt.  */
                                                                                
      if (op_type == binary_op)
        expr = build2 (code, vectype, loop_vec_def0, reduc_def);
      else
        expr = build3 (code, vectype, loop_vec_def0, loop_vec_def1, 
                                                                reduc_def);
      new_stmt = build2 (MODIFY_EXPR, void_type_node, vec_dest, expr);
      new_temp = make_ssa_name (vec_dest, new_stmt);
      TREE_OPERAND (new_stmt, 0) = new_temp;
      vect_finish_stmt_generation (stmt, new_stmt, bsi);
                                                                                
      if (j == 0)
        STMT_VINFO_VEC_STMT (stmt_info) = *vec_stmt = new_stmt;
      else
        STMT_VINFO_RELATED_STMT (prev_stmt_info) = new_stmt;
      prev_stmt_info = vinfo_for_stmt (new_stmt);
    }
                                                                                
  /* Finalize the reduction-phi (set it's arguments) and create the
     epilog reduction code.  */
  vect_create_epilog_for_reduction (new_temp, stmt, epilog_reduc_code, new_phi);                                                                                
  return true;
}


/* Function vectorizable_assignment.

   Check if STMT performs an assignment (copy) that can be vectorized. 
   If VEC_STMT is also passed, vectorize the STMT: create a vectorized 
   stmt to replace it, put it in VEC_STMT, and insert it at BSI.
   Return FALSE if not a vectorizable STMT, TRUE otherwise.  */

bool
vectorizable_assignment (tree stmt, block_stmt_iterator *bsi, tree *vec_stmt)
{
  tree vec_dest;
  tree scalar_dest;
  tree op;
  tree vec_oprnd;
  stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
  tree vectype = STMT_VINFO_VECTYPE (stmt_info);
  loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
  tree new_temp;
  tree def, def_stmt;
  enum vect_def_type dt;
  int nunits = TYPE_VECTOR_SUBPARTS (vectype);
  int ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits;

  gcc_assert (ncopies >= 1);
  if (ncopies > 1)
    return false; /* FORNOW */

  /* Is vectorizable assignment?  */
  if (!STMT_VINFO_RELEVANT_P (stmt_info))
    return false;

  gcc_assert (STMT_VINFO_DEF_TYPE (stmt_info) == vect_loop_def);

  if (TREE_CODE (stmt) != MODIFY_EXPR)
    return false;

  scalar_dest = TREE_OPERAND (stmt, 0);
  if (TREE_CODE (scalar_dest) != SSA_NAME)
    return false;

  op = TREE_OPERAND (stmt, 1);
  if (!vect_is_simple_use (op, loop_vinfo, &def_stmt, &def, &dt))
    {
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "use not simple.");
      return false;
    }

  if (!vec_stmt) /* transformation not required.  */
    {
      STMT_VINFO_TYPE (stmt_info) = assignment_vec_info_type;
      return true;
    }

  /** Transform.  **/
  if (vect_print_dump_info (REPORT_DETAILS))
    fprintf (vect_dump, "transform assignment.");

  /* Handle def.  */
  vec_dest = vect_create_destination_var (scalar_dest, vectype);

  /* Handle use.  */
  op = TREE_OPERAND (stmt, 1);
  vec_oprnd = vect_get_vec_def_for_operand (op, stmt, NULL);

  /* Arguments are ready. create the new vector stmt.  */
  *vec_stmt = build2 (MODIFY_EXPR, vectype, vec_dest, vec_oprnd);
  new_temp = make_ssa_name (vec_dest, *vec_stmt);
  TREE_OPERAND (*vec_stmt, 0) = new_temp;
  vect_finish_stmt_generation (stmt, *vec_stmt, bsi);
  
  return true;
}


/* Function vect_min_worthwhile_factor.

   For a loop where we could vectorize the operation indicated by CODE,
   return the minimum vectorization factor that makes it worthwhile
   to use generic vectors.  */
static int
vect_min_worthwhile_factor (enum tree_code code)
{
  switch (code)
    {
    case PLUS_EXPR:
    case MINUS_EXPR:
    case NEGATE_EXPR:
      return 4;

    case BIT_AND_EXPR:
    case BIT_IOR_EXPR:
    case BIT_XOR_EXPR:
    case BIT_NOT_EXPR:
      return 2;

    default:
      return INT_MAX;
    }
}


/* Function vectorizable_operation.

   Check if STMT performs a binary or unary operation that can be vectorized. 
   If VEC_STMT is also passed, vectorize the STMT: create a vectorized 
   stmt to replace it, put it in VEC_STMT, and insert it at BSI.
   Return FALSE if not a vectorizable STMT, TRUE otherwise.  */

bool
vectorizable_operation (tree stmt, block_stmt_iterator *bsi, tree *vec_stmt)
{
  tree vec_dest;
  tree scalar_dest;
  tree operation;
  tree op0, op1 = NULL;
  tree vec_oprnd0 = NULL_TREE, vec_oprnd1 = NULL_TREE;
  stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
  tree vectype = STMT_VINFO_VECTYPE (stmt_info);
  loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
  enum tree_code code;
  enum machine_mode vec_mode;
  tree new_temp;
  int op_type;
  optab optab;
  int icode;
  enum machine_mode optab_op2_mode;
  tree def, def_stmt;
  enum vect_def_type dt0, dt1;
  tree new_stmt;
  stmt_vec_info prev_stmt_info;
  int nunits_in = TYPE_VECTOR_SUBPARTS (vectype);
  int nunits_out;
  tree vectype_out;
  int ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits_in;
  int j;

  gcc_assert (ncopies >= 1);

  /* Is STMT a vectorizable binary/unary operation?   */
  if (!STMT_VINFO_RELEVANT_P (stmt_info))
    return false;

  gcc_assert (STMT_VINFO_DEF_TYPE (stmt_info) == vect_loop_def);

  if (STMT_VINFO_LIVE_P (stmt_info))
    {
      /* FORNOW: not yet supported.  */
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "value used after loop.");
      return false;
    }

  if (TREE_CODE (stmt) != MODIFY_EXPR)
    return false;

  if (TREE_CODE (TREE_OPERAND (stmt, 0)) != SSA_NAME)
    return false;

  scalar_dest = TREE_OPERAND (stmt, 0);
  vectype_out = get_vectype_for_scalar_type (TREE_TYPE (scalar_dest));
  nunits_out = TYPE_VECTOR_SUBPARTS (vectype_out);
  if (nunits_out != nunits_in)
    return false;

  operation = TREE_OPERAND (stmt, 1);
  code = TREE_CODE (operation);
  optab = optab_for_tree_code (code, vectype);

  /* Support only unary or binary operations.  */
  op_type = TREE_CODE_LENGTH (code);
  if (op_type != unary_op && op_type != binary_op)
    {
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "num. args = %d (not unary/binary op).", op_type);
      return false;
    }

  op0 = TREE_OPERAND (operation, 0);
  if (!vect_is_simple_use (op0, loop_vinfo, &def_stmt, &def, &dt0))
    {
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "use not simple.");
      return false;
    }
                                                                                
  if (op_type == binary_op)
    {
      op1 = TREE_OPERAND (operation, 1);
      if (!vect_is_simple_use (op1, loop_vinfo, &def_stmt, &def, &dt1))
        {
          if (vect_print_dump_info (REPORT_DETAILS))
            fprintf (vect_dump, "use not simple.");
          return false;
        }
    }

  /* Supportable by target?  */
  if (!optab)
    {
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "no optab.");
      return false;
    }
  vec_mode = TYPE_MODE (vectype);
  icode = (int) optab->handlers[(int) vec_mode].insn_code;
  if (icode == CODE_FOR_nothing)
    {
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "op not supported by target.");
      if (GET_MODE_SIZE (vec_mode) != UNITS_PER_WORD
          || LOOP_VINFO_VECT_FACTOR (loop_vinfo)
             < vect_min_worthwhile_factor (code))
        return false;
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "proceeding using word mode.");
    }

  /* Worthwhile without SIMD support?  */
  if (!VECTOR_MODE_P (TYPE_MODE (vectype))
      && LOOP_VINFO_VECT_FACTOR (loop_vinfo)
         < vect_min_worthwhile_factor (code))
    {
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "not worthwhile without SIMD support.");
      return false;
    }

  if (code == LSHIFT_EXPR || code == RSHIFT_EXPR)
    {
      /* FORNOW: not yet supported.  */
      if (!VECTOR_MODE_P (vec_mode))
        return false;

      /* Invariant argument is needed for a vector shift
         by a scalar shift operand.  */
      optab_op2_mode = insn_data[icode].operand[2].mode;
      if (! (VECTOR_MODE_P (optab_op2_mode)
             || dt1 == vect_constant_def
             || dt1 == vect_invariant_def))
        {
          if (vect_print_dump_info (REPORT_DETAILS))
            fprintf (vect_dump, "operand mode requires invariant argument.");
          return false;
        }
    }

  if (!vec_stmt) /* transformation not required.  */
    {
      STMT_VINFO_TYPE (stmt_info) = op_vec_info_type;
      return true;
    }

  /** Transform.  **/

  if (vect_print_dump_info (REPORT_DETAILS))
    fprintf (vect_dump, "transform binary/unary operation.");

  /* Handle def.  */
  vec_dest = vect_create_destination_var (scalar_dest, vectype);

  /* In case the vectorization factor (VF) is bigger than the number
     of elements that we can fit in a vectype (nunits), we have to generate
     more than one vector stmt - i.e - we need to "unroll" the
     vector stmt by a factor VF/nunits. In doing so, we record a pointer
     from one copy of the vector stmt to the next, in the field
     STMT_VINFO_RELATED_STMT. This is necessary in order to allow following
     stages to find the correct vector defs to be used when vectorizing
     stmts that use the defs of the current stmt. The example below illustrates
     the vectorization process when VF=16 and nunits=4 (i.e - we need to create
     4 vectorized stmts):
                                                                                
     before vectorization:
                                RELATED_STMT    VEC_STMT
        S1:     x = memref      -               -
        S2:     z = x + 1       -               -
                                                                                
     step 1: vectorize stmt S1 (done in vectorizable_load. See more details
             there):
                                RELATED_STMT    VEC_STMT
        VS1_0:  vx0 = memref0   VS1_1           -
        VS1_1:  vx1 = memref1   VS1_2           -
        VS1_2:  vx2 = memref2   VS1_3           -
        VS1_3:  vx3 = memref3   -               -
        S1:     x = load        -               VS1_0
        S2:     z = x + 1       -               -
                                                                                
     step2: vectorize stmt S2 (done here):
        To vectorize stmt S2 we first need to find the relevant vector
        def for the first operand 'x'. This is, as usual, obtained from
        the vector stmt recorded in the STMT_VINFO_VEC_STMT of the stmt
        that defines 'x' (S1). This way we find the stmt VS1_0, and the
        relevant vector def 'vx0'. Having found 'vx0' we can generate
        the vector stmt VS2_0, and as usual, record it in the
        STMT_VINFO_VEC_STMT of stmt S2.
        When creating the second copy (VS2_1), we obtain the relevant vector
        def from the vector stmt recorded in the STMT_VINFO_RELATED_STMT of
        stmt VS1_0. This way we find the stmt VS1_1 and the relevant
        vector def 'vx1'. Using 'vx1' we create stmt VS2_1 and record a
        pointer to it in the STMT_VINFO_RELATED_STMT of the vector stmt VS2_0.
        Similarly when creating stmts VS2_2 and VS2_3. This is the resulting
        chain of stmts and pointers:
                                RELATED_STMT    VEC_STMT
        VS1_0:  vx0 = memref0   VS1_1           -
        VS1_1:  vx1 = memref1   VS1_2           -
        VS1_2:  vx2 = memref2   VS1_3           -
        VS1_3:  vx3 = memref3   -               -
        S1:     x = load        -               VS1_0
        VS2_0:  vz0 = vx0 + v1  VS2_1           -
        VS2_1:  vz1 = vx1 + v1  VS2_2           -
        VS2_2:  vz2 = vx2 + v1  VS2_3           -
        VS2_3:  vz3 = vx3 + v1  -               -
        S2:     z = x + 1       -               VS2_0  */
                                                                                
  prev_stmt_info = NULL;
  for (j = 0; j < ncopies; j++)
    {
      /* Handle uses.  */
      if (j == 0)
        {
          vec_oprnd0 = vect_get_vec_def_for_operand (op0, stmt, NULL);
          if (op_type == binary_op)
            {
              if (code == LSHIFT_EXPR || code == RSHIFT_EXPR)
                {
                  /* Vector shl and shr insn patterns can be defined with
                     scalar operand 2 (shift operand).  In this case, use
                     constant or loop invariant op1 directly, without
                     extending it to vector mode first.  */
                  optab_op2_mode = insn_data[icode].operand[2].mode;
                  if (!VECTOR_MODE_P (optab_op2_mode))
                    {
                      if (vect_print_dump_info (REPORT_DETAILS))
                        fprintf (vect_dump, "operand 1 using scalar mode.");
                      vec_oprnd1 = op1;
                    }
                }
              if (!vec_oprnd1)
                vec_oprnd1 = vect_get_vec_def_for_operand (op1, stmt, NULL);
            }
        }
      else
        {
          vec_oprnd0 = vect_get_vec_def_for_stmt_copy (dt0, vec_oprnd0);
          if (op_type == binary_op)
            vec_oprnd1 = vect_get_vec_def_for_stmt_copy (dt1, vec_oprnd1);
        }

      /* Arguments are ready. create the new vector stmt.  */
                                                                                
      if (op_type == binary_op)
        new_stmt = build2 (MODIFY_EXPR, void_type_node, vec_dest,
                    build2 (code, vectype, vec_oprnd0, vec_oprnd1));
      else
        new_stmt = build2 (MODIFY_EXPR, void_type_node, vec_dest,
                    build1 (code, vectype, vec_oprnd0));
      new_temp = make_ssa_name (vec_dest, new_stmt);
      TREE_OPERAND (new_stmt, 0) = new_temp;
      vect_finish_stmt_generation (stmt, new_stmt, bsi);
                                                                                
      if (j == 0)
        STMT_VINFO_VEC_STMT (stmt_info) = *vec_stmt = new_stmt;
      else
        STMT_VINFO_RELATED_STMT (prev_stmt_info) = new_stmt;
      prev_stmt_info = vinfo_for_stmt (new_stmt);
    }

  return true;
}


/* Function vectorizable_type_demotion
                                                                                
   Check if STMT performs a binary or unary operation that involves
   type demotion, and if it can be vectorized.
   If VEC_STMT is also passed, vectorize the STMT: create a vectorized
   stmt to replace it, put it in VEC_STMT, and insert it at BSI.
   Return FALSE if not a vectorizable STMT, TRUE otherwise.  */
                                                                                
bool
vectorizable_type_demotion (tree stmt, block_stmt_iterator *bsi,
                             tree *vec_stmt)
{
  tree vec_dest;
  tree scalar_dest;
  tree operation;
  tree op0;
  tree vec_oprnd0=NULL, vec_oprnd1=NULL;
  stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
  loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
  enum tree_code code;
  tree new_temp;
  tree def, def_stmt;
  enum vect_def_type dt0;
  tree new_stmt;
  stmt_vec_info prev_stmt_info;
  int nunits_in;
  int nunits_out;
  tree vectype_out;
  int ncopies;
  int j;
  tree expr;
  tree vectype_in;
  tree scalar_type;
  optab optab;
  enum machine_mode vec_mode;
                                                                                
  /* Is STMT a vectorizable type-demotion operation?  */
                                                                                
  if (!STMT_VINFO_RELEVANT_P (stmt_info))
    return false;
                                                                                
  gcc_assert (STMT_VINFO_DEF_TYPE (stmt_info) == vect_loop_def);
                                                                                
  if (STMT_VINFO_LIVE_P (stmt_info))
    {
      /* FORNOW: not yet supported.  */
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "value used after loop.");
      return false;
    }
                                                                                
  if (TREE_CODE (stmt) != MODIFY_EXPR)
    return false;
                                                                                
  if (TREE_CODE (TREE_OPERAND (stmt, 0)) != SSA_NAME)
    return false;
                                                                                
  operation = TREE_OPERAND (stmt, 1);
  code = TREE_CODE (operation);
  if (code != NOP_EXPR && code != CONVERT_EXPR)
    return false;
                                                                                
  op0 = TREE_OPERAND (operation, 0);
  vectype_in = get_vectype_for_scalar_type (TREE_TYPE (op0));
  nunits_in = TYPE_VECTOR_SUBPARTS (vectype_in);
                                                                                
  scalar_dest = TREE_OPERAND (stmt, 0);
  scalar_type = TREE_TYPE (scalar_dest);
  vectype_out = get_vectype_for_scalar_type (scalar_type);
  nunits_out = TYPE_VECTOR_SUBPARTS (vectype_out);
  if (nunits_in != nunits_out / 2) /* FORNOW */
    return false;
                                                                                
  ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits_out;
  gcc_assert (ncopies >= 1);
                                                                                
  /* Check the operands of the operation.  */
  if (!vect_is_simple_use (op0, loop_vinfo, &def_stmt, &def, &dt0))
    {
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "use not simple.");
      return false;
    }
                                                                                
  /* Supportable by target?  */
  code = VEC_PACK_MOD_EXPR;
  optab = optab_for_tree_code (VEC_PACK_MOD_EXPR, vectype_in);
  if (!optab)
    return false;
                                                                                
  vec_mode = TYPE_MODE (vectype_in);
  if (optab->handlers[(int) vec_mode].insn_code == CODE_FOR_nothing)
    return false;
                                                                                
  STMT_VINFO_VECTYPE (stmt_info) = vectype_in;
                                                                                
  if (!vec_stmt) /* transformation not required.  */
    {
      STMT_VINFO_TYPE (stmt_info) = type_demotion_vec_info_type;
      return true;
    }
                                                                                
  /** Transform.  **/
                                                                                
  if (vect_print_dump_info (REPORT_DETAILS))
    fprintf (vect_dump, "transform type demotion operation. ncopies = %d.",
                        ncopies);
                                                                                
  /* Handle def.  */
  vec_dest = vect_create_destination_var (scalar_dest, vectype_out);
  
  /* In case the vectorization factor (VF) is bigger than the number
     of elements that we can fit in a vectype (nunits), we have to generate
     more than one vector stmt - i.e - we need to "unroll" the
     vector stmt by a factor VF/nunits.   */
  prev_stmt_info = NULL;
  for (j = 0; j < ncopies; j++)
    {
      /* Handle uses.  */
      if (j == 0)
        {
          enum vect_def_type dt = vect_unknown_def_type; /* Dummy */
          vec_oprnd0 = vect_get_vec_def_for_operand (op0, stmt, NULL);
          vec_oprnd1 = vect_get_vec_def_for_stmt_copy (dt, vec_oprnd0);
        }
      else
        {
          vec_oprnd0 = vect_get_vec_def_for_stmt_copy (dt0, vec_oprnd1);
          vec_oprnd1 = vect_get_vec_def_for_stmt_copy (dt0, vec_oprnd0);
        }
                                                                                
      /* Arguments are ready. Create the new vector stmt.  */
      expr = build2 (code, vectype_out, vec_oprnd0, vec_oprnd1);
      new_stmt = build2 (MODIFY_EXPR, void_type_node, vec_dest, expr);
      new_temp = make_ssa_name (vec_dest, new_stmt);
      TREE_OPERAND (new_stmt, 0) = new_temp;
      vect_finish_stmt_generation (stmt, new_stmt, bsi);
                                                                                
      if (j == 0)
        STMT_VINFO_VEC_STMT (stmt_info) = new_stmt;
      else
        STMT_VINFO_RELATED_STMT (prev_stmt_info) = new_stmt;
                                                                                
      prev_stmt_info = vinfo_for_stmt (new_stmt);
    }
                                                                                
  *vec_stmt = STMT_VINFO_VEC_STMT (stmt_info);
  return true;
}


/* Function vect_gen_widened_results_half

   Create a vector stmt whose code, type, number of arguments, and result
   variable are CODE, VECTYPE, OP_TYPE, and VEC_DEST, and its arguments are 
   VEC_OPRND0 and VEC_OPRND1. The new vector stmt is to be inserted at BSI.
   In the case that CODE is a CALL_EXPR, this means that a call to DECL
   needs to be created (DECL is a function-decl of a target-builtin).
   STMT is the original scalar stmt that we are vectorizing.  */

static tree
vect_gen_widened_results_half (enum tree_code code, tree vectype, tree decl,
                               tree vec_oprnd0, tree vec_oprnd1, int op_type,
                               tree vec_dest, block_stmt_iterator *bsi,
                               tree stmt)
{ 
  tree vec_params;
  tree expr; 
  tree new_stmt; 
  tree new_temp; 
  tree sym; 
  ssa_op_iter iter;
 
  /* Generate half of the widened result:  */ 
  if (code == CALL_EXPR) 
    {  
      /* Target specific support  */ 
      vec_params = build_tree_list (NULL_TREE, vec_oprnd0); 
      if (op_type == binary_op) 
        vec_params = tree_cons (NULL_TREE, vec_oprnd1, vec_params); 
      expr = build_function_call_expr (decl, vec_params); 
    } 
  else 
    { 
      /* Generic support */ 
      gcc_assert (op_type == TREE_CODE_LENGTH (code)); 
      if (op_type == binary_op) 
        expr = build2 (code, vectype, vec_oprnd0, vec_oprnd1); 
      else  
        expr = build1 (code, vectype, vec_oprnd0); 
    } 
  new_stmt = build2 (MODIFY_EXPR, void_type_node, vec_dest, expr); 
  new_temp = make_ssa_name (vec_dest, new_stmt); 
  TREE_OPERAND (new_stmt, 0) = new_temp; 
  vect_finish_stmt_generation (stmt, new_stmt, bsi); 

  if (code == CALL_EXPR)
    {
      FOR_EACH_SSA_TREE_OPERAND (sym, new_stmt, iter, SSA_OP_ALL_VIRTUALS)
        {
          if (TREE_CODE (sym) == SSA_NAME)
            sym = SSA_NAME_VAR (sym);
          mark_sym_for_renaming (sym);
        }
    }

  return new_stmt;
}


/* Function vectorizable_type_promotion

   Check if STMT performs a binary or unary operation that involves
   type promotion, and if it can be vectorized.
   If VEC_STMT is also passed, vectorize the STMT: create a vectorized
   stmt to replace it, put it in VEC_STMT, and insert it at BSI.
   Return FALSE if not a vectorizable STMT, TRUE otherwise.  */

bool
vectorizable_type_promotion (tree stmt, block_stmt_iterator *bsi,
                             tree *vec_stmt)
{
  tree vec_dest;
  tree scalar_dest;
  tree operation;
  tree op0, op1 = NULL;
  tree vec_oprnd0=NULL, vec_oprnd1=NULL;
  stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
  loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
  enum tree_code code, code1 = CODE_FOR_nothing, code2 = CODE_FOR_nothing;
  tree decl1 = NULL_TREE, decl2 = NULL_TREE;
  int op_type; 
  tree def, def_stmt;
  enum vect_def_type dt0, dt1;
  tree new_stmt;
  stmt_vec_info prev_stmt_info;
  int nunits_in;
  int nunits_out;
  tree vectype_out;
  int ncopies;
  int j;
  tree vectype_in;
  
  /* Is STMT a vectorizable type-promotion operation?  */

  if (!STMT_VINFO_RELEVANT_P (stmt_info))
    return false;

  gcc_assert (STMT_VINFO_DEF_TYPE (stmt_info) == vect_loop_def);

  if (STMT_VINFO_LIVE_P (stmt_info))
    {
      /* FORNOW: not yet supported.  */
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "value used after loop.");
      return false;
    }

  if (TREE_CODE (stmt) != MODIFY_EXPR)
    return false;

  if (TREE_CODE (TREE_OPERAND (stmt, 0)) != SSA_NAME)
    return false;

  operation = TREE_OPERAND (stmt, 1);
  code = TREE_CODE (operation);
  if (code != NOP_EXPR && code != WIDEN_MULT_EXPR)
    return false;

  op0 = TREE_OPERAND (operation, 0);
  vectype_in = get_vectype_for_scalar_type (TREE_TYPE (op0));
  nunits_in = TYPE_VECTOR_SUBPARTS (vectype_in);
  ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits_in;
  gcc_assert (ncopies >= 1);

  scalar_dest = TREE_OPERAND (stmt, 0);
  vectype_out = get_vectype_for_scalar_type (TREE_TYPE (scalar_dest));
  nunits_out = TYPE_VECTOR_SUBPARTS (vectype_out);
  if (nunits_out != nunits_in / 2) /* FORNOW */
    return false;

  /* Check the operands of the operation.  */
  if (!vect_is_simple_use (op0, loop_vinfo, &def_stmt, &def, &dt0))
    {
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "use not simple.");
      return false;
    }

  op_type = TREE_CODE_LENGTH (code);
  if (op_type == binary_op)
    {
      op1 = TREE_OPERAND (operation, 1);
      if (!vect_is_simple_use (op1, loop_vinfo, &def_stmt, &def, &dt1))
        {
          if (vect_print_dump_info (REPORT_DETAILS))
            fprintf (vect_dump, "use not simple.");
          return false;
        }
    }

  /* Supportable by target?  */
  if (!supportable_widening_operation (code, stmt, vectype_in,
                                       &decl1, &decl2, &code1, &code2))
    return false;

  STMT_VINFO_VECTYPE (stmt_info) = vectype_in;

  if (!vec_stmt) /* transformation not required.  */
    {
      STMT_VINFO_TYPE (stmt_info) = type_promotion_vec_info_type;
      return true;
    }

  /** Transform.  **/

  if (vect_print_dump_info (REPORT_DETAILS))
    fprintf (vect_dump, "transform type promotion operation. ncopies = %d.",
                        ncopies);

  /* Handle def.  */
  vec_dest = vect_create_destination_var (scalar_dest, vectype_out);

  /* In case the vectorization factor (VF) is bigger than the number
     of elements that we can fit in a vectype (nunits), we have to generate
     more than one vector stmt - i.e - we need to "unroll" the
     vector stmt by a factor VF/nunits.   */

  prev_stmt_info = NULL;
  for (j = 0; j < ncopies; j++)
    {
      /* Handle uses.  */
      if (j == 0)
        {
          vec_oprnd0 = vect_get_vec_def_for_operand (op0, stmt, NULL);
          if (op_type == binary_op)
            vec_oprnd1 = vect_get_vec_def_for_operand (op1, stmt, NULL);
        }
      else
        {
          vec_oprnd0 = vect_get_vec_def_for_stmt_copy (dt0, vec_oprnd0);
          if (op_type == binary_op)
            vec_oprnd1 = vect_get_vec_def_for_stmt_copy (dt1, vec_oprnd1);
        }

      /* Arguments are ready. Create the new vector stmt.  We are creating 
         two vector defs because the widened result does not fit in one vector.
         The vectorized stmt can be expressed as a call to a taregt builtin,
         or a using a tree-code.  */
      /* Generate first half of the widened result:  */
      new_stmt = vect_gen_widened_results_half (code1, vectype_out, decl1, 
                        vec_oprnd0, vec_oprnd1, op_type, vec_dest, bsi, stmt);
      if (j == 0)
        STMT_VINFO_VEC_STMT (stmt_info) = new_stmt;
      else
        STMT_VINFO_RELATED_STMT (prev_stmt_info) = new_stmt;
      prev_stmt_info = vinfo_for_stmt (new_stmt);

      /* Generate second half of the widened result:  */
      new_stmt = vect_gen_widened_results_half (code2, vectype_out, decl2,
                        vec_oprnd0, vec_oprnd1, op_type, vec_dest, bsi, stmt);
      STMT_VINFO_RELATED_STMT (prev_stmt_info) = new_stmt;
      prev_stmt_info = vinfo_for_stmt (new_stmt);

    }

  *vec_stmt = STMT_VINFO_VEC_STMT (stmt_info);
  return true;
}


/* Function vect_strided_store_supported.

   Returns TRUE is INTERLEAVE_HIGH and INTERLEAVE_LOW operations are supported,
   and FALSE otherwise.  */

static bool
vect_strided_store_supported (tree vectype)
{
  optab interleave_high_optab, interleave_low_optab;
  int mode;

  mode = (int) TYPE_MODE (vectype);
      
  /* Check that the operation is supported.  */
  interleave_high_optab = optab_for_tree_code (VEC_INTERLEAVE_HIGH_EXPR, 
                                               vectype);
  interleave_low_optab = optab_for_tree_code (VEC_INTERLEAVE_LOW_EXPR, 
                                              vectype);
  if (!interleave_high_optab || !interleave_low_optab)
    {
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "no optab for interleave.");
      return false;
    }

  if (interleave_high_optab->handlers[(int) mode].insn_code 
      == CODE_FOR_nothing
      || interleave_low_optab->handlers[(int) mode].insn_code 
      == CODE_FOR_nothing)
    {
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "interleave op not supported by target.");
      return false;
    }
  return true;
}


/* Function vect_permute_store_chain.

   Given a chain of interleaved strores in DR_CHAIN of LENGTH that must be
   a power of 2, generate interleave_high/low stmts to reorder the data 
   correctly for the stores. Return the final references for stores in
   RESULT_CHAIN.

   E.g., LENGTH is 4 and the scalar type is short, i.e., VF is 8.
   The input is 4 vectors each containg 8 elements. We assign a number to each 
   element, the input sequence is:

   1st vec:   0  1  2  3  4  5  6  7
   2nd vec:   8  9 10 11 12 13 14 15
   3rd vec:  16 17 18 19 20 21 22 23 
   4th vec:  24 25 26 27 28 29 30 31

   The output sequence should be:

   1st vec:  0  8 16 24  1  9 17 25
   2nd vec:  2 10 18 26  3 11 19 27
   3rd vec:  4 12 20 28  5 13 21 30
   4th vec:  6 14 22 30  7 15 23 31

   i.e., we interleave the contents of the four vectors in their order.

   We use interleave_high/low instructions to create such output. The input of 
   each interleave_high/low operation is two vectors:
   1st vec    2nd vec 
   0 1 2 3    4 5 6 7 
   the even elements of the result vector are obtained left-to-right from the 
   high/low elements of the first vector. The odd elements of the result are 
   obtained left-to-right from the high/low elements of the second vector.
   The output of interleave_high will be:   0 4 1 5
   and of interleave_low:                   2 6 3 7

   
   The permutaion is done in log LENGTH stages. In each stage interleave_high 
   and interleave_low stmts are created for each pair of vectors in DR_CHAIN, 
   where the first argument is taken from the first half of DR_CHAIN and the 
   second argument from it's second half. 
   In our example, 

   I1: interleave_high (1st vec, 3rd vec)
   I2: interleave_low (1st vec, 3rd vec)
   I3: interleave_high (2nd vec, 4th vec)
   I4: interleave_low (2nd vec, 4th vec)

   The output for the first stage is:

   I1:  0 16  1 17  2 18  3 19
   I2:  4 20  5 21  6 22  7 23
   I3:  8 24  9 25 10 26 11 27
   I4: 12 28 13 29 14 30 15 31

   The output of the second stage, i.e. the final result is:

   I1:  0  8 16 24  1  9 17 25
   I2:  2 10 18 26  3 11 19 27
   I3:  4 12 20 28  5 13 21 30
   I4:  6 14 22 30  7 15 23 31.  */
 
static bool
vect_permute_store_chain (VEC(tree,heap) *dr_chain, 
                          unsigned int length, 
                          tree stmt, 
                          block_stmt_iterator *bsi,
                          VEC(tree,heap) **result_chain)
{
  tree perm_dest, perm_stmt, vect1, vect2, high, low;
  tree vectype = STMT_VINFO_VECTYPE (vinfo_for_stmt (stmt));
  tree scalar_dest;
  int i;
  unsigned int j;
  VEC(tree,heap) *first, *second;
  
  scalar_dest = TREE_OPERAND (stmt, 0);
  first = VEC_alloc (tree, heap, length/2);
  second = VEC_alloc (tree, heap, length/2);

  /* Check that the operation is supported.  */
  if (!vect_strided_store_supported (vectype))
    return false;

  *result_chain = VEC_copy (tree, heap, dr_chain);

  for (i = 0; i < exact_log2 (length); i++)
    {
      for (j = 0; j < length/2; j++)
        {
          vect1 = VEC_index (tree, dr_chain, j);
          vect2 = VEC_index (tree, dr_chain, j+length/2);

          /* high = interleave_high (vect1, vect2);  */
          perm_dest = create_tmp_var (vectype, "vect_inter_high");
          add_referenced_var (perm_dest);
          perm_stmt = build2 (MODIFY_EXPR, vectype, perm_dest,
                              build2 (VEC_INTERLEAVE_HIGH_EXPR, vectype, vect1, 
                                      vect2));
          high = make_ssa_name (perm_dest, perm_stmt);
          TREE_OPERAND (perm_stmt, 0) = high;
          vect_finish_stmt_generation (stmt, perm_stmt, bsi);
          VEC_replace (tree, *result_chain, 2*j, high);

          /* low = interleave_low (vect1, vect2);  */
          perm_dest = create_tmp_var (vectype, "vect_inter_low");
          add_referenced_var (perm_dest);
          perm_stmt = build2 (MODIFY_EXPR, vectype, perm_dest,
                              build2 (VEC_INTERLEAVE_LOW_EXPR, vectype, vect1, 
                                      vect2));
          low = make_ssa_name (perm_dest, perm_stmt);
          TREE_OPERAND (perm_stmt, 0) = low;
          vect_finish_stmt_generation (stmt, perm_stmt, bsi);
          VEC_replace (tree, *result_chain, 2*j+1, low);
        }
      dr_chain = VEC_copy (tree, heap, *result_chain);
    }
  return true;
}


/* Function vectorizable_store.

   Check if STMT defines a non scalar data-ref (array/pointer/structure) that 
   can be vectorized. 
   If VEC_STMT is also passed, vectorize the STMT: create a vectorized 
   stmt to replace it, put it in VEC_STMT, and insert it at BSI.
   Return FALSE if not a vectorizable STMT, TRUE otherwise.  */

bool
vectorizable_store (tree stmt, block_stmt_iterator *bsi, tree *vec_stmt)
{
  tree scalar_dest;
  tree data_ref;
  tree op;
  tree vec_oprnd = NULL_TREE;
  stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
  struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info), *first_dr = NULL;
  tree vectype = STMT_VINFO_VECTYPE (stmt_info);
  loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
  enum machine_mode vec_mode;
  tree dummy;
  enum dr_alignment_support alignment_support_cheme;
  ssa_op_iter iter;
  def_operand_p def_p;
  tree def, def_stmt;
  enum vect_def_type dt;
  stmt_vec_info prev_stmt_info = NULL;
  tree dataref_ptr = NULL_TREE;
  int nunits = TYPE_VECTOR_SUBPARTS (vectype);
  int ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits;
  int j;
  tree next_stmt, first_stmt;
  bool strided_store = false;
  unsigned int group_size, i;
  VEC(tree,heap) *dr_chain = NULL, *oprnds = NULL, *result_chain = NULL;
  gcc_assert (ncopies >= 1);

  /* Is vectorizable store? */

  if (TREE_CODE (stmt) != MODIFY_EXPR)
    return false;

  scalar_dest = TREE_OPERAND (stmt, 0);
  if (TREE_CODE (scalar_dest) != ARRAY_REF
      && TREE_CODE (scalar_dest) != INDIRECT_REF
      && !DR_GROUP_FIRST_DR (stmt_info))
    return false;

  op = TREE_OPERAND (stmt, 1);
  if (!vect_is_simple_use (op, loop_vinfo, &def_stmt, &def, &dt))
    {
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "use not simple.");
      return false;
    }

  vec_mode = TYPE_MODE (vectype);
  /* FORNOW. In some cases can vectorize even if data-type not supported
     (e.g. - array initialization with 0).  */
  if (mov_optab->handlers[(int)vec_mode].insn_code == CODE_FOR_nothing)
    return false;

  if (!STMT_VINFO_DATA_REF (stmt_info))
    return false;

  if (DR_GROUP_FIRST_DR (stmt_info))
    {
      strided_store = true;
      if (!vect_strided_store_supported (vectype))
        return false;      
    }

  if (!vec_stmt) /* transformation not required.  */
    {
      STMT_VINFO_TYPE (stmt_info) = store_vec_info_type;
      return true;
    }

  /** Transform.  **/

  if (vect_print_dump_info (REPORT_DETAILS))
    fprintf (vect_dump, "transform store. ncopies = %d",ncopies);

  if (strided_store)
    {
      first_stmt = DR_GROUP_FIRST_DR (stmt_info);
      first_dr = STMT_VINFO_DATA_REF (vinfo_for_stmt (first_stmt));
      group_size = DR_GROUP_SIZE (vinfo_for_stmt (first_stmt));

      DR_GROUP_STORE_COUNT (vinfo_for_stmt (first_stmt))++;

      /* We vectorize all the stmts of the interleaving group when we
         reach the last stmt in the group.  */
      if (DR_GROUP_STORE_COUNT (vinfo_for_stmt (first_stmt)) 
          < DR_GROUP_SIZE (vinfo_for_stmt (first_stmt)))
        {
          *vec_stmt = NULL_TREE;
          return true;
        }
    }
  else 
    {
      first_stmt = stmt;
      first_dr = dr;
      group_size = 1;
    }
  
  dr_chain = VEC_alloc (tree, heap, group_size);
  oprnds = VEC_alloc (tree, heap, group_size);

  alignment_support_cheme = vect_supportable_dr_alignment (first_dr);
  gcc_assert (alignment_support_cheme);
  gcc_assert (alignment_support_cheme == dr_aligned);  /* FORNOW */

  /* In case the vectorization factor (VF) is bigger than the number
     of elements that we can fit in a vectype (nunits), we have to generate
     more than one vector stmt - i.e - we need to "unroll" the
     vector stmt by a factor VF/nunits.  For more details see documentation in 
     vect_get_vec_def_for_copy_stmt.  */

  /* In case of interleaving (non-unit strided access):

        S1:  &base + 2 = x2
        S2:  &base = x0
        S3:  &base + 1 = x1
        S4:  &base + 3 = x3

     We create vectorized storess starting from base address (the access of the
     first stmt in the chain (S2 in the above example), when the last store stmt
     of the chain (S4) is reached:

        VS1: &base = vx2
        VS2: &base + vec_size*1 = vx0
        VS3: &base + vec_size*2 = vx1
        VS4: &base + vec_size*3 = vx3

     Then permutation statements are generated:

        VS5: vx5 = VEC_INTERLEAVE_HIGH_EXPR < vx0, vx3 >
        VS6: vx6 = VEC_INTERLEAVE_LOW_EXPR < vx0, vx3 >
        ...
        
     And they are put in STMT_VINFO_VEC_STMT of the corresponding scalar stmts
     (the order of the data-refs in the output of vect_permute_store_chain
     corresponds to the order of scalar stmts in the interleaving chain - see
     the documentaion of vect_permute_store_chain()).

     In case of both multiple types and interleaving, above vector stores and
     permutation stmts are created for every copy. The result vector stmts are
     put in STMT_VINFO_VEC_STMT for the first copy and in the corresponding
     STMT_VINFO_RELATED_STMT for the next copies.     
  */

  prev_stmt_info = NULL;
  for (j = 0; j < ncopies; j++)
    {
      tree new_stmt;
      tree ptr_incr;

      if (j == 0)
        {
          /* For interleaved stores we collect vectorized defs for all the 
             stores in the group in DR_CHAIN and OPRNDS. DR_CHAIN is then used
             as an input to vect_permute_store_chain(), and OPRNDS as an input
             to vect_get_vec_def_for_stmt_copy() for the next copy.
             If the store is not strided, GROUP_SIZE is 1, and DR_CHAIN and
             OPRNDS are of size 1.
          */
          next_stmt = first_stmt;         
          for (i = 0; i < group_size; i++)
            {
              /* Since gaps are not supported for interleaved stores, GROUP_SIZE
                 is the exact number of stmts in the chain. Therefore, NEXT_STMT
                 can't be NULL_TREE.  In case that there is no interleaving, 
                 GROUP_SIZE is 1, and only one iteration of the loop will be 
                 executed.
              */
              gcc_assert (next_stmt);
              op = TREE_OPERAND (next_stmt, 1);
              vec_oprnd = vect_get_vec_def_for_operand (op, next_stmt, NULL);
              VEC_quick_push(tree, dr_chain, vec_oprnd); 
              VEC_quick_push(tree, oprnds, vec_oprnd); 
              next_stmt = DR_GROUP_NEXT_DR (vinfo_for_stmt (next_stmt));
            }
          dataref_ptr = vect_create_data_ref_ptr (first_stmt, bsi, NULL_TREE, 
                                                  &dummy, &ptr_incr, false);
        }
      else 
        {
          /* For interleaved stores we created vectorized defs for all the 
             defs stored in OPRNDS in the previous iteration (previous copy). 
             DR_CHAIN is then used as an input to vect_permute_store_chain(), 
             and OPRNDS as an input to vect_get_vec_def_for_stmt_copy() for the 
             next copy.
             If the store is not strided, GROUP_SIZE is 1, and DR_CHAIN and
             OPRNDS are of size 1.
          */
          for (i = 0; i < group_size; i++)
            {
              vec_oprnd = vect_get_vec_def_for_stmt_copy (dt, 
                                                   VEC_index (tree, oprnds, i));
              VEC_replace(tree, dr_chain, i, vec_oprnd);
              VEC_replace(tree, oprnds, i, vec_oprnd);
            }
          dataref_ptr = bump_vector_ptr (dataref_ptr, ptr_incr, bsi, stmt);
        }

      if (strided_store)
        {
          result_chain = VEC_alloc (tree, heap, group_size);     
          /* Permute.  */
          if (!vect_permute_store_chain (dr_chain, group_size, stmt, bsi, 
                                         &result_chain))
            return false;
        }

      next_stmt = first_stmt;
      for (i = 0; i < group_size; i++)
        {
          /* For strided stores vectorized defs are interleaved in 
             vect_permute_store_chain().  */
          if (strided_store)
            vec_oprnd = VEC_index(tree, result_chain, i);

          data_ref = build_fold_indirect_ref (dataref_ptr);
          /* Arguments are ready. Create the new vector stmt.  */
          new_stmt = build2 (MODIFY_EXPR, vectype, data_ref, vec_oprnd);
          vect_finish_stmt_generation (stmt, new_stmt, bsi);

          /* Set the V_MAY_DEFS for the vector pointer. If this virtual def has a 
             use outside the loop and a loop peel is performed then the def may be 
             renamed by the peel.  Mark it for renaming so the later use will also 
             be renamed.  */
          copy_virtual_operands (new_stmt, next_stmt);
          if (j == 0)
            {
              /* The original store is deleted so the same SSA_NAMEs can be used.  
               */
              FOR_EACH_SSA_TREE_OPERAND (def, next_stmt, iter, SSA_OP_VMAYDEF)
                {
                  SSA_NAME_DEF_STMT (def) = new_stmt;
                  mark_sym_for_renaming (SSA_NAME_VAR (def));
                }
              
              STMT_VINFO_VEC_STMT (stmt_info) = *vec_stmt =  new_stmt;
            }
          else
            {
              /* Create new names for all the definitions created by COPY and
                 add replacement mappings for each new name.  */
              FOR_EACH_SSA_DEF_OPERAND (def_p, new_stmt, iter, SSA_OP_VMAYDEF)
                {
                  create_new_def_for (DEF_FROM_PTR (def_p), new_stmt, def_p);
                  mark_sym_for_renaming (SSA_NAME_VAR (DEF_FROM_PTR (def_p)));
                }
              
              STMT_VINFO_RELATED_STMT (prev_stmt_info) = new_stmt;
            }

          prev_stmt_info = vinfo_for_stmt (new_stmt);
                  next_stmt = DR_GROUP_NEXT_DR (vinfo_for_stmt (next_stmt));
          if (!next_stmt)
            break;
          /* Bump the vector pointer.  */
          dataref_ptr = bump_vector_ptr (dataref_ptr, ptr_incr, bsi, stmt);
        }
    }

  return true;
}


/* Function vect_setup_realignment
  
   This function is called when vectorizing an unaligned load using
   the dr_unaligned_software_pipeline scheme.
   This function generates the following code at the loop prolog:

      p = initial_addr;
      msq_init = *(floor(p));   # prolog load
      realignment_token = call target_builtin; 
    loop:
      msq = phi (msq_init, ---)

   The code above sets up a new (vector) pointer, pointing to the first 
   location accessed by STMT, and a "floor-aligned" load using that pointer.
   It also generates code to compute the "realignment-token" (if the relevant
   target hook was defined), and creates a phi-node at the loop-header bb
   whose arguments are the result of the prolog-load (created by this
   function) and the result of a load that takes place in the loop (to be
   created by the caller to this function).
   The caller to this function uses the phi-result (msq) to create the 
   realignment code inside the loop, and sets up the missing phi argument,
   as follows:

    loop: 
      msq = phi (msq_init, lsq)
      lsq = *(floor(p'));        # load in loop
      result = realign_load (msq, lsq, realignment_token);

   Input:
   STMT - (scalar) load stmt to be vectorized. This load accesses
          a memory location that may be unaligned.
   BSI - place where new code is to be inserted.
   
   Output:
   REALIGNMENT_TOKEN - the result of a call to the builtin_mask_for_load
                       target hook, if defined.
   Return value - the result of the loop-header phi node.  */

static tree
vect_setup_realignment (tree stmt, block_stmt_iterator *bsi,
                        tree *realignment_token)
{
  stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
  tree vectype = STMT_VINFO_VECTYPE (stmt_info);
  loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
  struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
  edge pe = loop_preheader_edge (loop);
  tree scalar_dest = TREE_OPERAND (stmt, 0);
  tree vec_dest;
  tree init_addr;
  tree inc;
  tree ptr;
  tree data_ref;
  tree new_stmt;
  basic_block new_bb;
  tree msq_init;
  tree new_temp;
  tree phi_stmt;
  tree msq;

  /* 1. Create msq_init = *(floor(p1)) in the loop preheader  */
  vec_dest = vect_create_destination_var (scalar_dest, vectype);
  ptr = vect_create_data_ref_ptr (stmt, bsi, NULL_TREE, &init_addr, &inc, true);
  data_ref = build1 (ALIGN_INDIRECT_REF, vectype, ptr);
  new_stmt = build2 (MODIFY_EXPR, vectype, vec_dest, data_ref);
  new_temp = make_ssa_name (vec_dest, new_stmt);
  TREE_OPERAND (new_stmt, 0) = new_temp;
  new_bb = bsi_insert_on_edge_immediate (pe, new_stmt);
  gcc_assert (!new_bb);
  msq_init = TREE_OPERAND (new_stmt, 0);
  copy_virtual_operands (new_stmt, stmt);
  update_vuses_to_preheader (new_stmt, loop);

  /* 2. Create permutation mask, if required, in loop preheader.  */
  if (targetm.vectorize.builtin_mask_for_load)
    {
      tree builtin_decl;
      tree params = build_tree_list (NULL_TREE, init_addr);

      vec_dest = vect_create_destination_var (scalar_dest, 
                                                        TREE_TYPE (new_stmt));
      builtin_decl = targetm.vectorize.builtin_mask_for_load ();
      new_stmt = build_function_call_expr (builtin_decl, params);
      new_stmt = build2 (MODIFY_EXPR, void_type_node, vec_dest, new_stmt);
      new_temp = make_ssa_name (vec_dest, new_stmt);
      TREE_OPERAND (new_stmt, 0) = new_temp;
      new_bb = bsi_insert_on_edge_immediate (pe, new_stmt);
      gcc_assert (!new_bb);
      *realignment_token = TREE_OPERAND (new_stmt, 0);

      /* The result of the CALL_EXPR to this builtin is determined from
         the value of the parameter and no global variables are touched
         which makes the builtin a "const" function.  Requiring the
         builtin to have the "const" attribute makes it unnecessary
         to call mark_call_clobbered.  */
      gcc_assert (TREE_READONLY (builtin_decl));
    }

  /* 3. Create msq = phi <msq_init, lsq> in loop  */
  vec_dest = vect_create_destination_var (scalar_dest, vectype);
  msq = make_ssa_name (vec_dest, NULL_TREE);
  phi_stmt = create_phi_node (msq, loop->header); 
  SSA_NAME_DEF_STMT (msq) = phi_stmt;
  add_phi_arg (phi_stmt, msq_init, loop_preheader_edge (loop));

  return msq;
}


/* Function vect_strided_load_supported.

   Returns TRUE is EXTRACT_EVEN and EXTRACT_ODD operations are supported,
   and FALSE otherwise.  */

static bool
vect_strided_load_supported (tree vectype)
{
  optab perm_even_optab, perm_odd_optab;
  int mode;

  mode = (int) TYPE_MODE (vectype);

  perm_even_optab = optab_for_tree_code (VEC_EXTRACT_EVEN_EXPR, vectype);
  if (!perm_even_optab)
    {
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "no optab for perm_even.");
      return false;
    }

  if (perm_even_optab->handlers[mode].insn_code == CODE_FOR_nothing)
    {
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "perm_even op not supported by target.");
      return false;
    }

  perm_odd_optab = optab_for_tree_code (VEC_EXTRACT_ODD_EXPR, vectype);
  if (!perm_odd_optab)
    {
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "no optab for perm_odd.");
      return false;
    }

  if (perm_odd_optab->handlers[mode].insn_code == CODE_FOR_nothing)
    {
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "perm_odd op not supported by target.");
      return false;
    }
  return true;
}


/* Function vect_permute_load_chain.

   Given a chain of interleaved loads in DR_CHAIN of LENGTH that must be
   a power of 2, generate extract_even/odd stmts to reorder the input data 
   correctly. Return the final references for loads in RESULT_CHAIN.

   E.g., LENGTH is 4 and the scalar type is short, i.e., VF is 8.
   The input is 4 vectors each containg 8 elements. We assign a number to each 
   element, the input sequence is:

   1st vec:   0  1  2  3  4  5  6  7
   2nd vec:   8  9 10 11 12 13 14 15
   3rd vec:  16 17 18 19 20 21 22 23 
   4th vec:  24 25 26 27 28 29 30 31

   The output sequence should be:

   1st vec:  0 4  8 12 16 20 24 28
   2nd vec:  1 5  9 13 17 21 25 29
   3rd vec:  2 6 10 14 18 22 26 30 
   4th vec:  3 7 11 15 19 23 27 31

   i.e., the first output vector should contain the first elements of each
   interleaving group, etc.

   We use extract_even/odd instructions to create such output. The input of each
   extract_even/odd operation is two vectors
   1st vec    2nd vec 
   0 1 2 3    4 5 6 7 

   and the output is the vector of extracted even/odd elements. The output of 
   extract_even will be:   0 2 4 6
   and of extract_odd:     1 3 5 7

   
   The permutaion is done in log LENGTH stages. In each stage extract_even and 
   extract_odd stmts are created for each pair of vectors in DR_CHAIN in their 
   order. In our example, 

   E1: extract_even (1st vec, 2nd vec)
   E2: extract_odd (1st vec, 2nd vec)
   E3: extract_even (3rd vec, 4th vec)
   E4: extract_odd (3rd vec, 4th vec)

   The output for the first stage will be:

   E1:  0  2  4  6  8 10 12 14
   E2:  1  3  5  7  9 11 13 15
   E3: 16 18 20 22 24 26 28 30 
   E4: 17 19 21 23 25 27 29 31

   In order to proceed and create the correct sequence for the next stage (or
   for the correct output, if the second stage is the last one, as in our 
   example), we first put the output of extract_even operation and then the 
   output of extract_odd in RESULT_CHAIN (which is then copied to DR_CHAIN).
   The input for the second stage is:

   1st vec (E1):  0  2  4  6  8 10 12 14
   2nd vec (E3): 16 18 20 22 24 26 28 30  
   3rd vec (E2):  1  3  5  7  9 11 13 15    
   4th vec (E4): 17 19 21 23 25 27 29 31

   The output of the second stage:

   E1: 0 4  8 12 16 20 24 28
   E2: 2 6 10 14 18 22 26 30
   E3: 1 5  9 13 17 21 25 29
   E4: 3 7 11 15 19 23 27 31

   And RESULT_CHAIN after reordering:

   1st vec (E1):  0 4  8 12 16 20 24 28
   2nd vec (E3):  1 5  9 13 17 21 25 29
   3rd vec (E2):  2 6 10 14 18 22 26 30 
   4th vec (E4):  3 7 11 15 19 23 27 31.  */

static bool
vect_permute_load_chain (VEC(tree,heap) *dr_chain, 
                         unsigned int length, 
                         tree stmt, 
                         block_stmt_iterator *bsi,
                         VEC(tree,heap) **result_chain)
{
  tree perm_dest, perm_stmt, data_ref, first_vect, second_vect;
  tree vectype = STMT_VINFO_VECTYPE (vinfo_for_stmt (stmt));
  int i;
  unsigned int j;

  /* Check that the operation is supported.  */
  if (!vect_strided_load_supported (vectype))
    return false;

  *result_chain = VEC_copy (tree, heap, dr_chain);
  for (i = 0; i < exact_log2 (length); i++)
    {
      for (j = 0; j < length; j +=2)
        {
          first_vect = VEC_index (tree, dr_chain, j);
          second_vect = VEC_index (tree, dr_chain, j+1);

          /* data_ref = permute_even (first_data_ref, second_data_ref);  */
          perm_dest = create_tmp_var (vectype, "vect_perm_even");
          add_referenced_var (perm_dest);
         
          perm_stmt = build2 (MODIFY_EXPR, vectype, perm_dest,
                              build2 (VEC_EXTRACT_EVEN_EXPR, vectype, 
                                      first_vect, second_vect));

          data_ref = make_ssa_name (perm_dest, perm_stmt);
          TREE_OPERAND (perm_stmt, 0) = data_ref;
          vect_finish_stmt_generation (stmt, perm_stmt, bsi);
          mark_new_vars_to_rename (perm_stmt);

          VEC_replace (tree, *result_chain, j/2, data_ref);           
              
          /* data_ref = permute_odd (first_data_ref, second_data_ref);  */
          perm_dest = create_tmp_var (vectype, "vect_perm_odd");
          add_referenced_var (perm_dest);

          perm_stmt = build2 (MODIFY_EXPR, vectype, perm_dest,
                              build2 (VEC_EXTRACT_ODD_EXPR, vectype, 
                                      first_vect, second_vect));
          data_ref = make_ssa_name (perm_dest, perm_stmt);
          TREE_OPERAND (perm_stmt, 0) = data_ref;
          vect_finish_stmt_generation (stmt, perm_stmt, bsi);
          mark_new_vars_to_rename (perm_stmt);

          VEC_replace (tree, *result_chain, j/2+length/2, data_ref);
        }
      dr_chain = VEC_copy (tree, heap, *result_chain);
    }
  return true;
}


/* Function vect_transform_strided_load.

   Given a chain of input interleaved data-refs (in DR_CHAIN), build statements
   to perform their permutation and ascribe the result vectorized statements to
   the scalar statements.
*/

static bool
vect_transform_strided_load (tree stmt, VEC(tree,heap) *dr_chain, int size,
                             block_stmt_iterator *bsi)
{
  stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
  tree first_stmt = DR_GROUP_FIRST_DR (stmt_info);
  tree next_stmt, new_stmt;
  VEC(tree,heap) *result_chain = NULL;
  unsigned int i, gap_count;
  tree tmp_data_ref;

  /* DR_CHAIN contains input data-refs that are a part of the interleaving. 
     RESULT_CHAIN is the output of vect_permute_load_chain, it contains permuted 
     vectors, that are ready for vector computation.  */
  result_chain = VEC_alloc (tree, heap, size);
  /* Permute.  */
  if (!vect_permute_load_chain (dr_chain, size, stmt, bsi, &result_chain))
    return false;

  /* Put a permuted data-ref in the VECTORIZED_STMT field.  
     Since we scan the chain starting from it's first node, their order 
     corresponds the order of data-refs in RESULT_CHAIN.  */
  next_stmt = first_stmt;
  gap_count = 1;
  for (i = 0; VEC_iterate(tree, result_chain, i, tmp_data_ref); i++)
    {
      if (!next_stmt)
        break;

      /* Skip the gaps. Loads created for the gaps will be removed by dead
       code elimination pass later.
       DR_GROUP_GAP is the number of steps in elements from the previous
       access (if there is no gap DR_GROUP_GAP is 1). We skip loads that
       correspond to the gaps.
      */
      if (gap_count < DR_GROUP_GAP (vinfo_for_stmt (next_stmt)))
      {
        gap_count++;
        continue;
      }

      while (next_stmt)
        {
          new_stmt = SSA_NAME_DEF_STMT (tmp_data_ref);
          /* We assume that if VEC_STMT is not NULL, this is a case of multiple
             copies, and we put the new vector statement in the first available
             RELATED_STMT.  */
          if (!STMT_VINFO_VEC_STMT (vinfo_for_stmt (next_stmt)))
            STMT_VINFO_VEC_STMT (vinfo_for_stmt (next_stmt)) = new_stmt;
          else
            {
              tree prev_stmt = STMT_VINFO_VEC_STMT (vinfo_for_stmt (next_stmt));
              tree rel_stmt = STMT_VINFO_RELATED_STMT (
                                                       vinfo_for_stmt (prev_stmt));
              while (rel_stmt)
                {
                  prev_stmt = rel_stmt;
                  rel_stmt = STMT_VINFO_RELATED_STMT (vinfo_for_stmt (rel_stmt));
                }
              STMT_VINFO_RELATED_STMT (vinfo_for_stmt (prev_stmt)) = new_stmt;
            }
          next_stmt = DR_GROUP_NEXT_DR (vinfo_for_stmt (next_stmt));
          gap_count = 1;
          /* If NEXT_STMT accesses the same DR as the previous statement,
             put the same TMP_DATA_REF as its vectorized statement; otherwise
             get the next data-ref from RESULT_CHAIN.  */
          if (!next_stmt || !DR_GROUP_SAME_DR_STMT (vinfo_for_stmt (next_stmt)))
            break;
        }
    }
  return true;
}


/* vectorizable_load.

   Check if STMT reads a non scalar data-ref (array/pointer/structure) that 
   can be vectorized. 
   If VEC_STMT is also passed, vectorize the STMT: create a vectorized 
   stmt to replace it, put it in VEC_STMT, and insert it at BSI.
   Return FALSE if not a vectorizable STMT, TRUE otherwise.  */

bool
vectorizable_load (tree stmt, block_stmt_iterator *bsi, tree *vec_stmt)
{
  tree scalar_dest;
  tree vec_dest = NULL;
  tree data_ref = NULL;
  tree op;
  stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
  stmt_vec_info prev_stmt_info; 
  loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
  struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
  struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info), *first_dr;
  tree vectype = STMT_VINFO_VECTYPE (stmt_info);
  tree new_temp;
  int mode;
  tree new_stmt = NULL_TREE;
  tree dummy;
  enum dr_alignment_support alignment_support_cheme;
  tree dataref_ptr = NULL_TREE;
  tree ptr_incr;
  int nunits = TYPE_VECTOR_SUBPARTS (vectype);
  int ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits;
  int i, j, group_size;
  tree msq = NULL_TREE, lsq;
  tree offset = NULL_TREE;
  tree realignment_token = NULL_TREE;
  tree phi_stmt = NULL_TREE;
  VEC(tree,heap) *dr_chain = NULL;
  bool strided_load = false;
  tree first_stmt;

  /* Is vectorizable load? */
  if (!STMT_VINFO_RELEVANT_P (stmt_info))
    return false;

  gcc_assert (STMT_VINFO_DEF_TYPE (stmt_info) == vect_loop_def);

  if (STMT_VINFO_LIVE_P (stmt_info))
    {
      /* FORNOW: not yet supported.  */
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "value used after loop.");
      return false;
    }

  if (TREE_CODE (stmt) != MODIFY_EXPR)
    return false;

  scalar_dest = TREE_OPERAND (stmt, 0);
  if (TREE_CODE (scalar_dest) != SSA_NAME)
    return false;

  op = TREE_OPERAND (stmt, 1);
  if (TREE_CODE (op) != ARRAY_REF 
      && TREE_CODE (op) != INDIRECT_REF
      && !DR_GROUP_FIRST_DR (stmt_info))
    return false;

  if (!STMT_VINFO_DATA_REF (stmt_info))
    return false;

  mode = (int) TYPE_MODE (vectype);

  /* FORNOW. In some cases can vectorize even if data-type not supported
    (e.g. - data copies).  */
  if (mov_optab->handlers[mode].insn_code == CODE_FOR_nothing)
    {
      if (vect_print_dump_info (REPORT_DETAILS))
        fprintf (vect_dump, "Aligned load, but unsupported type.");
      return false;
    }

  /* Check if the load is a part of an interleaving chain.  */
  if (DR_GROUP_FIRST_DR (stmt_info))
    {
      strided_load = true;

      /* Check if interleaving is supported.  */
      if (!vect_strided_load_supported (vectype))
        return false;
    }

  if (!vec_stmt) /* transformation not required.  */
    {
      STMT_VINFO_TYPE (stmt_info) = load_vec_info_type;
      return true;
    }

  /** Transform.  **/

  if (vect_print_dump_info (REPORT_DETAILS))
    fprintf (vect_dump, "transform load.");

  if (strided_load)
    {
      first_stmt = DR_GROUP_FIRST_DR (stmt_info);
      /* Check if the chain of loads is already vectorized.  */
      if (STMT_VINFO_VEC_STMT (vinfo_for_stmt (first_stmt)))
        {
          *vec_stmt = STMT_VINFO_VEC_STMT (stmt_info);
          return true;
        }
      first_dr = STMT_VINFO_DATA_REF (vinfo_for_stmt (first_stmt));
      group_size = DR_GROUP_SIZE (vinfo_for_stmt (first_stmt));
      dr_chain = VEC_alloc (tree, heap, group_size);
    }
  else
    {
      first_stmt = stmt;
      first_dr = dr;
      group_size = 1;
    }

  alignment_support_cheme = vect_supportable_dr_alignment (first_dr);
  gcc_assert (alignment_support_cheme);


  /* In case the vectorization factor (VF) is bigger than the number
     of elements that we can fit in a vectype (nunits), we have to generate
     more than one vector stmt - i.e - we need to "unroll" the
     vector stmt by a factor VF/nunits. In doing so, we record a pointer
     from one copy of the vector stmt to the next, in the field
     STMT_VINFO_RELATED_STMT. This is necessary in order to allow following
     stages to find the correct vector defs to be used when vectorizing
     stmts that use the defs of the current stmt. The example below illustrates
     the vectorization process when VF=16 and nunits=4 (i.e - we need to create
     4 vectorized stmts):

     before vectorization:
                                RELATED_STMT    VEC_STMT
        S1:     x = memref      -               -
        S2:     z = x + 1       -               -

     step 1: vectorize stmt S1:
        We first create the vector stmt VS1_0, and, as usual, record a
        pointer to it in the STMT_VINFO_VEC_STMT of the scalar stmt S1.
        Next, we create the vector stmt VS1_1, and record a pointer to
        it in the STMT_VINFO_RELATED_STMT of the vector stmt VS1_0.
        Similarly, for VS1_2 and VS1_3. This is the resulting chain of
        stmts and pointers:
                                RELATED_STMT    VEC_STMT
        VS1_0:  vx0 = memref0   VS1_1           -
        VS1_1:  vx1 = memref1   VS1_2           -
        VS1_2:  vx2 = memref2   VS1_3           -
        VS1_3:  vx3 = memref3   -               -
        S1:     x = load        -               VS1_0
        S2:     z = x + 1       -               -

     See in documentation in vect_get_vec_def_for_stmt_copy for how the 
     information we recorded in RELATED_STMT field is used to vectorize 
     stmt S2.  */

  /* In case of interleaving (non-unit strided access):

     S1:  x2 = &base + 2
     S2:  x0 = &base
     S3:  x1 = &base + 1
     S4:  x3 = &base + 3

     Vectorized loads are created in the order of memory accesses 
     starting from the access of the first stmt of the chain:

     VS1: vx0 = &base
     VS2: vx1 = &base + vec_size*1
     VS3: vx3 = &base + vec_size*2
     VS4: vx4 = &base + vec_size*3

     Then permutation statements are generated:

     VS5: vx5 = VEC_EXTRACT_EVEN_EXPR < vx0, vx1 >
     VS6: vx6 = VEC_EXTRACT_ODD_EXPR < vx0, vx1 >
       ...

     And they are put in STMT_VINFO_VEC_STMT of the corresponding scalar stmts
     (the order of the data-refs in the output of vect_permute_load_chain
     corresponds to the order of scalar stmts in the interleaving chain - see
     the documentaion of vect_permute_load_chain()).
     The generation of permutation stmts and recording them in
     STMT_VINFO_VEC_STMT is done in vect_transform_strided_load().

     In case of both multiple types and interleaving, the vector loads and 
     permutation stmts above are created for every copy. The result vector stmts
     are put in STMT_VINFO_VEC_STMT for the first copy and in the corresponding
     STMT_VINFO_RELATED_STMT for the next copies.  */

  /* If the data reference is aligned (dr_aligned) or potentially unaligned
     on a target that supports unaligned accesses (dr_unaligned_supported)
     we generate the following code:
         p = initial_addr;
         indx = 0;
         loop {
           p = p + indx * vectype_size;
           vec_dest = *(p);
           indx = indx + 1;
         }

     Otherwise, the data reference is potentially unaligned on a target that
     does not support unaligned accesses (dr_unaligned_software_pipeline) - 
     then generate the following code, in which the data in each iteration is
     obtained by two vector loads, one from the previous iteration, and one
     from the current iteration:
         p1 = initial_addr;
         msq_init = *(floor(p1))
         p2 = initial_addr + VS - 1;
         realignment_token = call target_builtin;
         indx = 0;
         loop {
           p2 = p2 + indx * vectype_size
           lsq = *(floor(p2))
           vec_dest = realign_load (msq, lsq, realignment_token)
           indx = indx + 1;
           msq = lsq;