#include "real.h"
/* Utility functions for the code transformation. */
-static bool vect_transform_stmt (tree, block_stmt_iterator *, bool *, slp_tree);
+static bool vect_transform_stmt (gimple, gimple_stmt_iterator *, bool *,
+ slp_tree);
static tree vect_create_destination_var (tree, tree);
static tree vect_create_data_ref_ptr
- (tree, struct loop*, tree, tree *, tree *, bool, tree, bool *);
+ (gimple, struct loop*, tree, tree *, gimple *, bool, bool *);
static tree vect_create_addr_base_for_vector_ref
- (tree, tree *, tree, struct loop *);
+ (gimple, gimple_seq *, tree, struct loop *);
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, tree, block_stmt_iterator *);
+static tree vect_get_vec_def_for_operand (tree, gimple, tree *);
+static tree vect_init_vector (gimple, tree, tree, gimple_stmt_iterator *);
static void vect_finish_stmt_generation
- (tree stmt, tree vec_stmt, block_stmt_iterator *);
+ (gimple stmt, gimple vec_stmt, gimple_stmt_iterator *);
static bool vect_is_simple_cond (tree, loop_vec_info);
-static void vect_create_epilog_for_reduction (tree, tree, enum tree_code, tree);
-static tree get_initial_def_for_reduction (tree, tree, tree *);
+static void vect_create_epilog_for_reduction
+ (tree, gimple, int, enum tree_code, gimple);
+static tree get_initial_def_for_reduction (gimple, tree, tree *);
/* Utility function dealing with loop peeling (not peeling itself). */
static void vect_generate_tmps_on_preheader
static int
-cost_for_stmt (tree stmt)
+cost_for_stmt (gimple stmt)
{
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
/* Requires loop versioning tests to handle misalignment. */
- if (VEC_length (tree, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo)))
+ if (VEC_length (gimple, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo)))
{
/* FIXME: Make cost depend on complexity of individual check. */
vec_outside_cost +=
- VEC_length (tree, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo));
+ VEC_length (gimple, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo));
if (vect_print_dump_info (REPORT_COST))
fprintf (vect_dump, "cost model: Adding cost of checks for loop "
"versioning to treat misalignment.\n");
"versioning aliasing.\n");
}
- if (VEC_length (tree, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo))
+ if (VEC_length (gimple, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo))
|| VEC_length (ddr_p, LOOP_VINFO_MAY_ALIAS_DDRS (loop_vinfo)))
{
vec_outside_cost += TARG_COND_TAKEN_BRANCH_COST;
for (i = 0; i < nbbs; i++)
{
- block_stmt_iterator si;
+ gimple_stmt_iterator si;
basic_block bb = bbs[i];
if (bb->loop_father == loop->inner)
else
factor = 1;
- for (si = bsi_start (bb); !bsi_end_p (si); bsi_next (&si))
+ for (si = gsi_start_bb (bb); !gsi_end_p (si); gsi_next (&si))
{
- tree stmt = bsi_stmt (si);
+ gimple stmt = gsi_stmt (si);
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
/* Skip stmts that are not vectorized inside the loop. */
if (!STMT_VINFO_RELEVANT_P (stmt_info)
/* Add additional cost for the peeled instructions in prologue and epilogue
loop.
- FORNOW: If we dont know the value of peel_iters for prologue or epilogue
+ FORNOW: If we don't know the value of peel_iters for prologue or epilogue
at compile-time - we assume it's vf/2 (the worst would be vf-1).
TODO: Build an expression that represents peel_iters for prologue and
branches.
TODO: The back end may reorder the BBS's differently and reverse
- conditions/branch directions. Change the stimates below to
+ conditions/branch directions. Change the estimates below to
something more reasonable. */
if (runtime_test)
{
/* Cost model check occurs at versioning. */
- if (VEC_length (tree, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo))
+ if (VEC_length (gimple, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo))
|| VEC_length (ddr_p, LOOP_VINFO_MAY_ALIAS_DDRS (loop_vinfo)))
scalar_outside_cost += TARG_COND_NOT_TAKEN_BRANCH_COST;
else
enum tree_code code;
optab optab;
tree vectype;
- tree orig_stmt;
+ gimple stmt, orig_stmt;
tree reduction_op;
enum machine_mode mode;
- tree operation = GIMPLE_STMT_OPERAND (STMT_VINFO_STMT (stmt_info), 1);
- int op_type = TREE_CODE_LENGTH (TREE_CODE (operation));
loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
+
/* Cost of reduction op inside loop. */
STMT_VINFO_INSIDE_OF_LOOP_COST (stmt_info) += ncopies * TARG_VEC_STMT_COST;
- reduction_op = TREE_OPERAND (operation, op_type-1);
+ stmt = STMT_VINFO_STMT (stmt_info);
+
+ switch (get_gimple_rhs_class (gimple_assign_rhs_code (stmt)))
+ {
+ case GIMPLE_SINGLE_RHS:
+ gcc_assert (TREE_OPERAND_LENGTH (gimple_assign_rhs1 (stmt)) == ternary_op);
+ reduction_op = TREE_OPERAND (gimple_assign_rhs1 (stmt), 2);
+ break;
+ case GIMPLE_UNARY_RHS:
+ reduction_op = gimple_assign_rhs1 (stmt);
+ break;
+ case GIMPLE_BINARY_RHS:
+ reduction_op = gimple_assign_rhs2 (stmt);
+ break;
+ default:
+ gcc_unreachable ();
+ }
+
vectype = get_vectype_for_scalar_type (TREE_TYPE (reduction_op));
if (!vectype)
{
if (!orig_stmt)
orig_stmt = STMT_VINFO_STMT (stmt_info);
- code = TREE_CODE (GIMPLE_STMT_OPERAND (orig_stmt, 1));
+ code = gimple_assign_rhs_code (orig_stmt);
/* Add in cost for initial definition. */
outer_cost += TARG_SCALAR_TO_VEC_COST;
{
int vec_size_in_bits = tree_low_cst (TYPE_SIZE (vectype), 1);
tree bitsize =
- TYPE_SIZE (TREE_TYPE ( GIMPLE_STMT_OPERAND (orig_stmt, 0)));
+ TYPE_SIZE (TREE_TYPE (gimple_assign_lhs (orig_stmt)));
int element_bitsize = tree_low_cst (bitsize, 1);
int nelements = vec_size_in_bits / element_bitsize;
int i;
int inside_cost = 0, outside_cost = 0;
+ /* The SLP costs were already calculated during SLP tree build. */
+ if (PURE_SLP_STMT (stmt_info))
+ return;
+
inside_cost = ncopies * TARG_VEC_STMT_COST;
/* FORNOW: Assuming maximum 2 args per stmts. */
static int
vect_cost_strided_group_size (stmt_vec_info stmt_info)
{
- tree first_stmt = DR_GROUP_FIRST_DR (stmt_info);
+ gimple first_stmt = DR_GROUP_FIRST_DR (stmt_info);
if (first_stmt == STMT_VINFO_STMT (stmt_info))
return DR_GROUP_SIZE (stmt_info);
int group_size;
int inside_cost = 0, outside_cost = 0;
+ /* The SLP costs were already calculated during SLP tree build. */
+ if (PURE_SLP_STMT (stmt_info))
+ return;
+
if (dt == vect_constant_def || dt == vect_invariant_def)
outside_cost = TARG_SCALAR_TO_VEC_COST;
/* Strided access? */
- if (DR_GROUP_FIRST_DR (stmt_info))
+ if (DR_GROUP_FIRST_DR (stmt_info) && !slp_node)
group_size = vect_cost_strided_group_size (stmt_info);
/* Not a strided access. */
else
{
int group_size;
int alignment_support_cheme;
- tree first_stmt;
+ gimple first_stmt;
struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info), *first_dr;
int inside_cost = 0, outside_cost = 0;
+ /* The SLP costs were already calculated during SLP tree build. */
+ if (PURE_SLP_STMT (stmt_info))
+ return;
+
/* Strided accesses? */
first_stmt = DR_GROUP_FIRST_DR (stmt_info);
if (first_stmt && !slp_node)
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,
+vect_create_addr_base_for_vector_ref (gimple stmt,
+ gimple_seq *new_stmt_list,
tree offset,
struct loop *loop)
{
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info);
- struct loop *containing_loop = (bb_for_stmt (stmt))->loop_father;
+ struct loop *containing_loop = (gimple_bb (stmt))->loop_father;
tree data_ref_base = unshare_expr (DR_BASE_ADDRESS (dr));
tree base_name;
tree data_ref_base_var;
- tree new_base_stmt;
tree vec_stmt;
tree addr_base, addr_expr;
- tree dest, new_stmt;
+ tree dest;
+ gimple_seq seq = NULL;
tree base_offset = unshare_expr (DR_OFFSET (dr));
tree init = unshare_expr (DR_INIT (dr));
tree vect_ptr_type, addr_expr2;
base_name = build_fold_indirect_ref (data_ref_base);
data_ref_base_var = create_tmp_var (TREE_TYPE (data_ref_base), "batmp");
add_referenced_var (data_ref_base_var);
- data_ref_base = force_gimple_operand (data_ref_base, &new_base_stmt,
- true, data_ref_base_var);
- append_to_statement_list_force(new_base_stmt, new_stmt_list);
+ data_ref_base = force_gimple_operand (data_ref_base, &seq, true,
+ data_ref_base_var);
+ gimple_seq_add_seq (new_stmt_list, seq);
/* Create base_offset */
base_offset = size_binop (PLUS_EXPR, base_offset, init);
base_offset = fold_convert (sizetype, base_offset);
dest = create_tmp_var (TREE_TYPE (base_offset), "base_off");
add_referenced_var (dest);
- base_offset = force_gimple_operand (base_offset, &new_stmt, true, dest);
- append_to_statement_list_force (new_stmt, new_stmt_list);
+ base_offset = force_gimple_operand (base_offset, &seq, true, dest);
+ gimple_seq_add_seq (new_stmt_list, seq);
if (offset)
{
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_offset = force_gimple_operand (base_offset, &seq, false, tmp);
+ gimple_seq_add_seq (new_stmt_list, seq);
}
/* base + base_offset */
addr_expr2 = vect_get_new_vect_var (vect_ptr_type, vect_pointer_var,
get_name (base_name));
add_referenced_var (addr_expr2);
- vec_stmt = force_gimple_operand (vec_stmt, &new_stmt, false, addr_expr2);
- append_to_statement_list_force (new_stmt, new_stmt_list);
+ vec_stmt = force_gimple_operand (vec_stmt, &seq, false, addr_expr2);
+ gimple_seq_add_seq (new_stmt_list, seq);
if (vect_print_dump_info (REPORT_DETAILS))
{
Input:
1. STMT: a stmt that references memory. Expected to be of the form
- GIMPLE_MODIFY_STMT <name, data-ref> or
- GIMPLE_MODIFY_STMT <data-ref, name>.
+ GIMPLE_ASSIGN <name, data-ref> or
+ GIMPLE_ASSIGN <data-ref, name>.
2. AT_LOOP: the loop where the vector memref is to be created.
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.
- 5. TYPE: if not NULL indicates the required type of the data-ref
Output:
1. Declare a new ptr to vector_type, and have it point to the base of the
4. Return the pointer. */
static tree
-vect_create_data_ref_ptr (tree stmt, struct loop *at_loop,
- tree offset, tree *initial_address, tree *ptr_incr,
- bool only_init, tree type, bool *inv_p)
+vect_create_data_ref_ptr (gimple stmt, struct loop *at_loop,
+ tree offset, tree *initial_address, gimple *ptr_incr,
+ bool only_init, bool *inv_p)
{
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);
bool nested_in_vect_loop = nested_in_vect_loop_p (loop, stmt);
- struct loop *containing_loop = (bb_for_stmt (stmt))->loop_father;
+ struct loop *containing_loop = (gimple_bb (stmt))->loop_father;
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;
+ gimple vec_stmt;
+ gimple_seq new_stmt_list = NULL;
edge pe;
basic_block new_bb;
tree vect_ptr_init;
struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info);
tree vptr;
- block_stmt_iterator incr_bsi;
+ gimple_stmt_iterator incr_gsi;
bool insert_after;
tree indx_before_incr, indx_after_incr;
- tree incr;
+ gimple incr;
tree step;
/* Check the step (evolution) of the load in LOOP, and record
/* Create an expression for the first address accessed by this load
in LOOP. */
- base_name = build_fold_indirect_ref (unshare_expr (DR_BASE_ADDRESS (dr)));
+ base_name = build_fold_indirect_ref (unshare_expr (DR_BASE_ADDRESS (dr)));
if (vect_print_dump_info (REPORT_DETAILS))
{
}
/** (1) Create the new vector-pointer variable: **/
- if (type)
- vect_ptr_type = build_pointer_type (type);
- else
- vect_ptr_type = build_pointer_type (vectype);
+ 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);
pe = loop_preheader_edge (loop);
if (new_stmt_list)
{
- new_bb = bsi_insert_on_edge_immediate (pe, new_stmt_list);
+ new_bb = gsi_insert_seq_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 = build_gimple_modify_stmt (vect_ptr, vec_stmt);
+ vec_stmt = gimple_build_assign (vect_ptr,
+ fold_convert (vect_ptr_type, new_temp));
vect_ptr_init = make_ssa_name (vect_ptr, vec_stmt);
- GIMPLE_STMT_OPERAND (vec_stmt, 0) = vect_ptr_init;
- new_bb = bsi_insert_on_edge_immediate (pe, vec_stmt);
+ gimple_assign_set_lhs (vec_stmt, vect_ptr_init);
+ new_bb = gsi_insert_on_edge_immediate (pe, vec_stmt);
gcc_assert (!new_bb);
if (*inv_p)
step = size_zero_node;
- standard_iv_increment_position (loop, &incr_bsi, &insert_after);
+ standard_iv_increment_position (loop, &incr_gsi, &insert_after);
create_iv (vect_ptr_init,
fold_convert (vect_ptr_type, step),
- NULL_TREE, loop, &incr_bsi, insert_after,
+ NULL_TREE, loop, &incr_gsi, 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));
+ incr = gsi_stmt (incr_gsi);
+ set_vinfo_for_stmt (incr, new_stmt_vec_info (incr, loop_vinfo));
/* Copy the points-to information if it exists. */
if (DR_PTR_INFO (dr))
gcc_assert (nested_in_vect_loop);
if (!only_init)
{
- standard_iv_increment_position (containing_loop, &incr_bsi,
+ standard_iv_increment_position (containing_loop, &incr_gsi,
&insert_after);
create_iv (vptr, fold_convert (vect_ptr_type, DR_STEP (dr)), NULL_TREE,
- containing_loop, &incr_bsi, insert_after, &indx_before_incr,
+ containing_loop, &incr_gsi, 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));
+ incr = gsi_stmt (incr_gsi);
+ set_vinfo_for_stmt (incr, new_stmt_vec_info (incr, loop_vinfo));
/* Copy the points-to information if it exists. */
if (DR_PTR_INFO (dr))
*/
static tree
-bump_vector_ptr (tree dataref_ptr, tree ptr_incr, block_stmt_iterator *bsi,
- tree stmt, tree bump)
+bump_vector_ptr (tree dataref_ptr, gimple ptr_incr, gimple_stmt_iterator *gsi,
+ gimple stmt, tree bump)
{
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 = TYPE_SIZE_UNIT (vectype);
- tree incr_stmt;
+ gimple incr_stmt;
ssa_op_iter iter;
use_operand_p use_p;
tree new_dataref_ptr;
if (bump)
update = bump;
- incr_stmt = build_gimple_modify_stmt (ptr_var,
- build2 (POINTER_PLUS_EXPR, vptr_type,
- dataref_ptr, update));
+ incr_stmt = gimple_build_assign_with_ops (POINTER_PLUS_EXPR, ptr_var,
+ dataref_ptr, update);
new_dataref_ptr = make_ssa_name (ptr_var, incr_stmt);
- GIMPLE_STMT_OPERAND (incr_stmt, 0) = new_dataref_ptr;
- vect_finish_stmt_generation (stmt, incr_stmt, bsi);
+ gimple_assign_set_lhs (incr_stmt, new_dataref_ptr);
+ vect_finish_stmt_generation (stmt, incr_stmt, gsi);
/* Copy the points-to information if it exists. */
if (DR_PTR_INFO (dr))
It will be used in the vectorization of STMT. */
static tree
-vect_init_vector (tree stmt, tree vector_var, tree vector_type,
- block_stmt_iterator *bsi)
+vect_init_vector (gimple stmt, tree vector_var, tree vector_type,
+ gimple_stmt_iterator *gsi)
{
stmt_vec_info stmt_vinfo = vinfo_for_stmt (stmt);
tree new_var;
- tree init_stmt;
+ gimple init_stmt;
tree vec_oprnd;
edge pe;
tree new_temp;
new_var = vect_get_new_vect_var (vector_type, vect_simple_var, "cst_");
add_referenced_var (new_var);
- init_stmt = build_gimple_modify_stmt (new_var, vector_var);
+ init_stmt = gimple_build_assign (new_var, vector_var);
new_temp = make_ssa_name (new_var, init_stmt);
- GIMPLE_STMT_OPERAND (init_stmt, 0) = new_temp;
+ gimple_assign_set_lhs (init_stmt, new_temp);
- if (bsi)
- vect_finish_stmt_generation (stmt, init_stmt, bsi);
+ if (gsi)
+ vect_finish_stmt_generation (stmt, init_stmt, gsi);
else
{
loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_vinfo);
if (nested_in_vect_loop_p (loop, stmt))
loop = loop->inner;
pe = loop_preheader_edge (loop);
- new_bb = bsi_insert_on_edge_immediate (pe, init_stmt);
+ new_bb = gsi_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);
+ print_gimple_stmt (vect_dump, init_stmt, 0, TDF_SLIM);
}
- vec_oprnd = GIMPLE_STMT_OPERAND (init_stmt, 0);
+ vec_oprnd = gimple_assign_lhs (init_stmt);
return vec_oprnd;
}
vect_get_constant_vectors (slp_tree slp_node, VEC(tree,heap) **vec_oprnds,
unsigned int op_num)
{
- VEC (tree, heap) *stmts = SLP_TREE_SCALAR_STMTS (slp_node);
- tree stmt = VEC_index (tree, stmts, 0);
+ VEC (gimple, heap) *stmts = SLP_TREE_SCALAR_STMTS (slp_node);
+ gimple stmt = VEC_index (gimple, stmts, 0);
stmt_vec_info stmt_vinfo = vinfo_for_stmt (stmt);
tree vectype = STMT_VINFO_VECTYPE (stmt_vinfo);
int nunits = TYPE_VECTOR_SUBPARTS (vectype);
tree t = NULL_TREE;
int j, number_of_places_left_in_vector;
tree vector_type;
- tree op, vop, operation;
- int group_size = VEC_length (tree, stmts);
+ tree op, vop;
+ int group_size = VEC_length (gimple, stmts);
unsigned int vec_num, i;
int number_of_copies = 1;
bool is_store = false;
created vectors. It is greater than 1 if unrolling is performed.
For example, we have two scalar operands, s1 and s2 (e.g., group of
- strided accesses of size two), while NUINTS is four (i.e., four scalars
+ strided accesses of size two), while NUNITS is four (i.e., four scalars
of this type can be packed in a vector). The output vector will contain
two copies of each scalar operand: {s1, s2, s1, s2}. (NUMBER_OF_COPIES
will be 2).
If GROUP_SIZE > NUNITS, the scalars will be split into several vectors
containing the operands.
- For example, NUINTS is four as before, and the group size is 8
+ For example, NUNITS is four as before, and the group size is 8
(s1, s2, ..., s8). We will create two vectors {s1, s2, s3, s4} and
{s5, s6, s7, s8}. */
constant_p = true;
for (j = 0; j < number_of_copies; j++)
{
- for (i = group_size - 1; VEC_iterate (tree, stmts, i, stmt); i--)
+ for (i = group_size - 1; VEC_iterate (gimple, stmts, i, stmt); i--)
{
- operation = GIMPLE_STMT_OPERAND (stmt, 1);
if (is_store)
- op = operation;
+ op = gimple_assign_rhs1 (stmt);
else
- op = TREE_OPERAND (operation, op_num);
+ op = gimple_op (stmt, op_num + 1);
if (!CONSTANT_CLASS_P (op))
constant_p = false;
/* Get vectorized definitions from SLP_NODE that contains corresponding
vectorized def-stmts. */
-
+
static void
vect_get_slp_vect_defs (slp_tree slp_node, VEC (tree,heap) **vec_oprnds)
{
tree vec_oprnd;
- tree vec_def_stmt;
+ gimple vec_def_stmt;
unsigned int i;
gcc_assert (SLP_TREE_VEC_STMTS (slp_node));
- for (i = 0;
- VEC_iterate (tree, SLP_TREE_VEC_STMTS (slp_node), i, vec_def_stmt);
+ for (i = 0;
+ VEC_iterate (gimple, SLP_TREE_VEC_STMTS (slp_node), i, vec_def_stmt);
i++)
{
gcc_assert (vec_def_stmt);
- vec_oprnd = GIMPLE_STMT_OPERAND (vec_def_stmt, 0);
+ vec_oprnd = gimple_get_lhs (vec_def_stmt);
VEC_quick_push (tree, *vec_oprnds, vec_oprnd);
}
}
must be stored in the LEFT/RIGHT node of SLP_NODE, and we call
vect_get_slp_vect_defs() to retrieve them.
If VEC_OPRNDS1 is NULL, don't get vector defs for the second operand (from
- the right node. This is used when the second operand must remain scalar. */
+ the right node. This is used when the second operand must remain scalar. */
static void
vect_get_slp_defs (slp_tree slp_node, VEC (tree,heap) **vec_oprnds0,
VEC (tree,heap) **vec_oprnds1)
{
- tree operation, first_stmt;
+ gimple first_stmt;
+ enum tree_code code;
+ int number_of_vects;
+
+ /* The number of vector defs is determined by the number of vector statements
+ in the node from which we get those statements. */
+ if (SLP_TREE_LEFT (slp_node))
+ number_of_vects = SLP_TREE_NUMBER_OF_VEC_STMTS (SLP_TREE_LEFT (slp_node));
+ else
+ number_of_vects = SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node);
/* Allocate memory for vectorized defs. */
- *vec_oprnds0 = VEC_alloc (tree, heap,
- SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node));
+ *vec_oprnds0 = VEC_alloc (tree, heap, number_of_vects);
- /* SLP_NODE corresponds either to a group of stores or to a group of
+ /* SLP_NODE corresponds either to a group of stores or to a group of
unary/binary operations. We don't call this function for loads. */
- if (SLP_TREE_LEFT (slp_node))
- /* The defs are already vectorized. */
+ if (SLP_TREE_LEFT (slp_node))
+ /* The defs are already vectorized. */
vect_get_slp_vect_defs (SLP_TREE_LEFT (slp_node), vec_oprnds0);
else
/* Build vectors from scalar defs. */
vect_get_constant_vectors (slp_node, vec_oprnds0, 0);
- first_stmt = VEC_index (tree, SLP_TREE_SCALAR_STMTS (slp_node), 0);
+ first_stmt = VEC_index (gimple, SLP_TREE_SCALAR_STMTS (slp_node), 0);
if (STMT_VINFO_DATA_REF (vinfo_for_stmt (first_stmt)))
- /* Since we don't call this function with loads, this is a group of
+ /* Since we don't call this function with loads, this is a group of
stores. */
return;
- operation = GIMPLE_STMT_OPERAND (first_stmt, 1);
- if (TREE_OPERAND_LENGTH (operation) == unary_op || !vec_oprnds1)
+ code = gimple_assign_rhs_code (first_stmt);
+ if (get_gimple_rhs_class (code) != GIMPLE_BINARY_RHS || !vec_oprnds1)
return;
- *vec_oprnds1 = VEC_alloc (tree, heap,
- SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node));
+ /* The number of vector defs is determined by the number of vector statements
+ in the node from which we get those statements. */
+ if (SLP_TREE_RIGHT (slp_node))
+ number_of_vects = SLP_TREE_NUMBER_OF_VEC_STMTS (SLP_TREE_RIGHT (slp_node));
+ else
+ number_of_vects = SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node);
+
+ *vec_oprnds1 = VEC_alloc (tree, heap, number_of_vects);
if (SLP_TREE_RIGHT (slp_node))
- /* The defs are already vectorized. */
+ /* The defs are already vectorized. */
vect_get_slp_vect_defs (SLP_TREE_RIGHT (slp_node), vec_oprnds1);
else
/* Build vectors from scalar defs. */
[X, X + S, X + 2*S, X + 3*S]. */
static tree
-get_initial_def_for_induction (tree iv_phi)
+get_initial_def_for_induction (gimple iv_phi)
{
stmt_vec_info stmt_vinfo = vinfo_for_stmt (iv_phi);
loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_vinfo);
struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
- tree scalar_type = TREE_TYPE (PHI_RESULT_TREE (iv_phi));
+ tree scalar_type = TREE_TYPE (gimple_phi_result (iv_phi));
tree vectype;
int nunits;
edge pe = loop_preheader_edge (loop);
tree access_fn;
tree new_var;
tree new_name;
- tree init_stmt;
- tree induction_phi, induc_def, new_stmt, vec_def, vec_dest;
+ gimple init_stmt, induction_phi, new_stmt;
+ tree induc_def, vec_def, vec_dest;
tree init_expr, step_expr;
int vf = LOOP_VINFO_VECT_FACTOR (loop_vinfo);
int i;
tree expr;
stmt_vec_info phi_info = vinfo_for_stmt (iv_phi);
bool nested_in_vect_loop = false;
- tree stmts;
+ gimple_seq stmts = NULL;
imm_use_iterator imm_iter;
use_operand_p use_p;
- tree exit_phi;
+ gimple exit_phi;
edge latch_e;
tree loop_arg;
- block_stmt_iterator si;
- basic_block bb = bb_for_stmt (iv_phi);
+ gimple_stmt_iterator si;
+ basic_block bb = gimple_bb (iv_phi);
vectype = get_vectype_for_scalar_type (scalar_type);
gcc_assert (vectype);
gcc_assert (ncopies >= 1);
/* Find the first insertion point in the BB. */
- si = bsi_after_labels (bb);
+ si = gsi_after_labels (bb);
- if (INTEGRAL_TYPE_P (scalar_type))
+ if (INTEGRAL_TYPE_P (scalar_type) || POINTER_TYPE_P (scalar_type))
step_expr = build_int_cst (scalar_type, 0);
else
step_expr = build_real (scalar_type, dconst0);
}
else
iv_loop = loop;
- gcc_assert (iv_loop == (bb_for_stmt (iv_phi))->loop_father);
+ gcc_assert (iv_loop == (gimple_bb (iv_phi))->loop_father);
latch_e = loop_latch_edge (iv_loop);
loop_arg = PHI_ARG_DEF_FROM_EDGE (iv_phi, latch_e);
new_name = force_gimple_operand (init_expr, &stmts, false, new_var);
if (stmts)
{
- new_bb = bsi_insert_on_edge_immediate (pe, stmts);
+ new_bb = gsi_insert_seq_on_edge_immediate (pe, stmts);
gcc_assert (!new_bb);
}
t = tree_cons (NULL_TREE, init_expr, t);
for (i = 1; i < nunits; i++)
{
- tree tmp;
-
/* Create: new_name_i = new_name + step_expr */
- tmp = fold_build2 (PLUS_EXPR, scalar_type, new_name, step_expr);
- init_stmt = build_gimple_modify_stmt (new_var, tmp);
+ enum tree_code code = POINTER_TYPE_P (scalar_type)
+ ? POINTER_PLUS_EXPR : PLUS_EXPR;
+ init_stmt = gimple_build_assign_with_ops (code, new_var,
+ new_name, step_expr);
new_name = make_ssa_name (new_var, init_stmt);
- GIMPLE_STMT_OPERAND (init_stmt, 0) = new_name;
+ gimple_assign_set_lhs (init_stmt, new_name);
- new_bb = bsi_insert_on_edge_immediate (pe, init_stmt);
+ new_bb = gsi_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);
+ print_gimple_stmt (vect_dump, init_stmt, 0, TDF_SLIM);
}
t = tree_cons (NULL_TREE, new_name, t);
}
vec_dest = vect_get_new_vect_var (vectype, vect_simple_var, "vec_iv_");
add_referenced_var (vec_dest);
induction_phi = create_phi_node (vec_dest, iv_loop->header);
- set_stmt_info (get_stmt_ann (induction_phi),
- new_stmt_vec_info (induction_phi, loop_vinfo));
+ set_vinfo_for_stmt (induction_phi,
+ new_stmt_vec_info (induction_phi, loop_vinfo));
induc_def = PHI_RESULT (induction_phi);
/* Create the iv update inside the loop */
- new_stmt = build_gimple_modify_stmt (NULL_TREE,
- build2 (PLUS_EXPR, vectype,
- induc_def, vec_step));
+ new_stmt = gimple_build_assign_with_ops (PLUS_EXPR, vec_dest,
+ induc_def, vec_step);
vec_def = make_ssa_name (vec_dest, new_stmt);
- GIMPLE_STMT_OPERAND (new_stmt, 0) = vec_def;
- bsi_insert_before (&si, new_stmt, BSI_SAME_STMT);
- set_stmt_info (get_stmt_ann (new_stmt),
- new_stmt_vec_info (new_stmt, loop_vinfo));
+ gimple_assign_set_lhs (new_stmt, vec_def);
+ gsi_insert_before (&si, new_stmt, GSI_SAME_STMT);
+ set_vinfo_for_stmt (new_stmt, new_stmt_vec_info (new_stmt, loop_vinfo));
/* Set the arguments of the phi node: */
add_phi_arg (induction_phi, vec_init, pe);
prev_stmt_vinfo = vinfo_for_stmt (induction_phi);
for (i = 1; i < ncopies; i++)
{
- tree tmp;
-
/* vec_i = vec_prev + vec_step */
- tmp = build2 (PLUS_EXPR, vectype, vec_def, vec_step);
- new_stmt = build_gimple_modify_stmt (NULL_TREE, tmp);
+ new_stmt = gimple_build_assign_with_ops (PLUS_EXPR, vec_dest,
+ vec_def, vec_step);
vec_def = make_ssa_name (vec_dest, new_stmt);
- GIMPLE_STMT_OPERAND (new_stmt, 0) = vec_def;
- bsi_insert_before (&si, new_stmt, BSI_SAME_STMT);
- set_stmt_info (get_stmt_ann (new_stmt),
- new_stmt_vec_info (new_stmt, loop_vinfo));
+ gimple_assign_set_lhs (new_stmt, vec_def);
+
+ gsi_insert_before (&si, new_stmt, GSI_SAME_STMT);
+ set_vinfo_for_stmt (new_stmt,
+ new_stmt_vec_info (new_stmt, loop_vinfo));
STMT_VINFO_RELATED_STMT (prev_stmt_vinfo) = new_stmt;
prev_stmt_vinfo = vinfo_for_stmt (new_stmt);
}
exit_phi = NULL;
FOR_EACH_IMM_USE_FAST (use_p, imm_iter, loop_arg)
{
- if (!flow_bb_inside_loop_p (iv_loop, bb_for_stmt (USE_STMT (use_p))))
+ if (!flow_bb_inside_loop_p (iv_loop, gimple_bb (USE_STMT (use_p))))
{
exit_phi = USE_STMT (use_p);
break;
if (vect_print_dump_info (REPORT_DETAILS))
{
fprintf (vect_dump, "vector of inductions after inner-loop:");
- print_generic_expr (vect_dump, new_stmt, TDF_SLIM);
+ print_gimple_stmt (vect_dump, new_stmt, 0, TDF_SLIM);
}
}
}
if (vect_print_dump_info (REPORT_DETAILS))
{
- fprintf (vect_dump, "transform induction: created def-use cycle:");
- print_generic_expr (vect_dump, induction_phi, TDF_SLIM);
+ fprintf (vect_dump, "transform induction: created def-use cycle: ");
+ print_gimple_stmt (vect_dump, induction_phi, 0, TDF_SLIM);
fprintf (vect_dump, "\n");
- print_generic_expr (vect_dump, SSA_NAME_DEF_STMT (vec_def), TDF_SLIM);
+ print_gimple_stmt (vect_dump, SSA_NAME_DEF_STMT (vec_def), 0, TDF_SLIM);
}
STMT_VINFO_VEC_STMT (phi_info) = induction_phi;
needs to be introduced. */
static tree
-vect_get_vec_def_for_operand (tree op, tree stmt, tree *scalar_def)
+vect_get_vec_def_for_operand (tree op, gimple stmt, tree *scalar_def)
{
tree vec_oprnd;
- tree vec_stmt;
- tree def_stmt;
+ gimple vec_stmt;
+ gimple 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);
if (def_stmt)
{
fprintf (vect_dump, " def_stmt = ");
- print_generic_expr (vect_dump, def_stmt, TDF_SLIM);
+ print_gimple_stmt (vect_dump, def_stmt, 0, TDF_SLIM);
}
}
case vect_loop_def:
{
if (scalar_def)
- *scalar_def = def_stmt;
+ *scalar_def = NULL/* FIXME tuples: 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);
- if (TREE_CODE (vec_stmt) == PHI_NODE)
+ if (gimple_code (vec_stmt) == GIMPLE_PHI)
vec_oprnd = PHI_RESULT (vec_stmt);
+ else if (is_gimple_call (vec_stmt))
+ vec_oprnd = gimple_call_lhs (vec_stmt);
else
- vec_oprnd = GIMPLE_STMT_OPERAND (vec_stmt, 0);
+ vec_oprnd = gimple_assign_lhs (vec_stmt);
return vec_oprnd;
}
{
struct loop *loop;
- gcc_assert (TREE_CODE (def_stmt) == PHI_NODE);
- loop = (bb_for_stmt (def_stmt))->loop_father;
+ gcc_assert (gimple_code (def_stmt) == GIMPLE_PHI);
+ loop = (gimple_bb (def_stmt))->loop_father;
/* Get the def before the loop */
op = PHI_ARG_DEF_FROM_EDGE (def_stmt, loop_preheader_edge (loop));
/* Case 5: operand is defined by loop-header phi - induction. */
case vect_induction_def:
{
- gcc_assert (TREE_CODE (def_stmt) == PHI_NODE);
+ gcc_assert (gimple_code (def_stmt) == GIMPLE_PHI);
/* 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 && (TREE_CODE (vec_stmt) == PHI_NODE));
+ gcc_assert (vec_stmt && gimple_code (vec_stmt) == GIMPLE_PHI);
vec_oprnd = PHI_RESULT (vec_stmt);
return vec_oprnd;
}
static tree
vect_get_vec_def_for_stmt_copy (enum vect_def_type dt, tree vec_oprnd)
{
- tree vec_stmt_for_operand;
+ gimple vec_stmt_for_operand;
stmt_vec_info def_stmt_info;
/* Do nothing; can reuse same def. */
gcc_assert (def_stmt_info);
vec_stmt_for_operand = STMT_VINFO_RELATED_STMT (def_stmt_info);
gcc_assert (vec_stmt_for_operand);
- vec_oprnd = GIMPLE_STMT_OPERAND (vec_stmt_for_operand, 0);
+ vec_oprnd = gimple_get_lhs (vec_stmt_for_operand);
+ if (gimple_code (vec_stmt_for_operand) == GIMPLE_PHI)
+ vec_oprnd = PHI_RESULT (vec_stmt_for_operand);
+ else
+ vec_oprnd = gimple_get_lhs (vec_stmt_for_operand);
return vec_oprnd;
}
/* Get vectorized definitions for OP0 and OP1, or SLP_NODE if it is not NULL. */
static void
-vect_get_vec_defs (tree op0, tree op1, tree stmt, VEC(tree,heap) **vec_oprnds0,
- VEC(tree,heap) **vec_oprnds1, slp_tree slp_node)
+vect_get_vec_defs (tree op0, tree op1, gimple stmt,
+ VEC(tree,heap) **vec_oprnds0, VEC(tree,heap) **vec_oprnds1,
+ slp_tree slp_node)
{
if (slp_node)
vect_get_slp_defs (slp_node, vec_oprnds0, vec_oprnds1);
Insert a new stmt. */
static void
-vect_finish_stmt_generation (tree stmt, tree vec_stmt,
- block_stmt_iterator *bsi)
+vect_finish_stmt_generation (gimple stmt, gimple vec_stmt,
+ gimple_stmt_iterator *gsi)
{
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
- gcc_assert (stmt == bsi_stmt (*bsi));
- gcc_assert (TREE_CODE (stmt) != LABEL_EXPR);
+ gcc_assert (stmt == gsi_stmt (*gsi));
+ gcc_assert (gimple_code (stmt) != GIMPLE_LABEL);
- bsi_insert_before (bsi, vec_stmt, BSI_SAME_STMT);
+ gsi_insert_before (gsi, vec_stmt, GSI_SAME_STMT);
- set_stmt_info (get_stmt_ann (vec_stmt),
- new_stmt_vec_info (vec_stmt, loop_vinfo));
+ set_vinfo_for_stmt (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);
+ print_gimple_stmt (vect_dump, vec_stmt, 0, TDF_SLIM);
}
- /* Make sure bsi points to the stmt that is being vectorized. */
- gcc_assert (stmt == bsi_stmt (*bsi));
+ /* Make sure gsi points to the stmt that is being vectorized. */
+ gcc_assert (stmt == gsi_stmt (*gsi));
- SET_EXPR_LOCATION (vec_stmt, EXPR_LOCATION (stmt));
+ gimple_set_location (vec_stmt, gimple_location (stmt));
}
A cost model should help decide between these two schemes. */
static tree
-get_initial_def_for_reduction (tree stmt, tree init_val, tree *adjustment_def)
+get_initial_def_for_reduction (gimple stmt, tree init_val, tree *adjustment_def)
{
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 vectype = STMT_VINFO_VECTYPE (stmt_vinfo);
int nunits = TYPE_VECTOR_SUBPARTS (vectype);
- enum tree_code code = TREE_CODE (GIMPLE_STMT_OPERAND (stmt, 1));
+ enum tree_code code = gimple_assign_rhs_code (stmt);
tree type = TREE_TYPE (init_val);
tree vecdef;
tree def_for_init;
if (nested_in_vect_loop_p (loop, stmt))
nested_in_vect_loop = true;
else
- gcc_assert (loop == (bb_for_stmt (stmt))->loop_father);
+ gcc_assert (loop == (gimple_bb (stmt))->loop_father);
vecdef = vect_get_vec_def_for_operand (init_val, stmt, NULL);
VECT_DEF is a vector of partial results.
REDUC_CODE is the tree-code for the epilog reduction.
+ NCOPIES is > 1 in case the vectorization factor (VF) is bigger than the
+ number of elements that we can fit in a vectype (nunits). In this case
+ 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.
STMT is the scalar reduction stmt that is being vectorized.
REDUCTION_PHI is the phi-node that carries the reduction computation.
*/
static void
-vect_create_epilog_for_reduction (tree vect_def, tree stmt,
- enum tree_code reduc_code, tree reduction_phi)
+vect_create_epilog_for_reduction (tree vect_def, gimple stmt,
+ int ncopies,
+ enum tree_code reduc_code,
+ gimple reduction_phi)
{
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
+ stmt_vec_info prev_phi_info;
tree vectype;
enum machine_mode mode;
loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
basic_block exit_bb;
tree scalar_dest;
tree scalar_type;
- tree new_phi;
- block_stmt_iterator exit_bsi;
+ gimple new_phi = NULL, phi;
+ gimple_stmt_iterator exit_gsi;
tree vec_dest;
tree new_temp = NULL_TREE;
tree new_name;
- tree epilog_stmt = NULL_TREE;
- tree new_scalar_dest, exit_phi, new_dest;
+ gimple epilog_stmt = NULL;
+ tree new_scalar_dest, new_dest;
+ gimple exit_phi;
tree bitsize, bitpos, bytesize;
- enum tree_code code = TREE_CODE (GIMPLE_STMT_OPERAND (stmt, 1));
+ enum tree_code code = gimple_assign_rhs_code (stmt);
tree adjustment_def;
- tree vec_initial_def;
+ tree vec_initial_def, def;
tree orig_name;
imm_use_iterator imm_iter;
use_operand_p use_p;
bool extract_scalar_result = false;
tree reduction_op, expr;
- tree orig_stmt;
- tree use_stmt;
- tree operation = GIMPLE_STMT_OPERAND (stmt, 1);
+ gimple orig_stmt;
+ gimple use_stmt;
bool nested_in_vect_loop = false;
- int op_type;
- VEC(tree,heap) *phis = NULL;
- int i;
+ VEC(gimple,heap) *phis = NULL;
+ enum vect_def_type dt = vect_unknown_def_type;
+ int j, i;
if (nested_in_vect_loop_p (loop, stmt))
{
nested_in_vect_loop = true;
}
- op_type = TREE_OPERAND_LENGTH (operation);
- reduction_op = TREE_OPERAND (operation, op_type-1);
+ switch (get_gimple_rhs_class (gimple_assign_rhs_code (stmt)))
+ {
+ case GIMPLE_SINGLE_RHS:
+ gcc_assert (TREE_OPERAND_LENGTH (gimple_assign_rhs1 (stmt)) == ternary_op);
+ reduction_op = TREE_OPERAND (gimple_assign_rhs1 (stmt), 2);
+ break;
+ case GIMPLE_UNARY_RHS:
+ reduction_op = gimple_assign_rhs1 (stmt);
+ break;
+ case GIMPLE_BINARY_RHS:
+ reduction_op = gimple_assign_rhs2 (stmt);
+ break;
+ default:
+ gcc_unreachable ();
+ }
+
vectype = get_vectype_for_scalar_type (TREE_TYPE (reduction_op));
gcc_assert (vectype);
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,
&adjustment_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))
+ phi = reduction_phi;
+ def = vect_def;
+ for (j = 0; j < ncopies; j++)
{
- 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);
- }
+ /* 1.1 set the loop-entry arg of the reduction-phi: */
+ add_phi_arg (phi, vec_initial_def, loop_preheader_edge (loop));
+ /* 1.2 set the loop-latch arg for the reduction-phi: */
+ if (j > 0)
+ def = vect_get_vec_def_for_stmt_copy (dt, def);
+ add_phi_arg (phi, def, loop_latch_edge (loop));
+
+ if (vect_print_dump_info (REPORT_DETAILS))
+ {
+ fprintf (vect_dump, "transform reduction: created def-use cycle: ");
+ print_gimple_stmt (vect_dump, phi, 0, TDF_SLIM);
+ fprintf (vect_dump, "\n");
+ print_gimple_stmt (vect_dump, SSA_NAME_DEF_STMT (def), 0, TDF_SLIM);
+ }
+
+ phi = STMT_VINFO_RELATED_STMT (vinfo_for_stmt (phi));
+ }
/*** 2. Create epilog code
The reduction epilog code operates across the elements of the vector
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).
+ (step 3 is optional, and steps 1 and 2 may be combined).
Lastly, the uses of s_out0 are replaced by s_out4.
***/
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_after_labels (exit_bb);
+ def = vect_def;
+ prev_phi_info = NULL;
+ for (j = 0; j < ncopies; j++)
+ {
+ phi = create_phi_node (SSA_NAME_VAR (vect_def), exit_bb);
+ set_vinfo_for_stmt (phi, new_stmt_vec_info (phi, loop_vinfo));
+ if (j == 0)
+ new_phi = phi;
+ else
+ {
+ def = vect_get_vec_def_for_stmt_copy (dt, def);
+ STMT_VINFO_RELATED_STMT (prev_phi_info) = phi;
+ }
+ SET_PHI_ARG_DEF (phi, single_exit (loop)->dest_idx, def);
+ prev_phi_info = vinfo_for_stmt (phi);
+ }
+ exit_gsi = gsi_after_labels (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
gcc_assert (STMT_VINFO_IN_PATTERN_P (stmt_vinfo));
gcc_assert (STMT_VINFO_RELATED_STMT (stmt_vinfo) == stmt);
}
- code = TREE_CODE (GIMPLE_STMT_OPERAND (orig_stmt, 1));
- scalar_dest = GIMPLE_STMT_OPERAND (orig_stmt, 0);
+ code = gimple_assign_rhs_code (orig_stmt);
+ scalar_dest = gimple_assign_lhs (orig_stmt);
scalar_type = TREE_TYPE (scalar_dest);
new_scalar_dest = vect_create_destination_var (scalar_dest, NULL);
bitsize = TYPE_SIZE (scalar_type);
if (nested_in_vect_loop)
goto vect_finalize_reduction;
+ /* FORNOW */
+ gcc_assert (ncopies == 1);
+
/* 2.3 Create the reduction code, using one of the three schemes described
above. */
vec_dest = vect_create_destination_var (scalar_dest, vectype);
tmp = build1 (reduc_code, vectype, PHI_RESULT (new_phi));
- epilog_stmt = build_gimple_modify_stmt (vec_dest, tmp);
+ epilog_stmt = gimple_build_assign (vec_dest, tmp);
new_temp = make_ssa_name (vec_dest, epilog_stmt);
- GIMPLE_STMT_OPERAND (epilog_stmt, 0) = new_temp;
- bsi_insert_before (&exit_bsi, epilog_stmt, BSI_SAME_STMT);
+ gimple_assign_set_lhs (epilog_stmt, new_temp);
+ gsi_insert_before (&exit_gsi, epilog_stmt, GSI_SAME_STMT);
extract_scalar_result = true;
}
bit_offset /= 2)
{
tree bitpos = size_int (bit_offset);
- tree tmp = build2 (shift_code, vectype, new_temp, bitpos);
- epilog_stmt = build_gimple_modify_stmt (vec_dest, tmp);
+ epilog_stmt = gimple_build_assign_with_ops (shift_code, vec_dest,
+ new_temp, bitpos);
new_name = make_ssa_name (vec_dest, epilog_stmt);
- GIMPLE_STMT_OPERAND (epilog_stmt, 0) = new_name;
- bsi_insert_before (&exit_bsi, epilog_stmt, BSI_SAME_STMT);
+ gimple_assign_set_lhs (epilog_stmt, new_name);
+ gsi_insert_before (&exit_gsi, epilog_stmt, GSI_SAME_STMT);
- tmp = build2 (code, vectype, new_name, new_temp);
- epilog_stmt = build_gimple_modify_stmt (vec_dest, tmp);
+ epilog_stmt = gimple_build_assign_with_ops (code, vec_dest,
+ new_name, new_temp);
new_temp = make_ssa_name (vec_dest, epilog_stmt);
- GIMPLE_STMT_OPERAND (epilog_stmt, 0) = new_temp;
- bsi_insert_before (&exit_bsi, epilog_stmt, BSI_SAME_STMT);
+ gimple_assign_set_lhs (epilog_stmt, new_temp);
+ gsi_insert_before (&exit_gsi, epilog_stmt, GSI_SAME_STMT);
}
extract_scalar_result = true;
vec_size_in_bits = tree_low_cst (TYPE_SIZE (vectype), 1);
rhs = build3 (BIT_FIELD_REF, scalar_type, vec_temp, bitsize,
bitsize_zero_node);
- epilog_stmt = build_gimple_modify_stmt (new_scalar_dest, rhs);
+ epilog_stmt = gimple_build_assign (new_scalar_dest, rhs);
new_temp = make_ssa_name (new_scalar_dest, epilog_stmt);
- GIMPLE_STMT_OPERAND (epilog_stmt, 0) = new_temp;
- bsi_insert_before (&exit_bsi, epilog_stmt, BSI_SAME_STMT);
+ gimple_assign_set_lhs (epilog_stmt, new_temp);
+ gsi_insert_before (&exit_gsi, epilog_stmt, GSI_SAME_STMT);
for (bit_offset = element_bitsize;
bit_offset < vec_size_in_bits;
bit_offset += element_bitsize)
{
- tree tmp;
tree bitpos = bitsize_int (bit_offset);
tree rhs = build3 (BIT_FIELD_REF, scalar_type, vec_temp, bitsize,
bitpos);
- epilog_stmt = build_gimple_modify_stmt (new_scalar_dest, rhs);
+ epilog_stmt = gimple_build_assign (new_scalar_dest, rhs);
new_name = make_ssa_name (new_scalar_dest, epilog_stmt);
- GIMPLE_STMT_OPERAND (epilog_stmt, 0) = new_name;
- bsi_insert_before (&exit_bsi, epilog_stmt, BSI_SAME_STMT);
+ gimple_assign_set_lhs (epilog_stmt, new_name);
+ gsi_insert_before (&exit_gsi, epilog_stmt, GSI_SAME_STMT);
- tmp = build2 (code, scalar_type, new_name, new_temp);
- epilog_stmt = build_gimple_modify_stmt (new_scalar_dest, tmp);
+ epilog_stmt = gimple_build_assign_with_ops (code,
+ new_scalar_dest,
+ new_name, new_temp);
new_temp = make_ssa_name (new_scalar_dest, epilog_stmt);
- GIMPLE_STMT_OPERAND (epilog_stmt, 0) = new_temp;
- bsi_insert_before (&exit_bsi, epilog_stmt, BSI_SAME_STMT);
+ gimple_assign_set_lhs (epilog_stmt, new_temp);
+ gsi_insert_before (&exit_gsi, epilog_stmt, GSI_SAME_STMT);
}
extract_scalar_result = false;
bitpos = bitsize_zero_node;
rhs = build3 (BIT_FIELD_REF, scalar_type, new_temp, bitsize, bitpos);
- epilog_stmt = build_gimple_modify_stmt (new_scalar_dest, rhs);
+ epilog_stmt = gimple_build_assign (new_scalar_dest, rhs);
new_temp = make_ssa_name (new_scalar_dest, epilog_stmt);
- GIMPLE_STMT_OPERAND (epilog_stmt, 0) = new_temp;
- bsi_insert_before (&exit_bsi, epilog_stmt, BSI_SAME_STMT);
+ gimple_assign_set_lhs (epilog_stmt, new_temp);
+ gsi_insert_before (&exit_gsi, epilog_stmt, GSI_SAME_STMT);
}
vect_finalize_reduction:
expr = build2 (code, scalar_type, new_temp, adjustment_def);
new_dest = vect_create_destination_var (scalar_dest, scalar_type);
}
- epilog_stmt = build_gimple_modify_stmt (new_dest, expr);
+ epilog_stmt = gimple_build_assign (new_dest, expr);
new_temp = make_ssa_name (new_dest, epilog_stmt);
- GIMPLE_STMT_OPERAND (epilog_stmt, 0) = new_temp;
- bsi_insert_before (&exit_bsi, epilog_stmt, BSI_SAME_STMT);
+ gimple_assign_set_lhs (epilog_stmt, new_temp);
+ SSA_NAME_DEF_STMT (new_temp) = epilog_stmt;
+ gsi_insert_before (&exit_gsi, epilog_stmt, GSI_SAME_STMT);
}
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). */
- phis = VEC_alloc (tree, heap, 10);
+ phis = VEC_alloc (gimple, heap, 10);
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))))
+ if (!flow_bb_inside_loop_p (loop, gimple_bb (USE_STMT (use_p))))
{
exit_phi = USE_STMT (use_p);
- VEC_quick_push (tree, phis, exit_phi);
+ VEC_quick_push (gimple, phis, exit_phi);
}
}
/* We expect to have found an exit_phi because of loop-closed-ssa form. */
- gcc_assert (!VEC_empty (tree, phis));
+ gcc_assert (!VEC_empty (gimple, phis));
- for (i = 0; VEC_iterate (tree, phis, i, exit_phi); i++)
+ for (i = 0; VEC_iterate (gimple, phis, i, exit_phi); i++)
{
if (nested_in_vect_loop)
{
stmt_vec_info stmt_vinfo = vinfo_for_stmt (exit_phi);
- /* FORNOW. Currently not supporting the case that an inner-loop reduction
- is not used in the outer-loop (but only outside the outer-loop). */
+ /* FORNOW. Currently not supporting the case that an inner-loop
+ reduction is not used in the outer-loop (but only outside the
+ outer-loop). */
gcc_assert (STMT_VINFO_RELEVANT_P (stmt_vinfo)
&& !STMT_VINFO_LIVE_P (stmt_vinfo));
- epilog_stmt = adjustment_def ? epilog_stmt : new_phi;
+ epilog_stmt = adjustment_def ? epilog_stmt : new_phi;
STMT_VINFO_VEC_STMT (stmt_vinfo) = epilog_stmt;
- set_stmt_info (get_stmt_ann (epilog_stmt),
- new_stmt_vec_info (epilog_stmt, loop_vinfo));
+ set_vinfo_for_stmt (epilog_stmt,
+ new_stmt_vec_info (epilog_stmt, loop_vinfo));
+ if (adjustment_def)
+ STMT_VINFO_RELATED_STMT (vinfo_for_stmt (epilog_stmt)) =
+ STMT_VINFO_RELATED_STMT (vinfo_for_stmt (new_phi));
continue;
}
FOR_EACH_IMM_USE_ON_STMT (use_p, imm_iter)
SET_USE (use_p, new_temp);
}
- VEC_free (tree, heap, phis);
+ VEC_free (gimple, heap, phis);
}
does *NOT* necessarily hold for reduction patterns. */
bool
-vectorizable_reduction (tree stmt, block_stmt_iterator *bsi, tree *vec_stmt)
+vectorizable_reduction (gimple stmt, gimple_stmt_iterator *gsi,
+ gimple *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;
+ tree def;
+ gimple def_stmt;
enum vect_def_type dt;
- tree new_phi;
+ gimple new_phi = NULL;
tree scalar_type;
bool is_simple_use;
- tree orig_stmt;
+ gimple 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;
+ int epilog_copies;
+ stmt_vec_info prev_stmt_info, prev_phi_info;
+ gimple first_phi = NULL;
+ bool single_defuse_cycle = false;
tree reduc_def;
- tree new_stmt = NULL_TREE;
+ gimple new_stmt = NULL;
int j;
+ tree ops[3];
if (nested_in_vect_loop_p (loop, stmt))
- {
- loop = loop->inner;
- /* FORNOW. This restriction should be relaxed. */
- if (ncopies > 1)
- {
- if (vect_print_dump_info (REPORT_DETAILS))
- fprintf (vect_dump, "multiple types in nested loop.");
- return false;
- }
- }
+ loop = loop->inner;
gcc_assert (ncopies >= 1);
inside the loop body. The last operand is the reduction variable,
which is defined by the loop-header-phi. */
- gcc_assert (TREE_CODE (stmt) == GIMPLE_MODIFY_STMT);
+ gcc_assert (is_gimple_assign (stmt));
- operation = GIMPLE_STMT_OPERAND (stmt, 1);
- code = TREE_CODE (operation);
- op_type = TREE_OPERAND_LENGTH (operation);
- if (op_type != binary_op && op_type != ternary_op)
- return false;
- scalar_dest = GIMPLE_STMT_OPERAND (stmt, 0);
+ /* Flatten RHS */
+ switch (get_gimple_rhs_class (gimple_assign_rhs_code (stmt)))
+ {
+ case GIMPLE_SINGLE_RHS:
+ op_type = TREE_OPERAND_LENGTH (gimple_assign_rhs1 (stmt));
+ if (op_type == ternary_op)
+ {
+ tree rhs = gimple_assign_rhs1 (stmt);
+ ops[0] = TREE_OPERAND (rhs, 0);
+ ops[1] = TREE_OPERAND (rhs, 1);
+ ops[2] = TREE_OPERAND (rhs, 2);
+ code = TREE_CODE (rhs);
+ }
+ else
+ return false;
+ break;
+
+ case GIMPLE_BINARY_RHS:
+ code = gimple_assign_rhs_code (stmt);
+ op_type = TREE_CODE_LENGTH (code);
+ gcc_assert (op_type == binary_op);
+ ops[0] = gimple_assign_rhs1 (stmt);
+ ops[1] = gimple_assign_rhs2 (stmt);
+ break;
+
+ case GIMPLE_UNARY_RHS:
+ return false;
+
+ default:
+ gcc_unreachable ();
+ }
+
+ scalar_dest = gimple_assign_lhs (stmt);
scalar_type = TREE_TYPE (scalar_dest);
if (!POINTER_TYPE_P (scalar_type) && !INTEGRAL_TYPE_P (scalar_type)
&& !SCALAR_FLOAT_TYPE_P (scalar_type))
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);
+ is_simple_use = vect_is_simple_use (ops[i], loop_vinfo, &def_stmt,
+ &def, &dt);
gcc_assert (is_simple_use);
if (dt != vect_loop_def
&& dt != vect_invariant_def
return false;
}
- op = TREE_OPERAND (operation, i);
- is_simple_use = vect_is_simple_use (op, loop_vinfo, &def_stmt, &def, &dt);
+ is_simple_use = vect_is_simple_use (ops[i], 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);
+ gcc_assert (gimple_code (def_stmt) == GIMPLE_PHI);
if (orig_stmt)
gcc_assert (orig_stmt == vect_is_simple_reduction (loop_vinfo, def_stmt));
else
{
/* 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 (GIMPLE_STMT_OPERAND (orig_stmt, 1));
+ orig_code = gimple_assign_rhs_code (orig_stmt);
vectype = get_vectype_for_scalar_type (TREE_TYPE (def));
if (!vectype)
{
/* 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. */
+ /* If the reduction is used in an outer loop we need to generate
+ VF intermediate results, like so (e.g. for ncopies=2):
+ r0 = phi (init, r0)
+ r1 = phi (init, r1)
+ r0 = x0 + r0;
+ r1 = x1 + r1;
+ (i.e. we generate VF results in 2 registers).
+ In this case we have a separate def-use cycle for each copy, and therefore
+ for each copy we get the vector def for the reduction variable from the
+ respective phi node created for this copy.
+
+ Otherwise (the reduction is unused in the loop nest), we can combine
+ together intermediate results, like so (e.g. for ncopies=2):
+ r = phi (init, r)
+ r = x0 + r;
+ r = x1 + r;
+ (i.e. we generate VF/2 results in a single register).
+ In this case for each copy we get the vector def for the reduction variable
+ from the vectorized reduction operation generated in the previous iteration.
+ */
+
+ if (STMT_VINFO_RELEVANT (stmt_info) == vect_unused_in_loop)
+ {
+ single_defuse_cycle = true;
+ epilog_copies = 1;
+ }
+ else
+ epilog_copies = ncopies;
+
prev_stmt_info = NULL;
+ prev_phi_info = NULL;
for (j = 0; j < ncopies; j++)
{
+ if (j == 0 || !single_defuse_cycle)
+ {
+ /* Create the reduction-phi that defines the reduction-operand. */
+ new_phi = create_phi_node (vec_dest, loop->header);
+ set_vinfo_for_stmt (new_phi, new_stmt_vec_info (new_phi, loop_vinfo));
+ }
+
/* Handle uses. */
if (j == 0)
{
- op = TREE_OPERAND (operation, 0);
- loop_vec_def0 = vect_get_vec_def_for_operand (op, stmt, NULL);
+ loop_vec_def0 = vect_get_vec_def_for_operand (ops[0], stmt, NULL);
if (op_type == ternary_op)
{
- op = TREE_OPERAND (operation, 1);
- loop_vec_def1 = vect_get_vec_def_for_operand (op, stmt, NULL);
+ loop_vec_def1 = vect_get_vec_def_for_operand (ops[1], stmt, NULL);
}
/* Get the vector def for the reduction variable from the phi node */
reduc_def = PHI_RESULT (new_phi);
+ first_phi = new_phi;
}
else
{
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 = GIMPLE_STMT_OPERAND (new_stmt ,0);
+ if (single_defuse_cycle)
+ reduc_def = gimple_assign_lhs (new_stmt);
+ else
+ reduc_def = PHI_RESULT (new_phi);
+
+ STMT_VINFO_RELATED_STMT (prev_phi_info) = new_phi;
}
/* Arguments are ready. create the new vector stmt. */
else
expr = build3 (code, vectype, loop_vec_def0, loop_vec_def1,
reduc_def);
- new_stmt = build_gimple_modify_stmt (vec_dest, expr);
+ new_stmt = gimple_build_assign (vec_dest, expr);
new_temp = make_ssa_name (vec_dest, new_stmt);
- GIMPLE_STMT_OPERAND (new_stmt, 0) = new_temp;
- vect_finish_stmt_generation (stmt, new_stmt, bsi);
+ gimple_assign_set_lhs (new_stmt, new_temp);
+ vect_finish_stmt_generation (stmt, new_stmt, gsi);
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);
+ prev_phi_info = vinfo_for_stmt (new_phi);
}
- /* Finalize the reduction-phi (set it's arguments) and create the
+ /* Finalize the reduction-phi (set its arguments) and create the
epilog reduction code. */
- vect_create_epilog_for_reduction (new_temp, stmt, epilog_reduc_code, new_phi);
+ if (!single_defuse_cycle)
+ new_temp = gimple_assign_lhs (*vec_stmt);
+ vect_create_epilog_for_reduction (new_temp, stmt, epilog_copies,
+ epilog_reduc_code, first_phi);
return true;
}
of the function, or NULL_TREE if the function cannot be vectorized. */
tree
-vectorizable_function (tree call, tree vectype_out, tree vectype_in)
+vectorizable_function (gimple call, tree vectype_out, tree vectype_in)
{
- tree fndecl = get_callee_fndecl (call);
+ tree fndecl = gimple_call_fndecl (call);
enum built_in_function code;
/* We only handle functions that do not read or clobber memory -- i.e.
const or novops ones. */
- if (!(call_expr_flags (call) & (ECF_CONST | ECF_NOVOPS)))
+ if (!(gimple_call_flags (call) & (ECF_CONST | ECF_NOVOPS)))
return NULL_TREE;
if (!fndecl
Return FALSE if not a vectorizable STMT, TRUE otherwise. */
bool
-vectorizable_call (tree stmt, block_stmt_iterator *bsi, tree *vec_stmt)
+vectorizable_call (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt)
{
tree vec_dest;
tree scalar_dest;
- tree operation;
tree op, type;
tree vec_oprnd0 = NULL_TREE, vec_oprnd1 = NULL_TREE;
stmt_vec_info stmt_info = vinfo_for_stmt (stmt), prev_stmt_info;
int nunits_in;
int nunits_out;
loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
- struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
- tree fndecl, rhs, new_temp, def, def_stmt, rhs_type, lhs_type;
+ tree fndecl, new_temp, def, rhs_type, lhs_type;
+ gimple def_stmt;
enum vect_def_type dt[2] = {vect_unknown_def_type, vect_unknown_def_type};
- tree new_stmt;
- int ncopies, j, nargs;
- call_expr_arg_iterator iter;
- tree vargs;
+ gimple new_stmt;
+ int ncopies, j;
+ VEC(tree, heap) *vargs = NULL;
enum { NARROW, NONE, WIDEN } modifier;
+ size_t i, nargs;
if (!STMT_VINFO_RELEVANT_P (stmt_info))
return false;
return false;
/* Is STMT a vectorizable call? */
- if (TREE_CODE (stmt) != GIMPLE_MODIFY_STMT)
- return false;
-
- if (TREE_CODE (GIMPLE_STMT_OPERAND (stmt, 0)) != SSA_NAME)
+ if (!is_gimple_call (stmt))
return false;
- operation = GIMPLE_STMT_OPERAND (stmt, 1);
- if (TREE_CODE (operation) != CALL_EXPR)
+ if (TREE_CODE (gimple_call_lhs (stmt)) != SSA_NAME)
return false;
/* Process function arguments. */
rhs_type = NULL_TREE;
- nargs = 0;
- FOR_EACH_CALL_EXPR_ARG (op, iter, operation)
+ nargs = gimple_call_num_args (stmt);
+
+ /* Bail out if the function has more than two arguments, we
+ do not have interesting builtin functions to vectorize with
+ more than two arguments. No arguments is also not good. */
+ if (nargs == 0 || nargs > 2)
+ return false;
+
+ for (i = 0; i < nargs; i++)
{
- /* Bail out if the function has more than two arguments, we
- do not have interesting builtin functions to vectorize with
- more than two arguments. */
- if (nargs >= 2)
- return false;
+ op = gimple_call_arg (stmt, i);
/* We can only handle calls with arguments of the same type. */
if (rhs_type
}
rhs_type = TREE_TYPE (op);
- if (!vect_is_simple_use (op, loop_vinfo, &def_stmt, &def, &dt[nargs]))
+ if (!vect_is_simple_use (op, loop_vinfo, &def_stmt, &def, &dt[i]))
{
if (vect_print_dump_info (REPORT_DETAILS))
fprintf (vect_dump, "use not simple.");
return false;
}
-
- ++nargs;
}
- /* No arguments is also not good. */
- if (nargs == 0)
- return false;
-
vectype_in = get_vectype_for_scalar_type (rhs_type);
if (!vectype_in)
return false;
nunits_in = TYPE_VECTOR_SUBPARTS (vectype_in);
- lhs_type = TREE_TYPE (GIMPLE_STMT_OPERAND (stmt, 0));
+ lhs_type = TREE_TYPE (gimple_call_lhs (stmt));
vectype_out = get_vectype_for_scalar_type (lhs_type);
if (!vectype_out)
return false;
is available. TODO -- in some cases, it might be profitable to
insert the calls for pieces of the vector, in order to be able
to vectorize other operations in the loop. */
- fndecl = vectorizable_function (operation, vectype_out, vectype_in);
+ fndecl = vectorizable_function (stmt, vectype_out, vectype_in);
if (fndecl == NULL_TREE)
{
if (vect_print_dump_info (REPORT_DETAILS))
needs to be generated. */
gcc_assert (ncopies >= 1);
- /* FORNOW. This restriction should be relaxed. */
- if (nested_in_vect_loop_p (loop, stmt) && ncopies > 1)
- {
- if (vect_print_dump_info (REPORT_DETAILS))
- fprintf (vect_dump, "multiple types in nested loop.");
- return false;
- }
-
if (!vec_stmt) /* transformation not required. */
{
STMT_VINFO_TYPE (stmt_info) = call_vec_info_type;
if (vect_print_dump_info (REPORT_DETAILS))
fprintf (vect_dump, "transform operation.");
- /* FORNOW. This restriction should be relaxed. */
- if (nested_in_vect_loop_p (loop, stmt) && ncopies > 1)
- {
- if (vect_print_dump_info (REPORT_DETAILS))
- fprintf (vect_dump, "multiple types in nested loop.");
- return false;
- }
-
/* Handle def. */
- scalar_dest = GIMPLE_STMT_OPERAND (stmt, 0);
+ scalar_dest = gimple_call_lhs (stmt);
vec_dest = vect_create_destination_var (scalar_dest, vectype_out);
prev_stmt_info = NULL;
for (j = 0; j < ncopies; ++j)
{
/* Build argument list for the vectorized call. */
- /* FIXME: Rewrite this so that it doesn't
- construct a temporary list. */
- vargs = NULL_TREE;
- nargs = 0;
- FOR_EACH_CALL_EXPR_ARG (op, iter, operation)
+ if (j == 0)
+ vargs = VEC_alloc (tree, heap, nargs);
+ else
+ VEC_truncate (tree, vargs, 0);
+
+ for (i = 0; i < nargs; i++)
{
+ op = gimple_call_arg (stmt, i);
if (j == 0)
vec_oprnd0
= vect_get_vec_def_for_operand (op, stmt, NULL);
vec_oprnd0
= vect_get_vec_def_for_stmt_copy (dt[nargs], vec_oprnd0);
- vargs = tree_cons (NULL_TREE, vec_oprnd0, vargs);
-
- ++nargs;
+ VEC_quick_push (tree, vargs, vec_oprnd0);
}
- vargs = nreverse (vargs);
- rhs = build_function_call_expr (fndecl, vargs);
- new_stmt = build_gimple_modify_stmt (vec_dest, rhs);
+ new_stmt = gimple_build_call_vec (fndecl, vargs);
new_temp = make_ssa_name (vec_dest, new_stmt);
- GIMPLE_STMT_OPERAND (new_stmt, 0) = new_temp;
+ gimple_call_set_lhs (new_stmt, new_temp);
- vect_finish_stmt_generation (stmt, new_stmt, bsi);
+ vect_finish_stmt_generation (stmt, new_stmt, gsi);
if (j == 0)
STMT_VINFO_VEC_STMT (stmt_info) = *vec_stmt = new_stmt;
for (j = 0; j < ncopies; ++j)
{
/* Build argument list for the vectorized call. */
- /* FIXME: Rewrite this so that it doesn't
- construct a temporary list. */
- vargs = NULL_TREE;
- nargs = 0;
- FOR_EACH_CALL_EXPR_ARG (op, iter, operation)
+ if (j == 0)
+ vargs = VEC_alloc (tree, heap, nargs * 2);
+ else
+ VEC_truncate (tree, vargs, 0);
+
+ for (i = 0; i < nargs; i++)
{
+ op = gimple_call_arg (stmt, i);
if (j == 0)
{
vec_oprnd0
= vect_get_vec_def_for_stmt_copy (dt[nargs], vec_oprnd0);
}
- vargs = tree_cons (NULL_TREE, vec_oprnd0, vargs);
- vargs = tree_cons (NULL_TREE, vec_oprnd1, vargs);
-
- ++nargs;
+ VEC_quick_push (tree, vargs, vec_oprnd0);
+ VEC_quick_push (tree, vargs, vec_oprnd1);
}
- vargs = nreverse (vargs);
- rhs = build_function_call_expr (fndecl, vargs);
- new_stmt = build_gimple_modify_stmt (vec_dest, rhs);
+ new_stmt = gimple_build_call_vec (fndecl, vargs);
new_temp = make_ssa_name (vec_dest, new_stmt);
- GIMPLE_STMT_OPERAND (new_stmt, 0) = new_temp;
+ gimple_call_set_lhs (new_stmt, new_temp);
- vect_finish_stmt_generation (stmt, new_stmt, bsi);
+ vect_finish_stmt_generation (stmt, new_stmt, gsi);
if (j == 0)
STMT_VINFO_VEC_STMT (stmt_info) = new_stmt;
return false;
}
+ VEC_free (tree, heap, vargs);
+
/* The call in STMT might prevent it from being removed in dce.
We however cannot remove it here, due to the way the ssa name
it defines is mapped to the new definition. So just replace
rhs of the statement with something harmless. */
+
type = TREE_TYPE (scalar_dest);
- GIMPLE_STMT_OPERAND (stmt, 1) = fold_convert (type, integer_zero_node);
- update_stmt (stmt);
+ new_stmt = gimple_build_assign (gimple_call_lhs (stmt),
+ fold_convert (type, integer_zero_node));
+ set_vinfo_for_stmt (new_stmt, stmt_info);
+ set_vinfo_for_stmt (stmt, NULL);
+ STMT_VINFO_STMT (stmt_info) = new_stmt;
+ gsi_replace (gsi, new_stmt, false);
+ SSA_NAME_DEF_STMT (gimple_assign_lhs (new_stmt)) = new_stmt;
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
+ variable are CODE, 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,
+static gimple
+vect_gen_widened_results_half (enum tree_code code,
+ tree decl,
tree vec_oprnd0, tree vec_oprnd1, int op_type,
- tree vec_dest, block_stmt_iterator *bsi,
- tree stmt)
+ tree vec_dest, gimple_stmt_iterator *gsi,
+ gimple stmt)
{
- tree expr;
- tree new_stmt;
+ gimple new_stmt;
tree new_temp;
tree sym;
ssa_op_iter iter;
{
/* Target specific support */
if (op_type == binary_op)
- expr = build_call_expr (decl, 2, vec_oprnd0, vec_oprnd1);
+ new_stmt = gimple_build_call (decl, 2, vec_oprnd0, vec_oprnd1);
else
- expr = build_call_expr (decl, 1, vec_oprnd0);
+ new_stmt = gimple_build_call (decl, 1, vec_oprnd0);
+ new_temp = make_ssa_name (vec_dest, new_stmt);
+ gimple_call_set_lhs (new_stmt, new_temp);
}
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);
+ if (op_type != binary_op)
+ vec_oprnd1 = NULL;
+ new_stmt = gimple_build_assign_with_ops (code, vec_dest, vec_oprnd0,
+ vec_oprnd1);
+ new_temp = make_ssa_name (vec_dest, new_stmt);
+ gimple_assign_set_lhs (new_stmt, new_temp);
}
- new_stmt = build_gimple_modify_stmt (vec_dest, expr);
- new_temp = make_ssa_name (vec_dest, new_stmt);
- GIMPLE_STMT_OPERAND (new_stmt, 0) = new_temp;
- vect_finish_stmt_generation (stmt, new_stmt, bsi);
+ vect_finish_stmt_generation (stmt, new_stmt, gsi);
if (code == CALL_EXPR)
{
Return FALSE if not a vectorizable STMT, TRUE otherwise. */
bool
-vectorizable_conversion (tree stmt, block_stmt_iterator *bsi,
- tree *vec_stmt, slp_tree slp_node)
+vectorizable_conversion (gimple stmt, gimple_stmt_iterator *gsi,
+ gimple *vec_stmt, slp_tree slp_node)
{
tree vec_dest;
tree scalar_dest;
- tree operation;
tree op0;
tree vec_oprnd0 = NULL_TREE, vec_oprnd1 = NULL_TREE;
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);
enum tree_code code, code1 = ERROR_MARK, code2 = ERROR_MARK;
tree decl1 = NULL_TREE, decl2 = NULL_TREE;
tree new_temp;
- tree def, def_stmt;
+ tree def;
+ gimple def_stmt;
enum vect_def_type dt[2] = {vect_unknown_def_type, vect_unknown_def_type};
- tree new_stmt = NULL_TREE;
+ gimple new_stmt = NULL;
stmt_vec_info prev_stmt_info;
int nunits_in;
int nunits_out;
int i;
VEC(tree,heap) *vec_oprnds0 = NULL;
tree vop0;
+ tree integral_type;
+ VEC(tree,heap) *dummy = NULL;
+ int dummy_int;
/* Is STMT a vectorizable conversion? */
if (STMT_VINFO_DEF_TYPE (stmt_info) != vect_loop_def)
return false;
- if (TREE_CODE (stmt) != GIMPLE_MODIFY_STMT)
+ if (!is_gimple_assign (stmt))
return false;
- if (TREE_CODE (GIMPLE_STMT_OPERAND (stmt, 0)) != SSA_NAME)
+ if (TREE_CODE (gimple_assign_lhs (stmt)) != SSA_NAME)
return false;
- operation = GIMPLE_STMT_OPERAND (stmt, 1);
- code = TREE_CODE (operation);
+ code = gimple_assign_rhs_code (stmt);
if (code != FIX_TRUNC_EXPR && code != FLOAT_EXPR)
return false;
/* Check types of lhs and rhs. */
- op0 = TREE_OPERAND (operation, 0);
+ op0 = gimple_assign_rhs1 (stmt);
rhs_type = TREE_TYPE (op0);
vectype_in = get_vectype_for_scalar_type (rhs_type);
if (!vectype_in)
return false;
nunits_in = TYPE_VECTOR_SUBPARTS (vectype_in);
- scalar_dest = GIMPLE_STMT_OPERAND (stmt, 0);
+ scalar_dest = gimple_assign_lhs (stmt);
lhs_type = TREE_TYPE (scalar_dest);
vectype_out = get_vectype_for_scalar_type (lhs_type);
if (!vectype_out)
|| (!INTEGRAL_TYPE_P (rhs_type) && !INTEGRAL_TYPE_P (lhs_type)))
return false;
+ integral_type = INTEGRAL_TYPE_P (rhs_type) ? vectype_in : vectype_out;
+
if (modifier == NARROW)
ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits_out;
else
needs to be generated. */
gcc_assert (ncopies >= 1);
- /* FORNOW. This restriction should be relaxed. */
- if (nested_in_vect_loop_p (loop, stmt) && ncopies > 1)
- {
- if (vect_print_dump_info (REPORT_DETAILS))
- fprintf (vect_dump, "multiple types in nested loop.");
- return false;
- }
-
/* Check the operands of the operation. */
if (!vect_is_simple_use (op0, loop_vinfo, &def_stmt, &def, &dt[0]))
{
/* Supportable by target? */
if ((modifier == NONE
- && !targetm.vectorize.builtin_conversion (code, vectype_in))
+ && !targetm.vectorize.builtin_conversion (code, integral_type))
|| (modifier == WIDEN
&& !supportable_widening_operation (code, stmt, vectype_in,
&decl1, &decl2,
- &code1, &code2))
+ &code1, &code2,
+ &dummy_int, &dummy))
|| (modifier == NARROW
&& !supportable_narrowing_operation (code, stmt, vectype_in,
- &code1)))
+ &code1, &dummy_int, &dummy)))
{
if (vect_print_dump_info (REPORT_DETAILS))
- fprintf (vect_dump, "op not supported by target.");
+ fprintf (vect_dump, "conversion not supported by target.");
return false;
}
ssa_op_iter iter;
if (j == 0)
- vect_get_vec_defs (op0, NULL, stmt, &vec_oprnds0, NULL, slp_node);
+ vect_get_vec_defs (op0, NULL, stmt, &vec_oprnds0, NULL, slp_node);
else
vect_get_vec_defs_for_stmt_copy (dt, &vec_oprnds0, NULL);
builtin_decl =
- targetm.vectorize.builtin_conversion (code, vectype_in);
+ targetm.vectorize.builtin_conversion (code, integral_type);
for (i = 0; VEC_iterate (tree, vec_oprnds0, i, vop0); i++)
{
- new_stmt = build_call_expr (builtin_decl, 1, vop0);
-
/* Arguments are ready. create the new vector stmt. */
- new_stmt = build_gimple_modify_stmt (vec_dest, new_stmt);
+ new_stmt = gimple_build_call (builtin_decl, 1, vop0);
new_temp = make_ssa_name (vec_dest, new_stmt);
- GIMPLE_STMT_OPERAND (new_stmt, 0) = new_temp;
- vect_finish_stmt_generation (stmt, new_stmt, bsi);
+ gimple_call_set_lhs (new_stmt, new_temp);
+ vect_finish_stmt_generation (stmt, new_stmt, gsi);
FOR_EACH_SSA_TREE_OPERAND (sym, new_stmt, iter,
SSA_OP_ALL_VIRTUALS)
{
mark_sym_for_renaming (sym);
}
if (slp_node)
- VEC_quick_push (tree, SLP_TREE_VEC_STMTS (slp_node), new_stmt);
+ VEC_quick_push (gimple, SLP_TREE_VEC_STMTS (slp_node), new_stmt);
}
if (j == 0)
/* Generate first half of the widened result: */
new_stmt
- = vect_gen_widened_results_half (code1, vectype_out, decl1,
+ = vect_gen_widened_results_half (code1, decl1,
vec_oprnd0, vec_oprnd1,
- unary_op, vec_dest, bsi, stmt);
+ unary_op, vec_dest, gsi, stmt);
if (j == 0)
STMT_VINFO_VEC_STMT (stmt_info) = new_stmt;
else
/* Generate second half of the widened result: */
new_stmt
- = vect_gen_widened_results_half (code2, vectype_out, decl2,
+ = vect_gen_widened_results_half (code2, decl2,
vec_oprnd0, vec_oprnd1,
- unary_op, vec_dest, bsi, stmt);
+ unary_op, vec_dest, gsi, stmt);
STMT_VINFO_RELATED_STMT (prev_stmt_info) = new_stmt;
prev_stmt_info = vinfo_for_stmt (new_stmt);
}
/* Arguments are ready. Create the new vector stmt. */
expr = build2 (code1, vectype_out, vec_oprnd0, vec_oprnd1);
- new_stmt = build_gimple_modify_stmt (vec_dest, expr);
+ new_stmt = gimple_build_assign_with_ops (code1, vec_dest, vec_oprnd0,
+ vec_oprnd1);
new_temp = make_ssa_name (vec_dest, new_stmt);
- GIMPLE_STMT_OPERAND (new_stmt, 0) = new_temp;
- vect_finish_stmt_generation (stmt, new_stmt, bsi);
+ gimple_assign_set_lhs (new_stmt, new_temp);
+ vect_finish_stmt_generation (stmt, new_stmt, gsi);
if (j == 0)
STMT_VINFO_VEC_STMT (stmt_info) = new_stmt;
Return FALSE if not a vectorizable STMT, TRUE otherwise. */
bool
-vectorizable_assignment (tree stmt, block_stmt_iterator *bsi, tree *vec_stmt,
- slp_tree slp_node)
+vectorizable_assignment (gimple stmt, gimple_stmt_iterator *gsi,
+ gimple *vec_stmt, slp_tree slp_node)
{
tree vec_dest;
tree scalar_dest;
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;
+ tree def;
+ gimple def_stmt;
enum vect_def_type dt[2] = {vect_unknown_def_type, vect_unknown_def_type};
int nunits = TYPE_VECTOR_SUBPARTS (vectype);
- int ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits;
+ int ncopies;
int i;
VEC(tree,heap) *vec_oprnds = NULL;
tree vop;
- /* FORNOW: SLP with multiple types is not supported. The SLP analysis
- verifies this, so we can safely override NCOPIES with 1 here. */
+ /* Multiple types in SLP are handled by creating the appropriate number of
+ vectorized stmts for each SLP node. Hence, NCOPIES is always 1 in
+ case of SLP. */
if (slp_node)
ncopies = 1;
+ else
+ ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits;
gcc_assert (ncopies >= 1);
if (ncopies > 1)
return false;
/* Is vectorizable assignment? */
- if (TREE_CODE (stmt) != GIMPLE_MODIFY_STMT)
+ if (!is_gimple_assign (stmt))
return false;
- scalar_dest = GIMPLE_STMT_OPERAND (stmt, 0);
+ scalar_dest = gimple_assign_lhs (stmt);
if (TREE_CODE (scalar_dest) != SSA_NAME)
return false;
- op = GIMPLE_STMT_OPERAND (stmt, 1);
+ if (gimple_assign_single_p (stmt)
+ || gimple_assign_rhs_code (stmt) == PAREN_EXPR)
+ op = gimple_assign_rhs1 (stmt);
+ else
+ return false;
+
if (!vect_is_simple_use (op, loop_vinfo, &def_stmt, &def, &dt[0]))
{
if (vect_print_dump_info (REPORT_DETAILS))
/* Arguments are ready. create the new vector stmt. */
for (i = 0; VEC_iterate (tree, vec_oprnds, i, vop); i++)
{
- *vec_stmt = build_gimple_modify_stmt (vec_dest, vop);
+ *vec_stmt = gimple_build_assign (vec_dest, vop);
new_temp = make_ssa_name (vec_dest, *vec_stmt);
- GIMPLE_STMT_OPERAND (*vec_stmt, 0) = new_temp;
- vect_finish_stmt_generation (stmt, *vec_stmt, bsi);
+ gimple_assign_set_lhs (*vec_stmt, new_temp);
+ vect_finish_stmt_generation (stmt, *vec_stmt, gsi);
STMT_VINFO_VEC_STMT (stmt_info) = *vec_stmt;
if (slp_node)
- VEC_quick_push (tree, SLP_TREE_VEC_STMTS (slp_node), *vec_stmt);
+ VEC_quick_push (gimple, SLP_TREE_VEC_STMTS (slp_node), *vec_stmt);
}
VEC_free (tree, heap, vec_oprnds);
Return FALSE if not a vectorizable STMT, TRUE otherwise. */
bool
-vectorizable_induction (tree phi, block_stmt_iterator *bsi ATTRIBUTE_UNUSED,
- tree *vec_stmt)
+vectorizable_induction (gimple phi, gimple_stmt_iterator *gsi ATTRIBUTE_UNUSED,
+ gimple *vec_stmt)
{
stmt_vec_info stmt_info = vinfo_for_stmt (phi);
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);
int nunits = TYPE_VECTOR_SUBPARTS (vectype);
int ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits;
tree vec_def;
gcc_assert (ncopies >= 1);
+ /* FORNOW. This restriction should be relaxed. */
+ if (nested_in_vect_loop_p (loop, phi) && ncopies > 1)
+ {
+ if (vect_print_dump_info (REPORT_DETAILS))
+ fprintf (vect_dump, "multiple types in nested loop.");
+ return false;
+ }
if (!STMT_VINFO_RELEVANT_P (stmt_info))
return false;
gcc_assert (STMT_VINFO_DEF_TYPE (stmt_info) == vect_induction_def);
- if (TREE_CODE (phi) != PHI_NODE)
+ if (gimple_code (phi) != GIMPLE_PHI)
return false;
if (!vec_stmt) /* transformation not required. */
Return FALSE if not a vectorizable STMT, TRUE otherwise. */
bool
-vectorizable_operation (tree stmt, block_stmt_iterator *bsi, tree *vec_stmt,
- slp_tree slp_node)
+vectorizable_operation (gimple stmt, gimple_stmt_iterator *gsi,
+ gimple *vec_stmt, slp_tree slp_node)
{
tree vec_dest;
tree scalar_dest;
- tree operation;
tree op0, op1 = 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);
- struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
enum tree_code code;
enum machine_mode vec_mode;
tree new_temp;
optab optab;
int icode;
enum machine_mode optab_op2_mode;
- tree def, def_stmt;
+ tree def;
+ gimple def_stmt;
enum vect_def_type dt[2] = {vect_unknown_def_type, vect_unknown_def_type};
- tree new_stmt = NULL_TREE;
+ gimple new_stmt = NULL;
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 ncopies;
int j, i;
VEC(tree,heap) *vec_oprnds0 = NULL, *vec_oprnds1 = NULL;
tree vop0, vop1;
bool shift_p = false;
bool scalar_shift_arg = false;
- /* FORNOW: SLP with multiple types is not supported. The SLP analysis verifies
- this, so we can safely override NCOPIES with 1 here. */
+ /* Multiple types in SLP are handled by creating the appropriate number of
+ vectorized stmts for each SLP node. Hence, NCOPIES is always 1 in
+ case of SLP. */
if (slp_node)
ncopies = 1;
+ else
+ ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits_in;
+
gcc_assert (ncopies >= 1);
- /* FORNOW. This restriction should be relaxed. */
- if (nested_in_vect_loop_p (loop, stmt) && ncopies > 1)
- {
- if (vect_print_dump_info (REPORT_DETAILS))
- fprintf (vect_dump, "multiple types in nested loop.");
- return false;
- }
if (!STMT_VINFO_RELEVANT_P (stmt_info))
return false;
return false;
/* Is STMT a vectorizable binary/unary operation? */
- if (TREE_CODE (stmt) != GIMPLE_MODIFY_STMT)
+ if (!is_gimple_assign (stmt))
return false;
- if (TREE_CODE (GIMPLE_STMT_OPERAND (stmt, 0)) != SSA_NAME)
+ if (TREE_CODE (gimple_assign_lhs (stmt)) != SSA_NAME)
return false;
- scalar_dest = GIMPLE_STMT_OPERAND (stmt, 0);
+ scalar_dest = gimple_assign_lhs (stmt);
vectype_out = get_vectype_for_scalar_type (TREE_TYPE (scalar_dest));
if (!vectype_out)
return false;
if (nunits_out != nunits_in)
return false;
- operation = GIMPLE_STMT_OPERAND (stmt, 1);
- code = TREE_CODE (operation);
+ code = gimple_assign_rhs_code (stmt);
/* For pointer addition, we should use the normal plus for
the vector addition. */
code = PLUS_EXPR;
/* Support only unary or binary operations. */
- op_type = TREE_OPERAND_LENGTH (operation);
+ op_type = TREE_CODE_LENGTH (code);
if (op_type != unary_op && op_type != binary_op)
{
if (vect_print_dump_info (REPORT_DETAILS))
return false;
}
- op0 = TREE_OPERAND (operation, 0);
+ op0 = gimple_assign_rhs1 (stmt);
if (!vect_is_simple_use (op0, loop_vinfo, &def_stmt, &def, &dt[0]))
{
if (vect_print_dump_info (REPORT_DETAILS))
if (op_type == binary_op)
{
- op1 = TREE_OPERAND (operation, 1);
+ op1 = gimple_assign_rhs2 (stmt);
if (!vect_is_simple_use (op1, loop_vinfo, &def_stmt, &def, &dt[1]))
{
if (vect_print_dump_info (REPORT_DETAILS))
/* Arguments are ready. Create the new vector stmt. */
for (i = 0; VEC_iterate (tree, vec_oprnds0, i, vop0); i++)
{
- if (op_type == binary_op)
- {
- vop1 = VEC_index (tree, vec_oprnds1, i);
- new_stmt = build_gimple_modify_stmt (vec_dest,
- build2 (code, vectype, vop0, vop1));
- }
- else
- new_stmt = build_gimple_modify_stmt (vec_dest,
- build1 (code, vectype, vop0));
-
+ vop1 = ((op_type == binary_op)
+ ? VEC_index (tree, vec_oprnds1, i) : NULL);
+ new_stmt = gimple_build_assign_with_ops (code, vec_dest, vop0, vop1);
new_temp = make_ssa_name (vec_dest, new_stmt);
- GIMPLE_STMT_OPERAND (new_stmt, 0) = new_temp;
- vect_finish_stmt_generation (stmt, new_stmt, bsi);
+ gimple_assign_set_lhs (new_stmt, new_temp);
+ vect_finish_stmt_generation (stmt, new_stmt, gsi);
if (slp_node)
- VEC_quick_push (tree, SLP_TREE_VEC_STMTS (slp_node), new_stmt);
+ VEC_quick_push (gimple, SLP_TREE_VEC_STMTS (slp_node), new_stmt);
}
+ if (slp_node)
+ continue;
+
if (j == 0)
STMT_VINFO_VEC_STMT (stmt_info) = *vec_stmt = new_stmt;
else
}
+/* Get vectorized definitions for loop-based vectorization. For the first
+ operand we call vect_get_vec_def_for_operand() (with OPRND containing
+ scalar operand), and for the rest we get a copy with
+ vect_get_vec_def_for_stmt_copy() using the previous vector definition
+ (stored in OPRND). See vect_get_vec_def_for_stmt_copy() for details.
+ The vectors are collected into VEC_OPRNDS. */
+
+static void
+vect_get_loop_based_defs (tree *oprnd, gimple stmt, enum vect_def_type dt,
+ VEC (tree, heap) **vec_oprnds, int multi_step_cvt)
+{
+ tree vec_oprnd;
+
+ /* Get first vector operand. */
+ /* All the vector operands except the very first one (that is scalar oprnd)
+ are stmt copies. */
+ if (TREE_CODE (TREE_TYPE (*oprnd)) != VECTOR_TYPE)
+ vec_oprnd = vect_get_vec_def_for_operand (*oprnd, stmt, NULL);
+ else
+ vec_oprnd = vect_get_vec_def_for_stmt_copy (dt, *oprnd);
+
+ VEC_quick_push (tree, *vec_oprnds, vec_oprnd);
+
+ /* Get second vector operand. */
+ vec_oprnd = vect_get_vec_def_for_stmt_copy (dt, vec_oprnd);
+ VEC_quick_push (tree, *vec_oprnds, vec_oprnd);
+
+ *oprnd = vec_oprnd;
+
+ /* For conversion in multiple steps, continue to get operands
+ recursively. */
+ if (multi_step_cvt)
+ vect_get_loop_based_defs (oprnd, stmt, dt, vec_oprnds, multi_step_cvt - 1);
+}
+
+
+/* Create vectorized demotion statements for vector operands from VEC_OPRNDS.
+ For multi-step conversions store the resulting vectors and call the function
+ recursively. */
+
+static void
+vect_create_vectorized_demotion_stmts (VEC (tree, heap) **vec_oprnds,
+ int multi_step_cvt, gimple stmt,
+ VEC (tree, heap) *vec_dsts,
+ gimple_stmt_iterator *gsi,
+ slp_tree slp_node, enum tree_code code,
+ stmt_vec_info *prev_stmt_info)
+{
+ unsigned int i;
+ tree vop0, vop1, new_tmp, vec_dest;
+ gimple new_stmt;
+ stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
+
+ vec_dest = VEC_pop (tree, vec_dsts);
+
+ for (i = 0; i < VEC_length (tree, *vec_oprnds); i += 2)
+ {
+ /* Create demotion operation. */
+ vop0 = VEC_index (tree, *vec_oprnds, i);
+ vop1 = VEC_index (tree, *vec_oprnds, i + 1);
+ new_stmt = gimple_build_assign_with_ops (code, vec_dest, vop0, vop1);
+ new_tmp = make_ssa_name (vec_dest, new_stmt);
+ gimple_assign_set_lhs (new_stmt, new_tmp);
+ vect_finish_stmt_generation (stmt, new_stmt, gsi);
+
+ if (multi_step_cvt)
+ /* Store the resulting vector for next recursive call. */
+ VEC_replace (tree, *vec_oprnds, i/2, new_tmp);
+ else
+ {
+ /* This is the last step of the conversion sequence. Store the
+ vectors in SLP_NODE or in vector info of the scalar statement
+ (or in STMT_VINFO_RELATED_STMT chain). */
+ if (slp_node)
+ VEC_quick_push (gimple, SLP_TREE_VEC_STMTS (slp_node), new_stmt);
+ else
+ {
+ if (!*prev_stmt_info)
+ 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);
+ }
+ }
+ }
+
+ /* For multi-step demotion operations we first generate demotion operations
+ from the source type to the intermediate types, and then combine the
+ results (stored in VEC_OPRNDS) in demotion operation to the destination
+ type. */
+ if (multi_step_cvt)
+ {
+ /* At each level of recursion we have have of the operands we had at the
+ previous level. */
+ VEC_truncate (tree, *vec_oprnds, (i+1)/2);
+ vect_create_vectorized_demotion_stmts (vec_oprnds, multi_step_cvt - 1,
+ stmt, vec_dsts, gsi, slp_node,
+ code, prev_stmt_info);
+ }
+}
+
+
/* Function vectorizable_type_demotion
Check if STMT performs a binary or unary operation that involves
Return FALSE if not a vectorizable STMT, TRUE otherwise. */
bool
-vectorizable_type_demotion (tree stmt, block_stmt_iterator *bsi,
- tree *vec_stmt)
+vectorizable_type_demotion (gimple stmt, gimple_stmt_iterator *gsi,
+ gimple *vec_stmt, slp_tree slp_node)
{
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);
- struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
enum tree_code code, code1 = ERROR_MARK;
- tree new_temp;
- tree def, def_stmt;
+ tree def;
+ gimple def_stmt;
enum vect_def_type dt[2] = {vect_unknown_def_type, vect_unknown_def_type};
- tree new_stmt;
stmt_vec_info prev_stmt_info;
int nunits_in;
int nunits_out;
tree vectype_out;
int ncopies;
- int j;
- tree expr;
+ int j, i;
tree vectype_in;
+ int multi_step_cvt = 0;
+ VEC (tree, heap) *vec_oprnds0 = NULL;
+ VEC (tree, heap) *vec_dsts = NULL, *interm_types = NULL, *tmp_vec_dsts = NULL;
+ tree last_oprnd, intermediate_type;
if (!STMT_VINFO_RELEVANT_P (stmt_info))
return false;
return false;
/* Is STMT a vectorizable type-demotion operation? */
- if (TREE_CODE (stmt) != GIMPLE_MODIFY_STMT)
+ if (!is_gimple_assign (stmt))
return false;
- if (TREE_CODE (GIMPLE_STMT_OPERAND (stmt, 0)) != SSA_NAME)
+ if (TREE_CODE (gimple_assign_lhs (stmt)) != SSA_NAME)
return false;
- operation = GIMPLE_STMT_OPERAND (stmt, 1);
- code = TREE_CODE (operation);
- if (code != NOP_EXPR && code != CONVERT_EXPR)
+ code = gimple_assign_rhs_code (stmt);
+ if (!CONVERT_EXPR_CODE_P (code))
return false;
- op0 = TREE_OPERAND (operation, 0);
+ op0 = gimple_assign_rhs1 (stmt);
vectype_in = get_vectype_for_scalar_type (TREE_TYPE (op0));
if (!vectype_in)
return false;
nunits_in = TYPE_VECTOR_SUBPARTS (vectype_in);
- scalar_dest = GIMPLE_STMT_OPERAND (stmt, 0);
+ scalar_dest = gimple_assign_lhs (stmt);
vectype_out = get_vectype_for_scalar_type (TREE_TYPE (scalar_dest));
if (!vectype_out)
return false;
nunits_out = TYPE_VECTOR_SUBPARTS (vectype_out);
- if (nunits_in != nunits_out / 2) /* FORNOW */
+ if (nunits_in >= nunits_out)
return false;
- ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits_out;
+ /* Multiple types in SLP are handled by creating the appropriate number of
+ vectorized stmts for each SLP node. Hence, NCOPIES is always 1 in
+ case of SLP. */
+ if (slp_node)
+ ncopies = 1;
+ else
+ ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits_out;
+
gcc_assert (ncopies >= 1);
- /* FORNOW. This restriction should be relaxed. */
- if (nested_in_vect_loop_p (loop, stmt) && ncopies > 1)
- {
- if (vect_print_dump_info (REPORT_DETAILS))
- fprintf (vect_dump, "multiple types in nested loop.");
- return false;
- }
if (! ((INTEGRAL_TYPE_P (TREE_TYPE (scalar_dest))
&& INTEGRAL_TYPE_P (TREE_TYPE (op0)))
|| (SCALAR_FLOAT_TYPE_P (TREE_TYPE (scalar_dest))
&& SCALAR_FLOAT_TYPE_P (TREE_TYPE (op0))
- && (code == NOP_EXPR || code == CONVERT_EXPR))))
+ && CONVERT_EXPR_CODE_P (code))))
return false;
/* Check the operands of the operation. */
}
/* Supportable by target? */
- if (!supportable_narrowing_operation (code, stmt, vectype_in, &code1))
+ if (!supportable_narrowing_operation (code, stmt, vectype_in, &code1,
+ &multi_step_cvt, &interm_types))
return false;
STMT_VINFO_VECTYPE (stmt_info) = vectype_in;
fprintf (vect_dump, "transform type demotion operation. ncopies = %d.",
ncopies);
- /* Handle def. */
+ /* In case of multi-step demotion, we first generate demotion operations to
+ the intermediate types, and then from that types to the final one.
+ We create vector destinations for the intermediate type (TYPES) received
+ from supportable_narrowing_operation, and store them in the correct order
+ for future use in vect_create_vectorized_demotion_stmts(). */
+ if (multi_step_cvt)
+ vec_dsts = VEC_alloc (tree, heap, multi_step_cvt + 1);
+ else
+ vec_dsts = VEC_alloc (tree, heap, 1);
+
vec_dest = vect_create_destination_var (scalar_dest, vectype_out);
-
+ VEC_quick_push (tree, vec_dsts, vec_dest);
+
+ if (multi_step_cvt)
+ {
+ for (i = VEC_length (tree, interm_types) - 1;
+ VEC_iterate (tree, interm_types, i, intermediate_type); i--)
+ {
+ vec_dest = vect_create_destination_var (scalar_dest,
+ intermediate_type);
+ VEC_quick_push (tree, vec_dsts, vec_dest);
+ }
+ }
+
/* 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. */
+ last_oprnd = op0;
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);
- vec_oprnd1 = vect_get_vec_def_for_stmt_copy (dt[0], vec_oprnd0);
- }
+ if (slp_node)
+ vect_get_slp_defs (slp_node, &vec_oprnds0, NULL);
else
- {
- vec_oprnd0 = vect_get_vec_def_for_stmt_copy (dt[0], vec_oprnd1);
- vec_oprnd1 = vect_get_vec_def_for_stmt_copy (dt[0], vec_oprnd0);
- }
+ {
+ VEC_free (tree, heap, vec_oprnds0);
+ vec_oprnds0 = VEC_alloc (tree, heap,
+ (multi_step_cvt ? vect_pow2 (multi_step_cvt) * 2 : 2));
+ vect_get_loop_based_defs (&last_oprnd, stmt, dt[0], &vec_oprnds0,
+ vect_pow2 (multi_step_cvt) - 1);
+ }
- /* Arguments are ready. Create the new vector stmt. */
- expr = build2 (code1, vectype_out, vec_oprnd0, vec_oprnd1);
- new_stmt = build_gimple_modify_stmt (vec_dest, expr);
- new_temp = make_ssa_name (vec_dest, new_stmt);
- GIMPLE_STMT_OPERAND (new_stmt, 0) = new_temp;
- vect_finish_stmt_generation (stmt, new_stmt, bsi);
+ /* Arguments are ready. Create the new vector stmts. */
+ tmp_vec_dsts = VEC_copy (tree, heap, vec_dsts);
+ vect_create_vectorized_demotion_stmts (&vec_oprnds0,
+ multi_step_cvt, stmt, tmp_vec_dsts,
+ gsi, slp_node, code1,
+ &prev_stmt_info);
+ }
- if (j == 0)
- STMT_VINFO_VEC_STMT (stmt_info) = new_stmt;
+ VEC_free (tree, heap, vec_oprnds0);
+ VEC_free (tree, heap, vec_dsts);
+ VEC_free (tree, heap, tmp_vec_dsts);
+ VEC_free (tree, heap, interm_types);
+
+ *vec_stmt = STMT_VINFO_VEC_STMT (stmt_info);
+ return true;
+}
+
+
+/* Create vectorized promotion statements for vector operands from VEC_OPRNDS0
+ and VEC_OPRNDS1 (for binary operations). For multi-step conversions store
+ the resulting vectors and call the function recursively. */
+
+static void
+vect_create_vectorized_promotion_stmts (VEC (tree, heap) **vec_oprnds0,
+ VEC (tree, heap) **vec_oprnds1,
+ int multi_step_cvt, gimple stmt,
+ VEC (tree, heap) *vec_dsts,
+ gimple_stmt_iterator *gsi,
+ slp_tree slp_node, enum tree_code code1,
+ enum tree_code code2, tree decl1,
+ tree decl2, int op_type,
+ stmt_vec_info *prev_stmt_info)
+{
+ int i;
+ tree vop0, vop1, new_tmp1, new_tmp2, vec_dest;
+ gimple new_stmt1, new_stmt2;
+ stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
+ VEC (tree, heap) *vec_tmp;
+
+ vec_dest = VEC_pop (tree, vec_dsts);
+ vec_tmp = VEC_alloc (tree, heap, VEC_length (tree, *vec_oprnds0) * 2);
+
+ for (i = 0; VEC_iterate (tree, *vec_oprnds0, i, vop0); i++)
+ {
+ if (op_type == binary_op)
+ vop1 = VEC_index (tree, *vec_oprnds1, i);
else
- STMT_VINFO_RELATED_STMT (prev_stmt_info) = new_stmt;
+ vop1 = NULL_TREE;
+
+ /* Generate the two halves of promotion operation. */
+ new_stmt1 = vect_gen_widened_results_half (code1, decl1, vop0, vop1,
+ op_type, vec_dest, gsi, stmt);
+ new_stmt2 = vect_gen_widened_results_half (code2, decl2, vop0, vop1,
+ op_type, vec_dest, gsi, stmt);
+ if (is_gimple_call (new_stmt1))
+ {
+ new_tmp1 = gimple_call_lhs (new_stmt1);
+ new_tmp2 = gimple_call_lhs (new_stmt2);
+ }
+ else
+ {
+ new_tmp1 = gimple_assign_lhs (new_stmt1);
+ new_tmp2 = gimple_assign_lhs (new_stmt2);
+ }
- prev_stmt_info = vinfo_for_stmt (new_stmt);
+ if (multi_step_cvt)
+ {
+ /* Store the results for the recursive call. */
+ VEC_quick_push (tree, vec_tmp, new_tmp1);
+ VEC_quick_push (tree, vec_tmp, new_tmp2);
+ }
+ else
+ {
+ /* Last step of promotion sequience - store the results. */
+ if (slp_node)
+ {
+ VEC_quick_push (gimple, SLP_TREE_VEC_STMTS (slp_node), new_stmt1);
+ VEC_quick_push (gimple, SLP_TREE_VEC_STMTS (slp_node), new_stmt2);
+ }
+ else
+ {
+ if (!*prev_stmt_info)
+ STMT_VINFO_VEC_STMT (stmt_info) = new_stmt1;
+ else
+ STMT_VINFO_RELATED_STMT (*prev_stmt_info) = new_stmt1;
+
+ *prev_stmt_info = vinfo_for_stmt (new_stmt1);
+ STMT_VINFO_RELATED_STMT (*prev_stmt_info) = new_stmt2;
+ *prev_stmt_info = vinfo_for_stmt (new_stmt2);
+ }
+ }
}
- *vec_stmt = STMT_VINFO_VEC_STMT (stmt_info);
- return true;
+ if (multi_step_cvt)
+ {
+ /* For multi-step promotion operation we first generate we call the
+ function recurcively for every stage. We start from the input type,
+ create promotion operations to the intermediate types, and then
+ create promotions to the output type. */
+ *vec_oprnds0 = VEC_copy (tree, heap, vec_tmp);
+ VEC_free (tree, heap, vec_tmp);
+ vect_create_vectorized_promotion_stmts (vec_oprnds0, vec_oprnds1,
+ multi_step_cvt - 1, stmt,
+ vec_dsts, gsi, slp_node, code1,
+ code2, decl2, decl2, op_type,
+ prev_stmt_info);
+ }
}
Return FALSE if not a vectorizable STMT, TRUE otherwise. */
bool
-vectorizable_type_promotion (tree stmt, block_stmt_iterator *bsi,
- tree *vec_stmt)
+vectorizable_type_promotion (gimple stmt, gimple_stmt_iterator *gsi,
+ gimple *vec_stmt, slp_tree slp_node)
{
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);
- struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
enum tree_code code, code1 = ERROR_MARK, code2 = ERROR_MARK;
tree decl1 = NULL_TREE, decl2 = NULL_TREE;
int op_type;
- tree def, def_stmt;
+ tree def;
+ gimple def_stmt;
enum vect_def_type dt[2] = {vect_unknown_def_type, vect_unknown_def_type};
- tree new_stmt;
stmt_vec_info prev_stmt_info;
int nunits_in;
int nunits_out;
tree vectype_out;
int ncopies;
- int j;
+ int j, i;
tree vectype_in;
+ tree intermediate_type = NULL_TREE;
+ int multi_step_cvt = 0;
+ VEC (tree, heap) *vec_oprnds0 = NULL, *vec_oprnds1 = NULL;
+ VEC (tree, heap) *vec_dsts = NULL, *interm_types = NULL, *tmp_vec_dsts = NULL;
if (!STMT_VINFO_RELEVANT_P (stmt_info))
return false;
return false;
/* Is STMT a vectorizable type-promotion operation? */
- if (TREE_CODE (stmt) != GIMPLE_MODIFY_STMT)
+ if (!is_gimple_assign (stmt))
return false;
- if (TREE_CODE (GIMPLE_STMT_OPERAND (stmt, 0)) != SSA_NAME)
+ if (TREE_CODE (gimple_assign_lhs (stmt)) != SSA_NAME)
return false;
- operation = GIMPLE_STMT_OPERAND (stmt, 1);
- code = TREE_CODE (operation);
- if (code != NOP_EXPR && code != CONVERT_EXPR
+ code = gimple_assign_rhs_code (stmt);
+ if (!CONVERT_EXPR_CODE_P (code)
&& code != WIDEN_MULT_EXPR)
return false;
- op0 = TREE_OPERAND (operation, 0);
+ op0 = gimple_assign_rhs1 (stmt);
vectype_in = get_vectype_for_scalar_type (TREE_TYPE (op0));
if (!vectype_in)
return false;
nunits_in = TYPE_VECTOR_SUBPARTS (vectype_in);
- scalar_dest = GIMPLE_STMT_OPERAND (stmt, 0);
+ scalar_dest = gimple_assign_lhs (stmt);
vectype_out = get_vectype_for_scalar_type (TREE_TYPE (scalar_dest));
if (!vectype_out)
return false;
nunits_out = TYPE_VECTOR_SUBPARTS (vectype_out);
- if (nunits_out != nunits_in / 2) /* FORNOW */
+ if (nunits_in <= nunits_out)
return false;
- ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits_in;
+ /* Multiple types in SLP are handled by creating the appropriate number of
+ vectorized stmts for each SLP node. Hence, NCOPIES is always 1 in
+ case of SLP. */
+ if (slp_node)
+ ncopies = 1;
+ else
+ ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits_in;
+
gcc_assert (ncopies >= 1);
- /* FORNOW. This restriction should be relaxed. */
- if (nested_in_vect_loop_p (loop, stmt) && ncopies > 1)
- {
- if (vect_print_dump_info (REPORT_DETAILS))
- fprintf (vect_dump, "multiple types in nested loop.");
- return false;
- }
if (! ((INTEGRAL_TYPE_P (TREE_TYPE (scalar_dest))
&& INTEGRAL_TYPE_P (TREE_TYPE (op0)))
|| (SCALAR_FLOAT_TYPE_P (TREE_TYPE (scalar_dest))
&& SCALAR_FLOAT_TYPE_P (TREE_TYPE (op0))
- && (code == CONVERT_EXPR || code == NOP_EXPR))))
+ && CONVERT_EXPR_CODE_P (code))))
return false;
/* Check the operands of the operation. */
op_type = TREE_CODE_LENGTH (code);
if (op_type == binary_op)
{
- op1 = TREE_OPERAND (operation, 1);
+ op1 = gimple_assign_rhs2 (stmt);
if (!vect_is_simple_use (op1, loop_vinfo, &def_stmt, &def, &dt[1]))
{
if (vect_print_dump_info (REPORT_DETAILS))
/* Supportable by target? */
if (!supportable_widening_operation (code, stmt, vectype_in,
- &decl1, &decl2, &code1, &code2))
+ &decl1, &decl2, &code1, &code2,
+ &multi_step_cvt, &interm_types))
return false;
+ /* Binary widening operation can only be supported directly by the
+ architecture. */
+ gcc_assert (!(multi_step_cvt && op_type == binary_op));
+
STMT_VINFO_VECTYPE (stmt_info) = vectype_in;
if (!vec_stmt) /* transformation not required. */
ncopies);
/* Handle def. */
+ /* In case of multi-step promotion, we first generate promotion operations
+ to the intermediate types, and then from that types to the final one.
+ We store vector destination in VEC_DSTS in the correct order for
+ recursive creation of promotion operations in
+ vect_create_vectorized_promotion_stmts(). Vector destinations are created
+ according to TYPES recieved from supportable_widening_operation(). */
+ if (multi_step_cvt)
+ vec_dsts = VEC_alloc (tree, heap, multi_step_cvt + 1);
+ else
+ vec_dsts = VEC_alloc (tree, heap, 1);
+
vec_dest = vect_create_destination_var (scalar_dest, vectype_out);
+ VEC_quick_push (tree, vec_dsts, vec_dest);
+
+ if (multi_step_cvt)
+ {
+ for (i = VEC_length (tree, interm_types) - 1;
+ VEC_iterate (tree, interm_types, i, intermediate_type); i--)
+ {
+ vec_dest = vect_create_destination_var (scalar_dest,
+ intermediate_type);
+ VEC_quick_push (tree, vec_dsts, vec_dest);
+ }
+ }
+
+ if (!slp_node)
+ {
+ vec_oprnds0 = VEC_alloc (tree, heap,
+ (multi_step_cvt ? vect_pow2 (multi_step_cvt) : 1));
+ if (op_type == binary_op)
+ vec_oprnds1 = VEC_alloc (tree, heap, 1);
+ }
/* 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
/* 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);
+ if (slp_node)
+ vect_get_slp_defs (slp_node, &vec_oprnds0, &vec_oprnds1);
+ else
+ {
+ vec_oprnd0 = vect_get_vec_def_for_operand (op0, stmt, NULL);
+ VEC_quick_push (tree, vec_oprnds0, vec_oprnd0);
+ if (op_type == binary_op)
+ {
+ vec_oprnd1 = vect_get_vec_def_for_operand (op1, stmt, NULL);
+ VEC_quick_push (tree, vec_oprnds1, vec_oprnd1);
+ }
+ }
}
else
{
- vec_oprnd0 = vect_get_vec_def_for_stmt_copy (dt[0], vec_oprnd0);
- if (op_type == binary_op)
- vec_oprnd1 = vect_get_vec_def_for_stmt_copy (dt[1], vec_oprnd1);
+ vec_oprnd0 = vect_get_vec_def_for_stmt_copy (dt[0], vec_oprnd0);
+ VEC_replace (tree, vec_oprnds0, 0, vec_oprnd0);
+ if (op_type == binary_op)
+ {
+ vec_oprnd1 = vect_get_vec_def_for_stmt_copy (dt[1], vec_oprnd1);
+ VEC_replace (tree, vec_oprnds1, 0, 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);
-
+ /* Arguments are ready. Create the new vector stmts. */
+ tmp_vec_dsts = VEC_copy (tree, heap, vec_dsts);
+ vect_create_vectorized_promotion_stmts (&vec_oprnds0, &vec_oprnds1,
+ multi_step_cvt, stmt,
+ tmp_vec_dsts,
+ gsi, slp_node, code1, code2,
+ decl1, decl2, op_type,
+ &prev_stmt_info);
}
+ VEC_free (tree, heap, vec_dsts);
+ VEC_free (tree, heap, tmp_vec_dsts);
+ VEC_free (tree, heap, interm_types);
+ VEC_free (tree, heap, vec_oprnds0);
+ VEC_free (tree, heap, vec_oprnds1);
+
*vec_stmt = STMT_VINFO_VEC_STMT (stmt_info);
return true;
}
static bool
vect_permute_store_chain (VEC(tree,heap) *dr_chain,
unsigned int length,
- tree stmt,
- block_stmt_iterator *bsi,
+ gimple stmt,
+ gimple_stmt_iterator *gsi,
VEC(tree,heap) **result_chain)
{
- tree perm_dest, perm_stmt, vect1, vect2, high, low;
+ tree perm_dest, vect1, vect2, high, low;
+ gimple perm_stmt;
tree vectype = STMT_VINFO_VECTYPE (vinfo_for_stmt (stmt));
- tree scalar_dest, tmp;
+ tree scalar_dest;
int i;
unsigned int j;
+ enum tree_code high_code, low_code;
- scalar_dest = GIMPLE_STMT_OPERAND (stmt, 0);
+ scalar_dest = gimple_assign_lhs (stmt);
/* Check that the operation is supported. */
if (!vect_strided_store_supported (vectype))
DECL_GIMPLE_REG_P (perm_dest) = 1;
add_referenced_var (perm_dest);
if (BYTES_BIG_ENDIAN)
- tmp = build2 (VEC_INTERLEAVE_HIGH_EXPR, vectype, vect1, vect2);
+ {
+ high_code = VEC_INTERLEAVE_HIGH_EXPR;
+ low_code = VEC_INTERLEAVE_LOW_EXPR;
+ }
else
- tmp = build2 (VEC_INTERLEAVE_LOW_EXPR, vectype, vect1, vect2);
- perm_stmt = build_gimple_modify_stmt (perm_dest, tmp);
+ {
+ low_code = VEC_INTERLEAVE_HIGH_EXPR;
+ high_code = VEC_INTERLEAVE_LOW_EXPR;
+ }
+ perm_stmt = gimple_build_assign_with_ops (high_code, perm_dest,
+ vect1, vect2);
high = make_ssa_name (perm_dest, perm_stmt);
- GIMPLE_STMT_OPERAND (perm_stmt, 0) = high;
- vect_finish_stmt_generation (stmt, perm_stmt, bsi);
+ gimple_assign_set_lhs (perm_stmt, high);
+ vect_finish_stmt_generation (stmt, perm_stmt, gsi);
VEC_replace (tree, *result_chain, 2*j, high);
/* Create interleaving stmt:
perm_dest = create_tmp_var (vectype, "vect_inter_low");
DECL_GIMPLE_REG_P (perm_dest) = 1;
add_referenced_var (perm_dest);
- if (BYTES_BIG_ENDIAN)
- tmp = build2 (VEC_INTERLEAVE_LOW_EXPR, vectype, vect1, vect2);
- else
- tmp = build2 (VEC_INTERLEAVE_HIGH_EXPR, vectype, vect1, vect2);
- perm_stmt = build_gimple_modify_stmt (perm_dest, tmp);
+ perm_stmt = gimple_build_assign_with_ops (low_code, perm_dest,
+ vect1, vect2);
low = make_ssa_name (perm_dest, perm_stmt);
- GIMPLE_STMT_OPERAND (perm_stmt, 0) = low;
- vect_finish_stmt_generation (stmt, perm_stmt, bsi);
+ gimple_assign_set_lhs (perm_stmt, low);
+ vect_finish_stmt_generation (stmt, perm_stmt, gsi);
VEC_replace (tree, *result_chain, 2*j+1, low);
}
dr_chain = VEC_copy (tree, heap, *result_chain);
Return FALSE if not a vectorizable STMT, TRUE otherwise. */
bool
-vectorizable_store (tree stmt, block_stmt_iterator *bsi, tree *vec_stmt,
+vectorizable_store (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt,
slp_tree slp_node)
{
tree scalar_dest;
enum machine_mode vec_mode;
tree dummy;
enum dr_alignment_support alignment_support_scheme;
- tree def, def_stmt;
+ tree def;
+ gimple 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 ncopies;
int j;
- tree next_stmt, first_stmt = NULL_TREE;
+ gimple next_stmt, first_stmt = NULL;
bool strided_store = false;
unsigned int group_size, i;
VEC(tree,heap) *dr_chain = NULL, *oprnds = NULL, *result_chain = NULL;
stmt_vec_info first_stmt_vinfo;
unsigned int vec_num;
- /* FORNOW: SLP with multiple types is not supported. The SLP analysis verifies
- this, so we can safely override NCOPIES with 1 here. */
+ /* Multiple types in SLP are handled by creating the appropriate number of
+ vectorized stmts for each SLP node. Hence, NCOPIES is always 1 in
+ case of SLP. */
if (slp)
ncopies = 1;
+ else
+ ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits;
gcc_assert (ncopies >= 1);
/* Is vectorizable store? */
- if (TREE_CODE (stmt) != GIMPLE_MODIFY_STMT)
+ if (!is_gimple_assign (stmt))
return false;
- scalar_dest = GIMPLE_STMT_OPERAND (stmt, 0);
+ scalar_dest = gimple_assign_lhs (stmt);
if (TREE_CODE (scalar_dest) != ARRAY_REF
&& TREE_CODE (scalar_dest) != INDIRECT_REF
&& !STMT_VINFO_STRIDED_ACCESS (stmt_info))
return false;
- op = GIMPLE_STMT_OPERAND (stmt, 1);
+ gcc_assert (gimple_assign_single_p (stmt));
+ op = gimple_assign_rhs1 (stmt);
if (!vect_is_simple_use (op, loop_vinfo, &def_stmt, &def, &dt))
{
if (vect_print_dump_info (REPORT_DETAILS))
return false;
}
+ /* If accesses through a pointer to vectype do not alias the original
+ memory reference we have a problem. */
+ if (get_alias_set (vectype) != get_alias_set (TREE_TYPE (scalar_dest))
+ && !alias_set_subset_of (get_alias_set (vectype),
+ get_alias_set (TREE_TYPE (scalar_dest))))
+ {
+ if (vect_print_dump_info (REPORT_DETAILS))
+ fprintf (vect_dump, "vector type does not alias scalar type");
+ return false;
+ }
+
+ if (!useless_type_conversion_p (TREE_TYPE (op), TREE_TYPE (scalar_dest)))
+ {
+ if (vect_print_dump_info (REPORT_DETAILS))
+ fprintf (vect_dump, "operands of different types");
+ 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). */
next_stmt = DR_GROUP_NEXT_DR (stmt_info);
while (next_stmt)
{
- op = GIMPLE_STMT_OPERAND (next_stmt, 1);
+ gcc_assert (gimple_assign_single_p (next_stmt));
+ op = gimple_assign_rhs1 (next_stmt);
if (!vect_is_simple_use (op, loop_vinfo, &def_stmt, &def, &dt))
{
if (vect_print_dump_info (REPORT_DETAILS))
if (!vec_stmt) /* transformation not required. */
{
STMT_VINFO_TYPE (stmt_info) = store_vec_info_type;
- if (!PURE_SLP_STMT (stmt_info))
- vect_model_store_cost (stmt_info, ncopies, dt, NULL);
+ vect_model_store_cost (stmt_info, ncopies, dt, NULL);
return true;
}
< DR_GROUP_SIZE (vinfo_for_stmt (first_stmt))
&& !slp)
{
- *vec_stmt = NULL_TREE;
+ *vec_stmt = NULL;
return true;
}
strided_store = false;
/* VEC_NUM is the number of vect stmts to be created for this group. */
- if (slp && SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node) < group_size)
+ if (slp)
vec_num = SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node);
else
vec_num = group_size;
prev_stmt_info = NULL;
for (j = 0; j < ncopies; j++)
{
- tree new_stmt;
- tree ptr_incr;
+ gimple new_stmt;
+ gimple ptr_incr;
if (j == 0)
{
there is no interleaving, GROUP_SIZE is 1, and only one
iteration of the loop will be executed. */
gcc_assert (next_stmt);
- op = GIMPLE_STMT_OPERAND (next_stmt, 1);
+ gcc_assert (gimple_assign_single_p (next_stmt));
+ op = gimple_assign_rhs1 (next_stmt);
vec_oprnd = vect_get_vec_def_for_operand (op, next_stmt,
NULL);
next_stmt = DR_GROUP_NEXT_DR (vinfo_for_stmt (next_stmt));
}
}
+
dataref_ptr = vect_create_data_ref_ptr (first_stmt, NULL, NULL_TREE,
- &dummy, &ptr_incr, false,
- TREE_TYPE (vec_oprnd), &inv_p);
+ &dummy, &ptr_incr, false,
+ &inv_p);
gcc_assert (!inv_p);
}
else
{
- /* FORNOW SLP doesn't work for multiple types. */
- gcc_assert (!slp);
-
/* 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(),
VEC_replace(tree, oprnds, i, vec_oprnd);
}
dataref_ptr =
- bump_vector_ptr (dataref_ptr, ptr_incr, bsi, stmt, NULL_TREE);
+ bump_vector_ptr (dataref_ptr, ptr_incr, gsi, stmt, NULL_TREE);
}
if (strided_store)
{
result_chain = VEC_alloc (tree, heap, group_size);
/* Permute. */
- if (!vect_permute_store_chain (dr_chain, group_size, stmt, bsi,
+ if (!vect_permute_store_chain (dr_chain, group_size, stmt, gsi,
&result_chain))
return false;
}
{
if (i > 0)
/* Bump the vector pointer. */
- dataref_ptr = bump_vector_ptr (dataref_ptr, ptr_incr, bsi, stmt,
+ dataref_ptr = bump_vector_ptr (dataref_ptr, ptr_incr, gsi, stmt,
NULL_TREE);
if (slp)
data_ref = build_fold_indirect_ref (dataref_ptr);
/* Arguments are ready. Create the new vector stmt. */
- new_stmt = build_gimple_modify_stmt (data_ref, vec_oprnd);
- vect_finish_stmt_generation (stmt, new_stmt, bsi);
+ new_stmt = gimple_build_assign (data_ref, vec_oprnd);
+ vect_finish_stmt_generation (stmt, new_stmt, gsi);
mark_symbols_for_renaming (new_stmt);
+
+ if (slp)
+ continue;
if (j == 0)
STMT_VINFO_VEC_STMT (stmt_info) = *vec_stmt = new_stmt;
Return value - the result of the loop-header phi node. */
static tree
-vect_setup_realignment (tree stmt, block_stmt_iterator *bsi,
+vect_setup_realignment (gimple stmt, gimple_stmt_iterator *gsi,
tree *realignment_token,
enum dr_alignment_support alignment_support_scheme,
tree init_addr,
loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
edge pe;
- tree scalar_dest = GIMPLE_STMT_OPERAND (stmt, 0);
+ tree scalar_dest = gimple_assign_lhs (stmt);
tree vec_dest;
- tree inc;
+ gimple inc;
tree ptr;
tree data_ref;
- tree new_stmt;
+ gimple new_stmt;
basic_block new_bb;
tree msq_init = NULL_TREE;
tree new_temp;
- tree phi_stmt;
+ gimple phi_stmt;
tree msq = NULL_TREE;
- tree stmts = NULL_TREE;
+ gimple_seq stmts = NULL;
bool inv_p;
bool compute_in_loop = false;
bool nested_in_vect_loop = nested_in_vect_loop_p (loop, stmt);
- struct loop *containing_loop = (bb_for_stmt (stmt))->loop_father;
+ struct loop *containing_loop = (gimple_bb (stmt))->loop_father;
struct loop *loop_for_initial_load;
gcc_assert (alignment_support_scheme == dr_explicit_realign
pe = loop_preheader_edge (loop_for_initial_load);
vec_dest = vect_create_destination_var (scalar_dest, vectype);
ptr = vect_create_data_ref_ptr (stmt, loop_for_initial_load, NULL_TREE,
- &init_addr, &inc, true, NULL_TREE, &inv_p);
+ &init_addr, &inc, true, &inv_p);
data_ref = build1 (ALIGN_INDIRECT_REF, vectype, ptr);
- new_stmt = build_gimple_modify_stmt (vec_dest, data_ref);
+ new_stmt = gimple_build_assign (vec_dest, data_ref);
new_temp = make_ssa_name (vec_dest, new_stmt);
- GIMPLE_STMT_OPERAND (new_stmt, 0) = new_temp;
+ gimple_assign_set_lhs (new_stmt, new_temp);
mark_symbols_for_renaming (new_stmt);
- new_bb = bsi_insert_on_edge_immediate (pe, new_stmt);
+ new_bb = gsi_insert_on_edge_immediate (pe, new_stmt);
gcc_assert (!new_bb);
- msq_init = GIMPLE_STMT_OPERAND (new_stmt, 0);
+ msq_init = gimple_assign_lhs (new_stmt);
}
/* 4. Create realignment token using a target builtin, if available.
init_addr = vect_create_addr_base_for_vector_ref (stmt, &stmts,
NULL_TREE, loop);
pe = loop_preheader_edge (loop);
- new_bb = bsi_insert_on_edge_immediate (pe, stmts);
+ new_bb = gsi_insert_seq_on_edge_immediate (pe, stmts);
gcc_assert (!new_bb);
}
builtin_decl = targetm.vectorize.builtin_mask_for_load ();
- new_stmt = build_call_expr (builtin_decl, 1, init_addr);
- vec_dest = vect_create_destination_var (scalar_dest,
- TREE_TYPE (new_stmt));
- new_stmt = build_gimple_modify_stmt (vec_dest, new_stmt);
+ new_stmt = gimple_build_call (builtin_decl, 1, init_addr);
+ vec_dest =
+ vect_create_destination_var (scalar_dest,
+ gimple_call_return_type (new_stmt));
new_temp = make_ssa_name (vec_dest, new_stmt);
- GIMPLE_STMT_OPERAND (new_stmt, 0) = new_temp;
+ gimple_call_set_lhs (new_stmt, new_temp);
if (compute_in_loop)
- bsi_insert_before (bsi, new_stmt, BSI_SAME_STMT);
+ gsi_insert_before (gsi, new_stmt, GSI_SAME_STMT);
else
{
/* Generate the misalignment computation outside LOOP. */
pe = loop_preheader_edge (loop);
- new_bb = bsi_insert_on_edge_immediate (pe, new_stmt);
+ new_bb = gsi_insert_on_edge_immediate (pe, new_stmt);
gcc_assert (!new_bb);
}
- *realignment_token = GIMPLE_STMT_OPERAND (new_stmt, 0);
+ *realignment_token = gimple_call_lhs (new_stmt);
/* The result of the CALL_EXPR to this builtin is determined from
the value of the parameter and no global variables are touched
pe = loop_preheader_edge (containing_loop);
vec_dest = vect_create_destination_var (scalar_dest, vectype);
- msq = make_ssa_name (vec_dest, NULL_TREE);
+ msq = make_ssa_name (vec_dest, NULL);
phi_stmt = create_phi_node (msq, containing_loop->header);
SSA_NAME_DEF_STMT (msq) = phi_stmt;
add_phi_arg (phi_stmt, msq_init, pe);
static bool
vect_permute_load_chain (VEC(tree,heap) *dr_chain,
unsigned int length,
- tree stmt,
- block_stmt_iterator *bsi,
+ gimple stmt,
+ gimple_stmt_iterator *gsi,
VEC(tree,heap) **result_chain)
{
- tree perm_dest, perm_stmt, data_ref, first_vect, second_vect;
+ tree perm_dest, data_ref, first_vect, second_vect;
+ gimple perm_stmt;
tree vectype = STMT_VINFO_VECTYPE (vinfo_for_stmt (stmt));
- tree tmp;
int i;
unsigned int j;
DECL_GIMPLE_REG_P (perm_dest) = 1;
add_referenced_var (perm_dest);
- tmp = build2 (VEC_EXTRACT_EVEN_EXPR, vectype,
- first_vect, second_vect);
- perm_stmt = build_gimple_modify_stmt (perm_dest, tmp);
+ perm_stmt = gimple_build_assign_with_ops (VEC_EXTRACT_EVEN_EXPR,
+ perm_dest, first_vect,
+ second_vect);
data_ref = make_ssa_name (perm_dest, perm_stmt);
- GIMPLE_STMT_OPERAND (perm_stmt, 0) = data_ref;
- vect_finish_stmt_generation (stmt, perm_stmt, bsi);
+ gimple_assign_set_lhs (perm_stmt, data_ref);
+ vect_finish_stmt_generation (stmt, perm_stmt, gsi);
mark_symbols_for_renaming (perm_stmt);
VEC_replace (tree, *result_chain, j/2, data_ref);
DECL_GIMPLE_REG_P (perm_dest) = 1;
add_referenced_var (perm_dest);
- tmp = build2 (VEC_EXTRACT_ODD_EXPR, vectype,
- first_vect, second_vect);
- perm_stmt = build_gimple_modify_stmt (perm_dest, tmp);
+ perm_stmt = gimple_build_assign_with_ops (VEC_EXTRACT_ODD_EXPR,
+ perm_dest, first_vect,
+ second_vect);
data_ref = make_ssa_name (perm_dest, perm_stmt);
- GIMPLE_STMT_OPERAND (perm_stmt, 0) = data_ref;
- vect_finish_stmt_generation (stmt, perm_stmt, bsi);
+ gimple_assign_set_lhs (perm_stmt, data_ref);
+ vect_finish_stmt_generation (stmt, perm_stmt, gsi);
mark_symbols_for_renaming (perm_stmt);
VEC_replace (tree, *result_chain, j/2+length/2, data_ref);
*/
static bool
-vect_transform_strided_load (tree stmt, VEC(tree,heap) *dr_chain, int size,
- block_stmt_iterator *bsi)
+vect_transform_strided_load (gimple stmt, VEC(tree,heap) *dr_chain, int size,
+ gimple_stmt_iterator *gsi)
{
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
- tree first_stmt = DR_GROUP_FIRST_DR (stmt_info);
- tree next_stmt, new_stmt;
+ gimple first_stmt = DR_GROUP_FIRST_DR (stmt_info);
+ gimple next_stmt, new_stmt;
VEC(tree,heap) *result_chain = NULL;
unsigned int i, gap_count;
tree tmp_data_ref;
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))
+ if (!vect_permute_load_chain (dr_chain, size, stmt, gsi, &result_chain))
return false;
/* Put a permuted data-ref in the VECTORIZED_STMT field.
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));
+ gimple prev_stmt =
+ STMT_VINFO_VEC_STMT (vinfo_for_stmt (next_stmt));
+ gimple rel_stmt =
+ STMT_VINFO_RELATED_STMT (vinfo_for_stmt (prev_stmt));
while (rel_stmt)
{
prev_stmt = rel_stmt;
Return FALSE if not a vectorizable STMT, TRUE otherwise. */
bool
-vectorizable_load (tree stmt, block_stmt_iterator *bsi, tree *vec_stmt,
+vectorizable_load (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt,
slp_tree slp_node)
{
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 loop *containing_loop = (bb_for_stmt (stmt))->loop_father;
+ struct loop *containing_loop = (gimple_bb (stmt))->loop_father;
bool nested_in_vect_loop = nested_in_vect_loop_p (loop, stmt);
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;
+ gimple new_stmt = NULL;
tree dummy;
enum dr_alignment_support alignment_support_scheme;
tree dataref_ptr = NULL_TREE;
- tree ptr_incr;
+ gimple ptr_incr;
int nunits = TYPE_VECTOR_SUBPARTS (vectype);
- int ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits;
+ int ncopies;
int i, j, group_size;
tree msq = NULL_TREE, lsq;
tree offset = NULL_TREE;
tree realignment_token = NULL_TREE;
- tree phi = NULL_TREE;
+ gimple phi = NULL;
VEC(tree,heap) *dr_chain = NULL;
bool strided_load = false;
- tree first_stmt;
+ gimple first_stmt;
tree scalar_type;
bool inv_p;
bool compute_in_loop = false;
struct loop *at_loop;
int vec_num;
bool slp = (slp_node != NULL);
+ enum tree_code code;
- /* FORNOW: SLP with multiple types is not supported. The SLP analysis verifies
- this, so we can safely override NCOPIES with 1 here. */
+ /* Multiple types in SLP are handled by creating the appropriate number of
+ vectorized stmts for each SLP node. Hence, NCOPIES is always 1 in
+ case of SLP. */
if (slp)
ncopies = 1;
+ else
+ ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits;
gcc_assert (ncopies >= 1);
return false;
/* Is vectorizable load? */
- if (TREE_CODE (stmt) != GIMPLE_MODIFY_STMT)
+ if (!is_gimple_assign (stmt))
return false;
- scalar_dest = GIMPLE_STMT_OPERAND (stmt, 0);
+ scalar_dest = gimple_assign_lhs (stmt);
if (TREE_CODE (scalar_dest) != SSA_NAME)
return false;
- op = GIMPLE_STMT_OPERAND (stmt, 1);
- if (TREE_CODE (op) != ARRAY_REF
- && TREE_CODE (op) != INDIRECT_REF
+ code = gimple_assign_rhs_code (stmt);
+ if (code != ARRAY_REF
+ && code != INDIRECT_REF
&& !STMT_VINFO_STRIDED_ACCESS (stmt_info))
return false;
return false;
}
+ /* If accesses through a pointer to vectype do not alias the original
+ memory reference we have a problem. */
+ if (get_alias_set (vectype) != get_alias_set (scalar_type)
+ && !alias_set_subset_of (get_alias_set (vectype),
+ get_alias_set (scalar_type)))
+ {
+ if (vect_print_dump_info (REPORT_DETAILS))
+ fprintf (vect_dump, "vector type does not alias scalar type");
+ return false;
+ }
+
/* Check if the load is a part of an interleaving chain. */
if (STMT_VINFO_STRIDED_ACCESS (stmt_info))
{
}
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);
/* VEC_NUM is the number of vect stmts to be created for this group. */
if (slp)
}
else
vec_num = group_size;
+
+ dr_chain = VEC_alloc (tree, heap, vec_num);
}
else
{
nested within an outer-loop that is being vectorized. */
if (nested_in_vect_loop_p (loop, stmt)
- && (TREE_INT_CST_LOW (DR_STEP (dr)) % UNITS_PER_SIMD_WORD != 0))
+ && (TREE_INT_CST_LOW (DR_STEP (dr))
+ % GET_MODE_SIZE (TYPE_MODE (vectype)) != 0))
{
gcc_assert (alignment_support_scheme != dr_explicit_realign_optimized);
compute_in_loop = true;
|| alignment_support_scheme == dr_explicit_realign)
&& !compute_in_loop)
{
- msq = vect_setup_realignment (first_stmt, bsi, &realignment_token,
+ msq = vect_setup_realignment (first_stmt, gsi, &realignment_token,
alignment_support_scheme, NULL_TREE,
&at_loop);
if (alignment_support_scheme == dr_explicit_realign_optimized)
dataref_ptr = vect_create_data_ref_ptr (first_stmt,
at_loop, offset,
&dummy, &ptr_incr, false,
- NULL_TREE, &inv_p);
+ &inv_p);
else
dataref_ptr =
- bump_vector_ptr (dataref_ptr, ptr_incr, bsi, stmt, NULL_TREE);
+ bump_vector_ptr (dataref_ptr, ptr_incr, gsi, stmt, NULL_TREE);
for (i = 0; i < vec_num; i++)
{
if (i > 0)
- dataref_ptr = bump_vector_ptr (dataref_ptr, ptr_incr, bsi, stmt,
+ dataref_ptr = bump_vector_ptr (dataref_ptr, ptr_incr, gsi, stmt,
NULL_TREE);
/* 2. Create the vector-load in the loop. */
tree vs_minus_1 = size_int (TYPE_VECTOR_SUBPARTS (vectype) - 1);
if (compute_in_loop)
- msq = vect_setup_realignment (first_stmt, bsi,
+ msq = vect_setup_realignment (first_stmt, gsi,
&realignment_token,
dr_explicit_realign,
dataref_ptr, NULL);
data_ref = build1 (ALIGN_INDIRECT_REF, vectype, dataref_ptr);
vec_dest = vect_create_destination_var (scalar_dest, vectype);
- new_stmt = build_gimple_modify_stmt (vec_dest, data_ref);
+ new_stmt = gimple_build_assign (vec_dest, data_ref);
new_temp = make_ssa_name (vec_dest, new_stmt);
- GIMPLE_STMT_OPERAND (new_stmt, 0) = new_temp;
- vect_finish_stmt_generation (stmt, new_stmt, bsi);
+ gimple_assign_set_lhs (new_stmt, new_temp);
+ vect_finish_stmt_generation (stmt, new_stmt, gsi);
copy_virtual_operands (new_stmt, stmt);
mark_symbols_for_renaming (new_stmt);
msq = new_temp;
bump = size_binop (MULT_EXPR, vs_minus_1,
TYPE_SIZE_UNIT (scalar_type));
- ptr = bump_vector_ptr (dataref_ptr, NULL_TREE, bsi, stmt, bump);
+ ptr = bump_vector_ptr (dataref_ptr, NULL, gsi, stmt, bump);
data_ref = build1 (ALIGN_INDIRECT_REF, vectype, ptr);
break;
}
gcc_unreachable ();
}
vec_dest = vect_create_destination_var (scalar_dest, vectype);
- new_stmt = build_gimple_modify_stmt (vec_dest, data_ref);
+ new_stmt = gimple_build_assign (vec_dest, data_ref);
new_temp = make_ssa_name (vec_dest, new_stmt);
- GIMPLE_STMT_OPERAND (new_stmt, 0) = new_temp;
- vect_finish_stmt_generation (stmt, new_stmt, bsi);
+ gimple_assign_set_lhs (new_stmt, new_temp);
+ vect_finish_stmt_generation (stmt, new_stmt, gsi);
mark_symbols_for_renaming (new_stmt);
/* 3. Handle explicit realignment if necessary/supported. Create in
if (alignment_support_scheme == dr_explicit_realign_optimized
|| alignment_support_scheme == dr_explicit_realign)
{
- lsq = GIMPLE_STMT_OPERAND (new_stmt, 0);
+ tree tmp;
+
+ lsq = gimple_assign_lhs (new_stmt);
if (!realignment_token)
realignment_token = dataref_ptr;
vec_dest = vect_create_destination_var (scalar_dest, vectype);
- new_stmt = build3 (REALIGN_LOAD_EXPR, vectype, msq, lsq,
- realignment_token);
- new_stmt = build_gimple_modify_stmt (vec_dest, new_stmt);
+ tmp = build3 (REALIGN_LOAD_EXPR, vectype, msq, lsq,
+ realignment_token);
+ new_stmt = gimple_build_assign (vec_dest, tmp);
new_temp = make_ssa_name (vec_dest, new_stmt);
- GIMPLE_STMT_OPERAND (new_stmt, 0) = new_temp;
- vect_finish_stmt_generation (stmt, new_stmt, bsi);
+ gimple_assign_set_lhs (new_stmt, new_temp);
+ vect_finish_stmt_generation (stmt, new_stmt, gsi);
if (alignment_support_scheme == dr_explicit_realign_optimized)
{
+ gcc_assert (phi);
if (i == vec_num - 1 && j == ncopies - 1)
add_phi_arg (phi, lsq, loop_latch_edge (containing_loop));
msq = lsq;
/* CHECKME: bitpos depends on endianess? */
bitpos = bitsize_zero_node;
vec_inv = build3 (BIT_FIELD_REF, scalar_type, new_temp,
- bitsize, bitpos);
+ bitsize, bitpos);
vec_dest =
vect_create_destination_var (scalar_dest, NULL_TREE);
- new_stmt = build_gimple_modify_stmt (vec_dest, vec_inv);
+ new_stmt = gimple_build_assign (vec_dest, vec_inv);
new_temp = make_ssa_name (vec_dest, new_stmt);
- GIMPLE_STMT_OPERAND (new_stmt, 0) = new_temp;
- vect_finish_stmt_generation (stmt, new_stmt, bsi);
+ gimple_assign_set_lhs (new_stmt, new_temp);
+ vect_finish_stmt_generation (stmt, new_stmt, gsi);
for (k = nunits - 1; k >= 0; --k)
t = tree_cons (NULL_TREE, new_temp, t);
/* FIXME: use build_constructor directly. */
vec_inv = build_constructor_from_list (vectype, t);
- new_temp = vect_init_vector (stmt, vec_inv, vectype, bsi);
+ new_temp = vect_init_vector (stmt, vec_inv, vectype, gsi);
new_stmt = SSA_NAME_DEF_STMT (new_temp);
}
else
/* Store vector loads in the corresponding SLP_NODE. */
if (slp)
- VEC_quick_push (tree, SLP_TREE_VEC_STMTS (slp_node), new_stmt);
+ VEC_quick_push (gimple, SLP_TREE_VEC_STMTS (slp_node), new_stmt);
}
- /* FORNOW: SLP with multiple types is unsupported. */
if (slp)
- return true;
+ continue;
if (strided_load)
{
- if (!vect_transform_strided_load (stmt, dr_chain, group_size, bsi))
+ if (!vect_transform_strided_load (stmt, dr_chain, group_size, gsi))
return false;
*vec_stmt = STMT_VINFO_VEC_STMT (stmt_info);
VEC_free (tree, heap, dr_chain);
it can be supported. */
bool
-vectorizable_live_operation (tree stmt,
- block_stmt_iterator *bsi ATTRIBUTE_UNUSED,
- tree *vec_stmt ATTRIBUTE_UNUSED)
+vectorizable_live_operation (gimple stmt,
+ gimple_stmt_iterator *gsi ATTRIBUTE_UNUSED,
+ gimple *vec_stmt ATTRIBUTE_UNUSED)
{
- tree operation;
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);
int i;
int op_type;
tree op;
- tree def, def_stmt;
+ tree def;
+ gimple def_stmt;
enum vect_def_type dt;
+ enum tree_code code;
+ enum gimple_rhs_class rhs_class;
gcc_assert (STMT_VINFO_LIVE_P (stmt_info));
if (STMT_VINFO_DEF_TYPE (stmt_info) == vect_reduction_def)
return false;
- if (TREE_CODE (stmt) != GIMPLE_MODIFY_STMT)
+ if (!is_gimple_assign (stmt))
return false;
- if (TREE_CODE (GIMPLE_STMT_OPERAND (stmt, 0)) != SSA_NAME)
+ if (TREE_CODE (gimple_assign_lhs (stmt)) != SSA_NAME)
return false;
/* FORNOW. CHECKME. */
if (nested_in_vect_loop_p (loop, stmt))
return false;
- operation = GIMPLE_STMT_OPERAND (stmt, 1);
- op_type = TREE_OPERAND_LENGTH (operation);
+ code = gimple_assign_rhs_code (stmt);
+ op_type = TREE_CODE_LENGTH (code);
+ rhs_class = get_gimple_rhs_class (code);
+ gcc_assert (rhs_class != GIMPLE_UNARY_RHS || op_type == unary_op);
+ gcc_assert (rhs_class != GIMPLE_BINARY_RHS || op_type == binary_op);
/* FORNOW: support only if all uses are invariant. This means
that the scalar operations can remain in place, unvectorized.
for (i = 0; i < op_type; i++)
{
- op = TREE_OPERAND (operation, i);
+ if (rhs_class == GIMPLE_SINGLE_RHS)
+ op = TREE_OPERAND (gimple_op (stmt, 1), i);
+ else
+ op = gimple_op (stmt, i + 1);
if (op && !vect_is_simple_use (op, loop_vinfo, &def_stmt, &def, &dt))
{
if (vect_print_dump_info (REPORT_DETAILS))
if (TREE_CODE (lhs) == SSA_NAME)
{
- tree lhs_def_stmt = SSA_NAME_DEF_STMT (lhs);
+ gimple lhs_def_stmt = SSA_NAME_DEF_STMT (lhs);
if (!vect_is_simple_use (lhs, loop_vinfo, &lhs_def_stmt, &def, &dt))
return false;
}
if (TREE_CODE (rhs) == SSA_NAME)
{
- tree rhs_def_stmt = SSA_NAME_DEF_STMT (rhs);
+ gimple rhs_def_stmt = SSA_NAME_DEF_STMT (rhs);
if (!vect_is_simple_use (rhs, loop_vinfo, &rhs_def_stmt, &def, &dt))
return false;
}
Return FALSE if not a vectorizable STMT, TRUE otherwise. */
bool
-vectorizable_condition (tree stmt, block_stmt_iterator *bsi, tree *vec_stmt)
+vectorizable_condition (gimple stmt, gimple_stmt_iterator *gsi,
+ gimple *vec_stmt)
{
tree scalar_dest = NULL_TREE;
tree vec_dest = NULL_TREE;
enum vect_def_type dt;
int nunits = TYPE_VECTOR_SUBPARTS (vectype);
int ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits;
+ enum tree_code code;
gcc_assert (ncopies >= 1);
if (ncopies > 1)
}
/* Is vectorizable conditional operation? */
- if (TREE_CODE (stmt) != GIMPLE_MODIFY_STMT)
+ if (!is_gimple_assign (stmt))
return false;
- op = GIMPLE_STMT_OPERAND (stmt, 1);
+ code = gimple_assign_rhs_code (stmt);
- if (TREE_CODE (op) != COND_EXPR)
+ if (code != COND_EXPR)
return false;
+ gcc_assert (gimple_assign_single_p (stmt));
+ op = gimple_assign_rhs1 (stmt);
cond_expr = TREE_OPERAND (op, 0);
then_clause = TREE_OPERAND (op, 1);
else_clause = TREE_OPERAND (op, 2);
if (TREE_CODE (then_clause) == SSA_NAME)
{
- tree then_def_stmt = SSA_NAME_DEF_STMT (then_clause);
+ gimple then_def_stmt = SSA_NAME_DEF_STMT (then_clause);
if (!vect_is_simple_use (then_clause, loop_vinfo,
&then_def_stmt, &def, &dt))
return false;
if (TREE_CODE (else_clause) == SSA_NAME)
{
- tree else_def_stmt = SSA_NAME_DEF_STMT (else_clause);
+ gimple else_def_stmt = SSA_NAME_DEF_STMT (else_clause);
if (!vect_is_simple_use (else_clause, loop_vinfo,
&else_def_stmt, &def, &dt))
return false;
/* Transform */
/* Handle def. */
- scalar_dest = GIMPLE_STMT_OPERAND (stmt, 0);
+ scalar_dest = gimple_assign_lhs (stmt);
vec_dest = vect_create_destination_var (scalar_dest, vectype);
/* Handle cond expr. */
vec_then_clause = vect_get_vec_def_for_operand (then_clause, stmt, NULL);
vec_else_clause = vect_get_vec_def_for_operand (else_clause, stmt, NULL);
- /* Arguments are ready. create the new vector stmt. */
+ /* Arguments are ready. Create the new vector stmt. */
vec_compare = build2 (TREE_CODE (cond_expr), vectype,
vec_cond_lhs, vec_cond_rhs);
vec_cond_expr = build3 (VEC_COND_EXPR, vectype,
vec_compare, vec_then_clause, vec_else_clause);
- *vec_stmt = build_gimple_modify_stmt (vec_dest, vec_cond_expr);
+ *vec_stmt = gimple_build_assign (vec_dest, vec_cond_expr);
new_temp = make_ssa_name (vec_dest, *vec_stmt);
- GIMPLE_STMT_OPERAND (*vec_stmt, 0) = new_temp;
- vect_finish_stmt_generation (stmt, *vec_stmt, bsi);
+ gimple_assign_set_lhs (*vec_stmt, new_temp);
+ vect_finish_stmt_generation (stmt, *vec_stmt, gsi);
return true;
}
Create a vectorized stmt to replace STMT, and insert it at BSI. */
static bool
-vect_transform_stmt (tree stmt, block_stmt_iterator *bsi, bool *strided_store,
- slp_tree slp_node)
+vect_transform_stmt (gimple stmt, gimple_stmt_iterator *gsi,
+ bool *strided_store, slp_tree slp_node)
{
bool is_store = false;
- tree vec_stmt = NULL_TREE;
+ gimple vec_stmt = NULL;
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
- tree orig_stmt_in_pattern;
+ gimple orig_stmt_in_pattern;
bool done;
switch (STMT_VINFO_TYPE (stmt_info))
{
case type_demotion_vec_info_type:
- gcc_assert (!slp_node);
- done = vectorizable_type_demotion (stmt, bsi, &vec_stmt);
+ done = vectorizable_type_demotion (stmt, gsi, &vec_stmt, slp_node);
gcc_assert (done);
break;
case type_promotion_vec_info_type:
- gcc_assert (!slp_node);
- done = vectorizable_type_promotion (stmt, bsi, &vec_stmt);
+ done = vectorizable_type_promotion (stmt, gsi, &vec_stmt, slp_node);
gcc_assert (done);
break;
case type_conversion_vec_info_type:
- done = vectorizable_conversion (stmt, bsi, &vec_stmt, slp_node);
+ done = vectorizable_conversion (stmt, gsi, &vec_stmt, slp_node);
gcc_assert (done);
break;
case induc_vec_info_type:
gcc_assert (!slp_node);
- done = vectorizable_induction (stmt, bsi, &vec_stmt);
+ done = vectorizable_induction (stmt, gsi, &vec_stmt);
gcc_assert (done);
break;
case op_vec_info_type:
- done = vectorizable_operation (stmt, bsi, &vec_stmt, slp_node);
+ done = vectorizable_operation (stmt, gsi, &vec_stmt, slp_node);
gcc_assert (done);
break;
case assignment_vec_info_type:
- done = vectorizable_assignment (stmt, bsi, &vec_stmt, slp_node);
+ done = vectorizable_assignment (stmt, gsi, &vec_stmt, slp_node);
gcc_assert (done);
break;
case load_vec_info_type:
- done = vectorizable_load (stmt, bsi, &vec_stmt, slp_node);
+ done = vectorizable_load (stmt, gsi, &vec_stmt, slp_node);
gcc_assert (done);
break;
case store_vec_info_type:
- done = vectorizable_store (stmt, bsi, &vec_stmt, slp_node);
+ done = vectorizable_store (stmt, gsi, &vec_stmt, slp_node);
gcc_assert (done);
- if (STMT_VINFO_STRIDED_ACCESS (stmt_info))
+ if (STMT_VINFO_STRIDED_ACCESS (stmt_info) && !slp_node)
{
/* In case of interleaving, the whole chain is vectorized when the
last store in the chain is reached. Store stmts before the last
case condition_vec_info_type:
gcc_assert (!slp_node);
- done = vectorizable_condition (stmt, bsi, &vec_stmt);
+ done = vectorizable_condition (stmt, gsi, &vec_stmt);
gcc_assert (done);
break;
case call_vec_info_type:
gcc_assert (!slp_node);
- done = vectorizable_call (stmt, bsi, &vec_stmt);
+ done = vectorizable_call (stmt, gsi, &vec_stmt);
break;
case reduc_vec_info_type:
gcc_assert (!slp_node);
- done = vectorizable_reduction (stmt, bsi, &vec_stmt);
+ done = vectorizable_reduction (stmt, gsi, &vec_stmt);
gcc_assert (done);
break;
if (STMT_VINFO_LIVE_P (stmt_info)
&& STMT_VINFO_TYPE (stmt_info) != reduc_vec_info_type)
{
- done = vectorizable_live_operation (stmt, bsi, &vec_stmt);
+ done = vectorizable_live_operation (stmt, gsi, &vec_stmt);
gcc_assert (done);
}
static tree
vect_build_loop_niters (loop_vec_info loop_vinfo)
{
- tree ni_name, stmt, var;
+ tree ni_name, var;
+ gimple_seq stmts = NULL;
edge pe;
struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
tree ni = unshare_expr (LOOP_VINFO_NITERS (loop_vinfo));
var = create_tmp_var (TREE_TYPE (ni), "niters");
add_referenced_var (var);
- ni_name = force_gimple_operand (ni, &stmt, false, var);
+ ni_name = force_gimple_operand (ni, &stmts, false, var);
pe = loop_preheader_edge (loop);
- if (stmt)
+ if (stmts)
{
- basic_block new_bb = bsi_insert_on_edge_immediate (pe, stmt);
+ basic_block new_bb = gsi_insert_seq_on_edge_immediate (pe, stmts);
gcc_assert (!new_bb);
}
edge pe;
basic_block new_bb;
- tree stmt, ni_name;
+ gimple_seq stmts;
+ tree ni_name;
tree var;
tree ratio_name;
tree ratio_mult_vf_name;
var = create_tmp_var (TREE_TYPE (ni), "bnd");
add_referenced_var (var);
- ratio_name = force_gimple_operand (ratio_name, &stmt, true, var);
+ stmts = NULL;
+ ratio_name = force_gimple_operand (ratio_name, &stmts, true, var);
pe = loop_preheader_edge (loop);
- new_bb = bsi_insert_on_edge_immediate (pe, stmt);
+ new_bb = gsi_insert_seq_on_edge_immediate (pe, stmts);
gcc_assert (!new_bb);
}
var = create_tmp_var (TREE_TYPE (ni), "ratio_mult_vf");
add_referenced_var (var);
- ratio_mult_vf_name = force_gimple_operand (ratio_mult_vf_name, &stmt,
+ stmts = NULL;
+ ratio_mult_vf_name = force_gimple_operand (ratio_mult_vf_name, &stmts,
true, var);
pe = loop_preheader_edge (loop);
- new_bb = bsi_insert_on_edge_immediate (pe, stmt);
+ new_bb = gsi_insert_seq_on_edge_immediate (pe, stmts);
gcc_assert (!new_bb);
}
{
struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
basic_block exit_bb = single_exit (loop)->dest;
- tree phi, phi1;
+ gimple phi, phi1;
+ gimple_stmt_iterator gsi, gsi1;
basic_block update_bb = update_e->dest;
/* gcc_assert (vect_can_advance_ivs_p (loop_vinfo)); */
/* Make sure there exists a single-predecessor exit bb: */
gcc_assert (single_pred_p (exit_bb));
- for (phi = phi_nodes (loop->header), phi1 = phi_nodes (update_bb);
- phi && phi1;
- phi = PHI_CHAIN (phi), phi1 = PHI_CHAIN (phi1))
+ for (gsi = gsi_start_phis (loop->header), gsi1 = gsi_start_phis (update_bb);
+ !gsi_end_p (gsi) && !gsi_end_p (gsi1);
+ gsi_next (&gsi), gsi_next (&gsi1))
{
tree access_fn = NULL;
tree evolution_part;
tree init_expr;
tree step_expr;
tree var, ni, ni_name;
- block_stmt_iterator last_bsi;
+ gimple_stmt_iterator last_gsi;
+ phi = gsi_stmt (gsi);
+ phi1 = gsi_stmt (gsi1);
if (vect_print_dump_info (REPORT_DETAILS))
{
fprintf (vect_dump, "vect_update_ivs_after_vectorizer: phi: ");
- print_generic_expr (vect_dump, phi, TDF_SLIM);
+ print_gimple_stmt (vect_dump, phi, 0, TDF_SLIM);
}
/* Skip virtual phi's. */
var = create_tmp_var (TREE_TYPE (init_expr), "tmp");
add_referenced_var (var);
- last_bsi = bsi_last (exit_bb);
- ni_name = force_gimple_operand_bsi (&last_bsi, ni, false, var,
- true, BSI_SAME_STMT);
+ last_gsi = gsi_last_bb (exit_bb);
+ ni_name = force_gimple_operand_gsi (&last_gsi, ni, false, var,
+ true, GSI_SAME_STMT);
/* Fix phi expressions in the successor bb. */
SET_PHI_ARG_DEF (phi1, update_e->dest_idx, ni_name);
/* If cost model check not done during versioning and
peeling for alignment. */
- if (!VEC_length (tree, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo))
+ if (!VEC_length (gimple, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo))
&& !VEC_length (ddr_p, LOOP_VINFO_MAY_ALIAS_DDRS (loop_vinfo))
&& !LOOP_PEELING_FOR_ALIGNMENT (loop_vinfo))
{
Else, compute address misalignment in bytes:
addr_mis = addr & (vectype_size - 1)
- prolog_niters = min ( LOOP_NITERS , (VF - addr_mis/elem_size)&(VF-1) )
-
- (elem_size = element type size; an element is the scalar element
- whose type is the inner type of the vectype)
+ prolog_niters = min (LOOP_NITERS, ((VF - addr_mis/elem_size)&(VF-1))/step)
- For interleaving,
+ (elem_size = element type size; an element is the scalar element whose type
+ is the inner type of the vectype)
- prolog_niters = min ( LOOP_NITERS ,
- (VF/group_size - addr_mis/elem_size)&(VF/group_size-1) )
- where group_size is the size of the interleaved group.
+ When the step of the data-ref in the loop is not 1 (as in interleaved data
+ and SLP), the number of iterations of the prolog must be divided by the step
+ (which is equal to the size of interleaved group).
The above formulas assume that VF == number of elements in the vector. This
may not hold when there are multiple-types in the loop.
{
struct data_reference *dr = LOOP_VINFO_UNALIGNED_DR (loop_vinfo);
struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
- tree var, stmt;
+ tree var;
+ gimple_seq stmts;
tree iters, iters_name;
edge pe;
basic_block new_bb;
- tree dr_stmt = DR_STMT (dr);
+ gimple dr_stmt = DR_STMT (dr);
stmt_vec_info stmt_info = vinfo_for_stmt (dr_stmt);
tree vectype = STMT_VINFO_VECTYPE (stmt_info);
int vectype_align = TYPE_ALIGN (vectype) / BITS_PER_UNIT;
tree niters_type = TREE_TYPE (loop_niters);
- int group_size = 1;
+ int step = 1;
int element_size = GET_MODE_SIZE (TYPE_MODE (TREE_TYPE (DR_REF (dr))));
int nelements = TYPE_VECTOR_SUBPARTS (vectype);
if (STMT_VINFO_STRIDED_ACCESS (stmt_info))
- {
- /* For interleaved access element size must be multiplied by the size of
- the interleaved group. */
- group_size = DR_GROUP_SIZE (vinfo_for_stmt (
- DR_GROUP_FIRST_DR (stmt_info)));
- element_size *= group_size;
- }
+ step = DR_GROUP_SIZE (vinfo_for_stmt (DR_GROUP_FIRST_DR (stmt_info)));
pe = loop_preheader_edge (loop);
if (vect_print_dump_info (REPORT_DETAILS))
fprintf (vect_dump, "known alignment = %d.", byte_misalign);
- iters = build_int_cst (niters_type,
- (nelements - elem_misalign)&(nelements/group_size-1));
+
+ iters = build_int_cst (niters_type,
+ (((nelements - elem_misalign) & (nelements - 1)) / step));
}
else
{
- tree new_stmts = NULL_TREE;
+ gimple_seq new_stmts = NULL;
tree start_addr = vect_create_addr_base_for_vector_ref (dr_stmt,
&new_stmts, NULL_TREE, loop);
tree ptr_type = TREE_TYPE (start_addr);
tree byte_misalign;
tree elem_misalign;
- new_bb = bsi_insert_on_edge_immediate (pe, new_stmts);
+ new_bb = gsi_insert_seq_on_edge_immediate (pe, new_stmts);
gcc_assert (!new_bb);
/* Create: byte_misalign = addr & (vectype_size - 1) */
var = create_tmp_var (niters_type, "prolog_loop_niters");
add_referenced_var (var);
- iters_name = force_gimple_operand (iters, &stmt, false, var);
+ stmts = NULL;
+ iters_name = force_gimple_operand (iters, &stmts, false, var);
/* Insert stmt on loop preheader edge. */
- if (stmt)
+ if (stmts)
{
- basic_block new_bb = bsi_insert_on_edge_immediate (pe, stmt);
+ basic_block new_bb = gsi_insert_seq_on_edge_immediate (pe, stmts);
gcc_assert (!new_bb);
}
/* If cost model check not done during versioning. */
- if (!VEC_length (tree, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo))
+ if (!VEC_length (gimple, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo))
&& !VEC_length (ddr_p, LOOP_VINFO_MAY_ALIAS_DDRS (loop_vinfo)))
{
check_profitability = true;
static void
vect_create_cond_for_align_checks (loop_vec_info loop_vinfo,
tree *cond_expr,
- tree *cond_expr_stmt_list)
+ gimple_seq *cond_expr_stmt_list)
{
struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
- VEC(tree,heap) *may_misalign_stmts
+ VEC(gimple,heap) *may_misalign_stmts
= LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo);
- tree ref_stmt, tmp;
+ gimple ref_stmt;
int mask = LOOP_VINFO_PTR_MASK (loop_vinfo);
tree mask_cst;
unsigned int i;
tree int_ptrsize_type;
char tmp_name[20];
tree or_tmp_name = NULL_TREE;
- tree and_tmp, and_tmp_name, and_stmt;
+ tree and_tmp, and_tmp_name;
+ gimple and_stmt;
tree ptrsize_zero;
tree part_cond_expr;
/* Create expression (mask & (dr_1 || ... || dr_n)) where dr_i is the address
of the first vector of the i'th data reference. */
- for (i = 0; VEC_iterate (tree, may_misalign_stmts, i, ref_stmt); i++)
+ for (i = 0; VEC_iterate (gimple, may_misalign_stmts, i, ref_stmt); i++)
{
- tree new_stmt_list = NULL_TREE;
+ gimple_seq new_stmt_list = NULL;
tree addr_base;
- tree addr_tmp, addr_tmp_name, addr_stmt;
- tree or_tmp, new_or_tmp_name, or_stmt;
+ tree addr_tmp, addr_tmp_name;
+ tree or_tmp, new_or_tmp_name;
+ gimple addr_stmt, or_stmt;
/* create: addr_tmp = (int)(address_of_first_vector) */
- addr_base = vect_create_addr_base_for_vector_ref (ref_stmt,
- &new_stmt_list, NULL_TREE, loop);
-
- if (new_stmt_list != NULL_TREE)
- append_to_statement_list_force (new_stmt_list, cond_expr_stmt_list);
+ addr_base =
+ vect_create_addr_base_for_vector_ref (ref_stmt, &new_stmt_list,
+ NULL_TREE, loop);
+ if (new_stmt_list != NULL)
+ gimple_seq_add_seq (cond_expr_stmt_list, new_stmt_list);
sprintf (tmp_name, "%s%d", "addr2int", i);
addr_tmp = create_tmp_var (int_ptrsize_type, tmp_name);
add_referenced_var (addr_tmp);
- addr_tmp_name = make_ssa_name (addr_tmp, NULL_TREE);
- addr_stmt = fold_convert (int_ptrsize_type, addr_base);
- addr_stmt = build_gimple_modify_stmt (addr_tmp_name, addr_stmt);
+ addr_tmp_name = make_ssa_name (addr_tmp, NULL);
+ addr_stmt = gimple_build_assign_with_ops (NOP_EXPR, addr_tmp_name,
+ addr_base, NULL_TREE);
SSA_NAME_DEF_STMT (addr_tmp_name) = addr_stmt;
- append_to_statement_list_force (addr_stmt, cond_expr_stmt_list);
+ gimple_seq_add_stmt (cond_expr_stmt_list, addr_stmt);
/* The addresses are OR together. */
sprintf (tmp_name, "%s%d", "orptrs", i);
or_tmp = create_tmp_var (int_ptrsize_type, tmp_name);
add_referenced_var (or_tmp);
- new_or_tmp_name = make_ssa_name (or_tmp, NULL_TREE);
- tmp = build2 (BIT_IOR_EXPR, int_ptrsize_type,
- or_tmp_name, addr_tmp_name);
- or_stmt = build_gimple_modify_stmt (new_or_tmp_name, tmp);
+ new_or_tmp_name = make_ssa_name (or_tmp, NULL);
+ or_stmt = gimple_build_assign_with_ops (BIT_IOR_EXPR,
+ new_or_tmp_name,
+ or_tmp_name, addr_tmp_name);
SSA_NAME_DEF_STMT (new_or_tmp_name) = or_stmt;
- append_to_statement_list_force (or_stmt, cond_expr_stmt_list);
+ gimple_seq_add_stmt (cond_expr_stmt_list, or_stmt);
or_tmp_name = new_or_tmp_name;
}
else
/* create: and_tmp = or_tmp & mask */
and_tmp = create_tmp_var (int_ptrsize_type, "andmask" );
add_referenced_var (and_tmp);
- and_tmp_name = make_ssa_name (and_tmp, NULL_TREE);
+ and_tmp_name = make_ssa_name (and_tmp, NULL);
- tmp = build2 (BIT_AND_EXPR, int_ptrsize_type, or_tmp_name, mask_cst);
- and_stmt = build_gimple_modify_stmt (and_tmp_name, tmp);
+ and_stmt = gimple_build_assign_with_ops (BIT_AND_EXPR, and_tmp_name,
+ or_tmp_name, mask_cst);
SSA_NAME_DEF_STMT (and_tmp_name) = and_stmt;
- append_to_statement_list_force (and_stmt, cond_expr_stmt_list);
+ gimple_seq_add_stmt (cond_expr_stmt_list, and_stmt);
/* Make and_tmp the left operand of the conditional test against zero.
if and_tmp has a nonzero bit then some address is unaligned. */
static void
vect_create_cond_for_alias_checks (loop_vec_info loop_vinfo,
tree * cond_expr,
- tree * cond_expr_stmt_list)
+ gimple_seq * cond_expr_stmt_list)
{
struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
VEC (ddr_p, heap) * may_alias_ddrs =
for (i = 0; VEC_iterate (ddr_p, may_alias_ddrs, i, ddr); i++)
{
struct data_reference *dr_a, *dr_b;
- tree dr_group_first_a, dr_group_first_b;
+ gimple dr_group_first_a, dr_group_first_b;
tree addr_base_a, addr_base_b;
tree segment_length_a, segment_length_b;
- tree stmt_a, stmt_b;
+ gimple stmt_a, stmt_b;
dr_a = DDR_A (ddr);
stmt_a = DR_STMT (DDR_A (ddr));
if (*cond_expr)
*cond_expr = fold_build2 (TRUTH_AND_EXPR, boolean_type_node,
- *cond_expr, part_cond_expr);
+ *cond_expr, part_cond_expr);
else
*cond_expr = part_cond_expr;
}
struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
struct loop *nloop;
tree cond_expr = NULL_TREE;
- tree cond_expr_stmt_list = NULL_TREE;
+ gimple_seq cond_expr_stmt_list = NULL;
basic_block condition_bb;
- block_stmt_iterator cond_exp_bsi;
+ gimple_stmt_iterator gsi, cond_exp_gsi;
basic_block merge_bb;
basic_block new_exit_bb;
edge new_exit_e, e;
- tree orig_phi, new_phi, arg;
+ gimple orig_phi, new_phi;
+ tree arg;
unsigned prob = 4 * REG_BR_PROB_BASE / 5;
- tree gimplify_stmt_list;
+ gimple_seq gimplify_stmt_list = NULL;
tree scalar_loop_iters = LOOP_VINFO_NITERS (loop_vinfo);
int min_profitable_iters = 0;
unsigned int th;
cond_expr = force_gimple_operand (cond_expr, &cond_expr_stmt_list,
false, NULL_TREE);
- if (VEC_length (tree, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo)))
+ if (VEC_length (gimple, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo)))
vect_create_cond_for_align_checks (loop_vinfo, &cond_expr,
&cond_expr_stmt_list);
cond_expr =
fold_build2 (NE_EXPR, boolean_type_node, cond_expr, integer_zero_node);
cond_expr =
- force_gimple_operand (cond_expr, &gimplify_stmt_list, true,
- NULL_TREE);
- append_to_statement_list (gimplify_stmt_list, &cond_expr_stmt_list);
+ force_gimple_operand (cond_expr, &gimplify_stmt_list, true, NULL_TREE);
+ gimple_seq_add_seq (&cond_expr_stmt_list, gimplify_stmt_list);
initialize_original_copy_tables ();
nloop = loop_version (loop, cond_expr, &condition_bb,
new_exit_e = single_exit (loop);
e = EDGE_SUCC (new_exit_bb, 0);
- for (orig_phi = phi_nodes (merge_bb); orig_phi;
- orig_phi = PHI_CHAIN (orig_phi))
+ for (gsi = gsi_start_phis (merge_bb); !gsi_end_p (gsi); gsi_next (&gsi))
{
+ orig_phi = gsi_stmt (gsi);
new_phi = create_phi_node (SSA_NAME_VAR (PHI_RESULT (orig_phi)),
new_exit_bb);
arg = PHI_ARG_DEF_FROM_EDGE (orig_phi, e);
update_ssa (TODO_update_ssa);
if (cond_expr_stmt_list)
{
- cond_exp_bsi = bsi_last (condition_bb);
- bsi_insert_before (&cond_exp_bsi, cond_expr_stmt_list, BSI_SAME_STMT);
+ cond_exp_gsi = gsi_last_bb (condition_bb);
+ gsi_insert_seq_before (&cond_exp_gsi, cond_expr_stmt_list, GSI_SAME_STMT);
}
}
stmt_vec_info. */
static void
-vect_remove_stores (tree first_stmt)
+vect_remove_stores (gimple first_stmt)
{
- tree next = first_stmt;
- tree tmp;
- block_stmt_iterator next_si;
+ gimple next = first_stmt;
+ gimple tmp;
+ gimple_stmt_iterator next_si;
while (next)
{
/* Free the attached stmt_vec_info and remove the stmt. */
- next_si = bsi_for_stmt (next);
- bsi_remove (&next_si, true);
+ next_si = gsi_for_stmt (next);
+ gsi_remove (&next_si, true);
tmp = DR_GROUP_NEXT_DR (vinfo_for_stmt (next));
free_stmt_vec_info (next);
next = tmp;
/* Vectorize SLP instance tree in postorder. */
static bool
-vect_schedule_slp_instance (slp_tree node, unsigned int vec_stmts_size)
+vect_schedule_slp_instance (slp_tree node, slp_instance instance,
+ unsigned int vectorization_factor)
{
- tree stmt;
+ gimple stmt;
bool strided_store, is_store;
- block_stmt_iterator si;
+ gimple_stmt_iterator si;
stmt_vec_info stmt_info;
+ unsigned int vec_stmts_size, nunits, group_size;
+ tree vectype;
if (!node)
return false;
- vect_schedule_slp_instance (SLP_TREE_LEFT (node), vec_stmts_size);
- vect_schedule_slp_instance (SLP_TREE_RIGHT (node), vec_stmts_size);
+ vect_schedule_slp_instance (SLP_TREE_LEFT (node), instance,
+ vectorization_factor);
+ vect_schedule_slp_instance (SLP_TREE_RIGHT (node), instance,
+ vectorization_factor);
- stmt = VEC_index(tree, SLP_TREE_SCALAR_STMTS (node), 0);
+ stmt = VEC_index (gimple, SLP_TREE_SCALAR_STMTS (node), 0);
stmt_info = vinfo_for_stmt (stmt);
- SLP_TREE_VEC_STMTS (node) = VEC_alloc (tree, heap, vec_stmts_size);
+ /* VECTYPE is the type of the destination. */
+ vectype = get_vectype_for_scalar_type (TREE_TYPE (gimple_assign_lhs (stmt)));
+ nunits = (unsigned int) TYPE_VECTOR_SUBPARTS (vectype);
+ group_size = SLP_INSTANCE_GROUP_SIZE (instance);
+
+ /* For each SLP instance calculate number of vector stmts to be created
+ for the scalar stmts in each node of the SLP tree. Number of vector
+ elements in one vector iteration is the number of scalar elements in
+ one scalar iteration (GROUP_SIZE) multiplied by VF divided by vector
+ size. */
+ vec_stmts_size = (vectorization_factor * group_size) / nunits;
+
+ SLP_TREE_VEC_STMTS (node) = VEC_alloc (gimple, heap, vec_stmts_size);
SLP_TREE_NUMBER_OF_VEC_STMTS (node) = vec_stmts_size;
if (vect_print_dump_info (REPORT_DETAILS))
{
fprintf (vect_dump, "------>vectorizing SLP node starting from: ");
- print_generic_expr (vect_dump, stmt, TDF_SLIM);
+ print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM);
}
- si = bsi_for_stmt (stmt);
+ si = gsi_for_stmt (stmt);
is_store = vect_transform_stmt (stmt, &si, &strided_store, node);
if (is_store)
{
static bool
-vect_schedule_slp (loop_vec_info loop_vinfo, unsigned int nunits)
+vect_schedule_slp (loop_vec_info loop_vinfo)
{
VEC (slp_instance, heap) *slp_instances =
LOOP_VINFO_SLP_INSTANCES (loop_vinfo);
slp_instance instance;
- unsigned int vec_stmts_size;
- unsigned int group_size, i;
- unsigned int vectorization_factor = LOOP_VINFO_VECT_FACTOR (loop_vinfo);
+ unsigned int i;
bool is_store = false;
for (i = 0; VEC_iterate (slp_instance, slp_instances, i, instance); i++)
{
- group_size = SLP_INSTANCE_GROUP_SIZE (instance);
- /* For each SLP instance calculate number of vector stmts to be created
- for the scalar stmts in each node of the SLP tree. Number of vector
- elements in one vector iteration is the number of scalar elements in
- one scalar iteration (GROUP_SIZE) multiplied by VF divided by vector
- size. */
- vec_stmts_size = vectorization_factor * group_size / nunits;
-
/* Schedule the tree of INSTANCE. */
- is_store = vect_schedule_slp_instance (SLP_INSTANCE_TREE (instance),
- vec_stmts_size);
-
+ is_store = vect_schedule_slp_instance (SLP_INSTANCE_TREE (instance),
+ instance,
+ LOOP_VINFO_VECT_FACTOR (loop_vinfo));
+
if (vect_print_dump_info (REPORT_VECTORIZED_LOOPS)
|| vect_print_dump_info (REPORT_UNVECTORIZED_LOOPS))
fprintf (vect_dump, "vectorizing stmts using SLP.");
struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
basic_block *bbs = LOOP_VINFO_BBS (loop_vinfo);
int nbbs = loop->num_nodes;
- block_stmt_iterator si;
+ gimple_stmt_iterator si;
int i;
tree ratio = NULL;
int vectorization_factor = LOOP_VINFO_VECT_FACTOR (loop_vinfo);
if (vect_print_dump_info (REPORT_DETAILS))
fprintf (vect_dump, "=== vec_transform_loop ===");
- if (VEC_length (tree, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo))
+ if (VEC_length (gimple, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo))
|| VEC_length (ddr_p, LOOP_VINFO_MAY_ALIAS_DDRS (loop_vinfo)))
vect_loop_versioning (loop_vinfo);
{
basic_block bb = bbs[i];
stmt_vec_info stmt_info;
- tree phi;
+ gimple phi;
- for (phi = phi_nodes (bb); phi; phi = PHI_CHAIN (phi))
+ for (si = gsi_start_phis (bb); !gsi_end_p (si); gsi_next (&si))
{
+ phi = gsi_stmt (si);
if (vect_print_dump_info (REPORT_DETAILS))
{
fprintf (vect_dump, "------>vectorizing phi: ");
- print_generic_expr (vect_dump, phi, TDF_SLIM);
+ print_gimple_stmt (vect_dump, phi, 0, TDF_SLIM);
}
stmt_info = vinfo_for_stmt (phi);
if (!stmt_info)
}
}
- for (si = bsi_start (bb); !bsi_end_p (si);)
+ for (si = gsi_start_bb (bb); !gsi_end_p (si);)
{
- tree stmt = bsi_stmt (si);
+ gimple stmt = gsi_stmt (si);
bool is_store;
if (vect_print_dump_info (REPORT_DETAILS))
{
fprintf (vect_dump, "------>vectorizing statement: ");
- print_generic_expr (vect_dump, stmt, TDF_SLIM);
+ print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM);
}
stmt_info = vinfo_for_stmt (stmt);
need to be vectorized. */
if (!stmt_info)
{
- bsi_next (&si);
+ gsi_next (&si);
continue;
}
if (!STMT_VINFO_RELEVANT_P (stmt_info)
&& !STMT_VINFO_LIVE_P (stmt_info))
{
- bsi_next (&si);
+ gsi_next (&si);
continue;
}
if (vect_print_dump_info (REPORT_DETAILS))
fprintf (vect_dump, "=== scheduling SLP instances ===");
- is_store = vect_schedule_slp (loop_vinfo, nunits);
+ is_store = vect_schedule_slp (loop_vinfo);
/* IS_STORE is true if STMT is a store. Stores cannot be of
hybrid SLP type. They are removed in
vect_schedule_slp_instance and their vinfo is destroyed. */
if (is_store)
{
- bsi_next (&si);
+ gsi_next (&si);
continue;
}
}
/* Hybrid SLP stmts must be vectorized in addition to SLP. */
if (PURE_SLP_STMT (stmt_info))
{
- bsi_next (&si);
+ gsi_next (&si);
continue;
}
}
interleaving chain was completed - free all the stores in
the chain. */
vect_remove_stores (DR_GROUP_FIRST_DR (stmt_info));
- bsi_remove (&si, true);
+ gsi_remove (&si, true);
continue;
}
else
{
/* Free the attached stmt_vec_info and remove the stmt. */
free_stmt_vec_info (stmt);
- bsi_remove (&si, true);
+ gsi_remove (&si, true);
continue;
}
}
- bsi_next (&si);
+ gsi_next (&si);
} /* stmts in BB */
} /* BBs in loop */
OSDN Git Service
