#include "real.h"
/* Utility functions for the code transformation. */
-static bool vect_transform_stmt (tree, block_stmt_iterator *);
+static bool vect_transform_stmt (tree, block_stmt_iterator *, bool *);
static tree vect_create_destination_var (tree, tree);
static tree vect_create_data_ref_ptr
- (tree, block_stmt_iterator *, tree, tree *, tree *, bool);
+ (tree, block_stmt_iterator *, tree, tree *, tree *, bool, tree);
static tree vect_create_addr_base_for_vector_ref (tree, tree *, tree);
static tree vect_setup_realignment (tree, block_stmt_iterator *, tree *);
static tree vect_get_new_vect_var (tree, enum vect_var_kind, const char *);
static tree vect_get_vec_def_for_operand (tree, tree, tree *);
-static tree vect_init_vector (tree, tree);
+static tree vect_init_vector (tree, tree, tree);
static void vect_finish_stmt_generation
(tree stmt, tree vec_stmt, block_stmt_iterator *bsi);
static bool vect_is_simple_cond (tree, loop_vec_info);
static tree vect_gen_niters_for_prolog_loop (loop_vec_info, tree);
static void vect_update_init_of_dr (struct data_reference *, tree niters);
static void vect_update_inits_of_drs (loop_vec_info, tree);
-static void vect_do_peeling_for_alignment (loop_vec_info, struct loops *);
-static void vect_do_peeling_for_loop_bound
- (loop_vec_info, tree *, struct loops *);
static int vect_min_worthwhile_factor (enum tree_code);
struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info);
tree data_ref_base = unshare_expr (DR_BASE_ADDRESS (dr));
tree base_name = build_fold_indirect_ref (data_ref_base);
- tree ref = DR_REF (dr);
- tree scalar_type = TREE_TYPE (ref);
- tree scalar_ptr_type = build_pointer_type (scalar_type);
tree vec_stmt;
- tree new_temp;
tree addr_base, addr_expr;
tree dest, new_stmt;
tree base_offset = unshare_expr (DR_OFFSET (dr));
tree init = unshare_expr (DR_INIT (dr));
+ tree vect_ptr_type, addr_expr2;
/* Create base_offset */
base_offset = size_binop (PLUS_EXPR, base_offset, init);
if (offset)
{
tree tmp = create_tmp_var (TREE_TYPE (base_offset), "offset");
+ tree step;
+
+ /* For interleaved access step we divide STEP by the size of the
+ interleaving group. */
+ if (DR_GROUP_SIZE (stmt_info))
+ step = fold_build2 (TRUNC_DIV_EXPR, TREE_TYPE (offset), DR_STEP (dr),
+ build_int_cst (TREE_TYPE (offset),
+ DR_GROUP_SIZE (stmt_info)));
+ else
+ step = DR_STEP (dr);
+
add_referenced_var (tmp);
- offset = fold_build2 (MULT_EXPR, TREE_TYPE (offset), offset,
- DR_STEP (dr));
+ 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);
addr_base = fold_build2 (PLUS_EXPR, TREE_TYPE (data_ref_base), data_ref_base,
base_offset);
+ vect_ptr_type = build_pointer_type (STMT_VINFO_VECTYPE (stmt_info));
+
/* addr_expr = addr_base */
- addr_expr = vect_get_new_vect_var (scalar_ptr_type, vect_pointer_var,
+ addr_expr = vect_get_new_vect_var (vect_ptr_type, vect_pointer_var,
get_name (base_name));
add_referenced_var (addr_expr);
- vec_stmt = build2 (MODIFY_EXPR, void_type_node, addr_expr, addr_base);
- new_temp = make_ssa_name (addr_expr, vec_stmt);
- TREE_OPERAND (vec_stmt, 0) = new_temp;
- append_to_statement_list_force (vec_stmt, new_stmt_list);
+ vec_stmt = fold_convert (vect_ptr_type, addr_base);
+ 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);
if (vect_print_dump_info (REPORT_DETAILS))
{
fprintf (vect_dump, "created ");
print_generic_expr (vect_dump, vec_stmt, TDF_SLIM);
}
- return new_temp;
+ return vec_stmt;
}
Input:
1. STMT: a stmt that references memory. Expected to be of the form
- MODIFY_EXPR <name, data-ref> or MODIFY_EXPR <data-ref, name>.
+ GIMPLE_MODIFY_STMT <name, data-ref> or
+ GIMPLE_MODIFY_STMT <data-ref, name>.
2. BSI: block_stmt_iterator where new stmts can be added.
3. OFFSET (optional): an offset to be added to the initial address accessed
by the data-ref in STMT.
4. ONLY_INIT: indicate if vp is to be updated in the loop, or remain
pointing to the initial address.
+ 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
vect_create_data_ref_ptr (tree stmt,
block_stmt_iterator *bsi ATTRIBUTE_UNUSED,
tree offset, tree *initial_address, tree *ptr_incr,
- bool only_init)
+ bool only_init, tree type)
{
tree base_name;
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
}
/** (1) Create the new vector-pointer variable: **/
-
- vect_ptr_type = build_pointer_type (vectype);
+ if (type)
+ vect_ptr_type = build_pointer_type (type);
+ else
+ vect_ptr_type = build_pointer_type (vectype);
vect_ptr = vect_get_new_vect_var (vect_ptr_type, vect_pointer_var,
get_name (base_name));
add_referenced_var (vect_ptr);
-
-
+
/** (2) Add aliasing information to the new vector-pointer:
(The points-to info (DR_PTR_INFO) may be defined later.) **/
/* Create: p = (vectype *) initial_base */
vec_stmt = fold_convert (vect_ptr_type, new_temp);
- vec_stmt = build2 (MODIFY_EXPR, void_type_node, vect_ptr, vec_stmt);
+ vec_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node, vect_ptr, vec_stmt);
vect_ptr_init = make_ssa_name (vect_ptr, vec_stmt);
- TREE_OPERAND (vec_stmt, 0) = vect_ptr_init;
+ GIMPLE_STMT_OPERAND (vec_stmt, 0) = vect_ptr_init;
new_bb = bsi_insert_on_edge_immediate (pe, vec_stmt);
gcc_assert (!new_bb);
/* Function bump_vector_ptr
Increment a pointer (to a vector type) by vector-size. Connect the new
- increment stmt to the exising def-use update-chain of the pointer.
+ increment stmt to the existing def-use update-chain of the pointer.
The pointer def-use update-chain before this function:
DATAREF_PTR = phi (p_0, p_2)
use_operand_p use_p;
tree new_dataref_ptr;
- incr_stmt = build2 (MODIFY_EXPR, void_type_node, ptr_var,
+ incr_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node, ptr_var,
build2 (PLUS_EXPR, vptr_type, dataref_ptr, update));
new_dataref_ptr = make_ssa_name (ptr_var, incr_stmt);
- TREE_OPERAND (incr_stmt, 0) = new_dataref_ptr;
+ GIMPLE_STMT_OPERAND (incr_stmt, 0) = new_dataref_ptr;
vect_finish_stmt_generation (stmt, incr_stmt, bsi);
/* Update the vector-pointer's cross-iteration increment. */
used in the vectorization of STMT. */
static tree
-vect_init_vector (tree stmt, tree vector_var)
+vect_init_vector (tree stmt, tree vector_var, tree vector_type)
{
stmt_vec_info stmt_vinfo = vinfo_for_stmt (stmt);
loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_vinfo);
struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
tree new_var;
tree init_stmt;
- tree vectype = STMT_VINFO_VECTYPE (stmt_vinfo);
tree vec_oprnd;
edge pe;
tree new_temp;
basic_block new_bb;
- new_var = vect_get_new_vect_var (vectype, vect_simple_var, "cst_");
+ new_var = vect_get_new_vect_var (vector_type, vect_simple_var, "cst_");
add_referenced_var (new_var);
- init_stmt = build2 (MODIFY_EXPR, vectype, new_var, vector_var);
+ init_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node, new_var, vector_var);
new_temp = make_ssa_name (new_var, init_stmt);
- TREE_OPERAND (init_stmt, 0) = new_temp;
+ GIMPLE_STMT_OPERAND (init_stmt, 0) = new_temp;
pe = loop_preheader_edge (loop);
new_bb = bsi_insert_on_edge_immediate (pe, init_stmt);
print_generic_expr (vect_dump, init_stmt, TDF_SLIM);
}
- vec_oprnd = TREE_OPERAND (init_stmt, 0);
+ vec_oprnd = GIMPLE_STMT_OPERAND (init_stmt, 0);
return vec_oprnd;
}
int i;
enum vect_def_type dt;
bool is_simple_use;
+ tree vector_type;
if (vect_print_dump_info (REPORT_DETAILS))
{
{
t = tree_cons (NULL_TREE, op, t);
}
- vec_cst = build_vector (vectype, t);
- return vect_init_vector (stmt, vec_cst);
+ vector_type = get_vectype_for_scalar_type (TREE_TYPE (op));
+ vec_cst = build_vector (vector_type, t);
+
+ return vect_init_vector (stmt, vec_cst, vector_type);
}
/* Case 2: operand is defined outside the loop - loop invariant. */
}
/* FIXME: use build_constructor directly. */
- vec_inv = build_constructor_from_list (vectype, t);
- return vect_init_vector (stmt, vec_inv);
+ vector_type = get_vectype_for_scalar_type (TREE_TYPE (def));
+ vec_inv = build_constructor_from_list (vector_type, t);
+ return vect_init_vector (stmt, vec_inv, vector_type);
}
/* Case 3: operand is defined inside the loop. */
def_stmt_info = vinfo_for_stmt (def_stmt);
vec_stmt = STMT_VINFO_VEC_STMT (def_stmt_info);
gcc_assert (vec_stmt);
- vec_oprnd = TREE_OPERAND (vec_stmt, 0);
+ vec_oprnd = GIMPLE_STMT_OPERAND (vec_stmt, 0);
return vec_oprnd;
}
stmts operating on wider types we need to create 'VF/nunits' "copies" of the
vector stmt (each computing a vector of 'nunits' results, and together
computing 'VF' results in each iteration). This function is called when
- vectorizing such a stmt (e.g. vectorizing S2 in the illusration below, in
+ vectorizing such a stmt (e.g. vectorizing S2 in the illustration below, in
which VF=16 and nuniti=4, so the number of copies required is 4):
scalar stmt: vectorized into: STMT_VINFO_RELATED_STMT
gcc_assert (def_stmt_info);
vec_stmt_for_operand = STMT_VINFO_RELATED_STMT (def_stmt_info);
gcc_assert (vec_stmt_for_operand);
- vec_oprnd = TREE_OPERAND (vec_stmt_for_operand, 0);
+ vec_oprnd = GIMPLE_STMT_OPERAND (vec_stmt_for_operand, 0);
return vec_oprnd;
}
tree vectype = STMT_VINFO_VECTYPE (stmt_vinfo);
int nunits = GET_MODE_NUNITS (TYPE_MODE (vectype));
int nelements;
- enum tree_code code = TREE_CODE (TREE_OPERAND (stmt, 1));
+ enum tree_code code = TREE_CODE (GIMPLE_STMT_OPERAND (stmt, 1));
tree type = TREE_TYPE (init_val);
tree def;
tree vec, t = NULL_TREE;
bool need_epilog_adjust;
int i;
+ tree vector_type;
gcc_assert (INTEGRAL_TYPE_P (type) || SCALAR_FLOAT_TYPE_P (type));
nelements += 1;
}
gcc_assert (nelements == nunits);
-
+
+ vector_type = get_vectype_for_scalar_type (TREE_TYPE (def));
if (TREE_CODE (init_val) == INTEGER_CST || TREE_CODE (init_val) == REAL_CST)
- vec = build_vector (vectype, t);
+ vec = build_vector (vector_type, t);
else
- vec = build_constructor_from_list (vectype, t);
+ vec = build_constructor_from_list (vector_type, t);
if (!need_epilog_adjust)
*scalar_def = NULL_TREE;
else
*scalar_def = init_val;
- return vect_init_vector (stmt, vec);
+ return vect_init_vector (stmt, vec, vector_type);
}
tree epilog_stmt;
tree new_scalar_dest, exit_phi;
tree bitsize, bitpos, bytesize;
- enum tree_code code = TREE_CODE (TREE_OPERAND (stmt, 1));
+ enum tree_code code = TREE_CODE (GIMPLE_STMT_OPERAND (stmt, 1));
tree scalar_initial_def;
tree vec_initial_def;
tree orig_name;
tree reduction_op;
tree orig_stmt;
tree use_stmt;
- tree operation = TREE_OPERAND (stmt, 1);
+ tree operation = GIMPLE_STMT_OPERAND (stmt, 1);
int op_type;
op_type = TREE_CODE_LENGTH (TREE_CODE (operation));
/* 2.1 Create new loop-exit-phi to preserve loop-closed form:
v_out1 = phi <v_loop> */
- exit_bb = loop->single_exit->dest;
+ exit_bb = single_exit (loop)->dest;
new_phi = create_phi_node (SSA_NAME_VAR (vect_def), exit_bb);
- SET_PHI_ARG_DEF (new_phi, loop->single_exit->dest_idx, vect_def);
+ SET_PHI_ARG_DEF (new_phi, single_exit (loop)->dest_idx, vect_def);
exit_bsi = bsi_start (exit_bb);
/* 2.2 Get the relevant tree-code to use in the epilog for schemes 2,3
gcc_assert (STMT_VINFO_IN_PATTERN_P (stmt_vinfo));
gcc_assert (STMT_VINFO_RELATED_STMT (stmt_vinfo) == stmt);
}
- code = TREE_CODE (TREE_OPERAND (orig_stmt, 1));
- scalar_dest = TREE_OPERAND (orig_stmt, 0);
+ code = TREE_CODE (GIMPLE_STMT_OPERAND (orig_stmt, 1));
+ scalar_dest = GIMPLE_STMT_OPERAND (orig_stmt, 0);
scalar_type = TREE_TYPE (scalar_dest);
new_scalar_dest = vect_create_destination_var (scalar_dest, NULL);
bitsize = TYPE_SIZE (scalar_type);
fprintf (vect_dump, "Reduce using direct vector reduction.");
vec_dest = vect_create_destination_var (scalar_dest, vectype);
- epilog_stmt = build2 (MODIFY_EXPR, vectype, vec_dest,
+ epilog_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node, vec_dest,
build1 (reduc_code, vectype, PHI_RESULT (new_phi)));
new_temp = make_ssa_name (vec_dest, epilog_stmt);
- TREE_OPERAND (epilog_stmt, 0) = new_temp;
+ GIMPLE_STMT_OPERAND (epilog_stmt, 0) = new_temp;
bsi_insert_after (&exit_bsi, epilog_stmt, BSI_NEW_STMT);
extract_scalar_result = true;
{
tree bitpos = size_int (bit_offset);
- epilog_stmt = build2 (MODIFY_EXPR, vectype, vec_dest,
- build2 (shift_code, vectype, new_temp, bitpos));
+ epilog_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node,
+ vec_dest,
+ build2 (shift_code, vectype,
+ new_temp, bitpos));
new_name = make_ssa_name (vec_dest, epilog_stmt);
- TREE_OPERAND (epilog_stmt, 0) = new_name;
+ GIMPLE_STMT_OPERAND (epilog_stmt, 0) = new_name;
bsi_insert_after (&exit_bsi, epilog_stmt, BSI_NEW_STMT);
- epilog_stmt = build2 (MODIFY_EXPR, vectype, vec_dest,
- build2 (code, vectype, new_name, new_temp));
+ epilog_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node,
+ vec_dest,
+ build2 (code, vectype,
+ new_name, new_temp));
new_temp = make_ssa_name (vec_dest, epilog_stmt);
- TREE_OPERAND (epilog_stmt, 0) = new_temp;
+ GIMPLE_STMT_OPERAND (epilog_stmt, 0) = new_temp;
bsi_insert_after (&exit_bsi, epilog_stmt, BSI_NEW_STMT);
}
rhs = build3 (BIT_FIELD_REF, scalar_type, vec_temp, bitsize,
bitsize_zero_node);
BIT_FIELD_REF_UNSIGNED (rhs) = TYPE_UNSIGNED (scalar_type);
- epilog_stmt = build2 (MODIFY_EXPR, scalar_type, new_scalar_dest, rhs);
+ epilog_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node,
+ new_scalar_dest, rhs);
new_temp = make_ssa_name (new_scalar_dest, epilog_stmt);
- TREE_OPERAND (epilog_stmt, 0) = new_temp;
+ GIMPLE_STMT_OPERAND (epilog_stmt, 0) = new_temp;
bsi_insert_after (&exit_bsi, epilog_stmt, BSI_NEW_STMT);
for (bit_offset = element_bitsize;
bitpos);
BIT_FIELD_REF_UNSIGNED (rhs) = TYPE_UNSIGNED (scalar_type);
- epilog_stmt = build2 (MODIFY_EXPR, scalar_type, new_scalar_dest,
- rhs);
+ epilog_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node,
+ new_scalar_dest, rhs);
new_name = make_ssa_name (new_scalar_dest, epilog_stmt);
- TREE_OPERAND (epilog_stmt, 0) = new_name;
+ GIMPLE_STMT_OPERAND (epilog_stmt, 0) = new_name;
bsi_insert_after (&exit_bsi, epilog_stmt, BSI_NEW_STMT);
- epilog_stmt = build2 (MODIFY_EXPR, scalar_type, new_scalar_dest,
+ epilog_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node,
+ new_scalar_dest,
build2 (code, scalar_type, new_name, new_temp));
new_temp = make_ssa_name (new_scalar_dest, epilog_stmt);
- TREE_OPERAND (epilog_stmt, 0) = new_temp;
+ GIMPLE_STMT_OPERAND (epilog_stmt, 0) = new_temp;
bsi_insert_after (&exit_bsi, epilog_stmt, BSI_NEW_STMT);
}
rhs = build3 (BIT_FIELD_REF, scalar_type, new_temp, bitsize, bitpos);
BIT_FIELD_REF_UNSIGNED (rhs) = TYPE_UNSIGNED (scalar_type);
- epilog_stmt = build2 (MODIFY_EXPR, scalar_type, new_scalar_dest, rhs);
+ epilog_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node,
+ new_scalar_dest, rhs);
new_temp = make_ssa_name (new_scalar_dest, epilog_stmt);
- TREE_OPERAND (epilog_stmt, 0) = new_temp;
+ GIMPLE_STMT_OPERAND (epilog_stmt, 0) = new_temp;
bsi_insert_after (&exit_bsi, epilog_stmt, BSI_NEW_STMT);
}
if (scalar_initial_def)
{
- epilog_stmt = build2 (MODIFY_EXPR, scalar_type, new_scalar_dest,
+ epilog_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node,
+ new_scalar_dest,
build2 (code, scalar_type, new_temp, scalar_initial_def));
new_temp = make_ssa_name (new_scalar_dest, epilog_stmt);
- TREE_OPERAND (epilog_stmt, 0) = new_temp;
+ GIMPLE_STMT_OPERAND (epilog_stmt, 0) = new_temp;
bsi_insert_after (&exit_bsi, epilog_stmt, BSI_NEW_STMT);
}
inside the loop body. The last operand is the reduction variable,
which is defined by the loop-header-phi. */
- gcc_assert (TREE_CODE (stmt) == MODIFY_EXPR);
+ gcc_assert (TREE_CODE (stmt) == GIMPLE_MODIFY_STMT);
- operation = TREE_OPERAND (stmt, 1);
+ operation = GIMPLE_STMT_OPERAND (stmt, 1);
code = TREE_CODE (operation);
op_type = TREE_CODE_LENGTH (code);
if (op_type != binary_op && op_type != ternary_op)
return false;
- scalar_dest = TREE_OPERAND (stmt, 0);
+ scalar_dest = GIMPLE_STMT_OPERAND (stmt, 0);
scalar_type = TREE_TYPE (scalar_dest);
/* All uses but the last are expected to be defined in the loop.
{
/* This is a reduction pattern: get the vectype from the type of the
reduction variable, and get the tree-code from orig_stmt. */
- orig_code = TREE_CODE (TREE_OPERAND (orig_stmt, 1));
+ orig_code = TREE_CODE (GIMPLE_STMT_OPERAND (orig_stmt, 1));
vectype = get_vectype_for_scalar_type (TREE_TYPE (def));
vec_mode = TYPE_MODE (vectype);
}
/* Get the vector def for the reduction variable from the vectorized
reduction operation generated in the previous iteration (j-1) */
- reduc_def = TREE_OPERAND (new_stmt ,0);
+ reduc_def = GIMPLE_STMT_OPERAND (new_stmt ,0);
}
/* Arguments are ready. create the new vector stmt. */
else
expr = build3 (code, vectype, loop_vec_def0, loop_vec_def1,
reduc_def);
- new_stmt = build2 (MODIFY_EXPR, void_type_node, vec_dest, expr);
+ new_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node, vec_dest, expr);
new_temp = make_ssa_name (vec_dest, new_stmt);
- TREE_OPERAND (new_stmt, 0) = new_temp;
+ GIMPLE_STMT_OPERAND (new_stmt, 0) = new_temp;
vect_finish_stmt_generation (stmt, new_stmt, bsi);
if (j == 0)
return true;
}
+/* Checks if CALL can be vectorized in type VECTYPE. Returns
+ true if the target has a vectorized version of the function,
+ or false if the function cannot be vectorized. */
+
+bool
+vectorizable_function (tree call, tree vectype)
+{
+ tree fndecl = get_callee_fndecl (call);
+
+ /* 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)))
+ return false;
+
+ if (!fndecl
+ || TREE_CODE (fndecl) != FUNCTION_DECL
+ || !DECL_BUILT_IN (fndecl))
+ return false;
+
+ if (targetm.vectorize.builtin_vectorized_function (DECL_FUNCTION_CODE (fndecl), vectype))
+ return true;
+
+ return false;
+}
+
+/* Returns an expression that performs a call to vectorized version
+ of FNDECL in type VECTYPE, with the arguments given by ARGS.
+ If extra statements need to be generated, they are inserted
+ before BSI. */
+
+static tree
+build_vectorized_function_call (tree fndecl,
+ tree vectype, tree args)
+{
+ tree vfndecl;
+ enum built_in_function code = DECL_FUNCTION_CODE (fndecl);
+
+ /* The target specific builtin should be available. */
+ vfndecl = targetm.vectorize.builtin_vectorized_function (code, vectype);
+ gcc_assert (vfndecl != NULL_TREE);
+
+ return build_function_call_expr (vfndecl, args);
+}
+
+/* Function vectorizable_call.
+
+ Check if STMT performs a function call that can be vectorized.
+ If VEC_STMT is also passed, vectorize the STMT: create a vectorized
+ stmt to replace it, put it in VEC_STMT, and insert it at BSI.
+ Return FALSE if not a vectorizable STMT, TRUE otherwise. */
+
+bool
+vectorizable_call (tree stmt, block_stmt_iterator *bsi, tree *vec_stmt)
+{
+ tree vec_dest;
+ tree scalar_dest;
+ tree operation;
+ tree op, args, type;
+ tree vec_oprnd, vargs, *pvargs_end;
+ stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
+ tree vectype = STMT_VINFO_VECTYPE (stmt_info);
+ loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
+ tree fndecl, rhs, new_temp, def, def_stmt;
+ enum vect_def_type dt;
+
+ /* Is STMT a vectorizable call? */
+ if (TREE_CODE (stmt) != GIMPLE_MODIFY_STMT)
+ return false;
+
+ if (TREE_CODE (GIMPLE_STMT_OPERAND (stmt, 0)) != SSA_NAME)
+ return false;
+
+ operation = GIMPLE_STMT_OPERAND (stmt, 1);
+ if (TREE_CODE (operation) != CALL_EXPR)
+ return false;
+
+ /* For now, we only vectorize functions if a target specific builtin
+ 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. */
+ if (!vectorizable_function (operation, vectype))
+ {
+ if (vect_print_dump_info (REPORT_DETAILS))
+ fprintf (vect_dump, "function is not vectorizable.");
+
+ return false;
+ }
+ gcc_assert (!stmt_references_memory_p (stmt));
+
+ for (args = TREE_OPERAND (operation, 1); args; args = TREE_CHAIN (args))
+ {
+ op = TREE_VALUE (args);
+
+ if (!vect_is_simple_use (op, loop_vinfo, &def_stmt, &def, &dt))
+ {
+ if (vect_print_dump_info (REPORT_DETAILS))
+ fprintf (vect_dump, "use not simple.");
+ return false;
+ }
+ }
+
+ if (!vec_stmt) /* transformation not required. */
+ {
+ STMT_VINFO_TYPE (stmt_info) = call_vec_info_type;
+ return true;
+ }
+
+ /** Transform. **/
+
+ if (vect_print_dump_info (REPORT_DETAILS))
+ fprintf (vect_dump, "transform operation.");
+
+ /* Handle def. */
+ scalar_dest = GIMPLE_STMT_OPERAND (stmt, 0);
+ vec_dest = vect_create_destination_var (scalar_dest, vectype);
+
+ /* Handle uses. */
+ vargs = NULL_TREE;
+ pvargs_end = &vargs;
+ for (args = TREE_OPERAND (operation, 1); args; args = TREE_CHAIN (args))
+ {
+ op = TREE_VALUE (args);
+ vec_oprnd = vect_get_vec_def_for_operand (op, stmt, NULL);
+
+ *pvargs_end = tree_cons (NULL_TREE, vec_oprnd, NULL_TREE);
+ pvargs_end = &TREE_CHAIN (*pvargs_end);
+ }
+
+ fndecl = get_callee_fndecl (operation);
+ rhs = build_vectorized_function_call (fndecl, vectype, vargs);
+ *vec_stmt = build2 (GIMPLE_MODIFY_STMT, vectype, vec_dest, rhs);
+ 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);
+
+ /* 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);
+
+ return true;
+}
+
/* Function vectorizable_assignment.
gcc_assert (STMT_VINFO_DEF_TYPE (stmt_info) == vect_loop_def);
- if (TREE_CODE (stmt) != MODIFY_EXPR)
+ if (TREE_CODE (stmt) != GIMPLE_MODIFY_STMT)
return false;
- scalar_dest = TREE_OPERAND (stmt, 0);
+ scalar_dest = GIMPLE_STMT_OPERAND (stmt, 0);
if (TREE_CODE (scalar_dest) != SSA_NAME)
return false;
- op = TREE_OPERAND (stmt, 1);
+ op = GIMPLE_STMT_OPERAND (stmt, 1);
if (!vect_is_simple_use (op, loop_vinfo, &def_stmt, &def, &dt))
{
if (vect_print_dump_info (REPORT_DETAILS))
vec_dest = vect_create_destination_var (scalar_dest, vectype);
/* Handle use. */
- op = TREE_OPERAND (stmt, 1);
+ op = GIMPLE_STMT_OPERAND (stmt, 1);
vec_oprnd = vect_get_vec_def_for_operand (op, stmt, NULL);
/* Arguments are ready. create the new vector stmt. */
- *vec_stmt = build2 (MODIFY_EXPR, vectype, vec_dest, vec_oprnd);
+ *vec_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node, vec_dest, vec_oprnd);
new_temp = make_ssa_name (vec_dest, *vec_stmt);
- TREE_OPERAND (*vec_stmt, 0) = new_temp;
+ GIMPLE_STMT_OPERAND (*vec_stmt, 0) = new_temp;
vect_finish_stmt_generation (stmt, *vec_stmt, bsi);
return true;
return false;
}
- if (TREE_CODE (stmt) != MODIFY_EXPR)
+ if (TREE_CODE (stmt) != GIMPLE_MODIFY_STMT)
return false;
- if (TREE_CODE (TREE_OPERAND (stmt, 0)) != SSA_NAME)
+ if (TREE_CODE (GIMPLE_STMT_OPERAND (stmt, 0)) != SSA_NAME)
return false;
- scalar_dest = TREE_OPERAND (stmt, 0);
+ scalar_dest = GIMPLE_STMT_OPERAND (stmt, 0);
vectype_out = get_vectype_for_scalar_type (TREE_TYPE (scalar_dest));
nunits_out = TYPE_VECTOR_SUBPARTS (vectype_out);
if (nunits_out != nunits_in)
return false;
- operation = TREE_OPERAND (stmt, 1);
+ operation = GIMPLE_STMT_OPERAND (stmt, 1);
code = TREE_CODE (operation);
optab = optab_for_tree_code (code, vectype);
/* Arguments are ready. create the new vector stmt. */
if (op_type == binary_op)
- new_stmt = build2 (MODIFY_EXPR, void_type_node, vec_dest,
+ new_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node, vec_dest,
build2 (code, vectype, vec_oprnd0, vec_oprnd1));
else
- new_stmt = build2 (MODIFY_EXPR, void_type_node, vec_dest,
+ new_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node, vec_dest,
build1 (code, vectype, vec_oprnd0));
new_temp = make_ssa_name (vec_dest, new_stmt);
- TREE_OPERAND (new_stmt, 0) = new_temp;
+ GIMPLE_STMT_OPERAND (new_stmt, 0) = new_temp;
vect_finish_stmt_generation (stmt, new_stmt, bsi);
if (j == 0)
return false;
}
- if (TREE_CODE (stmt) != MODIFY_EXPR)
+ if (TREE_CODE (stmt) != GIMPLE_MODIFY_STMT)
return false;
- if (TREE_CODE (TREE_OPERAND (stmt, 0)) != SSA_NAME)
+ if (TREE_CODE (GIMPLE_STMT_OPERAND (stmt, 0)) != SSA_NAME)
return false;
- operation = TREE_OPERAND (stmt, 1);
+ operation = GIMPLE_STMT_OPERAND (stmt, 1);
code = TREE_CODE (operation);
if (code != NOP_EXPR && code != CONVERT_EXPR)
return false;
vectype_in = get_vectype_for_scalar_type (TREE_TYPE (op0));
nunits_in = TYPE_VECTOR_SUBPARTS (vectype_in);
- scalar_dest = TREE_OPERAND (stmt, 0);
+ scalar_dest = GIMPLE_STMT_OPERAND (stmt, 0);
scalar_type = TREE_TYPE (scalar_dest);
vectype_out = get_vectype_for_scalar_type (scalar_type);
nunits_out = TYPE_VECTOR_SUBPARTS (vectype_out);
/* Arguments are ready. Create the new vector stmt. */
expr = build2 (code, vectype_out, vec_oprnd0, vec_oprnd1);
- new_stmt = build2 (MODIFY_EXPR, void_type_node, vec_dest, expr);
+ new_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node, vec_dest, expr);
new_temp = make_ssa_name (vec_dest, new_stmt);
- TREE_OPERAND (new_stmt, 0) = new_temp;
+ GIMPLE_STMT_OPERAND (new_stmt, 0) = new_temp;
vect_finish_stmt_generation (stmt, new_stmt, bsi);
if (j == 0)
else
expr = build1 (code, vectype, vec_oprnd0);
}
- new_stmt = build2 (MODIFY_EXPR, void_type_node, vec_dest, expr);
+ new_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node, vec_dest, expr);
new_temp = make_ssa_name (vec_dest, new_stmt);
- TREE_OPERAND (new_stmt, 0) = new_temp;
+ GIMPLE_STMT_OPERAND (new_stmt, 0) = new_temp;
vect_finish_stmt_generation (stmt, new_stmt, bsi);
if (code == CALL_EXPR)
return false;
}
- if (TREE_CODE (stmt) != MODIFY_EXPR)
+ if (TREE_CODE (stmt) != GIMPLE_MODIFY_STMT)
return false;
- if (TREE_CODE (TREE_OPERAND (stmt, 0)) != SSA_NAME)
+ if (TREE_CODE (GIMPLE_STMT_OPERAND (stmt, 0)) != SSA_NAME)
return false;
- operation = TREE_OPERAND (stmt, 1);
+ operation = GIMPLE_STMT_OPERAND (stmt, 1);
code = TREE_CODE (operation);
if (code != NOP_EXPR && code != WIDEN_MULT_EXPR)
return false;
ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits_in;
gcc_assert (ncopies >= 1);
- scalar_dest = TREE_OPERAND (stmt, 0);
+ scalar_dest = GIMPLE_STMT_OPERAND (stmt, 0);
vectype_out = get_vectype_for_scalar_type (TREE_TYPE (scalar_dest));
nunits_out = TYPE_VECTOR_SUBPARTS (vectype_out);
if (nunits_out != nunits_in / 2) /* FORNOW */
}
+/* Function vect_strided_store_supported.
+
+ Returns TRUE is INTERLEAVE_HIGH and INTERLEAVE_LOW operations are supported,
+ and FALSE otherwise. */
+
+static bool
+vect_strided_store_supported (tree vectype)
+{
+ optab interleave_high_optab, interleave_low_optab;
+ int mode;
+
+ mode = (int) TYPE_MODE (vectype);
+
+ /* Check that the operation is supported. */
+ interleave_high_optab = optab_for_tree_code (VEC_INTERLEAVE_HIGH_EXPR,
+ vectype);
+ interleave_low_optab = optab_for_tree_code (VEC_INTERLEAVE_LOW_EXPR,
+ vectype);
+ if (!interleave_high_optab || !interleave_low_optab)
+ {
+ if (vect_print_dump_info (REPORT_DETAILS))
+ fprintf (vect_dump, "no optab for interleave.");
+ return false;
+ }
+
+ if (interleave_high_optab->handlers[(int) mode].insn_code
+ == CODE_FOR_nothing
+ || interleave_low_optab->handlers[(int) mode].insn_code
+ == CODE_FOR_nothing)
+ {
+ if (vect_print_dump_info (REPORT_DETAILS))
+ fprintf (vect_dump, "interleave op not supported by target.");
+ return false;
+ }
+ return true;
+}
+
+
+/* Function vect_permute_store_chain.
+
+ Given a chain of interleaved stores in DR_CHAIN of LENGTH that must be
+ a power of 2, generate interleave_high/low stmts to reorder the data
+ correctly for the stores. Return the final references for stores in
+ RESULT_CHAIN.
+
+ E.g., LENGTH is 4 and the scalar type is short, i.e., VF is 8.
+ The input is 4 vectors each containing 8 elements. We assign a number to each
+ element, the input sequence is:
+
+ 1st vec: 0 1 2 3 4 5 6 7
+ 2nd vec: 8 9 10 11 12 13 14 15
+ 3rd vec: 16 17 18 19 20 21 22 23
+ 4th vec: 24 25 26 27 28 29 30 31
+
+ The output sequence should be:
+
+ 1st vec: 0 8 16 24 1 9 17 25
+ 2nd vec: 2 10 18 26 3 11 19 27
+ 3rd vec: 4 12 20 28 5 13 21 30
+ 4th vec: 6 14 22 30 7 15 23 31
+
+ i.e., we interleave the contents of the four vectors in their order.
+
+ We use interleave_high/low instructions to create such output. The input of
+ each interleave_high/low operation is two vectors:
+ 1st vec 2nd vec
+ 0 1 2 3 4 5 6 7
+ the even elements of the result vector are obtained left-to-right from the
+ high/low elements of the first vector. The odd elements of the result are
+ obtained left-to-right from the high/low elements of the second vector.
+ The output of interleave_high will be: 0 4 1 5
+ and of interleave_low: 2 6 3 7
+
+
+ The permutation is done in log LENGTH stages. In each stage interleave_high
+ and interleave_low stmts are created for each pair of vectors in DR_CHAIN,
+ where the first argument is taken from the first half of DR_CHAIN and the
+ second argument from it's second half.
+ In our example,
+
+ I1: interleave_high (1st vec, 3rd vec)
+ I2: interleave_low (1st vec, 3rd vec)
+ I3: interleave_high (2nd vec, 4th vec)
+ I4: interleave_low (2nd vec, 4th vec)
+
+ The output for the first stage is:
+
+ I1: 0 16 1 17 2 18 3 19
+ I2: 4 20 5 21 6 22 7 23
+ I3: 8 24 9 25 10 26 11 27
+ I4: 12 28 13 29 14 30 15 31
+
+ The output of the second stage, i.e. the final result is:
+
+ I1: 0 8 16 24 1 9 17 25
+ I2: 2 10 18 26 3 11 19 27
+ I3: 4 12 20 28 5 13 21 30
+ I4: 6 14 22 30 7 15 23 31. */
+
+static bool
+vect_permute_store_chain (VEC(tree,heap) *dr_chain,
+ unsigned int length,
+ tree stmt,
+ block_stmt_iterator *bsi,
+ VEC(tree,heap) **result_chain)
+{
+ tree perm_dest, perm_stmt, vect1, vect2, high, low;
+ tree vectype = STMT_VINFO_VECTYPE (vinfo_for_stmt (stmt));
+ tree scalar_dest;
+ int i;
+ unsigned int j;
+ VEC(tree,heap) *first, *second;
+
+ scalar_dest = TREE_OPERAND (stmt, 0);
+ first = VEC_alloc (tree, heap, length/2);
+ second = VEC_alloc (tree, heap, length/2);
+
+ /* Check that the operation is supported. */
+ if (!vect_strided_store_supported (vectype))
+ return false;
+
+ *result_chain = VEC_copy (tree, heap, dr_chain);
+
+ for (i = 0; i < exact_log2 (length); i++)
+ {
+ for (j = 0; j < length/2; j++)
+ {
+ vect1 = VEC_index (tree, dr_chain, j);
+ vect2 = VEC_index (tree, dr_chain, j+length/2);
+
+ /* high = interleave_high (vect1, vect2); */
+ perm_dest = create_tmp_var (vectype, "vect_inter_high");
+ add_referenced_var (perm_dest);
+ perm_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node, perm_dest,
+ build2 (VEC_INTERLEAVE_HIGH_EXPR, vectype, 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);
+ VEC_replace (tree, *result_chain, 2*j, high);
+
+ /* low = interleave_low (vect1, vect2); */
+ perm_dest = create_tmp_var (vectype, "vect_inter_low");
+ add_referenced_var (perm_dest);
+ perm_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node, perm_dest,
+ build2 (VEC_INTERLEAVE_LOW_EXPR, vectype, 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);
+ VEC_replace (tree, *result_chain, 2*j+1, low);
+ }
+ dr_chain = VEC_copy (tree, heap, *result_chain);
+ }
+ return true;
+}
+
+
/* Function vectorizable_store.
Check if STMT defines a non scalar data-ref (array/pointer/structure) that
tree op;
tree vec_oprnd = NULL_TREE;
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
- struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info);
+ struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info), *first_dr = NULL;
tree vectype = STMT_VINFO_VECTYPE (stmt_info);
loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
enum machine_mode vec_mode;
def_operand_p def_p;
tree def, def_stmt;
enum vect_def_type dt;
- stmt_vec_info prev_stmt_info;
+ stmt_vec_info prev_stmt_info = NULL;
tree dataref_ptr = NULL_TREE;
int nunits = TYPE_VECTOR_SUBPARTS (vectype);
int ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits;
int j;
-
+ tree next_stmt, first_stmt;
+ bool strided_store = false;
+ unsigned int group_size, i;
+ VEC(tree,heap) *dr_chain = NULL, *oprnds = NULL, *result_chain = NULL;
gcc_assert (ncopies >= 1);
/* Is vectorizable store? */
- if (TREE_CODE (stmt) != MODIFY_EXPR)
+ if (TREE_CODE (stmt) != GIMPLE_MODIFY_STMT)
return false;
- scalar_dest = TREE_OPERAND (stmt, 0);
+ scalar_dest = GIMPLE_STMT_OPERAND (stmt, 0);
if (TREE_CODE (scalar_dest) != ARRAY_REF
- && TREE_CODE (scalar_dest) != INDIRECT_REF)
+ && TREE_CODE (scalar_dest) != INDIRECT_REF
+ && !DR_GROUP_FIRST_DR (stmt_info))
return false;
- op = TREE_OPERAND (stmt, 1);
+ op = GIMPLE_STMT_OPERAND (stmt, 1);
if (!vect_is_simple_use (op, loop_vinfo, &def_stmt, &def, &dt))
{
if (vect_print_dump_info (REPORT_DETAILS))
if (!STMT_VINFO_DATA_REF (stmt_info))
return false;
+ if (DR_GROUP_FIRST_DR (stmt_info))
+ {
+ strided_store = true;
+ if (!vect_strided_store_supported (vectype))
+ return false;
+ }
if (!vec_stmt) /* transformation not required. */
{
if (vect_print_dump_info (REPORT_DETAILS))
fprintf (vect_dump, "transform store. ncopies = %d",ncopies);
- alignment_support_cheme = vect_supportable_dr_alignment (dr);
+ if (strided_store)
+ {
+ first_stmt = DR_GROUP_FIRST_DR (stmt_info);
+ first_dr = STMT_VINFO_DATA_REF (vinfo_for_stmt (first_stmt));
+ group_size = DR_GROUP_SIZE (vinfo_for_stmt (first_stmt));
+
+ DR_GROUP_STORE_COUNT (vinfo_for_stmt (first_stmt))++;
+
+ /* We vectorize all the stmts of the interleaving group when we
+ reach the last stmt in the group. */
+ if (DR_GROUP_STORE_COUNT (vinfo_for_stmt (first_stmt))
+ < DR_GROUP_SIZE (vinfo_for_stmt (first_stmt)))
+ {
+ *vec_stmt = NULL_TREE;
+ return true;
+ }
+ }
+ else
+ {
+ first_stmt = stmt;
+ first_dr = dr;
+ group_size = 1;
+ }
+
+ dr_chain = VEC_alloc (tree, heap, group_size);
+ oprnds = VEC_alloc (tree, heap, group_size);
+
+ alignment_support_cheme = vect_supportable_dr_alignment (first_dr);
gcc_assert (alignment_support_cheme);
gcc_assert (alignment_support_cheme == dr_aligned); /* FORNOW */
vector stmt by a factor VF/nunits. For more details see documentation in
vect_get_vec_def_for_copy_stmt. */
+ /* In case of interleaving (non-unit strided access):
+
+ S1: &base + 2 = x2
+ S2: &base = x0
+ S3: &base + 1 = x1
+ S4: &base + 3 = x3
+
+ We create vectorized storess starting from base address (the access of the
+ first stmt in the chain (S2 in the above example), when the last store stmt
+ of the chain (S4) is reached:
+
+ VS1: &base = vx2
+ VS2: &base + vec_size*1 = vx0
+ VS3: &base + vec_size*2 = vx1
+ VS4: &base + vec_size*3 = vx3
+
+ Then permutation statements are generated:
+
+ VS5: vx5 = VEC_INTERLEAVE_HIGH_EXPR < vx0, vx3 >
+ VS6: vx6 = VEC_INTERLEAVE_LOW_EXPR < vx0, vx3 >
+ ...
+
+ And they are put in STMT_VINFO_VEC_STMT of the corresponding scalar stmts
+ (the order of the data-refs in the output of vect_permute_store_chain
+ corresponds to the order of scalar stmts in the interleaving chain - see
+ the documentation of vect_permute_store_chain()).
+
+ In case of both multiple types and interleaving, above vector stores and
+ permutation stmts are created for every copy. The result vector stmts are
+ put in STMT_VINFO_VEC_STMT for the first copy and in the corresponding
+ STMT_VINFO_RELATED_STMT for the next copies.
+ */
+
prev_stmt_info = NULL;
for (j = 0; j < ncopies; j++)
{
if (j == 0)
{
- vec_oprnd = vect_get_vec_def_for_operand (op, stmt, NULL);
- dataref_ptr = vect_create_data_ref_ptr (stmt, bsi, NULL_TREE, &dummy,
- &ptr_incr, false);
+ /* For interleaved stores we collect vectorized defs for all the
+ stores in the group in DR_CHAIN and OPRNDS. DR_CHAIN is then used
+ as an input to vect_permute_store_chain(), and OPRNDS as an input
+ to vect_get_vec_def_for_stmt_copy() for the next copy.
+ If the store is not strided, GROUP_SIZE is 1, and DR_CHAIN and
+ OPRNDS are of size 1.
+ */
+ next_stmt = first_stmt;
+ for (i = 0; i < group_size; i++)
+ {
+ /* Since gaps are not supported for interleaved stores, GROUP_SIZE
+ is the exact number of stmts in the chain. Therefore, NEXT_STMT
+ can't be NULL_TREE. In case that there is no interleaving,
+ GROUP_SIZE is 1, and only one iteration of the loop will be
+ executed.
+ */
+ gcc_assert (next_stmt);
+ op = GIMPLE_STMT_OPERAND (next_stmt, 1);
+ vec_oprnd = vect_get_vec_def_for_operand (op, next_stmt, NULL);
+ VEC_quick_push(tree, dr_chain, vec_oprnd);
+ VEC_quick_push(tree, oprnds, vec_oprnd);
+ next_stmt = DR_GROUP_NEXT_DR (vinfo_for_stmt (next_stmt));
+ }
+ dataref_ptr = vect_create_data_ref_ptr (first_stmt, bsi, NULL_TREE,
+ &dummy, &ptr_incr, false,
+ TREE_TYPE (vec_oprnd));
}
else
{
- vec_oprnd = vect_get_vec_def_for_stmt_copy (dt, vec_oprnd);
+ /* For interleaved stores we created vectorized defs for all the
+ defs stored in OPRNDS in the previous iteration (previous copy).
+ DR_CHAIN is then used as an input to vect_permute_store_chain(),
+ and OPRNDS as an input to vect_get_vec_def_for_stmt_copy() for the
+ next copy.
+ If the store is not strided, GROUP_SIZE is 1, and DR_CHAIN and
+ OPRNDS are of size 1.
+ */
+ for (i = 0; i < group_size; i++)
+ {
+ vec_oprnd = vect_get_vec_def_for_stmt_copy (dt,
+ VEC_index (tree, oprnds, i));
+ VEC_replace(tree, dr_chain, i, vec_oprnd);
+ VEC_replace(tree, oprnds, i, vec_oprnd);
+ }
dataref_ptr = bump_vector_ptr (dataref_ptr, ptr_incr, bsi, stmt);
}
- /* Arguments are ready. create the new vector stmt. */
- data_ref = build_fold_indirect_ref (dataref_ptr);
- new_stmt = build2 (MODIFY_EXPR, vectype, data_ref, vec_oprnd);
- vect_finish_stmt_generation (stmt, new_stmt, bsi);
+ if (strided_store)
+ {
+ result_chain = VEC_alloc (tree, heap, group_size);
+ /* Permute. */
+ if (!vect_permute_store_chain (dr_chain, group_size, stmt, bsi,
+ &result_chain))
+ return false;
+ }
- /* Set the V_MAY_DEFS for the vector pointer. If this virtual def has a
- use outside the loop and a loop peel is performed then the def may be
- renamed by the peel. Mark it for renaming so the later use will also
- be renamed. */
- copy_virtual_operands (new_stmt, stmt);
- if (j == 0)
+ next_stmt = first_stmt;
+ for (i = 0; i < group_size; i++)
{
- /* The original store is deleted so the same SSA_NAMEs can be used.
- */
- FOR_EACH_SSA_TREE_OPERAND (def, stmt, iter, SSA_OP_VMAYDEF)
+ /* For strided stores vectorized defs are interleaved in
+ vect_permute_store_chain(). */
+ if (strided_store)
+ vec_oprnd = VEC_index(tree, result_chain, i);
+
+ data_ref = build_fold_indirect_ref (dataref_ptr);
+ /* Arguments are ready. Create the new vector stmt. */
+ new_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node, data_ref,
+ vec_oprnd);
+ vect_finish_stmt_generation (stmt, new_stmt, bsi);
+
+ /* Set the V_MAY_DEFS for the vector pointer. If this virtual def has a
+ use outside the loop and a loop peel is performed then the def may be
+ renamed by the peel. Mark it for renaming so the later use will also
+ be renamed. */
+ copy_virtual_operands (new_stmt, next_stmt);
+ if (j == 0)
{
- SSA_NAME_DEF_STMT (def) = new_stmt;
- mark_sym_for_renaming (SSA_NAME_VAR (def));
+ /* The original store is deleted so the same SSA_NAMEs can be used.
+ */
+ FOR_EACH_SSA_TREE_OPERAND (def, next_stmt, iter, SSA_OP_VMAYDEF)
+ {
+ SSA_NAME_DEF_STMT (def) = new_stmt;
+ mark_sym_for_renaming (SSA_NAME_VAR (def));
+ }
+
+ STMT_VINFO_VEC_STMT (stmt_info) = *vec_stmt = new_stmt;
}
-
- STMT_VINFO_VEC_STMT (stmt_info) = *vec_stmt = new_stmt;
- }
- else
- {
- /* Create new names for all the definitions created by COPY and
- add replacement mappings for each new name. */
- FOR_EACH_SSA_DEF_OPERAND (def_p, new_stmt, iter, SSA_OP_VMAYDEF)
+ else
{
- create_new_def_for (DEF_FROM_PTR (def_p), new_stmt, def_p);
- mark_sym_for_renaming (SSA_NAME_VAR (DEF_FROM_PTR (def_p)));
+ /* Create new names for all the definitions created by COPY and
+ add replacement mappings for each new name. */
+ FOR_EACH_SSA_DEF_OPERAND (def_p, new_stmt, iter, SSA_OP_VMAYDEF)
+ {
+ create_new_def_for (DEF_FROM_PTR (def_p), new_stmt, def_p);
+ mark_sym_for_renaming (SSA_NAME_VAR (DEF_FROM_PTR (def_p)));
+ }
+
+ STMT_VINFO_RELATED_STMT (prev_stmt_info) = new_stmt;
}
- STMT_VINFO_RELATED_STMT (prev_stmt_info) = new_stmt;
+ prev_stmt_info = vinfo_for_stmt (new_stmt);
+ next_stmt = DR_GROUP_NEXT_DR (vinfo_for_stmt (next_stmt));
+ if (!next_stmt)
+ break;
+ /* Bump the vector pointer. */
+ dataref_ptr = bump_vector_ptr (dataref_ptr, ptr_incr, bsi, stmt);
}
-
- prev_stmt_info = vinfo_for_stmt (new_stmt);
}
return true;
Output:
REALIGNMENT_TOKEN - the result of a call to the builtin_mask_for_load
target hook, if defined.
- Return value - the result of the loop-header phi node.
-*/
+ Return value - the result of the loop-header phi node. */
static tree
vect_setup_realignment (tree stmt, block_stmt_iterator *bsi,
/* 1. Create msq_init = *(floor(p1)) in the loop preheader */
vec_dest = vect_create_destination_var (scalar_dest, vectype);
- ptr = vect_create_data_ref_ptr (stmt, bsi, NULL_TREE, &init_addr, &inc, true);
+ ptr = vect_create_data_ref_ptr (stmt, bsi, NULL_TREE, &init_addr, &inc, true,
+ NULL_TREE);
data_ref = build1 (ALIGN_INDIRECT_REF, vectype, ptr);
- new_stmt = build2 (MODIFY_EXPR, vectype, vec_dest, data_ref);
+ new_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node, vec_dest, data_ref);
new_temp = make_ssa_name (vec_dest, new_stmt);
- TREE_OPERAND (new_stmt, 0) = new_temp;
+ GIMPLE_STMT_OPERAND (new_stmt, 0) = new_temp;
new_bb = bsi_insert_on_edge_immediate (pe, new_stmt);
gcc_assert (!new_bb);
- msq_init = TREE_OPERAND (new_stmt, 0);
+ msq_init = GIMPLE_STMT_OPERAND (new_stmt, 0);
copy_virtual_operands (new_stmt, stmt);
update_vuses_to_preheader (new_stmt, loop);
tree builtin_decl;
tree params = build_tree_list (NULL_TREE, init_addr);
- vec_dest = vect_create_destination_var (scalar_dest,
- TREE_TYPE (new_stmt));
builtin_decl = targetm.vectorize.builtin_mask_for_load ();
new_stmt = build_function_call_expr (builtin_decl, params);
- new_stmt = build2 (MODIFY_EXPR, void_type_node, vec_dest, new_stmt);
+ vec_dest = vect_create_destination_var (scalar_dest,
+ TREE_TYPE (new_stmt));
+ new_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node, vec_dest,
+ new_stmt);
new_temp = make_ssa_name (vec_dest, new_stmt);
- TREE_OPERAND (new_stmt, 0) = new_temp;
+ GIMPLE_STMT_OPERAND (new_stmt, 0) = new_temp;
new_bb = bsi_insert_on_edge_immediate (pe, new_stmt);
gcc_assert (!new_bb);
- *realignment_token = TREE_OPERAND (new_stmt, 0);
+ *realignment_token = GIMPLE_STMT_OPERAND (new_stmt, 0);
/* The result of the CALL_EXPR to this builtin is determined from
the value of the parameter and no global variables are touched
}
+/* Function vect_strided_load_supported.
+
+ Returns TRUE is EXTRACT_EVEN and EXTRACT_ODD operations are supported,
+ and FALSE otherwise. */
+
+static bool
+vect_strided_load_supported (tree vectype)
+{
+ optab perm_even_optab, perm_odd_optab;
+ int mode;
+
+ mode = (int) TYPE_MODE (vectype);
+
+ perm_even_optab = optab_for_tree_code (VEC_EXTRACT_EVEN_EXPR, vectype);
+ if (!perm_even_optab)
+ {
+ if (vect_print_dump_info (REPORT_DETAILS))
+ fprintf (vect_dump, "no optab for perm_even.");
+ return false;
+ }
+
+ if (perm_even_optab->handlers[mode].insn_code == CODE_FOR_nothing)
+ {
+ if (vect_print_dump_info (REPORT_DETAILS))
+ fprintf (vect_dump, "perm_even op not supported by target.");
+ return false;
+ }
+
+ perm_odd_optab = optab_for_tree_code (VEC_EXTRACT_ODD_EXPR, vectype);
+ if (!perm_odd_optab)
+ {
+ if (vect_print_dump_info (REPORT_DETAILS))
+ fprintf (vect_dump, "no optab for perm_odd.");
+ return false;
+ }
+
+ if (perm_odd_optab->handlers[mode].insn_code == CODE_FOR_nothing)
+ {
+ if (vect_print_dump_info (REPORT_DETAILS))
+ fprintf (vect_dump, "perm_odd op not supported by target.");
+ return false;
+ }
+ return true;
+}
+
+
+/* Function vect_permute_load_chain.
+
+ Given a chain of interleaved loads in DR_CHAIN of LENGTH that must be
+ a power of 2, generate extract_even/odd stmts to reorder the input data
+ correctly. Return the final references for loads in RESULT_CHAIN.
+
+ E.g., LENGTH is 4 and the scalar type is short, i.e., VF is 8.
+ The input is 4 vectors each containing 8 elements. We assign a number to each
+ element, the input sequence is:
+
+ 1st vec: 0 1 2 3 4 5 6 7
+ 2nd vec: 8 9 10 11 12 13 14 15
+ 3rd vec: 16 17 18 19 20 21 22 23
+ 4th vec: 24 25 26 27 28 29 30 31
+
+ The output sequence should be:
+
+ 1st vec: 0 4 8 12 16 20 24 28
+ 2nd vec: 1 5 9 13 17 21 25 29
+ 3rd vec: 2 6 10 14 18 22 26 30
+ 4th vec: 3 7 11 15 19 23 27 31
+
+ i.e., the first output vector should contain the first elements of each
+ interleaving group, etc.
+
+ We use extract_even/odd instructions to create such output. The input of each
+ extract_even/odd operation is two vectors
+ 1st vec 2nd vec
+ 0 1 2 3 4 5 6 7
+
+ and the output is the vector of extracted even/odd elements. The output of
+ extract_even will be: 0 2 4 6
+ and of extract_odd: 1 3 5 7
+
+
+ The permutation is done in log LENGTH stages. In each stage extract_even and
+ extract_odd stmts are created for each pair of vectors in DR_CHAIN in their
+ order. In our example,
+
+ E1: extract_even (1st vec, 2nd vec)
+ E2: extract_odd (1st vec, 2nd vec)
+ E3: extract_even (3rd vec, 4th vec)
+ E4: extract_odd (3rd vec, 4th vec)
+
+ The output for the first stage will be:
+
+ E1: 0 2 4 6 8 10 12 14
+ E2: 1 3 5 7 9 11 13 15
+ E3: 16 18 20 22 24 26 28 30
+ E4: 17 19 21 23 25 27 29 31
+
+ In order to proceed and create the correct sequence for the next stage (or
+ for the correct output, if the second stage is the last one, as in our
+ example), we first put the output of extract_even operation and then the
+ output of extract_odd in RESULT_CHAIN (which is then copied to DR_CHAIN).
+ The input for the second stage is:
+
+ 1st vec (E1): 0 2 4 6 8 10 12 14
+ 2nd vec (E3): 16 18 20 22 24 26 28 30
+ 3rd vec (E2): 1 3 5 7 9 11 13 15
+ 4th vec (E4): 17 19 21 23 25 27 29 31
+
+ The output of the second stage:
+
+ E1: 0 4 8 12 16 20 24 28
+ E2: 2 6 10 14 18 22 26 30
+ E3: 1 5 9 13 17 21 25 29
+ E4: 3 7 11 15 19 23 27 31
+
+ And RESULT_CHAIN after reordering:
+
+ 1st vec (E1): 0 4 8 12 16 20 24 28
+ 2nd vec (E3): 1 5 9 13 17 21 25 29
+ 3rd vec (E2): 2 6 10 14 18 22 26 30
+ 4th vec (E4): 3 7 11 15 19 23 27 31. */
+
+static bool
+vect_permute_load_chain (VEC(tree,heap) *dr_chain,
+ unsigned int length,
+ tree stmt,
+ block_stmt_iterator *bsi,
+ VEC(tree,heap) **result_chain)
+{
+ tree perm_dest, perm_stmt, data_ref, first_vect, second_vect;
+ tree vectype = STMT_VINFO_VECTYPE (vinfo_for_stmt (stmt));
+ int i;
+ unsigned int j;
+
+ /* Check that the operation is supported. */
+ if (!vect_strided_load_supported (vectype))
+ return false;
+
+ *result_chain = VEC_copy (tree, heap, dr_chain);
+ for (i = 0; i < exact_log2 (length); i++)
+ {
+ for (j = 0; j < length; j +=2)
+ {
+ first_vect = VEC_index (tree, dr_chain, j);
+ second_vect = VEC_index (tree, dr_chain, j+1);
+
+ /* data_ref = permute_even (first_data_ref, second_data_ref); */
+ perm_dest = create_tmp_var (vectype, "vect_perm_even");
+ add_referenced_var (perm_dest);
+
+ perm_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node, perm_dest,
+ build2 (VEC_EXTRACT_EVEN_EXPR, vectype,
+ 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);
+ mark_new_vars_to_rename (perm_stmt);
+
+ VEC_replace (tree, *result_chain, j/2, data_ref);
+
+ /* data_ref = permute_odd (first_data_ref, second_data_ref); */
+ perm_dest = create_tmp_var (vectype, "vect_perm_odd");
+ add_referenced_var (perm_dest);
+
+ perm_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node, perm_dest,
+ build2 (VEC_EXTRACT_ODD_EXPR, vectype,
+ 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);
+ mark_new_vars_to_rename (perm_stmt);
+
+ VEC_replace (tree, *result_chain, j/2+length/2, data_ref);
+ }
+ dr_chain = VEC_copy (tree, heap, *result_chain);
+ }
+ return true;
+}
+
+
+/* Function vect_transform_strided_load.
+
+ Given a chain of input interleaved data-refs (in DR_CHAIN), build statements
+ to perform their permutation and ascribe the result vectorized statements to
+ the scalar statements.
+*/
+
+static bool
+vect_transform_strided_load (tree stmt, VEC(tree,heap) *dr_chain, int size,
+ block_stmt_iterator *bsi)
+{
+ stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
+ tree first_stmt = DR_GROUP_FIRST_DR (stmt_info);
+ tree next_stmt, new_stmt;
+ VEC(tree,heap) *result_chain = NULL;
+ unsigned int i, gap_count;
+ tree tmp_data_ref;
+
+ /* DR_CHAIN contains input data-refs that are a part of the interleaving.
+ RESULT_CHAIN is the output of vect_permute_load_chain, it contains permuted
+ vectors, that are ready for vector computation. */
+ result_chain = VEC_alloc (tree, heap, size);
+ /* Permute. */
+ if (!vect_permute_load_chain (dr_chain, size, stmt, bsi, &result_chain))
+ return false;
+
+ /* Put a permuted data-ref in the VECTORIZED_STMT field.
+ Since we scan the chain starting from it's first node, their order
+ corresponds the order of data-refs in RESULT_CHAIN. */
+ next_stmt = first_stmt;
+ gap_count = 1;
+ for (i = 0; VEC_iterate(tree, result_chain, i, tmp_data_ref); i++)
+ {
+ if (!next_stmt)
+ break;
+
+ /* Skip the gaps. Loads created for the gaps will be removed by dead
+ code elimination pass later.
+ DR_GROUP_GAP is the number of steps in elements from the previous
+ access (if there is no gap DR_GROUP_GAP is 1). We skip loads that
+ correspond to the gaps.
+ */
+ if (gap_count < DR_GROUP_GAP (vinfo_for_stmt (next_stmt)))
+ {
+ gap_count++;
+ continue;
+ }
+
+ while (next_stmt)
+ {
+ new_stmt = SSA_NAME_DEF_STMT (tmp_data_ref);
+ /* We assume that if VEC_STMT is not NULL, this is a case of multiple
+ copies, and we put the new vector statement in the first available
+ RELATED_STMT. */
+ if (!STMT_VINFO_VEC_STMT (vinfo_for_stmt (next_stmt)))
+ STMT_VINFO_VEC_STMT (vinfo_for_stmt (next_stmt)) = new_stmt;
+ else
+ {
+ tree prev_stmt = STMT_VINFO_VEC_STMT (vinfo_for_stmt (next_stmt));
+ tree rel_stmt = STMT_VINFO_RELATED_STMT (
+ vinfo_for_stmt (prev_stmt));
+ while (rel_stmt)
+ {
+ prev_stmt = rel_stmt;
+ rel_stmt = STMT_VINFO_RELATED_STMT (vinfo_for_stmt (rel_stmt));
+ }
+ STMT_VINFO_RELATED_STMT (vinfo_for_stmt (prev_stmt)) = new_stmt;
+ }
+ next_stmt = DR_GROUP_NEXT_DR (vinfo_for_stmt (next_stmt));
+ gap_count = 1;
+ /* If NEXT_STMT accesses the same DR as the previous statement,
+ put the same TMP_DATA_REF as its vectorized statement; otherwise
+ get the next data-ref from RESULT_CHAIN. */
+ if (!next_stmt || !DR_GROUP_SAME_DR_STMT (vinfo_for_stmt (next_stmt)))
+ break;
+ }
+ }
+ return true;
+}
+
+
/* vectorizable_load.
Check if STMT reads a non scalar data-ref (array/pointer/structure) that
stmt_vec_info prev_stmt_info;
loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
- struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info);
+ 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;
+ tree new_stmt = NULL_TREE;
tree dummy;
enum dr_alignment_support alignment_support_cheme;
tree dataref_ptr = NULL_TREE;
tree ptr_incr;
int nunits = TYPE_VECTOR_SUBPARTS (vectype);
int ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits;
- int j;
+ int i, j, group_size;
tree msq = NULL_TREE, lsq;
tree offset = NULL_TREE;
tree realignment_token = NULL_TREE;
tree phi_stmt = NULL_TREE;
+ VEC(tree,heap) *dr_chain = NULL;
+ bool strided_load = false;
+ tree first_stmt;
/* Is vectorizable load? */
if (!STMT_VINFO_RELEVANT_P (stmt_info))
return false;
}
- if (TREE_CODE (stmt) != MODIFY_EXPR)
+ if (TREE_CODE (stmt) != GIMPLE_MODIFY_STMT)
return false;
- scalar_dest = TREE_OPERAND (stmt, 0);
+ scalar_dest = GIMPLE_STMT_OPERAND (stmt, 0);
if (TREE_CODE (scalar_dest) != SSA_NAME)
return false;
- op = TREE_OPERAND (stmt, 1);
- if (TREE_CODE (op) != ARRAY_REF && TREE_CODE (op) != INDIRECT_REF)
+ op = GIMPLE_STMT_OPERAND (stmt, 1);
+ if (TREE_CODE (op) != ARRAY_REF
+ && TREE_CODE (op) != INDIRECT_REF
+ && !DR_GROUP_FIRST_DR (stmt_info))
return false;
if (!STMT_VINFO_DATA_REF (stmt_info))
return false;
}
+ /* Check if the load is a part of an interleaving chain. */
+ if (DR_GROUP_FIRST_DR (stmt_info))
+ {
+ strided_load = true;
+
+ /* Check if interleaving is supported. */
+ if (!vect_strided_load_supported (vectype))
+ return false;
+ }
+
if (!vec_stmt) /* transformation not required. */
{
STMT_VINFO_TYPE (stmt_info) = load_vec_info_type;
if (vect_print_dump_info (REPORT_DETAILS))
fprintf (vect_dump, "transform load.");
- alignment_support_cheme = vect_supportable_dr_alignment (dr);
+ if (strided_load)
+ {
+ first_stmt = DR_GROUP_FIRST_DR (stmt_info);
+ /* Check if the chain of loads is already vectorized. */
+ if (STMT_VINFO_VEC_STMT (vinfo_for_stmt (first_stmt)))
+ {
+ *vec_stmt = STMT_VINFO_VEC_STMT (stmt_info);
+ return true;
+ }
+ first_dr = STMT_VINFO_DATA_REF (vinfo_for_stmt (first_stmt));
+ group_size = DR_GROUP_SIZE (vinfo_for_stmt (first_stmt));
+ dr_chain = VEC_alloc (tree, heap, group_size);
+ }
+ else
+ {
+ first_stmt = stmt;
+ first_dr = dr;
+ group_size = 1;
+ }
+
+ alignment_support_cheme = vect_supportable_dr_alignment (first_dr);
gcc_assert (alignment_support_cheme);
+
/* In case the vectorization factor (VF) is bigger than the number
of elements that we can fit in a vectype (nunits), we have to generate
more than one vector stmt - i.e - we need to "unroll" the
information we recorded in RELATED_STMT field is used to vectorize
stmt S2. */
+ /* In case of interleaving (non-unit strided access):
+
+ S1: x2 = &base + 2
+ S2: x0 = &base
+ S3: x1 = &base + 1
+ S4: x3 = &base + 3
+
+ Vectorized loads are created in the order of memory accesses
+ starting from the access of the first stmt of the chain:
+
+ VS1: vx0 = &base
+ VS2: vx1 = &base + vec_size*1
+ VS3: vx3 = &base + vec_size*2
+ VS4: vx4 = &base + vec_size*3
+
+ Then permutation statements are generated:
+
+ VS5: vx5 = VEC_EXTRACT_EVEN_EXPR < vx0, vx1 >
+ VS6: vx6 = VEC_EXTRACT_ODD_EXPR < vx0, vx1 >
+ ...
+
+ And they are put in STMT_VINFO_VEC_STMT of the corresponding scalar stmts
+ (the order of the data-refs in the output of vect_permute_load_chain
+ corresponds to the order of scalar stmts in the interleaving chain - see
+ the documentation of vect_permute_load_chain()).
+ The generation of permutation stmts and recording them in
+ STMT_VINFO_VEC_STMT is done in vect_transform_strided_load().
+
+ In case of both multiple types and interleaving, the vector loads and
+ permutation stmts above are created for every copy. The result vector stmts
+ are put in STMT_VINFO_VEC_STMT for the first copy and in the corresponding
+ STMT_VINFO_RELATED_STMT for the next copies. */
+
/* If the data reference is aligned (dr_aligned) or potentially unaligned
on a target that supports unaligned accesses (dr_unaligned_supported)
we generate the following code:
vec_dest = realign_load (msq, lsq, realignment_token)
indx = indx + 1;
msq = lsq;
- }
- */
+ } */
if (alignment_support_cheme == dr_unaligned_software_pipeline)
{
- msq = vect_setup_realignment (stmt, bsi, &realignment_token);
+ msq = vect_setup_realignment (first_stmt, bsi, &realignment_token);
phi_stmt = SSA_NAME_DEF_STMT (msq);
offset = size_int (TYPE_VECTOR_SUBPARTS (vectype) - 1);
}
{
/* 1. Create the vector pointer update chain. */
if (j == 0)
- dataref_ptr = vect_create_data_ref_ptr (stmt, bsi, offset,
- &dummy, &ptr_incr, false);
+ dataref_ptr = vect_create_data_ref_ptr (first_stmt, bsi, offset, &dummy,
+ &ptr_incr, false, NULL_TREE);
else
dataref_ptr = bump_vector_ptr (dataref_ptr, ptr_incr, bsi, stmt);
- /* 2. Create the vector-load in the loop. */
- switch (alignment_support_cheme)
- {
- case dr_aligned:
- gcc_assert (aligned_access_p (dr));
- data_ref = build_fold_indirect_ref (dataref_ptr);
- break;
- case dr_unaligned_supported:
- {
- int mis = DR_MISALIGNMENT (dr);
- tree tmis = (mis == -1 ? size_zero_node : size_int (mis));
-
- gcc_assert (!aligned_access_p (dr));
- tmis = size_binop (MULT_EXPR, tmis, size_int(BITS_PER_UNIT));
- data_ref =
- build2 (MISALIGNED_INDIRECT_REF, vectype, dataref_ptr, tmis);
- break;
- }
- case dr_unaligned_software_pipeline:
- gcc_assert (!aligned_access_p (dr));
- data_ref = build1 (ALIGN_INDIRECT_REF, vectype, dataref_ptr);
- break;
- default:
- gcc_unreachable ();
- }
- vec_dest = vect_create_destination_var (scalar_dest, vectype);
- new_stmt = build2 (MODIFY_EXPR, vectype, vec_dest, data_ref);
- new_temp = make_ssa_name (vec_dest, new_stmt);
- TREE_OPERAND (new_stmt, 0) = new_temp;
- vect_finish_stmt_generation (stmt, new_stmt, bsi);
- copy_virtual_operands (new_stmt, stmt);
- mark_new_vars_to_rename (new_stmt);
-
- /* 3. Handle explicit realignment if necessary/supported. */
- if (alignment_support_cheme == dr_unaligned_software_pipeline)
- {
- /* Create in loop:
- <vec_dest = realign_load (msq, lsq, realignment_token)> */
- lsq = TREE_OPERAND (new_stmt, 0);
- 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 = build2 (MODIFY_EXPR, vectype, vec_dest, new_stmt);
- new_temp = make_ssa_name (vec_dest, new_stmt);
- TREE_OPERAND (new_stmt, 0) = new_temp;
- vect_finish_stmt_generation (stmt, new_stmt, bsi);
- if (j == ncopies - 1)
- add_phi_arg (phi_stmt, lsq, loop_latch_edge (loop));
- msq = lsq;
- }
+ for (i = 0; i < group_size; i++)
+ {
+ /* 2. Create the vector-load in the loop. */
+ switch (alignment_support_cheme)
+ {
+ case dr_aligned:
+ gcc_assert (aligned_access_p (first_dr));
+ data_ref = build_fold_indirect_ref (dataref_ptr);
+ break;
+ case dr_unaligned_supported:
+ {
+ int mis = DR_MISALIGNMENT (first_dr);
+ tree tmis = (mis == -1 ? size_zero_node : size_int (mis));
+
+ gcc_assert (!aligned_access_p (first_dr));
+ tmis = size_binop (MULT_EXPR, tmis, size_int(BITS_PER_UNIT));
+ data_ref =
+ build2 (MISALIGNED_INDIRECT_REF, vectype, dataref_ptr, tmis);
+ break;
+ }
+ case dr_unaligned_software_pipeline:
+ gcc_assert (!aligned_access_p (first_dr));
+ data_ref = build1 (ALIGN_INDIRECT_REF, vectype, dataref_ptr);
+ break;
+ default:
+ gcc_unreachable ();
+ }
+ vec_dest = vect_create_destination_var (scalar_dest, vectype);
+ new_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node, 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);
+ copy_virtual_operands (new_stmt, stmt);
+ mark_new_vars_to_rename (new_stmt);
+
+ /* 3. Handle explicit realignment if necessary/supported. */
+ if (alignment_support_cheme == dr_unaligned_software_pipeline)
+ {
+ /* Create in loop:
+ <vec_dest = realign_load (msq, lsq, realignment_token)> */
+ lsq = GIMPLE_STMT_OPERAND (new_stmt, 0);
+ 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 = build2 (GIMPLE_MODIFY_STMT, void_type_node, vec_dest,
+ new_stmt);
+ 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);
+ if (i == group_size - 1 && j == ncopies - 1)
+ add_phi_arg (phi_stmt, lsq, loop_latch_edge (loop));
+ msq = lsq;
+ }
+ if (strided_load)
+ VEC_quick_push (tree, dr_chain, new_temp);
+ if (i < group_size - 1)
+ dataref_ptr = bump_vector_ptr (dataref_ptr, ptr_incr, bsi, stmt);
+ }
- if (j == 0)
- STMT_VINFO_VEC_STMT (stmt_info) = *vec_stmt = new_stmt;
+ if (strided_load)
+ {
+ if (!vect_transform_strided_load (stmt, dr_chain, group_size, bsi))
+ return false;
+ *vec_stmt = STMT_VINFO_VEC_STMT (stmt_info);
+ dr_chain = VEC_alloc (tree, heap, group_size);
+ }
else
- STMT_VINFO_RELATED_STMT (prev_stmt_info) = new_stmt;
- prev_stmt_info = vinfo_for_stmt (new_stmt);
+ {
+ if (j == 0)
+ STMT_VINFO_VEC_STMT (stmt_info) = *vec_stmt = new_stmt;
+ else
+ STMT_VINFO_RELATED_STMT (prev_stmt_info) = new_stmt;
+ prev_stmt_info = vinfo_for_stmt (new_stmt);
+ }
}
return true;
if (!STMT_VINFO_LIVE_P (stmt_info))
return false;
- if (TREE_CODE (stmt) != MODIFY_EXPR)
+ if (TREE_CODE (stmt) != GIMPLE_MODIFY_STMT)
return false;
- if (TREE_CODE (TREE_OPERAND (stmt, 0)) != SSA_NAME)
+ if (TREE_CODE (GIMPLE_STMT_OPERAND (stmt, 0)) != SSA_NAME)
return false;
- operation = TREE_OPERAND (stmt, 1);
+ operation = GIMPLE_STMT_OPERAND (stmt, 1);
code = TREE_CODE (operation);
op_type = TREE_CODE_LENGTH (code);
return false;
}
- if (TREE_CODE (stmt) != MODIFY_EXPR)
+ if (TREE_CODE (stmt) != GIMPLE_MODIFY_STMT)
return false;
- op = TREE_OPERAND (stmt, 1);
+ op = GIMPLE_STMT_OPERAND (stmt, 1);
if (TREE_CODE (op) != COND_EXPR)
return false;
/* Transform */
/* Handle def. */
- scalar_dest = TREE_OPERAND (stmt, 0);
+ scalar_dest = GIMPLE_STMT_OPERAND (stmt, 0);
vec_dest = vect_create_destination_var (scalar_dest, vectype);
/* Handle cond expr. */
vec_cond_expr = build3 (VEC_COND_EXPR, vectype,
vec_compare, vec_then_clause, vec_else_clause);
- *vec_stmt = build2 (MODIFY_EXPR, vectype, vec_dest, vec_cond_expr);
+ *vec_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node, vec_dest,
+ vec_cond_expr);
new_temp = make_ssa_name (vec_dest, *vec_stmt);
- TREE_OPERAND (*vec_stmt, 0) = new_temp;
+ GIMPLE_STMT_OPERAND (*vec_stmt, 0) = new_temp;
vect_finish_stmt_generation (stmt, *vec_stmt, bsi);
return true;
Create a vectorized stmt to replace STMT, and insert it at BSI. */
bool
-vect_transform_stmt (tree stmt, block_stmt_iterator *bsi)
+vect_transform_stmt (tree stmt, block_stmt_iterator *bsi, bool *strided_store)
{
bool is_store = false;
tree vec_stmt = NULL_TREE;
case store_vec_info_type:
done = vectorizable_store (stmt, bsi, &vec_stmt);
gcc_assert (done);
- is_store = true;
+ if (DR_GROUP_FIRST_DR (stmt_info))
+ {
+ /* In case of interleaving, the whole chain is vectorized when the
+ last store in the chain is reached. Store stmts before the last
+ one are skipped, and there vec_stmt_info shouldn't be freed
+ meanwhile. */
+ *strided_store = true;
+ if (STMT_VINFO_VEC_STMT (stmt_info))
+ is_store = true;
+ }
+ else
+ is_store = true;
break;
case condition_vec_info_type:
gcc_assert (done);
break;
+ case call_vec_info_type:
+ done = vectorizable_call (stmt, bsi, &vec_stmt);
+ break;
+
default:
if (vect_print_dump_info (REPORT_DETAILS))
fprintf (vect_dump, "stmt not supported.");
gcc_unreachable ();
}
- gcc_assert (vec_stmt);
- STMT_VINFO_VEC_STMT (stmt_info) = vec_stmt;
- orig_stmt_in_pattern = STMT_VINFO_RELATED_STMT (stmt_info);
- if (orig_stmt_in_pattern)
- {
- stmt_vec_info stmt_vinfo = vinfo_for_stmt (orig_stmt_in_pattern);
- if (STMT_VINFO_IN_PATTERN_P (stmt_vinfo))
- {
- gcc_assert (STMT_VINFO_RELATED_STMT (stmt_vinfo) == stmt);
-
- /* STMT was inserted by the vectorizer to replace a computation
- idiom. ORIG_STMT_IN_PATTERN is a stmt in the original
- sequence that computed this idiom. We need to record a pointer
- to VEC_STMT in the stmt_info of ORIG_STMT_IN_PATTERN. See more
- detail in the documentation of vect_pattern_recog. */
-
- STMT_VINFO_VEC_STMT (stmt_vinfo) = vec_stmt;
- }
- }
+ gcc_assert (vec_stmt || *strided_store);
+ if (vec_stmt)
+ {
+ STMT_VINFO_VEC_STMT (stmt_info) = vec_stmt;
+ orig_stmt_in_pattern = STMT_VINFO_RELATED_STMT (stmt_info);
+ if (orig_stmt_in_pattern)
+ {
+ stmt_vec_info stmt_vinfo = vinfo_for_stmt (orig_stmt_in_pattern);
+ if (STMT_VINFO_IN_PATTERN_P (stmt_vinfo))
+ {
+ gcc_assert (STMT_VINFO_RELATED_STMT (stmt_vinfo) == stmt);
+
+ /* STMT was inserted by the vectorizer to replace a
+ computation idiom. ORIG_STMT_IN_PATTERN is a stmt in the
+ original sequence that computed this idiom. We need to
+ record a pointer to VEC_STMT in the stmt_info of
+ ORIG_STMT_IN_PATTERN. See more details in the
+ documentation of vect_pattern_recog. */
+
+ STMT_VINFO_VEC_STMT (stmt_vinfo) = vec_stmt;
+ }
+ }
+ }
}
if (STMT_VINFO_LIVE_P (stmt_info))
/* Create: ratio = ni >> log2(vf) */
- var = create_tmp_var (TREE_TYPE (ni), "bnd");
- add_referenced_var (var);
- ratio_name = make_ssa_name (var, NULL_TREE);
- stmt = build2 (MODIFY_EXPR, void_type_node, ratio_name,
- build2 (RSHIFT_EXPR, TREE_TYPE (ni_name), ni_name, log_vf));
- SSA_NAME_DEF_STMT (ratio_name) = stmt;
+ ratio_name = fold_build2 (RSHIFT_EXPR, TREE_TYPE (ni_name), ni_name, log_vf);
+ if (!is_gimple_val (ratio_name))
+ {
+ var = create_tmp_var (TREE_TYPE (ni), "bnd");
+ add_referenced_var (var);
- pe = loop_preheader_edge (loop);
- new_bb = bsi_insert_on_edge_immediate (pe, stmt);
- gcc_assert (!new_bb);
+ ratio_name = force_gimple_operand (ratio_name, &stmt, true, var);
+ pe = loop_preheader_edge (loop);
+ new_bb = bsi_insert_on_edge_immediate (pe, stmt);
+ gcc_assert (!new_bb);
+ }
/* Create: ratio_mult_vf = ratio << log2 (vf). */
- var = create_tmp_var (TREE_TYPE (ni), "ratio_mult_vf");
- add_referenced_var (var);
- ratio_mult_vf_name = make_ssa_name (var, NULL_TREE);
- stmt = build2 (MODIFY_EXPR, void_type_node, ratio_mult_vf_name,
- build2 (LSHIFT_EXPR, TREE_TYPE (ratio_name), ratio_name, log_vf));
- SSA_NAME_DEF_STMT (ratio_mult_vf_name) = stmt;
+ ratio_mult_vf_name = fold_build2 (LSHIFT_EXPR, TREE_TYPE (ratio_name),
+ ratio_name, log_vf);
+ if (!is_gimple_val (ratio_mult_vf_name))
+ {
+ var = create_tmp_var (TREE_TYPE (ni), "ratio_mult_vf");
+ add_referenced_var (var);
- pe = loop_preheader_edge (loop);
- new_bb = bsi_insert_on_edge_immediate (pe, stmt);
- gcc_assert (!new_bb);
+ ratio_mult_vf_name = force_gimple_operand (ratio_mult_vf_name, &stmt,
+ true, var);
+ pe = loop_preheader_edge (loop);
+ new_bb = bsi_insert_on_edge_immediate (pe, stmt);
+ gcc_assert (!new_bb);
+ }
*ni_name_ptr = ni_name;
*ratio_mult_vf_name_ptr = ratio_mult_vf_name;
edge update_e)
{
struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
- basic_block exit_bb = loop->single_exit->dest;
+ basic_block exit_bb = single_exit (loop)->dest;
tree phi, phi1;
basic_block update_bb = update_e->dest;
init_expr = unshare_expr (initial_condition_in_loop_num (access_fn,
loop->num));
- ni = build2 (PLUS_EXPR, TREE_TYPE (init_expr),
- build2 (MULT_EXPR, TREE_TYPE (niters),
- niters, step_expr), init_expr);
+ ni = fold_build2 (PLUS_EXPR, TREE_TYPE (init_expr),
+ fold_build2 (MULT_EXPR, TREE_TYPE (init_expr),
+ fold_convert (TREE_TYPE (init_expr),
+ niters),
+ step_expr),
+ init_expr);
var = create_tmp_var (TREE_TYPE (init_expr), "tmp");
add_referenced_var (var);
NITERS / VECTORIZATION_FACTOR times (this value is placed into RATIO). */
static void
-vect_do_peeling_for_loop_bound (loop_vec_info loop_vinfo, tree *ratio,
- struct loops *loops)
+vect_do_peeling_for_loop_bound (loop_vec_info loop_vinfo, tree *ratio)
{
tree ni_name, ratio_mult_vf_name;
struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
&ratio_mult_vf_name, ratio);
loop_num = loop->num;
- new_loop = slpeel_tree_peel_loop_to_edge (loop, loops, loop->single_exit,
+ new_loop = slpeel_tree_peel_loop_to_edge (loop, single_exit (loop),
ratio_mult_vf_name, ni_name, false);
gcc_assert (new_loop);
gcc_assert (loop_num == loop->num);
is on the path where the LOOP IVs are used and need to be updated. */
preheader = loop_preheader_edge (new_loop)->src;
- if (EDGE_PRED (preheader, 0)->src == loop->single_exit->dest)
+ if (EDGE_PRED (preheader, 0)->src == single_exit (loop)->dest)
update_e = EDGE_PRED (preheader, 0);
else
update_e = EDGE_PRED (preheader, 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) */
+ whose type is the inner type of the vectype)
+
+ For interleaving,
+
+ 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.
+*/
static tree
vect_gen_niters_for_prolog_loop (loop_vec_info loop_vinfo, tree loop_niters)
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 element_size = GET_MODE_SIZE (TYPE_MODE (TREE_TYPE (DR_REF (dr))));
+
+ if (DR_GROUP_FIRST_DR (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;
+ }
pe = loop_preheader_edge (loop);
if (LOOP_PEELING_FOR_ALIGNMENT (loop_vinfo) > 0)
{
int byte_misalign = LOOP_PEELING_FOR_ALIGNMENT (loop_vinfo);
- int element_size = GET_MODE_SIZE (TYPE_MODE (TREE_TYPE (DR_REF (dr))));
int elem_misalign = byte_misalign / element_size;
if (vect_print_dump_info (REPORT_DETAILS))
fprintf (vect_dump, "known alignment = %d.", byte_misalign);
- iters = build_int_cst (niters_type, (vf - elem_misalign)&(vf-1));
+ iters = build_int_cst (niters_type,
+ (vf - elem_misalign)&(vf/group_size-1));
}
else
{
/* Create: byte_misalign = addr & (vectype_size - 1) */
byte_misalign =
- build2 (BIT_AND_EXPR, type, start_addr, vectype_size_minus_1);
+ fold_build2 (BIT_AND_EXPR, type, start_addr, vectype_size_minus_1);
/* Create: elem_misalign = byte_misalign / element_size */
elem_misalign =
- build2 (RSHIFT_EXPR, type, byte_misalign, elem_size_log);
+ fold_build2 (RSHIFT_EXPR, type, byte_misalign, elem_size_log);
/* Create: (niters_type) (VF - elem_misalign)&(VF - 1) */
- iters = build2 (MINUS_EXPR, type, vf_tree, elem_misalign);
- iters = build2 (BIT_AND_EXPR, type, iters, vf_minus_1);
+ iters = fold_build2 (MINUS_EXPR, type, vf_tree, elem_misalign);
+ iters = fold_build2 (BIT_AND_EXPR, type, iters, vf_minus_1);
iters = fold_convert (niters_type, iters);
}
greater than vf; since the misalignment ('iters') is at most vf, there's
no need to generate the MIN_EXPR in this case. */
if (TREE_CODE (loop_niters) != INTEGER_CST)
- iters = build2 (MIN_EXPR, niters_type, iters, loop_niters);
+ iters = fold_build2 (MIN_EXPR, niters_type, iters, loop_niters);
if (vect_print_dump_info (REPORT_DETAILS))
{
peeling is recorded in LOOP_VINFO_UNALIGNED_DR. */
static void
-vect_do_peeling_for_alignment (loop_vec_info loop_vinfo, struct loops *loops)
+vect_do_peeling_for_alignment (loop_vec_info loop_vinfo)
{
struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
tree niters_of_prolog_loop, ni_name;
/* Peel the prolog loop and iterate it niters_of_prolog_loop. */
new_loop =
- slpeel_tree_peel_loop_to_edge (loop, loops, loop_preheader_edge (loop),
+ slpeel_tree_peel_loop_to_edge (loop, loop_preheader_edge (loop),
niters_of_prolog_loop, ni_name, true);
gcc_assert (new_loop);
#ifdef ENABLE_CHECKING
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 = build2 (MODIFY_EXPR, void_type_node,
+ addr_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node,
addr_tmp_name, addr_stmt);
SSA_NAME_DEF_STMT (addr_tmp_name) = addr_stmt;
append_to_statement_list_force (addr_stmt, cond_expr_stmt_list);
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);
- or_stmt = build2 (MODIFY_EXPR, void_type_node, new_or_tmp_name,
+ or_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node,
+ new_or_tmp_name,
build2 (BIT_IOR_EXPR, int_ptrsize_type,
or_tmp_name,
addr_tmp_name));
add_referenced_var (and_tmp);
and_tmp_name = make_ssa_name (and_tmp, NULL_TREE);
- and_stmt = build2 (MODIFY_EXPR, void_type_node,
+ and_stmt = build2 (GIMPLE_MODIFY_STMT, void_type_node,
and_tmp_name,
build2 (BIT_AND_EXPR, int_ptrsize_type,
or_tmp_name, mask_cst));
stmts in the loop, and update the loop exit condition. */
void
-vect_transform_loop (loop_vec_info loop_vinfo,
- struct loops *loops ATTRIBUTE_UNUSED)
+vect_transform_loop (loop_vec_info loop_vinfo)
{
struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
basic_block *bbs = LOOP_VINFO_BBS (loop_vinfo);
int vectorization_factor = LOOP_VINFO_VECT_FACTOR (loop_vinfo);
bitmap_iterator bi;
unsigned int j;
+ bool strided_store;
if (vect_print_dump_info (REPORT_DETAILS))
fprintf (vect_dump, "=== vec_transform_loop ===");
cond_expr = vect_create_cond_for_align_checks (loop_vinfo,
&cond_expr_stmt_list);
initialize_original_copy_tables ();
- nloop = loop_version (loops, loop, cond_expr, &condition_bb, true);
+ nloop = loop_version (loop, cond_expr, &condition_bb, true);
free_original_copy_tables();
/** Loop versioning violates an assumption we try to maintain during
here by adding a new (empty) block on the exit-edge of the loop,
with the proper loop-exit phis to maintain loop-closed-form. **/
- merge_bb = loop->single_exit->dest;
+ merge_bb = single_exit (loop)->dest;
gcc_assert (EDGE_COUNT (merge_bb->preds) == 2);
- new_exit_bb = split_edge (loop->single_exit);
- add_bb_to_loop (new_exit_bb, loop->outer);
- new_exit_e = loop->single_exit;
+ new_exit_bb = split_edge (single_exit (loop));
+ new_exit_e = single_exit (loop);
e = EDGE_SUCC (new_exit_bb, 0);
for (orig_phi = phi_nodes (merge_bb); orig_phi;
Only one data ref with unknown store is allowed. */
if (LOOP_PEELING_FOR_ALIGNMENT (loop_vinfo))
- vect_do_peeling_for_alignment (loop_vinfo, loops);
+ vect_do_peeling_for_alignment (loop_vinfo);
/* If the loop has a symbolic number of iterations 'n' (i.e. it's not a
compile time constant), or it is a constant that doesn't divide by the
if (!LOOP_VINFO_NITERS_KNOWN_P (loop_vinfo)
|| (LOOP_VINFO_NITERS_KNOWN_P (loop_vinfo)
&& LOOP_VINFO_INT_NITERS (loop_vinfo) % vectorization_factor != 0))
- vect_do_peeling_for_loop_bound (loop_vinfo, &ratio, loops);
+ vect_do_peeling_for_loop_bound (loop_vinfo, &ratio);
else
ratio = build_int_cst (TREE_TYPE (LOOP_VINFO_NITERS (loop_vinfo)),
LOOP_VINFO_INT_NITERS (loop_vinfo) / vectorization_factor);
gcc_assert (EDGE_COUNT (loop->header->preds) == 2);
- loop_split_edge_with (loop_preheader_edge (loop), NULL);
-
+ split_edge (loop_preheader_edge (loop));
/* FORNOW: the vectorizer supports only loops which body consist
of one basic block (header + empty latch). When the vectorizer will
if (vect_print_dump_info (REPORT_DETAILS))
fprintf (vect_dump, "transform statement.");
- is_store = vect_transform_stmt (stmt, &si);
- if (is_store)
- {
- /* Free the attached stmt_vec_info and remove the stmt. */
- stmt_ann_t ann = stmt_ann (stmt);
- free (stmt_info);
- set_stmt_info (ann, NULL);
- bsi_remove (&si, true);
- continue;
+ strided_store = false;
+ is_store = vect_transform_stmt (stmt, &si, &strided_store);
+ if (is_store)
+ {
+ stmt_ann_t ann;
+ if (DR_GROUP_FIRST_DR (stmt_info))
+ {
+ /* Interleaving. If IS_STORE is TRUE, the vectorization of the
+ interleaving chain was completed - free all the stores in
+ the chain. */
+ tree next = DR_GROUP_FIRST_DR (stmt_info);
+ tree tmp;
+ stmt_vec_info next_stmt_info;
+
+ while (next)
+ {
+ next_stmt_info = vinfo_for_stmt (next);
+ /* Free the attached stmt_vec_info and remove the stmt. */
+ ann = stmt_ann (next);
+ tmp = DR_GROUP_NEXT_DR (next_stmt_info);
+ free (next_stmt_info);
+ set_stmt_info (ann, NULL);
+ next = tmp;
+ }
+ bsi_remove (&si, true);
+ continue;
+ }
+ else
+ {
+ /* Free the attached stmt_vec_info and remove the stmt. */
+ ann = stmt_ann (stmt);
+ free (stmt_info);
+ set_stmt_info (ann, NULL);
+ bsi_remove (&si, true);
+ continue;
+ }
}
-
+ else
+ {
+ if (strided_store)
+ {
+ /* This is case of skipped interleaved store. We don't free
+ its stmt_vec_info. */
+ bsi_remove (&si, true);
+ continue;
+ }
+ }
bsi_next (&si);
} /* stmts in BB */
} /* BBs in loop */
OSDN Git Service
