/* Statement Analysis and Transformation for Vectorization
- Copyright (C) 2003, 2004, 2005, 2006, 2007, 2008, 2009 Free Software
- Foundation, Inc.
- Contributed by Dorit Naishlos <dorit@il.ibm.com>
+ Copyright (C) 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010
+ Free Software Foundation, Inc.
+ Contributed by Dorit Naishlos <dorit@il.ibm.com>
and Ira Rosen <irar@il.ibm.com>
This file is part of GCC.
{
gimple pattern_stmt;
- /* This is the last stmt in a sequence that was detected as a
+ /* This is the last stmt in a sequence that was detected as a
pattern that can potentially be vectorized. Don't mark the stmt
as relevant/live because it's not going to be vectorized.
Instead mark the pattern-stmt that replaces it. */
*live_p = false;
/* cond stmt other than loop exit cond. */
- if (is_ctrl_stmt (stmt)
- && STMT_VINFO_TYPE (vinfo_for_stmt (stmt))
- != loop_exit_ctrl_vec_info_type)
+ if (is_ctrl_stmt (stmt)
+ && STMT_VINFO_TYPE (vinfo_for_stmt (stmt))
+ != loop_exit_ctrl_vec_info_type)
*relevant = vect_used_in_scope;
/* changing memory. */
if (vect_print_dump_info (REPORT_DETAILS))
fprintf (vect_dump, "vec_stmt_relevant_p: used out of loop.");
+ if (is_gimple_debug (USE_STMT (use_p)))
+ continue;
+
/* We expect all such uses to be in the loop exit phis
(because of loop closed form) */
gcc_assert (gimple_code (USE_STMT (use_p)) == GIMPLE_PHI);
}
-/* Function exist_non_indexing_operands_for_use_p
+/* Function exist_non_indexing_operands_for_use_p
- USE is one of the uses attached to STMT. Check if USE is
+ USE is one of the uses attached to STMT. Check if USE is
used in STMT for anything other than indexing an array. */
static bool
{
tree operand;
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
-
+
/* USE corresponds to some operand in STMT. If there is no data
reference in STMT, then any operand that corresponds to USE
is not indexing an array. */
if (!STMT_VINFO_DATA_REF (stmt_info))
return true;
-
+
/* STMT has a data_ref. FORNOW this means that its of one of
the following forms:
-1- ARRAY_REF = var
(This should have been verified in analyze_data_refs).
'var' in the second case corresponds to a def, not a use,
- so USE cannot correspond to any operands that are not used
+ so USE cannot correspond to any operands that are not used
for array indexing.
Therefore, all we need to check is if STMT falls into the
first case, and whether var corresponds to USE. */
-
- if (TREE_CODE (gimple_assign_lhs (stmt)) == SSA_NAME)
- return false;
if (!gimple_assign_copy_p (stmt))
return false;
+ if (TREE_CODE (gimple_assign_lhs (stmt)) == SSA_NAME)
+ return false;
operand = gimple_assign_rhs1 (stmt);
-
if (TREE_CODE (operand) != SSA_NAME)
return false;
}
-/*
+/*
Function process_use.
Inputs:
- a USE in STMT in a loop represented by LOOP_VINFO
- - LIVE_P, RELEVANT - enum values to be set in the STMT_VINFO of the stmt
+ - LIVE_P, RELEVANT - enum values to be set in the STMT_VINFO of the stmt
that defined USE. This is done by calling mark_relevant and passing it
the WORKLIST (to add DEF_STMT to the WORKLIST in case it is relevant).
STMT_VINFO_RELEVANT (DEF_STMT_info) <-- relevant
Exceptions:
- case 1: If USE is used only for address computations (e.g. array indexing),
- which does not need to be directly vectorized, then the liveness/relevance
+ which does not need to be directly vectorized, then the liveness/relevance
of the respective DEF_STMT is left unchanged.
- - case 2: If STMT is a reduction phi and DEF_STMT is a reduction stmt, we
- skip DEF_STMT cause it had already been processed.
+ - case 2: If STMT is a reduction phi and DEF_STMT is a reduction stmt, we
+ skip DEF_STMT cause it had already been processed.
- case 3: If DEF_STMT and STMT are in different nests, then "relevant" will
be modified accordingly.
Return true if everything is as expected. Return false otherwise. */
static bool
-process_use (gimple stmt, tree use, loop_vec_info loop_vinfo, bool live_p,
+process_use (gimple stmt, tree use, loop_vec_info loop_vinfo, bool live_p,
enum vect_relevant relevant, VEC(gimple,heap) **worklist)
{
struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
gimple def_stmt;
enum vect_def_type dt;
- /* case 1: we are only interested in uses that need to be vectorized. Uses
+ /* case 1: we are only interested in uses that need to be vectorized. Uses
that are used for address computation are not considered relevant. */
if (!exist_non_indexing_operands_for_use_p (use, stmt))
return true;
if (!vect_is_simple_use (use, loop_vinfo, NULL, &def_stmt, &def, &dt))
- {
+ {
if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS))
fprintf (vect_dump, "not vectorized: unsupported use in stmt.");
return false;
return true;
}
- /* case 2: A reduction phi (STMT) defined by a reduction stmt (DEF_STMT).
- DEF_STMT must have already been processed, because this should be the
- only way that STMT, which is a reduction-phi, was put in the worklist,
- as there should be no other uses for DEF_STMT in the loop. So we just
+ /* case 2: A reduction phi (STMT) defined by a reduction stmt (DEF_STMT).
+ DEF_STMT must have already been processed, because this should be the
+ only way that STMT, which is a reduction-phi, was put in the worklist,
+ as there should be no other uses for DEF_STMT in the loop. So we just
check that everything is as expected, and we are done. */
dstmt_vinfo = vinfo_for_stmt (def_stmt);
bb = gimple_bb (stmt);
if (STMT_VINFO_IN_PATTERN_P (dstmt_vinfo))
dstmt_vinfo = vinfo_for_stmt (STMT_VINFO_RELATED_STMT (dstmt_vinfo));
gcc_assert (STMT_VINFO_RELEVANT (dstmt_vinfo) < vect_used_by_reduction);
- gcc_assert (STMT_VINFO_LIVE_P (dstmt_vinfo)
+ gcc_assert (STMT_VINFO_LIVE_P (dstmt_vinfo)
|| STMT_VINFO_RELEVANT (dstmt_vinfo) > vect_unused_in_scope);
return true;
}
default:
gcc_unreachable ();
- }
+ }
}
/* case 3b: inner-loop stmt defining an outer-loop stmt:
switch (relevant)
{
case vect_unused_in_scope:
- relevant = (STMT_VINFO_DEF_TYPE (stmt_vinfo) == vect_reduction_def
+ relevant = (STMT_VINFO_DEF_TYPE (stmt_vinfo) == vect_reduction_def
|| STMT_VINFO_DEF_TYPE (stmt_vinfo) == vect_double_reduction_def) ?
vect_used_in_outer_by_reduction : vect_unused_in_scope;
break;
{
bb = bbs[i];
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, "init: stmt relevant? ");
print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM);
- }
+ }
if (vect_stmt_relevant_p (stmt, loop_vinfo, &relevant, &live_p))
vect_mark_relevant (&worklist, stmt, relevant, live_p);
print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM);
}
- /* Examine the USEs of STMT. For each USE, mark the stmt that defines it
- (DEF_STMT) as relevant/irrelevant and live/dead according to the
+ /* Examine the USEs of STMT. For each USE, mark the stmt that defines it
+ (DEF_STMT) as relevant/irrelevant and live/dead according to the
liveness and relevance properties of STMT. */
stmt_vinfo = vinfo_for_stmt (stmt);
relevant = STMT_VINFO_RELEVANT (stmt_vinfo);
live_p = false
relevant = vect_used_by_reduction
This is because we distinguish between two kinds of relevant stmts -
- those that are used by a reduction computation, and those that are
- (also) used by a regular computation. This allows us later on to
- identify stmts that are used solely by a reduction, and therefore the
+ those that are used by a reduction computation, and those that are
+ (also) used by a regular computation. This allows us later on to
+ identify stmts that are used solely by a reduction, and therefore the
order of the results that they produce does not have to be kept. */
def_type = STMT_VINFO_DEF_TYPE (stmt_vinfo);
return false;
}
- live_p = false;
+ live_p = false;
break;
-
+
case vect_nested_cycle:
if (tmp_relevant != vect_unused_in_scope
&& tmp_relevant != vect_used_in_outer_by_reduction
return false;
}
- live_p = false;
- break;
-
+ live_p = false;
+ break;
+
case vect_double_reduction_def:
if (tmp_relevant != vect_unused_in_scope
&& tmp_relevant != vect_used_by_reduction)
}
live_p = false;
- break;
+ break;
default:
break;
}
-
+
FOR_EACH_PHI_OR_STMT_USE (use_p, stmt, iter, SSA_OP_USE)
{
tree op = USE_FROM_PTR (use_p);
}
}
-/* Function vect_model_simple_cost.
+/* Function vect_model_simple_cost.
- Models cost for simple operations, i.e. those that only emit ncopies of a
+ Models cost for simple operations, i.e. those that only emit ncopies of a
single op. Right now, this does not account for multiple insns that could
be generated for the single vector op. We will handle that shortly. */
void
-vect_model_simple_cost (stmt_vec_info stmt_info, int ncopies,
+vect_model_simple_cost (stmt_vec_info stmt_info, int ncopies,
enum vect_def_type *dt, slp_tree slp_node)
{
int i;
for (i = 0; i < 2; i++)
{
if (dt[i] == vect_constant_def || dt[i] == vect_external_def)
- outside_cost += TARG_SCALAR_TO_VEC_COST;
+ outside_cost += TARG_SCALAR_TO_VEC_COST;
}
-
+
if (vect_print_dump_info (REPORT_COST))
fprintf (vect_dump, "vect_model_simple_cost: inside_cost = %d, "
"outside_cost = %d .", inside_cost, outside_cost);
}
-/* Function vect_cost_strided_group_size
-
+/* Function vect_cost_strided_group_size
+
For strided load or store, return the group_size only if it is the first
load or store of a group, else return 1. This ensures that group size is
only returned once per group. */
has the overhead of the strided access attributed to it. */
void
-vect_model_store_cost (stmt_vec_info stmt_info, int ncopies,
+vect_model_store_cost (stmt_vec_info stmt_info, int ncopies,
enum vect_def_type dt, slp_tree slp_node)
{
int group_size;
outside_cost = TARG_SCALAR_TO_VEC_COST;
/* Strided access? */
- if (DR_GROUP_FIRST_DR (stmt_info) && !slp_node)
+ if (DR_GROUP_FIRST_DR (stmt_info) && !slp_node)
group_size = vect_cost_strided_group_size (stmt_info);
/* Not a strided access. */
else
group_size = 1;
- /* Is this an access in a group of stores, which provide strided access?
+ /* Is this an access in a group of stores, which provide strided access?
If so, add in the cost of the permutes. */
- if (group_size > 1)
+ if (group_size > 1)
{
/* Uses a high and low interleave operation for each needed permute. */
- inside_cost = ncopies * exact_log2(group_size) * group_size
+ inside_cost = ncopies * exact_log2(group_size) * group_size
* TARG_VEC_STMT_COST;
if (vect_print_dump_info (REPORT_COST))
Models cost for loads. In the case of strided accesses, the last access
has the overhead of the strided access attributed to it. Since unaligned
- accesses are supported for loads, we also account for the costs of the
+ accesses are supported for loads, we also account for the costs of the
access scheme chosen. */
void
vect_model_load_cost (stmt_vec_info stmt_info, int ncopies, slp_tree slp_node)
-
+
{
int group_size;
int alignment_support_cheme;
alignment_support_cheme = vect_supportable_dr_alignment (first_dr);
- /* Is this an access in a group of loads providing strided access?
+ /* Is this an access in a group of loads providing strided access?
If so, add in the cost of the permutes. */
- if (group_size > 1)
+ if (group_size > 1)
{
/* Uses an even and odd extract operations for each needed permute. */
inside_cost = ncopies * exact_log2(group_size) * group_size
default:
gcc_unreachable ();
}
-
+
if (vect_print_dump_info (REPORT_COST))
fprintf (vect_dump, "vect_model_load_cost: inside_cost = %d, "
"outside_cost = %d .", inside_cost, outside_cost);
Insert a new stmt (INIT_STMT) that initializes a new vector variable with
the vector elements of VECTOR_VAR. Place the initialization at BSI if it
is not NULL. Otherwise, place the initialization at the loop preheader.
- Return the DEF of INIT_STMT.
+ Return the DEF of INIT_STMT.
It will be used in the vectorization of STMT. */
tree
edge pe;
tree new_temp;
basic_block new_bb;
-
+
new_var = vect_get_new_vect_var (vector_type, vect_simple_var, "cst_");
- add_referenced_var (new_var);
+ add_referenced_var (new_var);
init_stmt = gimple_build_assign (new_var, vector_var);
new_temp = make_ssa_name (new_var, init_stmt);
gimple_assign_set_lhs (init_stmt, new_temp);
else
{
loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_vinfo);
-
+
if (loop_vinfo)
{
struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
if (nested_in_vect_loop_p (loop, stmt))
loop = loop->inner;
-
+
pe = loop_preheader_edge (loop);
new_bb = gsi_insert_on_edge_immediate (pe, init_stmt);
gcc_assert (!new_bb);
print_generic_expr (vect_dump, op, TDF_SLIM);
}
- is_simple_use = vect_is_simple_use (op, loop_vinfo, NULL, &def_stmt, &def,
+ is_simple_use = vect_is_simple_use (op, loop_vinfo, NULL, &def_stmt, &def,
&dt);
gcc_assert (is_simple_use);
if (vect_print_dump_info (REPORT_DETAILS))
vector_type = get_vectype_for_scalar_type (TREE_TYPE (op));
gcc_assert (vector_type);
- if (scalar_def)
+ if (scalar_def)
*scalar_def = op;
/* Create 'vect_cst_ = {cst,cst,...,cst}' */
gcc_assert (vector_type);
nunits = TYPE_VECTOR_SUBPARTS (vector_type);
- if (scalar_def)
+ if (scalar_def)
*scalar_def = def;
/* Create 'vec_inv = {inv,inv,..,inv}' */
/* Case 3: operand is defined inside the loop. */
case vect_internal_def:
{
- if (scalar_def)
+ if (scalar_def)
*scalar_def = NULL/* FIXME tuples: def_stmt*/;
/* Get the def from the vectorized stmt. */
struct loop *loop;
gcc_assert (gimple_code (def_stmt) == GIMPLE_PHI);
- loop = (gimple_bb (def_stmt))->loop_father;
+ 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));
/* Function vect_get_vec_def_for_stmt_copy
- Return a vector-def for an operand. This function is used when the
- vectorized stmt to be created (by the caller to this function) is a "copy"
- created in case the vectorized result cannot fit in one vector, and several
- copies of the vector-stmt are required. In this case the vector-def is
+ Return a vector-def for an operand. This function is used when the
+ vectorized stmt to be created (by the caller to this function) is a "copy"
+ created in case the vectorized result cannot fit in one vector, and several
+ copies of the vector-stmt are required. In this case the vector-def is
retrieved from the vector stmt recorded in the STMT_VINFO_RELATED_STMT field
- of the stmt that defines VEC_OPRND.
+ of the stmt that defines VEC_OPRND.
DT is the type of the vector def VEC_OPRND.
Context:
In case the vectorization factor (VF) is bigger than the number
of elements that can fit in a vectype (nunits), we have to generate
more than one vector stmt to vectorize the scalar stmt. This situation
- arises when there are multiple data-types operated upon in the loop; the
+ arises when there are multiple data-types operated upon in the loop; the
smallest data-type determines the VF, and as a result, when vectorizing
stmts operating on wider types we need to create 'VF/nunits' "copies" of the
vector stmt (each computing a vector of 'nunits' results, and together
- computing 'VF' results in each iteration). This function is called when
+ computing 'VF' results in each iteration). This function is called when
vectorizing such a stmt (e.g. vectorizing S2 in the illustration below, in
which VF=16 and nunits=4, so the number of copies required is 4):
scalar stmt: vectorized into: STMT_VINFO_RELATED_STMT
-
+
S1: x = load VS1.0: vx.0 = memref0 VS1.1
VS1.1: vx.1 = memref1 VS1.2
VS1.2: vx.2 = memref2 VS1.3
- VS1.3: vx.3 = memref3
+ VS1.3: vx.3 = memref3
S2: z = x + ... VSnew.0: vz0 = vx.0 + ... VSnew.1
VSnew.1: vz1 = vx.1 + ... VSnew.2
The vectorization of S1 is explained in vectorizable_load.
The vectorization of S2:
- To create the first vector-stmt out of the 4 copies - VSnew.0 -
- the function 'vect_get_vec_def_for_operand' is called to
+ To create the first vector-stmt out of the 4 copies - VSnew.0 -
+ the function 'vect_get_vec_def_for_operand' is called to
get the relevant vector-def for each operand of S2. For operand x it
returns the vector-def 'vx.0'.
- To create the remaining copies of the vector-stmt (VSnew.j), this
- function is called to get the relevant vector-def for each operand. It is
- obtained from the respective VS1.j stmt, which is recorded in the
+ To create the remaining copies of the vector-stmt (VSnew.j), this
+ function is called to get the relevant vector-def for each operand. It is
+ obtained from the respective VS1.j stmt, which is recorded in the
STMT_VINFO_RELATED_STMT field of the stmt that defines VEC_OPRND.
- For example, to obtain the vector-def 'vx.1' in order to create the
- vector stmt 'VSnew.1', this function is called with VEC_OPRND='vx.0'.
- Given 'vx0' we obtain the stmt that defines it ('VS1.0'); from the
+ For example, to obtain the vector-def 'vx.1' in order to create the
+ vector stmt 'VSnew.1', this function is called with VEC_OPRND='vx.0'.
+ Given 'vx0' we obtain the stmt that defines it ('VS1.0'); from the
STMT_VINFO_RELATED_STMT field of 'VS1.0' we obtain the next copy - 'VS1.1',
and return its def ('vx.1').
Overall, to create the above sequence this function will be called 3 times:
stmt. See vect_get_vec_def_for_stmt_copy() for details. */
static void
-vect_get_vec_defs_for_stmt_copy (enum vect_def_type *dt,
- VEC(tree,heap) **vec_oprnds0,
+vect_get_vec_defs_for_stmt_copy (enum vect_def_type *dt,
+ VEC(tree,heap) **vec_oprnds0,
VEC(tree,heap) **vec_oprnds1)
{
tree vec_oprnd = VEC_pop (tree, *vec_oprnds0);
slp_tree slp_node)
{
if (slp_node)
- vect_get_slp_defs (slp_node, vec_oprnds0, vec_oprnds1);
+ vect_get_slp_defs (slp_node, vec_oprnds0, vec_oprnds1, -1);
else
{
tree vec_oprnd;
- *vec_oprnds0 = VEC_alloc (tree, heap, 1);
- vec_oprnd = vect_get_vec_def_for_operand (op0, stmt, NULL);
+ *vec_oprnds0 = VEC_alloc (tree, heap, 1);
+ vec_oprnd = vect_get_vec_def_for_operand (op0, stmt, NULL);
VEC_quick_push (tree, *vec_oprnds0, vec_oprnd);
if (op1)
{
- *vec_oprnds1 = VEC_alloc (tree, heap, 1);
- vec_oprnd = vect_get_vec_def_for_operand (op1, stmt, NULL);
+ *vec_oprnds1 = VEC_alloc (tree, heap, 1);
+ vec_oprnd = vect_get_vec_def_for_operand (op1, stmt, NULL);
VEC_quick_push (tree, *vec_oprnds1, vec_oprnd);
}
}
gsi_insert_before (gsi, vec_stmt, GSI_SAME_STMT);
- set_vinfo_for_stmt (vec_stmt, new_stmt_vec_info (vec_stmt, loop_vinfo,
+ set_vinfo_for_stmt (vec_stmt, new_stmt_vec_info (vec_stmt, loop_vinfo,
bb_vinfo));
if (vect_print_dump_info (REPORT_DETAILS))
vectorizable_function (gimple call, tree vectype_out, tree vectype_in)
{
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. */
|| !DECL_BUILT_IN (fndecl))
return NULL_TREE;
- code = DECL_FUNCTION_CODE (fndecl);
- return targetm.vectorize.builtin_vectorized_function (code, vectype_out,
+ return targetm.vectorize.builtin_vectorized_function (fndecl, vectype_out,
vectype_in);
}
/* 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
+ 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. */
int nunits_in;
int nunits_out;
loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
- tree fndecl, new_temp, def, rhs_type, lhs_type;
+ tree fndecl, new_temp, def, rhs_type;
gimple def_stmt;
enum vect_def_type dt[2] = {vect_unknown_def_type, vect_unknown_def_type};
- gimple new_stmt;
+ gimple new_stmt = NULL;
int ncopies, j;
VEC(tree, heap) *vargs = NULL;
enum { NARROW, NONE, WIDEN } modifier;
/* FORNOW: unsupported in basic block SLP. */
gcc_assert (loop_vinfo);
-
+
if (!STMT_VINFO_RELEVANT_P (stmt_info))
return false;
if (TREE_CODE (gimple_call_lhs (stmt)) != SSA_NAME)
return false;
+ vectype_out = STMT_VINFO_VECTYPE (stmt_info);
+
/* Process function arguments. */
rhs_type = NULL_TREE;
+ vectype_in = NULL_TREE;
nargs = gimple_call_num_args (stmt);
/* Bail out if the function has more than two arguments, we
for (i = 0; i < nargs; i++)
{
+ tree opvectype;
+
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))
+ && !types_compatible_p (rhs_type, TREE_TYPE (op)))
{
if (vect_print_dump_info (REPORT_DETAILS))
fprintf (vect_dump, "argument types differ.");
return false;
}
- rhs_type = TREE_TYPE (op);
+ if (!rhs_type)
+ rhs_type = TREE_TYPE (op);
- if (!vect_is_simple_use (op, loop_vinfo, NULL, &def_stmt, &def, &dt[i]))
+ if (!vect_is_simple_use_1 (op, loop_vinfo, NULL,
+ &def_stmt, &def, &dt[i], &opvectype))
{
if (vect_print_dump_info (REPORT_DETAILS))
fprintf (vect_dump, "use not simple.");
return false;
}
- }
- vectype_in = get_vectype_for_scalar_type (rhs_type);
+ if (!vectype_in)
+ vectype_in = opvectype;
+ else if (opvectype
+ && opvectype != vectype_in)
+ {
+ if (vect_print_dump_info (REPORT_DETAILS))
+ fprintf (vect_dump, "argument vector types differ.");
+ return false;
+ }
+ }
+ /* If all arguments are external or constant defs use a vector type with
+ the same size as the output vector type. */
if (!vectype_in)
- return false;
- nunits_in = TYPE_VECTOR_SUBPARTS (vectype_in);
-
- lhs_type = TREE_TYPE (gimple_call_lhs (stmt));
- vectype_out = get_vectype_for_scalar_type (lhs_type);
- if (!vectype_out)
- return false;
- nunits_out = TYPE_VECTOR_SUBPARTS (vectype_out);
+ vectype_in = get_same_sized_vectype (rhs_type, vectype_out);
/* FORNOW */
+ nunits_in = TYPE_VECTOR_SUBPARTS (vectype_in);
+ nunits_out = TYPE_VECTOR_SUBPARTS (vectype_out);
if (nunits_in == nunits_out / 2)
modifier = NARROW;
else if (nunits_out == nunits_in)
vec_oprnd0
= vect_get_vec_def_for_operand (op, stmt, NULL);
else
- vec_oprnd0
- = vect_get_vec_def_for_stmt_copy (dt[nargs], vec_oprnd0);
+ {
+ vec_oprnd0 = gimple_call_arg (new_stmt, i);
+ vec_oprnd0
+ = vect_get_vec_def_for_stmt_copy (dt[i], vec_oprnd0);
+ }
VEC_quick_push (tree, vargs, vec_oprnd0);
}
gimple_call_set_lhs (new_stmt, new_temp);
vect_finish_stmt_generation (stmt, new_stmt, gsi);
+ mark_symbols_for_renaming (new_stmt);
if (j == 0)
STMT_VINFO_VEC_STMT (stmt_info) = *vec_stmt = new_stmt;
vec_oprnd0
= vect_get_vec_def_for_operand (op, stmt, NULL);
vec_oprnd1
- = vect_get_vec_def_for_stmt_copy (dt[nargs], vec_oprnd0);
+ = vect_get_vec_def_for_stmt_copy (dt[i], vec_oprnd0);
}
else
{
+ vec_oprnd1 = gimple_call_arg (new_stmt, 2*i);
vec_oprnd0
- = vect_get_vec_def_for_stmt_copy (dt[nargs], vec_oprnd1);
+ = vect_get_vec_def_for_stmt_copy (dt[i], vec_oprnd1);
vec_oprnd1
- = vect_get_vec_def_for_stmt_copy (dt[nargs], vec_oprnd0);
+ = vect_get_vec_def_for_stmt_copy (dt[i], vec_oprnd0);
}
VEC_quick_push (tree, vargs, vec_oprnd0);
gimple_call_set_lhs (new_stmt, new_temp);
vect_finish_stmt_generation (stmt, new_stmt, gsi);
+ mark_symbols_for_renaming (new_stmt);
if (j == 0)
STMT_VINFO_VEC_STMT (stmt_info) = new_stmt;
/* Function vect_gen_widened_results_half
Create a vector stmt whose code, type, number of arguments, and result
- variable are CODE, 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).
tree vec_oprnd0, tree vec_oprnd1, int op_type,
tree vec_dest, gimple_stmt_iterator *gsi,
gimple stmt)
-{
+{
gimple new_stmt;
- tree new_temp;
-
- /* Generate half of the widened result: */
- if (code == CALL_EXPR)
- {
- /* Target specific support */
+ tree new_temp;
+
+ /* Generate half of the widened result: */
+ if (code == CALL_EXPR)
+ {
+ /* Target specific support */
if (op_type == binary_op)
new_stmt = gimple_build_call (decl, 2, vec_oprnd0, vec_oprnd1);
else
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
+ }
+ else
{
- /* Generic support */
- gcc_assert (op_type == TREE_CODE_LENGTH (code));
+ /* Generic support */
+ gcc_assert (op_type == TREE_CODE_LENGTH (code));
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);
- }
+ }
vect_finish_stmt_generation (stmt, new_stmt, gsi);
return new_stmt;
}
-/* Check if STMT performs a conversion operation, that can be vectorized.
- If VEC_STMT is also passed, vectorize the STMT: create a vectorized
+/* Check if STMT performs a conversion operation, that can be vectorized.
+ If VEC_STMT is also passed, vectorize the STMT: create a vectorized
stmt to replace it, put it in VEC_STMT, and insert it at BSI.
Return FALSE if not a vectorizable STMT, TRUE otherwise. */
int nunits_out;
tree vectype_out, vectype_in;
int ncopies, j;
- tree expr;
- tree rhs_type, lhs_type;
+ tree rhs_type;
tree builtin_decl;
enum { NARROW, NONE, WIDEN } modifier;
int i;
VEC(tree,heap) *vec_oprnds0 = NULL;
tree vop0;
- tree integral_type;
VEC(tree,heap) *dummy = NULL;
int dummy_int;
/* FORNOW: unsupported in basic block SLP. */
gcc_assert (loop_vinfo);
-
+
if (!STMT_VINFO_RELEVANT_P (stmt_info))
return false;
return false;
/* Check types of lhs and rhs. */
+ scalar_dest = gimple_assign_lhs (stmt);
+ vectype_out = STMT_VINFO_VECTYPE (stmt_info);
+
op0 = gimple_assign_rhs1 (stmt);
rhs_type = TREE_TYPE (op0);
- vectype_in = get_vectype_for_scalar_type (rhs_type);
+ /* Check the operands of the operation. */
+ if (!vect_is_simple_use_1 (op0, loop_vinfo, NULL,
+ &def_stmt, &def, &dt[0], &vectype_in))
+ {
+ if (vect_print_dump_info (REPORT_DETAILS))
+ fprintf (vect_dump, "use not simple.");
+ return false;
+ }
+ /* If op0 is an external or constant defs use a vector type of
+ the same size as the output vector type. */
if (!vectype_in)
- return false;
- nunits_in = TYPE_VECTOR_SUBPARTS (vectype_in);
-
- scalar_dest = gimple_assign_lhs (stmt);
- lhs_type = TREE_TYPE (scalar_dest);
- vectype_out = get_vectype_for_scalar_type (lhs_type);
- if (!vectype_out)
- return false;
- nunits_out = TYPE_VECTOR_SUBPARTS (vectype_out);
+ vectype_in = get_same_sized_vectype (rhs_type, vectype_out);
/* FORNOW */
+ nunits_in = TYPE_VECTOR_SUBPARTS (vectype_in);
+ nunits_out = TYPE_VECTOR_SUBPARTS (vectype_out);
if (nunits_in == nunits_out / 2)
modifier = NARROW;
else if (nunits_out == nunits_in)
else
return false;
- if (modifier == NONE)
- gcc_assert (STMT_VINFO_VECTYPE (stmt_info) == vectype_out);
-
- /* Bail out if the types are both integral or non-integral. */
- if ((INTEGRAL_TYPE_P (rhs_type) && INTEGRAL_TYPE_P (lhs_type))
- || (!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
this, so we can safely override NCOPIES with 1 here. */
if (slp_node)
ncopies = 1;
-
+
/* Sanity check: make sure that at least one copy of the vectorized stmt
needs to be generated. */
gcc_assert (ncopies >= 1);
- /* Check the operands of the operation. */
- if (!vect_is_simple_use (op0, loop_vinfo, NULL, &def_stmt, &def, &dt[0]))
- {
- if (vect_print_dump_info (REPORT_DETAILS))
- fprintf (vect_dump, "use not simple.");
- return false;
- }
-
/* Supportable by target? */
if ((modifier == NONE
- && !targetm.vectorize.builtin_conversion (code, integral_type))
+ && !targetm.vectorize.builtin_conversion (code, vectype_out, vectype_in))
|| (modifier == WIDEN
- && !supportable_widening_operation (code, stmt, vectype_in,
+ && !supportable_widening_operation (code, stmt,
+ vectype_out, vectype_in,
&decl1, &decl2,
&code1, &code2,
&dummy_int, &dummy))
|| (modifier == NARROW
- && !supportable_narrowing_operation (code, stmt, vectype_in,
+ && !supportable_narrowing_operation (code, vectype_out, vectype_in,
&code1, &dummy_int, &dummy)))
{
if (vect_print_dump_info (REPORT_DETAILS))
if (modifier != NONE)
{
- STMT_VINFO_VECTYPE (stmt_info) = vectype_in;
/* FORNOW: SLP not supported. */
if (STMT_SLP_TYPE (stmt_info))
- return false;
+ return false;
}
if (!vec_stmt) /* transformation not required. */
for (j = 0; j < ncopies; j++)
{
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, integral_type);
+ targetm.vectorize.builtin_conversion (code,
+ vectype_out, vectype_in);
for (i = 0; VEC_iterate (tree, vec_oprnds0, i, vop0); i++)
- {
+ {
/* Arguments are ready. create the new vector stmt. */
new_stmt = gimple_build_call (builtin_decl, 1, vop0);
new_temp = make_ssa_name (vec_dest, new_stmt);
else
vec_oprnd0 = vect_get_vec_def_for_stmt_copy (dt[0], vec_oprnd0);
- STMT_VINFO_VECTYPE (stmt_info) = vectype_in;
-
/* Generate first half of the widened result: */
new_stmt
- = vect_gen_widened_results_half (code1, decl1,
+ = vect_gen_widened_results_half (code1, decl1,
vec_oprnd0, vec_oprnd1,
unary_op, vec_dest, gsi, stmt);
if (j == 0)
}
/* Arguments are ready. Create the new vector stmt. */
- expr = build2 (code1, vectype_out, vec_oprnd0, vec_oprnd1);
new_stmt = gimple_build_assign_with_ops (code1, vec_dest, vec_oprnd0,
vec_oprnd1);
new_temp = make_ssa_name (vec_dest, new_stmt);
}
if (vec_oprnds0)
- VEC_free (tree, heap, vec_oprnds0);
+ VEC_free (tree, heap, vec_oprnds0);
return true;
}
/* Function vectorizable_assignment.
- Check if STMT performs an assignment (copy) that can be vectorized.
- If VEC_STMT is also passed, vectorize the STMT: create a vectorized
+ Check if STMT performs an assignment (copy) that can be vectorized.
+ If VEC_STMT is also passed, vectorize the STMT: create a vectorized
stmt to replace it, put it in VEC_STMT, and insert it at BSI.
Return FALSE if not a vectorizable STMT, TRUE otherwise. */
enum vect_def_type dt[2] = {vect_unknown_def_type, vect_unknown_def_type};
int nunits = TYPE_VECTOR_SUBPARTS (vectype);
int ncopies;
- int i;
+ int i, j;
VEC(tree,heap) *vec_oprnds = NULL;
tree vop;
bb_vec_info bb_vinfo = STMT_VINFO_BB_VINFO (stmt_info);
+ gimple new_stmt = NULL;
+ stmt_vec_info prev_stmt_info = NULL;
/* Multiple types in SLP are handled by creating the appropriate number of
vectorized stmts for each SLP node. Hence, NCOPIES is always 1 in
ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits;
gcc_assert (ncopies >= 1);
- if (ncopies > 1)
- return false; /* FORNOW */
if (!STMT_VINFO_RELEVANT_P (stmt_info) && !bb_vinfo)
return false;
vec_dest = vect_create_destination_var (scalar_dest, vectype);
/* Handle use. */
- vect_get_vec_defs (op, NULL, stmt, &vec_oprnds, NULL, slp_node);
-
- /* Arguments are ready. create the new vector stmt. */
- for (i = 0; VEC_iterate (tree, vec_oprnds, i, vop); i++)
+ for (j = 0; j < ncopies; j++)
{
- *vec_stmt = gimple_build_assign (vec_dest, vop);
- new_temp = make_ssa_name (vec_dest, *vec_stmt);
- gimple_assign_set_lhs (*vec_stmt, new_temp);
- vect_finish_stmt_generation (stmt, *vec_stmt, gsi);
- STMT_VINFO_VEC_STMT (stmt_info) = *vec_stmt;
+ /* Handle uses. */
+ if (j == 0)
+ vect_get_vec_defs (op, NULL, stmt, &vec_oprnds, NULL, slp_node);
+ else
+ vect_get_vec_defs_for_stmt_copy (dt, &vec_oprnds, NULL);
+
+ /* Arguments are ready. create the new vector stmt. */
+ for (i = 0; VEC_iterate (tree, vec_oprnds, i, vop); i++)
+ {
+ new_stmt = gimple_build_assign (vec_dest, vop);
+ new_temp = make_ssa_name (vec_dest, new_stmt);
+ gimple_assign_set_lhs (new_stmt, new_temp);
+ vect_finish_stmt_generation (stmt, new_stmt, gsi);
+ if (slp_node)
+ VEC_quick_push (gimple, SLP_TREE_VEC_STMTS (slp_node), new_stmt);
+ }
if (slp_node)
- VEC_quick_push (gimple, SLP_TREE_VEC_STMTS (slp_node), *vec_stmt);
- }
-
- VEC_free (tree, heap, vec_oprnds);
+ continue;
+
+ 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);
+ }
+
+ VEC_free (tree, heap, vec_oprnds);
return true;
}
/* Function vectorizable_operation.
- Check if STMT performs a binary or unary operation that can be vectorized.
- If VEC_STMT is also passed, vectorize the STMT: create a vectorized
+ Check if STMT performs a binary or unary operation that can be vectorized.
+ If VEC_STMT is also passed, vectorize the STMT: create a vectorized
stmt to replace it, put it in VEC_STMT, and insert it at BSI.
Return FALSE if not a vectorizable STMT, TRUE otherwise. */
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);
+ tree vectype;
loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
enum tree_code code;
enum machine_mode vec_mode;
enum vect_def_type dt[2] = {vect_unknown_def_type, vect_unknown_def_type};
gimple new_stmt = NULL;
stmt_vec_info prev_stmt_info;
- int nunits_in = TYPE_VECTOR_SUBPARTS (vectype);
+ int nunits_in;
int nunits_out;
tree vectype_out;
int ncopies;
VEC(tree,heap) *vec_oprnds0 = NULL, *vec_oprnds1 = NULL;
tree vop0, vop1;
unsigned int k;
- bool shift_p = false;
bool scalar_shift_arg = false;
bb_vec_info bb_vinfo = STMT_VINFO_BB_VINFO (stmt_info);
int vf;
- if (loop_vinfo)
- vf = LOOP_VINFO_VECT_FACTOR (loop_vinfo);
- else
- /* FORNOW: multiple types are not supported in basic block SLP. */
- vf = 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);
-
if (!STMT_VINFO_RELEVANT_P (stmt_info) && !bb_vinfo)
return false;
if (TREE_CODE (gimple_assign_lhs (stmt)) != SSA_NAME)
return false;
- 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)
- return false;
-
code = gimple_assign_rhs_code (stmt);
/* For pointer addition, we should use the normal plus for
return false;
}
+ scalar_dest = gimple_assign_lhs (stmt);
+ vectype_out = STMT_VINFO_VECTYPE (stmt_info);
+
op0 = gimple_assign_rhs1 (stmt);
- if (!vect_is_simple_use (op0, loop_vinfo, bb_vinfo, &def_stmt, &def, &dt[0]))
+ if (!vect_is_simple_use_1 (op0, loop_vinfo, bb_vinfo,
+ &def_stmt, &def, &dt[0], &vectype))
{
if (vect_print_dump_info (REPORT_DETAILS))
fprintf (vect_dump, "use not simple.");
return false;
}
+ /* If op0 is an external or constant def use a vector type with
+ the same size as the output vector type. */
+ if (!vectype)
+ vectype = get_same_sized_vectype (TREE_TYPE (op0), vectype_out);
+ gcc_assert (vectype);
+
+ nunits_out = TYPE_VECTOR_SUBPARTS (vectype_out);
+ nunits_in = TYPE_VECTOR_SUBPARTS (vectype);
+ if (nunits_out != nunits_in)
+ return false;
if (op_type == binary_op)
{
op1 = gimple_assign_rhs2 (stmt);
- if (!vect_is_simple_use (op1, loop_vinfo, bb_vinfo, &def_stmt, &def,
+ if (!vect_is_simple_use (op1, loop_vinfo, bb_vinfo, &def_stmt, &def,
&dt[1]))
{
if (vect_print_dump_info (REPORT_DETAILS))
}
}
+ if (loop_vinfo)
+ vf = LOOP_VINFO_VECT_FACTOR (loop_vinfo);
+ else
+ vf = 1;
+
+ /* 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);
+
/* If this is a shift/rotate, determine whether the shift amount is a vector,
or scalar. If the shift/rotate amount is a vector, use the vector/vector
shift optabs. */
if (code == LSHIFT_EXPR || code == RSHIFT_EXPR || code == LROTATE_EXPR
|| code == RROTATE_EXPR)
{
- shift_p = true;
-
/* vector shifted by vector */
if (dt[1] == vect_internal_def)
{
/* Handle def. */
vec_dest = vect_create_destination_var (scalar_dest, vectype);
- /* Allocate VECs for vector operands. In case of SLP, vector operands are
+ /* Allocate VECs for vector operands. In case of SLP, vector operands are
created in the previous stages of the recursion, so no allocation is
needed, except for the case of shift with scalar shift argument. In that
case we store the scalar operand in VEC_OPRNDS1 for every vector stmt to
be created to vectorize the SLP group, i.e., SLP_NODE->VEC_STMTS_SIZE.
- In case of loop-based vectorization we allocate VECs of size 1. We
- allocate VEC_OPRNDS1 only in case of binary operation. */
+ In case of loop-based vectorization we allocate VECs of size 1. We
+ allocate VEC_OPRNDS1 only in case of binary operation. */
if (!slp_node)
{
vec_oprnds0 = VEC_alloc (tree, heap, 1);
vec_oprnds1 = VEC_alloc (tree, heap, 1);
}
else if (scalar_shift_arg)
- vec_oprnds1 = VEC_alloc (tree, heap, slp_node->vec_stmts_size);
+ vec_oprnds1 = VEC_alloc (tree, heap, slp_node->vec_stmts_size);
/* 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
{
if (op_type == binary_op && scalar_shift_arg)
{
- /* Vector shl and shr insn patterns can be defined with scalar
- operand 2 (shift operand). In this case, use constant or loop
- invariant op1 directly, without extending it to vector mode
+ /* Vector shl and shr insn patterns can be defined with scalar
+ operand 2 (shift operand). In this case, use constant or loop
+ invariant op1 directly, without extending it to vector mode
first. */
optab_op2_mode = insn_data[icode].operand[2].mode;
if (!VECTOR_MODE_P (optab_op2_mode))
{
/* Store vec_oprnd1 for every vector stmt to be created
for SLP_NODE. We check during the analysis that all the
- shift arguments are the same.
- TODO: Allow different constants for different vector
- stmts generated for an SLP instance. */
+ shift arguments are the same.
+ TODO: Allow different constants for different vector
+ stmts generated for an SLP instance. */
for (k = 0; k < slp_node->vec_stmts_size - 1; k++)
VEC_quick_push (tree, vec_oprnds1, vec_oprnd1);
}
}
}
-
- /* vec_oprnd1 is available if operand 1 should be of a scalar-type
- (a special case for certain kind of vector shifts); otherwise,
+
+ /* vec_oprnd1 is available if operand 1 should be of a scalar-type
+ (a special case for certain kind of vector shifts); otherwise,
operand 1 should be of a vector type (the usual case). */
if (op_type == binary_op && !vec_oprnd1)
- vect_get_vec_defs (op0, op1, stmt, &vec_oprnds0, &vec_oprnds1,
+ vect_get_vec_defs (op0, op1, stmt, &vec_oprnds0, &vec_oprnds1,
slp_node);
else
- vect_get_vec_defs (op0, NULL_TREE, stmt, &vec_oprnds0, NULL,
+ vect_get_vec_defs (op0, NULL_TREE, stmt, &vec_oprnds0, NULL,
slp_node);
}
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
+ 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,
+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)
+ 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);
/* 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
+ /* 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);
+ 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
+ For multi-step conversions store the resulting vectors and call the function
recursively. */
static void
gimple new_stmt;
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
- vec_dest = VEC_pop (tree, vec_dsts);
+ vec_dest = VEC_pop (tree, vec_dsts);
for (i = 0; i < VEC_length (tree, *vec_oprnds); i += 2)
{
if (multi_step_cvt)
/* Store the resulting vector for next recursive call. */
- VEC_replace (tree, *vec_oprnds, i/2, new_tmp);
+ VEC_replace (tree, *vec_oprnds, i/2, new_tmp);
else
{
- /* This is the last step of the conversion sequence. Store the
+ /* 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)
}
/* For multi-step demotion operations we first generate demotion operations
- from the source type to the intermediate types, and then combine the
+ 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,
+ vect_create_vectorized_demotion_stmts (vec_oprnds, multi_step_cvt - 1,
stmt, vec_dsts, gsi, slp_node,
code, prev_stmt_info);
}
if (!CONVERT_EXPR_CODE_P (code))
return false;
+ scalar_dest = gimple_assign_lhs (stmt);
+ vectype_out = STMT_VINFO_VECTYPE (stmt_info);
+
+ /* Check the operands of the operation. */
op0 = gimple_assign_rhs1 (stmt);
- vectype_in = get_vectype_for_scalar_type (TREE_TYPE (op0));
+ 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))
+ && CONVERT_EXPR_CODE_P (code))))
+ return false;
+ if (!vect_is_simple_use_1 (op0, loop_vinfo, NULL,
+ &def_stmt, &def, &dt[0], &vectype_in))
+ {
+ if (vect_print_dump_info (REPORT_DETAILS))
+ fprintf (vect_dump, "use not simple.");
+ return false;
+ }
+ /* If op0 is an external def use a vector type with the
+ same size as the output vector type if possible. */
+ if (!vectype_in)
+ vectype_in = get_same_sized_vectype (TREE_TYPE (op0), vectype_out);
if (!vectype_in)
return false;
- nunits_in = TYPE_VECTOR_SUBPARTS (vectype_in);
- scalar_dest = gimple_assign_lhs (stmt);
- vectype_out = get_vectype_for_scalar_type (TREE_TYPE (scalar_dest));
- if (!vectype_out)
- return false;
+ nunits_in = TYPE_VECTOR_SUBPARTS (vectype_in);
nunits_out = TYPE_VECTOR_SUBPARTS (vectype_out);
if (nunits_in >= nunits_out)
return false;
ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits_out;
gcc_assert (ncopies >= 1);
- 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))
- && CONVERT_EXPR_CODE_P (code))))
- return false;
-
- /* Check the operands of the operation. */
- if (!vect_is_simple_use (op0, loop_vinfo, NULL, &def_stmt, &def, &dt[0]))
- {
- if (vect_print_dump_info (REPORT_DETAILS))
- fprintf (vect_dump, "use not simple.");
- return false;
- }
-
/* Supportable by target? */
- if (!supportable_narrowing_operation (code, stmt, vectype_in, &code1,
- &multi_step_cvt, &interm_types))
+ if (!supportable_narrowing_operation (code, vectype_out, vectype_in,
+ &code1, &multi_step_cvt, &interm_types))
return false;
- STMT_VINFO_VECTYPE (stmt_info) = vectype_in;
-
if (!vec_stmt) /* transformation not required. */
{
STMT_VINFO_TYPE (stmt_info) = type_demotion_vec_info_type;
fprintf (vect_dump, "transform type demotion operation. ncopies = %d.",
ncopies);
- /* In case of multi-step demotion, we first generate demotion operations to
- the intermediate types, and then from that types to the final one.
+ /* 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
+ 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;
+ 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,
+ vec_dest = vect_create_destination_var (scalar_dest,
intermediate_type);
VEC_quick_push (tree, vec_dsts, vec_dest);
}
{
/* Handle uses. */
if (slp_node)
- vect_get_slp_defs (slp_node, &vec_oprnds0, NULL);
+ vect_get_slp_defs (slp_node, &vec_oprnds0, NULL, -1);
else
{
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_get_loop_based_defs (&last_oprnd, stmt, dt[0], &vec_oprnds0,
vect_pow2 (multi_step_cvt) - 1);
}
/* Arguments are ready. Create the new vector stmts. */
tmp_vec_dsts = VEC_copy (tree, heap, vec_dsts);
- vect_create_vectorized_demotion_stmts (&vec_oprnds0,
+ vect_create_vectorized_demotion_stmts (&vec_oprnds0,
multi_step_cvt, stmt, tmp_vec_dsts,
- gsi, slp_node, code1,
+ gsi, slp_node, code1,
&prev_stmt_info);
}
/* Create vectorized promotion statements for vector operands from VEC_OPRNDS0
- and VEC_OPRNDS1 (for binary operations). For multi-step conversions store
+ and VEC_OPRNDS1 (for binary operations). For multi-step conversions store
the resulting vectors and call the function recursively. */
static void
VEC (tree, heap) *vec_dsts,
gimple_stmt_iterator *gsi,
slp_tree slp_node, enum tree_code code1,
- enum tree_code code2, tree decl1,
+ enum tree_code code2, tree decl1,
tree decl2, int op_type,
stmt_vec_info *prev_stmt_info)
{
vop1 = NULL_TREE;
/* Generate the two halves of promotion operation. */
- new_stmt1 = vect_gen_widened_results_half (code1, decl1, vop0, vop1,
+ 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 (multi_step_cvt)
{
- /* For multi-step promotion operation we first generate we call the
+ /* 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. */
prev_stmt_info);
}
}
-
+
/* Function vectorizable_type_promotion
loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
enum tree_code code, code1 = ERROR_MARK, code2 = ERROR_MARK;
tree decl1 = NULL_TREE, decl2 = NULL_TREE;
- int op_type;
+ int op_type;
tree def;
gimple def_stmt;
enum vect_def_type dt[2] = {vect_unknown_def_type, vect_unknown_def_type};
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;
-
+
/* FORNOW: not supported by basic block SLP vectorization. */
gcc_assert (loop_vinfo);
-
+
if (!STMT_VINFO_RELEVANT_P (stmt_info))
return false;
&& code != WIDEN_MULT_EXPR)
return false;
+ scalar_dest = gimple_assign_lhs (stmt);
+ vectype_out = STMT_VINFO_VECTYPE (stmt_info);
+
+ /* Check the operands of the operation. */
op0 = gimple_assign_rhs1 (stmt);
- vectype_in = get_vectype_for_scalar_type (TREE_TYPE (op0));
+ 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))
+ && CONVERT_EXPR_CODE_P (code))))
+ return false;
+ if (!vect_is_simple_use_1 (op0, loop_vinfo, NULL,
+ &def_stmt, &def, &dt[0], &vectype_in))
+ {
+ if (vect_print_dump_info (REPORT_DETAILS))
+ fprintf (vect_dump, "use not simple.");
+ return false;
+ }
+ /* If op0 is an external or constant def use a vector type with
+ the same size as the output vector type. */
+ if (!vectype_in)
+ vectype_in = get_same_sized_vectype (TREE_TYPE (op0), vectype_out);
if (!vectype_in)
return false;
- nunits_in = TYPE_VECTOR_SUBPARTS (vectype_in);
- scalar_dest = gimple_assign_lhs (stmt);
- vectype_out = get_vectype_for_scalar_type (TREE_TYPE (scalar_dest));
- if (!vectype_out)
- return false;
+ nunits_in = TYPE_VECTOR_SUBPARTS (vectype_in);
nunits_out = TYPE_VECTOR_SUBPARTS (vectype_out);
if (nunits_in <= nunits_out)
return false;
gcc_assert (ncopies >= 1);
- 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))
- && CONVERT_EXPR_CODE_P (code))))
- return false;
-
- /* Check the operands of the operation. */
- if (!vect_is_simple_use (op0, loop_vinfo, NULL, &def_stmt, &def, &dt[0]))
- {
- if (vect_print_dump_info (REPORT_DETAILS))
- fprintf (vect_dump, "use not simple.");
- return false;
- }
-
op_type = TREE_CODE_LENGTH (code);
if (op_type == binary_op)
{
}
/* Supportable by target? */
- if (!supportable_widening_operation (code, stmt, vectype_in,
+ if (!supportable_widening_operation (code, stmt, vectype_out, vectype_in,
&decl1, &decl2, &code1, &code2,
&multi_step_cvt, &interm_types))
return false;
architecture. */
gcc_assert (!(multi_step_cvt && op_type == binary_op));
- STMT_VINFO_VECTYPE (stmt_info) = vectype_in;
-
if (!vec_stmt) /* transformation not required. */
{
STMT_VINFO_TYPE (stmt_info) = type_promotion_vec_info_type;
ncopies);
/* Handle def. */
- /* In case of multi-step promotion, we first generate promotion operations
+ /* 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
+ 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_quick_push (tree, vec_dsts, vec_dest);
}
}
-
+
if (!slp_node)
{
- vec_oprnds0 = VEC_alloc (tree, heap,
+ 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);
if (j == 0)
{
if (slp_node)
- vect_get_slp_defs (slp_node, &vec_oprnds0, &vec_oprnds1);
+ vect_get_slp_defs (slp_node, &vec_oprnds0, &vec_oprnds1, -1);
else
{
vec_oprnd0 = vect_get_vec_def_for_operand (op0, stmt, NULL);
/* 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,
+ multi_step_cvt, stmt,
tmp_vec_dsts,
gsi, slp_node, code1, code2,
decl1, decl2, op_type,
/* Function vectorizable_store.
- Check if STMT defines a non scalar data-ref (array/pointer/structure) that
- can be vectorized.
- If VEC_STMT is also passed, vectorize the STMT: create a vectorized
+ Check if STMT defines a non scalar data-ref (array/pointer/structure) that
+ can be vectorized.
+ If VEC_STMT is also passed, vectorize the STMT: create a vectorized
stmt to replace it, put it in VEC_STMT, and insert it at BSI.
Return FALSE if not a vectorizable STMT, TRUE otherwise. */
bool inv_p;
VEC(tree,heap) *vec_oprnds = NULL;
bool slp = (slp_node != NULL);
- stmt_vec_info first_stmt_vinfo;
unsigned int vec_num;
bb_vec_info bb_vinfo = STMT_VINFO_BB_VINFO (stmt_info);
/* The scalar rhs type needs to be trivially convertible to the vector
component type. This should always be the case. */
if (!useless_type_conversion_p (TREE_TYPE (vectype), TREE_TYPE (op)))
- {
+ {
if (vect_print_dump_info (REPORT_DETAILS))
fprintf (vect_dump, "??? operands of different types");
return false;
if (!vect_strided_store_supported (vectype)
&& !PURE_SLP_STMT (stmt_info) && !slp)
return false;
-
+
if (first_stmt == stmt)
{
/* STMT is the leader of the group. Check the operands of all the
{
gcc_assert (gimple_assign_single_p (next_stmt));
op = gimple_assign_rhs1 (next_stmt);
- if (!vect_is_simple_use (op, loop_vinfo, bb_vinfo, &def_stmt,
+ if (!vect_is_simple_use (op, loop_vinfo, bb_vinfo, &def_stmt,
&def, &dt))
{
if (vect_print_dump_info (REPORT_DETAILS))
/* 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))
+ if (DR_GROUP_STORE_COUNT (vinfo_for_stmt (first_stmt))
< DR_GROUP_SIZE (vinfo_for_stmt (first_stmt))
&& !slp)
{
}
if (slp)
- strided_store = false;
-
- /* VEC_NUM is the number of vect stmts to be created for this group. */
- if (slp)
- vec_num = SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node);
+ {
+ strided_store = false;
+ /* VEC_NUM is the number of vect stmts to be created for this
+ group. */
+ vec_num = SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node);
+ first_stmt = VEC_index (gimple, SLP_TREE_SCALAR_STMTS (slp_node), 0);
+ first_dr = STMT_VINFO_DATA_REF (vinfo_for_stmt (first_stmt));
+ }
else
+ /* VEC_NUM is the number of vect stmts to be created for this
+ group. */
vec_num = group_size;
}
- else
+ else
{
first_stmt = stmt;
first_dr = dr;
group_size = vec_num = 1;
- first_stmt_vinfo = stmt_info;
}
-
+
if (vect_print_dump_info (REPORT_DETAILS))
fprintf (vect_dump, "transform store. ncopies = %d",ncopies);
/* 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
+ 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):
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
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.
+ STMT_VINFO_RELATED_STMT for the next copies.
*/
prev_stmt_info = NULL;
if (slp)
{
/* Get vectorized arguments for SLP_NODE. */
- vect_get_slp_defs (slp_node, &vec_oprnds, NULL);
+ vect_get_slp_defs (slp_node, &vec_oprnds, NULL, -1);
vec_oprnd = VEC_index (tree, vec_oprnds, 0);
}
else
{
- /* 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
+ /* 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;
+ 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
+ /* 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
&& gimple_assign_single_p (next_stmt));
op = gimple_assign_rhs1 (next_stmt);
- vec_oprnd = vect_get_vec_def_for_operand (op, next_stmt,
+ 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);
+ 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));
}
}
/* We should have catched mismatched types earlier. */
gcc_assert (useless_type_conversion_p (vectype,
TREE_TYPE (vec_oprnd)));
- dataref_ptr = vect_create_data_ref_ptr (first_stmt, NULL, NULL_TREE,
- &dummy, &ptr_incr, false,
+ dataref_ptr = vect_create_data_ref_ptr (first_stmt, NULL, NULL_TREE,
+ &dummy, &ptr_incr, false,
&inv_p);
gcc_assert (bb_vinfo || !inv_p);
}
- else
+ else
{
- /* For interleaved stores we created vectorized defs for all the
- defs stored in OPRNDS in the previous iteration (previous copy).
- DR_CHAIN is then used as an input to vect_permute_store_chain(),
+ /* 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
for (i = 0; i < group_size; i++)
{
op = VEC_index (tree, oprnds, i);
- vect_is_simple_use (op, loop_vinfo, bb_vinfo, &def_stmt, &def,
+ vect_is_simple_use (op, loop_vinfo, bb_vinfo, &def_stmt, &def,
&dt);
- vec_oprnd = vect_get_vec_def_for_stmt_copy (dt, op);
+ vec_oprnd = vect_get_vec_def_for_stmt_copy (dt, op);
VEC_replace(tree, dr_chain, i, vec_oprnd);
VEC_replace(tree, oprnds, i, vec_oprnd);
}
- dataref_ptr =
+ dataref_ptr =
bump_vector_ptr (dataref_ptr, ptr_incr, gsi, stmt, NULL_TREE);
}
if (strided_store)
{
- result_chain = VEC_alloc (tree, heap, group_size);
+ result_chain = VEC_alloc (tree, heap, group_size);
/* Permute. */
if (!vect_permute_store_chain (dr_chain, group_size, stmt, gsi,
&result_chain))
if (slp)
vec_oprnd = VEC_index (tree, vec_oprnds, i);
else if (strided_store)
- /* For strided stores vectorized defs are interleaved in
+ /* For strided stores vectorized defs are interleaved in
vect_permute_store_chain(). */
vec_oprnd = VEC_index (tree, result_chain, i);
if (slp)
continue;
-
+
if (j == 0)
STMT_VINFO_VEC_STMT (stmt_info) = *vec_stmt = new_stmt;
else
}
}
- VEC_free (tree, heap, dr_chain);
- VEC_free (tree, heap, oprnds);
+ VEC_free (tree, heap, dr_chain);
+ VEC_free (tree, heap, oprnds);
if (result_chain)
- VEC_free (tree, heap, result_chain);
+ VEC_free (tree, heap, result_chain);
return true;
}
/* vectorizable_load.
- Check if STMT reads a non scalar data-ref (array/pointer/structure) that
- can be vectorized.
- If VEC_STMT is also passed, vectorize the STMT: create a vectorized
+ Check if STMT reads a non scalar data-ref (array/pointer/structure) that
+ can be vectorized.
+ If VEC_STMT is also passed, vectorize the STMT: create a vectorized
stmt to replace it, put it in VEC_STMT, and insert it at BSI.
Return FALSE if not a vectorizable STMT, TRUE otherwise. */
tree vec_dest = NULL;
tree data_ref = NULL;
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
- stmt_vec_info prev_stmt_info;
+ stmt_vec_info prev_stmt_info;
loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
struct loop *loop = NULL;
struct loop *containing_loop = (gimple_bb (stmt))->loop_father;
vf = LOOP_VINFO_VECT_FACTOR (loop_vinfo);
}
else
- /* FORNOW: multiple types are not supported in basic block SLP. */
- vf = nunits;
+ vf = 1;
/* Multiple types in SLP are handled by creating the appropriate number of
vectorized stmts for each SLP node. Hence, NCOPIES is always 1 in
/* The vector component type needs to be trivially convertible to the
scalar lhs. This should always be the case. */
if (!useless_type_conversion_p (TREE_TYPE (scalar_dest), TREE_TYPE (vectype)))
- {
+ {
if (vect_print_dump_info (REPORT_DETAILS))
fprintf (vect_dump, "??? operands of different types");
return false;
S1: x = load - VS1_0
S2: z = x + 1 - -
- See in documentation in vect_get_vec_def_for_stmt_copy for how the
- information we recorded in RELATED_STMT field is used to vectorize
+ See in documentation in vect_get_vec_def_for_stmt_copy for how the
+ information we recorded in RELATED_STMT field is used to vectorize
stmt S2. */
/* In case of interleaving (non-unit strided access):
S3: x1 = &base + 1
S4: x3 = &base + 3
- Vectorized loads are created in the order of memory accesses
+ Vectorized loads are created in the order of memory accesses
starting from the access of the first stmt of the chain:
VS1: vx0 = &base
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
+ 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. */
}
Otherwise, the data reference is potentially unaligned on a target that
- does not support unaligned accesses (dr_explicit_realign_optimized) -
+ does not support unaligned accesses (dr_explicit_realign_optimized) -
then generate the following code, in which the data in each iteration is
obtained by two vector loads, one from the previous iteration, and one
from the current iteration:
prev_stmt_info = NULL;
for (j = 0; j < ncopies; j++)
- {
+ {
/* 1. Create the vector pointer update chain. */
if (j == 0)
dataref_ptr = vect_create_data_ref_ptr (first_stmt,
- at_loop, offset,
- &dummy, &ptr_incr, false,
+ at_loop, offset,
+ &dummy, &ptr_incr, false,
&inv_p);
else
- dataref_ptr =
+ dataref_ptr =
bump_vector_ptr (dataref_ptr, ptr_incr, gsi, stmt, NULL_TREE);
for (i = 0; i < vec_num; i++)
if (compute_in_loop)
msq = vect_setup_realignment (first_stmt, gsi,
&realignment_token,
- dr_explicit_realign,
+ dr_explicit_realign,
dataref_ptr, NULL);
data_ref = build1 (ALIGN_INDIRECT_REF, vectype, dataref_ptr);
{
gcc_assert (phi);
if (i == vec_num - 1 && j == ncopies - 1)
- add_phi_arg (phi, lsq, loop_latch_edge (containing_loop));
+ add_phi_arg (phi, lsq, loop_latch_edge (containing_loop),
+ UNKNOWN_LOCATION);
msq = lsq;
}
}
/* CHECKME: bitpos depends on endianess? */
bitpos = bitsize_zero_node;
- vec_inv = build3 (BIT_FIELD_REF, scalar_type, new_temp,
+ vec_inv = build3 (BIT_FIELD_REF, scalar_type, new_temp,
bitsize, bitpos);
- vec_dest =
+ vec_dest =
vect_create_destination_var (scalar_dest, NULL_TREE);
new_stmt = gimple_build_assign (vec_dest, vec_inv);
new_temp = make_ssa_name (vec_dest, new_stmt);
if (strided_load)
{
if (!vect_transform_strided_load (stmt, dr_chain, group_size, gsi))
- return false;
+ return false;
*vec_stmt = STMT_VINFO_VEC_STMT (stmt_info);
VEC_free (tree, heap, dr_chain);
}
/* Function vect_is_simple_cond.
-
+
Input:
LOOP - the loop that is being vectorized.
COND - Condition that is checked for simple use.
if (TREE_CODE (lhs) == SSA_NAME)
{
gimple lhs_def_stmt = SSA_NAME_DEF_STMT (lhs);
- if (!vect_is_simple_use (lhs, loop_vinfo, NULL, &lhs_def_stmt, &def,
+ if (!vect_is_simple_use (lhs, loop_vinfo, NULL, &lhs_def_stmt, &def,
&dt))
return false;
}
if (TREE_CODE (rhs) == SSA_NAME)
{
gimple rhs_def_stmt = SSA_NAME_DEF_STMT (rhs);
- if (!vect_is_simple_use (rhs, loop_vinfo, NULL, &rhs_def_stmt, &def,
+ if (!vect_is_simple_use (rhs, loop_vinfo, NULL, &rhs_def_stmt, &def,
&dt))
return false;
}
/* vectorizable_condition.
- Check if STMT is conditional modify expression that can be vectorized.
- If VEC_STMT is also passed, vectorize the STMT: create a vectorized
- stmt using VEC_COND_EXPR to replace it, put it in VEC_STMT, and insert it
- at BSI.
+ Check if STMT is conditional modify expression that can be vectorized.
+ If VEC_STMT is also passed, vectorize the STMT: create a vectorized
+ stmt using VEC_COND_EXPR to replace it, put it in VEC_STMT, and insert it
+ at GSI.
+
+ When STMT is vectorized as nested cycle, REDUC_DEF is the vector variable
+ to be used at REDUC_INDEX (in then clause if REDUC_INDEX is 1, and in
+ else caluse if it is 2).
Return FALSE if not a vectorizable STMT, TRUE otherwise. */
-static bool
+bool
vectorizable_condition (gimple stmt, gimple_stmt_iterator *gsi,
- gimple *vec_stmt)
+ gimple *vec_stmt, tree reduc_def, int reduc_index)
{
tree scalar_dest = NULL_TREE;
tree vec_dest = NULL_TREE;
/* FORNOW: unsupported in basic block SLP. */
gcc_assert (loop_vinfo);
-
+
gcc_assert (ncopies >= 1);
if (ncopies > 1)
return false; /* FORNOW */
if (!STMT_VINFO_RELEVANT_P (stmt_info))
return false;
- if (STMT_VINFO_DEF_TYPE (stmt_info) != vect_internal_def)
+ if (STMT_VINFO_DEF_TYPE (stmt_info) != vect_internal_def
+ && !(STMT_VINFO_DEF_TYPE (stmt_info) == vect_nested_cycle
+ && reduc_def))
return false;
/* FORNOW: SLP not supported. */
/* We do not handle two different vector types for the condition
and the values. */
- if (TREE_TYPE (TREE_OPERAND (cond_expr, 0)) != TREE_TYPE (vectype))
+ if (!types_compatible_p (TREE_TYPE (TREE_OPERAND (cond_expr, 0)),
+ TREE_TYPE (vectype)))
return false;
if (TREE_CODE (then_clause) == SSA_NAME)
{
gimple then_def_stmt = SSA_NAME_DEF_STMT (then_clause);
- if (!vect_is_simple_use (then_clause, loop_vinfo, NULL,
+ if (!vect_is_simple_use (then_clause, loop_vinfo, NULL,
&then_def_stmt, &def, &dt))
return false;
}
- else if (TREE_CODE (then_clause) != INTEGER_CST
+ else if (TREE_CODE (then_clause) != INTEGER_CST
&& TREE_CODE (then_clause) != REAL_CST
&& TREE_CODE (then_clause) != FIXED_CST)
return false;
&else_def_stmt, &def, &dt))
return false;
}
- else if (TREE_CODE (else_clause) != INTEGER_CST
+ else if (TREE_CODE (else_clause) != INTEGER_CST
&& TREE_CODE (else_clause) != REAL_CST
&& TREE_CODE (else_clause) != FIXED_CST)
return false;
vec_mode = TYPE_MODE (vectype);
- if (!vec_stmt)
+ if (!vec_stmt)
{
STMT_VINFO_TYPE (stmt_info) = condition_vec_info_type;
- return expand_vec_cond_expr_p (op, vec_mode);
+ return expand_vec_cond_expr_p (TREE_TYPE (op), vec_mode);
}
/* Transform */
vec_dest = vect_create_destination_var (scalar_dest, vectype);
/* Handle cond expr. */
- vec_cond_lhs =
+ vec_cond_lhs =
vect_get_vec_def_for_operand (TREE_OPERAND (cond_expr, 0), stmt, NULL);
- vec_cond_rhs =
+ vec_cond_rhs =
vect_get_vec_def_for_operand (TREE_OPERAND (cond_expr, 1), stmt, NULL);
- 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);
+ if (reduc_index == 1)
+ vec_then_clause = reduc_def;
+ else
+ vec_then_clause = vect_get_vec_def_for_operand (then_clause, stmt, NULL);
+ if (reduc_index == 2)
+ vec_else_clause = reduc_def;
+ else
+ vec_else_clause = vect_get_vec_def_for_operand (else_clause, stmt, NULL);
/* Arguments are ready. Create the new vector stmt. */
- vec_compare = build2 (TREE_CODE (cond_expr), vectype,
+ vec_compare = build2 (TREE_CODE (cond_expr), vectype,
vec_cond_lhs, vec_cond_rhs);
- vec_cond_expr = build3 (VEC_COND_EXPR, vectype,
+ vec_cond_expr = build3 (VEC_COND_EXPR, vectype,
vec_compare, vec_then_clause, vec_else_clause);
*vec_stmt = gimple_build_assign (vec_dest, vec_cond_expr);
new_temp = make_ssa_name (vec_dest, *vec_stmt);
gimple_assign_set_lhs (*vec_stmt, new_temp);
vect_finish_stmt_generation (stmt, *vec_stmt, gsi);
-
+
return true;
}
{
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
bb_vec_info bb_vinfo = STMT_VINFO_BB_VINFO (stmt_info);
- enum vect_relevant relevance = STMT_VINFO_RELEVANT (stmt_info);
+ enum vect_relevant relevance = STMT_VINFO_RELEVANT (stmt_info);
bool ok;
- HOST_WIDE_INT dummy;
tree scalar_type, vectype;
if (vect_print_dump_info (REPORT_DETAILS))
}
if (gimple_has_volatile_ops (stmt))
- {
+ {
if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS))
fprintf (vect_dump, "not vectorized: stmt has volatile operands");
return false;
}
-
- /* Skip stmts that do not need to be vectorized. In loops this is expected
+
+ /* Skip stmts that do not need to be vectorized. In loops this is expected
to include:
- the COND_EXPR which is the loop exit condition
- any LABEL_EXPRs in the loop
- - computations that are used only for array indexing or loop control.
+ - computations that are used only for array indexing or loop control.
In basic blocks we only analyze statements that are a part of some SLP
instance, therefore, all the statements are relevant. */
- if (!STMT_VINFO_RELEVANT_P (stmt_info)
+ if (!STMT_VINFO_RELEVANT_P (stmt_info)
&& !STMT_VINFO_LIVE_P (stmt_info))
{
if (vect_print_dump_info (REPORT_DETAILS))
{
gcc_assert (PURE_SLP_STMT (stmt_info));
- scalar_type = vect_get_smallest_scalar_type (stmt, &dummy, &dummy);
+ scalar_type = TREE_TYPE (gimple_get_lhs (stmt));
if (vect_print_dump_info (REPORT_DETAILS))
{
fprintf (vect_dump, "get vectype for scalar type: ");
}
ok = true;
- if (!bb_vinfo
+ if (!bb_vinfo
&& (STMT_VINFO_RELEVANT_P (stmt_info)
|| STMT_VINFO_DEF_TYPE (stmt_info) == vect_reduction_def))
ok = (vectorizable_type_promotion (stmt, NULL, NULL, NULL)
|| vectorizable_load (stmt, NULL, NULL, NULL, NULL)
|| vectorizable_call (stmt, NULL, NULL)
|| vectorizable_store (stmt, NULL, NULL, NULL)
- || vectorizable_condition (stmt, NULL, NULL)
- || vectorizable_reduction (stmt, NULL, NULL));
+ || vectorizable_reduction (stmt, NULL, NULL, NULL)
+ || vectorizable_condition (stmt, NULL, NULL, NULL, 0));
else
{
if (bb_vinfo)
|| vectorizable_assignment (stmt, NULL, NULL, node)
|| vectorizable_load (stmt, NULL, NULL, node, NULL)
|| vectorizable_store (stmt, NULL, NULL, node));
- }
+ }
if (!ok)
{
fprintf (vect_dump, "supported: ");
print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM);
}
-
+
return false;
}
fprintf (vect_dump, "supported: ");
print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM);
}
-
+
return false;
}
if (!PURE_SLP_STMT (stmt_info))
{
- /* Groups of strided accesses whose size is not a power of 2 are not
- vectorizable yet using loop-vectorization. Therefore, if this stmt
- feeds non-SLP-able stmts (i.e., this stmt has to be both SLPed and
+ /* Groups of strided accesses whose size is not a power of 2 are not
+ vectorizable yet using loop-vectorization. Therefore, if this stmt
+ feeds non-SLP-able stmts (i.e., this stmt has to be both SLPed and
loop-based vectorized), the loop cannot be vectorized. */
if (STMT_VINFO_STRIDED_ACCESS (stmt_info)
&& exact_log2 (DR_GROUP_SIZE (vinfo_for_stmt (
return false;
}
}
-
+
return true;
}
bool
vect_transform_stmt (gimple stmt, gimple_stmt_iterator *gsi,
- bool *strided_store, slp_tree slp_node,
+ bool *strided_store, slp_tree slp_node,
slp_instance slp_node_instance)
{
bool is_store = false;
break;
case load_vec_info_type:
- done = vectorizable_load (stmt, gsi, &vec_stmt, slp_node,
+ done = vectorizable_load (stmt, gsi, &vec_stmt, slp_node,
slp_node_instance);
gcc_assert (done);
break;
case condition_vec_info_type:
gcc_assert (!slp_node);
- done = vectorizable_condition (stmt, gsi, &vec_stmt);
+ done = vectorizable_condition (stmt, gsi, &vec_stmt, NULL, 0);
gcc_assert (done);
break;
break;
case reduc_vec_info_type:
- gcc_assert (!slp_node);
- done = vectorizable_reduction (stmt, gsi, &vec_stmt);
+ done = vectorizable_reduction (stmt, gsi, &vec_stmt, slp_node);
gcc_assert (done);
break;
STMT_VINFO_LOOP_VINFO (stmt_info)), stmt)
&& STMT_VINFO_TYPE (stmt_info) != reduc_vec_info_type
&& (STMT_VINFO_RELEVANT (stmt_info) == vect_used_in_outer
- || STMT_VINFO_RELEVANT (stmt_info) ==
+ || STMT_VINFO_RELEVANT (stmt_info) ==
vect_used_in_outer_by_reduction))
{
struct loop *innerloop = LOOP_VINFO_LOOP (
{
stmt_vec_info stmt_vinfo = vinfo_for_stmt (orig_stmt_in_pattern);
/* 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
+ 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. */
if (STMT_VINFO_IN_PATTERN_P (stmt_vinfo))
{
}
}
- return is_store;
+ return is_store;
}
-/* Remove a group of stores (for SLP or interleaving), free their
+/* Remove a group of stores (for SLP or interleaving), free their
stmt_vec_info. */
void
Create and initialize a new stmt_vec_info struct for STMT. */
stmt_vec_info
-new_stmt_vec_info (gimple stmt, loop_vec_info loop_vinfo,
+new_stmt_vec_info (gimple stmt, loop_vec_info loop_vinfo,
bb_vec_info bb_vinfo)
{
stmt_vec_info res;
STMT_VINFO_LIVE_P (res) = false;
STMT_VINFO_VECTYPE (res) = NULL;
STMT_VINFO_VEC_STMT (res) = NULL;
+ STMT_VINFO_VECTORIZABLE (res) = true;
STMT_VINFO_IN_PATTERN_P (res) = false;
STMT_VINFO_RELATED_STMT (res) = NULL;
STMT_VINFO_DATA_REF (res) = NULL;
get_vectype_for_scalar_type (tree scalar_type)
{
enum machine_mode inner_mode = TYPE_MODE (scalar_type);
- int nbytes = GET_MODE_SIZE (inner_mode);
+ unsigned int nbytes = GET_MODE_SIZE (inner_mode);
int nunits;
tree vectype;
if (nbytes == 0 || nbytes >= UNITS_PER_SIMD_WORD (inner_mode))
return NULL_TREE;
+ /* We can't build a vector type of elements with alignment bigger than
+ their size. */
+ if (nbytes < TYPE_ALIGN_UNIT (scalar_type))
+ return NULL_TREE;
+
+ /* If we'd build a vector type of elements whose mode precision doesn't
+ match their types precision we'll get mismatched types on vector
+ extracts via BIT_FIELD_REFs. This effectively means we disable
+ vectorization of bool and/or enum types in some languages. */
+ if (INTEGRAL_TYPE_P (scalar_type)
+ && GET_MODE_BITSIZE (inner_mode) != TYPE_PRECISION (scalar_type))
+ return NULL_TREE;
+
/* FORNOW: Only a single vector size per mode (UNITS_PER_SIMD_WORD)
is expected. */
nunits = UNITS_PER_SIMD_WORD (inner_mode) / nbytes;
return vectype;
}
+/* Function get_same_sized_vectype
+
+ Returns a vector type corresponding to SCALAR_TYPE of size
+ VECTOR_TYPE if supported by the target. */
+
+tree
+get_same_sized_vectype (tree scalar_type, tree vector_type ATTRIBUTE_UNUSED)
+{
+ return get_vectype_for_scalar_type (scalar_type);
+}
+
/* Function vect_is_simple_use.
Input:
DEF - the defining stmt in case OPERAND is an SSA_NAME.
Returns whether a stmt with OPERAND can be vectorized.
- For loops, supportable operands are constants, loop invariants, and operands
- that are defined by the current iteration of the loop. Unsupportable
- operands are those that are defined by a previous iteration of the loop (as
+ For loops, supportable operands are constants, loop invariants, and operands
+ that are defined by the current iteration of the loop. Unsupportable
+ operands are those that are defined by a previous iteration of the loop (as
is the case in reduction/induction computations).
For basic blocks, supportable operands are constants and bb invariants.
For now, operands defined outside the basic block are not supported. */
bool
-vect_is_simple_use (tree operand, loop_vec_info loop_vinfo,
+vect_is_simple_use (tree operand, loop_vec_info loop_vinfo,
bb_vec_info bb_vinfo, gimple *def_stmt,
tree *def, enum vect_def_type *dt)
-{
+{
basic_block bb;
stmt_vec_info stmt_vinfo;
struct loop *loop = NULL;
-
+
if (loop_vinfo)
loop = LOOP_VINFO_LOOP (loop_vinfo);
*def_stmt = NULL;
*def = NULL_TREE;
-
+
if (vect_print_dump_info (REPORT_DETAILS))
{
fprintf (vect_dump, "vect_is_simple_use: operand ");
print_generic_expr (vect_dump, operand, TDF_SLIM);
}
-
+
if (TREE_CODE (operand) == INTEGER_CST || TREE_CODE (operand) == REAL_CST)
{
*dt = vect_constant_def;
return true;
}
-
+
if (is_gimple_min_invariant (operand))
{
*def = operand;
fprintf (vect_dump, "non-associatable copy.");
operand = TREE_OPERAND (operand, 0);
}
-
+
if (TREE_CODE (operand) != SSA_NAME)
{
if (vect_print_dump_info (REPORT_DETAILS))
fprintf (vect_dump, "not ssa-name.");
return false;
}
-
+
*def_stmt = SSA_NAME_DEF_STMT (operand);
if (*def_stmt == NULL)
{
if ((loop && !flow_bb_inside_loop_p (loop, bb))
|| (!loop && bb != BB_VINFO_BB (bb_vinfo))
- || (!loop && gimple_code (*def_stmt) == GIMPLE_PHI))
+ || (!loop && gimple_code (*def_stmt) == GIMPLE_PHI))
*dt = vect_external_def;
else
{
return true;
}
+/* Function vect_is_simple_use_1.
+
+ Same as vect_is_simple_use_1 but also determines the vector operand
+ type of OPERAND and stores it to *VECTYPE. If the definition of
+ OPERAND is vect_uninitialized_def, vect_constant_def or
+ vect_external_def *VECTYPE will be set to NULL_TREE and the caller
+ is responsible to compute the best suited vector type for the
+ scalar operand. */
+
+bool
+vect_is_simple_use_1 (tree operand, loop_vec_info loop_vinfo,
+ bb_vec_info bb_vinfo, gimple *def_stmt,
+ tree *def, enum vect_def_type *dt, tree *vectype)
+{
+ if (!vect_is_simple_use (operand, loop_vinfo, bb_vinfo, def_stmt, def, dt))
+ return false;
+
+ /* Now get a vector type if the def is internal, otherwise supply
+ NULL_TREE and leave it up to the caller to figure out a proper
+ type for the use stmt. */
+ if (*dt == vect_internal_def
+ || *dt == vect_induction_def
+ || *dt == vect_reduction_def
+ || *dt == vect_double_reduction_def
+ || *dt == vect_nested_cycle)
+ {
+ stmt_vec_info stmt_info = vinfo_for_stmt (*def_stmt);
+ if (STMT_VINFO_IN_PATTERN_P (stmt_info))
+ stmt_info = vinfo_for_stmt (STMT_VINFO_RELATED_STMT (stmt_info));
+ *vectype = STMT_VINFO_VECTYPE (stmt_info);
+ gcc_assert (*vectype != NULL_TREE);
+ }
+ else if (*dt == vect_uninitialized_def
+ || *dt == vect_constant_def
+ || *dt == vect_external_def)
+ *vectype = NULL_TREE;
+ else
+ gcc_unreachable ();
+
+ return true;
+}
+
/* Function supportable_widening_operation
- Check whether an operation represented by the code CODE is a
- widening operation that is supported by the target platform in
- vector form (i.e., when operating on arguments of type VECTYPE).
-
+ Check whether an operation represented by the code CODE is a
+ widening operation that is supported by the target platform in
+ vector form (i.e., when operating on arguments of type VECTYPE_IN
+ producing a result of type VECTYPE_OUT).
+
Widening operations we currently support are NOP (CONVERT), FLOAT
and WIDEN_MULT. This function checks if these operations are supported
by the target platform either directly (via vector tree-codes), or via
target builtins.
Output:
- - CODE1 and CODE2 are codes of vector operations to be used when
- vectorizing the operation, if available.
+ - CODE1 and CODE2 are codes of vector operations to be used when
+ vectorizing the operation, if available.
- DECL1 and DECL2 are decls of target builtin functions to be used
when vectorizing the operation, if available. In this case,
- CODE1 and CODE2 are CALL_EXPR.
+ CODE1 and CODE2 are CALL_EXPR.
- MULTI_STEP_CVT determines the number of required intermediate steps in
case of multi-step conversion (like char->short->int - in that case
MULTI_STEP_CVT will be 1).
- - INTERM_TYPES contains the intermediate type required to perform the
- widening operation (short in the above example). */
+ - INTERM_TYPES contains the intermediate type required to perform the
+ widening operation (short in the above example). */
bool
-supportable_widening_operation (enum tree_code code, gimple stmt, tree vectype,
+supportable_widening_operation (enum tree_code code, gimple stmt,
+ tree vectype_out, tree vectype_in,
tree *decl1, tree *decl2,
enum tree_code *code1, enum tree_code *code2,
int *multi_step_cvt,
enum machine_mode vec_mode;
enum insn_code icode1, icode2;
optab optab1, optab2;
- tree type = gimple_expr_type (stmt);
- tree wide_vectype = get_vectype_for_scalar_type (type);
+ tree vectype = vectype_in;
+ tree wide_vectype = vectype_out;
enum tree_code c1, c2;
/* The result of a vectorized widening operation usually requires two vectors
- (because the widened results do not fit int one vector). The generated
- vector results would normally be expected to be generated in the same
+ (because the widened results do not fit int one vector). The generated
+ vector results would normally be expected to be generated in the same
order as in the original scalar computation, i.e. if 8 results are
generated in each vector iteration, they are to be organized as follows:
- vect1: [res1,res2,res3,res4], vect2: [res5,res6,res7,res8].
+ vect1: [res1,res2,res3,res4], vect2: [res5,res6,res7,res8].
- However, in the special case that the result of the widening operation is
+ However, in the special case that the result of the widening operation is
used in a reduction computation only, the order doesn't matter (because
- when vectorizing a reduction we change the order of the computation).
+ when vectorizing a reduction we change the order of the computation).
Some targets can take advantage of this and generate more efficient code.
For example, targets like Altivec, that support widen_mult using a sequence
of {mult_even,mult_odd} generate the following vectors:
vect1: [res1,res3,res5,res7], vect2: [res2,res4,res6,res8].
When vectorizing outer-loops, we execute the inner-loop sequentially
- (each vectorized inner-loop iteration contributes to VF outer-loop
- iterations in parallel). We therefore don't allow to change the order
+ (each vectorized inner-loop iteration contributes to VF outer-loop
+ iterations in parallel). We therefore don't allow to change the order
of the computation in the inner-loop during outer-loop vectorization. */
if (STMT_VINFO_RELEVANT (stmt_info) == vect_used_by_reduction
if (code == FIX_TRUNC_EXPR)
{
/* The signedness is determined from output operand. */
- optab1 = optab_for_tree_code (c1, type, optab_default);
- optab2 = optab_for_tree_code (c2, type, optab_default);
+ optab1 = optab_for_tree_code (c1, vectype_out, optab_default);
+ optab2 = optab_for_tree_code (c2, vectype_out, optab_default);
}
else
{
== CODE_FOR_nothing)
return false;
- /* Check if it's a multi-step conversion that can be done using intermediate
+ /* Check if it's a multi-step conversion that can be done using intermediate
types. */
if (insn_data[icode1].operand[0].mode != TYPE_MODE (wide_vectype)
|| insn_data[icode2].operand[0].mode != TYPE_MODE (wide_vectype))
if (!CONVERT_EXPR_CODE_P (code))
return false;
-
+
*code1 = c1;
*code2 = c2;
-
+
/* We assume here that there will not be more than MAX_INTERM_CVT_STEPS
intermediate steps in promotion sequence. We try MAX_INTERM_CVT_STEPS
to get to NARROW_VECTYPE, and fail if we do not. */
|| (icode2 = optab2->handlers[(int) prev_mode].insn_code)
== CODE_FOR_nothing
|| insn_data[icode2].operand[0].mode != intermediate_mode
- || (icode1 = optab3->handlers[(int) intermediate_mode].insn_code)
+ || (icode1 = optab3->handlers[(int) intermediate_mode].insn_code)
== CODE_FOR_nothing
|| (icode2 = optab4->handlers[(int) intermediate_mode].insn_code)
== CODE_FOR_nothing)
/* Function supportable_narrowing_operation
- Check whether an operation represented by the code CODE is a
- narrowing operation that is supported by the target platform in
- vector form (i.e., when operating on arguments of type VECTYPE).
-
+ Check whether an operation represented by the code CODE is a
+ narrowing operation that is supported by the target platform in
+ vector form (i.e., when operating on arguments of type VECTYPE_IN
+ and producing a result of type VECTYPE_OUT).
+
Narrowing operations we currently support are NOP (CONVERT) and
FIX_TRUNC. This function checks if these operations are supported by
the target platform directly via vector tree-codes.
Output:
- - CODE1 is the code of a vector operation to be used when
- vectorizing the operation, if available.
+ - CODE1 is the code of a vector operation to be used when
+ vectorizing the operation, if available.
- MULTI_STEP_CVT determines the number of required intermediate steps in
case of multi-step conversion (like int->short->char - in that case
MULTI_STEP_CVT will be 1).
- INTERM_TYPES contains the intermediate type required to perform the
- narrowing operation (short in the above example). */
+ narrowing operation (short in the above example). */
bool
supportable_narrowing_operation (enum tree_code code,
- const_gimple stmt, tree vectype,
+ tree vectype_out, tree vectype_in,
enum tree_code *code1, int *multi_step_cvt,
VEC (tree, heap) **interm_types)
{
enum machine_mode vec_mode;
enum insn_code icode1;
optab optab1, interm_optab;
- tree type = gimple_expr_type (stmt);
- tree narrow_vectype = get_vectype_for_scalar_type (type);
+ tree vectype = vectype_in;
+ tree narrow_vectype = vectype_out;
enum tree_code c1;
tree intermediate_type, prev_type;
int i;
if (code == FIX_TRUNC_EXPR)
/* The signedness is determined from output operand. */
- optab1 = optab_for_tree_code (c1, type, optab_default);
+ optab1 = optab_for_tree_code (c1, vectype_out, optab_default);
else
optab1 = optab_for_tree_code (c1, vectype, optab_default);
return false;
vec_mode = TYPE_MODE (vectype);
- if ((icode1 = optab_handler (optab1, vec_mode)->insn_code)
+ if ((icode1 = optab_handler (optab1, vec_mode)->insn_code)
== CODE_FOR_nothing)
return false;
intermediate_mode = insn_data[icode1].operand[0].mode;
intermediate_type = lang_hooks.types.type_for_mode (intermediate_mode,
TYPE_UNSIGNED (prev_type));
- interm_optab = optab_for_tree_code (c1, intermediate_type,
+ interm_optab = optab_for_tree_code (c1, intermediate_type,
optab_default);
- if (!interm_optab
+ if (!interm_optab
|| (icode1 = optab1->handlers[(int) prev_mode].insn_code)
== CODE_FOR_nothing
|| insn_data[icode1].operand[0].mode != intermediate_mode
- || (icode1
+ || (icode1
= interm_optab->handlers[(int) intermediate_mode].insn_code)
== CODE_FOR_nothing)
return false;