/* Vectorizer
Copyright (C) 2003-2013 Free Software Foundation, Inc.
Contributed by Dorit Naishlos <dorit@il.ibm.com>
This file is part of GCC.
GCC is free software; you can redistribute it and/or modify it under
the terms of the GNU General Public License as published by the Free
Software Foundation; either version 3, or (at your option) any later
version.
GCC is distributed in the hope that it will be useful, but WITHOUT ANY
WARRANTY; without even the implied warranty of MERCHANTABILITY or
FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
for more details.
You should have received a copy of the GNU General Public License
along with GCC; see the file COPYING3. If not see
<http://www.gnu.org/licenses/>. */
/* Loop and basic block vectorizer.
This file contains drivers for the three vectorizers:
(1) loop vectorizer (inter-iteration parallelism),
(2) loop-aware SLP (intra-iteration parallelism) (invoked by the loop
vectorizer)
(3) BB vectorizer (out-of-loops), aka SLP
The rest of the vectorizer's code is organized as follows:
- tree-vect-loop.c - loop specific parts such as reductions, etc. These are
used by drivers (1) and (2).
- tree-vect-loop-manip.c - vectorizer's loop control-flow utilities, used by
drivers (1) and (2).
- tree-vect-slp.c - BB vectorization specific analysis and transformation,
used by drivers (2) and (3).
- tree-vect-stmts.c - statements analysis and transformation (used by all).
- tree-vect-data-refs.c - vectorizer specific data-refs analysis and
manipulations (used by all).
- tree-vect-patterns.c - vectorizable code patterns detector (used by all)
Here's a poor attempt at illustrating that:
tree-vectorizer.c:
loop_vect() loop_aware_slp() slp_vect()
| / \ /
| / \ /
tree-vect-loop.c tree-vect-slp.c
| \ \ / / |
| \ \/ / |
| \ /\ / |
| \ / \ / |
tree-vect-stmts.c tree-vect-data-refs.c
\ /
tree-vect-patterns.c
*/
#include "config.h"
#include "system.h"
#include "coretypes.h"
#include "dumpfile.h"
#include "tm.h"
#include "ggc.h"
#include "tree.h"
#include "tree-pretty-print.h"
#include "gimple.h"
#include "gimple-iterator.h"
#include "gimple-walk.h"
#include "gimple-ssa.h"
#include "cgraph.h"
#include "tree-phinodes.h"
#include "ssa-iterators.h"
#include "tree-ssa-loop-manip.h"
#include "cfgloop.h"
#include "tree-vectorizer.h"
#include "tree-pass.h"
#include "hash-table.h"
#include "tree-ssa-propagate.h"
#include "dbgcnt.h"
/* Loop or bb location. */
LOC vect_location;
/* Vector mapping GIMPLE stmt to stmt_vec_info. */
vec<vec_void_p> stmt_vec_info_vec;
/* For mapping simduid to vectorization factor. */
struct simduid_to_vf : typed_free_remove<simduid_to_vf>
{
unsigned int simduid;
int vf;
/* hash_table support. */
typedef simduid_to_vf value_type;
typedef simduid_to_vf compare_type;
static inline hashval_t hash (const value_type *);
static inline int equal (const value_type *, const compare_type *);
};
inline hashval_t
simduid_to_vf::hash (const value_type *p)
{
return p->simduid;
}
inline int
simduid_to_vf::equal (const value_type *p1, const value_type *p2)
{
return p1->simduid == p2->simduid;
}
/* This hash maps the OMP simd array to the corresponding simduid used
to index into it. Like thus,
_7 = GOMP_SIMD_LANE (simduid.0)
...
...
D.1737[_7] = stuff;
This hash maps from the OMP simd array (D.1737[]) to DECL_UID of
simduid.0. */
struct simd_array_to_simduid : typed_free_remove<simd_array_to_simduid>
{
tree decl;
unsigned int simduid;
/* hash_table support. */
typedef simd_array_to_simduid value_type;
typedef simd_array_to_simduid compare_type;
static inline hashval_t hash (const value_type *);
static inline int equal (const value_type *, const compare_type *);
};
inline hashval_t
simd_array_to_simduid::hash (const value_type *p)
{
return DECL_UID (p->decl);
}
inline int
simd_array_to_simduid::equal (const value_type *p1, const value_type *p2)
{
return p1->decl == p2->decl;
}
/* Fold IFN_GOMP_SIMD_LANE, IFN_GOMP_SIMD_VF and IFN_GOMP_SIMD_LAST_LANE
into their corresponding constants. */
static void
adjust_simduid_builtins (hash_table <simduid_to_vf> &htab)
{
basic_block bb;
FOR_EACH_BB (bb)
{
gimple_stmt_iterator i;
for (i = gsi_start_bb (bb); !gsi_end_p (i); gsi_next (&i))
{
unsigned int vf = 1;
enum internal_fn ifn;
gimple stmt = gsi_stmt (i);
tree t;
if (!is_gimple_call (stmt)
|| !gimple_call_internal_p (stmt))
continue;
ifn = gimple_call_internal_fn (stmt);
switch (ifn)
{
case IFN_GOMP_SIMD_LANE:
case IFN_GOMP_SIMD_VF:
case IFN_GOMP_SIMD_LAST_LANE:
break;
default:
continue;
}
tree arg = gimple_call_arg (stmt, 0);
gcc_assert (arg != NULL_TREE);
gcc_assert (TREE_CODE (arg) == SSA_NAME);
simduid_to_vf *p = NULL, data;
data.simduid = DECL_UID (SSA_NAME_VAR (arg));
if (htab.is_created ())
p = htab.find (&data);
if (p)
vf = p->vf;
switch (ifn)
{
case IFN_GOMP_SIMD_VF:
t = build_int_cst (unsigned_type_node, vf);
break;
case IFN_GOMP_SIMD_LANE:
t = build_int_cst (unsigned_type_node, 0);
break;
case IFN_GOMP_SIMD_LAST_LANE:
t = gimple_call_arg (stmt, 1);
break;
default:
gcc_unreachable ();
}
update_call_from_tree (&i, t);
}
}
}
/* Helper structure for note_simd_array_uses. */
struct note_simd_array_uses_struct
{
hash_table <simd_array_to_simduid> *htab;
unsigned int simduid;
};
/* Callback for note_simd_array_uses, called through walk_gimple_op. */
static tree
note_simd_array_uses_cb (tree *tp, int *walk_subtrees, void *data)
{
struct walk_stmt_info *wi = (struct walk_stmt_info *) data;
struct note_simd_array_uses_struct *ns
= (struct note_simd_array_uses_struct *) wi->info;
if (TYPE_P (*tp))
*walk_subtrees = 0;
else if (VAR_P (*tp)
&& lookup_attribute ("omp simd array", DECL_ATTRIBUTES (*tp))
&& DECL_CONTEXT (*tp) == current_function_decl)
{
simd_array_to_simduid data;
if (!ns->htab->is_created ())
ns->htab->create (15);
data.decl = *tp;
data.simduid = ns->simduid;
simd_array_to_simduid **slot = ns->htab->find_slot (&data, INSERT);
if (*slot == NULL)
{
simd_array_to_simduid *p = XNEW (simd_array_to_simduid);
*p = data;
*slot = p;
}
else if ((*slot)->simduid != ns->simduid)
(*slot)->simduid = -1U;
*walk_subtrees = 0;
}
return NULL_TREE;
}
/* Find "omp simd array" temporaries and map them to corresponding
simduid. */
static void
note_simd_array_uses (hash_table <simd_array_to_simduid> *htab)
{
basic_block bb;
gimple_stmt_iterator gsi;
struct walk_stmt_info wi;
struct note_simd_array_uses_struct ns;
memset (&wi, 0, sizeof (wi));
wi.info = &ns;
ns.htab = htab;
FOR_EACH_BB (bb)
for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi))
{
gimple stmt = gsi_stmt (gsi);
if (!is_gimple_call (stmt) || !gimple_call_internal_p (stmt))
continue;
switch (gimple_call_internal_fn (stmt))
{
case IFN_GOMP_SIMD_LANE:
case IFN_GOMP_SIMD_VF:
case IFN_GOMP_SIMD_LAST_LANE:
break;
default:
continue;
}
tree lhs = gimple_call_lhs (stmt);
if (lhs == NULL_TREE)
continue;
imm_use_iterator use_iter;
gimple use_stmt;
ns.simduid = DECL_UID (SSA_NAME_VAR (gimple_call_arg (stmt, 0)));
FOR_EACH_IMM_USE_STMT (use_stmt, use_iter, lhs)
if (!is_gimple_debug (use_stmt))
walk_gimple_op (use_stmt, note_simd_array_uses_cb, &wi);
}
}
/* A helper function to free data refs. */
void
vect_destroy_datarefs (loop_vec_info loop_vinfo, bb_vec_info bb_vinfo)
{
vec<data_reference_p> datarefs;
struct data_reference *dr;
unsigned int i;
if (loop_vinfo)
datarefs = LOOP_VINFO_DATAREFS (loop_vinfo);
else
datarefs = BB_VINFO_DATAREFS (bb_vinfo);
FOR_EACH_VEC_ELT (datarefs, i, dr)
if (dr->aux)
{
free (dr->aux);
dr->aux = NULL;
}
free_data_refs (datarefs);
}
/* Function vectorize_loops.
Entry point to loop vectorization phase. */
unsigned
vectorize_loops (void)
{
unsigned int i;
unsigned int num_vectorized_loops = 0;
unsigned int vect_loops_num;
loop_iterator li;
struct loop *loop;
hash_table <simduid_to_vf> simduid_to_vf_htab;
hash_table <simd_array_to_simduid> simd_array_to_simduid_htab;
vect_loops_num = number_of_loops (cfun);
/* Bail out if there are no loops. */
if (vect_loops_num <= 1)
{
if (cfun->has_simduid_loops)
adjust_simduid_builtins (simduid_to_vf_htab);
return 0;
}
if (cfun->has_simduid_loops)
note_simd_array_uses (&simd_array_to_simduid_htab);
init_stmt_vec_info_vec ();
/* ----------- Analyze loops. ----------- */
/* If some loop was duplicated, it gets bigger number
than all previously defined loops. This fact allows us to run
only over initial loops skipping newly generated ones. */
FOR_EACH_LOOP (li, loop, 0)
if ((flag_tree_loop_vectorize && optimize_loop_nest_for_speed_p (loop))
|| loop->force_vect)
{
loop_vec_info loop_vinfo;
vect_location = find_loop_location (loop);
if (LOCATION_LOCUS (vect_location) != UNKNOWN_LOC
&& dump_enabled_p ())
dump_printf (MSG_NOTE, "\nAnalyzing loop at %s:%d\n",
LOC_FILE (vect_location), LOC_LINE (vect_location));
loop_vinfo = vect_analyze_loop (loop);
loop->aux = loop_vinfo;
if (!loop_vinfo || !LOOP_VINFO_VECTORIZABLE_P (loop_vinfo))
continue;
if (!dbg_cnt (vect_loop))
break;
if (LOCATION_LOCUS (vect_location) != UNKNOWN_LOC
&& dump_enabled_p ())
dump_printf_loc (MSG_OPTIMIZED_LOCATIONS, vect_location,
"loop vectorized\n");
vect_transform_loop (loop_vinfo);
num_vectorized_loops++;
/* Now that the loop has been vectorized, allow it to be unrolled
etc. */
loop->force_vect = false;
if (loop->simduid)
{
simduid_to_vf *simduid_to_vf_data = XNEW (simduid_to_vf);
if (!simduid_to_vf_htab.is_created ())
simduid_to_vf_htab.create (15);
simduid_to_vf_data->simduid = DECL_UID (loop->simduid);
simduid_to_vf_data->vf = loop_vinfo->vectorization_factor;
*simduid_to_vf_htab.find_slot (simduid_to_vf_data, INSERT)
= simduid_to_vf_data;
}
}
vect_location = UNKNOWN_LOC;
statistics_counter_event (cfun, "Vectorized loops", num_vectorized_loops);
if (dump_enabled_p ()
|| (num_vectorized_loops > 0 && dump_enabled_p ()))
dump_printf_loc (MSG_NOTE, vect_location,
"vectorized %u loops in function.\n",
num_vectorized_loops);
/* ----------- Finalize. ----------- */
for (i = 1; i < vect_loops_num; i++)
{
loop_vec_info loop_vinfo;
loop = get_loop (cfun, i);
if (!loop)
continue;
loop_vinfo = (loop_vec_info) loop->aux;
destroy_loop_vec_info (loop_vinfo, true);
loop->aux = NULL;
}
free_stmt_vec_info_vec ();
/* Fold IFN_GOMP_SIMD_{VF,LANE,LAST_LANE} builtins. */
if (cfun->has_simduid_loops)
adjust_simduid_builtins (simduid_to_vf_htab);
/* Shrink any "omp array simd" temporary arrays to the
actual vectorization factors. */
if (simd_array_to_simduid_htab.is_created ())
{
for (hash_table <simd_array_to_simduid>::iterator iter
= simd_array_to_simduid_htab.begin ();
iter != simd_array_to_simduid_htab.end (); ++iter)
if ((*iter).simduid != -1U)
{
tree decl = (*iter).decl;
int vf = 1;
if (simduid_to_vf_htab.is_created ())
{
simduid_to_vf *p = NULL, data;
data.simduid = (*iter).simduid;
p = simduid_to_vf_htab.find (&data);
if (p)
vf = p->vf;
}
tree atype
= build_array_type_nelts (TREE_TYPE (TREE_TYPE (decl)), vf);
TREE_TYPE (decl) = atype;
relayout_decl (decl);
}
simd_array_to_simduid_htab.dispose ();
}
if (simduid_to_vf_htab.is_created ())
simduid_to_vf_htab.dispose ();
if (num_vectorized_loops > 0)
{
/* If we vectorized any loop only virtual SSA form needs to be updated.
??? Also while we try hard to update loop-closed SSA form we fail
to properly do this in some corner-cases (see PR56286). */
rewrite_into_loop_closed_ssa (NULL, TODO_update_ssa_only_virtuals);
return TODO_cleanup_cfg;
}
return 0;
}
/* Entry point to basic block SLP phase. */
static unsigned int
execute_vect_slp (void)
{
basic_block bb;
init_stmt_vec_info_vec ();
FOR_EACH_BB (bb)
{
vect_location = find_bb_location (bb);
if (vect_slp_analyze_bb (bb))
{
if (!dbg_cnt (vect_slp))
break;
vect_slp_transform_bb (bb);
if (dump_enabled_p ())
dump_printf_loc (MSG_OPTIMIZED_LOCATIONS, vect_location,
"basic block vectorized\n");
}
}
free_stmt_vec_info_vec ();
return 0;
}
static bool
gate_vect_slp (void)
{
return flag_tree_slp_vectorize != 0;
}
namespace {
const pass_data pass_data_slp_vectorize =
{
GIMPLE_PASS, /* type */
"slp", /* name */
OPTGROUP_LOOP | OPTGROUP_VEC, /* optinfo_flags */
true, /* has_gate */
true, /* has_execute */
TV_TREE_SLP_VECTORIZATION, /* tv_id */
( PROP_ssa | PROP_cfg ), /* properties_required */
0, /* properties_provided */
0, /* properties_destroyed */
0, /* todo_flags_start */
( TODO_verify_ssa | TODO_update_ssa
| TODO_verify_stmts ), /* todo_flags_finish */
};
class pass_slp_vectorize : public gimple_opt_pass
{
public:
pass_slp_vectorize (gcc::context *ctxt)
: gimple_opt_pass (pass_data_slp_vectorize, ctxt)
{}
/* opt_pass methods: */
bool gate () { return gate_vect_slp (); }
unsigned int execute () { return execute_vect_slp (); }
}; // class pass_slp_vectorize
} // anon namespace
gimple_opt_pass *
make_pass_slp_vectorize (gcc::context *ctxt)
{
return new pass_slp_vectorize (ctxt);
}
/* Increase alignment of global arrays to improve vectorization potential.
TODO:
- Consider also structs that have an array field.
- Use ipa analysis to prune arrays that can't be vectorized?
This should involve global alignment analysis and in the future also
array padding. */
static unsigned int
increase_alignment (void)
{
struct varpool_node *vnode;
vect_location = UNKNOWN_LOC;
/* Increase the alignment of all global arrays for vectorization. */
FOR_EACH_DEFINED_VARIABLE (vnode)
{
tree vectype, decl = vnode->decl;
tree t;
unsigned int alignment;
t = TREE_TYPE (decl);
if (TREE_CODE (t) != ARRAY_TYPE)
continue;
vectype = get_vectype_for_scalar_type (strip_array_types (t));
if (!vectype)
continue;
alignment = TYPE_ALIGN (vectype);
if (DECL_ALIGN (decl) >= alignment)
continue;
if (vect_can_force_dr_alignment_p (decl, alignment))
{
DECL_ALIGN (decl) = TYPE_ALIGN (vectype);
DECL_USER_ALIGN (decl) = 1;
dump_printf (MSG_NOTE, "Increasing alignment of decl: ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, decl);
dump_printf (MSG_NOTE, "\n");
}
}
return 0;
}
static bool
gate_increase_alignment (void)
{
return flag_section_anchors && flag_tree_loop_vectorize;
}
namespace {
const pass_data pass_data_ipa_increase_alignment =
{
SIMPLE_IPA_PASS, /* type */
"increase_alignment", /* name */
OPTGROUP_LOOP | OPTGROUP_VEC, /* optinfo_flags */
true, /* has_gate */
true, /* has_execute */
TV_IPA_OPT, /* tv_id */
0, /* properties_required */
0, /* properties_provided */
0, /* properties_destroyed */
0, /* todo_flags_start */
0, /* todo_flags_finish */
};
class pass_ipa_increase_alignment : public simple_ipa_opt_pass
{
public:
pass_ipa_increase_alignment (gcc::context *ctxt)
: simple_ipa_opt_pass (pass_data_ipa_increase_alignment, ctxt)
{}
/* opt_pass methods: */
bool gate () { return gate_increase_alignment (); }
unsigned int execute () { return increase_alignment (); }
}; // class pass_ipa_increase_alignment
} // anon namespace
simple_ipa_opt_pass *
make_pass_ipa_increase_alignment (gcc::context *ctxt)
{
return new pass_ipa_increase_alignment (ctxt);
}