/* Function splitting pass
|
/* Function splitting pass
|
Copyright (C) 2010, 2011, 2012
|
Copyright (C) 2010, 2011, 2012
|
Free Software Foundation, Inc.
|
Free Software Foundation, Inc.
|
Contributed by Jan Hubicka <jh@suse.cz>
|
Contributed by Jan Hubicka <jh@suse.cz>
|
|
|
This file is part of GCC.
|
This file is part of GCC.
|
|
|
GCC is free software; you can redistribute it and/or modify it under
|
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
|
the terms of the GNU General Public License as published by the Free
|
Software Foundation; either version 3, or (at your option) any later
|
Software Foundation; either version 3, or (at your option) any later
|
version.
|
version.
|
|
|
GCC is distributed in the hope that it will be useful, but WITHOUT ANY
|
GCC is distributed in the hope that it will be useful, but WITHOUT ANY
|
WARRANTY; without even the implied warranty of MERCHANTABILITY or
|
WARRANTY; without even the implied warranty of MERCHANTABILITY or
|
FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
for more details.
|
for more details.
|
|
|
You should have received a copy of the GNU General Public License
|
You should have received a copy of the GNU General Public License
|
along with GCC; see the file COPYING3. If not see
|
along with GCC; see the file COPYING3. If not see
|
<http://www.gnu.org/licenses/>. */
|
<http://www.gnu.org/licenses/>. */
|
|
|
/* The purpose of this pass is to split function bodies to improve
|
/* The purpose of this pass is to split function bodies to improve
|
inlining. I.e. for function of the form:
|
inlining. I.e. for function of the form:
|
|
|
func (...)
|
func (...)
|
{
|
{
|
if (cheap_test)
|
if (cheap_test)
|
something_small
|
something_small
|
else
|
else
|
something_big
|
something_big
|
}
|
}
|
|
|
Produce:
|
Produce:
|
|
|
func.part (...)
|
func.part (...)
|
{
|
{
|
something_big
|
something_big
|
}
|
}
|
|
|
func (...)
|
func (...)
|
{
|
{
|
if (cheap_test)
|
if (cheap_test)
|
something_small
|
something_small
|
else
|
else
|
func.part (...);
|
func.part (...);
|
}
|
}
|
|
|
When func becomes inlinable and when cheap_test is often true, inlining func,
|
When func becomes inlinable and when cheap_test is often true, inlining func,
|
but not fund.part leads to performance improvement similar as inlining
|
but not fund.part leads to performance improvement similar as inlining
|
original func while the code size growth is smaller.
|
original func while the code size growth is smaller.
|
|
|
The pass is organized in three stages:
|
The pass is organized in three stages:
|
1) Collect local info about basic block into BB_INFO structure and
|
1) Collect local info about basic block into BB_INFO structure and
|
compute function body estimated size and time.
|
compute function body estimated size and time.
|
2) Via DFS walk find all possible basic blocks where we can split
|
2) Via DFS walk find all possible basic blocks where we can split
|
and chose best one.
|
and chose best one.
|
3) If split point is found, split at the specified BB by creating a clone
|
3) If split point is found, split at the specified BB by creating a clone
|
and updating function to call it.
|
and updating function to call it.
|
|
|
The decisions what functions to split are in execute_split_functions
|
The decisions what functions to split are in execute_split_functions
|
and consider_split.
|
and consider_split.
|
|
|
There are several possible future improvements for this pass including:
|
There are several possible future improvements for this pass including:
|
|
|
1) Splitting to break up large functions
|
1) Splitting to break up large functions
|
2) Splitting to reduce stack frame usage
|
2) Splitting to reduce stack frame usage
|
3) Allow split part of function to use values computed in the header part.
|
3) Allow split part of function to use values computed in the header part.
|
The values needs to be passed to split function, perhaps via same
|
The values needs to be passed to split function, perhaps via same
|
interface as for nested functions or as argument.
|
interface as for nested functions or as argument.
|
4) Support for simple rematerialization. I.e. when split part use
|
4) Support for simple rematerialization. I.e. when split part use
|
value computed in header from function parameter in very cheap way, we
|
value computed in header from function parameter in very cheap way, we
|
can just recompute it.
|
can just recompute it.
|
5) Support splitting of nested functions.
|
5) Support splitting of nested functions.
|
6) Support non-SSA arguments.
|
6) Support non-SSA arguments.
|
7) There is nothing preventing us from producing multiple parts of single function
|
7) There is nothing preventing us from producing multiple parts of single function
|
when needed or splitting also the parts. */
|
when needed or splitting also the parts. */
|
|
|
#include "config.h"
|
#include "config.h"
|
#include "system.h"
|
#include "system.h"
|
#include "coretypes.h"
|
#include "coretypes.h"
|
#include "tree.h"
|
#include "tree.h"
|
#include "target.h"
|
#include "target.h"
|
#include "cgraph.h"
|
#include "cgraph.h"
|
#include "ipa-prop.h"
|
#include "ipa-prop.h"
|
#include "tree-flow.h"
|
#include "tree-flow.h"
|
#include "tree-pass.h"
|
#include "tree-pass.h"
|
#include "flags.h"
|
#include "flags.h"
|
#include "timevar.h"
|
#include "timevar.h"
|
#include "diagnostic.h"
|
#include "diagnostic.h"
|
#include "tree-dump.h"
|
#include "tree-dump.h"
|
#include "tree-inline.h"
|
#include "tree-inline.h"
|
#include "fibheap.h"
|
#include "fibheap.h"
|
#include "params.h"
|
#include "params.h"
|
#include "gimple-pretty-print.h"
|
#include "gimple-pretty-print.h"
|
#include "ipa-inline.h"
|
#include "ipa-inline.h"
|
|
|
/* Per basic block info. */
|
/* Per basic block info. */
|
|
|
typedef struct
|
typedef struct
|
{
|
{
|
unsigned int size;
|
unsigned int size;
|
unsigned int time;
|
unsigned int time;
|
} bb_info;
|
} bb_info;
|
DEF_VEC_O(bb_info);
|
DEF_VEC_O(bb_info);
|
DEF_VEC_ALLOC_O(bb_info,heap);
|
DEF_VEC_ALLOC_O(bb_info,heap);
|
|
|
static VEC(bb_info, heap) *bb_info_vec;
|
static VEC(bb_info, heap) *bb_info_vec;
|
|
|
/* Description of split point. */
|
/* Description of split point. */
|
|
|
struct split_point
|
struct split_point
|
{
|
{
|
/* Size of the partitions. */
|
/* Size of the partitions. */
|
unsigned int header_time, header_size, split_time, split_size;
|
unsigned int header_time, header_size, split_time, split_size;
|
|
|
/* SSA names that need to be passed into spit function. */
|
/* SSA names that need to be passed into spit function. */
|
bitmap ssa_names_to_pass;
|
bitmap ssa_names_to_pass;
|
|
|
/* Basic block where we split (that will become entry point of new function. */
|
/* Basic block where we split (that will become entry point of new function. */
|
basic_block entry_bb;
|
basic_block entry_bb;
|
|
|
/* Basic blocks we are splitting away. */
|
/* Basic blocks we are splitting away. */
|
bitmap split_bbs;
|
bitmap split_bbs;
|
|
|
/* True when return value is computed on split part and thus it needs
|
/* True when return value is computed on split part and thus it needs
|
to be returned. */
|
to be returned. */
|
bool split_part_set_retval;
|
bool split_part_set_retval;
|
};
|
};
|
|
|
/* Best split point found. */
|
/* Best split point found. */
|
|
|
struct split_point best_split_point;
|
struct split_point best_split_point;
|
|
|
/* Set of basic blocks that are not allowed to dominate a split point. */
|
/* Set of basic blocks that are not allowed to dominate a split point. */
|
|
|
static bitmap forbidden_dominators;
|
static bitmap forbidden_dominators;
|
|
|
static tree find_retval (basic_block return_bb);
|
static tree find_retval (basic_block return_bb);
|
|
|
/* Callback for walk_stmt_load_store_addr_ops. If T is non-SSA automatic
|
/* Callback for walk_stmt_load_store_addr_ops. If T is non-SSA automatic
|
variable, check it if it is present in bitmap passed via DATA. */
|
variable, check it if it is present in bitmap passed via DATA. */
|
|
|
static bool
|
static bool
|
test_nonssa_use (gimple stmt ATTRIBUTE_UNUSED, tree t, void *data)
|
test_nonssa_use (gimple stmt ATTRIBUTE_UNUSED, tree t, void *data)
|
{
|
{
|
t = get_base_address (t);
|
t = get_base_address (t);
|
|
|
if (!t || is_gimple_reg (t))
|
if (!t || is_gimple_reg (t))
|
return false;
|
return false;
|
|
|
if (TREE_CODE (t) == PARM_DECL
|
if (TREE_CODE (t) == PARM_DECL
|
|| (TREE_CODE (t) == VAR_DECL
|
|| (TREE_CODE (t) == VAR_DECL
|
&& auto_var_in_fn_p (t, current_function_decl))
|
&& auto_var_in_fn_p (t, current_function_decl))
|
|| TREE_CODE (t) == RESULT_DECL
|
|| TREE_CODE (t) == RESULT_DECL
|
|| TREE_CODE (t) == LABEL_DECL)
|
|| TREE_CODE (t) == LABEL_DECL)
|
return bitmap_bit_p ((bitmap)data, DECL_UID (t));
|
return bitmap_bit_p ((bitmap)data, DECL_UID (t));
|
|
|
/* For DECL_BY_REFERENCE, the return value is actually a pointer. We want
|
/* For DECL_BY_REFERENCE, the return value is actually a pointer. We want
|
to pretend that the value pointed to is actual result decl. */
|
to pretend that the value pointed to is actual result decl. */
|
if ((TREE_CODE (t) == MEM_REF || INDIRECT_REF_P (t))
|
if ((TREE_CODE (t) == MEM_REF || INDIRECT_REF_P (t))
|
&& TREE_CODE (TREE_OPERAND (t, 0)) == SSA_NAME
|
&& TREE_CODE (TREE_OPERAND (t, 0)) == SSA_NAME
|
&& TREE_CODE (SSA_NAME_VAR (TREE_OPERAND (t, 0))) == RESULT_DECL
|
&& TREE_CODE (SSA_NAME_VAR (TREE_OPERAND (t, 0))) == RESULT_DECL
|
&& DECL_BY_REFERENCE (DECL_RESULT (current_function_decl)))
|
&& DECL_BY_REFERENCE (DECL_RESULT (current_function_decl)))
|
return
|
return
|
bitmap_bit_p ((bitmap)data,
|
bitmap_bit_p ((bitmap)data,
|
DECL_UID (DECL_RESULT (current_function_decl)));
|
DECL_UID (DECL_RESULT (current_function_decl)));
|
|
|
return false;
|
return false;
|
}
|
}
|
|
|
/* Dump split point CURRENT. */
|
/* Dump split point CURRENT. */
|
|
|
static void
|
static void
|
dump_split_point (FILE * file, struct split_point *current)
|
dump_split_point (FILE * file, struct split_point *current)
|
{
|
{
|
fprintf (file,
|
fprintf (file,
|
"Split point at BB %i\n"
|
"Split point at BB %i\n"
|
" header time: %i header size: %i\n"
|
" header time: %i header size: %i\n"
|
" split time: %i split size: %i\n bbs: ",
|
" split time: %i split size: %i\n bbs: ",
|
current->entry_bb->index, current->header_time,
|
current->entry_bb->index, current->header_time,
|
current->header_size, current->split_time, current->split_size);
|
current->header_size, current->split_time, current->split_size);
|
dump_bitmap (file, current->split_bbs);
|
dump_bitmap (file, current->split_bbs);
|
fprintf (file, " SSA names to pass: ");
|
fprintf (file, " SSA names to pass: ");
|
dump_bitmap (file, current->ssa_names_to_pass);
|
dump_bitmap (file, current->ssa_names_to_pass);
|
}
|
}
|
|
|
/* Look for all BBs in header that might lead to the split part and verify
|
/* Look for all BBs in header that might lead to the split part and verify
|
that they are not defining any non-SSA var used by the split part.
|
that they are not defining any non-SSA var used by the split part.
|
Parameters are the same as for consider_split. */
|
Parameters are the same as for consider_split. */
|
|
|
static bool
|
static bool
|
verify_non_ssa_vars (struct split_point *current, bitmap non_ssa_vars,
|
verify_non_ssa_vars (struct split_point *current, bitmap non_ssa_vars,
|
basic_block return_bb)
|
basic_block return_bb)
|
{
|
{
|
bitmap seen = BITMAP_ALLOC (NULL);
|
bitmap seen = BITMAP_ALLOC (NULL);
|
VEC (basic_block,heap) *worklist = NULL;
|
VEC (basic_block,heap) *worklist = NULL;
|
edge e;
|
edge e;
|
edge_iterator ei;
|
edge_iterator ei;
|
bool ok = true;
|
bool ok = true;
|
|
|
FOR_EACH_EDGE (e, ei, current->entry_bb->preds)
|
FOR_EACH_EDGE (e, ei, current->entry_bb->preds)
|
if (e->src != ENTRY_BLOCK_PTR
|
if (e->src != ENTRY_BLOCK_PTR
|
&& !bitmap_bit_p (current->split_bbs, e->src->index))
|
&& !bitmap_bit_p (current->split_bbs, e->src->index))
|
{
|
{
|
VEC_safe_push (basic_block, heap, worklist, e->src);
|
VEC_safe_push (basic_block, heap, worklist, e->src);
|
bitmap_set_bit (seen, e->src->index);
|
bitmap_set_bit (seen, e->src->index);
|
}
|
}
|
|
|
while (!VEC_empty (basic_block, worklist))
|
while (!VEC_empty (basic_block, worklist))
|
{
|
{
|
gimple_stmt_iterator bsi;
|
gimple_stmt_iterator bsi;
|
basic_block bb = VEC_pop (basic_block, worklist);
|
basic_block bb = VEC_pop (basic_block, worklist);
|
|
|
FOR_EACH_EDGE (e, ei, bb->preds)
|
FOR_EACH_EDGE (e, ei, bb->preds)
|
if (e->src != ENTRY_BLOCK_PTR
|
if (e->src != ENTRY_BLOCK_PTR
|
&& bitmap_set_bit (seen, e->src->index))
|
&& bitmap_set_bit (seen, e->src->index))
|
{
|
{
|
gcc_checking_assert (!bitmap_bit_p (current->split_bbs,
|
gcc_checking_assert (!bitmap_bit_p (current->split_bbs,
|
e->src->index));
|
e->src->index));
|
VEC_safe_push (basic_block, heap, worklist, e->src);
|
VEC_safe_push (basic_block, heap, worklist, e->src);
|
}
|
}
|
for (bsi = gsi_start_bb (bb); !gsi_end_p (bsi); gsi_next (&bsi))
|
for (bsi = gsi_start_bb (bb); !gsi_end_p (bsi); gsi_next (&bsi))
|
{
|
{
|
gimple stmt = gsi_stmt (bsi);
|
gimple stmt = gsi_stmt (bsi);
|
if (is_gimple_debug (stmt))
|
if (is_gimple_debug (stmt))
|
continue;
|
continue;
|
if (walk_stmt_load_store_addr_ops
|
if (walk_stmt_load_store_addr_ops
|
(stmt, non_ssa_vars, test_nonssa_use, test_nonssa_use,
|
(stmt, non_ssa_vars, test_nonssa_use, test_nonssa_use,
|
test_nonssa_use))
|
test_nonssa_use))
|
{
|
{
|
ok = false;
|
ok = false;
|
goto done;
|
goto done;
|
}
|
}
|
if (gimple_code (stmt) == GIMPLE_LABEL
|
if (gimple_code (stmt) == GIMPLE_LABEL
|
&& test_nonssa_use (stmt, gimple_label_label (stmt),
|
&& test_nonssa_use (stmt, gimple_label_label (stmt),
|
non_ssa_vars))
|
non_ssa_vars))
|
{
|
{
|
ok = false;
|
ok = false;
|
goto done;
|
goto done;
|
}
|
}
|
}
|
}
|
for (bsi = gsi_start_phis (bb); !gsi_end_p (bsi); gsi_next (&bsi))
|
for (bsi = gsi_start_phis (bb); !gsi_end_p (bsi); gsi_next (&bsi))
|
{
|
{
|
if (walk_stmt_load_store_addr_ops
|
if (walk_stmt_load_store_addr_ops
|
(gsi_stmt (bsi), non_ssa_vars, test_nonssa_use, test_nonssa_use,
|
(gsi_stmt (bsi), non_ssa_vars, test_nonssa_use, test_nonssa_use,
|
test_nonssa_use))
|
test_nonssa_use))
|
{
|
{
|
ok = false;
|
ok = false;
|
goto done;
|
goto done;
|
}
|
}
|
}
|
}
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
{
|
{
|
if (e->dest != return_bb)
|
if (e->dest != return_bb)
|
continue;
|
continue;
|
for (bsi = gsi_start_phis (return_bb); !gsi_end_p (bsi);
|
for (bsi = gsi_start_phis (return_bb); !gsi_end_p (bsi);
|
gsi_next (&bsi))
|
gsi_next (&bsi))
|
{
|
{
|
gimple stmt = gsi_stmt (bsi);
|
gimple stmt = gsi_stmt (bsi);
|
tree op = gimple_phi_arg_def (stmt, e->dest_idx);
|
tree op = gimple_phi_arg_def (stmt, e->dest_idx);
|
|
|
if (!is_gimple_reg (gimple_phi_result (stmt)))
|
if (!is_gimple_reg (gimple_phi_result (stmt)))
|
continue;
|
continue;
|
if (TREE_CODE (op) != SSA_NAME
|
if (TREE_CODE (op) != SSA_NAME
|
&& test_nonssa_use (stmt, op, non_ssa_vars))
|
&& test_nonssa_use (stmt, op, non_ssa_vars))
|
{
|
{
|
ok = false;
|
ok = false;
|
goto done;
|
goto done;
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
done:
|
done:
|
BITMAP_FREE (seen);
|
BITMAP_FREE (seen);
|
VEC_free (basic_block, heap, worklist);
|
VEC_free (basic_block, heap, worklist);
|
return ok;
|
return ok;
|
}
|
}
|
|
|
/* If STMT is a call, check the callee against a list of forbidden
|
/* If STMT is a call, check the callee against a list of forbidden
|
predicate functions. If a match is found, look for uses of the
|
predicate functions. If a match is found, look for uses of the
|
call result in condition statements that compare against zero.
|
call result in condition statements that compare against zero.
|
For each such use, find the block targeted by the condition
|
For each such use, find the block targeted by the condition
|
statement for the nonzero result, and set the bit for this block
|
statement for the nonzero result, and set the bit for this block
|
in the forbidden dominators bitmap. The purpose of this is to avoid
|
in the forbidden dominators bitmap. The purpose of this is to avoid
|
selecting a split point where we are likely to lose the chance
|
selecting a split point where we are likely to lose the chance
|
to optimize away an unused function call. */
|
to optimize away an unused function call. */
|
|
|
static void
|
static void
|
check_forbidden_calls (gimple stmt)
|
check_forbidden_calls (gimple stmt)
|
{
|
{
|
imm_use_iterator use_iter;
|
imm_use_iterator use_iter;
|
use_operand_p use_p;
|
use_operand_p use_p;
|
tree lhs;
|
tree lhs;
|
|
|
/* At the moment, __builtin_constant_p is the only forbidden
|
/* At the moment, __builtin_constant_p is the only forbidden
|
predicate function call (see PR49642). */
|
predicate function call (see PR49642). */
|
if (!gimple_call_builtin_p (stmt, BUILT_IN_CONSTANT_P))
|
if (!gimple_call_builtin_p (stmt, BUILT_IN_CONSTANT_P))
|
return;
|
return;
|
|
|
lhs = gimple_call_lhs (stmt);
|
lhs = gimple_call_lhs (stmt);
|
|
|
if (!lhs || TREE_CODE (lhs) != SSA_NAME)
|
if (!lhs || TREE_CODE (lhs) != SSA_NAME)
|
return;
|
return;
|
|
|
FOR_EACH_IMM_USE_FAST (use_p, use_iter, lhs)
|
FOR_EACH_IMM_USE_FAST (use_p, use_iter, lhs)
|
{
|
{
|
tree op1;
|
tree op1;
|
basic_block use_bb, forbidden_bb;
|
basic_block use_bb, forbidden_bb;
|
enum tree_code code;
|
enum tree_code code;
|
edge true_edge, false_edge;
|
edge true_edge, false_edge;
|
gimple use_stmt = USE_STMT (use_p);
|
gimple use_stmt = USE_STMT (use_p);
|
|
|
if (gimple_code (use_stmt) != GIMPLE_COND)
|
if (gimple_code (use_stmt) != GIMPLE_COND)
|
continue;
|
continue;
|
|
|
/* Assuming canonical form for GIMPLE_COND here, with constant
|
/* Assuming canonical form for GIMPLE_COND here, with constant
|
in second position. */
|
in second position. */
|
op1 = gimple_cond_rhs (use_stmt);
|
op1 = gimple_cond_rhs (use_stmt);
|
code = gimple_cond_code (use_stmt);
|
code = gimple_cond_code (use_stmt);
|
use_bb = gimple_bb (use_stmt);
|
use_bb = gimple_bb (use_stmt);
|
|
|
extract_true_false_edges_from_block (use_bb, &true_edge, &false_edge);
|
extract_true_false_edges_from_block (use_bb, &true_edge, &false_edge);
|
|
|
/* We're only interested in comparisons that distinguish
|
/* We're only interested in comparisons that distinguish
|
unambiguously from zero. */
|
unambiguously from zero. */
|
if (!integer_zerop (op1) || code == LE_EXPR || code == GE_EXPR)
|
if (!integer_zerop (op1) || code == LE_EXPR || code == GE_EXPR)
|
continue;
|
continue;
|
|
|
if (code == EQ_EXPR)
|
if (code == EQ_EXPR)
|
forbidden_bb = false_edge->dest;
|
forbidden_bb = false_edge->dest;
|
else
|
else
|
forbidden_bb = true_edge->dest;
|
forbidden_bb = true_edge->dest;
|
|
|
bitmap_set_bit (forbidden_dominators, forbidden_bb->index);
|
bitmap_set_bit (forbidden_dominators, forbidden_bb->index);
|
}
|
}
|
}
|
}
|
|
|
/* If BB is dominated by any block in the forbidden dominators set,
|
/* If BB is dominated by any block in the forbidden dominators set,
|
return TRUE; else FALSE. */
|
return TRUE; else FALSE. */
|
|
|
static bool
|
static bool
|
dominated_by_forbidden (basic_block bb)
|
dominated_by_forbidden (basic_block bb)
|
{
|
{
|
unsigned dom_bb;
|
unsigned dom_bb;
|
bitmap_iterator bi;
|
bitmap_iterator bi;
|
|
|
EXECUTE_IF_SET_IN_BITMAP (forbidden_dominators, 1, dom_bb, bi)
|
EXECUTE_IF_SET_IN_BITMAP (forbidden_dominators, 1, dom_bb, bi)
|
{
|
{
|
if (dominated_by_p (CDI_DOMINATORS, bb, BASIC_BLOCK (dom_bb)))
|
if (dominated_by_p (CDI_DOMINATORS, bb, BASIC_BLOCK (dom_bb)))
|
return true;
|
return true;
|
}
|
}
|
|
|
return false;
|
return false;
|
}
|
}
|
|
|
/* We found an split_point CURRENT. NON_SSA_VARS is bitmap of all non ssa
|
/* We found an split_point CURRENT. NON_SSA_VARS is bitmap of all non ssa
|
variables used and RETURN_BB is return basic block.
|
variables used and RETURN_BB is return basic block.
|
See if we can split function here. */
|
See if we can split function here. */
|
|
|
static void
|
static void
|
consider_split (struct split_point *current, bitmap non_ssa_vars,
|
consider_split (struct split_point *current, bitmap non_ssa_vars,
|
basic_block return_bb)
|
basic_block return_bb)
|
{
|
{
|
tree parm;
|
tree parm;
|
unsigned int num_args = 0;
|
unsigned int num_args = 0;
|
unsigned int call_overhead;
|
unsigned int call_overhead;
|
edge e;
|
edge e;
|
edge_iterator ei;
|
edge_iterator ei;
|
gimple_stmt_iterator bsi;
|
gimple_stmt_iterator bsi;
|
unsigned int i;
|
unsigned int i;
|
int incoming_freq = 0;
|
int incoming_freq = 0;
|
tree retval;
|
tree retval;
|
|
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
dump_split_point (dump_file, current);
|
dump_split_point (dump_file, current);
|
|
|
FOR_EACH_EDGE (e, ei, current->entry_bb->preds)
|
FOR_EACH_EDGE (e, ei, current->entry_bb->preds)
|
if (!bitmap_bit_p (current->split_bbs, e->src->index))
|
if (!bitmap_bit_p (current->split_bbs, e->src->index))
|
incoming_freq += EDGE_FREQUENCY (e);
|
incoming_freq += EDGE_FREQUENCY (e);
|
|
|
/* Do not split when we would end up calling function anyway. */
|
/* Do not split when we would end up calling function anyway. */
|
if (incoming_freq
|
if (incoming_freq
|
>= (ENTRY_BLOCK_PTR->frequency
|
>= (ENTRY_BLOCK_PTR->frequency
|
* PARAM_VALUE (PARAM_PARTIAL_INLINING_ENTRY_PROBABILITY) / 100))
|
* PARAM_VALUE (PARAM_PARTIAL_INLINING_ENTRY_PROBABILITY) / 100))
|
{
|
{
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
fprintf (dump_file,
|
fprintf (dump_file,
|
" Refused: incoming frequency is too large.\n");
|
" Refused: incoming frequency is too large.\n");
|
return;
|
return;
|
}
|
}
|
|
|
if (!current->header_size)
|
if (!current->header_size)
|
{
|
{
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
fprintf (dump_file, " Refused: header empty\n");
|
fprintf (dump_file, " Refused: header empty\n");
|
return;
|
return;
|
}
|
}
|
|
|
/* Verify that PHI args on entry are either virtual or all their operands
|
/* Verify that PHI args on entry are either virtual or all their operands
|
incoming from header are the same. */
|
incoming from header are the same. */
|
for (bsi = gsi_start_phis (current->entry_bb); !gsi_end_p (bsi); gsi_next (&bsi))
|
for (bsi = gsi_start_phis (current->entry_bb); !gsi_end_p (bsi); gsi_next (&bsi))
|
{
|
{
|
gimple stmt = gsi_stmt (bsi);
|
gimple stmt = gsi_stmt (bsi);
|
tree val = NULL;
|
tree val = NULL;
|
|
|
if (!is_gimple_reg (gimple_phi_result (stmt)))
|
if (!is_gimple_reg (gimple_phi_result (stmt)))
|
continue;
|
continue;
|
for (i = 0; i < gimple_phi_num_args (stmt); i++)
|
for (i = 0; i < gimple_phi_num_args (stmt); i++)
|
{
|
{
|
edge e = gimple_phi_arg_edge (stmt, i);
|
edge e = gimple_phi_arg_edge (stmt, i);
|
if (!bitmap_bit_p (current->split_bbs, e->src->index))
|
if (!bitmap_bit_p (current->split_bbs, e->src->index))
|
{
|
{
|
tree edge_val = gimple_phi_arg_def (stmt, i);
|
tree edge_val = gimple_phi_arg_def (stmt, i);
|
if (val && edge_val != val)
|
if (val && edge_val != val)
|
{
|
{
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
fprintf (dump_file,
|
fprintf (dump_file,
|
" Refused: entry BB has PHI with multiple variants\n");
|
" Refused: entry BB has PHI with multiple variants\n");
|
return;
|
return;
|
}
|
}
|
val = edge_val;
|
val = edge_val;
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
|
|
/* See what argument we will pass to the split function and compute
|
/* See what argument we will pass to the split function and compute
|
call overhead. */
|
call overhead. */
|
call_overhead = eni_size_weights.call_cost;
|
call_overhead = eni_size_weights.call_cost;
|
for (parm = DECL_ARGUMENTS (current_function_decl); parm;
|
for (parm = DECL_ARGUMENTS (current_function_decl); parm;
|
parm = DECL_CHAIN (parm))
|
parm = DECL_CHAIN (parm))
|
{
|
{
|
if (!is_gimple_reg (parm))
|
if (!is_gimple_reg (parm))
|
{
|
{
|
if (bitmap_bit_p (non_ssa_vars, DECL_UID (parm)))
|
if (bitmap_bit_p (non_ssa_vars, DECL_UID (parm)))
|
{
|
{
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
fprintf (dump_file,
|
fprintf (dump_file,
|
" Refused: need to pass non-ssa param values\n");
|
" Refused: need to pass non-ssa param values\n");
|
return;
|
return;
|
}
|
}
|
}
|
}
|
else if (gimple_default_def (cfun, parm)
|
else if (gimple_default_def (cfun, parm)
|
&& bitmap_bit_p (current->ssa_names_to_pass,
|
&& bitmap_bit_p (current->ssa_names_to_pass,
|
SSA_NAME_VERSION (gimple_default_def
|
SSA_NAME_VERSION (gimple_default_def
|
(cfun, parm))))
|
(cfun, parm))))
|
{
|
{
|
if (!VOID_TYPE_P (TREE_TYPE (parm)))
|
if (!VOID_TYPE_P (TREE_TYPE (parm)))
|
call_overhead += estimate_move_cost (TREE_TYPE (parm));
|
call_overhead += estimate_move_cost (TREE_TYPE (parm));
|
num_args++;
|
num_args++;
|
}
|
}
|
}
|
}
|
if (!VOID_TYPE_P (TREE_TYPE (current_function_decl)))
|
if (!VOID_TYPE_P (TREE_TYPE (current_function_decl)))
|
call_overhead += estimate_move_cost (TREE_TYPE (current_function_decl));
|
call_overhead += estimate_move_cost (TREE_TYPE (current_function_decl));
|
|
|
if (current->split_size <= call_overhead)
|
if (current->split_size <= call_overhead)
|
{
|
{
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
fprintf (dump_file,
|
fprintf (dump_file,
|
" Refused: split size is smaller than call overhead\n");
|
" Refused: split size is smaller than call overhead\n");
|
return;
|
return;
|
}
|
}
|
if (current->header_size + call_overhead
|
if (current->header_size + call_overhead
|
>= (unsigned int)(DECL_DECLARED_INLINE_P (current_function_decl)
|
>= (unsigned int)(DECL_DECLARED_INLINE_P (current_function_decl)
|
? MAX_INLINE_INSNS_SINGLE
|
? MAX_INLINE_INSNS_SINGLE
|
: MAX_INLINE_INSNS_AUTO))
|
: MAX_INLINE_INSNS_AUTO))
|
{
|
{
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
fprintf (dump_file,
|
fprintf (dump_file,
|
" Refused: header size is too large for inline candidate\n");
|
" Refused: header size is too large for inline candidate\n");
|
return;
|
return;
|
}
|
}
|
|
|
/* FIXME: we currently can pass only SSA function parameters to the split
|
/* FIXME: we currently can pass only SSA function parameters to the split
|
arguments. Once parm_adjustment infrastructure is supported by cloning,
|
arguments. Once parm_adjustment infrastructure is supported by cloning,
|
we can pass more than that. */
|
we can pass more than that. */
|
if (num_args != bitmap_count_bits (current->ssa_names_to_pass))
|
if (num_args != bitmap_count_bits (current->ssa_names_to_pass))
|
{
|
{
|
|
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
fprintf (dump_file,
|
fprintf (dump_file,
|
" Refused: need to pass non-param values\n");
|
" Refused: need to pass non-param values\n");
|
return;
|
return;
|
}
|
}
|
|
|
/* When there are non-ssa vars used in the split region, see if they
|
/* When there are non-ssa vars used in the split region, see if they
|
are used in the header region. If so, reject the split.
|
are used in the header region. If so, reject the split.
|
FIXME: we can use nested function support to access both. */
|
FIXME: we can use nested function support to access both. */
|
if (!bitmap_empty_p (non_ssa_vars)
|
if (!bitmap_empty_p (non_ssa_vars)
|
&& !verify_non_ssa_vars (current, non_ssa_vars, return_bb))
|
&& !verify_non_ssa_vars (current, non_ssa_vars, return_bb))
|
{
|
{
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
fprintf (dump_file,
|
fprintf (dump_file,
|
" Refused: split part has non-ssa uses\n");
|
" Refused: split part has non-ssa uses\n");
|
return;
|
return;
|
}
|
}
|
|
|
/* If the split point is dominated by a forbidden block, reject
|
/* If the split point is dominated by a forbidden block, reject
|
the split. */
|
the split. */
|
if (!bitmap_empty_p (forbidden_dominators)
|
if (!bitmap_empty_p (forbidden_dominators)
|
&& dominated_by_forbidden (current->entry_bb))
|
&& dominated_by_forbidden (current->entry_bb))
|
{
|
{
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
fprintf (dump_file,
|
fprintf (dump_file,
|
" Refused: split point dominated by forbidden block\n");
|
" Refused: split point dominated by forbidden block\n");
|
return;
|
return;
|
}
|
}
|
|
|
/* See if retval used by return bb is computed by header or split part.
|
/* See if retval used by return bb is computed by header or split part.
|
When it is computed by split part, we need to produce return statement
|
When it is computed by split part, we need to produce return statement
|
in the split part and add code to header to pass it around.
|
in the split part and add code to header to pass it around.
|
|
|
This is bit tricky to test:
|
This is bit tricky to test:
|
1) When there is no return_bb or no return value, we always pass
|
1) When there is no return_bb or no return value, we always pass
|
value around.
|
value around.
|
2) Invariants are always computed by caller.
|
2) Invariants are always computed by caller.
|
3) For SSA we need to look if defining statement is in header or split part
|
3) For SSA we need to look if defining statement is in header or split part
|
4) For non-SSA we need to look where the var is computed. */
|
4) For non-SSA we need to look where the var is computed. */
|
retval = find_retval (return_bb);
|
retval = find_retval (return_bb);
|
if (!retval)
|
if (!retval)
|
current->split_part_set_retval = true;
|
current->split_part_set_retval = true;
|
else if (is_gimple_min_invariant (retval))
|
else if (is_gimple_min_invariant (retval))
|
current->split_part_set_retval = false;
|
current->split_part_set_retval = false;
|
/* Special case is value returned by reference we record as if it was non-ssa
|
/* Special case is value returned by reference we record as if it was non-ssa
|
set to result_decl. */
|
set to result_decl. */
|
else if (TREE_CODE (retval) == SSA_NAME
|
else if (TREE_CODE (retval) == SSA_NAME
|
&& TREE_CODE (SSA_NAME_VAR (retval)) == RESULT_DECL
|
&& TREE_CODE (SSA_NAME_VAR (retval)) == RESULT_DECL
|
&& DECL_BY_REFERENCE (DECL_RESULT (current_function_decl)))
|
&& DECL_BY_REFERENCE (DECL_RESULT (current_function_decl)))
|
current->split_part_set_retval
|
current->split_part_set_retval
|
= bitmap_bit_p (non_ssa_vars, DECL_UID (SSA_NAME_VAR (retval)));
|
= bitmap_bit_p (non_ssa_vars, DECL_UID (SSA_NAME_VAR (retval)));
|
else if (TREE_CODE (retval) == SSA_NAME)
|
else if (TREE_CODE (retval) == SSA_NAME)
|
current->split_part_set_retval
|
current->split_part_set_retval
|
= (!SSA_NAME_IS_DEFAULT_DEF (retval)
|
= (!SSA_NAME_IS_DEFAULT_DEF (retval)
|
&& (bitmap_bit_p (current->split_bbs,
|
&& (bitmap_bit_p (current->split_bbs,
|
gimple_bb (SSA_NAME_DEF_STMT (retval))->index)
|
gimple_bb (SSA_NAME_DEF_STMT (retval))->index)
|
|| gimple_bb (SSA_NAME_DEF_STMT (retval)) == return_bb));
|
|| gimple_bb (SSA_NAME_DEF_STMT (retval)) == return_bb));
|
else if (TREE_CODE (retval) == PARM_DECL)
|
else if (TREE_CODE (retval) == PARM_DECL)
|
current->split_part_set_retval = false;
|
current->split_part_set_retval = false;
|
else if (TREE_CODE (retval) == VAR_DECL
|
else if (TREE_CODE (retval) == VAR_DECL
|
|| TREE_CODE (retval) == RESULT_DECL)
|
|| TREE_CODE (retval) == RESULT_DECL)
|
current->split_part_set_retval
|
current->split_part_set_retval
|
= bitmap_bit_p (non_ssa_vars, DECL_UID (retval));
|
= bitmap_bit_p (non_ssa_vars, DECL_UID (retval));
|
else
|
else
|
current->split_part_set_retval = true;
|
current->split_part_set_retval = true;
|
|
|
/* split_function fixes up at most one PHI non-virtual PHI node in return_bb,
|
/* split_function fixes up at most one PHI non-virtual PHI node in return_bb,
|
for the return value. If there are other PHIs, give up. */
|
for the return value. If there are other PHIs, give up. */
|
if (return_bb != EXIT_BLOCK_PTR)
|
if (return_bb != EXIT_BLOCK_PTR)
|
{
|
{
|
gimple_stmt_iterator psi;
|
gimple_stmt_iterator psi;
|
|
|
for (psi = gsi_start_phis (return_bb); !gsi_end_p (psi); gsi_next (&psi))
|
for (psi = gsi_start_phis (return_bb); !gsi_end_p (psi); gsi_next (&psi))
|
if (is_gimple_reg (gimple_phi_result (gsi_stmt (psi)))
|
if (is_gimple_reg (gimple_phi_result (gsi_stmt (psi)))
|
&& !(retval
|
&& !(retval
|
&& current->split_part_set_retval
|
&& current->split_part_set_retval
|
&& TREE_CODE (retval) == SSA_NAME
|
&& TREE_CODE (retval) == SSA_NAME
|
&& !DECL_BY_REFERENCE (DECL_RESULT (current_function_decl))
|
&& !DECL_BY_REFERENCE (DECL_RESULT (current_function_decl))
|
&& SSA_NAME_DEF_STMT (retval) == gsi_stmt (psi)))
|
&& SSA_NAME_DEF_STMT (retval) == gsi_stmt (psi)))
|
{
|
{
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
fprintf (dump_file,
|
fprintf (dump_file,
|
" Refused: return bb has extra PHIs\n");
|
" Refused: return bb has extra PHIs\n");
|
return;
|
return;
|
}
|
}
|
}
|
}
|
|
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
fprintf (dump_file, " Accepted!\n");
|
fprintf (dump_file, " Accepted!\n");
|
|
|
/* At the moment chose split point with lowest frequency and that leaves
|
/* At the moment chose split point with lowest frequency and that leaves
|
out smallest size of header.
|
out smallest size of header.
|
In future we might re-consider this heuristics. */
|
In future we might re-consider this heuristics. */
|
if (!best_split_point.split_bbs
|
if (!best_split_point.split_bbs
|
|| best_split_point.entry_bb->frequency > current->entry_bb->frequency
|
|| best_split_point.entry_bb->frequency > current->entry_bb->frequency
|
|| (best_split_point.entry_bb->frequency == current->entry_bb->frequency
|
|| (best_split_point.entry_bb->frequency == current->entry_bb->frequency
|
&& best_split_point.split_size < current->split_size))
|
&& best_split_point.split_size < current->split_size))
|
|
|
{
|
{
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
fprintf (dump_file, " New best split point!\n");
|
fprintf (dump_file, " New best split point!\n");
|
if (best_split_point.ssa_names_to_pass)
|
if (best_split_point.ssa_names_to_pass)
|
{
|
{
|
BITMAP_FREE (best_split_point.ssa_names_to_pass);
|
BITMAP_FREE (best_split_point.ssa_names_to_pass);
|
BITMAP_FREE (best_split_point.split_bbs);
|
BITMAP_FREE (best_split_point.split_bbs);
|
}
|
}
|
best_split_point = *current;
|
best_split_point = *current;
|
best_split_point.ssa_names_to_pass = BITMAP_ALLOC (NULL);
|
best_split_point.ssa_names_to_pass = BITMAP_ALLOC (NULL);
|
bitmap_copy (best_split_point.ssa_names_to_pass,
|
bitmap_copy (best_split_point.ssa_names_to_pass,
|
current->ssa_names_to_pass);
|
current->ssa_names_to_pass);
|
best_split_point.split_bbs = BITMAP_ALLOC (NULL);
|
best_split_point.split_bbs = BITMAP_ALLOC (NULL);
|
bitmap_copy (best_split_point.split_bbs, current->split_bbs);
|
bitmap_copy (best_split_point.split_bbs, current->split_bbs);
|
}
|
}
|
}
|
}
|
|
|
/* Return basic block containing RETURN statement. We allow basic blocks
|
/* Return basic block containing RETURN statement. We allow basic blocks
|
of the form:
|
of the form:
|
<retval> = tmp_var;
|
<retval> = tmp_var;
|
return <retval>
|
return <retval>
|
but return_bb can not be more complex than this.
|
but return_bb can not be more complex than this.
|
If nothing is found, return EXIT_BLOCK_PTR.
|
If nothing is found, return EXIT_BLOCK_PTR.
|
|
|
When there are multiple RETURN statement, chose one with return value,
|
When there are multiple RETURN statement, chose one with return value,
|
since that one is more likely shared by multiple code paths.
|
since that one is more likely shared by multiple code paths.
|
|
|
Return BB is special, because for function splitting it is the only
|
Return BB is special, because for function splitting it is the only
|
basic block that is duplicated in between header and split part of the
|
basic block that is duplicated in between header and split part of the
|
function.
|
function.
|
|
|
TODO: We might support multiple return blocks. */
|
TODO: We might support multiple return blocks. */
|
|
|
static basic_block
|
static basic_block
|
find_return_bb (void)
|
find_return_bb (void)
|
{
|
{
|
edge e;
|
edge e;
|
basic_block return_bb = EXIT_BLOCK_PTR;
|
basic_block return_bb = EXIT_BLOCK_PTR;
|
gimple_stmt_iterator bsi;
|
gimple_stmt_iterator bsi;
|
bool found_return = false;
|
bool found_return = false;
|
tree retval = NULL_TREE;
|
tree retval = NULL_TREE;
|
|
|
if (!single_pred_p (EXIT_BLOCK_PTR))
|
if (!single_pred_p (EXIT_BLOCK_PTR))
|
return return_bb;
|
return return_bb;
|
|
|
e = single_pred_edge (EXIT_BLOCK_PTR);
|
e = single_pred_edge (EXIT_BLOCK_PTR);
|
for (bsi = gsi_last_bb (e->src); !gsi_end_p (bsi); gsi_prev (&bsi))
|
for (bsi = gsi_last_bb (e->src); !gsi_end_p (bsi); gsi_prev (&bsi))
|
{
|
{
|
gimple stmt = gsi_stmt (bsi);
|
gimple stmt = gsi_stmt (bsi);
|
if (gimple_code (stmt) == GIMPLE_LABEL
|
if (gimple_code (stmt) == GIMPLE_LABEL
|
|| is_gimple_debug (stmt)
|
|| is_gimple_debug (stmt)
|
|| gimple_clobber_p (stmt))
|
|| gimple_clobber_p (stmt))
|
;
|
;
|
else if (gimple_code (stmt) == GIMPLE_ASSIGN
|
else if (gimple_code (stmt) == GIMPLE_ASSIGN
|
&& found_return
|
&& found_return
|
&& gimple_assign_single_p (stmt)
|
&& gimple_assign_single_p (stmt)
|
&& (auto_var_in_fn_p (gimple_assign_rhs1 (stmt),
|
&& (auto_var_in_fn_p (gimple_assign_rhs1 (stmt),
|
current_function_decl)
|
current_function_decl)
|
|| is_gimple_min_invariant (gimple_assign_rhs1 (stmt)))
|
|| is_gimple_min_invariant (gimple_assign_rhs1 (stmt)))
|
&& retval == gimple_assign_lhs (stmt))
|
&& retval == gimple_assign_lhs (stmt))
|
;
|
;
|
else if (gimple_code (stmt) == GIMPLE_RETURN)
|
else if (gimple_code (stmt) == GIMPLE_RETURN)
|
{
|
{
|
found_return = true;
|
found_return = true;
|
retval = gimple_return_retval (stmt);
|
retval = gimple_return_retval (stmt);
|
}
|
}
|
else
|
else
|
break;
|
break;
|
}
|
}
|
if (gsi_end_p (bsi) && found_return)
|
if (gsi_end_p (bsi) && found_return)
|
return_bb = e->src;
|
return_bb = e->src;
|
|
|
return return_bb;
|
return return_bb;
|
}
|
}
|
|
|
/* Given return basic block RETURN_BB, see where return value is really
|
/* Given return basic block RETURN_BB, see where return value is really
|
stored. */
|
stored. */
|
static tree
|
static tree
|
find_retval (basic_block return_bb)
|
find_retval (basic_block return_bb)
|
{
|
{
|
gimple_stmt_iterator bsi;
|
gimple_stmt_iterator bsi;
|
for (bsi = gsi_start_bb (return_bb); !gsi_end_p (bsi); gsi_next (&bsi))
|
for (bsi = gsi_start_bb (return_bb); !gsi_end_p (bsi); gsi_next (&bsi))
|
if (gimple_code (gsi_stmt (bsi)) == GIMPLE_RETURN)
|
if (gimple_code (gsi_stmt (bsi)) == GIMPLE_RETURN)
|
return gimple_return_retval (gsi_stmt (bsi));
|
return gimple_return_retval (gsi_stmt (bsi));
|
else if (gimple_code (gsi_stmt (bsi)) == GIMPLE_ASSIGN
|
else if (gimple_code (gsi_stmt (bsi)) == GIMPLE_ASSIGN
|
&& !gimple_clobber_p (gsi_stmt (bsi)))
|
&& !gimple_clobber_p (gsi_stmt (bsi)))
|
return gimple_assign_rhs1 (gsi_stmt (bsi));
|
return gimple_assign_rhs1 (gsi_stmt (bsi));
|
return NULL;
|
return NULL;
|
}
|
}
|
|
|
/* Callback for walk_stmt_load_store_addr_ops. If T is non-SSA automatic
|
/* Callback for walk_stmt_load_store_addr_ops. If T is non-SSA automatic
|
variable, mark it as used in bitmap passed via DATA.
|
variable, mark it as used in bitmap passed via DATA.
|
Return true when access to T prevents splitting the function. */
|
Return true when access to T prevents splitting the function. */
|
|
|
static bool
|
static bool
|
mark_nonssa_use (gimple stmt ATTRIBUTE_UNUSED, tree t, void *data)
|
mark_nonssa_use (gimple stmt ATTRIBUTE_UNUSED, tree t, void *data)
|
{
|
{
|
t = get_base_address (t);
|
t = get_base_address (t);
|
|
|
if (!t || is_gimple_reg (t))
|
if (!t || is_gimple_reg (t))
|
return false;
|
return false;
|
|
|
/* At present we can't pass non-SSA arguments to split function.
|
/* At present we can't pass non-SSA arguments to split function.
|
FIXME: this can be relaxed by passing references to arguments. */
|
FIXME: this can be relaxed by passing references to arguments. */
|
if (TREE_CODE (t) == PARM_DECL)
|
if (TREE_CODE (t) == PARM_DECL)
|
{
|
{
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
fprintf (dump_file,
|
fprintf (dump_file,
|
"Cannot split: use of non-ssa function parameter.\n");
|
"Cannot split: use of non-ssa function parameter.\n");
|
return true;
|
return true;
|
}
|
}
|
|
|
if ((TREE_CODE (t) == VAR_DECL
|
if ((TREE_CODE (t) == VAR_DECL
|
&& auto_var_in_fn_p (t, current_function_decl))
|
&& auto_var_in_fn_p (t, current_function_decl))
|
|| TREE_CODE (t) == RESULT_DECL
|
|| TREE_CODE (t) == RESULT_DECL
|
|| TREE_CODE (t) == LABEL_DECL)
|
|| TREE_CODE (t) == LABEL_DECL)
|
bitmap_set_bit ((bitmap)data, DECL_UID (t));
|
bitmap_set_bit ((bitmap)data, DECL_UID (t));
|
|
|
/* For DECL_BY_REFERENCE, the return value is actually a pointer. We want
|
/* For DECL_BY_REFERENCE, the return value is actually a pointer. We want
|
to pretend that the value pointed to is actual result decl. */
|
to pretend that the value pointed to is actual result decl. */
|
if ((TREE_CODE (t) == MEM_REF || INDIRECT_REF_P (t))
|
if ((TREE_CODE (t) == MEM_REF || INDIRECT_REF_P (t))
|
&& TREE_CODE (TREE_OPERAND (t, 0)) == SSA_NAME
|
&& TREE_CODE (TREE_OPERAND (t, 0)) == SSA_NAME
|
&& TREE_CODE (SSA_NAME_VAR (TREE_OPERAND (t, 0))) == RESULT_DECL
|
&& TREE_CODE (SSA_NAME_VAR (TREE_OPERAND (t, 0))) == RESULT_DECL
|
&& DECL_BY_REFERENCE (DECL_RESULT (current_function_decl)))
|
&& DECL_BY_REFERENCE (DECL_RESULT (current_function_decl)))
|
return
|
return
|
bitmap_bit_p ((bitmap)data,
|
bitmap_bit_p ((bitmap)data,
|
DECL_UID (DECL_RESULT (current_function_decl)));
|
DECL_UID (DECL_RESULT (current_function_decl)));
|
|
|
return false;
|
return false;
|
}
|
}
|
|
|
/* Compute local properties of basic block BB we collect when looking for
|
/* Compute local properties of basic block BB we collect when looking for
|
split points. We look for ssa defs and store them in SET_SSA_NAMES,
|
split points. We look for ssa defs and store them in SET_SSA_NAMES,
|
for ssa uses and store them in USED_SSA_NAMES and for any non-SSA automatic
|
for ssa uses and store them in USED_SSA_NAMES and for any non-SSA automatic
|
vars stored in NON_SSA_VARS.
|
vars stored in NON_SSA_VARS.
|
|
|
When BB has edge to RETURN_BB, collect uses in RETURN_BB too.
|
When BB has edge to RETURN_BB, collect uses in RETURN_BB too.
|
|
|
Return false when BB contains something that prevents it from being put into
|
Return false when BB contains something that prevents it from being put into
|
split function. */
|
split function. */
|
|
|
static bool
|
static bool
|
visit_bb (basic_block bb, basic_block return_bb,
|
visit_bb (basic_block bb, basic_block return_bb,
|
bitmap set_ssa_names, bitmap used_ssa_names,
|
bitmap set_ssa_names, bitmap used_ssa_names,
|
bitmap non_ssa_vars)
|
bitmap non_ssa_vars)
|
{
|
{
|
gimple_stmt_iterator bsi;
|
gimple_stmt_iterator bsi;
|
edge e;
|
edge e;
|
edge_iterator ei;
|
edge_iterator ei;
|
bool can_split = true;
|
bool can_split = true;
|
|
|
for (bsi = gsi_start_bb (bb); !gsi_end_p (bsi); gsi_next (&bsi))
|
for (bsi = gsi_start_bb (bb); !gsi_end_p (bsi); gsi_next (&bsi))
|
{
|
{
|
gimple stmt = gsi_stmt (bsi);
|
gimple stmt = gsi_stmt (bsi);
|
tree op;
|
tree op;
|
ssa_op_iter iter;
|
ssa_op_iter iter;
|
tree decl;
|
tree decl;
|
|
|
if (is_gimple_debug (stmt))
|
if (is_gimple_debug (stmt))
|
continue;
|
continue;
|
|
|
if (gimple_clobber_p (stmt))
|
if (gimple_clobber_p (stmt))
|
continue;
|
continue;
|
|
|
/* FIXME: We can split regions containing EH. We can not however
|
/* FIXME: We can split regions containing EH. We can not however
|
split RESX, EH_DISPATCH and EH_POINTER referring to same region
|
split RESX, EH_DISPATCH and EH_POINTER referring to same region
|
into different partitions. This would require tracking of
|
into different partitions. This would require tracking of
|
EH regions and checking in consider_split_point if they
|
EH regions and checking in consider_split_point if they
|
are not used elsewhere. */
|
are not used elsewhere. */
|
if (gimple_code (stmt) == GIMPLE_RESX)
|
if (gimple_code (stmt) == GIMPLE_RESX)
|
{
|
{
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
fprintf (dump_file, "Cannot split: resx.\n");
|
fprintf (dump_file, "Cannot split: resx.\n");
|
can_split = false;
|
can_split = false;
|
}
|
}
|
if (gimple_code (stmt) == GIMPLE_EH_DISPATCH)
|
if (gimple_code (stmt) == GIMPLE_EH_DISPATCH)
|
{
|
{
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
fprintf (dump_file, "Cannot split: eh dispatch.\n");
|
fprintf (dump_file, "Cannot split: eh dispatch.\n");
|
can_split = false;
|
can_split = false;
|
}
|
}
|
|
|
/* Check builtins that prevent splitting. */
|
/* Check builtins that prevent splitting. */
|
if (gimple_code (stmt) == GIMPLE_CALL
|
if (gimple_code (stmt) == GIMPLE_CALL
|
&& (decl = gimple_call_fndecl (stmt)) != NULL_TREE
|
&& (decl = gimple_call_fndecl (stmt)) != NULL_TREE
|
&& DECL_BUILT_IN (decl)
|
&& DECL_BUILT_IN (decl)
|
&& DECL_BUILT_IN_CLASS (decl) == BUILT_IN_NORMAL)
|
&& DECL_BUILT_IN_CLASS (decl) == BUILT_IN_NORMAL)
|
switch (DECL_FUNCTION_CODE (decl))
|
switch (DECL_FUNCTION_CODE (decl))
|
{
|
{
|
/* FIXME: once we will allow passing non-parm values to split part,
|
/* FIXME: once we will allow passing non-parm values to split part,
|
we need to be sure to handle correct builtin_stack_save and
|
we need to be sure to handle correct builtin_stack_save and
|
builtin_stack_restore. At the moment we are safe; there is no
|
builtin_stack_restore. At the moment we are safe; there is no
|
way to store builtin_stack_save result in non-SSA variable
|
way to store builtin_stack_save result in non-SSA variable
|
since all calls to those are compiler generated. */
|
since all calls to those are compiler generated. */
|
case BUILT_IN_APPLY:
|
case BUILT_IN_APPLY:
|
case BUILT_IN_APPLY_ARGS:
|
case BUILT_IN_APPLY_ARGS:
|
case BUILT_IN_VA_START:
|
case BUILT_IN_VA_START:
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
fprintf (dump_file,
|
fprintf (dump_file,
|
"Cannot split: builtin_apply and va_start.\n");
|
"Cannot split: builtin_apply and va_start.\n");
|
can_split = false;
|
can_split = false;
|
break;
|
break;
|
case BUILT_IN_EH_POINTER:
|
case BUILT_IN_EH_POINTER:
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
fprintf (dump_file, "Cannot split: builtin_eh_pointer.\n");
|
fprintf (dump_file, "Cannot split: builtin_eh_pointer.\n");
|
can_split = false;
|
can_split = false;
|
break;
|
break;
|
default:
|
default:
|
break;
|
break;
|
}
|
}
|
|
|
FOR_EACH_SSA_TREE_OPERAND (op, stmt, iter, SSA_OP_DEF)
|
FOR_EACH_SSA_TREE_OPERAND (op, stmt, iter, SSA_OP_DEF)
|
bitmap_set_bit (set_ssa_names, SSA_NAME_VERSION (op));
|
bitmap_set_bit (set_ssa_names, SSA_NAME_VERSION (op));
|
FOR_EACH_SSA_TREE_OPERAND (op, stmt, iter, SSA_OP_USE)
|
FOR_EACH_SSA_TREE_OPERAND (op, stmt, iter, SSA_OP_USE)
|
bitmap_set_bit (used_ssa_names, SSA_NAME_VERSION (op));
|
bitmap_set_bit (used_ssa_names, SSA_NAME_VERSION (op));
|
can_split &= !walk_stmt_load_store_addr_ops (stmt, non_ssa_vars,
|
can_split &= !walk_stmt_load_store_addr_ops (stmt, non_ssa_vars,
|
mark_nonssa_use,
|
mark_nonssa_use,
|
mark_nonssa_use,
|
mark_nonssa_use,
|
mark_nonssa_use);
|
mark_nonssa_use);
|
}
|
}
|
for (bsi = gsi_start_phis (bb); !gsi_end_p (bsi); gsi_next (&bsi))
|
for (bsi = gsi_start_phis (bb); !gsi_end_p (bsi); gsi_next (&bsi))
|
{
|
{
|
gimple stmt = gsi_stmt (bsi);
|
gimple stmt = gsi_stmt (bsi);
|
unsigned int i;
|
unsigned int i;
|
|
|
if (is_gimple_debug (stmt))
|
if (is_gimple_debug (stmt))
|
continue;
|
continue;
|
if (!is_gimple_reg (gimple_phi_result (stmt)))
|
if (!is_gimple_reg (gimple_phi_result (stmt)))
|
continue;
|
continue;
|
bitmap_set_bit (set_ssa_names,
|
bitmap_set_bit (set_ssa_names,
|
SSA_NAME_VERSION (gimple_phi_result (stmt)));
|
SSA_NAME_VERSION (gimple_phi_result (stmt)));
|
for (i = 0; i < gimple_phi_num_args (stmt); i++)
|
for (i = 0; i < gimple_phi_num_args (stmt); i++)
|
{
|
{
|
tree op = gimple_phi_arg_def (stmt, i);
|
tree op = gimple_phi_arg_def (stmt, i);
|
if (TREE_CODE (op) == SSA_NAME)
|
if (TREE_CODE (op) == SSA_NAME)
|
bitmap_set_bit (used_ssa_names, SSA_NAME_VERSION (op));
|
bitmap_set_bit (used_ssa_names, SSA_NAME_VERSION (op));
|
}
|
}
|
can_split &= !walk_stmt_load_store_addr_ops (stmt, non_ssa_vars,
|
can_split &= !walk_stmt_load_store_addr_ops (stmt, non_ssa_vars,
|
mark_nonssa_use,
|
mark_nonssa_use,
|
mark_nonssa_use,
|
mark_nonssa_use,
|
mark_nonssa_use);
|
mark_nonssa_use);
|
}
|
}
|
/* Record also uses coming from PHI operand in return BB. */
|
/* Record also uses coming from PHI operand in return BB. */
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
FOR_EACH_EDGE (e, ei, bb->succs)
|
if (e->dest == return_bb)
|
if (e->dest == return_bb)
|
{
|
{
|
for (bsi = gsi_start_phis (return_bb); !gsi_end_p (bsi); gsi_next (&bsi))
|
for (bsi = gsi_start_phis (return_bb); !gsi_end_p (bsi); gsi_next (&bsi))
|
{
|
{
|
gimple stmt = gsi_stmt (bsi);
|
gimple stmt = gsi_stmt (bsi);
|
tree op = gimple_phi_arg_def (stmt, e->dest_idx);
|
tree op = gimple_phi_arg_def (stmt, e->dest_idx);
|
|
|
if (is_gimple_debug (stmt))
|
if (is_gimple_debug (stmt))
|
continue;
|
continue;
|
if (!is_gimple_reg (gimple_phi_result (stmt)))
|
if (!is_gimple_reg (gimple_phi_result (stmt)))
|
continue;
|
continue;
|
if (TREE_CODE (op) == SSA_NAME)
|
if (TREE_CODE (op) == SSA_NAME)
|
bitmap_set_bit (used_ssa_names, SSA_NAME_VERSION (op));
|
bitmap_set_bit (used_ssa_names, SSA_NAME_VERSION (op));
|
else
|
else
|
can_split &= !mark_nonssa_use (stmt, op, non_ssa_vars);
|
can_split &= !mark_nonssa_use (stmt, op, non_ssa_vars);
|
}
|
}
|
}
|
}
|
return can_split;
|
return can_split;
|
}
|
}
|
|
|
/* Stack entry for recursive DFS walk in find_split_point. */
|
/* Stack entry for recursive DFS walk in find_split_point. */
|
|
|
typedef struct
|
typedef struct
|
{
|
{
|
/* Basic block we are examining. */
|
/* Basic block we are examining. */
|
basic_block bb;
|
basic_block bb;
|
|
|
/* SSA names set and used by the BB and all BBs reachable
|
/* SSA names set and used by the BB and all BBs reachable
|
from it via DFS walk. */
|
from it via DFS walk. */
|
bitmap set_ssa_names, used_ssa_names;
|
bitmap set_ssa_names, used_ssa_names;
|
bitmap non_ssa_vars;
|
bitmap non_ssa_vars;
|
|
|
/* All BBS visited from this BB via DFS walk. */
|
/* All BBS visited from this BB via DFS walk. */
|
bitmap bbs_visited;
|
bitmap bbs_visited;
|
|
|
/* Last examined edge in DFS walk. Since we walk unoriented graph,
|
/* Last examined edge in DFS walk. Since we walk unoriented graph,
|
the value is up to sum of incoming and outgoing edges of BB. */
|
the value is up to sum of incoming and outgoing edges of BB. */
|
unsigned int edge_num;
|
unsigned int edge_num;
|
|
|
/* Stack entry index of earliest BB reachable from current BB
|
/* Stack entry index of earliest BB reachable from current BB
|
or any BB visited later in DFS walk. */
|
or any BB visited later in DFS walk. */
|
int earliest;
|
int earliest;
|
|
|
/* Overall time and size of all BBs reached from this BB in DFS walk. */
|
/* Overall time and size of all BBs reached from this BB in DFS walk. */
|
int overall_time, overall_size;
|
int overall_time, overall_size;
|
|
|
/* When false we can not split on this BB. */
|
/* When false we can not split on this BB. */
|
bool can_split;
|
bool can_split;
|
} stack_entry;
|
} stack_entry;
|
DEF_VEC_O(stack_entry);
|
DEF_VEC_O(stack_entry);
|
DEF_VEC_ALLOC_O(stack_entry,heap);
|
DEF_VEC_ALLOC_O(stack_entry,heap);
|
|
|
|
|
/* Find all articulations and call consider_split on them.
|
/* Find all articulations and call consider_split on them.
|
OVERALL_TIME and OVERALL_SIZE is time and size of the function.
|
OVERALL_TIME and OVERALL_SIZE is time and size of the function.
|
|
|
We perform basic algorithm for finding an articulation in a graph
|
We perform basic algorithm for finding an articulation in a graph
|
created from CFG by considering it to be an unoriented graph.
|
created from CFG by considering it to be an unoriented graph.
|
|
|
The articulation is discovered via DFS walk. We collect earliest
|
The articulation is discovered via DFS walk. We collect earliest
|
basic block on stack that is reachable via backward edge. Articulation
|
basic block on stack that is reachable via backward edge. Articulation
|
is any basic block such that there is no backward edge bypassing it.
|
is any basic block such that there is no backward edge bypassing it.
|
To reduce stack usage we maintain heap allocated stack in STACK vector.
|
To reduce stack usage we maintain heap allocated stack in STACK vector.
|
AUX pointer of BB is set to index it appears in the stack or -1 once
|
AUX pointer of BB is set to index it appears in the stack or -1 once
|
it is visited and popped off the stack.
|
it is visited and popped off the stack.
|
|
|
The algorithm finds articulation after visiting the whole component
|
The algorithm finds articulation after visiting the whole component
|
reachable by it. This makes it convenient to collect information about
|
reachable by it. This makes it convenient to collect information about
|
the component used by consider_split. */
|
the component used by consider_split. */
|
|
|
static void
|
static void
|
find_split_points (int overall_time, int overall_size)
|
find_split_points (int overall_time, int overall_size)
|
{
|
{
|
stack_entry first;
|
stack_entry first;
|
VEC(stack_entry, heap) *stack = NULL;
|
VEC(stack_entry, heap) *stack = NULL;
|
basic_block bb;
|
basic_block bb;
|
basic_block return_bb = find_return_bb ();
|
basic_block return_bb = find_return_bb ();
|
struct split_point current;
|
struct split_point current;
|
|
|
current.header_time = overall_time;
|
current.header_time = overall_time;
|
current.header_size = overall_size;
|
current.header_size = overall_size;
|
current.split_time = 0;
|
current.split_time = 0;
|
current.split_size = 0;
|
current.split_size = 0;
|
current.ssa_names_to_pass = BITMAP_ALLOC (NULL);
|
current.ssa_names_to_pass = BITMAP_ALLOC (NULL);
|
|
|
first.bb = ENTRY_BLOCK_PTR;
|
first.bb = ENTRY_BLOCK_PTR;
|
first.edge_num = 0;
|
first.edge_num = 0;
|
first.overall_time = 0;
|
first.overall_time = 0;
|
first.overall_size = 0;
|
first.overall_size = 0;
|
first.earliest = INT_MAX;
|
first.earliest = INT_MAX;
|
first.set_ssa_names = 0;
|
first.set_ssa_names = 0;
|
first.used_ssa_names = 0;
|
first.used_ssa_names = 0;
|
first.bbs_visited = 0;
|
first.bbs_visited = 0;
|
VEC_safe_push (stack_entry, heap, stack, &first);
|
VEC_safe_push (stack_entry, heap, stack, &first);
|
ENTRY_BLOCK_PTR->aux = (void *)(intptr_t)-1;
|
ENTRY_BLOCK_PTR->aux = (void *)(intptr_t)-1;
|
|
|
while (!VEC_empty (stack_entry, stack))
|
while (!VEC_empty (stack_entry, stack))
|
{
|
{
|
stack_entry *entry = VEC_last (stack_entry, stack);
|
stack_entry *entry = VEC_last (stack_entry, stack);
|
|
|
/* We are walking an acyclic graph, so edge_num counts
|
/* We are walking an acyclic graph, so edge_num counts
|
succ and pred edges together. However when considering
|
succ and pred edges together. However when considering
|
articulation, we want to have processed everything reachable
|
articulation, we want to have processed everything reachable
|
from articulation but nothing that reaches into it. */
|
from articulation but nothing that reaches into it. */
|
if (entry->edge_num == EDGE_COUNT (entry->bb->succs)
|
if (entry->edge_num == EDGE_COUNT (entry->bb->succs)
|
&& entry->bb != ENTRY_BLOCK_PTR)
|
&& entry->bb != ENTRY_BLOCK_PTR)
|
{
|
{
|
int pos = VEC_length (stack_entry, stack);
|
int pos = VEC_length (stack_entry, stack);
|
entry->can_split &= visit_bb (entry->bb, return_bb,
|
entry->can_split &= visit_bb (entry->bb, return_bb,
|
entry->set_ssa_names,
|
entry->set_ssa_names,
|
entry->used_ssa_names,
|
entry->used_ssa_names,
|
entry->non_ssa_vars);
|
entry->non_ssa_vars);
|
if (pos <= entry->earliest && !entry->can_split
|
if (pos <= entry->earliest && !entry->can_split
|
&& dump_file && (dump_flags & TDF_DETAILS))
|
&& dump_file && (dump_flags & TDF_DETAILS))
|
fprintf (dump_file,
|
fprintf (dump_file,
|
"found articulation at bb %i but can not split\n",
|
"found articulation at bb %i but can not split\n",
|
entry->bb->index);
|
entry->bb->index);
|
if (pos <= entry->earliest && entry->can_split)
|
if (pos <= entry->earliest && entry->can_split)
|
{
|
{
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
fprintf (dump_file, "found articulation at bb %i\n",
|
fprintf (dump_file, "found articulation at bb %i\n",
|
entry->bb->index);
|
entry->bb->index);
|
current.entry_bb = entry->bb;
|
current.entry_bb = entry->bb;
|
current.ssa_names_to_pass = BITMAP_ALLOC (NULL);
|
current.ssa_names_to_pass = BITMAP_ALLOC (NULL);
|
bitmap_and_compl (current.ssa_names_to_pass,
|
bitmap_and_compl (current.ssa_names_to_pass,
|
entry->used_ssa_names, entry->set_ssa_names);
|
entry->used_ssa_names, entry->set_ssa_names);
|
current.header_time = overall_time - entry->overall_time;
|
current.header_time = overall_time - entry->overall_time;
|
current.header_size = overall_size - entry->overall_size;
|
current.header_size = overall_size - entry->overall_size;
|
current.split_time = entry->overall_time;
|
current.split_time = entry->overall_time;
|
current.split_size = entry->overall_size;
|
current.split_size = entry->overall_size;
|
current.split_bbs = entry->bbs_visited;
|
current.split_bbs = entry->bbs_visited;
|
consider_split (¤t, entry->non_ssa_vars, return_bb);
|
consider_split (¤t, entry->non_ssa_vars, return_bb);
|
BITMAP_FREE (current.ssa_names_to_pass);
|
BITMAP_FREE (current.ssa_names_to_pass);
|
}
|
}
|
}
|
}
|
/* Do actual DFS walk. */
|
/* Do actual DFS walk. */
|
if (entry->edge_num
|
if (entry->edge_num
|
< (EDGE_COUNT (entry->bb->succs)
|
< (EDGE_COUNT (entry->bb->succs)
|
+ EDGE_COUNT (entry->bb->preds)))
|
+ EDGE_COUNT (entry->bb->preds)))
|
{
|
{
|
edge e;
|
edge e;
|
basic_block dest;
|
basic_block dest;
|
if (entry->edge_num < EDGE_COUNT (entry->bb->succs))
|
if (entry->edge_num < EDGE_COUNT (entry->bb->succs))
|
{
|
{
|
e = EDGE_SUCC (entry->bb, entry->edge_num);
|
e = EDGE_SUCC (entry->bb, entry->edge_num);
|
dest = e->dest;
|
dest = e->dest;
|
}
|
}
|
else
|
else
|
{
|
{
|
e = EDGE_PRED (entry->bb, entry->edge_num
|
e = EDGE_PRED (entry->bb, entry->edge_num
|
- EDGE_COUNT (entry->bb->succs));
|
- EDGE_COUNT (entry->bb->succs));
|
dest = e->src;
|
dest = e->src;
|
}
|
}
|
|
|
entry->edge_num++;
|
entry->edge_num++;
|
|
|
/* New BB to visit, push it to the stack. */
|
/* New BB to visit, push it to the stack. */
|
if (dest != return_bb && dest != EXIT_BLOCK_PTR
|
if (dest != return_bb && dest != EXIT_BLOCK_PTR
|
&& !dest->aux)
|
&& !dest->aux)
|
{
|
{
|
stack_entry new_entry;
|
stack_entry new_entry;
|
|
|
new_entry.bb = dest;
|
new_entry.bb = dest;
|
new_entry.edge_num = 0;
|
new_entry.edge_num = 0;
|
new_entry.overall_time
|
new_entry.overall_time
|
= VEC_index (bb_info, bb_info_vec, dest->index)->time;
|
= VEC_index (bb_info, bb_info_vec, dest->index)->time;
|
new_entry.overall_size
|
new_entry.overall_size
|
= VEC_index (bb_info, bb_info_vec, dest->index)->size;
|
= VEC_index (bb_info, bb_info_vec, dest->index)->size;
|
new_entry.earliest = INT_MAX;
|
new_entry.earliest = INT_MAX;
|
new_entry.set_ssa_names = BITMAP_ALLOC (NULL);
|
new_entry.set_ssa_names = BITMAP_ALLOC (NULL);
|
new_entry.used_ssa_names = BITMAP_ALLOC (NULL);
|
new_entry.used_ssa_names = BITMAP_ALLOC (NULL);
|
new_entry.bbs_visited = BITMAP_ALLOC (NULL);
|
new_entry.bbs_visited = BITMAP_ALLOC (NULL);
|
new_entry.non_ssa_vars = BITMAP_ALLOC (NULL);
|
new_entry.non_ssa_vars = BITMAP_ALLOC (NULL);
|
new_entry.can_split = true;
|
new_entry.can_split = true;
|
bitmap_set_bit (new_entry.bbs_visited, dest->index);
|
bitmap_set_bit (new_entry.bbs_visited, dest->index);
|
VEC_safe_push (stack_entry, heap, stack, &new_entry);
|
VEC_safe_push (stack_entry, heap, stack, &new_entry);
|
dest->aux = (void *)(intptr_t)VEC_length (stack_entry, stack);
|
dest->aux = (void *)(intptr_t)VEC_length (stack_entry, stack);
|
}
|
}
|
/* Back edge found, record the earliest point. */
|
/* Back edge found, record the earliest point. */
|
else if ((intptr_t)dest->aux > 0
|
else if ((intptr_t)dest->aux > 0
|
&& (intptr_t)dest->aux < entry->earliest)
|
&& (intptr_t)dest->aux < entry->earliest)
|
entry->earliest = (intptr_t)dest->aux;
|
entry->earliest = (intptr_t)dest->aux;
|
}
|
}
|
/* We are done with examining the edges. Pop off the value from stack
|
/* We are done with examining the edges. Pop off the value from stack
|
and merge stuff we accumulate during the walk. */
|
and merge stuff we accumulate during the walk. */
|
else if (entry->bb != ENTRY_BLOCK_PTR)
|
else if (entry->bb != ENTRY_BLOCK_PTR)
|
{
|
{
|
stack_entry *prev = VEC_index (stack_entry, stack,
|
stack_entry *prev = VEC_index (stack_entry, stack,
|
VEC_length (stack_entry, stack) - 2);
|
VEC_length (stack_entry, stack) - 2);
|
|
|
entry->bb->aux = (void *)(intptr_t)-1;
|
entry->bb->aux = (void *)(intptr_t)-1;
|
prev->can_split &= entry->can_split;
|
prev->can_split &= entry->can_split;
|
if (prev->set_ssa_names)
|
if (prev->set_ssa_names)
|
{
|
{
|
bitmap_ior_into (prev->set_ssa_names, entry->set_ssa_names);
|
bitmap_ior_into (prev->set_ssa_names, entry->set_ssa_names);
|
bitmap_ior_into (prev->used_ssa_names, entry->used_ssa_names);
|
bitmap_ior_into (prev->used_ssa_names, entry->used_ssa_names);
|
bitmap_ior_into (prev->bbs_visited, entry->bbs_visited);
|
bitmap_ior_into (prev->bbs_visited, entry->bbs_visited);
|
bitmap_ior_into (prev->non_ssa_vars, entry->non_ssa_vars);
|
bitmap_ior_into (prev->non_ssa_vars, entry->non_ssa_vars);
|
}
|
}
|
if (prev->earliest > entry->earliest)
|
if (prev->earliest > entry->earliest)
|
prev->earliest = entry->earliest;
|
prev->earliest = entry->earliest;
|
prev->overall_time += entry->overall_time;
|
prev->overall_time += entry->overall_time;
|
prev->overall_size += entry->overall_size;
|
prev->overall_size += entry->overall_size;
|
BITMAP_FREE (entry->set_ssa_names);
|
BITMAP_FREE (entry->set_ssa_names);
|
BITMAP_FREE (entry->used_ssa_names);
|
BITMAP_FREE (entry->used_ssa_names);
|
BITMAP_FREE (entry->bbs_visited);
|
BITMAP_FREE (entry->bbs_visited);
|
BITMAP_FREE (entry->non_ssa_vars);
|
BITMAP_FREE (entry->non_ssa_vars);
|
VEC_pop (stack_entry, stack);
|
VEC_pop (stack_entry, stack);
|
}
|
}
|
else
|
else
|
VEC_pop (stack_entry, stack);
|
VEC_pop (stack_entry, stack);
|
}
|
}
|
ENTRY_BLOCK_PTR->aux = NULL;
|
ENTRY_BLOCK_PTR->aux = NULL;
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
bb->aux = NULL;
|
bb->aux = NULL;
|
VEC_free (stack_entry, heap, stack);
|
VEC_free (stack_entry, heap, stack);
|
BITMAP_FREE (current.ssa_names_to_pass);
|
BITMAP_FREE (current.ssa_names_to_pass);
|
}
|
}
|
|
|
/* Split function at SPLIT_POINT. */
|
/* Split function at SPLIT_POINT. */
|
|
|
static void
|
static void
|
split_function (struct split_point *split_point)
|
split_function (struct split_point *split_point)
|
{
|
{
|
VEC (tree, heap) *args_to_pass = NULL;
|
VEC (tree, heap) *args_to_pass = NULL;
|
bitmap args_to_skip;
|
bitmap args_to_skip;
|
tree parm;
|
tree parm;
|
int num = 0;
|
int num = 0;
|
struct cgraph_node *node, *cur_node = cgraph_get_node (current_function_decl);
|
struct cgraph_node *node, *cur_node = cgraph_get_node (current_function_decl);
|
basic_block return_bb = find_return_bb ();
|
basic_block return_bb = find_return_bb ();
|
basic_block call_bb;
|
basic_block call_bb;
|
gimple_stmt_iterator gsi;
|
gimple_stmt_iterator gsi;
|
gimple call;
|
gimple call;
|
edge e;
|
edge e;
|
edge_iterator ei;
|
edge_iterator ei;
|
tree retval = NULL, real_retval = NULL;
|
tree retval = NULL, real_retval = NULL;
|
bool split_part_return_p = false;
|
bool split_part_return_p = false;
|
gimple last_stmt = NULL;
|
gimple last_stmt = NULL;
|
unsigned int i;
|
unsigned int i;
|
tree arg;
|
tree arg;
|
|
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file, "\n\nSplitting function at:\n");
|
fprintf (dump_file, "\n\nSplitting function at:\n");
|
dump_split_point (dump_file, split_point);
|
dump_split_point (dump_file, split_point);
|
}
|
}
|
|
|
if (cur_node->local.can_change_signature)
|
if (cur_node->local.can_change_signature)
|
args_to_skip = BITMAP_ALLOC (NULL);
|
args_to_skip = BITMAP_ALLOC (NULL);
|
else
|
else
|
args_to_skip = NULL;
|
args_to_skip = NULL;
|
|
|
/* Collect the parameters of new function and args_to_skip bitmap. */
|
/* Collect the parameters of new function and args_to_skip bitmap. */
|
for (parm = DECL_ARGUMENTS (current_function_decl);
|
for (parm = DECL_ARGUMENTS (current_function_decl);
|
parm; parm = DECL_CHAIN (parm), num++)
|
parm; parm = DECL_CHAIN (parm), num++)
|
if (args_to_skip
|
if (args_to_skip
|
&& (!is_gimple_reg (parm)
|
&& (!is_gimple_reg (parm)
|
|| !gimple_default_def (cfun, parm)
|
|| !gimple_default_def (cfun, parm)
|
|| !bitmap_bit_p (split_point->ssa_names_to_pass,
|
|| !bitmap_bit_p (split_point->ssa_names_to_pass,
|
SSA_NAME_VERSION (gimple_default_def (cfun,
|
SSA_NAME_VERSION (gimple_default_def (cfun,
|
parm)))))
|
parm)))))
|
bitmap_set_bit (args_to_skip, num);
|
bitmap_set_bit (args_to_skip, num);
|
else
|
else
|
{
|
{
|
/* This parm might not have been used up to now, but is going to be
|
/* This parm might not have been used up to now, but is going to be
|
used, hence register it. */
|
used, hence register it. */
|
add_referenced_var (parm);
|
add_referenced_var (parm);
|
if (is_gimple_reg (parm))
|
if (is_gimple_reg (parm))
|
{
|
{
|
arg = gimple_default_def (cfun, parm);
|
arg = gimple_default_def (cfun, parm);
|
if (!arg)
|
if (!arg)
|
{
|
{
|
arg = make_ssa_name (parm, gimple_build_nop ());
|
arg = make_ssa_name (parm, gimple_build_nop ());
|
set_default_def (parm, arg);
|
set_default_def (parm, arg);
|
}
|
}
|
}
|
}
|
else
|
else
|
arg = parm;
|
arg = parm;
|
|
|
if (!useless_type_conversion_p (DECL_ARG_TYPE (parm), TREE_TYPE (arg)))
|
if (!useless_type_conversion_p (DECL_ARG_TYPE (parm), TREE_TYPE (arg)))
|
arg = fold_convert (DECL_ARG_TYPE (parm), arg);
|
arg = fold_convert (DECL_ARG_TYPE (parm), arg);
|
VEC_safe_push (tree, heap, args_to_pass, arg);
|
VEC_safe_push (tree, heap, args_to_pass, arg);
|
}
|
}
|
|
|
/* See if the split function will return. */
|
/* See if the split function will return. */
|
FOR_EACH_EDGE (e, ei, return_bb->preds)
|
FOR_EACH_EDGE (e, ei, return_bb->preds)
|
if (bitmap_bit_p (split_point->split_bbs, e->src->index))
|
if (bitmap_bit_p (split_point->split_bbs, e->src->index))
|
break;
|
break;
|
if (e)
|
if (e)
|
split_part_return_p = true;
|
split_part_return_p = true;
|
|
|
/* Add return block to what will become the split function.
|
/* Add return block to what will become the split function.
|
We do not return; no return block is needed. */
|
We do not return; no return block is needed. */
|
if (!split_part_return_p)
|
if (!split_part_return_p)
|
;
|
;
|
/* We have no return block, so nothing is needed. */
|
/* We have no return block, so nothing is needed. */
|
else if (return_bb == EXIT_BLOCK_PTR)
|
else if (return_bb == EXIT_BLOCK_PTR)
|
;
|
;
|
/* When we do not want to return value, we need to construct
|
/* When we do not want to return value, we need to construct
|
new return block with empty return statement.
|
new return block with empty return statement.
|
FIXME: Once we are able to change return type, we should change function
|
FIXME: Once we are able to change return type, we should change function
|
to return void instead of just outputting function with undefined return
|
to return void instead of just outputting function with undefined return
|
value. For structures this affects quality of codegen. */
|
value. For structures this affects quality of codegen. */
|
else if (!split_point->split_part_set_retval
|
else if (!split_point->split_part_set_retval
|
&& find_retval (return_bb))
|
&& find_retval (return_bb))
|
{
|
{
|
bool redirected = true;
|
bool redirected = true;
|
basic_block new_return_bb = create_basic_block (NULL, 0, return_bb);
|
basic_block new_return_bb = create_basic_block (NULL, 0, return_bb);
|
gimple_stmt_iterator gsi = gsi_start_bb (new_return_bb);
|
gimple_stmt_iterator gsi = gsi_start_bb (new_return_bb);
|
gsi_insert_after (&gsi, gimple_build_return (NULL), GSI_NEW_STMT);
|
gsi_insert_after (&gsi, gimple_build_return (NULL), GSI_NEW_STMT);
|
while (redirected)
|
while (redirected)
|
{
|
{
|
redirected = false;
|
redirected = false;
|
FOR_EACH_EDGE (e, ei, return_bb->preds)
|
FOR_EACH_EDGE (e, ei, return_bb->preds)
|
if (bitmap_bit_p (split_point->split_bbs, e->src->index))
|
if (bitmap_bit_p (split_point->split_bbs, e->src->index))
|
{
|
{
|
new_return_bb->count += e->count;
|
new_return_bb->count += e->count;
|
new_return_bb->frequency += EDGE_FREQUENCY (e);
|
new_return_bb->frequency += EDGE_FREQUENCY (e);
|
redirect_edge_and_branch (e, new_return_bb);
|
redirect_edge_and_branch (e, new_return_bb);
|
redirected = true;
|
redirected = true;
|
break;
|
break;
|
}
|
}
|
}
|
}
|
e = make_edge (new_return_bb, EXIT_BLOCK_PTR, 0);
|
e = make_edge (new_return_bb, EXIT_BLOCK_PTR, 0);
|
e->probability = REG_BR_PROB_BASE;
|
e->probability = REG_BR_PROB_BASE;
|
e->count = new_return_bb->count;
|
e->count = new_return_bb->count;
|
bitmap_set_bit (split_point->split_bbs, new_return_bb->index);
|
bitmap_set_bit (split_point->split_bbs, new_return_bb->index);
|
}
|
}
|
/* When we pass around the value, use existing return block. */
|
/* When we pass around the value, use existing return block. */
|
else
|
else
|
bitmap_set_bit (split_point->split_bbs, return_bb->index);
|
bitmap_set_bit (split_point->split_bbs, return_bb->index);
|
|
|
/* If RETURN_BB has virtual operand PHIs, they must be removed and the
|
/* If RETURN_BB has virtual operand PHIs, they must be removed and the
|
virtual operand marked for renaming as we change the CFG in a way that
|
virtual operand marked for renaming as we change the CFG in a way that
|
tree-inline is not able to compensate for.
|
tree-inline is not able to compensate for.
|
|
|
Note this can happen whether or not we have a return value. If we have
|
Note this can happen whether or not we have a return value. If we have
|
a return value, then RETURN_BB may have PHIs for real operands too. */
|
a return value, then RETURN_BB may have PHIs for real operands too. */
|
if (return_bb != EXIT_BLOCK_PTR)
|
if (return_bb != EXIT_BLOCK_PTR)
|
{
|
{
|
bool phi_p = false;
|
bool phi_p = false;
|
for (gsi = gsi_start_phis (return_bb); !gsi_end_p (gsi);)
|
for (gsi = gsi_start_phis (return_bb); !gsi_end_p (gsi);)
|
{
|
{
|
gimple stmt = gsi_stmt (gsi);
|
gimple stmt = gsi_stmt (gsi);
|
if (is_gimple_reg (gimple_phi_result (stmt)))
|
if (is_gimple_reg (gimple_phi_result (stmt)))
|
{
|
{
|
gsi_next (&gsi);
|
gsi_next (&gsi);
|
continue;
|
continue;
|
}
|
}
|
mark_virtual_phi_result_for_renaming (stmt);
|
mark_virtual_phi_result_for_renaming (stmt);
|
remove_phi_node (&gsi, true);
|
remove_phi_node (&gsi, true);
|
phi_p = true;
|
phi_p = true;
|
}
|
}
|
/* In reality we have to rename the reaching definition of the
|
/* In reality we have to rename the reaching definition of the
|
virtual operand at return_bb as we will eventually release it
|
virtual operand at return_bb as we will eventually release it
|
when we remove the code region we outlined.
|
when we remove the code region we outlined.
|
So we have to rename all immediate virtual uses of that region
|
So we have to rename all immediate virtual uses of that region
|
if we didn't see a PHI definition yet. */
|
if we didn't see a PHI definition yet. */
|
/* ??? In real reality we want to set the reaching vdef of the
|
/* ??? In real reality we want to set the reaching vdef of the
|
entry of the SESE region as the vuse of the call and the reaching
|
entry of the SESE region as the vuse of the call and the reaching
|
vdef of the exit of the SESE region as the vdef of the call. */
|
vdef of the exit of the SESE region as the vdef of the call. */
|
if (!phi_p)
|
if (!phi_p)
|
for (gsi = gsi_start_bb (return_bb); !gsi_end_p (gsi); gsi_next (&gsi))
|
for (gsi = gsi_start_bb (return_bb); !gsi_end_p (gsi); gsi_next (&gsi))
|
{
|
{
|
gimple stmt = gsi_stmt (gsi);
|
gimple stmt = gsi_stmt (gsi);
|
if (gimple_vuse (stmt))
|
if (gimple_vuse (stmt))
|
{
|
{
|
gimple_set_vuse (stmt, NULL_TREE);
|
gimple_set_vuse (stmt, NULL_TREE);
|
update_stmt (stmt);
|
update_stmt (stmt);
|
}
|
}
|
if (gimple_vdef (stmt))
|
if (gimple_vdef (stmt))
|
break;
|
break;
|
}
|
}
|
}
|
}
|
|
|
/* Now create the actual clone. */
|
/* Now create the actual clone. */
|
rebuild_cgraph_edges ();
|
rebuild_cgraph_edges ();
|
node = cgraph_function_versioning (cur_node, NULL, NULL, args_to_skip,
|
node = cgraph_function_versioning (cur_node, NULL, NULL, args_to_skip,
|
!split_part_return_p,
|
!split_part_return_p,
|
split_point->split_bbs,
|
split_point->split_bbs,
|
split_point->entry_bb, "part");
|
split_point->entry_bb, "part");
|
/* For usual cloning it is enough to clear builtin only when signature
|
/* For usual cloning it is enough to clear builtin only when signature
|
changes. For partial inlining we however can not expect the part
|
changes. For partial inlining we however can not expect the part
|
of builtin implementation to have same semantic as the whole. */
|
of builtin implementation to have same semantic as the whole. */
|
if (DECL_BUILT_IN (node->decl))
|
if (DECL_BUILT_IN (node->decl))
|
{
|
{
|
DECL_BUILT_IN_CLASS (node->decl) = NOT_BUILT_IN;
|
DECL_BUILT_IN_CLASS (node->decl) = NOT_BUILT_IN;
|
DECL_FUNCTION_CODE (node->decl) = (enum built_in_function) 0;
|
DECL_FUNCTION_CODE (node->decl) = (enum built_in_function) 0;
|
}
|
}
|
cgraph_node_remove_callees (cur_node);
|
cgraph_node_remove_callees (cur_node);
|
if (!split_part_return_p)
|
if (!split_part_return_p)
|
TREE_THIS_VOLATILE (node->decl) = 1;
|
TREE_THIS_VOLATILE (node->decl) = 1;
|
if (dump_file)
|
if (dump_file)
|
dump_function_to_file (node->decl, dump_file, dump_flags);
|
dump_function_to_file (node->decl, dump_file, dump_flags);
|
|
|
/* Create the basic block we place call into. It is the entry basic block
|
/* Create the basic block we place call into. It is the entry basic block
|
split after last label. */
|
split after last label. */
|
call_bb = split_point->entry_bb;
|
call_bb = split_point->entry_bb;
|
for (gsi = gsi_start_bb (call_bb); !gsi_end_p (gsi);)
|
for (gsi = gsi_start_bb (call_bb); !gsi_end_p (gsi);)
|
if (gimple_code (gsi_stmt (gsi)) == GIMPLE_LABEL)
|
if (gimple_code (gsi_stmt (gsi)) == GIMPLE_LABEL)
|
{
|
{
|
last_stmt = gsi_stmt (gsi);
|
last_stmt = gsi_stmt (gsi);
|
gsi_next (&gsi);
|
gsi_next (&gsi);
|
}
|
}
|
else
|
else
|
break;
|
break;
|
e = split_block (split_point->entry_bb, last_stmt);
|
e = split_block (split_point->entry_bb, last_stmt);
|
remove_edge (e);
|
remove_edge (e);
|
|
|
/* Produce the call statement. */
|
/* Produce the call statement. */
|
gsi = gsi_last_bb (call_bb);
|
gsi = gsi_last_bb (call_bb);
|
FOR_EACH_VEC_ELT (tree, args_to_pass, i, arg)
|
FOR_EACH_VEC_ELT (tree, args_to_pass, i, arg)
|
if (!is_gimple_val (arg))
|
if (!is_gimple_val (arg))
|
{
|
{
|
arg = force_gimple_operand_gsi (&gsi, arg, true, NULL_TREE,
|
arg = force_gimple_operand_gsi (&gsi, arg, true, NULL_TREE,
|
false, GSI_CONTINUE_LINKING);
|
false, GSI_CONTINUE_LINKING);
|
VEC_replace (tree, args_to_pass, i, arg);
|
VEC_replace (tree, args_to_pass, i, arg);
|
}
|
}
|
call = gimple_build_call_vec (node->decl, args_to_pass);
|
call = gimple_build_call_vec (node->decl, args_to_pass);
|
gimple_set_block (call, DECL_INITIAL (current_function_decl));
|
gimple_set_block (call, DECL_INITIAL (current_function_decl));
|
|
|
/* We avoid address being taken on any variable used by split part,
|
/* We avoid address being taken on any variable used by split part,
|
so return slot optimization is always possible. Moreover this is
|
so return slot optimization is always possible. Moreover this is
|
required to make DECL_BY_REFERENCE work. */
|
required to make DECL_BY_REFERENCE work. */
|
if (aggregate_value_p (DECL_RESULT (current_function_decl),
|
if (aggregate_value_p (DECL_RESULT (current_function_decl),
|
TREE_TYPE (current_function_decl)))
|
TREE_TYPE (current_function_decl)))
|
gimple_call_set_return_slot_opt (call, true);
|
gimple_call_set_return_slot_opt (call, true);
|
|
|
/* Update return value. This is bit tricky. When we do not return,
|
/* Update return value. This is bit tricky. When we do not return,
|
do nothing. When we return we might need to update return_bb
|
do nothing. When we return we might need to update return_bb
|
or produce a new return statement. */
|
or produce a new return statement. */
|
if (!split_part_return_p)
|
if (!split_part_return_p)
|
gsi_insert_after (&gsi, call, GSI_NEW_STMT);
|
gsi_insert_after (&gsi, call, GSI_NEW_STMT);
|
else
|
else
|
{
|
{
|
e = make_edge (call_bb, return_bb,
|
e = make_edge (call_bb, return_bb,
|
return_bb == EXIT_BLOCK_PTR ? 0 : EDGE_FALLTHRU);
|
return_bb == EXIT_BLOCK_PTR ? 0 : EDGE_FALLTHRU);
|
e->count = call_bb->count;
|
e->count = call_bb->count;
|
e->probability = REG_BR_PROB_BASE;
|
e->probability = REG_BR_PROB_BASE;
|
|
|
/* If there is return basic block, see what value we need to store
|
/* If there is return basic block, see what value we need to store
|
return value into and put call just before it. */
|
return value into and put call just before it. */
|
if (return_bb != EXIT_BLOCK_PTR)
|
if (return_bb != EXIT_BLOCK_PTR)
|
{
|
{
|
real_retval = retval = find_retval (return_bb);
|
real_retval = retval = find_retval (return_bb);
|
|
|
if (real_retval && split_point->split_part_set_retval)
|
if (real_retval && split_point->split_part_set_retval)
|
{
|
{
|
gimple_stmt_iterator psi;
|
gimple_stmt_iterator psi;
|
|
|
/* See if we need new SSA_NAME for the result.
|
/* See if we need new SSA_NAME for the result.
|
When DECL_BY_REFERENCE is true, retval is actually pointer to
|
When DECL_BY_REFERENCE is true, retval is actually pointer to
|
return value and it is constant in whole function. */
|
return value and it is constant in whole function. */
|
if (TREE_CODE (retval) == SSA_NAME
|
if (TREE_CODE (retval) == SSA_NAME
|
&& !DECL_BY_REFERENCE (DECL_RESULT (current_function_decl)))
|
&& !DECL_BY_REFERENCE (DECL_RESULT (current_function_decl)))
|
{
|
{
|
retval = make_ssa_name (SSA_NAME_VAR (retval), call);
|
retval = make_ssa_name (SSA_NAME_VAR (retval), call);
|
|
|
/* See if there is PHI defining return value. */
|
/* See if there is PHI defining return value. */
|
for (psi = gsi_start_phis (return_bb);
|
for (psi = gsi_start_phis (return_bb);
|
!gsi_end_p (psi); gsi_next (&psi))
|
!gsi_end_p (psi); gsi_next (&psi))
|
if (is_gimple_reg (gimple_phi_result (gsi_stmt (psi))))
|
if (is_gimple_reg (gimple_phi_result (gsi_stmt (psi))))
|
break;
|
break;
|
|
|
/* When there is PHI, just update its value. */
|
/* When there is PHI, just update its value. */
|
if (TREE_CODE (retval) == SSA_NAME
|
if (TREE_CODE (retval) == SSA_NAME
|
&& !gsi_end_p (psi))
|
&& !gsi_end_p (psi))
|
add_phi_arg (gsi_stmt (psi), retval, e, UNKNOWN_LOCATION);
|
add_phi_arg (gsi_stmt (psi), retval, e, UNKNOWN_LOCATION);
|
/* Otherwise update the return BB itself.
|
/* Otherwise update the return BB itself.
|
find_return_bb allows at most one assignment to return value,
|
find_return_bb allows at most one assignment to return value,
|
so update first statement. */
|
so update first statement. */
|
else
|
else
|
{
|
{
|
gimple_stmt_iterator bsi;
|
gimple_stmt_iterator bsi;
|
for (bsi = gsi_start_bb (return_bb); !gsi_end_p (bsi);
|
for (bsi = gsi_start_bb (return_bb); !gsi_end_p (bsi);
|
gsi_next (&bsi))
|
gsi_next (&bsi))
|
if (gimple_code (gsi_stmt (bsi)) == GIMPLE_RETURN)
|
if (gimple_code (gsi_stmt (bsi)) == GIMPLE_RETURN)
|
{
|
{
|
gimple_return_set_retval (gsi_stmt (bsi), retval);
|
gimple_return_set_retval (gsi_stmt (bsi), retval);
|
break;
|
break;
|
}
|
}
|
else if (gimple_code (gsi_stmt (bsi)) == GIMPLE_ASSIGN
|
else if (gimple_code (gsi_stmt (bsi)) == GIMPLE_ASSIGN
|
&& !gimple_clobber_p (gsi_stmt (bsi)))
|
&& !gimple_clobber_p (gsi_stmt (bsi)))
|
{
|
{
|
gimple_assign_set_rhs1 (gsi_stmt (bsi), retval);
|
gimple_assign_set_rhs1 (gsi_stmt (bsi), retval);
|
break;
|
break;
|
}
|
}
|
update_stmt (gsi_stmt (bsi));
|
update_stmt (gsi_stmt (bsi));
|
}
|
}
|
}
|
}
|
if (DECL_BY_REFERENCE (DECL_RESULT (current_function_decl)))
|
if (DECL_BY_REFERENCE (DECL_RESULT (current_function_decl)))
|
{
|
{
|
gimple_call_set_lhs (call, build_simple_mem_ref (retval));
|
gimple_call_set_lhs (call, build_simple_mem_ref (retval));
|
gsi_insert_after (&gsi, call, GSI_NEW_STMT);
|
gsi_insert_after (&gsi, call, GSI_NEW_STMT);
|
}
|
}
|
else
|
else
|
{
|
{
|
tree restype;
|
tree restype;
|
restype = TREE_TYPE (DECL_RESULT (current_function_decl));
|
restype = TREE_TYPE (DECL_RESULT (current_function_decl));
|
gsi_insert_after (&gsi, call, GSI_NEW_STMT);
|
gsi_insert_after (&gsi, call, GSI_NEW_STMT);
|
if (!useless_type_conversion_p (TREE_TYPE (retval), restype))
|
if (!useless_type_conversion_p (TREE_TYPE (retval), restype))
|
{
|
{
|
gimple cpy;
|
gimple cpy;
|
tree tem = create_tmp_reg (restype, NULL);
|
tree tem = create_tmp_reg (restype, NULL);
|
tem = make_ssa_name (tem, call);
|
tem = make_ssa_name (tem, call);
|
cpy = gimple_build_assign_with_ops (NOP_EXPR, retval,
|
cpy = gimple_build_assign_with_ops (NOP_EXPR, retval,
|
tem, NULL_TREE);
|
tem, NULL_TREE);
|
gsi_insert_after (&gsi, cpy, GSI_NEW_STMT);
|
gsi_insert_after (&gsi, cpy, GSI_NEW_STMT);
|
retval = tem;
|
retval = tem;
|
}
|
}
|
gimple_call_set_lhs (call, retval);
|
gimple_call_set_lhs (call, retval);
|
update_stmt (call);
|
update_stmt (call);
|
}
|
}
|
}
|
}
|
else
|
else
|
gsi_insert_after (&gsi, call, GSI_NEW_STMT);
|
gsi_insert_after (&gsi, call, GSI_NEW_STMT);
|
}
|
}
|
/* We don't use return block (there is either no return in function or
|
/* We don't use return block (there is either no return in function or
|
multiple of them). So create new basic block with return statement.
|
multiple of them). So create new basic block with return statement.
|
*/
|
*/
|
else
|
else
|
{
|
{
|
gimple ret;
|
gimple ret;
|
if (split_point->split_part_set_retval
|
if (split_point->split_part_set_retval
|
&& !VOID_TYPE_P (TREE_TYPE (TREE_TYPE (current_function_decl))))
|
&& !VOID_TYPE_P (TREE_TYPE (TREE_TYPE (current_function_decl))))
|
{
|
{
|
retval = DECL_RESULT (current_function_decl);
|
retval = DECL_RESULT (current_function_decl);
|
|
|
/* We use temporary register to hold value when aggregate_value_p
|
/* We use temporary register to hold value when aggregate_value_p
|
is false. Similarly for DECL_BY_REFERENCE we must avoid extra
|
is false. Similarly for DECL_BY_REFERENCE we must avoid extra
|
copy. */
|
copy. */
|
if (!aggregate_value_p (retval, TREE_TYPE (current_function_decl))
|
if (!aggregate_value_p (retval, TREE_TYPE (current_function_decl))
|
&& !DECL_BY_REFERENCE (retval))
|
&& !DECL_BY_REFERENCE (retval))
|
retval = create_tmp_reg (TREE_TYPE (retval), NULL);
|
retval = create_tmp_reg (TREE_TYPE (retval), NULL);
|
if (is_gimple_reg (retval))
|
if (is_gimple_reg (retval))
|
{
|
{
|
/* When returning by reference, there is only one SSA name
|
/* When returning by reference, there is only one SSA name
|
assigned to RESULT_DECL (that is pointer to return value).
|
assigned to RESULT_DECL (that is pointer to return value).
|
Look it up or create new one if it is missing. */
|
Look it up or create new one if it is missing. */
|
if (DECL_BY_REFERENCE (retval))
|
if (DECL_BY_REFERENCE (retval))
|
{
|
{
|
tree retval_name;
|
tree retval_name;
|
if ((retval_name = gimple_default_def (cfun, retval))
|
if ((retval_name = gimple_default_def (cfun, retval))
|
!= NULL)
|
!= NULL)
|
retval = retval_name;
|
retval = retval_name;
|
else
|
else
|
{
|
{
|
retval_name = make_ssa_name (retval,
|
retval_name = make_ssa_name (retval,
|
gimple_build_nop ());
|
gimple_build_nop ());
|
set_default_def (retval, retval_name);
|
set_default_def (retval, retval_name);
|
retval = retval_name;
|
retval = retval_name;
|
}
|
}
|
}
|
}
|
/* Otherwise produce new SSA name for return value. */
|
/* Otherwise produce new SSA name for return value. */
|
else
|
else
|
retval = make_ssa_name (retval, call);
|
retval = make_ssa_name (retval, call);
|
}
|
}
|
if (DECL_BY_REFERENCE (DECL_RESULT (current_function_decl)))
|
if (DECL_BY_REFERENCE (DECL_RESULT (current_function_decl)))
|
gimple_call_set_lhs (call, build_simple_mem_ref (retval));
|
gimple_call_set_lhs (call, build_simple_mem_ref (retval));
|
else
|
else
|
gimple_call_set_lhs (call, retval);
|
gimple_call_set_lhs (call, retval);
|
}
|
}
|
gsi_insert_after (&gsi, call, GSI_NEW_STMT);
|
gsi_insert_after (&gsi, call, GSI_NEW_STMT);
|
ret = gimple_build_return (retval);
|
ret = gimple_build_return (retval);
|
gsi_insert_after (&gsi, ret, GSI_NEW_STMT);
|
gsi_insert_after (&gsi, ret, GSI_NEW_STMT);
|
}
|
}
|
}
|
}
|
free_dominance_info (CDI_DOMINATORS);
|
free_dominance_info (CDI_DOMINATORS);
|
free_dominance_info (CDI_POST_DOMINATORS);
|
free_dominance_info (CDI_POST_DOMINATORS);
|
compute_inline_parameters (node, true);
|
compute_inline_parameters (node, true);
|
}
|
}
|
|
|
/* Execute function splitting pass. */
|
/* Execute function splitting pass. */
|
|
|
static unsigned int
|
static unsigned int
|
execute_split_functions (void)
|
execute_split_functions (void)
|
{
|
{
|
gimple_stmt_iterator bsi;
|
gimple_stmt_iterator bsi;
|
basic_block bb;
|
basic_block bb;
|
int overall_time = 0, overall_size = 0;
|
int overall_time = 0, overall_size = 0;
|
int todo = 0;
|
int todo = 0;
|
struct cgraph_node *node = cgraph_get_node (current_function_decl);
|
struct cgraph_node *node = cgraph_get_node (current_function_decl);
|
|
|
if (flags_from_decl_or_type (current_function_decl)
|
if (flags_from_decl_or_type (current_function_decl)
|
& (ECF_NORETURN|ECF_MALLOC))
|
& (ECF_NORETURN|ECF_MALLOC))
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "Not splitting: noreturn/malloc function.\n");
|
fprintf (dump_file, "Not splitting: noreturn/malloc function.\n");
|
return 0;
|
return 0;
|
}
|
}
|
if (MAIN_NAME_P (DECL_NAME (current_function_decl)))
|
if (MAIN_NAME_P (DECL_NAME (current_function_decl)))
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "Not splitting: main function.\n");
|
fprintf (dump_file, "Not splitting: main function.\n");
|
return 0;
|
return 0;
|
}
|
}
|
/* This can be relaxed; function might become inlinable after splitting
|
/* This can be relaxed; function might become inlinable after splitting
|
away the uninlinable part. */
|
away the uninlinable part. */
|
if (!inline_summary (node)->inlinable)
|
if (!inline_summary (node)->inlinable)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "Not splitting: not inlinable.\n");
|
fprintf (dump_file, "Not splitting: not inlinable.\n");
|
return 0;
|
return 0;
|
}
|
}
|
if (DECL_DISREGARD_INLINE_LIMITS (node->decl))
|
if (DECL_DISREGARD_INLINE_LIMITS (node->decl))
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "Not splitting: disregarding inline limits.\n");
|
fprintf (dump_file, "Not splitting: disregarding inline limits.\n");
|
return 0;
|
return 0;
|
}
|
}
|
/* This can be relaxed; most of versioning tests actually prevents
|
/* This can be relaxed; most of versioning tests actually prevents
|
a duplication. */
|
a duplication. */
|
if (!tree_versionable_function_p (current_function_decl))
|
if (!tree_versionable_function_p (current_function_decl))
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "Not splitting: not versionable.\n");
|
fprintf (dump_file, "Not splitting: not versionable.\n");
|
return 0;
|
return 0;
|
}
|
}
|
/* FIXME: we could support this. */
|
/* FIXME: we could support this. */
|
if (DECL_STRUCT_FUNCTION (current_function_decl)->static_chain_decl)
|
if (DECL_STRUCT_FUNCTION (current_function_decl)->static_chain_decl)
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "Not splitting: nested function.\n");
|
fprintf (dump_file, "Not splitting: nested function.\n");
|
return 0;
|
return 0;
|
}
|
}
|
|
|
/* See if it makes sense to try to split.
|
/* See if it makes sense to try to split.
|
It makes sense to split if we inline, that is if we have direct calls to
|
It makes sense to split if we inline, that is if we have direct calls to
|
handle or direct calls are possibly going to appear as result of indirect
|
handle or direct calls are possibly going to appear as result of indirect
|
inlining or LTO. Also handle -fprofile-generate as LTO to allow non-LTO
|
inlining or LTO. Also handle -fprofile-generate as LTO to allow non-LTO
|
training for LTO -fprofile-use build.
|
training for LTO -fprofile-use build.
|
|
|
Note that we are not completely conservative about disqualifying functions
|
Note that we are not completely conservative about disqualifying functions
|
called once. It is possible that the caller is called more then once and
|
called once. It is possible that the caller is called more then once and
|
then inlining would still benefit. */
|
then inlining would still benefit. */
|
if ((!node->callers || !node->callers->next_caller)
|
if ((!node->callers || !node->callers->next_caller)
|
&& !node->address_taken
|
&& !node->address_taken
|
&& (!flag_lto || !node->local.externally_visible))
|
&& (!flag_lto || !node->local.externally_visible))
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "Not splitting: not called directly "
|
fprintf (dump_file, "Not splitting: not called directly "
|
"or called once.\n");
|
"or called once.\n");
|
return 0;
|
return 0;
|
}
|
}
|
|
|
/* FIXME: We can actually split if splitting reduces call overhead. */
|
/* FIXME: We can actually split if splitting reduces call overhead. */
|
if (!flag_inline_small_functions
|
if (!flag_inline_small_functions
|
&& !DECL_DECLARED_INLINE_P (current_function_decl))
|
&& !DECL_DECLARED_INLINE_P (current_function_decl))
|
{
|
{
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "Not splitting: not autoinlining and function"
|
fprintf (dump_file, "Not splitting: not autoinlining and function"
|
" is not inline.\n");
|
" is not inline.\n");
|
return 0;
|
return 0;
|
}
|
}
|
|
|
/* Initialize bitmap to track forbidden calls. */
|
/* Initialize bitmap to track forbidden calls. */
|
forbidden_dominators = BITMAP_ALLOC (NULL);
|
forbidden_dominators = BITMAP_ALLOC (NULL);
|
calculate_dominance_info (CDI_DOMINATORS);
|
calculate_dominance_info (CDI_DOMINATORS);
|
|
|
/* Compute local info about basic blocks and determine function size/time. */
|
/* Compute local info about basic blocks and determine function size/time. */
|
VEC_safe_grow_cleared (bb_info, heap, bb_info_vec, last_basic_block + 1);
|
VEC_safe_grow_cleared (bb_info, heap, bb_info_vec, last_basic_block + 1);
|
memset (&best_split_point, 0, sizeof (best_split_point));
|
memset (&best_split_point, 0, sizeof (best_split_point));
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
{
|
{
|
int time = 0;
|
int time = 0;
|
int size = 0;
|
int size = 0;
|
int freq = compute_call_stmt_bb_frequency (current_function_decl, bb);
|
int freq = compute_call_stmt_bb_frequency (current_function_decl, bb);
|
|
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
fprintf (dump_file, "Basic block %i\n", bb->index);
|
fprintf (dump_file, "Basic block %i\n", bb->index);
|
|
|
for (bsi = gsi_start_bb (bb); !gsi_end_p (bsi); gsi_next (&bsi))
|
for (bsi = gsi_start_bb (bb); !gsi_end_p (bsi); gsi_next (&bsi))
|
{
|
{
|
int this_time, this_size;
|
int this_time, this_size;
|
gimple stmt = gsi_stmt (bsi);
|
gimple stmt = gsi_stmt (bsi);
|
|
|
this_size = estimate_num_insns (stmt, &eni_size_weights);
|
this_size = estimate_num_insns (stmt, &eni_size_weights);
|
this_time = estimate_num_insns (stmt, &eni_time_weights) * freq;
|
this_time = estimate_num_insns (stmt, &eni_time_weights) * freq;
|
size += this_size;
|
size += this_size;
|
time += this_time;
|
time += this_time;
|
check_forbidden_calls (stmt);
|
check_forbidden_calls (stmt);
|
|
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
{
|
{
|
fprintf (dump_file, " freq:%6i size:%3i time:%3i ",
|
fprintf (dump_file, " freq:%6i size:%3i time:%3i ",
|
freq, this_size, this_time);
|
freq, this_size, this_time);
|
print_gimple_stmt (dump_file, stmt, 0, 0);
|
print_gimple_stmt (dump_file, stmt, 0, 0);
|
}
|
}
|
}
|
}
|
overall_time += time;
|
overall_time += time;
|
overall_size += size;
|
overall_size += size;
|
VEC_index (bb_info, bb_info_vec, bb->index)->time = time;
|
VEC_index (bb_info, bb_info_vec, bb->index)->time = time;
|
VEC_index (bb_info, bb_info_vec, bb->index)->size = size;
|
VEC_index (bb_info, bb_info_vec, bb->index)->size = size;
|
}
|
}
|
find_split_points (overall_time, overall_size);
|
find_split_points (overall_time, overall_size);
|
if (best_split_point.split_bbs)
|
if (best_split_point.split_bbs)
|
{
|
{
|
split_function (&best_split_point);
|
split_function (&best_split_point);
|
BITMAP_FREE (best_split_point.ssa_names_to_pass);
|
BITMAP_FREE (best_split_point.ssa_names_to_pass);
|
BITMAP_FREE (best_split_point.split_bbs);
|
BITMAP_FREE (best_split_point.split_bbs);
|
todo = TODO_update_ssa | TODO_cleanup_cfg;
|
todo = TODO_update_ssa | TODO_cleanup_cfg;
|
}
|
}
|
BITMAP_FREE (forbidden_dominators);
|
BITMAP_FREE (forbidden_dominators);
|
VEC_free (bb_info, heap, bb_info_vec);
|
VEC_free (bb_info, heap, bb_info_vec);
|
bb_info_vec = NULL;
|
bb_info_vec = NULL;
|
return todo;
|
return todo;
|
}
|
}
|
|
|
/* Gate function splitting pass. When doing profile feedback, we want
|
/* Gate function splitting pass. When doing profile feedback, we want
|
to execute the pass after profiling is read. So disable one in
|
to execute the pass after profiling is read. So disable one in
|
early optimization. */
|
early optimization. */
|
|
|
static bool
|
static bool
|
gate_split_functions (void)
|
gate_split_functions (void)
|
{
|
{
|
return (flag_partial_inlining
|
return (flag_partial_inlining
|
&& !profile_arc_flag && !flag_branch_probabilities);
|
&& !profile_arc_flag && !flag_branch_probabilities);
|
}
|
}
|
|
|
struct gimple_opt_pass pass_split_functions =
|
struct gimple_opt_pass pass_split_functions =
|
{
|
{
|
{
|
{
|
GIMPLE_PASS,
|
GIMPLE_PASS,
|
"fnsplit", /* name */
|
"fnsplit", /* name */
|
gate_split_functions, /* gate */
|
gate_split_functions, /* gate */
|
execute_split_functions, /* execute */
|
execute_split_functions, /* execute */
|
NULL, /* sub */
|
NULL, /* sub */
|
NULL, /* next */
|
NULL, /* next */
|
0, /* static_pass_number */
|
0, /* static_pass_number */
|
TV_IPA_FNSPLIT, /* tv_id */
|
TV_IPA_FNSPLIT, /* tv_id */
|
PROP_cfg, /* properties_required */
|
PROP_cfg, /* properties_required */
|
0, /* properties_provided */
|
0, /* properties_provided */
|
0, /* properties_destroyed */
|
0, /* properties_destroyed */
|
0, /* todo_flags_start */
|
0, /* todo_flags_start */
|
TODO_verify_all /* todo_flags_finish */
|
TODO_verify_all /* todo_flags_finish */
|
}
|
}
|
};
|
};
|
|
|
/* Gate feedback driven function splitting pass.
|
/* Gate feedback driven function splitting pass.
|
We don't need to split when profiling at all, we are producing
|
We don't need to split when profiling at all, we are producing
|
lousy code anyway. */
|
lousy code anyway. */
|
|
|
static bool
|
static bool
|
gate_feedback_split_functions (void)
|
gate_feedback_split_functions (void)
|
{
|
{
|
return (flag_partial_inlining
|
return (flag_partial_inlining
|
&& flag_branch_probabilities);
|
&& flag_branch_probabilities);
|
}
|
}
|
|
|
/* Execute function splitting pass. */
|
/* Execute function splitting pass. */
|
|
|
static unsigned int
|
static unsigned int
|
execute_feedback_split_functions (void)
|
execute_feedback_split_functions (void)
|
{
|
{
|
unsigned int retval = execute_split_functions ();
|
unsigned int retval = execute_split_functions ();
|
if (retval)
|
if (retval)
|
retval |= TODO_rebuild_cgraph_edges;
|
retval |= TODO_rebuild_cgraph_edges;
|
return retval;
|
return retval;
|
}
|
}
|
|
|
struct gimple_opt_pass pass_feedback_split_functions =
|
struct gimple_opt_pass pass_feedback_split_functions =
|
{
|
{
|
{
|
{
|
GIMPLE_PASS,
|
GIMPLE_PASS,
|
"feedback_fnsplit", /* name */
|
"feedback_fnsplit", /* name */
|
gate_feedback_split_functions, /* gate */
|
gate_feedback_split_functions, /* gate */
|
execute_feedback_split_functions, /* execute */
|
execute_feedback_split_functions, /* execute */
|
NULL, /* sub */
|
NULL, /* sub */
|
NULL, /* next */
|
NULL, /* next */
|
0, /* static_pass_number */
|
0, /* static_pass_number */
|
TV_IPA_FNSPLIT, /* tv_id */
|
TV_IPA_FNSPLIT, /* tv_id */
|
PROP_cfg, /* properties_required */
|
PROP_cfg, /* properties_required */
|
0, /* properties_provided */
|
0, /* properties_provided */
|
0, /* properties_destroyed */
|
0, /* properties_destroyed */
|
0, /* todo_flags_start */
|
0, /* todo_flags_start */
|
TODO_verify_all /* todo_flags_finish */
|
TODO_verify_all /* todo_flags_finish */
|
}
|
}
|
};
|
};
|
|
|
