blob: 83d0eeb7d11a3775a7ab03c2b2d104d08ead5512 [file] [log] [blame]
/* Tree inlining.
Copyright 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009
Free Software Foundation, Inc.
Contributed by Alexandre Oliva <aoliva@redhat.com>
This file is part of GCC.
GCC is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 3, or (at your option)
any later version.
GCC is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with GCC; see the file COPYING3. If not see
<http://www.gnu.org/licenses/>. */
#include "config.h"
#include "system.h"
#include "coretypes.h"
#include "tm.h"
#include "toplev.h"
#include "tree.h"
#include "tree-inline.h"
#include "rtl.h"
#include "expr.h"
#include "flags.h"
#include "params.h"
#include "input.h"
#include "insn-config.h"
#include "varray.h"
#include "hashtab.h"
#include "langhooks.h"
#include "basic-block.h"
#include "tree-iterator.h"
#include "cgraph.h"
#include "intl.h"
#include "tree-mudflap.h"
#include "tree-flow.h"
#include "function.h"
#include "ggc.h"
#include "tree-flow.h"
#include "diagnostic.h"
#include "except.h"
#include "debug.h"
#include "pointer-set.h"
#include "ipa-prop.h"
#include "value-prof.h"
#include "tree-pass.h"
#include "target.h"
#include "integrate.h"
/* I'm not real happy about this, but we need to handle gimple and
non-gimple trees. */
#include "gimple.h"
/* Inlining, Cloning, Versioning, Parallelization
Inlining: a function body is duplicated, but the PARM_DECLs are
remapped into VAR_DECLs, and non-void RETURN_EXPRs become
MODIFY_EXPRs that store to a dedicated returned-value variable.
The duplicated eh_region info of the copy will later be appended
to the info for the caller; the eh_region info in copied throwing
statements and RESX_EXPRs is adjusted accordingly.
Cloning: (only in C++) We have one body for a con/de/structor, and
multiple function decls, each with a unique parameter list.
Duplicate the body, using the given splay tree; some parameters
will become constants (like 0 or 1).
Versioning: a function body is duplicated and the result is a new
function rather than into blocks of an existing function as with
inlining. Some parameters will become constants.
Parallelization: a region of a function is duplicated resulting in
a new function. Variables may be replaced with complex expressions
to enable shared variable semantics.
All of these will simultaneously lookup any callgraph edges. If
we're going to inline the duplicated function body, and the given
function has some cloned callgraph nodes (one for each place this
function will be inlined) those callgraph edges will be duplicated.
If we're cloning the body, those callgraph edges will be
updated to point into the new body. (Note that the original
callgraph node and edge list will not be altered.)
See the CALL_EXPR handling case in copy_tree_body_r (). */
/* To Do:
o In order to make inlining-on-trees work, we pessimized
function-local static constants. In particular, they are now
always output, even when not addressed. Fix this by treating
function-local static constants just like global static
constants; the back-end already knows not to output them if they
are not needed.
o Provide heuristics to clamp inlining of recursive template
calls? */
/* Weights that estimate_num_insns uses for heuristics in inlining. */
eni_weights eni_inlining_weights;
/* Weights that estimate_num_insns uses to estimate the size of the
produced code. */
eni_weights eni_size_weights;
/* Weights that estimate_num_insns uses to estimate the time necessary
to execute the produced code. */
eni_weights eni_time_weights;
/* Prototypes. */
static tree declare_return_variable (copy_body_data *, tree, tree, tree *);
static bool inlinable_function_p (tree);
static void remap_block (tree *, copy_body_data *);
static void copy_bind_expr (tree *, int *, copy_body_data *);
static tree mark_local_for_remap_r (tree *, int *, void *);
static void unsave_expr_1 (tree);
static tree unsave_r (tree *, int *, void *);
static void declare_inline_vars (tree, tree);
static void remap_save_expr (tree *, void *, int *);
static void prepend_lexical_block (tree current_block, tree new_block);
static tree copy_decl_to_var (tree, copy_body_data *);
static tree copy_result_decl_to_var (tree, copy_body_data *);
static tree copy_decl_maybe_to_var (tree, copy_body_data *);
static gimple remap_gimple_stmt (gimple, copy_body_data *);
/* Insert a tree->tree mapping for ID. Despite the name suggests
that the trees should be variables, it is used for more than that. */
void
insert_decl_map (copy_body_data *id, tree key, tree value)
{
*pointer_map_insert (id->decl_map, key) = value;
/* Always insert an identity map as well. If we see this same new
node again, we won't want to duplicate it a second time. */
if (key != value)
*pointer_map_insert (id->decl_map, value) = value;
}
/* Construct new SSA name for old NAME. ID is the inline context. */
static tree
remap_ssa_name (tree name, copy_body_data *id)
{
tree new_tree;
tree *n;
gcc_assert (TREE_CODE (name) == SSA_NAME);
n = (tree *) pointer_map_contains (id->decl_map, name);
if (n)
return unshare_expr (*n);
/* Do not set DEF_STMT yet as statement is not copied yet. We do that
in copy_bb. */
new_tree = remap_decl (SSA_NAME_VAR (name), id);
/* We might've substituted constant or another SSA_NAME for
the variable.
Replace the SSA name representing RESULT_DECL by variable during
inlining: this saves us from need to introduce PHI node in a case
return value is just partly initialized. */
if ((TREE_CODE (new_tree) == VAR_DECL || TREE_CODE (new_tree) == PARM_DECL)
&& (TREE_CODE (SSA_NAME_VAR (name)) != RESULT_DECL
|| !id->transform_return_to_modify))
{
new_tree = make_ssa_name (new_tree, NULL);
insert_decl_map (id, name, new_tree);
SSA_NAME_OCCURS_IN_ABNORMAL_PHI (new_tree)
= SSA_NAME_OCCURS_IN_ABNORMAL_PHI (name);
TREE_TYPE (new_tree) = TREE_TYPE (SSA_NAME_VAR (new_tree));
if (gimple_nop_p (SSA_NAME_DEF_STMT (name)))
{
/* By inlining function having uninitialized variable, we might
extend the lifetime (variable might get reused). This cause
ICE in the case we end up extending lifetime of SSA name across
abnormal edge, but also increase register pressure.
We simply initialize all uninitialized vars by 0 except
for case we are inlining to very first BB. We can avoid
this for all BBs that are not inside strongly connected
regions of the CFG, but this is expensive to test. */
if (id->entry_bb
&& is_gimple_reg (SSA_NAME_VAR (name))
&& TREE_CODE (SSA_NAME_VAR (name)) != PARM_DECL
&& (id->entry_bb != EDGE_SUCC (ENTRY_BLOCK_PTR, 0)->dest
|| EDGE_COUNT (id->entry_bb->preds) != 1))
{
gimple_stmt_iterator gsi = gsi_last_bb (id->entry_bb);
gimple init_stmt;
init_stmt = gimple_build_assign (new_tree,
fold_convert (TREE_TYPE (new_tree),
integer_zero_node));
gsi_insert_after (&gsi, init_stmt, GSI_NEW_STMT);
SSA_NAME_IS_DEFAULT_DEF (new_tree) = 0;
}
else
{
SSA_NAME_DEF_STMT (new_tree) = gimple_build_nop ();
if (gimple_default_def (id->src_cfun, SSA_NAME_VAR (name))
== name)
set_default_def (SSA_NAME_VAR (new_tree), new_tree);
}
}
}
else
insert_decl_map (id, name, new_tree);
return new_tree;
}
/* Remap DECL during the copying of the BLOCK tree for the function. */
tree
remap_decl (tree decl, copy_body_data *id)
{
tree *n;
tree fn;
/* We only remap local variables in the current function. */
fn = id->src_fn;
/* See if we have remapped this declaration. */
n = (tree *) pointer_map_contains (id->decl_map, decl);
/* If we didn't already have an equivalent for this declaration,
create one now. */
if (!n)
{
/* Make a copy of the variable or label. */
tree t = id->copy_decl (decl, id);
/* Remember it, so that if we encounter this local entity again
we can reuse this copy. Do this early because remap_type may
need this decl for TYPE_STUB_DECL. */
insert_decl_map (id, decl, t);
if (!DECL_P (t))
return t;
/* Remap types, if necessary. */
TREE_TYPE (t) = remap_type (TREE_TYPE (t), id);
if (TREE_CODE (t) == TYPE_DECL)
DECL_ORIGINAL_TYPE (t) = remap_type (DECL_ORIGINAL_TYPE (t), id);
/* Remap sizes as necessary. */
walk_tree (&DECL_SIZE (t), copy_tree_body_r, id, NULL);
walk_tree (&DECL_SIZE_UNIT (t), copy_tree_body_r, id, NULL);
/* If fields, do likewise for offset and qualifier. */
if (TREE_CODE (t) == FIELD_DECL)
{
walk_tree (&DECL_FIELD_OFFSET (t), copy_tree_body_r, id, NULL);
if (TREE_CODE (DECL_CONTEXT (t)) == QUAL_UNION_TYPE)
walk_tree (&DECL_QUALIFIER (t), copy_tree_body_r, id, NULL);
}
if (cfun && gimple_in_ssa_p (cfun)
&& (TREE_CODE (t) == VAR_DECL
|| TREE_CODE (t) == RESULT_DECL || TREE_CODE (t) == PARM_DECL))
{
tree def = gimple_default_def (id->src_cfun, decl);
get_var_ann (t);
if (TREE_CODE (decl) != PARM_DECL && def)
{
tree map = remap_ssa_name (def, id);
/* Watch out RESULT_DECLs whose SSA names map directly
to them. */
if (TREE_CODE (map) == SSA_NAME
&& gimple_nop_p (SSA_NAME_DEF_STMT (map)))
set_default_def (t, map);
}
add_referenced_var (t);
}
return t;
}
return unshare_expr (*n);
}
static tree
remap_type_1 (tree type, copy_body_data *id)
{
tree new_tree, t;
/* We do need a copy. build and register it now. If this is a pointer or
reference type, remap the designated type and make a new pointer or
reference type. */
if (TREE_CODE (type) == POINTER_TYPE)
{
new_tree = build_pointer_type_for_mode (remap_type (TREE_TYPE (type), id),
TYPE_MODE (type),
TYPE_REF_CAN_ALIAS_ALL (type));
insert_decl_map (id, type, new_tree);
return new_tree;
}
else if (TREE_CODE (type) == REFERENCE_TYPE)
{
new_tree = build_reference_type_for_mode (remap_type (TREE_TYPE (type), id),
TYPE_MODE (type),
TYPE_REF_CAN_ALIAS_ALL (type));
insert_decl_map (id, type, new_tree);
return new_tree;
}
else
new_tree = copy_node (type);
insert_decl_map (id, type, new_tree);
/* This is a new type, not a copy of an old type. Need to reassociate
variants. We can handle everything except the main variant lazily. */
t = TYPE_MAIN_VARIANT (type);
if (type != t)
{
t = remap_type (t, id);
TYPE_MAIN_VARIANT (new_tree) = t;
TYPE_NEXT_VARIANT (new_tree) = TYPE_NEXT_VARIANT (t);
TYPE_NEXT_VARIANT (t) = new_tree;
}
else
{
TYPE_MAIN_VARIANT (new_tree) = new_tree;
TYPE_NEXT_VARIANT (new_tree) = NULL;
}
if (TYPE_STUB_DECL (type))
TYPE_STUB_DECL (new_tree) = remap_decl (TYPE_STUB_DECL (type), id);
/* Lazily create pointer and reference types. */
TYPE_POINTER_TO (new_tree) = NULL;
TYPE_REFERENCE_TO (new_tree) = NULL;
switch (TREE_CODE (new_tree))
{
case INTEGER_TYPE:
case REAL_TYPE:
case FIXED_POINT_TYPE:
case ENUMERAL_TYPE:
case BOOLEAN_TYPE:
t = TYPE_MIN_VALUE (new_tree);
if (t && TREE_CODE (t) != INTEGER_CST)
walk_tree (&TYPE_MIN_VALUE (new_tree), copy_tree_body_r, id, NULL);
t = TYPE_MAX_VALUE (new_tree);
if (t && TREE_CODE (t) != INTEGER_CST)
walk_tree (&TYPE_MAX_VALUE (new_tree), copy_tree_body_r, id, NULL);
return new_tree;
case FUNCTION_TYPE:
TREE_TYPE (new_tree) = remap_type (TREE_TYPE (new_tree), id);
walk_tree (&TYPE_ARG_TYPES (new_tree), copy_tree_body_r, id, NULL);
return new_tree;
case ARRAY_TYPE:
TREE_TYPE (new_tree) = remap_type (TREE_TYPE (new_tree), id);
TYPE_DOMAIN (new_tree) = remap_type (TYPE_DOMAIN (new_tree), id);
break;
case RECORD_TYPE:
case UNION_TYPE:
case QUAL_UNION_TYPE:
{
tree f, nf = NULL;
for (f = TYPE_FIELDS (new_tree); f ; f = TREE_CHAIN (f))
{
t = remap_decl (f, id);
DECL_CONTEXT (t) = new_tree;
TREE_CHAIN (t) = nf;
nf = t;
}
TYPE_FIELDS (new_tree) = nreverse (nf);
}
break;
case OFFSET_TYPE:
default:
/* Shouldn't have been thought variable sized. */
gcc_unreachable ();
}
walk_tree (&TYPE_SIZE (new_tree), copy_tree_body_r, id, NULL);
walk_tree (&TYPE_SIZE_UNIT (new_tree), copy_tree_body_r, id, NULL);
return new_tree;
}
tree
remap_type (tree type, copy_body_data *id)
{
tree *node;
tree tmp;
if (type == NULL)
return type;
/* See if we have remapped this type. */
node = (tree *) pointer_map_contains (id->decl_map, type);
if (node)
return *node;
/* The type only needs remapping if it's variably modified. */
if (! variably_modified_type_p (type, id->src_fn))
{
insert_decl_map (id, type, type);
return type;
}
id->remapping_type_depth++;
tmp = remap_type_1 (type, id);
id->remapping_type_depth--;
return tmp;
}
/* Return previously remapped type of TYPE in ID. Return NULL if TYPE
is NULL or TYPE has not been remapped before. */
static tree
remapped_type (tree type, copy_body_data *id)
{
tree *node;
if (type == NULL)
return type;
/* See if we have remapped this type. */
node = (tree *) pointer_map_contains (id->decl_map, type);
if (node)
return *node;
else
return NULL;
}
/* The type only needs remapping if it's variably modified. */
/* Decide if DECL can be put into BLOCK_NONLOCAL_VARs. */
static bool
can_be_nonlocal (tree decl, copy_body_data *id)
{
/* We can not duplicate function decls. */
if (TREE_CODE (decl) == FUNCTION_DECL)
return true;
/* Local static vars must be non-local or we get multiple declaration
problems. */
if (TREE_CODE (decl) == VAR_DECL
&& !auto_var_in_fn_p (decl, id->src_fn))
return true;
/* At the moment dwarf2out can handle only these types of nodes. We
can support more later. */
if (TREE_CODE (decl) != VAR_DECL && TREE_CODE (decl) != PARM_DECL)
return false;
/* We must use global type. We call remapped_type instead of
remap_type since we don't want to remap this type here if it
hasn't been remapped before. */
if (TREE_TYPE (decl) != remapped_type (TREE_TYPE (decl), id))
return false;
/* Wihtout SSA we can't tell if variable is used. */
if (!gimple_in_ssa_p (cfun))
return false;
/* Live variables must be copied so we can attach DECL_RTL. */
if (var_ann (decl))
return false;
return true;
}
static tree
remap_decls (tree decls, VEC(tree,gc) **nonlocalized_list, copy_body_data *id)
{
tree old_var;
tree new_decls = NULL_TREE;
/* Remap its variables. */
for (old_var = decls; old_var; old_var = TREE_CHAIN (old_var))
{
tree new_var;
tree origin_var = DECL_ORIGIN (old_var);
if (can_be_nonlocal (old_var, id))
{
if (TREE_CODE (old_var) == VAR_DECL
&& (var_ann (old_var) || !gimple_in_ssa_p (cfun)))
cfun->local_decls = tree_cons (NULL_TREE, old_var,
cfun->local_decls);
if ((!optimize || debug_info_level > DINFO_LEVEL_TERSE)
&& !DECL_IGNORED_P (old_var)
&& nonlocalized_list)
VEC_safe_push (tree, gc, *nonlocalized_list, origin_var);
continue;
}
/* Remap the variable. */
new_var = remap_decl (old_var, id);
/* If we didn't remap this variable, we can't mess with its
TREE_CHAIN. If we remapped this variable to the return slot, it's
already declared somewhere else, so don't declare it here. */
if (new_var == id->retvar)
;
else if (!new_var)
{
if ((!optimize || debug_info_level > DINFO_LEVEL_TERSE)
&& !DECL_IGNORED_P (old_var)
&& nonlocalized_list)
VEC_safe_push (tree, gc, *nonlocalized_list, origin_var);
}
else
{
gcc_assert (DECL_P (new_var));
TREE_CHAIN (new_var) = new_decls;
new_decls = new_var;
}
}
return nreverse (new_decls);
}
/* Copy the BLOCK to contain remapped versions of the variables
therein. And hook the new block into the block-tree. */
static void
remap_block (tree *block, copy_body_data *id)
{
tree old_block;
tree new_block;
tree fn;
/* Make the new block. */
old_block = *block;
new_block = make_node (BLOCK);
TREE_USED (new_block) = TREE_USED (old_block);
BLOCK_ABSTRACT_ORIGIN (new_block) = old_block;
BLOCK_SOURCE_LOCATION (new_block) = BLOCK_SOURCE_LOCATION (old_block);
BLOCK_NONLOCALIZED_VARS (new_block)
= VEC_copy (tree, gc, BLOCK_NONLOCALIZED_VARS (old_block));
*block = new_block;
/* Remap its variables. */
BLOCK_VARS (new_block) = remap_decls (BLOCK_VARS (old_block),
&BLOCK_NONLOCALIZED_VARS (new_block),
id);
fn = id->dst_fn;
if (id->transform_lang_insert_block)
id->transform_lang_insert_block (new_block);
/* Remember the remapped block. */
insert_decl_map (id, old_block, new_block);
}
/* Copy the whole block tree and root it in id->block. */
static tree
remap_blocks (tree block, copy_body_data *id)
{
tree t;
tree new_tree = block;
if (!block)
return NULL;
remap_block (&new_tree, id);
gcc_assert (new_tree != block);
for (t = BLOCK_SUBBLOCKS (block); t ; t = BLOCK_CHAIN (t))
prepend_lexical_block (new_tree, remap_blocks (t, id));
/* Blocks are in arbitrary order, but make things slightly prettier and do
not swap order when producing a copy. */
BLOCK_SUBBLOCKS (new_tree) = blocks_nreverse (BLOCK_SUBBLOCKS (new_tree));
return new_tree;
}
static void
copy_statement_list (tree *tp)
{
tree_stmt_iterator oi, ni;
tree new_tree;
new_tree = alloc_stmt_list ();
ni = tsi_start (new_tree);
oi = tsi_start (*tp);
*tp = new_tree;
for (; !tsi_end_p (oi); tsi_next (&oi))
tsi_link_after (&ni, tsi_stmt (oi), TSI_NEW_STMT);
}
static void
copy_bind_expr (tree *tp, int *walk_subtrees, copy_body_data *id)
{
tree block = BIND_EXPR_BLOCK (*tp);
/* Copy (and replace) the statement. */
copy_tree_r (tp, walk_subtrees, NULL);
if (block)
{
remap_block (&block, id);
BIND_EXPR_BLOCK (*tp) = block;
}
if (BIND_EXPR_VARS (*tp))
/* This will remap a lot of the same decls again, but this should be
harmless. */
BIND_EXPR_VARS (*tp) = remap_decls (BIND_EXPR_VARS (*tp), NULL, id);
}
/* Create a new gimple_seq by remapping all the statements in BODY
using the inlining information in ID. */
gimple_seq
remap_gimple_seq (gimple_seq body, copy_body_data *id)
{
gimple_stmt_iterator si;
gimple_seq new_body = NULL;
for (si = gsi_start (body); !gsi_end_p (si); gsi_next (&si))
{
gimple new_stmt = remap_gimple_stmt (gsi_stmt (si), id);
gimple_seq_add_stmt (&new_body, new_stmt);
}
return new_body;
}
/* Copy a GIMPLE_BIND statement STMT, remapping all the symbols in its
block using the mapping information in ID. */
static gimple
copy_gimple_bind (gimple stmt, copy_body_data *id)
{
gimple new_bind;
tree new_block, new_vars;
gimple_seq body, new_body;
/* Copy the statement. Note that we purposely don't use copy_stmt
here because we need to remap statements as we copy. */
body = gimple_bind_body (stmt);
new_body = remap_gimple_seq (body, id);
new_block = gimple_bind_block (stmt);
if (new_block)
remap_block (&new_block, id);
/* This will remap a lot of the same decls again, but this should be
harmless. */
new_vars = gimple_bind_vars (stmt);
if (new_vars)
new_vars = remap_decls (new_vars, NULL, id);
new_bind = gimple_build_bind (new_vars, new_body, new_block);
return new_bind;
}
/* Remap the GIMPLE operand pointed to by *TP. DATA is really a
'struct walk_stmt_info *'. DATA->INFO is a 'copy_body_data *'.
WALK_SUBTREES is used to indicate walk_gimple_op whether to keep
recursing into the children nodes of *TP. */
static tree
remap_gimple_op_r (tree *tp, int *walk_subtrees, void *data)
{
struct walk_stmt_info *wi_p = (struct walk_stmt_info *) data;
copy_body_data *id = (copy_body_data *) wi_p->info;
tree fn = id->src_fn;
if (TREE_CODE (*tp) == SSA_NAME)
{
*tp = remap_ssa_name (*tp, id);
*walk_subtrees = 0;
return NULL;
}
else if (auto_var_in_fn_p (*tp, fn))
{
/* Local variables and labels need to be replaced by equivalent
variables. We don't want to copy static variables; there's
only one of those, no matter how many times we inline the
containing function. Similarly for globals from an outer
function. */
tree new_decl;
/* Remap the declaration. */
new_decl = remap_decl (*tp, id);
gcc_assert (new_decl);
/* Replace this variable with the copy. */
STRIP_TYPE_NOPS (new_decl);
*tp = new_decl;
*walk_subtrees = 0;
}
else if (TREE_CODE (*tp) == STATEMENT_LIST)
gcc_unreachable ();
else if (TREE_CODE (*tp) == SAVE_EXPR)
gcc_unreachable ();
else if (TREE_CODE (*tp) == LABEL_DECL
&& (!DECL_CONTEXT (*tp)
|| decl_function_context (*tp) == id->src_fn))
/* These may need to be remapped for EH handling. */
*tp = remap_decl (*tp, id);
else if (TYPE_P (*tp))
/* Types may need remapping as well. */
*tp = remap_type (*tp, id);
else if (CONSTANT_CLASS_P (*tp))
{
/* If this is a constant, we have to copy the node iff the type
will be remapped. copy_tree_r will not copy a constant. */
tree new_type = remap_type (TREE_TYPE (*tp), id);
if (new_type == TREE_TYPE (*tp))
*walk_subtrees = 0;
else if (TREE_CODE (*tp) == INTEGER_CST)
*tp = build_int_cst_wide (new_type, TREE_INT_CST_LOW (*tp),
TREE_INT_CST_HIGH (*tp));
else
{
*tp = copy_node (*tp);
TREE_TYPE (*tp) = new_type;
}
}
else
{
/* Otherwise, just copy the node. Note that copy_tree_r already
knows not to copy VAR_DECLs, etc., so this is safe. */
if (TREE_CODE (*tp) == INDIRECT_REF)
{
/* Get rid of *& from inline substitutions that can happen when a
pointer argument is an ADDR_EXPR. */
tree decl = TREE_OPERAND (*tp, 0);
tree *n;
n = (tree *) pointer_map_contains (id->decl_map, decl);
if (n)
{
tree type, new_tree, old;
/* If we happen to get an ADDR_EXPR in n->value, strip
it manually here as we'll eventually get ADDR_EXPRs
which lie about their types pointed to. In this case
build_fold_indirect_ref wouldn't strip the
INDIRECT_REF, but we absolutely rely on that. As
fold_indirect_ref does other useful transformations,
try that first, though. */
type = TREE_TYPE (TREE_TYPE (*n));
new_tree = unshare_expr (*n);
old = *tp;
*tp = gimple_fold_indirect_ref (new_tree);
if (!*tp)
{
if (TREE_CODE (new_tree) == ADDR_EXPR)
{
*tp = fold_indirect_ref_1 (type, new_tree);
/* ??? We should either assert here or build
a VIEW_CONVERT_EXPR instead of blindly leaking
incompatible types to our IL. */
if (! *tp)
*tp = TREE_OPERAND (new_tree, 0);
}
else
{
*tp = build1 (INDIRECT_REF, type, new_tree);
TREE_THIS_VOLATILE (*tp) = TREE_THIS_VOLATILE (old);
TREE_NO_WARNING (*tp) = TREE_NO_WARNING (old);
}
}
*walk_subtrees = 0;
return NULL;
}
}
/* Here is the "usual case". Copy this tree node, and then
tweak some special cases. */
copy_tree_r (tp, walk_subtrees, NULL);
/* Global variables we haven't seen yet need to go into referenced
vars. If not referenced from types only. */
if (gimple_in_ssa_p (cfun)
&& TREE_CODE (*tp) == VAR_DECL
&& id->remapping_type_depth == 0)
add_referenced_var (*tp);
/* We should never have TREE_BLOCK set on non-statements. */
if (EXPR_P (*tp))
gcc_assert (!TREE_BLOCK (*tp));
if (TREE_CODE (*tp) != OMP_CLAUSE)
TREE_TYPE (*tp) = remap_type (TREE_TYPE (*tp), id);
if (TREE_CODE (*tp) == TARGET_EXPR && TREE_OPERAND (*tp, 3))
{
/* The copied TARGET_EXPR has never been expanded, even if the
original node was expanded already. */
TREE_OPERAND (*tp, 1) = TREE_OPERAND (*tp, 3);
TREE_OPERAND (*tp, 3) = NULL_TREE;
}
else if (TREE_CODE (*tp) == ADDR_EXPR)
{
/* Variable substitution need not be simple. In particular,
the INDIRECT_REF substitution above. Make sure that
TREE_CONSTANT and friends are up-to-date. But make sure
to not improperly set TREE_BLOCK on some sub-expressions. */
int invariant = is_gimple_min_invariant (*tp);
tree block = id->block;
id->block = NULL_TREE;
walk_tree (&TREE_OPERAND (*tp, 0), copy_tree_body_r, id, NULL);
id->block = block;
/* Handle the case where we substituted an INDIRECT_REF
into the operand of the ADDR_EXPR. */
if (TREE_CODE (TREE_OPERAND (*tp, 0)) == INDIRECT_REF)
*tp = TREE_OPERAND (TREE_OPERAND (*tp, 0), 0);
else
recompute_tree_invariant_for_addr_expr (*tp);
/* If this used to be invariant, but is not any longer,
then regimplification is probably needed. */
if (invariant && !is_gimple_min_invariant (*tp))
id->regimplify = true;
*walk_subtrees = 0;
}
}
/* Keep iterating. */
return NULL_TREE;
}
/* Called from copy_body_id via walk_tree. DATA is really a
`copy_body_data *'. */
tree
copy_tree_body_r (tree *tp, int *walk_subtrees, void *data)
{
copy_body_data *id = (copy_body_data *) data;
tree fn = id->src_fn;
tree new_block;
/* Begin by recognizing trees that we'll completely rewrite for the
inlining context. Our output for these trees is completely
different from out input (e.g. RETURN_EXPR is deleted, and morphs
into an edge). Further down, we'll handle trees that get
duplicated and/or tweaked. */
/* When requested, RETURN_EXPRs should be transformed to just the
contained MODIFY_EXPR. The branch semantics of the return will
be handled elsewhere by manipulating the CFG rather than a statement. */
if (TREE_CODE (*tp) == RETURN_EXPR && id->transform_return_to_modify)
{
tree assignment = TREE_OPERAND (*tp, 0);
/* If we're returning something, just turn that into an
assignment into the equivalent of the original RESULT_DECL.
If the "assignment" is just the result decl, the result
decl has already been set (e.g. a recent "foo (&result_decl,
...)"); just toss the entire RETURN_EXPR. */
if (assignment && TREE_CODE (assignment) == MODIFY_EXPR)
{
/* Replace the RETURN_EXPR with (a copy of) the
MODIFY_EXPR hanging underneath. */
*tp = copy_node (assignment);
}
else /* Else the RETURN_EXPR returns no value. */
{
*tp = NULL;
return (tree) (void *)1;
}
}
else if (TREE_CODE (*tp) == SSA_NAME)
{
*tp = remap_ssa_name (*tp, id);
*walk_subtrees = 0;
return NULL;
}
/* Local variables and labels need to be replaced by equivalent
variables. We don't want to copy static variables; there's only
one of those, no matter how many times we inline the containing
function. Similarly for globals from an outer function. */
else if (auto_var_in_fn_p (*tp, fn))
{
tree new_decl;
/* Remap the declaration. */
new_decl = remap_decl (*tp, id);
gcc_assert (new_decl);
/* Replace this variable with the copy. */
STRIP_TYPE_NOPS (new_decl);
*tp = new_decl;
*walk_subtrees = 0;
}
else if (TREE_CODE (*tp) == STATEMENT_LIST)
copy_statement_list (tp);
else if (TREE_CODE (*tp) == SAVE_EXPR)
remap_save_expr (tp, id->decl_map, walk_subtrees);
else if (TREE_CODE (*tp) == LABEL_DECL
&& (! DECL_CONTEXT (*tp)
|| decl_function_context (*tp) == id->src_fn))
/* These may need to be remapped for EH handling. */
*tp = remap_decl (*tp, id);
else if (TREE_CODE (*tp) == BIND_EXPR)
copy_bind_expr (tp, walk_subtrees, id);
/* Types may need remapping as well. */
else if (TYPE_P (*tp))
*tp = remap_type (*tp, id);
/* If this is a constant, we have to copy the node iff the type will be
remapped. copy_tree_r will not copy a constant. */
else if (CONSTANT_CLASS_P (*tp))
{
tree new_type = remap_type (TREE_TYPE (*tp), id);
if (new_type == TREE_TYPE (*tp))
*walk_subtrees = 0;
else if (TREE_CODE (*tp) == INTEGER_CST)
*tp = build_int_cst_wide (new_type, TREE_INT_CST_LOW (*tp),
TREE_INT_CST_HIGH (*tp));
else
{
*tp = copy_node (*tp);
TREE_TYPE (*tp) = new_type;
}
}
/* Otherwise, just copy the node. Note that copy_tree_r already
knows not to copy VAR_DECLs, etc., so this is safe. */
else
{
/* Here we handle trees that are not completely rewritten.
First we detect some inlining-induced bogosities for
discarding. */
if (TREE_CODE (*tp) == MODIFY_EXPR
&& TREE_OPERAND (*tp, 0) == TREE_OPERAND (*tp, 1)
&& (auto_var_in_fn_p (TREE_OPERAND (*tp, 0), fn)))
{
/* Some assignments VAR = VAR; don't generate any rtl code
and thus don't count as variable modification. Avoid
keeping bogosities like 0 = 0. */
tree decl = TREE_OPERAND (*tp, 0), value;
tree *n;
n = (tree *) pointer_map_contains (id->decl_map, decl);
if (n)
{
value = *n;
STRIP_TYPE_NOPS (value);
if (TREE_CONSTANT (value) || TREE_READONLY (value))
{
*tp = build_empty_stmt ();
return copy_tree_body_r (tp, walk_subtrees, data);
}
}
}
else if (TREE_CODE (*tp) == INDIRECT_REF)
{
/* Get rid of *& from inline substitutions that can happen when a
pointer argument is an ADDR_EXPR. */
tree decl = TREE_OPERAND (*tp, 0);
tree *n;
n = (tree *) pointer_map_contains (id->decl_map, decl);
if (n)
{
tree new_tree;
tree old;
/* If we happen to get an ADDR_EXPR in n->value, strip
it manually here as we'll eventually get ADDR_EXPRs
which lie about their types pointed to. In this case
build_fold_indirect_ref wouldn't strip the INDIRECT_REF,
but we absolutely rely on that. As fold_indirect_ref
does other useful transformations, try that first, though. */
tree type = TREE_TYPE (TREE_TYPE (*n));
new_tree = unshare_expr (*n);
old = *tp;
*tp = gimple_fold_indirect_ref (new_tree);
if (! *tp)
{
if (TREE_CODE (new_tree) == ADDR_EXPR)
{
*tp = fold_indirect_ref_1 (type, new_tree);
/* ??? We should either assert here or build
a VIEW_CONVERT_EXPR instead of blindly leaking
incompatible types to our IL. */
if (! *tp)
*tp = TREE_OPERAND (new_tree, 0);
}
else
{
*tp = build1 (INDIRECT_REF, type, new_tree);
TREE_THIS_VOLATILE (*tp) = TREE_THIS_VOLATILE (old);
TREE_SIDE_EFFECTS (*tp) = TREE_SIDE_EFFECTS (old);
}
}
*walk_subtrees = 0;
return NULL;
}
}
/* Here is the "usual case". Copy this tree node, and then
tweak some special cases. */
copy_tree_r (tp, walk_subtrees, NULL);
/* Global variables we haven't seen yet needs to go into referenced
vars. If not referenced from types only. */
if (gimple_in_ssa_p (cfun)
&& TREE_CODE (*tp) == VAR_DECL
&& id->remapping_type_depth == 0)
add_referenced_var (*tp);
/* If EXPR has block defined, map it to newly constructed block.
When inlining we want EXPRs without block appear in the block
of function call. */
if (EXPR_P (*tp))
{
new_block = id->block;
if (TREE_BLOCK (*tp))
{
tree *n;
n = (tree *) pointer_map_contains (id->decl_map,
TREE_BLOCK (*tp));
gcc_assert (n);
new_block = *n;
}
TREE_BLOCK (*tp) = new_block;
}
if (TREE_CODE (*tp) == RESX_EXPR && id->eh_region_offset)
TREE_OPERAND (*tp, 0) =
build_int_cst (NULL_TREE,
id->eh_region_offset
+ TREE_INT_CST_LOW (TREE_OPERAND (*tp, 0)));
if (TREE_CODE (*tp) != OMP_CLAUSE)
TREE_TYPE (*tp) = remap_type (TREE_TYPE (*tp), id);
/* The copied TARGET_EXPR has never been expanded, even if the
original node was expanded already. */
if (TREE_CODE (*tp) == TARGET_EXPR && TREE_OPERAND (*tp, 3))
{
TREE_OPERAND (*tp, 1) = TREE_OPERAND (*tp, 3);
TREE_OPERAND (*tp, 3) = NULL_TREE;
}
/* Variable substitution need not be simple. In particular, the
INDIRECT_REF substitution above. Make sure that TREE_CONSTANT
and friends are up-to-date. */
else if (TREE_CODE (*tp) == ADDR_EXPR)
{
int invariant = is_gimple_min_invariant (*tp);
walk_tree (&TREE_OPERAND (*tp, 0), copy_tree_body_r, id, NULL);
/* Handle the case where we substituted an INDIRECT_REF
into the operand of the ADDR_EXPR. */
if (TREE_CODE (TREE_OPERAND (*tp, 0)) == INDIRECT_REF)
*tp = TREE_OPERAND (TREE_OPERAND (*tp, 0), 0);
else
recompute_tree_invariant_for_addr_expr (*tp);
/* If this used to be invariant, but is not any longer,
then regimplification is probably needed. */
if (invariant && !is_gimple_min_invariant (*tp))
id->regimplify = true;
*walk_subtrees = 0;
}
}
/* Keep iterating. */
return NULL_TREE;
}
/* Helper for copy_bb. Remap statement STMT using the inlining
information in ID. Return the new statement copy. */
static gimple
remap_gimple_stmt (gimple stmt, copy_body_data *id)
{
gimple copy = NULL;
struct walk_stmt_info wi;
tree new_block;
bool skip_first = false;
/* Begin by recognizing trees that we'll completely rewrite for the
inlining context. Our output for these trees is completely
different from out input (e.g. RETURN_EXPR is deleted, and morphs
into an edge). Further down, we'll handle trees that get
duplicated and/or tweaked. */
/* When requested, GIMPLE_RETURNs should be transformed to just the
contained GIMPLE_ASSIGN. The branch semantics of the return will
be handled elsewhere by manipulating the CFG rather than the
statement. */
if (gimple_code (stmt) == GIMPLE_RETURN && id->transform_return_to_modify)
{
tree retval = gimple_return_retval (stmt);
/* If we're returning something, just turn that into an
assignment into the equivalent of the original RESULT_DECL.
If RETVAL is just the result decl, the result decl has
already been set (e.g. a recent "foo (&result_decl, ...)");
just toss the entire GIMPLE_RETURN. */
if (retval && TREE_CODE (retval) != RESULT_DECL)
{
copy = gimple_build_assign (id->retvar, retval);
/* id->retvar is already substituted. Skip it on later remapping. */
skip_first = true;
}
else
return gimple_build_nop ();
}
else if (gimple_has_substatements (stmt))
{
gimple_seq s1, s2;
/* When cloning bodies from the C++ front end, we will be handed bodies
in High GIMPLE form. Handle here all the High GIMPLE statements that
have embedded statements. */
switch (gimple_code (stmt))
{
case GIMPLE_BIND:
copy = copy_gimple_bind (stmt, id);
break;
case GIMPLE_CATCH:
s1 = remap_gimple_seq (gimple_catch_handler (stmt), id);
copy = gimple_build_catch (gimple_catch_types (stmt), s1);
break;
case GIMPLE_EH_FILTER:
s1 = remap_gimple_seq (gimple_eh_filter_failure (stmt), id);
copy = gimple_build_eh_filter (gimple_eh_filter_types (stmt), s1);
break;
case GIMPLE_TRY:
s1 = remap_gimple_seq (gimple_try_eval (stmt), id);
s2 = remap_gimple_seq (gimple_try_cleanup (stmt), id);
copy = gimple_build_try (s1, s2, gimple_try_kind (stmt));
break;
case GIMPLE_WITH_CLEANUP_EXPR:
s1 = remap_gimple_seq (gimple_wce_cleanup (stmt), id);
copy = gimple_build_wce (s1);
break;
case GIMPLE_OMP_PARALLEL:
s1 = remap_gimple_seq (gimple_omp_body (stmt), id);
copy = gimple_build_omp_parallel
(s1,
gimple_omp_parallel_clauses (stmt),
gimple_omp_parallel_child_fn (stmt),
gimple_omp_parallel_data_arg (stmt));
break;
case GIMPLE_OMP_TASK:
s1 = remap_gimple_seq (gimple_omp_body (stmt), id);
copy = gimple_build_omp_task
(s1,
gimple_omp_task_clauses (stmt),
gimple_omp_task_child_fn (stmt),
gimple_omp_task_data_arg (stmt),
gimple_omp_task_copy_fn (stmt),
gimple_omp_task_arg_size (stmt),
gimple_omp_task_arg_align (stmt));
break;
case GIMPLE_OMP_FOR:
s1 = remap_gimple_seq (gimple_omp_body (stmt), id);
s2 = remap_gimple_seq (gimple_omp_for_pre_body (stmt), id);
copy = gimple_build_omp_for (s1, gimple_omp_for_clauses (stmt),
gimple_omp_for_collapse (stmt), s2);
{
size_t i;
for (i = 0; i < gimple_omp_for_collapse (stmt); i++)
{
gimple_omp_for_set_index (copy, i,
gimple_omp_for_index (stmt, i));
gimple_omp_for_set_initial (copy, i,
gimple_omp_for_initial (stmt, i));
gimple_omp_for_set_final (copy, i,
gimple_omp_for_final (stmt, i));
gimple_omp_for_set_incr (copy, i,
gimple_omp_for_incr (stmt, i));
gimple_omp_for_set_cond (copy, i,
gimple_omp_for_cond (stmt, i));
}
}
break;
case GIMPLE_OMP_MASTER:
s1 = remap_gimple_seq (gimple_omp_body (stmt), id);
copy = gimple_build_omp_master (s1);
break;
case GIMPLE_OMP_ORDERED:
s1 = remap_gimple_seq (gimple_omp_body (stmt), id);
copy = gimple_build_omp_ordered (s1);
break;
case GIMPLE_OMP_SECTION:
s1 = remap_gimple_seq (gimple_omp_body (stmt), id);
copy = gimple_build_omp_section (s1);
break;
case GIMPLE_OMP_SECTIONS:
s1 = remap_gimple_seq (gimple_omp_body (stmt), id);
copy = gimple_build_omp_sections
(s1, gimple_omp_sections_clauses (stmt));
break;
case GIMPLE_OMP_SINGLE:
s1 = remap_gimple_seq (gimple_omp_body (stmt), id);
copy = gimple_build_omp_single
(s1, gimple_omp_single_clauses (stmt));
break;
case GIMPLE_OMP_CRITICAL:
s1 = remap_gimple_seq (gimple_omp_body (stmt), id);
copy
= gimple_build_omp_critical (s1, gimple_omp_critical_name (stmt));
break;
default:
gcc_unreachable ();
}
}
else
{
if (gimple_assign_copy_p (stmt)
&& gimple_assign_lhs (stmt) == gimple_assign_rhs1 (stmt)
&& auto_var_in_fn_p (gimple_assign_lhs (stmt), id->src_fn))
{
/* Here we handle statements that are not completely rewritten.
First we detect some inlining-induced bogosities for
discarding. */
/* Some assignments VAR = VAR; don't generate any rtl code
and thus don't count as variable modification. Avoid
keeping bogosities like 0 = 0. */
tree decl = gimple_assign_lhs (stmt), value;
tree *n;
n = (tree *) pointer_map_contains (id->decl_map, decl);
if (n)
{
value = *n;
STRIP_TYPE_NOPS (value);
if (TREE_CONSTANT (value) || TREE_READONLY (value))
return gimple_build_nop ();
}
}
/* Create a new deep copy of the statement. */
copy = gimple_copy (stmt);
}
/* If STMT has a block defined, map it to the newly constructed
block. When inlining we want statements without a block to
appear in the block of the function call. */
new_block = id->block;
if (gimple_block (copy))
{
tree *n;
n = (tree *) pointer_map_contains (id->decl_map, gimple_block (copy));
gcc_assert (n);
new_block = *n;
}
gimple_set_block (copy, new_block);
/* Remap all the operands in COPY. */
memset (&wi, 0, sizeof (wi));
wi.info = id;
if (skip_first)
walk_tree (gimple_op_ptr (copy, 1), remap_gimple_op_r, &wi, NULL);
else
walk_gimple_op (copy, remap_gimple_op_r, &wi);
/* We have to handle EH region remapping of GIMPLE_RESX specially because
the region number is not an operand. */
if (gimple_code (stmt) == GIMPLE_RESX && id->eh_region_offset)
{
gimple_resx_set_region (copy, gimple_resx_region (stmt) + id->eh_region_offset);
}
return copy;
}
/* Copy basic block, scale profile accordingly. Edges will be taken care of
later */
static basic_block
copy_bb (copy_body_data *id, basic_block bb, int frequency_scale,
gcov_type count_scale)
{
gimple_stmt_iterator gsi, copy_gsi, seq_gsi;
basic_block copy_basic_block;
tree decl;
/* create_basic_block() will append every new block to
basic_block_info automatically. */
copy_basic_block = create_basic_block (NULL, (void *) 0,
(basic_block) bb->prev_bb->aux);
copy_basic_block->count = bb->count * count_scale / REG_BR_PROB_BASE;
/* We are going to rebuild frequencies from scratch. These values
have just small importance to drive canonicalize_loop_headers. */
copy_basic_block->frequency = ((gcov_type)bb->frequency
* frequency_scale / REG_BR_PROB_BASE);
if (copy_basic_block->frequency > BB_FREQ_MAX)
copy_basic_block->frequency = BB_FREQ_MAX;
copy_gsi = gsi_start_bb (copy_basic_block);
for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi))
{
gimple stmt = gsi_stmt (gsi);
gimple orig_stmt = stmt;
id->regimplify = false;
stmt = remap_gimple_stmt (stmt, id);
if (gimple_nop_p (stmt))
continue;
gimple_duplicate_stmt_histograms (cfun, stmt, id->src_cfun, orig_stmt);
seq_gsi = copy_gsi;
/* With return slot optimization we can end up with
non-gimple (foo *)&this->m, fix that here. */
if (is_gimple_assign (stmt)
&& gimple_assign_rhs_code (stmt) == NOP_EXPR
&& !is_gimple_val (gimple_assign_rhs1 (stmt)))
{
tree new_rhs;
new_rhs = force_gimple_operand_gsi (&seq_gsi,
gimple_assign_rhs1 (stmt),
true, NULL, false, GSI_NEW_STMT);
gimple_assign_set_rhs1 (stmt, new_rhs);
id->regimplify = false;
}
gsi_insert_after (&seq_gsi, stmt, GSI_NEW_STMT);
if (id->regimplify)
gimple_regimplify_operands (stmt, &seq_gsi);
/* If copy_basic_block has been empty at the start of this iteration,
call gsi_start_bb again to get at the newly added statements. */
if (gsi_end_p (copy_gsi))
copy_gsi = gsi_start_bb (copy_basic_block);
else
gsi_next (&copy_gsi);
/* Process the new statement. The call to gimple_regimplify_operands
possibly turned the statement into multiple statements, we
need to process all of them. */
do
{
stmt = gsi_stmt (copy_gsi);
if (is_gimple_call (stmt)
&& gimple_call_va_arg_pack_p (stmt)
&& id->gimple_call)
{
/* __builtin_va_arg_pack () should be replaced by
all arguments corresponding to ... in the caller. */
tree p;
gimple new_call;
VEC(tree, heap) *argarray;
size_t nargs = gimple_call_num_args (id->gimple_call);
size_t n;
for (p = DECL_ARGUMENTS (id->src_fn); p; p = TREE_CHAIN (p))
nargs--;
/* Create the new array of arguments. */
n = nargs + gimple_call_num_args (stmt);
argarray = VEC_alloc (tree, heap, n);
VEC_safe_grow (tree, heap, argarray, n);
/* Copy all the arguments before '...' */
memcpy (VEC_address (tree, argarray),
gimple_call_arg_ptr (stmt, 0),
gimple_call_num_args (stmt) * sizeof (tree));
/* Append the arguments passed in '...' */
memcpy (VEC_address(tree, argarray) + gimple_call_num_args (stmt),
gimple_call_arg_ptr (id->gimple_call, 0)
+ (gimple_call_num_args (id->gimple_call) - nargs),
nargs * sizeof (tree));
new_call = gimple_build_call_vec (gimple_call_fn (stmt),
argarray);
VEC_free (tree, heap, argarray);
/* Copy all GIMPLE_CALL flags, location and block, except
GF_CALL_VA_ARG_PACK. */
gimple_call_copy_flags (new_call, stmt);
gimple_call_set_va_arg_pack (new_call, false);
gimple_set_location (new_call, gimple_location (stmt));
gimple_set_block (new_call, gimple_block (stmt));
gimple_call_set_lhs (new_call, gimple_call_lhs (stmt));
gsi_replace (&copy_gsi, new_call, false);
gimple_set_bb (stmt, NULL);
stmt = new_call;
}
else if (is_gimple_call (stmt)
&& id->gimple_call
&& (decl = gimple_call_fndecl (stmt))
&& DECL_BUILT_IN_CLASS (decl) == BUILT_IN_NORMAL
&& DECL_FUNCTION_CODE (decl) == BUILT_IN_VA_ARG_PACK_LEN)
{
/* __builtin_va_arg_pack_len () should be replaced by
the number of anonymous arguments. */
size_t nargs = gimple_call_num_args (id->gimple_call);
tree count, p;
gimple new_stmt;
for (p = DECL_ARGUMENTS (id->src_fn); p; p = TREE_CHAIN (p))
nargs--;
count = build_int_cst (integer_type_node, nargs);
new_stmt = gimple_build_assign (gimple_call_lhs (stmt), count);
gsi_replace (&copy_gsi, new_stmt, false);
stmt = new_stmt;
}
/* Statements produced by inlining can be unfolded, especially
when we constant propagated some operands. We can't fold
them right now for two reasons:
1) folding require SSA_NAME_DEF_STMTs to be correct
2) we can't change function calls to builtins.
So we just mark statement for later folding. We mark
all new statements, instead just statements that has changed
by some nontrivial substitution so even statements made
foldable indirectly are updated. If this turns out to be
expensive, copy_body can be told to watch for nontrivial
changes. */
if (id->statements_to_fold)
pointer_set_insert (id->statements_to_fold, stmt);
/* We're duplicating a CALL_EXPR. Find any corresponding
callgraph edges and update or duplicate them. */
if (is_gimple_call (stmt))
{
struct cgraph_node *node;
struct cgraph_edge *edge;
int flags;
switch (id->transform_call_graph_edges)
{
case CB_CGE_DUPLICATE:
edge = cgraph_edge (id->src_node, orig_stmt);
if (edge)
cgraph_clone_edge (edge, id->dst_node, stmt,
REG_BR_PROB_BASE, 1,
edge->frequency, true);
break;
case CB_CGE_MOVE_CLONES:
for (node = id->dst_node->next_clone;
node;
node = node->next_clone)
{
edge = cgraph_edge (node, orig_stmt);
if (edge)
cgraph_set_call_stmt (edge, stmt);
}
/* FALLTHRU */
case CB_CGE_MOVE:
edge = cgraph_edge (id->dst_node, orig_stmt);
if (edge)
cgraph_set_call_stmt (edge, stmt);
break;
default:
gcc_unreachable ();
}
flags = gimple_call_flags (stmt);
if (flags & ECF_MAY_BE_ALLOCA)
cfun->calls_alloca = true;
if (flags & ECF_RETURNS_TWICE)
cfun->calls_setjmp = true;
}
/* If you think we can abort here, you are wrong.
There is no region 0 in gimple. */
gcc_assert (lookup_stmt_eh_region_fn (id->src_cfun, orig_stmt) != 0);
if (stmt_could_throw_p (stmt)
/* When we are cloning for inlining, we are supposed to
construct a clone that calls precisely the same functions
as original. However IPA optimizers might've proved
earlier some function calls as non-trapping that might
render some basic blocks dead that might become
unreachable.
We can't update SSA with unreachable blocks in CFG and thus
we prevent the scenario by preserving even the "dead" eh
edges until the point they are later removed by
fixup_cfg pass. */
|| (id->transform_call_graph_edges == CB_CGE_MOVE_CLONES
&& lookup_stmt_eh_region_fn (id->src_cfun, orig_stmt) > 0))
{
int region = lookup_stmt_eh_region_fn (id->src_cfun, orig_stmt);
/* Add an entry for the copied tree in the EH hashtable.
When cloning or versioning, use the hashtable in
cfun, and just copy the EH number. When inlining, use the
hashtable in the caller, and adjust the region number. */
if (region > 0)
add_stmt_to_eh_region (stmt, region + id->eh_region_offset);
/* If this tree doesn't have a region associated with it,
and there is a "current region,"
then associate this tree with the current region
and add edges associated with this region. */
if (lookup_stmt_eh_region_fn (id->src_cfun, orig_stmt) <= 0
&& id->eh_region > 0
&& stmt_could_throw_p (stmt))
add_stmt_to_eh_region (stmt, id->eh_region);
}
if (gimple_in_ssa_p (cfun))
{
ssa_op_iter i;
tree def;
find_new_referenced_vars (gsi_stmt (copy_gsi));
FOR_EACH_SSA_TREE_OPERAND (def, stmt, i, SSA_OP_DEF)
if (TREE_CODE (def) == SSA_NAME)
SSA_NAME_DEF_STMT (def) = stmt;
}
gsi_next (&copy_gsi);
}
while (!gsi_end_p (copy_gsi));
copy_gsi = gsi_last_bb (copy_basic_block);
}
return copy_basic_block;
}
/* Inserting Single Entry Multiple Exit region in SSA form into code in SSA
form is quite easy, since dominator relationship for old basic blocks does
not change.
There is however exception where inlining might change dominator relation
across EH edges from basic block within inlined functions destinating
to landing pads in function we inline into.
The function fills in PHI_RESULTs of such PHI nodes if they refer
to gimple regs. Otherwise, the function mark PHI_RESULT of such
PHI nodes for renaming. For non-gimple regs, renaming is safe: the
EH edges are abnormal and SSA_NAME_OCCURS_IN_ABNORMAL_PHI must be
set, and this means that there will be no overlapping live ranges
for the underlying symbol.
This might change in future if we allow redirecting of EH edges and
we might want to change way build CFG pre-inlining to include
all the possible edges then. */
static void
update_ssa_across_abnormal_edges (basic_block bb, basic_block ret_bb,
bool can_throw, bool nonlocal_goto)
{
edge e;
edge_iterator ei;
FOR_EACH_EDGE (e, ei, bb->succs)
if (!e->dest->aux
|| ((basic_block)e->dest->aux)->index == ENTRY_BLOCK)
{
gimple phi;
gimple_stmt_iterator si;
gcc_assert (e->flags & EDGE_ABNORMAL);
if (!nonlocal_goto)
gcc_assert (e->flags & EDGE_EH);
if (!can_throw)
gcc_assert (!(e->flags & EDGE_EH));
for (si = gsi_start_phis (e->dest); !gsi_end_p (si); gsi_next (&si))
{
edge re;
phi = gsi_stmt (si);
/* There shouldn't be any PHI nodes in the ENTRY_BLOCK. */
gcc_assert (!e->dest->aux);
gcc_assert (SSA_NAME_OCCURS_IN_ABNORMAL_PHI (PHI_RESULT (phi)));
if (!is_gimple_reg (PHI_RESULT (phi)))
{
mark_sym_for_renaming (SSA_NAME_VAR (PHI_RESULT (phi)));
continue;
}
re = find_edge (ret_bb, e->dest);
gcc_assert (re);
gcc_assert ((re->flags & (EDGE_EH | EDGE_ABNORMAL))
== (e->flags & (EDGE_EH | EDGE_ABNORMAL)));
SET_USE (PHI_ARG_DEF_PTR_FROM_EDGE (phi, e),
USE_FROM_PTR (PHI_ARG_DEF_PTR_FROM_EDGE (phi, re)));
}
}
}
/* Copy edges from BB into its copy constructed earlier, scale profile
accordingly. Edges will be taken care of later. Assume aux
pointers to point to the copies of each BB. */
static void
copy_edges_for_bb (basic_block bb, gcov_type count_scale, basic_block ret_bb)
{
basic_block new_bb = (basic_block) bb->aux;
edge_iterator ei;
edge old_edge;
gimple_stmt_iterator si;
int flags;
/* Use the indices from the original blocks to create edges for the
new ones. */
FOR_EACH_EDGE (old_edge, ei, bb->succs)
if (!(old_edge->flags & EDGE_EH))
{
edge new_edge;
flags = old_edge->flags;
/* Return edges do get a FALLTHRU flag when the get inlined. */
if (old_edge->dest->index == EXIT_BLOCK && !old_edge->flags
&& old_edge->dest->aux != EXIT_BLOCK_PTR)
flags |= EDGE_FALLTHRU;
new_edge = make_edge (new_bb, (basic_block) old_edge->dest->aux, flags);
new_edge->count = old_edge->count * count_scale / REG_BR_PROB_BASE;
new_edge->probability = old_edge->probability;
}
if (bb->index == ENTRY_BLOCK || bb->index == EXIT_BLOCK)
return;
for (si = gsi_start_bb (new_bb); !gsi_end_p (si);)
{
gimple copy_stmt;
bool can_throw, nonlocal_goto;
copy_stmt = gsi_stmt (si);
update_stmt (copy_stmt);
if (gimple_in_ssa_p (cfun))
mark_symbols_for_renaming (copy_stmt);
/* Do this before the possible split_block. */
gsi_next (&si);
/* If this tree could throw an exception, there are two
cases where we need to add abnormal edge(s): the
tree wasn't in a region and there is a "current
region" in the caller; or the original tree had
EH edges. In both cases split the block after the tree,
and add abnormal edge(s) as needed; we need both
those from the callee and the caller.
We check whether the copy can throw, because the const
propagation can change an INDIRECT_REF which throws
into a COMPONENT_REF which doesn't. If the copy
can throw, the original could also throw. */
can_throw = stmt_can_throw_internal (copy_stmt);
nonlocal_goto = stmt_can_make_abnormal_goto (copy_stmt);
if (can_throw || nonlocal_goto)
{
if (!gsi_end_p (si))
/* Note that bb's predecessor edges aren't necessarily
right at this point; split_block doesn't care. */
{
edge e = split_block (new_bb, copy_stmt);
new_bb = e->dest;
new_bb->aux = e->src->aux;
si = gsi_start_bb (new_bb);
}
}
if (can_throw)
make_eh_edges (copy_stmt);
if (nonlocal_goto)
make_abnormal_goto_edges (gimple_bb (copy_stmt), true);
if ((can_throw || nonlocal_goto)
&& gimple_in_ssa_p (cfun))
update_ssa_across_abnormal_edges (gimple_bb (copy_stmt), ret_bb,
can_throw, nonlocal_goto);
}
}
/* Copy the PHIs. All blocks and edges are copied, some blocks
was possibly split and new outgoing EH edges inserted.
BB points to the block of original function and AUX pointers links
the original and newly copied blocks. */
static void
copy_phis_for_bb (basic_block bb, copy_body_data *id)
{
basic_block const new_bb = (basic_block) bb->aux;
edge_iterator ei;
gimple phi;
gimple_stmt_iterator si;
for (si = gsi_start (phi_nodes (bb)); !gsi_end_p (si); gsi_next (&si))
{
tree res, new_res;
gimple new_phi;
edge new_edge;
phi = gsi_stmt (si);
res = PHI_RESULT (phi);
new_res = res;
if (is_gimple_reg (res))
{
walk_tree (&new_res, copy_tree_body_r, id, NULL);
SSA_NAME_DEF_STMT (new_res)
= new_phi = create_phi_node (new_res, new_bb);
FOR_EACH_EDGE (new_edge, ei, new_bb->preds)
{
edge const old_edge
= find_edge ((basic_block) new_edge->src->aux, bb);
tree arg = PHI_ARG_DEF_FROM_EDGE (phi, old_edge);
tree new_arg = arg;
tree block = id->block;
id->block = NULL_TREE;
walk_tree (&new_arg, copy_tree_body_r, id, NULL);
id->block = block;
gcc_assert (new_arg);
/* With return slot optimization we can end up with
non-gimple (foo *)&this->m, fix that here. */
if (TREE_CODE (new_arg) != SSA_NAME
&& TREE_CODE (new_arg) != FUNCTION_DECL
&& !is_gimple_val (new_arg))
{
gimple_seq stmts = NULL;
new_arg = force_gimple_operand (new_arg, &stmts, true, NULL);
gsi_insert_seq_on_edge_immediate (new_edge, stmts);
}
add_phi_arg (new_phi, new_arg, new_edge);
}
}
}
}
/* Wrapper for remap_decl so it can be used as a callback. */
static tree
remap_decl_1 (tree decl, void *data)
{
return remap_decl (decl, (copy_body_data *) data);
}
/* Build struct function and associated datastructures for the new clone
NEW_FNDECL to be build. CALLEE_FNDECL is the original */
static void
initialize_cfun (tree new_fndecl, tree callee_fndecl, gcov_type count,
int frequency)
{
struct function *src_cfun = DECL_STRUCT_FUNCTION (callee_fndecl);
gcov_type count_scale, frequency_scale;
if (ENTRY_BLOCK_PTR_FOR_FUNCTION (src_cfun)->count)
count_scale = (REG_BR_PROB_BASE * count
/ ENTRY_BLOCK_PTR_FOR_FUNCTION (src_cfun)->count);
else
count_scale = 1;
if (ENTRY_BLOCK_PTR_FOR_FUNCTION (src_cfun)->frequency)
frequency_scale = (REG_BR_PROB_BASE * frequency
/
ENTRY_BLOCK_PTR_FOR_FUNCTION (src_cfun)->frequency);
else
frequency_scale = count_scale;
/* Register specific tree functions. */
gimple_register_cfg_hooks ();
/* Get clean struct function. */
push_struct_function (new_fndecl);
/* We will rebuild these, so just sanity check that they are empty. */
gcc_assert (VALUE_HISTOGRAMS (cfun) == NULL);
gcc_assert (cfun->local_decls == NULL);
gcc_assert (cfun->cfg == NULL);
gcc_assert (cfun->decl == new_fndecl);
/* Copy items we preserve during clonning. */
cfun->static_chain_decl = src_cfun->static_chain_decl;
cfun->nonlocal_goto_save_area = src_cfun->nonlocal_goto_save_area;
cfun->function_end_locus = src_cfun->function_end_locus;
cfun->curr_properties = src_cfun->curr_properties;
cfun->last_verified = src_cfun->last_verified;
if (src_cfun->ipa_transforms_to_apply)
cfun->ipa_transforms_to_apply = VEC_copy (ipa_opt_pass, heap,
src_cfun->ipa_transforms_to_apply);
cfun->va_list_gpr_size = src_cfun->va_list_gpr_size;
cfun->va_list_fpr_size = src_cfun->va_list_fpr_size;
cfun->function_frequency = src_cfun->function_frequency;
cfun->has_nonlocal_label = src_cfun->has_nonlocal_label;
cfun->stdarg = src_cfun->stdarg;
cfun->dont_save_pending_sizes_p = src_cfun->dont_save_pending_sizes_p;
cfun->after_inlining = src_cfun->after_inlining;
cfun->returns_struct = src_cfun->returns_struct;
cfun->returns_pcc_struct = src_cfun->returns_pcc_struct;
cfun->after_tree_profile = src_cfun->after_tree_profile;
init_empty_tree_cfg ();
ENTRY_BLOCK_PTR->count =
(ENTRY_BLOCK_PTR_FOR_FUNCTION (src_cfun)->count * count_scale /
REG_BR_PROB_BASE);
ENTRY_BLOCK_PTR->frequency =
(ENTRY_BLOCK_PTR_FOR_FUNCTION (src_cfun)->frequency *
frequency_scale / REG_BR_PROB_BASE);
EXIT_BLOCK_PTR->count =
(EXIT_BLOCK_PTR_FOR_FUNCTION (src_cfun)->count * count_scale /
REG_BR_PROB_BASE);
EXIT_BLOCK_PTR->frequency =
(EXIT_BLOCK_PTR_FOR_FUNCTION (src_cfun)->frequency *
frequency_scale / REG_BR_PROB_BASE);
if (src_cfun->eh)
init_eh_for_function ();
if (src_cfun->gimple_df)
{
init_tree_ssa (cfun);
cfun->gimple_df->in_ssa_p = true;
init_ssa_operands ();
}
pop_cfun ();
}
/* Make a copy of the body of FN so that it can be inserted inline in
another function. Walks FN via CFG, returns new fndecl. */
static tree
copy_cfg_body (copy_body_data * id, gcov_type count, int frequency,
basic_block entry_block_map, basic_block exit_block_map)
{
tree callee_fndecl = id->src_fn;
/* Original cfun for the callee, doesn't change. */
struct function *src_cfun = DECL_STRUCT_FUNCTION (callee_fndecl);
struct function *cfun_to_copy;
basic_block bb;
tree new_fndecl = NULL;
gcov_type count_scale, frequency_scale;
int last;
if (ENTRY_BLOCK_PTR_FOR_FUNCTION (src_cfun)->count)
count_scale = (REG_BR_PROB_BASE * count
/ ENTRY_BLOCK_PTR_FOR_FUNCTION (src_cfun)->count);
else
count_scale = 1;
if (ENTRY_BLOCK_PTR_FOR_FUNCTION (src_cfun)->frequency)
frequency_scale = (REG_BR_PROB_BASE * frequency
/
ENTRY_BLOCK_PTR_FOR_FUNCTION (src_cfun)->frequency);
else
frequency_scale = count_scale;
/* Register specific tree functions. */
gimple_register_cfg_hooks ();
/* Must have a CFG here at this point. */
gcc_assert (ENTRY_BLOCK_PTR_FOR_FUNCTION
(DECL_STRUCT_FUNCTION (callee_fndecl)));
cfun_to_copy = id->src_cfun = DECL_STRUCT_FUNCTION (callee_fndecl);
ENTRY_BLOCK_PTR_FOR_FUNCTION (cfun_to_copy)->aux = entry_block_map;
EXIT_BLOCK_PTR_FOR_FUNCTION (cfun_to_copy)->aux = exit_block_map;
entry_block_map->aux = ENTRY_BLOCK_PTR_FOR_FUNCTION (cfun_to_copy);
exit_block_map->aux = EXIT_BLOCK_PTR_FOR_FUNCTION (cfun_to_copy);
/* Duplicate any exception-handling regions. */
if (cfun->eh)
{
id->eh_region_offset
= duplicate_eh_regions (cfun_to_copy, remap_decl_1, id,
0, id->eh_region);
}
/* Use aux pointers to map the original blocks to copy. */
FOR_EACH_BB_FN (bb, cfun_to_copy)
{
basic_block new_bb = copy_bb (id, bb, frequency_scale, count_scale);
bb->aux = new_bb;
new_bb->aux = bb;
}
last = last_basic_block;
/* Now that we've duplicated the blocks, duplicate their edges. */
FOR_ALL_BB_FN (bb, cfun_to_copy)
copy_edges_for_bb (bb, count_scale, exit_block_map);
if (gimple_in_ssa_p (cfun))
FOR_ALL_BB_FN (bb, cfun_to_copy)
copy_phis_for_bb (bb, id);
FOR_ALL_BB_FN (bb, cfun_to_copy)
{
((basic_block)bb->aux)->aux = NULL;
bb->aux = NULL;
}
/* Zero out AUX fields of newly created block during EH edge
insertion. */
for (; last < last_basic_block; last++)
BASIC_BLOCK (last)->aux = NULL;
entry_block_map->aux = NULL;
exit_block_map->aux = NULL;
return new_fndecl;
}
static tree
copy_body (copy_body_data *id, gcov_type count, int frequency,
basic_block entry_block_map, basic_block exit_block_map)
{
tree fndecl = id->src_fn;
tree body;
/* If this body has a CFG, walk CFG and copy. */
gcc_assert (ENTRY_BLOCK_PTR_FOR_FUNCTION (DECL_STRUCT_FUNCTION (fndecl)));
body = copy_cfg_body (id, count, frequency, entry_block_map, exit_block_map);
return body;
}
/* Return true if VALUE is an ADDR_EXPR of an automatic variable
defined in function FN, or of a data member thereof. */
static bool
self_inlining_addr_expr (tree value, tree fn)
{
tree var;
if (TREE_CODE (value) != ADDR_EXPR)
return false;
var = get_base_address (TREE_OPERAND (value, 0));
return var && auto_var_in_fn_p (var, fn);
}
static void
insert_init_stmt (basic_block bb, gimple init_stmt)
{
/* If VAR represents a zero-sized variable, it's possible that the
assignment statement may result in no gimple statements. */
if (init_stmt)
{
gimple_stmt_iterator si = gsi_last_bb (bb);
/* We can end up with init statements that store to a non-register
from a rhs with a conversion. Handle that here by forcing the
rhs into a temporary. gimple_regimplify_operands is not
prepared to do this for us. */
if (!is_gimple_reg (gimple_assign_lhs (init_stmt))
&& is_gimple_reg_type (TREE_TYPE (gimple_assign_lhs (init_stmt)))
&& gimple_assign_rhs_class (init_stmt) == GIMPLE_UNARY_RHS)
{
tree rhs = build1 (gimple_assign_rhs_code (init_stmt),
gimple_expr_type (init_stmt),
gimple_assign_rhs1 (init_stmt));
rhs = force_gimple_operand_gsi (&si, rhs, true, NULL_TREE, false,
GSI_NEW_STMT);
gimple_assign_set_rhs_code (init_stmt, TREE_CODE (rhs));
gimple_assign_set_rhs1 (init_stmt, rhs);
}
gsi_insert_after (&si, init_stmt, GSI_NEW_STMT);
gimple_regimplify_operands (init_stmt, &si);
mark_symbols_for_renaming (init_stmt);
}
}
/* Initialize parameter P with VALUE. If needed, produce init statement
at the end of BB. When BB is NULL, we return init statement to be
output later. */
static gimple
setup_one_parameter (copy_body_data *id, tree p, tree value, tree fn,
basic_block bb, tree *vars)
{
gimple init_stmt = NULL;
tree var;
tree rhs = value;
tree def = (gimple_in_ssa_p (cfun)
? gimple_default_def (id->src_cfun, p) : NULL);
if (value
&& value != error_mark_node
&& !useless_type_conversion_p (TREE_TYPE (p), TREE_TYPE (value)))
{
if (fold_convertible_p (TREE_TYPE (p), value))
rhs = fold_build1 (NOP_EXPR, TREE_TYPE (p), value);
else
/* ??? For valid (GIMPLE) programs we should not end up here.
Still if something has gone wrong and we end up with truly
mismatched types here, fall back to using a VIEW_CONVERT_EXPR
to not leak invalid GIMPLE to the following passes. */
rhs = fold_build1 (VIEW_CONVERT_EXPR, TREE_TYPE (p), value);
}
/* If the parameter is never assigned to, has no SSA_NAMEs created,
we may not need to create a new variable here at all. Instead, we may
be able to just use the argument value. */
if (TREE_READONLY (p)
&& !TREE_ADDRESSABLE (p)
&& value && !TREE_SIDE_EFFECTS (value)
&& !def)
{
/* We may produce non-gimple trees by adding NOPs or introduce
invalid sharing when operand is not really constant.
It is not big deal to prohibit constant propagation here as
we will constant propagate in DOM1 pass anyway. */
if (is_gimple_min_invariant (value)
&& useless_type_conversion_p (TREE_TYPE (p),
TREE_TYPE (value))
/* We have to be very careful about ADDR_EXPR. Make sure
the base variable isn't a local variable of the inlined
function, e.g., when doing recursive inlining, direct or
mutually-recursive or whatever, which is why we don't
just test whether fn == current_function_decl. */
&& ! self_inlining_addr_expr (value, fn))
{
insert_decl_map (id, p, value);
return NULL;
}
}
/* Make an equivalent VAR_DECL. Note that we must NOT remap the type
here since the type of this decl must be visible to the calling
function. */
var = copy_decl_to_var (p, id);
if (gimple_in_ssa_p (cfun) && TREE_CODE (var) == VAR_DECL)
{
get_var_ann (var);
add_referenced_var (var);
}
/* Register the VAR_DECL as the equivalent for the PARM_DECL;
that way, when the PARM_DECL is encountered, it will be
automatically replaced by the VAR_DECL. */
insert_decl_map (id, p, var);
/* Declare this new variable. */
TREE_CHAIN (var) = *vars;
*vars = var;
/* Make gimplifier happy about this variable. */
DECL_SEEN_IN_BIND_EXPR_P (var) = 1;
/* Even if P was TREE_READONLY, the new VAR should not be.
In the original code, we would have constructed a
temporary, and then the function body would have never
changed the value of P. However, now, we will be
constructing VAR directly. The constructor body may
change its value multiple times as it is being
constructed. Therefore, it must not be TREE_READONLY;
the back-end assumes that TREE_READONLY variable is
assigned to only once. */
if (TYPE_NEEDS_CONSTRUCTING (TREE_TYPE (p)))
TREE_READONLY (var) = 0;
/* If there is no setup required and we are in SSA, take the easy route
replacing all SSA names representing the function parameter by the
SSA name passed to function.
We need to construct map for the variable anyway as it might be used
in different SSA names when parameter is set in function.
Do replacement at -O0 for const arguments replaced by constant.
This is important for builtin_constant_p and other construct requiring
constant argument to be visible in inlined function body.
FIXME: This usually kills the last connection in between inlined
function parameter and the actual value in debug info. Can we do
better here? If we just inserted the statement, copy propagation
would kill it anyway as it always did in older versions of GCC.
We might want to introduce a notion that single SSA_NAME might
represent multiple variables for purposes of debugging. */
if (gimple_in_ssa_p (cfun) && rhs && def && is_gimple_reg (p)
&& (optimize
|| (TREE_READONLY (p)
&& is_gimple_min_invariant (rhs)))
&& (TREE_CODE (rhs) == SSA_NAME
|| is_gimple_min_invariant (rhs))
&& !SSA_NAME_OCCURS_IN_ABNORMAL_PHI (def))
{
insert_decl_map (id, def, rhs);
return NULL;
}
/* If the value of argument is never used, don't care about initializing
it. */
if (optimize && gimple_in_ssa_p (cfun) && !def && is_gimple_reg (p))
{
gcc_assert (!value || !TREE_SIDE_EFFECTS (value));
return NULL;
}
/* Initialize this VAR_DECL from the equivalent argument. Convert
the argument to the proper type in case it was promoted. */
if (value)
{
if (rhs == error_mark_node)
{
insert_decl_map (id, p, var);
return NULL;
}
STRIP_USELESS_TYPE_CONVERSION (rhs);
/* We want to use MODIFY_EXPR, not INIT_EXPR here so that we
keep our trees in gimple form. */
if (def && gimple_in_ssa_p (cfun) && is_gimple_reg (p))
{
def = remap_ssa_name (def, id);
init_stmt = gimple_build_assign (def, rhs);
SSA_NAME_IS_DEFAULT_DEF (def) = 0;
set_default_def (var, NULL);
}
else
init_stmt = gimple_build_assign (var, rhs);
if (bb && init_stmt)
insert_init_stmt (bb, init_stmt);
}
return init_stmt;
}
/* Generate code to initialize the parameters of the function at the
top of the stack in ID from the GIMPLE_CALL STMT. */
static void
initialize_inlined_parameters (copy_body_data *id, gimple stmt,
tree fn, basic_block bb)
{
tree parms;
size_t i;
tree p;
tree vars = NULL_TREE;
tree static_chain = gimple_call_chain (stmt);
/* Figure out what the parameters are. */
parms = DECL_ARGUMENTS (fn);
/* Loop through the parameter declarations, replacing each with an
equivalent VAR_DECL, appropriately initialized. */
for (p = parms, i = 0; p; p = TREE_CHAIN (p), i++)
{
tree val;
val = i < gimple_call_num_args (stmt) ? gimple_call_arg (stmt, i) : NULL;
setup_one_parameter (id, p, val, fn, bb, &vars);
}
/* Initialize the static chain. */
p = DECL_STRUCT_FUNCTION (fn)->static_chain_decl;
gcc_assert (fn != current_function_decl);
if (p)
{
/* No static chain? Seems like a bug in tree-nested.c. */
gcc_assert (static_chain);
setup_one_parameter (id, p, static_chain, fn, bb, &vars);
}
declare_inline_vars (id->block, vars);
}
/* Declare a return variable to replace the RESULT_DECL for the
function we are calling. An appropriate DECL_STMT is returned.
The USE_STMT is filled to contain a use of the declaration to
indicate the return value of the function.
RETURN_SLOT, if non-null is place where to store the result. It
is set only for CALL_EXPR_RETURN_SLOT_OPT. MODIFY_DEST, if non-null,
was the LHS of the MODIFY_EXPR to which this call is the RHS.
The return value is a (possibly null) value that is the result of the
function as seen by the callee. *USE_P is a (possibly null) value that
holds the result as seen by the caller. */
static tree
declare_return_variable (copy_body_data *id, tree return_slot, tree modify_dest,
tree *use_p)
{
tree callee = id->src_fn;
tree caller = id->dst_fn;
tree result = DECL_RESULT (callee);
tree callee_type = TREE_TYPE (result);
tree caller_type = TREE_TYPE (TREE_TYPE (callee));
tree var, use;
/* We don't need to do anything for functions that don't return
anything. */
if (!result || VOID_TYPE_P (callee_type))
{
*use_p = NULL_TREE;
return NULL_TREE;
}
/* If there was a return slot, then the return value is the
dereferenced address of that object. */
if (return_slot)
{
/* The front end shouldn't have used both return_slot and
a modify expression. */
gcc_assert (!modify_dest);
if (DECL_BY_REFERENCE (result))
{
tree return_slot_addr = build_fold_addr_expr (return_slot);
STRIP_USELESS_TYPE_CONVERSION (return_slot_addr);
/* We are going to construct *&return_slot and we can't do that
for variables believed to be not addressable.
FIXME: This check possibly can match, because values returned
via return slot optimization are not believed to have address
taken by alias analysis. */
gcc_assert (TREE_CODE (return_slot) != SSA_NAME);
if (gimple_in_ssa_p (cfun))
{
HOST_WIDE_INT bitsize;
HOST_WIDE_INT bitpos;
tree offset;
enum machine_mode mode;
int unsignedp;
int volatilep;
tree base;
base = get_inner_reference (return_slot, &bitsize, &bitpos,
&offset,
&mode, &unsignedp, &volatilep,
false);
if (TREE_CODE (base) == INDIRECT_REF)
base = TREE_OPERAND (base, 0);
if (TREE_CODE (base) == SSA_NAME)
base = SSA_NAME_VAR (base);
mark_sym_for_renaming (base);
}
var = return_slot_addr;
}
else
{
var = return_slot;
gcc_assert (TREE_CODE (var) != SSA_NAME);
TREE_ADDRESSABLE (var) |= TREE_ADDRESSABLE (result);
}
if ((TREE_CODE (TREE_TYPE (result)) == COMPLEX_TYPE
|| TREE_CODE (TREE_TYPE (result)) == VECTOR_TYPE)
&& !DECL_GIMPLE_REG_P (result)
&& DECL_P (var))
DECL_GIMPLE_REG_P (var) = 0;
use = NULL;
goto done;
}
/* All types requiring non-trivial constructors should have been handled. */
gcc_assert (!TREE_ADDRESSABLE (callee_type));
/* Attempt to avoid creating a new temporary variable. */
if (modify_dest
&& TREE_CODE (modify_dest) != SSA_NAME)
{
bool use_it = false;
/* We can't use MODIFY_DEST if there's type promotion involved. */
if (!useless_type_conversion_p (callee_type, caller_type))
use_it = false;
/* ??? If we're assigning to a variable sized type, then we must
reuse the destination variable, because we've no good way to
create variable sized temporaries at this point. */
else if (TREE_CODE (TYPE_SIZE_UNIT (caller_type)) != INTEGER_CST)
use_it = true;
/* If the callee cannot possibly modify MODIFY_DEST, then we can
reuse it as the result of the call directly. Don't do this if
it would promote MODIFY_DEST to addressable. */
else if (TREE_ADDRESSABLE (result))
use_it = false;
else
{
tree base_m = get_base_address (modify_dest);
/* If the base isn't a decl, then it's a pointer, and we don't
know where that's going to go. */
if (!DECL_P (base_m))
use_it = false;
else if (is_global_var (base_m))
use_it = false;
else if ((TREE_CODE (TREE_TYPE (result)) == COMPLEX_TYPE
|| TREE_CODE (TREE_TYPE (result)) == VECTOR_TYPE)
&& !DECL_GIMPLE_REG_P (result)
&& DECL_GIMPLE_REG_P (base_m))
use_it = false;
else if (!TREE_ADDRESSABLE (base_m))
use_it = true;
}
if (use_it)
{
var = modify_dest;
use = NULL;
goto done;
}
}
gcc_assert (TREE_CODE (TYPE_SIZE_UNIT (callee_type)) == INTEGER_CST);
var = copy_result_decl_to_var (result, id);
if (gimple_in_ssa_p (cfun))
{
get_var_ann (var);
add_referenced_var (var);
}
DECL_SEEN_IN_BIND_EXPR_P (var) = 1;
DECL_STRUCT_FUNCTION (caller)->local_decls
= tree_cons (NULL_TREE, var,
DECL_STRUCT_FUNCTION (caller)->local_decls);
/* Do not have the rest of GCC warn about this variable as it should
not be visible to the user. */
TREE_NO_WARNING (var) = 1;
declare_inline_vars (id->block, var);
/* Build the use expr. If the return type of the function was
promoted, convert it back to the expected type. */
use = var;
if (!useless_type_conversion_p (caller_type, TREE_TYPE (var)))
use = fold_convert (caller_type, var);
STRIP_USELESS_TYPE_CONVERSION (use);
if (DECL_BY_REFERENCE (result))
var = build_fold_addr_expr (var);
done:
/* Register the VAR_DECL as the equivalent for the RESULT_DECL; that
way, when the RESULT_DECL is encountered, it will be
automatically replaced by the VAR_DECL. */
insert_decl_map (id, result, var);
/* Remember this so we can ignore it in remap_decls. */
id->retvar = var;
*use_p = use;
return var;
}
/* Returns nonzero if a function can be inlined as a tree. */
bool
tree_inlinable_function_p (tree fn)
{
return inlinable_function_p (fn);
}
static const char *inline_forbidden_reason;
/* A callback for walk_gimple_seq to handle tree operands. Returns
NULL_TREE if a function can be inlined, otherwise sets the reason
why not and returns a tree representing the offending operand. */
static tree
inline_forbidden_p_op (tree *nodep, int *walk_subtrees ATTRIBUTE_UNUSED,
void *fnp ATTRIBUTE_UNUSED)
{
tree node = *nodep;
tree t;
if (TREE_CODE (node) == RECORD_TYPE || TREE_CODE (node) == UNION_TYPE)
{
/* We cannot inline a function of the form
void F (int i) { struct S { int ar[i]; } s; }
Attempting to do so produces a catch-22.
If walk_tree examines the TYPE_FIELDS chain of RECORD_TYPE/
UNION_TYPE nodes, then it goes into infinite recursion on a
structure containing a pointer to its own type. If it doesn't,
then the type node for S doesn't get adjusted properly when
F is inlined.
??? This is likely no longer true, but it's too late in the 4.0
cycle to try to find out. This should be checked for 4.1. */
for (t = TYPE_FIELDS (node); t; t = TREE_CHAIN (t))
if (variably_modified_type_p (TREE_TYPE (t), NULL))
{
inline_forbidden_reason
= G_("function %q+F can never be inlined "
"because it uses variable sized variables");
return node;
}
}
return NULL_TREE;
}
/* A callback for walk_gimple_seq to handle statements. Returns
non-NULL iff a function can not be inlined. Also sets the reason
why. */
static tree
inline_forbidden_p_stmt (gimple_stmt_iterator *gsi, bool *handled_ops_p,
struct walk_stmt_info *wip)
{
tree fn = (tree) wip->info;
tree t;
gimple stmt = gsi_stmt (*gsi);
switch (gimple_code (stmt))
{
case GIMPLE_CALL:
/* Refuse to inline alloca call unless user explicitly forced so as
this may change program's memory overhead drastically when the
function using alloca is called in loop. In GCC present in
SPEC2000 inlining into schedule_block cause it to require 2GB of
RAM instead of 256MB. */
if (gimple_alloca_call_p (stmt)
&& !lookup_attribute ("always_inline", DECL_ATTRIBUTES (fn)))
{
inline_forbidden_reason
= G_("function %q+F can never be inlined because it uses "
"alloca (override using the always_inline attribute)");
*handled_ops_p = true;
return fn;
}
t = gimple_call_fndecl (stmt);
if (t == NULL_TREE)
break;
/* We cannot inline functions that call setjmp. */
if (setjmp_call_p (t))
{
inline_forbidden_reason
= G_("function %q+F can never be inlined because it uses setjmp");
*handled_ops_p = true;
return t;
}
if (DECL_BUILT_IN_CLASS (t) == BUILT_IN_NORMAL)
switch (DECL_FUNCTION_CODE (t))
{
/* We cannot inline functions that take a variable number of
arguments. */
case BUILT_IN_VA_START:
case BUILT_IN_NEXT_ARG:
case BUILT_IN_VA_END:
inline_forbidden_reason
= G_("function %q+F can never be inlined because it "
"uses variable argument lists");
*handled_ops_p = true;
return t;
case BUILT_IN_LONGJMP:
/* We can't inline functions that call __builtin_longjmp at
all. The non-local goto machinery really requires the
destination be in a different function. If we allow the
function calling __builtin_longjmp to be inlined into the
function calling __builtin_setjmp, Things will Go Awry. */
inline_forbidden_reason
= G_("function %q+F can never be inlined because "
"it uses setjmp-longjmp exception handling");
*handled_ops_p = true;
return t;
case BUILT_IN_NONLOCAL_GOTO:
/* Similarly. */
inline_forbidden_reason
= G_("function %q+F can never be inlined because "
"it uses non-local goto");
*handled_ops_p = true;
return t;
case BUILT_IN_RETURN:
case BUILT_IN_APPLY_ARGS:
/* If a __builtin_apply_args caller would be inlined,
it would be saving arguments of the function it has
been inlined into. Similarly __builtin_return would
return from the function the inline has been inlined into. */
inline_forbidden_reason
= G_("function %q+F can never be inlined because "
"it uses __builtin_return or __builtin_apply_args");
*handled_ops_p = true;
return t;
default:
break;
}
break;
case GIMPLE_GOTO:
t = gimple_goto_dest (stmt);
/* We will not inline a function which uses computed goto. The
addresses of its local labels, which may be tucked into
global storage, are of course not constant across
instantiations, which causes unexpected behavior. */
if (TREE_CODE (t) != LABEL_DECL)
{
inline_forbidden_reason
= G_("function %q+F can never be inlined "
"because it contains a computed goto");
*handled_ops_p = true;
return t;
}
break;
case GIMPLE_LABEL:
t = gimple_label_label (stmt);
if (DECL_NONLOCAL (t))
{
/* We cannot inline a function that receives a non-local goto
because we cannot remap the destination label used in the
function that is performing the non-local goto. */
inline_forbidden_reason
= G_("function %q+F can never be inlined "
"because it receives a non-local goto");
*handled_ops_p = true;
return t;
}
break;
default:
break;
}
*handled_ops_p = false;
return NULL_TREE;
}
static tree
inline_forbidden_p_2 (tree *nodep, int *walk_subtrees,
void *fnp)
{
tree node = *nodep;
tree fn = (tree) fnp;
if (TREE_CODE (node) == LABEL_DECL && DECL_CONTEXT (node) == fn)
{
inline_forbidden_reason
= G_("function %q+F can never be inlined "
"because it saves address of local label in a static variable");
return node;
}
if (TYPE_P (node))
*walk_subtrees = 0;
return NULL_TREE;
}
/* Return true if FNDECL is a function that cannot be inlined into
another one. */
static bool
inline_forbidden_p (tree fndecl)
{
location_t saved_loc = input_location;
struct function *fun = DECL_STRUCT_FUNCTION (fndecl);
tree step;
struct walk_stmt_info wi;
struct pointer_set_t *visited_nodes;
basic_block bb;
bool forbidden_p = false;
visited_nodes = pointer_set_create ();
memset (&wi, 0, sizeof (wi));
wi.info = (void *) fndecl;
wi.pset = visited_nodes;
FOR_EACH_BB_FN (bb, fun)
{
gimple ret;
gimple_seq seq = bb_seq (bb);
ret = walk_gimple_seq (seq, inline_forbidden_p_stmt,
inline_forbidden_p_op, &wi);
forbidden_p = (ret != NULL);
if (forbidden_p)
goto egress;
}
for (step = fun->local_decls; step; step = TREE_CHAIN (step))
{
tree decl = TREE_VALUE (step);
if (TREE_CODE (decl) == VAR_DECL
&& TREE_STATIC (decl)
&& !DECL_EXTERNAL (decl)
&& DECL_INITIAL (decl))
{
tree ret;
ret = walk_tree_without_duplicates (&DECL_INITIAL (decl),
inline_forbidden_p_2, fndecl);
forbidden_p = (ret != NULL);
if (forbidden_p)
goto egress;
}
}
egress:
pointer_set_destroy (visited_nodes);
input_location = saved_loc;
return forbidden_p;
}
/* Returns nonzero if FN is a function that does not have any
fundamental inline blocking properties. */
static bool
inlinable_function_p (tree fn)
{
bool inlinable = true;
bool do_warning;
tree always_inline;
/* If we've already decided this function shouldn't be inlined,
there's no need to check again. */
if (DECL_UNINLINABLE (fn))
return false;
/* We only warn for functions declared `inline' by the user. */
do_warning = (warn_inline
&& DECL_DECLARED_INLINE_P (fn)
&& !DECL_NO_INLINE_WARNING_P (fn)
&& !DECL_IN_SYSTEM_HEADER (fn));
always_inline = lookup_attribute ("always_inline", DECL_ATTRIBUTES (fn));
if (flag_no_inline
&& always_inline == NULL)
{
if (do_warning)
warning (OPT_Winline, "function %q+F can never be inlined because it "
"is suppressed using -fno-inline", fn);
inlinable = false;
}
/* Don't auto-inline anything that might not be bound within
this unit of translation. */
else if (!DECL_DECLARED_INLINE_P (fn)
&& DECL_REPLACEABLE_P (fn))
inlinable = false;
else if (!function_attribute_inlinable_p (fn))
{
if (do_warning)
warning (OPT_Winline, "function %q+F can never be inlined because it "
"uses attributes conflicting with inlining", fn);
inlinable = false;
}
else if (inline_forbidden_p (fn))
{
/* See if we should warn about uninlinable functions. Previously,
some of these warnings would be issued while trying to expand
the function inline, but that would cause multiple warnings
about functions that would for example call alloca. But since
this a property of the function, just one warning is enough.
As a bonus we can now give more details about the reason why a
function is not inlinable. */
if (always_inline)
sorry (inline_forbidden_reason, fn);
else if (do_warning)
warning (OPT_Winline, inline_forbidden_reason, fn);
inlinable = false;
}
/* Squirrel away the result so that we don't have to check again. */
DECL_UNINLINABLE (fn) = !inlinable;
return inlinable;
}
/* Estimate the cost of a memory move. Use machine dependent
word size and take possible memcpy call into account. */
int
estimate_move_cost (tree type)
{
HOST_WIDE_INT size;
size = int_size_in_bytes (type);
if (size < 0 || size > MOVE_MAX_PIECES * MOVE_RATIO (!optimize_size))
/* Cost of a memcpy call, 3 arguments and the call. */
return 4;
else
return ((size + MOVE_MAX_PIECES - 1) / MOVE_MAX_PIECES);
}
/* Returns cost of operation CODE, according to WEIGHTS */
static int
estimate_operator_cost (enum tree_code code, eni_weights *weights)
{
switch (code)
{
/* These are "free" conversions, or their presumed cost
is folded into other operations. */
case RANGE_EXPR:
CASE_CONVERT:
case COMPLEX_EXPR:
case PAREN_EXPR:
return 0;
/* Assign cost of 1 to usual operations.
??? We may consider mapping RTL costs to this. */
case COND_EXPR:
case VEC_COND_EXPR:
case PLUS_EXPR:
case POINTER_PLUS_EXPR:
case MINUS_EXPR:
case MULT_EXPR:
case FIXED_CONVERT_EXPR:
case FIX_TRUNC_EXPR:
case NEGATE_EXPR:
case FLOAT_EXPR:
case MIN_EXPR:
case MAX_EXPR:
case ABS_EXPR:
case LSHIFT_EXPR:
case RSHIFT_EXPR:
case LROTATE_EXPR:
case RROTATE_EXPR:
case VEC_LSHIFT_EXPR:
case VEC_RSHIFT_EXPR:
case BIT_IOR_EXPR:
case BIT_XOR_EXPR:
case BIT_AND_EXPR:
case BIT_NOT_EXPR:
case TRUTH_ANDIF_EXPR:
case TRUTH_ORIF_EXPR:
case TRUTH_AND_EXPR:
case TRUTH_OR_EXPR:
case TRUTH_XOR_EXPR:
case TRUTH_NOT_EXPR:
case LT_EXPR:
case LE_EXPR:
case GT_EXPR:
case GE_EXPR:
case EQ_EXPR:
case NE_EXPR:
case ORDERED_EXPR:
case UNORDERED_EXPR:
case UNLT_EXPR:
case UNLE_EXPR:
case UNGT_EXPR:
case UNGE_EXPR:
case UNEQ_EXPR:
case LTGT_EXPR:
case CONJ_EXPR:
case PREDECREMENT_EXPR:
case PREINCREMENT_EXPR:
case POSTDECREMENT_EXPR:
case POSTINCREMENT_EXPR:
case REALIGN_LOAD_EXPR:
case REDUC_MAX_EXPR:
case REDUC_MIN_EXPR:
case REDUC_PLUS_EXPR:
case WIDEN_SUM_EXPR:
case WIDEN_MULT_EXPR:
case DOT_PROD_EXPR:
case VEC_WIDEN_MULT_HI_EXPR:
case VEC_WIDEN_MULT_LO_EXPR:
case VEC_UNPACK_HI_EXPR:
case VEC_UNPACK_LO_EXPR:
case VEC_UNPACK_FLOAT_HI_EXPR:
case VEC_UNPACK_FLOAT_LO_EXPR:
case VEC_PACK_TRUNC_EXPR:
case VEC_PACK_SAT_EXPR:
case VEC_PACK_FIX_TRUNC_EXPR:
case VEC_EXTRACT_EVEN_EXPR:
case VEC_EXTRACT_ODD_EXPR:
case VEC_INTERLEAVE_HIGH_EXPR:
case VEC_INTERLEAVE_LOW_EXPR:
return 1;
/* Few special cases of expensive operations. This is useful
to avoid inlining on functions having too many of these. */
case TRUNC_DIV_EXPR:
case CEIL_DIV_EXPR:
case FLOOR_DIV_EXPR:
case ROUND_DIV_EXPR:
case EXACT_DIV_EXPR:
case TRUNC_MOD_EXPR:
case CEIL_MOD_EXPR:
case FLOOR_MOD_EXPR:
case ROUND_MOD_EXPR:
case RDIV_EXPR:
return weights->div_mod_cost;
default:
/* We expect a copy assignment with no operator. */
gcc_assert (get_gimple_rhs_class (code) == GIMPLE_SINGLE_RHS);
return 0;
}
}
/* Estimate number of instructions that will be created by expanding
the statements in the statement sequence STMTS.
WEIGHTS contains weights attributed to various constructs. */
static
int estimate_num_insns_seq (gimple_seq stmts, eni_weights *weights)
{
int cost;
gimple_stmt_iterator gsi;
cost = 0;
for (gsi = gsi_start (stmts); !gsi_end_p (gsi); gsi_next (&gsi))
cost += estimate_num_insns (gsi_stmt (gsi), weights);
return cost;
}
/* Estimate number of instructions that will be created by expanding STMT.
WEIGHTS contains weights attributed to various constructs. */
int
estimate_num_insns (gimple stmt, eni_weights *weights)
{
unsigned cost, i;
enum gimple_code code = gimple_code (stmt);
tree lhs;
switch (code)
{
case GIMPLE_ASSIGN:
/* Try to estimate the cost of assignments. We have three cases to
deal with:
1) Simple assignments to registers;
2) Stores to things that must live in memory. This includes
"normal" stores to scalars, but also assignments of large
structures, or constructors of big arrays;
Let us look at the first two cases, assuming we have "a = b + C":
<GIMPLE_ASSIGN <var_decl "a">