-/* Callgraph based intraprocedural optimizations.
- Copyright (C) 2003, 2004, 2005 Free Software Foundation, Inc.
+/* Callgraph based interprocedural optimizations.
+ Copyright (C) 2003, 2004, 2005, 2006 Free Software Foundation, Inc.
Contributed by Jan Hubicka
This file is part of GCC.
You should have received a copy of the GNU General Public License
along with GCC; see the file COPYING. If not, write to the Free
-Software Foundation, 59 Temple Place - Suite 330, Boston, MA
-02111-1307, USA. */
+Software Foundation, 51 Franklin Street, Fifth Floor, Boston, MA
+02110-1301, USA. */
/* This module implements main driver of compilation process as well as
- few basic intraprocedural optimizers.
+ few basic interprocedural optimizers.
The main scope of this file is to act as an interface in between
tree based frontends and the backend (and middle end)
This function is called once front-end has parsed whole body of function
and it is certain that the function body nor the declaration will change.
- (There is one exception needed for implementing GCC extern inline function.)
+ (There is one exception needed for implementing GCC extern inline
+ function.)
- - cgraph_varpool_finalize_variable
+ - varpool_finalize_variable
This function has same behavior as the above but is used for static
variables.
- cgraph_finalize_compilation_unit
- This function is called once compilation unit is finalized and it will
- no longer change.
+ This function is called once (source level) compilation unit is finalized
+ and it will no longer change.
In the unit-at-a-time the call-graph construction and local function
analysis takes place here. Bodies of unreachable functions are released
to conserve memory usage.
- ??? The compilation unit in this point of view should be compilation
- unit as defined by the language - for instance C frontend allows multiple
- compilation units to be parsed at once and it should call function each
- time parsing is done so we save memory.
+ The function can be called multiple times when multiple source level
+ compilation units are combined (such as in C frontend)
- cgraph_optimize
taken are marked as local. Backend can then use this information to
modify calling conventions, do better inlining or similar optimizations.
- - cgraph_assemble_pending_functions
- - cgraph_varpool_assemble_pending_variables
-
- In non-unit-at-a-time mode these functions can be used to force compilation
- of functions or variables that are known to be needed at given stage
- of compilation
-
- cgraph_mark_needed_node
- - cgraph_varpool_mark_needed_node
+ - varpool_mark_needed_node
- When function or variable is referenced by some hidden way (for instance
- via assembly code and marked by attribute "used"), the call-graph data structure
- must be updated accordingly by this function.
+ When function or variable is referenced by some hidden way the call-graph
+ data structure must be updated accordingly by this function.
+ There should be little need to call this function and all the references
+ should be made explicit to cgraph code. At present these functions are
+ used by C++ frontend to explicitly mark the keyed methods.
- analyze_expr callback
This function is used to expand function and pass it into RTL back-end.
Front-end should not make any assumptions about when this function can be
called. In particular cgraph_assemble_pending_functions,
- cgraph_varpool_assemble_pending_variables, cgraph_finalize_function,
- cgraph_varpool_finalize_function, cgraph_optimize can cause arbitrarily
+ varpool_assemble_pending_variables, cgraph_finalize_function,
+ varpool_finalize_function, cgraph_optimize can cause arbitrarily
previously finalized functions to be expanded.
We implement two compilation modes.
decision on whether function is needed is made more conservative so
uninlininable static functions are needed too. During the call-graph
construction the edge destinations are not marked as reachable and it
- is completely relied upn assemble_variable to mark them.
-
- Inlining decision heuristics
- ??? Move this to separate file after tree-ssa merge.
-
- We separate inlining decisions from the inliner itself and store it
- inside callgraph as so called inline plan. Refer to cgraph.c
- documentation about particular representation of inline plans in the
- callgraph
-
- The implementation of particular heuristics is separated from
- the rest of code to make it easier to replace it with more complicated
- implementation in the future. The rest of inlining code acts as a
- library aimed to modify the callgraph and verify that the parameters
- on code size growth fits.
-
- To mark given call inline, use cgraph_mark_inline function, the
- verification is performed by cgraph_default_inline_p and
- cgraph_check_inline_limits.
-
- The heuristics implements simple knapsack style algorithm ordering
- all functions by their "profitability" (estimated by code size growth)
- and inlining them in priority order.
-
- cgraph_decide_inlining implements heuristics taking whole callgraph
- into account, while cgraph_decide_inlining_incrementally considers
- only one function at a time and is used in non-unit-at-a-time mode. */
+ is completely relied upn assemble_variable to mark them. */
#include "config.h"
#include "c-common.h"
#include "intl.h"
#include "function.h"
+#include "ipa-prop.h"
#include "tree-gimple.h"
-
-#define INSNS_PER_CALL 10
+#include "tree-pass.h"
+#include "output.h"
static void cgraph_expand_all_functions (void);
static void cgraph_mark_functions_to_output (void);
static void cgraph_expand_function (struct cgraph_node *);
-static tree record_call_1 (tree *, int *, void *);
-static void cgraph_mark_local_functions (void);
-static bool cgraph_default_inline_p (struct cgraph_node *n);
-static void cgraph_analyze_function (struct cgraph_node *node);
-static void cgraph_decide_inlining_incrementally (struct cgraph_node *);
-
-/* Statistics we collect about inlining algorithm. */
-static int ncalls_inlined;
-static int nfunctions_inlined;
-static int initial_insns;
-static int overall_insns;
-
-/* Records tree nodes seen in cgraph_create_edges. Simply using
+static tree record_reference (tree *, int *, void *);
+static void cgraph_output_pending_asms (void);
+static void cgraph_increase_alignment (void);
+static void initialize_inline_failed (struct cgraph_node *);
+
+/* Records tree nodes seen in record_reference. Simply using
walk_tree_without_duplicates doesn't guarantee each node is visited
once because it gets a new htab upon each recursive call from
- record_calls_1. */
+ record_reference itself. */
static struct pointer_set_t *visited_nodes;
static FILE *cgraph_dump_file;
decide_is_function_needed (struct cgraph_node *node, tree decl)
{
tree origin;
+ if (MAIN_NAME_P (DECL_NAME (decl))
+ && TREE_PUBLIC (decl))
+ {
+ node->local.externally_visible = true;
+ return true;
+ }
+
+ /* If the user told us it is used, then it must be so. */
+ if (node->local.externally_visible)
+ return true;
+
+ if (!flag_unit_at_a_time && lookup_attribute ("used", DECL_ATTRIBUTES (decl)))
+ return true;
+
+ /* ??? If the assembler name is set by hand, it is possible to assemble
+ the name later after finalizing the function and the fact is noticed
+ in assemble_name then. This is arguably a bug. */
+ if (DECL_ASSEMBLER_NAME_SET_P (decl)
+ && TREE_SYMBOL_REFERENCED (DECL_ASSEMBLER_NAME (decl)))
+ return true;
/* If we decided it was needed before, but at the time we didn't have
the body of the function available, then it's still needed. We have
return true;
/* Externally visible functions must be output. The exception is
- COMDAT functions that must be output only when they are needed. */
- if (TREE_PUBLIC (decl) && !DECL_COMDAT (decl) && !DECL_EXTERNAL (decl))
+ COMDAT functions that must be output only when they are needed.
+
+ When not optimizing, also output the static functions. (see
+ PR24561), but don't do so for always_inline functions, functions
+ declared inline and nested functions. These was optimized out
+ in the original implementation and it is unclear whether we want
+ to change the behavior here. */
+ if (((TREE_PUBLIC (decl)
+ || (!optimize && !node->local.disregard_inline_limits
+ && !DECL_DECLARED_INLINE_P (decl)
+ && !node->origin))
+ && !flag_whole_program)
+ && !DECL_COMDAT (decl) && !DECL_EXTERNAL (decl))
return true;
/* Constructors and destructors are reachable from the runtime by
if (DECL_STATIC_CONSTRUCTOR (decl) || DECL_STATIC_DESTRUCTOR (decl))
return true;
- /* If the user told us it is used, then it must be so. */
- if (lookup_attribute ("used", DECL_ATTRIBUTES (decl)))
- return true;
-
- /* ??? If the assembler name is set by hand, it is possible to assemble
- the name later after finalizing the function and the fact is noticed
- in assemble_name then. This is arguably a bug. */
- if (DECL_ASSEMBLER_NAME_SET_P (decl)
- && TREE_SYMBOL_REFERENCED (DECL_ASSEMBLER_NAME (decl)))
- return true;
-
if (flag_unit_at_a_time)
return false;
|| (!node->local.disregard_inline_limits
/* When declared inline, defer even the uninlinable functions.
This allows them to be eliminated when unused. */
- && !DECL_DECLARED_INLINE_P (decl)
- && (!node->local.inlinable || !cgraph_default_inline_p (node))))
+ && !DECL_DECLARED_INLINE_P (decl)
+ && (!node->local.inlinable || !cgraph_default_inline_p (node, NULL))))
return true;
return false;
}
+/* Process CGRAPH_NEW_FUNCTIONS and perform actions neccesary to add these
+ functions into callgraph in a way so they look like ordinary reachable
+ functions inserted into callgraph already at construction time. */
+
+bool
+cgraph_process_new_functions (void)
+{
+ bool output = false;
+ tree fndecl;
+ struct cgraph_node *node;
+
+ /* Note that this queue may grow as its being processed, as the new
+ functions may generate new ones. */
+ while (cgraph_new_nodes)
+ {
+ node = cgraph_new_nodes;
+ fndecl = node->decl;
+ cgraph_new_nodes = cgraph_new_nodes->next_needed;
+ switch (cgraph_state)
+ {
+ case CGRAPH_STATE_CONSTRUCTION:
+ /* At construction time we just need to finalize function and move
+ it into reachable functions list. */
+
+ node->next_needed = NULL;
+ cgraph_finalize_function (fndecl, false);
+ cgraph_mark_reachable_node (node);
+ output = true;
+ break;
+
+ case CGRAPH_STATE_IPA:
+ /* When IPA optimization already started, do all essential
+ transformations that has been already performed on the whole
+ cgraph but not on this function. */
+
+ tree_register_cfg_hooks ();
+ if (!node->analyzed)
+ cgraph_analyze_function (node);
+ push_cfun (DECL_STRUCT_FUNCTION (fndecl));
+ current_function_decl = fndecl;
+ node->local.inlinable = tree_inlinable_function_p (fndecl);
+ node->local.self_insns = estimate_num_insns (fndecl);
+ node->local.disregard_inline_limits
+ = lang_hooks.tree_inlining.disregard_inline_limits (fndecl);
+ /* Inlining characteristics are maintained by the
+ cgraph_mark_inline. */
+ node->global.insns = node->local.self_insns;
+ initialize_inline_failed (node);
+ if (flag_really_no_inline && !node->local.disregard_inline_limits)
+ node->local.inlinable = 0;
+ free_dominance_info (CDI_POST_DOMINATORS);
+ free_dominance_info (CDI_DOMINATORS);
+ pop_cfun ();
+ current_function_decl = NULL;
+ break;
+
+ case CGRAPH_STATE_EXPANSION:
+ /* Functions created during expansion shall be compiled
+ directly. */
+ node->output = 0;
+ cgraph_expand_function (node);
+ break;
+ default:
+ gcc_unreachable ();
+ break;
+ }
+ }
+ return output;
+}
/* When not doing unit-at-a-time, output all functions enqueued.
Return true when such a functions were found. */
-bool
+static bool
cgraph_assemble_pending_functions (void)
{
bool output = false;
if (flag_unit_at_a_time)
return false;
+ cgraph_output_pending_asms ();
+
while (cgraph_nodes_queue)
{
struct cgraph_node *n = cgraph_nodes_queue;
cgraph_nodes_queue = cgraph_nodes_queue->next_needed;
n->next_needed = NULL;
- if (!n->global.inlined_to && !DECL_EXTERNAL (n->decl))
+ if (!n->global.inlined_to
+ && !n->alias
+ && !DECL_EXTERNAL (n->decl))
{
cgraph_expand_function (n);
output = true;
}
+ output |= cgraph_process_new_functions ();
}
return output;
}
-/* DECL has been parsed. Take it, queue it, compile it at the whim of the
- logic in effect. If NESTED is true, then our caller cannot stand to have
- the garbage collector run at the moment. We would need to either create
- a new GC context, or just not compile right now. */
-void
-cgraph_finalize_function (tree decl, bool nested)
+/* As an GCC extension we allow redefinition of the function. The
+ semantics when both copies of bodies differ is not well defined.
+ We replace the old body with new body so in unit at a time mode
+ we always use new body, while in normal mode we may end up with
+ old body inlined into some functions and new body expanded and
+ inlined in others.
+
+ ??? It may make more sense to use one body for inlining and other
+ body for expanding the function but this is difficult to do. */
+
+static void
+cgraph_reset_node (struct cgraph_node *node)
{
- struct cgraph_node *node = cgraph_node (decl);
+ /* If node->output is set, then this is a unit-at-a-time compilation
+ and we have already begun whole-unit analysis. This is *not*
+ testing for whether we've already emitted the function. That
+ case can be sort-of legitimately seen with real function
+ redefinition errors. I would argue that the front end should
+ never present us with such a case, but don't enforce that for now. */
+ gcc_assert (!node->output);
+
+ /* Reset our data structures so we can analyze the function again. */
+ memset (&node->local, 0, sizeof (node->local));
+ memset (&node->global, 0, sizeof (node->global));
+ memset (&node->rtl, 0, sizeof (node->rtl));
+ node->analyzed = false;
+ node->local.redefined_extern_inline = true;
+ node->local.finalized = false;
- if (node->local.finalized)
+ if (!flag_unit_at_a_time)
{
- /* As an GCC extension we allow redefinition of the function. The
- semantics when both copies of bodies differ is not well defined.
- We replace the old body with new body so in unit at a time mode
- we always use new body, while in normal mode we may end up with
- old body inlined into some functions and new body expanded and
- inlined in others.
-
- ??? It may make more sense to use one body for inlining and other
- body for expanding the function but this is difficult to do. */
-
- /* If node->output is set, then this is a unit-at-a-time compilation
- and we have already begun whole-unit analysis. This is *not*
- testing for whether we've already emitted the function. That
- case can be sort-of legitimately seen with real function
- redefinition errors. I would argue that the front end should
- never present us with such a case, but don't enforce that for now. */
- gcc_assert (!node->output);
+ struct cgraph_node *n, *next;
- /* Reset our data structures so we can analyze the function again. */
- memset (&node->local, 0, sizeof (node->local));
- memset (&node->global, 0, sizeof (node->global));
- memset (&node->rtl, 0, sizeof (node->rtl));
- node->analyzed = false;
- node->local.redefined_extern_inline = true;
-
- if (!flag_unit_at_a_time)
+ for (n = cgraph_nodes; n; n = next)
{
- struct cgraph_node *n;
-
- for (n = cgraph_nodes; n; n = n->next)
- if (n->global.inlined_to == node)
- cgraph_remove_node (n);
+ next = n->next;
+ if (n->global.inlined_to == node)
+ cgraph_remove_node (n);
}
+ }
- cgraph_node_remove_callees (node);
+ cgraph_node_remove_callees (node);
- /* We may need to re-queue the node for assembling in case
- we already proceeded it and ignored as not needed. */
- if (node->reachable && !flag_unit_at_a_time)
- {
- struct cgraph_node *n;
+ /* We may need to re-queue the node for assembling in case
+ we already proceeded it and ignored as not needed. */
+ if (node->reachable && !flag_unit_at_a_time)
+ {
+ struct cgraph_node *n;
- for (n = cgraph_nodes_queue; n; n = n->next_needed)
- if (n == node)
- break;
- if (!n)
- node->reachable = 0;
- }
+ for (n = cgraph_nodes_queue; n; n = n->next_needed)
+ if (n == node)
+ break;
+ if (!n)
+ node->reachable = 0;
}
+}
+
+static void
+cgraph_lower_function (struct cgraph_node *node)
+{
+ if (node->lowered)
+ return;
+ tree_lowering_passes (node->decl);
+ node->lowered = true;
+}
+
+/* DECL has been parsed. Take it, queue it, compile it at the whim of the
+ logic in effect. If NESTED is true, then our caller cannot stand to have
+ the garbage collector run at the moment. We would need to either create
+ a new GC context, or just not compile right now. */
+
+void
+cgraph_finalize_function (tree decl, bool nested)
+{
+ struct cgraph_node *node = cgraph_node (decl);
+
+ if (node->local.finalized)
+ cgraph_reset_node (node);
notice_global_symbol (decl);
node->decl = decl;
node->local.finalized = true;
+ node->lowered = DECL_STRUCT_FUNCTION (decl)->cfg != NULL;
if (node->nested)
lower_nested_functions (decl);
gcc_assert (!node->nested);
if (!flag_unit_at_a_time)
{
cgraph_analyze_function (node);
- cgraph_decide_inlining_incrementally (node);
+ cgraph_decide_inlining_incrementally (node, false);
}
if (decide_is_function_needed (node, decl))
cgraph_mark_needed_node (node);
+ /* Since we reclaim unreachable nodes at the end of every language
+ level unit, we need to be conservative about possible entry points
+ there. */
+ if ((TREE_PUBLIC (decl) && !DECL_COMDAT (decl) && !DECL_EXTERNAL (decl)))
+ cgraph_mark_reachable_node (node);
+
/* If not unit at a time, go ahead and emit everything we've found
to be reachable at this time. */
if (!nested)
/* Walk tree and record all calls. Called via walk_tree. */
static tree
-record_call_1 (tree *tp, int *walk_subtrees, void *data)
+record_reference (tree *tp, int *walk_subtrees, void *data)
{
tree t = *tp;
/* ??? Really, we should mark this decl as *potentially* referenced
by this function and re-examine whether the decl is actually used
after rtl has been generated. */
- if (TREE_STATIC (t))
+ if (TREE_STATIC (t) || DECL_EXTERNAL (t))
{
- cgraph_varpool_mark_needed_node (cgraph_varpool_node (t));
+ varpool_mark_needed_node (varpool_node (t));
if (lang_hooks.callgraph.analyze_expr)
- return lang_hooks.callgraph.analyze_expr (tp, walk_subtrees,
+ return lang_hooks.callgraph.analyze_expr (tp, walk_subtrees,
data);
}
break;
+ case FDESC_EXPR:
case ADDR_EXPR:
if (flag_unit_at_a_time)
{
}
break;
- case CALL_EXPR:
- {
- tree decl = get_callee_fndecl (*tp);
- if (decl && TREE_CODE (decl) == FUNCTION_DECL)
- {
- cgraph_create_edge (data, cgraph_node (decl), *tp);
-
- /* When we see a function call, we don't want to look at the
- function reference in the ADDR_EXPR that is hanging from
- the CALL_EXPR we're examining here, because we would
- conclude incorrectly that the function's address could be
- taken by something that is not a function call. So only
- walk the function parameter list, skip the other subtrees. */
-
- walk_tree (&TREE_OPERAND (*tp, 1), record_call_1, data,
- visited_nodes);
- *walk_subtrees = 0;
- }
- break;
- }
-
default:
/* Save some cycles by not walking types and declaration as we
won't find anything useful there anyway. */
/* Create cgraph edges for function calls inside BODY from NODE. */
-void
+static void
cgraph_create_edges (struct cgraph_node *node, tree body)
{
- /* The nodes we're interested in are never shared, so walk
- the tree ignoring duplicates. */
+ basic_block bb;
+
+ struct function *this_cfun = DECL_STRUCT_FUNCTION (body);
+ block_stmt_iterator bsi;
+ tree step;
visited_nodes = pointer_set_create ();
- walk_tree (&body, record_call_1, node, visited_nodes);
+
+ /* Reach the trees by walking over the CFG, and note the
+ enclosing basic-blocks in the call edges. */
+ FOR_EACH_BB_FN (bb, this_cfun)
+ for (bsi = bsi_start (bb); !bsi_end_p (bsi); bsi_next (&bsi))
+ {
+ tree stmt = bsi_stmt (bsi);
+ tree call = get_call_expr_in (stmt);
+ tree decl;
+
+ if (call && (decl = get_callee_fndecl (call)))
+ {
+ cgraph_create_edge (node, cgraph_node (decl), stmt,
+ bb->count,
+ bb->loop_depth);
+ walk_tree (&TREE_OPERAND (call, 1),
+ record_reference, node, visited_nodes);
+ if (TREE_CODE (stmt) == GIMPLE_MODIFY_STMT)
+ walk_tree (&GIMPLE_STMT_OPERAND (stmt, 0),
+ record_reference, node, visited_nodes);
+ }
+ else
+ walk_tree (bsi_stmt_ptr (bsi), record_reference, node, visited_nodes);
+ }
+
+ /* Look for initializers of constant variables and private statics. */
+ for (step = DECL_STRUCT_FUNCTION (body)->unexpanded_var_list;
+ step;
+ step = TREE_CHAIN (step))
+ {
+ tree decl = TREE_VALUE (step);
+ if (TREE_CODE (decl) == VAR_DECL
+ && (TREE_STATIC (decl) && !DECL_EXTERNAL (decl))
+ && flag_unit_at_a_time)
+ varpool_finalize_decl (decl);
+ else if (TREE_CODE (decl) == VAR_DECL && DECL_INITIAL (decl))
+ walk_tree (&DECL_INITIAL (decl), record_reference, node, visited_nodes);
+ }
+
pointer_set_destroy (visited_nodes);
visited_nodes = NULL;
}
-static bool error_found;
+void
+record_references_in_initializer (tree decl)
+{
+ visited_nodes = pointer_set_create ();
+ walk_tree (&DECL_INITIAL (decl), record_reference, NULL, visited_nodes);
+ pointer_set_destroy (visited_nodes);
+ visited_nodes = NULL;
+}
-/* Callback of verify_cgraph_node. Check that all call_exprs have
- cgraph nodes. */
-static tree
-verify_cgraph_node_1 (tree *tp, int *walk_subtrees, void *data)
+/* Give initial reasons why inlining would fail. Those gets
+ either NULLified or usually overwritten by more precise reason
+ later. */
+static void
+initialize_inline_failed (struct cgraph_node *node)
{
- tree t = *tp;
- tree decl;
+ struct cgraph_edge *e;
- if (TREE_CODE (t) == CALL_EXPR && (decl = get_callee_fndecl (t)))
+ for (e = node->callers; e; e = e->next_caller)
{
- struct cgraph_edge *e = cgraph_edge (data, t);
- if (e)
- {
- if (e->aux)
- {
- error ("Shared call_expr:");
- debug_tree (t);
- error_found = true;
- }
- if (e->callee->decl != cgraph_node (decl)->decl)
- {
- error ("Edge points to wrong declaration:");
- debug_tree (e->callee->decl);
- fprintf (stderr," Instead of:");
- debug_tree (decl);
- }
- e->aux = (void *)1;
- }
+ gcc_assert (!e->callee->global.inlined_to);
+ gcc_assert (e->inline_failed);
+ if (node->local.redefined_extern_inline)
+ e->inline_failed = N_("redefined extern inline functions are not "
+ "considered for inlining");
+ else if (!node->local.inlinable)
+ e->inline_failed = N_("function not inlinable");
else
- {
- error ("Missing callgraph edge for call expr:");
- debug_tree (t);
- error_found = true;
- }
+ e->inline_failed = N_("function not considered for inlining");
}
+}
+
+/* Rebuild call edges from current function after a passes not aware
+ of cgraph updating. */
+static unsigned int
+rebuild_cgraph_edges (void)
+{
+ basic_block bb;
+ struct cgraph_node *node = cgraph_node (current_function_decl);
+ block_stmt_iterator bsi;
+
+ cgraph_node_remove_callees (node);
- /* Save some cycles by not walking types and declaration as we
- won't find anything useful there anyway. */
- if (IS_TYPE_OR_DECL_P (*tp))
- *walk_subtrees = 0;
+ node->count = ENTRY_BLOCK_PTR->count;
- return NULL_TREE;
+ FOR_EACH_BB (bb)
+ for (bsi = bsi_start (bb); !bsi_end_p (bsi); bsi_next (&bsi))
+ {
+ tree stmt = bsi_stmt (bsi);
+ tree call = get_call_expr_in (stmt);
+ tree decl;
+
+ if (call && (decl = get_callee_fndecl (call)))
+ cgraph_create_edge (node, cgraph_node (decl), stmt,
+ bb->count,
+ bb->loop_depth);
+ }
+ initialize_inline_failed (node);
+ gcc_assert (!node->global.inlined_to);
+ return 0;
}
+struct tree_opt_pass pass_rebuild_cgraph_edges =
+{
+ NULL, /* name */
+ NULL, /* gate */
+ rebuild_cgraph_edges, /* execute */
+ NULL, /* sub */
+ NULL, /* next */
+ 0, /* static_pass_number */
+ 0, /* tv_id */
+ PROP_cfg, /* properties_required */
+ 0, /* properties_provided */
+ 0, /* properties_destroyed */
+ 0, /* todo_flags_start */
+ 0, /* todo_flags_finish */
+ 0 /* letter */
+};
+
/* Verify cgraph nodes of given cgraph node. */
void
verify_cgraph_node (struct cgraph_node *node)
{
struct cgraph_edge *e;
struct cgraph_node *main_clone;
+ struct function *this_cfun = DECL_STRUCT_FUNCTION (node->decl);
+ basic_block this_block;
+ block_stmt_iterator bsi;
+ bool error_found = false;
+
+ if (errorcount || sorrycount)
+ return;
timevar_push (TV_CGRAPH_VERIFY);
- error_found = false;
for (e = node->callees; e; e = e->next_callee)
if (e->aux)
{
- error ("Aux field set for edge %s->%s",
+ error ("aux field set for edge %s->%s",
cgraph_node_name (e->caller), cgraph_node_name (e->callee));
error_found = true;
}
+ if (node->count < 0)
+ {
+ error ("Execution count is negative");
+ error_found = true;
+ }
for (e = node->callers; e; e = e->next_caller)
{
+ if (e->count < 0)
+ {
+ error ("caller edge count is negative");
+ error_found = true;
+ }
if (!e->inline_failed)
{
if (node->global.inlined_to
!= (e->caller->global.inlined_to
? e->caller->global.inlined_to : e->caller))
{
- error ("Inlined_to pointer is wrong");
+ error ("inlined_to pointer is wrong");
error_found = true;
}
if (node->callers->next_caller)
{
- error ("Multiple inline callers");
+ error ("multiple inline callers");
error_found = true;
}
}
else
if (node->global.inlined_to)
{
- error ("Inlined_to pointer set for noninline callers");
+ error ("inlined_to pointer set for noninline callers");
error_found = true;
}
}
if (!node->callers && node->global.inlined_to)
{
- error ("Inlined_to pointer is set but no predecesors found");
+ error ("inlined_to pointer is set but no predecessors found");
error_found = true;
}
if (node->global.inlined_to == node)
{
- error ("Inlined_to pointer reffers to itself");
+ error ("inlined_to pointer refers to itself");
error_found = true;
}
main_clone = main_clone->next_clone)
if (main_clone == node)
break;
- if (!node)
+ if (!cgraph_node (node->decl))
{
- error ("Node not found in DECL_ASSEMBLER_NAME hash");
+ error ("node not found in cgraph_hash");
error_found = true;
}
-
+
if (node->analyzed
&& DECL_SAVED_TREE (node->decl) && !TREE_ASM_WRITTEN (node->decl)
&& (!DECL_EXTERNAL (node->decl) || node->global.inlined_to))
{
- walk_tree_without_duplicates (&DECL_SAVED_TREE (node->decl),
- verify_cgraph_node_1, node);
+ if (this_cfun->cfg)
+ {
+ /* The nodes we're interested in are never shared, so walk
+ the tree ignoring duplicates. */
+ visited_nodes = pointer_set_create ();
+ /* Reach the trees by walking over the CFG, and note the
+ enclosing basic-blocks in the call edges. */
+ FOR_EACH_BB_FN (this_block, this_cfun)
+ for (bsi = bsi_start (this_block); !bsi_end_p (bsi); bsi_next (&bsi))
+ {
+ tree stmt = bsi_stmt (bsi);
+ tree call = get_call_expr_in (stmt);
+ tree decl;
+ if (call && (decl = get_callee_fndecl (call)))
+ {
+ struct cgraph_edge *e = cgraph_edge (node, stmt);
+ if (e)
+ {
+ if (e->aux)
+ {
+ error ("shared call_stmt:");
+ debug_generic_stmt (stmt);
+ error_found = true;
+ }
+ if (e->callee->decl != cgraph_node (decl)->decl
+ && e->inline_failed)
+ {
+ error ("edge points to wrong declaration:");
+ debug_tree (e->callee->decl);
+ fprintf (stderr," Instead of:");
+ debug_tree (decl);
+ }
+ e->aux = (void *)1;
+ }
+ else
+ {
+ error ("missing callgraph edge for call stmt:");
+ debug_generic_stmt (stmt);
+ error_found = true;
+ }
+ }
+ }
+ pointer_set_destroy (visited_nodes);
+ visited_nodes = NULL;
+ }
+ else
+ /* No CFG available?! */
+ gcc_unreachable ();
+
for (e = node->callees; e; e = e->next_callee)
{
if (!e->aux)
{
- error ("Edge %s->%s has no corresponding call_expr",
+ error ("edge %s->%s has no corresponding call_stmt",
cgraph_node_name (e->caller),
cgraph_node_name (e->callee));
+ debug_generic_stmt (e->call_stmt);
error_found = true;
}
e->aux = 0;
if (error_found)
{
dump_cgraph_node (stderr, node);
- internal_error ("verify_cgraph_node failed.");
+ internal_error ("verify_cgraph_node failed");
}
timevar_pop (TV_CGRAPH_VERIFY);
}
verify_cgraph_node (node);
}
-/* Analyze the function scheduled to be output. */
+/* Output all asm statements we have stored up to be output. */
+
static void
+cgraph_output_pending_asms (void)
+{
+ struct cgraph_asm_node *can;
+
+ if (errorcount || sorrycount)
+ return;
+
+ for (can = cgraph_asm_nodes; can; can = can->next)
+ assemble_asm (can->asm_str);
+ cgraph_asm_nodes = NULL;
+}
+
+/* Analyze the function scheduled to be output. */
+void
cgraph_analyze_function (struct cgraph_node *node)
{
tree decl = node->decl;
- struct cgraph_edge *e;
current_function_decl = decl;
+ push_cfun (DECL_STRUCT_FUNCTION (decl));
+ cgraph_lower_function (node);
/* First kill forward declaration so reverse inlining works properly. */
- cgraph_create_edges (node, DECL_SAVED_TREE (decl));
+ cgraph_create_edges (node, decl);
+ node->local.estimated_self_stack_size = estimated_stack_frame_size ();
+ node->global.estimated_stack_size = node->local.estimated_self_stack_size;
+ node->global.stack_frame_offset = 0;
node->local.inlinable = tree_inlinable_function_p (decl);
- node->local.self_insns = estimate_num_insns (DECL_SAVED_TREE (decl));
+ if (!flag_unit_at_a_time)
+ node->local.self_insns = estimate_num_insns (decl);
if (node->local.inlinable)
node->local.disregard_inline_limits
= lang_hooks.tree_inlining.disregard_inline_limits (decl);
- for (e = node->callers; e; e = e->next_caller)
- {
- if (node->local.redefined_extern_inline)
- e->inline_failed = N_("redefined extern inline functions are not "
- "considered for inlining");
- else if (!node->local.inlinable)
- e->inline_failed = N_("function not inlinable");
- else
- e->inline_failed = N_("function not considered for inlining");
- }
+ initialize_inline_failed (node);
if (flag_really_no_inline && !node->local.disregard_inline_limits)
node->local.inlinable = 0;
/* Inlining characteristics are maintained by the cgraph_mark_inline. */
node->global.insns = node->local.self_insns;
node->analyzed = true;
+ pop_cfun ();
current_function_decl = NULL;
}
+/* Look for externally_visible and used attributes and mark cgraph nodes
+ accordingly.
+
+ We cannot mark the nodes at the point the attributes are processed (in
+ handle_*_attribute) because the copy of the declarations available at that
+ point may not be canonical. For example, in:
+
+ void f();
+ void f() __attribute__((used));
+
+ the declaration we see in handle_used_attribute will be the second
+ declaration -- but the front end will subsequently merge that declaration
+ with the original declaration and discard the second declaration.
+
+ Furthermore, we can't mark these nodes in cgraph_finalize_function because:
+
+ void f() {}
+ void f() __attribute__((externally_visible));
+
+ is valid.
+
+ So, we walk the nodes at the end of the translation unit, applying the
+ attributes at that point. */
+
+static void
+process_function_and_variable_attributes (struct cgraph_node *first,
+ struct varpool_node *first_var)
+{
+ struct cgraph_node *node;
+ struct varpool_node *vnode;
+
+ for (node = cgraph_nodes; node != first; node = node->next)
+ {
+ tree decl = node->decl;
+ if (lookup_attribute ("used", DECL_ATTRIBUTES (decl)))
+ {
+ mark_decl_referenced (decl);
+ if (node->local.finalized)
+ cgraph_mark_needed_node (node);
+ }
+ if (lookup_attribute ("externally_visible", DECL_ATTRIBUTES (decl)))
+ {
+ if (! TREE_PUBLIC (node->decl))
+ warning (OPT_Wattributes,
+ "%J%<externally_visible%> attribute have effect only on public objects",
+ node->decl);
+ else
+ {
+ if (node->local.finalized)
+ cgraph_mark_needed_node (node);
+ node->local.externally_visible = true;
+ }
+ }
+ }
+ for (vnode = varpool_nodes; vnode != first_var; vnode = vnode->next)
+ {
+ tree decl = vnode->decl;
+ if (lookup_attribute ("used", DECL_ATTRIBUTES (decl)))
+ {
+ mark_decl_referenced (decl);
+ if (vnode->finalized)
+ varpool_mark_needed_node (vnode);
+ }
+ if (lookup_attribute ("externally_visible", DECL_ATTRIBUTES (decl)))
+ {
+ if (! TREE_PUBLIC (vnode->decl))
+ warning (OPT_Wattributes,
+ "%J%<externally_visible%> attribute have effect only on public objects",
+ vnode->decl);
+ else
+ {
+ if (vnode->finalized)
+ varpool_mark_needed_node (vnode);
+ vnode->externally_visible = true;
+ }
+ }
+ }
+}
+
/* Analyze the whole compilation unit once it is parsed completely. */
void
cgraph_finalize_compilation_unit (void)
{
- struct cgraph_node *node;
+ struct cgraph_node *node, *next;
+ /* Keep track of already processed nodes when called multiple times for
+ intermodule optimization. */
+ static struct cgraph_node *first_analyzed;
+ struct cgraph_node *first_processed = first_analyzed;
+ static struct varpool_node *first_analyzed_var;
+
+ if (errorcount || sorrycount)
+ return;
finish_aliases_1 ();
if (!flag_unit_at_a_time)
{
+ cgraph_output_pending_asms ();
cgraph_assemble_pending_functions ();
+ varpool_output_debug_info ();
return;
}
- cgraph_varpool_assemble_pending_decls ();
if (!quiet_flag)
- fprintf (stderr, "\nAnalyzing compilation unit\n");
+ {
+ fprintf (stderr, "\nAnalyzing compilation unit\n");
+ fflush (stderr);
+ }
timevar_push (TV_CGRAPH);
+ process_function_and_variable_attributes (first_processed,
+ first_analyzed_var);
+ first_processed = cgraph_nodes;
+ first_analyzed_var = varpool_nodes;
+ varpool_analyze_pending_decls ();
if (cgraph_dump_file)
{
fprintf (cgraph_dump_file, "Initial entry points:");
- for (node = cgraph_nodes; node; node = node->next)
+ for (node = cgraph_nodes; node != first_analyzed; node = node->next)
if (node->needed && DECL_SAVED_TREE (node->decl))
fprintf (cgraph_dump_file, " %s", cgraph_node_name (node));
fprintf (cgraph_dump_file, "\n");
node->next_needed = NULL;
/* ??? It is possible to create extern inline function and later using
- weak alas attribute to kill its body. See
+ weak alias attribute to kill its body. See
gcc.c-torture/compile/20011119-1.c */
if (!DECL_SAVED_TREE (decl))
- continue;
+ {
+ cgraph_reset_node (node);
+ continue;
+ }
gcc_assert (!node->analyzed && node->reachable);
gcc_assert (DECL_SAVED_TREE (decl));
if (!edge->callee->reachable)
cgraph_mark_reachable_node (edge->callee);
- cgraph_varpool_assemble_pending_decls ();
+ /* We finalize local static variables during constructing callgraph
+ edges. Process their attributes too. */
+ process_function_and_variable_attributes (first_processed,
+ first_analyzed_var);
+ first_processed = cgraph_nodes;
+ first_analyzed_var = varpool_nodes;
+ varpool_analyze_pending_decls ();
}
/* Collect entry points to the unit. */
-
if (cgraph_dump_file)
{
fprintf (cgraph_dump_file, "Unit entry points:");
- for (node = cgraph_nodes; node; node = node->next)
+ for (node = cgraph_nodes; node != first_analyzed; node = node->next)
if (node->needed && DECL_SAVED_TREE (node->decl))
fprintf (cgraph_dump_file, " %s", cgraph_node_name (node));
fprintf (cgraph_dump_file, "\n\nInitial ");
if (cgraph_dump_file)
fprintf (cgraph_dump_file, "\nReclaiming functions:");
- for (node = cgraph_nodes; node; node = node->next)
+ for (node = cgraph_nodes; node != first_analyzed; node = next)
{
tree decl = node->decl;
+ next = node->next;
+
+ if (node->local.finalized && !DECL_SAVED_TREE (decl))
+ cgraph_reset_node (node);
if (!node->reachable && DECL_SAVED_TREE (decl))
{
if (cgraph_dump_file)
fprintf (cgraph_dump_file, " %s", cgraph_node_name (node));
cgraph_remove_node (node);
+ continue;
}
else
node->next_needed = NULL;
+ gcc_assert (!node->local.finalized || DECL_SAVED_TREE (decl));
+ gcc_assert (node->analyzed == node->local.finalized);
}
if (cgraph_dump_file)
{
fprintf (cgraph_dump_file, "\n\nReclaimed ");
dump_cgraph (cgraph_dump_file);
}
+ first_analyzed = cgraph_nodes;
ggc_collect ();
timevar_pop (TV_CGRAPH);
}
{
tree decl = node->decl;
struct cgraph_edge *e;
-
+
gcc_assert (!node->output);
for (e = node->callers; e; e = e->next_caller)
|| DECL_EXTERNAL (decl));
}
-
+
}
}
if (flag_unit_at_a_time)
announce_function (decl);
+ cgraph_lower_function (node);
+
/* Generate RTL for the body of DECL. */
lang_hooks.callgraph.expand_function (decl);
points to the dead function body. */
cgraph_node_remove_callees (node);
}
+
+ cgraph_function_flags_ready = true;
}
-/* Fill array order with all nodes with output flag set in the reverse
- topological order. */
+/* Return true when CALLER_DECL should be inlined into CALLEE_DECL. */
-static int
-cgraph_postorder (struct cgraph_node **order)
+bool
+cgraph_inline_p (struct cgraph_edge *e, const char **reason)
{
- struct cgraph_node *node, *node2;
- int stack_size = 0;
- int order_pos = 0;
- struct cgraph_edge *edge, last;
-
- struct cgraph_node **stack =
- xcalloc (cgraph_n_nodes, sizeof (struct cgraph_node *));
-
- /* We have to deal with cycles nicely, so use a depth first traversal
- output algorithm. Ignore the fact that some functions won't need
- to be output and put them into order as well, so we get dependencies
- right through intline functions. */
- for (node = cgraph_nodes; node; node = node->next)
- node->aux = NULL;
- for (node = cgraph_nodes; node; node = node->next)
- if (!node->aux)
- {
- node2 = node;
- if (!node->callers)
- node->aux = &last;
- else
- node->aux = node->callers;
- while (node2)
- {
- while (node2->aux != &last)
- {
- edge = node2->aux;
- if (edge->next_caller)
- node2->aux = edge->next_caller;
- else
- node2->aux = &last;
- if (!edge->caller->aux)
- {
- if (!edge->caller->callers)
- edge->caller->aux = &last;
- else
- edge->caller->aux = edge->caller->callers;
- stack[stack_size++] = node2;
- node2 = edge->caller;
- break;
- }
- }
- if (node2->aux == &last)
- {
- order[order_pos++] = node2;
- if (stack_size)
- node2 = stack[--stack_size];
- else
- node2 = NULL;
- }
- }
- }
- free (stack);
- return order_pos;
-}
-
-
-/* Perform reachability analysis and reclaim all unreachable nodes.
- This function also remove unneeded bodies of extern inline functions
- and thus needs to be done only after inlining decisions has been made. */
-static bool
-cgraph_remove_unreachable_nodes (void)
-{
- struct cgraph_node *first = (void *) 1;
- struct cgraph_node *node;
- bool changed = false;
- int insns = 0;
-
-#ifdef ENABLE_CHECKING
- verify_cgraph ();
-#endif
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file, "\nReclaiming functions:");
-#ifdef ENABLE_CHECKING
- for (node = cgraph_nodes; node; node = node->next)
- gcc_assert (!node->aux);
-#endif
- for (node = cgraph_nodes; node; node = node->next)
- if (node->needed && !node->global.inlined_to
- && (!DECL_EXTERNAL (node->decl) || !node->analyzed))
- {
- node->aux = first;
- first = node;
- }
- else
- gcc_assert (!node->aux);
-
- /* Perform reachability analysis. As a special case do not consider
- extern inline functions not inlined as live because we won't output
- them at all. */
- while (first != (void *) 1)
- {
- struct cgraph_edge *e;
- node = first;
- first = first->aux;
-
- for (e = node->callees; e; e = e->next_callee)
- if (!e->callee->aux
- && node->analyzed
- && (!e->inline_failed || !e->callee->analyzed
- || !DECL_EXTERNAL (e->callee->decl)))
- {
- e->callee->aux = first;
- first = e->callee;
- }
- }
-
- /* Remove unreachable nodes. Extern inline functions need special care;
- Unreachable extern inline functions shall be removed.
- Reachable extern inline functions we never inlined shall get their bodies
- eliminated.
- Reachable extern inline functions we sometimes inlined will be turned into
- unanalyzed nodes so they look like for true extern functions to the rest
- of code. Body of such functions is released via remove_node once the
- inline clones are eliminated. */
- for (node = cgraph_nodes; node; node = node->next)
- {
- if (!node->aux)
- {
- int local_insns;
- tree decl = node->decl;
-
- node->global.inlined_to = NULL;
- if (DECL_STRUCT_FUNCTION (decl))
- local_insns = node->local.self_insns;
- else
- local_insns = 0;
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file, " %s", cgraph_node_name (node));
- if (!node->analyzed || !DECL_EXTERNAL (node->decl))
- cgraph_remove_node (node);
- else
- {
- struct cgraph_edge *e;
-
- for (e = node->callers; e; e = e->next_caller)
- if (e->caller->aux)
- break;
- if (e || node->needed)
- {
- struct cgraph_node *clone;
-
- for (clone = node->next_clone; clone;
- clone = clone->next_clone)
- if (clone->aux)
- break;
- if (!clone)
- {
- DECL_SAVED_TREE (node->decl) = NULL;
- DECL_STRUCT_FUNCTION (node->decl) = NULL;
- DECL_INITIAL (node->decl) = error_mark_node;
- }
- cgraph_node_remove_callees (node);
- node->analyzed = false;
- }
- else
- cgraph_remove_node (node);
- }
- if (!DECL_SAVED_TREE (decl))
- insns += local_insns;
- changed = true;
- }
- }
- for (node = cgraph_nodes; node; node = node->next)
- node->aux = NULL;
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file, "\nReclaimed %i insns", insns);
- return changed;
-}
-
-/* Estimate size of the function after inlining WHAT into TO. */
-
-static int
-cgraph_estimate_size_after_inlining (int times, struct cgraph_node *to,
- struct cgraph_node *what)
-{
- return (what->global.insns - INSNS_PER_CALL) * times + to->global.insns;
-}
-
-/* Estimate the growth caused by inlining NODE into all callees. */
-
-static int
-cgraph_estimate_growth (struct cgraph_node *node)
-{
- int growth = 0;
- struct cgraph_edge *e;
-
- for (e = node->callers; e; e = e->next_caller)
- if (e->inline_failed)
- growth += (cgraph_estimate_size_after_inlining (1, e->caller, node)
- - e->caller->global.insns);
-
- /* ??? Wrong for self recursive functions or cases where we decide to not
- inline for different reasons, but it is not big deal as in that case
- we will keep the body around, but we will also avoid some inlining. */
- if (!node->needed && !DECL_EXTERNAL (node->decl))
- growth -= node->global.insns;
-
- return growth;
+ *reason = e->inline_failed;
+ return !e->inline_failed;
}
-/* E is expected to be an edge being inlined. Clone destination node of
- the edge and redirect it to the new clone.
- DUPLICATE is used for bookkeeping on whether we are actually creating new
- clones or re-using node originally representing out-of-line function call.
- */
-void
-cgraph_clone_inlined_nodes (struct cgraph_edge *e, bool duplicate)
-{
- struct cgraph_node *n;
-
- /* We may eliminate the need for out-of-line copy to be output. In that
- case just go ahead and re-use it. */
- if (!e->callee->callers->next_caller
- && (!e->callee->needed || DECL_EXTERNAL (e->callee->decl))
- && duplicate
- && flag_unit_at_a_time)
- {
- gcc_assert (!e->callee->global.inlined_to);
- if (!DECL_EXTERNAL (e->callee->decl))
- overall_insns -= e->callee->global.insns, nfunctions_inlined++;
- duplicate = 0;
- }
- else if (duplicate)
- {
- n = cgraph_clone_node (e->callee);
- cgraph_redirect_edge_callee (e, n);
- }
- if (e->caller->global.inlined_to)
- e->callee->global.inlined_to = e->caller->global.inlined_to;
- else
- e->callee->global.inlined_to = e->caller;
- /* Recursively clone all bodies. */
- for (e = e->callee->callees; e; e = e->next_callee)
- if (!e->inline_failed)
- cgraph_clone_inlined_nodes (e, duplicate);
-}
+/* Expand all functions that must be output.
-/* Mark edge E as inlined and update callgraph accordingly. */
+ Attempt to topologically sort the nodes so function is output when
+ all called functions are already assembled to allow data to be
+ propagated across the callgraph. Use a stack to get smaller distance
+ between a function and its callees (later we may choose to use a more
+ sophisticated algorithm for function reordering; we will likely want
+ to use subsections to make the output functions appear in top-down
+ order). */
-void
-cgraph_mark_inline_edge (struct cgraph_edge *e)
+static void
+cgraph_expand_all_functions (void)
{
- int old_insns = 0, new_insns = 0;
- struct cgraph_node *to = NULL, *what;
-
- gcc_assert (e->inline_failed);
- e->inline_failed = NULL;
-
- if (!e->callee->global.inlined && flag_unit_at_a_time)
- DECL_POSSIBLY_INLINED (e->callee->decl) = true;
- e->callee->global.inlined = true;
-
- cgraph_clone_inlined_nodes (e, true);
-
- what = e->callee;
+ struct cgraph_node *node;
+ struct cgraph_node **order = XCNEWVEC (struct cgraph_node *, cgraph_n_nodes);
+ int order_pos = 0, new_order_pos = 0;
+ int i;
- /* Now update size of caller and all functions caller is inlined into. */
- for (;e && !e->inline_failed; e = e->caller->callers)
- {
- old_insns = e->caller->global.insns;
- new_insns = cgraph_estimate_size_after_inlining (1, e->caller,
- what);
- gcc_assert (new_insns >= 0);
- to = e->caller;
- to->global.insns = new_insns;
- }
- gcc_assert (what->global.inlined_to == to);
- overall_insns += new_insns - old_insns;
- ncalls_inlined++;
-}
+ order_pos = cgraph_postorder (order);
+ gcc_assert (order_pos == cgraph_n_nodes);
-/* Mark all calls of EDGE->CALLEE inlined into EDGE->CALLER.
- Return following unredirected edge in the list of callers
- of EDGE->CALLEE */
+ /* Garbage collector may remove inline clones we eliminate during
+ optimization. So we must be sure to not reference them. */
+ for (i = 0; i < order_pos; i++)
+ if (order[i]->output)
+ order[new_order_pos++] = order[i];
-static struct cgraph_edge *
-cgraph_mark_inline (struct cgraph_edge *edge)
-{
- struct cgraph_node *to = edge->caller;
- struct cgraph_node *what = edge->callee;
- struct cgraph_edge *e, *next;
- int times = 0;
-
- /* Look for all calls, mark them inline and clone recursively
- all inlined functions. */
- for (e = what->callers; e; e = next)
+ for (i = new_order_pos - 1; i >= 0; i--)
{
- next = e->next_caller;
- if (e->caller == to && e->inline_failed)
+ node = order[i];
+ if (node->output)
{
- cgraph_mark_inline_edge (e);
- if (e == edge)
- edge = next;
- times++;
+ gcc_assert (node->reachable);
+ node->output = 0;
+ cgraph_expand_function (node);
}
}
- gcc_assert (times);
- return edge;
-}
-
-/* Return false when inlining WHAT into TO is not good idea
- as it would cause too large growth of function bodies. */
-
-static bool
-cgraph_check_inline_limits (struct cgraph_node *to, struct cgraph_node *what,
- const char **reason)
-{
- int times = 0;
- struct cgraph_edge *e;
- int newsize;
- int limit;
-
- if (to->global.inlined_to)
- to = to->global.inlined_to;
-
- for (e = to->callees; e; e = e->next_callee)
- if (e->callee == what)
- times++;
-
- /* When inlining large function body called once into small function,
- take the inlined function as base for limiting the growth. */
- if (to->local.self_insns > what->local.self_insns)
- limit = to->local.self_insns;
- else
- limit = what->local.self_insns;
-
- limit += limit * PARAM_VALUE (PARAM_LARGE_FUNCTION_GROWTH) / 100;
-
- newsize = cgraph_estimate_size_after_inlining (times, to, what);
- if (newsize > PARAM_VALUE (PARAM_LARGE_FUNCTION_INSNS)
- && newsize > limit)
- {
- if (reason)
- *reason = N_("--param large-function-growth limit reached");
- return false;
- }
- return true;
-}
-
-/* Return true when function N is small enough to be inlined. */
+ cgraph_process_new_functions ();
-static bool
-cgraph_default_inline_p (struct cgraph_node *n)
-{
- if (!DECL_INLINE (n->decl) || !DECL_SAVED_TREE (n->decl))
- return false;
- if (DECL_DECLARED_INLINE_P (n->decl))
- return n->global.insns < MAX_INLINE_INSNS_SINGLE;
- else
- return n->global.insns < MAX_INLINE_INSNS_AUTO;
-}
-
-/* Return true when inlining WHAT would create recursive inlining.
- We call recursive inlining all cases where same function appears more than
- once in the single recursion nest path in the inline graph. */
-
-static bool
-cgraph_recursive_inlining_p (struct cgraph_node *to,
- struct cgraph_node *what,
- const char **reason)
-{
- bool recursive;
- if (to->global.inlined_to)
- recursive = what->decl == to->global.inlined_to->decl;
- else
- recursive = what->decl == to->decl;
- /* Marking recursive function inline has sane semantic and thus we should
- not warn on it. */
- if (recursive && reason)
- *reason = (what->local.disregard_inline_limits
- ? N_("recursive inlining") : "");
- return recursive;
-}
-
-/* Recompute heap nodes for each of callees. */
-static void
-update_callee_keys (fibheap_t heap, struct fibnode **heap_node,
- struct cgraph_node *node)
-{
- struct cgraph_edge *e;
+ free (order);
- for (e = node->callees; e; e = e->next_callee)
- if (e->inline_failed && heap_node[e->callee->uid])
- fibheap_replace_key (heap, heap_node[e->callee->uid],
- cgraph_estimate_growth (e->callee));
- else if (!e->inline_failed)
- update_callee_keys (heap, heap_node, e->callee);
}
-/* Enqueue all recursive calls from NODE into queue linked via aux pointers
- in between FIRST and LAST. WHERE is used for bookkeeping while looking
- int calls inlined within NODE. */
-static void
-lookup_recursive_calls (struct cgraph_node *node, struct cgraph_node *where,
- struct cgraph_edge **first, struct cgraph_edge **last)
-{
- struct cgraph_edge *e;
- for (e = where->callees; e; e = e->next_callee)
- if (e->callee == node)
- {
- if (!*first)
- *first = e;
- else
- (*last)->aux = e;
- *last = e;
- }
- for (e = where->callees; e; e = e->next_callee)
- if (!e->inline_failed)
- lookup_recursive_calls (node, e->callee, first, last);
-}
+/* This is used to sort the node types by the cgraph order number. */
-/* Decide on recursive inlining: in the case function has recursive calls,
- inline until body size reaches given argument. */
-static void
-cgraph_decide_recursive_inlining (struct cgraph_node *node)
+struct cgraph_order_sort
{
- int limit = PARAM_VALUE (PARAM_MAX_INLINE_INSNS_RECURSIVE_AUTO);
- int max_depth = PARAM_VALUE (PARAM_MAX_INLINE_RECURSIVE_DEPTH_AUTO);
- struct cgraph_edge *first_call = NULL, *last_call = NULL;
- struct cgraph_edge *last_in_current_depth;
- struct cgraph_edge *e;
- struct cgraph_node *master_clone;
- int depth = 0;
- int n = 0;
-
- if (DECL_DECLARED_INLINE_P (node->decl))
- {
- limit = PARAM_VALUE (PARAM_MAX_INLINE_INSNS_RECURSIVE);
- max_depth = PARAM_VALUE (PARAM_MAX_INLINE_RECURSIVE_DEPTH);
- }
-
- /* Make sure that function is small enough to be considered for inlining. */
- if (!max_depth
- || cgraph_estimate_size_after_inlining (1, node, node) >= limit)
- return;
- lookup_recursive_calls (node, node, &first_call, &last_call);
- if (!first_call)
- return;
-
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file,
- "\nPerforming recursive inlining on %s\n",
- cgraph_node_name (node));
-
- /* We need original clone to copy around. */
- master_clone = cgraph_clone_node (node);
- master_clone->needed = true;
- for (e = master_clone->callees; e; e = e->next_callee)
- if (!e->inline_failed)
- cgraph_clone_inlined_nodes (e, true);
-
- /* Do the inlining and update list of recursive call during process. */
- last_in_current_depth = last_call;
- while (first_call
- && cgraph_estimate_size_after_inlining (1, node, master_clone) <= limit)
- {
- struct cgraph_edge *curr = first_call;
-
- first_call = first_call->aux;
- curr->aux = NULL;
-
- cgraph_redirect_edge_callee (curr, master_clone);
- cgraph_mark_inline_edge (curr);
- lookup_recursive_calls (node, curr->callee, &first_call, &last_call);
-
- if (last_in_current_depth
- && ++depth >= max_depth)
- break;
- n++;
- }
-
- /* Cleanup queue pointers. */
- while (first_call)
- {
- struct cgraph_edge *next = first_call->aux;
- first_call->aux = NULL;
- first_call = next;
- }
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file,
- "\n Inlined %i times, body grown from %i to %i insns\n", n,
- master_clone->global.insns, node->global.insns);
-
- /* Remove master clone we used for inlining. We rely that clones inlined
- into master clone gets queued just before master clone so we don't
- need recursion. */
- for (node = cgraph_nodes; node != master_clone;
- node = node->next)
- if (node->global.inlined_to == master_clone)
- cgraph_remove_node (node);
- cgraph_remove_node (master_clone);
-}
-
-/* Set inline_failed for all callers of given function to REASON. */
+ enum { ORDER_UNDEFINED = 0, ORDER_FUNCTION, ORDER_VAR, ORDER_ASM } kind;
+ union
+ {
+ struct cgraph_node *f;
+ struct varpool_node *v;
+ struct cgraph_asm_node *a;
+ } u;
+};
+
+/* Output all functions, variables, and asm statements in the order
+ according to their order fields, which is the order in which they
+ appeared in the file. This implements -fno-toplevel-reorder. In
+ this mode we may output functions and variables which don't really
+ need to be output. */
static void
-cgraph_set_inline_failed (struct cgraph_node *node, const char *reason)
+cgraph_output_in_order (void)
{
- struct cgraph_edge *e;
-
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file, "Inlining failed: %s\n", reason);
- for (e = node->callers; e; e = e->next_caller)
- if (e->inline_failed)
- e->inline_failed = reason;
-}
-
-/* We use greedy algorithm for inlining of small functions:
- All inline candidates are put into prioritized heap based on estimated
- growth of the overall number of instructions and then update the estimates.
+ int max;
+ size_t size;
+ struct cgraph_order_sort *nodes;
+ int i;
+ struct cgraph_node *pf;
+ struct varpool_node *pv;
+ struct cgraph_asm_node *pa;
- INLINED and INLINED_CALEES are just pointers to arrays large enough
- to be passed to cgraph_inlined_into and cgraph_inlined_callees. */
+ max = cgraph_order;
+ size = max * sizeof (struct cgraph_order_sort);
+ nodes = (struct cgraph_order_sort *) alloca (size);
+ memset (nodes, 0, size);
-static void
-cgraph_decide_inlining_of_small_functions (void)
-{
- struct cgraph_node *node;
- fibheap_t heap = fibheap_new ();
- struct fibnode **heap_node =
- xcalloc (cgraph_max_uid, sizeof (struct fibnode *));
- int max_insns = ((HOST_WIDEST_INT) initial_insns
- * (100 + PARAM_VALUE (PARAM_INLINE_UNIT_GROWTH)) / 100);
+ varpool_analyze_pending_decls ();
- /* Put all inline candidates into the heap. */
-
- for (node = cgraph_nodes; node; node = node->next)
+ for (pf = cgraph_nodes; pf; pf = pf->next)
{
- if (!node->local.inlinable || !node->callers
- || node->local.disregard_inline_limits)
- continue;
-
- if (!cgraph_default_inline_p (node))
+ if (pf->output)
{
- cgraph_set_inline_failed (node,
- N_("--param max-inline-insns-single limit reached"));
- continue;
+ i = pf->order;
+ gcc_assert (nodes[i].kind == ORDER_UNDEFINED);
+ nodes[i].kind = ORDER_FUNCTION;
+ nodes[i].u.f = pf;
}
- heap_node[node->uid] =
- fibheap_insert (heap, cgraph_estimate_growth (node), node);
}
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file, "\nDeciding on smaller functions:\n");
- while (overall_insns <= max_insns && (node = fibheap_extract_min (heap)))
+ for (pv = varpool_nodes_queue; pv; pv = pv->next_needed)
{
- struct cgraph_edge *e, *next;
- int old_insns = overall_insns;
-
- heap_node[node->uid] = NULL;
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file,
- "\nConsidering %s with %i insns\n"
- " Estimated growth is %+i insns.\n",
- cgraph_node_name (node), node->global.insns,
- cgraph_estimate_growth (node));
- if (!cgraph_default_inline_p (node))
- {
- cgraph_set_inline_failed (node,
- N_("--param max-inline-insns-single limit reached after inlining into the callee"));
- continue;
- }
- for (e = node->callers; e; e = next)
- {
- next = e->next_caller;
- if (e->inline_failed)
- {
- struct cgraph_node *where;
-
- if (cgraph_recursive_inlining_p (e->caller, e->callee,
- &e->inline_failed)
- || !cgraph_check_inline_limits (e->caller, e->callee,
- &e->inline_failed))
- {
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file, " Not inlining into %s:%s.\n",
- cgraph_node_name (e->caller), e->inline_failed);
- continue;
- }
- next = cgraph_mark_inline (e);
- where = e->caller;
- if (where->global.inlined_to)
- where = where->global.inlined_to;
-
- if (heap_node[where->uid])
- fibheap_replace_key (heap, heap_node[where->uid],
- cgraph_estimate_growth (where));
-
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file,
- " Inlined into %s which now has %i insns.\n",
- cgraph_node_name (e->caller),
- e->caller->global.insns);
- }
- }
-
- cgraph_decide_recursive_inlining (node);
-
- /* Similarly all functions called by the function we just inlined
- are now called more times; update keys. */
- update_callee_keys (heap, heap_node, node);
-
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file,
- " Inlined for a net change of %+i insns.\n",
- overall_insns - old_insns);
+ i = pv->order;
+ gcc_assert (nodes[i].kind == ORDER_UNDEFINED);
+ nodes[i].kind = ORDER_VAR;
+ nodes[i].u.v = pv;
}
- while ((node = fibheap_extract_min (heap)) != NULL)
- if (!node->local.disregard_inline_limits)
- cgraph_set_inline_failed (node, N_("--param inline-unit-growth limit reached"));
- fibheap_delete (heap);
- free (heap_node);
-}
-
-/* Decide on the inlining. We do so in the topological order to avoid
- expenses on updating data structures. */
-
-static void
-cgraph_decide_inlining (void)
-{
- struct cgraph_node *node;
- int nnodes;
- struct cgraph_node **order =
- xcalloc (cgraph_n_nodes, sizeof (struct cgraph_node *));
- int old_insns = 0;
- int i;
-
- for (node = cgraph_nodes; node; node = node->next)
- initial_insns += node->local.self_insns;
- overall_insns = initial_insns;
-
- nnodes = cgraph_postorder (order);
-
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file,
- "\nDeciding on inlining. Starting with %i insns.\n",
- initial_insns);
-
- for (node = cgraph_nodes; node; node = node->next)
- node->aux = 0;
-
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file, "\nInlining always_inline functions:\n");
- /* In the first pass mark all always_inline edges. Do this with a priority
- so none of our later choices will make this impossible. */
- for (i = nnodes - 1; i >= 0; i--)
+ for (pa = cgraph_asm_nodes; pa; pa = pa->next)
{
- struct cgraph_edge *e, *next;
-
- node = order[i];
-
- if (!node->local.disregard_inline_limits)
- continue;
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file,
- "\nConsidering %s %i insns (always inline)\n",
- cgraph_node_name (node), node->global.insns);
- old_insns = overall_insns;
- for (e = node->callers; e; e = next)
- {
- next = e->next_caller;
- if (!e->inline_failed)
- continue;
- if (cgraph_recursive_inlining_p (e->caller, e->callee,
- &e->inline_failed))
- continue;
- cgraph_mark_inline_edge (e);
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file,
- " Inlined into %s which now has %i insns.\n",
- cgraph_node_name (e->caller),
- e->caller->global.insns);
- }
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file,
- " Inlined for a net change of %+i insns.\n",
- overall_insns - old_insns);
+ i = pa->order;
+ gcc_assert (nodes[i].kind == ORDER_UNDEFINED);
+ nodes[i].kind = ORDER_ASM;
+ nodes[i].u.a = pa;
}
- if (!flag_really_no_inline)
+ for (i = 0; i < max; ++i)
{
- cgraph_decide_inlining_of_small_functions ();
-
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file, "\nDeciding on functions called once:\n");
-
- /* And finally decide what functions are called once. */
-
- for (i = nnodes - 1; i >= 0; i--)
+ switch (nodes[i].kind)
{
- node = order[i];
-
- if (node->callers && !node->callers->next_caller && !node->needed
- && node->local.inlinable && node->callers->inline_failed
- && !DECL_EXTERNAL (node->decl) && !DECL_COMDAT (node->decl))
- {
- bool ok = true;
- struct cgraph_node *node1;
-
- /* Verify that we won't duplicate the caller. */
- for (node1 = node->callers->caller;
- node1->callers && !node1->callers->inline_failed
- && ok; node1 = node1->callers->caller)
- if (node1->callers->next_caller || node1->needed)
- ok = false;
- if (ok)
- {
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file,
- "\nConsidering %s %i insns.\n"
- " Called once from %s %i insns.\n",
- cgraph_node_name (node), node->global.insns,
- cgraph_node_name (node->callers->caller),
- node->callers->caller->global.insns);
-
- old_insns = overall_insns;
-
- if (cgraph_check_inline_limits (node->callers->caller, node,
- NULL))
- {
- cgraph_mark_inline (node->callers);
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file,
- " Inlined into %s which now has %i insns"
- " for a net change of %+i insns.\n",
- cgraph_node_name (node->callers->caller),
- node->callers->caller->global.insns,
- overall_insns - old_insns);
- }
- else
- {
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file,
- " Inline limit reached, not inlined.\n");
- }
- }
- }
- }
- }
-
- /* We will never output extern functions we didn't inline.
- ??? Perhaps we can prevent accounting of growth of external
- inline functions. */
- cgraph_remove_unreachable_nodes ();
+ case ORDER_FUNCTION:
+ nodes[i].u.f->output = 0;
+ cgraph_expand_function (nodes[i].u.f);
+ break;
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file,
- "\nInlined %i calls, eliminated %i functions, "
- "%i insns turned to %i insns.\n\n",
- ncalls_inlined, nfunctions_inlined, initial_insns,
- overall_insns);
- free (order);
-}
+ case ORDER_VAR:
+ varpool_assemble_decl (nodes[i].u.v);
+ break;
-/* Decide on the inlining. We do so in the topological order to avoid
- expenses on updating data structures. */
+ case ORDER_ASM:
+ assemble_asm (nodes[i].u.a->asm_str);
+ break;
-static void
-cgraph_decide_inlining_incrementally (struct cgraph_node *node)
-{
- struct cgraph_edge *e;
+ case ORDER_UNDEFINED:
+ break;
- /* First of all look for always inline functions. */
- for (e = node->callees; e; e = e->next_callee)
- if (e->callee->local.disregard_inline_limits
- && e->inline_failed
- && !cgraph_recursive_inlining_p (node, e->callee, &e->inline_failed)
- /* ??? It is possible that renaming variable removed the function body
- in duplicate_decls. See gcc.c-torture/compile/20011119-2.c */
- && DECL_SAVED_TREE (e->callee->decl))
- cgraph_mark_inline (e);
-
- /* Now do the automatic inlining. */
- if (!flag_really_no_inline)
- for (e = node->callees; e; e = e->next_callee)
- if (e->callee->local.inlinable
- && e->inline_failed
- && !e->callee->local.disregard_inline_limits
- && !cgraph_recursive_inlining_p (node, e->callee, &e->inline_failed)
- && cgraph_check_inline_limits (node, e->callee, &e->inline_failed)
- && DECL_SAVED_TREE (e->callee->decl))
- {
- if (cgraph_default_inline_p (e->callee))
- cgraph_mark_inline (e);
- else
- e->inline_failed
- = N_("--param max-inline-insns-single limit reached");
+ default:
+ gcc_unreachable ();
}
-}
-
-
-/* Return true when CALLER_DECL should be inlined into CALLEE_DECL. */
+ }
-bool
-cgraph_inline_p (struct cgraph_edge *e, const char **reason)
-{
- *reason = e->inline_failed;
- return !e->inline_failed;
+ cgraph_asm_nodes = NULL;
}
+/* Mark visibility of all functions.
-
-/* Expand all functions that must be output.
-
- Attempt to topologically sort the nodes so function is output when
- all called functions are already assembled to allow data to be
- propagated across the callgraph. Use a stack to get smaller distance
- between a function and its callees (later we may choose to use a more
- sophisticated algorithm for function reordering; we will likely want
- to use subsections to make the output functions appear in top-down
- order). */
-
-static void
-cgraph_expand_all_functions (void)
-{
- struct cgraph_node *node;
- struct cgraph_node **order =
- xcalloc (cgraph_n_nodes, sizeof (struct cgraph_node *));
- int order_pos = 0, new_order_pos = 0;
- int i;
-
- order_pos = cgraph_postorder (order);
- gcc_assert (order_pos == cgraph_n_nodes);
-
- /* Garbage collector may remove inline clones we eliminate during
- optimization. So we must be sure to not reference them. */
- for (i = 0; i < order_pos; i++)
- if (order[i]->output)
- order[new_order_pos++] = order[i];
-
- for (i = new_order_pos - 1; i >= 0; i--)
- {
- node = order[i];
- if (node->output)
- {
- gcc_assert (node->reachable);
- node->output = 0;
- cgraph_expand_function (node);
- }
- }
- free (order);
-}
-
-/* Mark all local functions.
-
A local function is one whose calls can occur only in the current
compilation unit and all its calls are explicit, so we can change
its calling convention. We simply mark all static functions whose
- address is not taken as local. */
+ address is not taken as local.
+
+ We also change the TREE_PUBLIC flag of all declarations that are public
+ in language point of view but we want to overwrite this default
+ via visibilities for the backend point of view. */
static void
-cgraph_mark_local_functions (void)
+cgraph_function_and_variable_visibility (void)
{
struct cgraph_node *node;
+ struct varpool_node *vnode;
- /* Figure out functions we want to assemble. */
for (node = cgraph_nodes; node; node = node->next)
{
+ if (node->reachable
+ && (DECL_COMDAT (node->decl)
+ || (!flag_whole_program
+ && TREE_PUBLIC (node->decl) && !DECL_EXTERNAL (node->decl))))
+ node->local.externally_visible = true;
+ if (!node->local.externally_visible && node->analyzed
+ && !DECL_EXTERNAL (node->decl))
+ {
+ gcc_assert (flag_whole_program || !TREE_PUBLIC (node->decl));
+ TREE_PUBLIC (node->decl) = 0;
+ }
node->local.local = (!node->needed
- && DECL_SAVED_TREE (node->decl)
- && !TREE_PUBLIC (node->decl));
+ && node->analyzed
+ && !DECL_EXTERNAL (node->decl)
+ && !node->local.externally_visible);
+ }
+ for (vnode = varpool_nodes_queue; vnode; vnode = vnode->next_needed)
+ {
+ if (vnode->needed
+ && !flag_whole_program
+ && (DECL_COMDAT (vnode->decl) || TREE_PUBLIC (vnode->decl)))
+ vnode->externally_visible = 1;
+ if (!vnode->externally_visible)
+ {
+ gcc_assert (flag_whole_program || !TREE_PUBLIC (vnode->decl));
+ TREE_PUBLIC (vnode->decl) = 0;
+ }
+ gcc_assert (TREE_STATIC (vnode->decl));
}
+ /* Because we have to be conservative on the boundaries of source
+ level units, it is possible that we marked some functions in
+ reachable just because they might be used later via external
+ linkage, but after making them local they are really unreachable
+ now. */
+ cgraph_remove_unreachable_nodes (true, cgraph_dump_file);
+
if (cgraph_dump_file)
{
fprintf (cgraph_dump_file, "\nMarking local functions:");
if (node->local.local)
fprintf (cgraph_dump_file, " %s", cgraph_node_name (node));
fprintf (cgraph_dump_file, "\n\n");
+ fprintf (cgraph_dump_file, "\nMarking externally visible functions:");
+ for (node = cgraph_nodes; node; node = node->next)
+ if (node->local.externally_visible)
+ fprintf (cgraph_dump_file, " %s", cgraph_node_name (node));
+ fprintf (cgraph_dump_file, "\n\n");
}
+ cgraph_function_flags_ready = true;
}
/* Return true when function body of DECL still needs to be kept around
cgraph_preserve_function_body_p (tree decl)
{
struct cgraph_node *node;
- /* Keep the body; we're going to dump it. */
- if (dump_enabled_p (TDI_tree_all))
- return true;
if (!cgraph_global_info_ready)
- return (DECL_INLINE (decl) && !flag_really_no_inline);
+ return (flag_really_no_inline
+ ? lang_hooks.tree_inlining.disregard_inline_limits (decl)
+ : DECL_INLINE (decl));
/* Look if there is any clone around. */
for (node = cgraph_node (decl); node; node = node->next_clone)
if (node->global.inlined_to)
return false;
}
+static void
+ipa_passes (void)
+{
+ cfun = NULL;
+ current_function_decl = NULL;
+ tree_register_cfg_hooks ();
+ bitmap_obstack_initialize (NULL);
+ execute_ipa_pass_list (all_ipa_passes);
+ bitmap_obstack_release (NULL);
+}
+
/* Perform simple optimizations based on callgraph. */
void
cgraph_optimize (void)
{
+ if (errorcount || sorrycount)
+ return;
+
#ifdef ENABLE_CHECKING
verify_cgraph ();
#endif
if (!flag_unit_at_a_time)
- return;
+ {
+ cgraph_assemble_pending_functions ();
+ cgraph_process_new_functions ();
+ cgraph_state = CGRAPH_STATE_FINISHED;
+ cgraph_output_pending_asms ();
+ varpool_assemble_pending_decls ();
+ varpool_output_debug_info ();
+ return;
+ }
- process_pending_assemble_externals ();
+ /* Frontend may output common variables after the unit has been finalized.
+ It is safe to deal with them here as they are always zero initialized. */
+ varpool_analyze_pending_decls ();
timevar_push (TV_CGRAPHOPT);
+ if (pre_ipa_mem_report)
+ {
+ fprintf (stderr, "Memory consumption before IPA\n");
+ dump_memory_report (false);
+ }
if (!quiet_flag)
- fprintf (stderr, "Performing intraprocedural optimizations\n");
+ fprintf (stderr, "Performing interprocedural optimizations\n");
- cgraph_mark_local_functions ();
+ cgraph_function_and_variable_visibility ();
if (cgraph_dump_file)
{
fprintf (cgraph_dump_file, "Marked ");
dump_cgraph (cgraph_dump_file);
}
-
- if (flag_inline_trees)
- cgraph_decide_inlining ();
+ cgraph_state = CGRAPH_STATE_IPA;
+
+ /* Don't run the IPA passes if there was any error or sorry messages. */
+ if (errorcount == 0 && sorrycount == 0)
+ ipa_passes ();
+
+ /* This pass remove bodies of extern inline functions we never inlined.
+ Do this later so other IPA passes see what is really going on. */
+ cgraph_remove_unreachable_nodes (false, dump_file);
+ cgraph_increase_alignment ();
cgraph_global_info_ready = true;
if (cgraph_dump_file)
{
fprintf (cgraph_dump_file, "Optimized ");
dump_cgraph (cgraph_dump_file);
+ dump_varpool (cgraph_dump_file);
+ }
+ if (post_ipa_mem_report)
+ {
+ fprintf (stderr, "Memory consumption after IPA\n");
+ dump_memory_report (false);
}
timevar_pop (TV_CGRAPHOPT);
#ifdef ENABLE_CHECKING
verify_cgraph ();
#endif
-
+
cgraph_mark_functions_to_output ();
-
- cgraph_expand_all_functions ();
+
+ cgraph_state = CGRAPH_STATE_EXPANSION;
+ if (!flag_toplevel_reorder)
+ cgraph_output_in_order ();
+ else
+ {
+ cgraph_output_pending_asms ();
+
+ cgraph_expand_all_functions ();
+ varpool_remove_unreferenced_decls ();
+
+ varpool_assemble_pending_decls ();
+ varpool_output_debug_info ();
+ }
+ cgraph_process_new_functions ();
+ cgraph_state = CGRAPH_STATE_FINISHED;
+
if (cgraph_dump_file)
{
fprintf (cgraph_dump_file, "\nFinal ");
/* Double check that all inline clones are gone and that all
function bodies have been released from memory. */
if (flag_unit_at_a_time
- && !dump_enabled_p (TDI_tree_all)
&& !(sorrycount || errorcount))
{
struct cgraph_node *node;
for (node = cgraph_nodes; node; node = node->next)
if (node->analyzed
&& (node->global.inlined_to
- || DECL_SAVED_TREE (node->decl)))
+ || DECL_SAVED_TREE (node->decl)))
{
error_found = true;
dump_cgraph_node (stderr, node);
- }
+ }
if (error_found)
- internal_error ("Nodes with no released memory found.");
+ internal_error ("nodes with no released memory found");
}
#endif
}
+/* Increase alignment of global arrays to improve vectorization potential.
+ TODO:
+ - Consider also structs that have an array field.
+ - Use ipa analysis to prune arrays that can't be vectorized?
+ This should involve global alignment analysis and in the future also
+ array padding. */
+
+static void
+cgraph_increase_alignment (void)
+{
+ if (flag_section_anchors && flag_tree_vectorize)
+ {
+ struct varpool_node *vnode;
+
+ /* Increase the alignment of all global arrays for vectorization. */
+ for (vnode = varpool_nodes_queue;
+ vnode;
+ vnode = vnode->next_needed)
+ {
+ tree vectype, decl = vnode->decl;
+ unsigned int alignment;
+
+ if (TREE_CODE (TREE_TYPE (decl)) != ARRAY_TYPE)
+ continue;
+ vectype = get_vectype_for_scalar_type (TREE_TYPE (TREE_TYPE (decl)));
+ if (!vectype)
+ continue;
+ alignment = TYPE_ALIGN (vectype);
+ if (DECL_ALIGN (decl) >= alignment)
+ continue;
+
+ if (vect_can_force_dr_alignment_p (decl, alignment))
+ {
+ DECL_ALIGN (decl) = TYPE_ALIGN (vectype);
+ DECL_USER_ALIGN (decl) = 1;
+ if (cgraph_dump_file)
+ {
+ fprintf (cgraph_dump_file, "Increasing alignment of decl: ");
+ print_generic_expr (cgraph_dump_file, decl, TDF_SLIM);
+ }
+ }
+ }
+ }
+}
+
/* Generate and emit a static constructor or destructor. WHICH must be
- one of 'I' or 'D'. BODY should be a STATEMENT_LIST containing
+ one of 'I' or 'D'. BODY should be a STATEMENT_LIST containing
GENERIC statements. */
void
tree decl, name, resdecl;
sprintf (which_buf, "%c_%d", which, counter++);
- name = get_file_function_name_long (which_buf);
+ name = get_file_function_name (which_buf);
decl = build_decl (FUNCTION_DECL, name,
build_function_type (void_type_node, void_list_node));
gimplify_function_tree (decl);
- /* ??? We will get called LATE in the compilation process. */
- if (cgraph_global_info_ready)
- tree_rest_of_compilation (decl);
- else
- cgraph_finalize_function (decl, 0);
-
+ cgraph_add_new_function (decl, false);
+ cgraph_mark_needed_node (cgraph_node (decl));
+
if (targetm.have_ctors_dtors)
{
void (*fn) (rtx, int);
{
cgraph_dump_file = dump_begin (TDI_cgraph, NULL);
}
+
+/* The edges representing the callers of the NEW_VERSION node were
+ fixed by cgraph_function_versioning (), now the call_expr in their
+ respective tree code should be updated to call the NEW_VERSION. */
+
+static void
+update_call_expr (struct cgraph_node *new_version)
+{
+ struct cgraph_edge *e;
+
+ gcc_assert (new_version);
+ for (e = new_version->callers; e; e = e->next_caller)
+ /* Update the call expr on the edges
+ to call the new version. */
+ TREE_OPERAND (TREE_OPERAND (get_call_expr_in (e->call_stmt), 0), 0) = new_version->decl;
+}
+
+
+/* Create a new cgraph node which is the new version of
+ OLD_VERSION node. REDIRECT_CALLERS holds the callers
+ edges which should be redirected to point to
+ NEW_VERSION. ALL the callees edges of OLD_VERSION
+ are cloned to the new version node. Return the new
+ version node. */
+
+static struct cgraph_node *
+cgraph_copy_node_for_versioning (struct cgraph_node *old_version,
+ tree new_decl,
+ VEC(cgraph_edge_p,heap) *redirect_callers)
+ {
+ struct cgraph_node *new_version;
+ struct cgraph_edge *e, *new_e;
+ struct cgraph_edge *next_callee;
+ unsigned i;
+
+ gcc_assert (old_version);
+
+ new_version = cgraph_node (new_decl);
+
+ new_version->analyzed = true;
+ new_version->local = old_version->local;
+ new_version->global = old_version->global;
+ new_version->rtl = new_version->rtl;
+ new_version->reachable = true;
+ new_version->count = old_version->count;
+
+ /* Clone the old node callees. Recursive calls are
+ also cloned. */
+ for (e = old_version->callees;e; e=e->next_callee)
+ {
+ new_e = cgraph_clone_edge (e, new_version, e->call_stmt, 0, e->loop_nest, true);
+ new_e->count = e->count;
+ }
+ /* Fix recursive calls.
+ If OLD_VERSION has a recursive call after the
+ previous edge cloning, the new version will have an edge
+ pointing to the old version, which is wrong;
+ Redirect it to point to the new version. */
+ for (e = new_version->callees ; e; e = next_callee)
+ {
+ next_callee = e->next_callee;
+ if (e->callee == old_version)
+ cgraph_redirect_edge_callee (e, new_version);
+
+ if (!next_callee)
+ break;
+ }
+ for (i = 0; VEC_iterate (cgraph_edge_p, redirect_callers, i, e); i++)
+ {
+ /* Redirect calls to the old version node to point to its new
+ version. */
+ cgraph_redirect_edge_callee (e, new_version);
+ }
+
+ return new_version;
+ }
+
+ /* Perform function versioning.
+ Function versioning includes copying of the tree and
+ a callgraph update (creating a new cgraph node and updating
+ its callees and callers).
+
+ REDIRECT_CALLERS varray includes the edges to be redirected
+ to the new version.
+
+ TREE_MAP is a mapping of tree nodes we want to replace with
+ new ones (according to results of prior analysis).
+ OLD_VERSION_NODE is the node that is versioned.
+ It returns the new version's cgraph node. */
+
+struct cgraph_node *
+cgraph_function_versioning (struct cgraph_node *old_version_node,
+ VEC(cgraph_edge_p,heap) *redirect_callers,
+ varray_type tree_map)
+{
+ tree old_decl = old_version_node->decl;
+ struct cgraph_node *new_version_node = NULL;
+ tree new_decl;
+
+ if (!tree_versionable_function_p (old_decl))
+ return NULL;
+
+ /* Make a new FUNCTION_DECL tree node for the
+ new version. */
+ new_decl = copy_node (old_decl);
+
+ /* Create the new version's call-graph node.
+ and update the edges of the new node. */
+ new_version_node =
+ cgraph_copy_node_for_versioning (old_version_node, new_decl,
+ redirect_callers);
+
+ /* Copy the OLD_VERSION_NODE function tree to the new version. */
+ tree_function_versioning (old_decl, new_decl, tree_map, false);
+ /* Update the call_expr on the edges to call the new version node. */
+ update_call_expr (new_version_node);
+
+ /* Update the new version's properties.
+ Make The new version visible only within this translation unit.
+ ??? We cannot use COMDAT linkage because there is no
+ ABI support for this. */
+ DECL_EXTERNAL (new_version_node->decl) = 0;
+ DECL_ONE_ONLY (new_version_node->decl) = 0;
+ TREE_PUBLIC (new_version_node->decl) = 0;
+ DECL_COMDAT (new_version_node->decl) = 0;
+ new_version_node->local.externally_visible = 0;
+ new_version_node->local.local = 1;
+ new_version_node->lowered = true;
+ return new_version_node;
+}
+
+/* Produce separate function body for inline clones so the offline copy can be
+ modified without affecting them. */
+struct cgraph_node *
+save_inline_function_body (struct cgraph_node *node)
+{
+ struct cgraph_node *first_clone;
+
+ gcc_assert (node == cgraph_node (node->decl));
+
+ cgraph_lower_function (node);
+
+ /* In non-unit-at-a-time we construct full fledged clone we never output to
+ assembly file. This clone is pointed out by inline_decl of original function
+ and inlining infrastructure knows how to deal with this. */
+ if (!flag_unit_at_a_time)
+ {
+ struct cgraph_edge *e;
+
+ first_clone = cgraph_clone_node (node, node->count, 0, false);
+ first_clone->needed = 0;
+ first_clone->reachable = 1;
+ /* Recursively clone all bodies. */
+ for (e = first_clone->callees; e; e = e->next_callee)
+ if (!e->inline_failed)
+ cgraph_clone_inlined_nodes (e, true, false);
+ }
+ else
+ first_clone = node->next_clone;
+
+ first_clone->decl = copy_node (node->decl);
+ node->next_clone = NULL;
+ if (!flag_unit_at_a_time)
+ node->inline_decl = first_clone->decl;
+ first_clone->prev_clone = NULL;
+ cgraph_insert_node_to_hashtable (first_clone);
+ gcc_assert (first_clone == cgraph_node (first_clone->decl));
+
+ /* Copy the OLD_VERSION_NODE function tree to the new version. */
+ tree_function_versioning (node->decl, first_clone->decl, NULL, true);
+
+ DECL_EXTERNAL (first_clone->decl) = 0;
+ DECL_ONE_ONLY (first_clone->decl) = 0;
+ TREE_PUBLIC (first_clone->decl) = 0;
+ DECL_COMDAT (first_clone->decl) = 0;
+
+ for (node = first_clone->next_clone; node; node = node->next_clone)
+ node->decl = first_clone->decl;
+#ifdef ENABLE_CHECKING
+ verify_cgraph_node (first_clone);
+#endif
+ return first_clone;
+}
OSDN Git Service
