/* Callgraph based intraprocedural optimizations.
- Copyright (C) 2003 Free Software Foundation, Inc.
+ Copyright (C) 2003, 2004, 2005 Free Software Foundation, Inc.
Contributed by Jan Hubicka
This file is part of GCC.
Software Foundation, 59 Temple Place - Suite 330, Boston, MA
02111-1307, USA. */
+/* This module implements main driver of compilation process as well as
+ few basic intraprocedural optimizers.
+
+ The main scope of this file is to act as an interface in between
+ tree based frontends and the backend (and middle end)
+
+ The front-end is supposed to use following functionality:
+
+ - cgraph_finalize_function
+
+ 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.)
+
+ - cgraph_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.
+
+ 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.
+
+ - cgraph_optimize
+
+ In this unit-at-a-time compilation the intra procedural analysis takes
+ place here. In particular the static functions whose address is never
+ 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
+
+ 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.
+
+ - analyze_expr callback
+
+ This function is responsible for lowering tree nodes not understood by
+ generic code into understandable ones or alternatively marking
+ callgraph and varpool nodes referenced by the as needed.
+
+ ??? On the tree-ssa genericizing should take place here and we will avoid
+ need for these hooks (replacing them by genericizing hook)
+
+ - expand_function 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
+ previously finalized functions to be expanded.
+
+ We implement two compilation modes.
+
+ - unit-at-a-time: In this mode analyzing of all functions is deferred
+ to cgraph_finalize_compilation_unit and expansion into cgraph_optimize.
+
+ In cgraph_finalize_compilation_unit the reachable functions are
+ analyzed. During analysis the call-graph edges from reachable
+ functions are constructed and their destinations are marked as
+ reachable. References to functions and variables are discovered too
+ and variables found to be needed output to the assembly file. Via
+ mark_referenced call in assemble_variable functions referenced by
+ static variables are noticed too.
+
+ The intra-procedural information is produced and its existence
+ indicated by global_info_ready. Once this flag is set it is impossible
+ to change function from !reachable to reachable and thus
+ assemble_variable no longer call mark_referenced.
+
+ Finally the call-graph is topologically sorted and all reachable functions
+ that has not been completely inlined or are not external are output.
+
+ ??? It is possible that reference to function or variable is optimized
+ out. We can not deal with this nicely because topological order is not
+ suitable for it. For tree-ssa we may consider another pass doing
+ optimization and re-discovering reachable functions.
+
+ ??? Reorganize code so variables are output very last and only if they
+ really has been referenced by produced code, so we catch more cases
+ where reference has been optimized out.
+
+ - non-unit-at-a-time
+
+ All functions are variables are output as early as possible to conserve
+ memory consumption. This may or may not result in less memory used but
+ it is still needed for some legacy code that rely on particular ordering
+ of things output from the compiler.
+
+ Varpool data structures are not used and variables are output directly.
+
+ Functions are output early using call of
+ cgraph_assemble_pending_function from cgraph_finalize_function. The
+ 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. */
+
+
#include "config.h"
#include "system.h"
#include "coretypes.h"
#include "tm.h"
#include "tree.h"
+#include "rtl.h"
+#include "tree-flow.h"
#include "tree-inline.h"
#include "langhooks.h"
-#include "hashtab.h"
+#include "pointer-set.h"
#include "toplev.h"
#include "flags.h"
#include "ggc.h"
#include "params.h"
#include "fibheap.h"
#include "c-common.h"
+#include "intl.h"
+#include "function.h"
+#include "tree-gimple.h"
+#include "tree-pass.h"
+#include "output.h"
-#define INSNS_PER_CALL 10
-
-static void cgraph_expand_functions (void);
+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 void cgraph_optimize_function (struct cgraph_node *);
-static bool cgraph_default_inline_p (struct cgraph_node *n);
static void cgraph_analyze_function (struct cgraph_node *node);
-
-/* Statistics we collect about inlining algorithm. */
-static int ncalls_inlined;
-static int nfunctions_inlined;
-static int initial_insns;
-static int overall_insns;
+static void cgraph_create_edges (struct cgraph_node *node, tree body);
/* Records tree nodes seen in cgraph_create_edges. 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. */
-static htab_t visited_nodes;
+static struct pointer_set_t *visited_nodes;
+
+static FILE *cgraph_dump_file;
/* Determine if function DECL is needed. That is, visible to something
either outside this translation unit, something magic in the system
static bool
decide_is_function_needed (struct cgraph_node *node, tree decl)
{
+ tree origin;
+
/* 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
to go back and re-check its dependencies now. */
/* "extern inline" functions are never output locally. */
if (DECL_EXTERNAL (decl))
return false;
- /* We want to emit COMDAT functions only when they turns out to be neccesary. */
+ /* Nested functions of extern inline function shall not be emit unless
+ we inlined the origin. */
+ for (origin = decl_function_context (decl); origin;
+ origin = decl_function_context (origin))
+ if (DECL_EXTERNAL (origin))
+ return false;
+ /* We want to emit COMDAT functions only when absolutely necessary. */
if (DECL_COMDAT (decl))
return false;
if (!DECL_INLINE (decl)
|| (!node->local.disregard_inline_limits
/* When declared inline, defer even the uninlinable functions.
- This allows them to be elliminated when unused. */
+ This allows them to be eliminated when unused. */
&& !DECL_DECLARED_INLINE_P (decl)
- && (node->local.inlinable || !cgraph_default_inline_p (node))))
+ && (!node->local.inlinable || !cgraph_default_inline_p (node))))
return true;
return false;
}
+/* Walk the decls we marked as necessary and see if they reference new
+ variables or functions and add them into the worklists. */
+static bool
+cgraph_varpool_analyze_pending_decls (void)
+{
+ bool changed = false;
+ timevar_push (TV_CGRAPH);
+
+ while (cgraph_varpool_first_unanalyzed_node)
+ {
+ tree decl = cgraph_varpool_first_unanalyzed_node->decl;
+
+ cgraph_varpool_first_unanalyzed_node->analyzed = true;
+
+ cgraph_varpool_first_unanalyzed_node = cgraph_varpool_first_unanalyzed_node->next_needed;
+
+ if (DECL_INITIAL (decl))
+ cgraph_create_edges (NULL, DECL_INITIAL (decl));
+ changed = true;
+ }
+ timevar_pop (TV_CGRAPH);
+ return changed;
+}
+
+/* Optimization of function bodies might've rendered some variables as
+ unnecessary so we want to avoid these from being compiled.
+
+ This is done by prunning the queue and keeping only the variables that
+ really appear needed (ie they are either externally visible or referenced
+ by compiled function). Re-doing the reachability analysis on variables
+ brings back the remaining variables referenced by these. */
+static void
+cgraph_varpool_remove_unreferenced_decls (void)
+{
+ struct cgraph_varpool_node *next, *node = cgraph_varpool_nodes_queue;
+
+ cgraph_varpool_reset_queue ();
+
+ if (errorcount || sorrycount)
+ return;
+
+ while (node)
+ {
+ tree decl = node->decl;
+ next = node->next_needed;
+ node->needed = 0;
+
+ if (node->finalized
+ && ((DECL_ASSEMBLER_NAME_SET_P (decl)
+ && TREE_SYMBOL_REFERENCED (DECL_ASSEMBLER_NAME (decl)))
+ || node->force_output
+ || decide_is_variable_needed (node, decl)))
+ cgraph_varpool_mark_needed_node (node);
+
+ node = next;
+ }
+ cgraph_varpool_analyze_pending_decls ();
+}
+
+
/* When not doing unit-at-a-time, output all functions enqueued.
Return true when such a functions were found. */
-static bool
+
+bool
cgraph_assemble_pending_functions (void)
{
bool output = false;
struct cgraph_node *n = cgraph_nodes_queue;
cgraph_nodes_queue = cgraph_nodes_queue->next_needed;
- if (!n->origin && !DECL_EXTERNAL (n->decl))
- cgraph_expand_function (n);
- output = true;
+ n->next_needed = NULL;
+ if (!n->global.inlined_to
+ && !n->alias
+ && !DECL_EXTERNAL (n->decl))
+ {
+ cgraph_expand_function (n);
+ output = true;
+ }
}
+
return output;
}
-/* Analyze function once it is parsed. Set up the local information
- available - create cgraph edges for function calls via BODY. */
+/* 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)
+cgraph_finalize_function (tree decl, bool nested)
{
struct cgraph_node *node = cgraph_node (decl);
if (node->local.finalized)
{
/* 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.
+ 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 dificult to do. */
- /* Reset our datastructures so we can analyze the function body
- again. */
+ ??? 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);
+
+ /* 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;
- if (node->output)
- abort ();
- while (node->callees)
- cgraph_remove_call (node->decl, node->callees->callee->decl);
+ node->local.redefined_extern_inline = true;
+
+ if (!flag_unit_at_a_time)
+ {
+ struct cgraph_node *n;
+
+ for (n = cgraph_nodes; n; n = n->next)
+ if (n->global.inlined_to == node)
+ cgraph_remove_node (n);
+ }
+
+ 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)
node->reachable = 0;
}
}
+
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 not unit at a time, then we need to create the call graph
now, so that called functions can be queued and emitted now. */
if (!flag_unit_at_a_time)
- cgraph_analyze_function (node);
+ {
+ cgraph_analyze_function (node);
+ cgraph_decide_inlining_incrementally (node);
+ }
if (decide_is_function_needed (node, decl))
cgraph_mark_needed_node (node);
- /* If not unit at a time, go ahead and emit everything we've
- found to be reachable at this time. Do this only at top-level. */
- if (!node->origin)
- cgraph_assemble_pending_functions ();
+ /* If not unit at a time, go ahead and emit everything we've found
+ to be reachable at this time. */
+ if (!nested)
+ {
+ if (!cgraph_assemble_pending_functions ())
+ ggc_collect ();
+ }
/* If we've not yet emitted decl, tell the debug info about it. */
- if (flag_unit_at_a_time || !node->reachable)
+ if (!TREE_ASM_WRITTEN (decl))
(*debug_hooks->deferred_inline_function) (decl);
+
+ /* Possibly warn about unused parameters. */
+ if (warn_unused_parameter)
+ do_warn_unused_parameter (decl);
+}
+
+/* Used only while constructing the callgraph. */
+static basic_block current_basic_block;
+
+void
+cgraph_lower_function (struct cgraph_node *node)
+{
+ if (node->lowered)
+ return;
+ tree_lowering_passes (node->decl);
+ node->lowered = true;
}
/* Walk tree and record all calls. Called via walk_tree. */
/* ??? 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))
- cgraph_varpool_mark_needed_node (cgraph_varpool_node (t));
+ if (TREE_STATIC (t) || DECL_EXTERNAL (t))
+ {
+ cgraph_varpool_mark_needed_node (cgraph_varpool_node (t));
+ if (lang_hooks.callgraph.analyze_expr)
+ return lang_hooks.callgraph.analyze_expr (tp, walk_subtrees,
+ data);
+ }
break;
+ case FDESC_EXPR:
case ADDR_EXPR:
if (flag_unit_at_a_time)
{
tree decl = get_callee_fndecl (*tp);
if (decl && TREE_CODE (decl) == FUNCTION_DECL)
{
- if (DECL_BUILT_IN (decl))
- return NULL;
- cgraph_record_call (data, decl);
+ cgraph_create_edge (data, cgraph_node (decl), *tp,
+ current_basic_block->count,
+ current_basic_block->loop_depth);
/* When we see a function call, we don't want to look at the
function reference in the ADDR_EXPR that is hanging from
default:
/* Save some cycles by not walking types and declaration as we
won't find anything useful there anyway. */
- if (DECL_P (*tp) || TYPE_P (*tp))
+ if (IS_TYPE_OR_DECL_P (*tp))
{
*walk_subtrees = 0;
break;
}
if ((unsigned int) TREE_CODE (t) >= LAST_AND_UNUSED_TREE_CODE)
- return (*lang_hooks.callgraph.analyze_expr) (tp, walk_subtrees, data);
+ return lang_hooks.callgraph.analyze_expr (tp, walk_subtrees, data);
break;
}
return NULL;
}
-/* Create cgraph edges for function calls inside BODY from DECL. */
+/* Create cgraph edges for function calls inside BODY from NODE. */
-void
-cgraph_create_edges (tree decl, tree body)
+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. */
- visited_nodes = htab_create (37, htab_hash_pointer,
- htab_eq_pointer, NULL);
- walk_tree (&body, record_call_1, decl, visited_nodes);
- htab_delete (visited_nodes);
+ visited_nodes = pointer_set_create ();
+ gcc_assert (current_basic_block == NULL);
+ if (TREE_CODE (body) == FUNCTION_DECL)
+ {
+ struct function *this_cfun = DECL_STRUCT_FUNCTION (body);
+ block_stmt_iterator bsi;
+ tree step;
+
+ /* Reach the trees by walking over the CFG, and note the
+ enclosing basic-blocks in the call edges. */
+ FOR_EACH_BB_FN (current_basic_block, this_cfun)
+ for (bsi = bsi_start (current_basic_block); !bsi_end_p (bsi); bsi_next (&bsi))
+ walk_tree (bsi_stmt_ptr (bsi), record_call_1, node, visited_nodes);
+ current_basic_block = NULL;
+
+ /* Walk over any private statics that may take addresses of functions. */
+ if (TREE_CODE (DECL_INITIAL (body)) == BLOCK)
+ {
+ for (step = BLOCK_VARS (DECL_INITIAL (body));
+ step;
+ step = TREE_CHAIN (step))
+ if (DECL_INITIAL (step))
+ walk_tree (&DECL_INITIAL (step), record_call_1, node, visited_nodes);
+ }
+
+ /* Also look here for private statics. */
+ if (DECL_STRUCT_FUNCTION (body))
+ for (step = DECL_STRUCT_FUNCTION (body)->unexpanded_var_list;
+ step;
+ step = TREE_CHAIN (step))
+ {
+ tree decl = TREE_VALUE (step);
+ if (DECL_INITIAL (decl) && TREE_STATIC (decl))
+ walk_tree (&DECL_INITIAL (decl), record_call_1, node, visited_nodes);
+ }
+ }
+ else
+ walk_tree (&body, record_call_1, node, visited_nodes);
+
+ pointer_set_destroy (visited_nodes);
visited_nodes = NULL;
}
+static bool error_found;
+
+/* 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)
+{
+ tree t = *tp;
+ tree decl;
+
+ if (TREE_CODE (t) == CALL_EXPR && (decl = get_callee_fndecl (t)))
+ {
+ 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;
+ }
+ else
+ {
+ error ("Missing callgraph edge for call expr:");
+ debug_tree (t);
+ error_found = true;
+ }
+ }
+
+ /* 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;
+
+ return NULL_TREE;
+}
+
+/* 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;
+
+ 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",
+ cgraph_node_name (e->caller), cgraph_node_name (e->callee));
+ error_found = true;
+ }
+ for (e = node->callers; e; e = e->next_caller)
+ {
+ 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_found = true;
+ }
+ if (node->callers->next_caller)
+ {
+ error ("Multiple inline callers");
+ error_found = true;
+ }
+ }
+ else
+ if (node->global.inlined_to)
+ {
+ 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_found = true;
+ }
+ if (node->global.inlined_to == node)
+ {
+ error ("Inlined_to pointer reffers to itself");
+ error_found = true;
+ }
+
+ for (main_clone = cgraph_node (node->decl); main_clone;
+ main_clone = main_clone->next_clone)
+ if (main_clone == node)
+ break;
+ if (!node)
+ {
+ error ("Node not found in DECL_ASSEMBLER_NAME 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))
+ {
+ 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))
+ walk_tree (bsi_stmt_ptr (bsi), verify_cgraph_node_1, node, visited_nodes);
+ 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",
+ cgraph_node_name (e->caller),
+ cgraph_node_name (e->callee));
+ error_found = true;
+ }
+ e->aux = 0;
+ }
+ }
+ if (error_found)
+ {
+ dump_cgraph_node (stderr, node);
+ internal_error ("verify_cgraph_node failed.");
+ }
+ timevar_pop (TV_CGRAPH_VERIFY);
+}
+
+/* Verify whole cgraph structure. */
+void
+verify_cgraph (void)
+{
+ struct cgraph_node *node;
+
+ if (sorrycount || errorcount)
+ return;
+
+ for (node = cgraph_nodes; node; node = node->next)
+ verify_cgraph_node (node);
+}
+
+
+/* Output all variables enqueued to be assembled. */
+bool
+cgraph_varpool_assemble_pending_decls (void)
+{
+ bool changed = false;
+
+ if (errorcount || sorrycount)
+ return false;
+
+ /* EH might mark decls as needed during expansion. This should be safe since
+ we don't create references to new function, but it should not be used
+ elsewhere. */
+ cgraph_varpool_analyze_pending_decls ();
+
+ while (cgraph_varpool_nodes_queue)
+ {
+ tree decl = cgraph_varpool_nodes_queue->decl;
+ struct cgraph_varpool_node *node = cgraph_varpool_nodes_queue;
+
+ cgraph_varpool_nodes_queue = cgraph_varpool_nodes_queue->next_needed;
+ if (!TREE_ASM_WRITTEN (decl) && !node->alias && !DECL_EXTERNAL (decl))
+ {
+ assemble_variable (decl, 0, 1, 0);
+ changed = true;
+ }
+ node->next_needed = NULL;
+ }
+ return changed;
+}
+
/* Analyze the function scheduled to be output. */
static 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 (decl, DECL_SAVED_TREE (decl));
+ cgraph_create_edges (node, decl);
node->local.inlinable = tree_inlinable_function_p (decl);
- if (!DECL_ESTIMATED_INSNS (decl))
- DECL_ESTIMATED_INSNS (decl)
- = (*lang_hooks.tree_inlining.estimate_num_insns) (decl);
- node->local.self_insns = DECL_ESTIMATED_INSNS (decl);
+ 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);
-
- /* Inlining characteristics are maintained by the cgraph_mark_inline. */
- node->global.insns = node->local.self_insns;
- if (!DECL_EXTERNAL (decl))
+ = lang_hooks.tree_inlining.disregard_inline_limits (decl);
+ for (e = node->callers; e; e = e->next_caller)
{
- node->global.cloned_times = 1;
- node->global.will_be_output = true;
+ 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");
}
+ 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;
}
cgraph_finalize_compilation_unit (void)
{
struct cgraph_node *node;
+ /* Keep track of already processed nodes when called multiple times for
+ intermodule optimization. */
+ static struct cgraph_node *first_analyzed;
+
+ finish_aliases_1 ();
if (!flag_unit_at_a_time)
{
return;
}
- cgraph_varpool_assemble_pending_decls ();
if (!quiet_flag)
- fprintf (stderr, "\nAnalyzing compilation unit\n");
+ {
+ fprintf (stderr, "\nAnalyzing compilation unit");
+ fflush (stderr);
+ }
timevar_push (TV_CGRAPH);
+ cgraph_varpool_analyze_pending_decls ();
if (cgraph_dump_file)
{
- fprintf (cgraph_dump_file, "\nInitial entry points:");
- for (node = cgraph_nodes; node; node = node->next)
+ fprintf (cgraph_dump_file, "Initial entry points:");
+ 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 = cgraph_nodes_queue;
cgraph_nodes_queue = cgraph_nodes_queue->next_needed;
+ node->next_needed = NULL;
/* ??? It is possible to create extern inline function and later using
- weak alas attribute to kill it's body. See
+ weak alias attribute to kill its body. See
gcc.c-torture/compile/20011119-1.c */
if (!DECL_SAVED_TREE (decl))
continue;
- if (node->analyzed || !node->reachable || !DECL_SAVED_TREE (decl))
- abort ();
+ gcc_assert (!node->analyzed && node->reachable);
+ gcc_assert (DECL_SAVED_TREE (decl));
cgraph_analyze_function (node);
if (!edge->callee->reachable)
cgraph_mark_reachable_node (edge->callee);
- cgraph_varpool_assemble_pending_decls ();
+ cgraph_varpool_analyze_pending_decls ();
}
/* Collect entry points to the unit. */
if (cgraph_dump_file)
{
- fprintf (cgraph_dump_file, "\nUnit entry points:");
- for (node = cgraph_nodes; node; node = node->next)
+ fprintf (cgraph_dump_file, "Unit entry points:");
+ 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");
+ fprintf (cgraph_dump_file, "\n\nInitial ");
dump_cgraph (cgraph_dump_file);
}
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 = node->next)
{
tree decl = node->decl;
if (!node->reachable && DECL_SAVED_TREE (decl))
{
- cgraph_remove_node (node);
if (cgraph_dump_file)
fprintf (cgraph_dump_file, " %s", cgraph_node_name (node));
+ cgraph_remove_node (node);
}
+ else
+ node->next_needed = NULL;
}
if (cgraph_dump_file)
- fprintf (cgraph_dump_file, "\n");
+ {
+ fprintf (cgraph_dump_file, "\n\nReclaimed ");
+ dump_cgraph (cgraph_dump_file);
+ }
+ first_analyzed = cgraph_nodes;
ggc_collect ();
timevar_pop (TV_CGRAPH);
}
-
/* Figure out what functions we want to assemble. */
static void
{
tree decl = node->decl;
struct cgraph_edge *e;
- if (node->output)
- abort ();
+
+ gcc_assert (!node->output);
for (e = node->callers; e; e = e->next_caller)
- if (!e->inline_call)
+ if (e->inline_failed)
break;
/* We need to output all local functions that are used and not
always inlined, as well as those that are reachable from
outside the current compilation unit. */
if (DECL_SAVED_TREE (decl)
+ && !node->global.inlined_to
&& (node->needed
|| (e && node->reachable))
- && !TREE_ASM_WRITTEN (decl) && !node->origin
+ && !TREE_ASM_WRITTEN (decl)
&& !DECL_EXTERNAL (decl))
node->output = 1;
- }
-}
-
-/* Optimize the function before expansion. */
-
-static void
-cgraph_optimize_function (struct cgraph_node *node)
-{
- tree decl = node->decl;
+ else
+ {
+ /* We should've reclaimed all functions that are not needed. */
+#ifdef ENABLE_CHECKING
+ if (!node->global.inlined_to && DECL_SAVED_TREE (decl)
+ && !DECL_EXTERNAL (decl))
+ {
+ dump_cgraph_node (stderr, node);
+ internal_error ("failed to reclaim unneeded function");
+ }
+#endif
+ gcc_assert (node->global.inlined_to || !DECL_SAVED_TREE (decl)
+ || DECL_EXTERNAL (decl));
- timevar_push (TV_INTEGRATION);
- /* optimize_inline_calls avoids inlining of current_function_decl. */
- current_function_decl = decl;
- if (flag_inline_trees)
- optimize_inline_calls (decl);
- if (node->nested)
- {
- for (node = node->nested; node; node = node->next_nested)
- cgraph_optimize_function (node);
+ }
+
}
- timevar_pop (TV_INTEGRATION);
}
/* Expand function specified by NODE. */
cgraph_expand_function (struct cgraph_node *node)
{
tree decl = node->decl;
- struct cgraph_edge *e;
- announce_function (decl);
+ /* We ought to not compile any inline clones. */
+ gcc_assert (!node->global.inlined_to);
- cgraph_optimize_function (node);
+ if (flag_unit_at_a_time)
+ announce_function (decl);
- /* Generate RTL for the body of DECL. Nested functions are expanded
- via lang_expand_decl_stmt. */
- (*lang_hooks.callgraph.expand_function) (decl);
+ cgraph_lower_function (node);
- if (!flag_unit_at_a_time)
- {
- if (!node->local.inlinable
- || (!node->local.disregard_inline_limits
- && !cgraph_default_inline_p (node)))
- DECL_SAVED_TREE (node->decl) = NULL;
- }
- else
+ /* Generate RTL for the body of DECL. */
+ lang_hooks.callgraph.expand_function (decl);
+
+ /* Make sure that BE didn't give up on compiling. */
+ /* ??? Can happen with nested function of extern inline. */
+ gcc_assert (TREE_ASM_WRITTEN (node->decl));
+
+ current_function_decl = NULL;
+ if (!cgraph_preserve_function_body_p (node->decl))
{
- for (e = node->callers; e; e = e->next_caller)
- if (e->inline_call)
- break;
- if (!e)
- DECL_SAVED_TREE (decl) = NULL;
+ DECL_SAVED_TREE (node->decl) = NULL;
+ DECL_STRUCT_FUNCTION (node->decl) = NULL;
+ DECL_INITIAL (node->decl) = error_mark_node;
+ /* Eliminate all call edges. This is important so the call_expr no longer
+ points to the dead function body. */
+ cgraph_node_remove_callees (node);
}
- current_function_decl = NULL;
}
-/* Fill array order with all nodes with output flag set in the reverse
- topological order. */
-static int
-cgraph_postorder (struct cgraph_node **order)
-{
- 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 *));
+/* Return true when CALLER_DECL should be inlined into CALLEE_DECL. */
- /* 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;
+bool
+cgraph_inline_p (struct cgraph_edge *e, const char **reason)
+{
+ *reason = e->inline_failed;
+ return !e->inline_failed;
}
-#define INLINED_TIMES(node) ((size_t)(node)->aux)
-#define SET_INLINED_TIMES(node,times) ((node)->aux = (void *)(times))
-/* Return list of nodes we decided to inline NODE into, set their output
- flag and compute INLINED_TIMES.
- We do simple backtracing to get INLINED_TIMES right. This should not be
- expensive as we limit the amount of inlining. Alternatively we may first
- discover set of nodes, topologically sort these and propagate
- INLINED_TIMES */
-
-static int
-cgraph_inlined_into (struct cgraph_node *node, struct cgraph_node **array)
-{
- int nfound = 0;
- struct cgraph_edge **stack;
- struct cgraph_edge *e, *e1;
- int sp;
- int i;
-
- /* Fast path: since we traverse in mostly topological order, we will likely
- find no edges. */
- for (e = node->callers; e; e = e->next_caller)
- if (e->inline_call)
- break;
-
- if (!e)
- return 0;
-
- /* Allocate stack for back-tracking up callgraph. */
- stack = xmalloc ((cgraph_n_nodes + 1) * sizeof (struct cgraph_edge));
- sp = 0;
-
- /* Push the first edge on to the stack. */
- stack[sp++] = e;
-
- while (sp)
- {
- struct cgraph_node *caller;
-
- /* Look at the edge on the top of the stack. */
- e = stack[sp - 1];
- caller = e->caller;
-
- /* Check if the caller destination has been visited yet. */
- if (!caller->output)
- {
- array[nfound++] = e->caller;
- /* Mark that we have visited the destination. */
- caller->output = true;
- SET_INLINED_TIMES (caller, 0);
- }
- SET_INLINED_TIMES (caller, INLINED_TIMES (caller) + 1);
-
- for (e1 = caller->callers; e1; e1 = e1->next_caller)
- if (e1->inline_call)
- break;
- if (e1)
- stack[sp++] = e1;
- else
- {
- while (true)
- {
- for (e1 = e->next_caller; e1; e1 = e1->next_caller)
- if (e1->inline_call)
- break;
-
- if (e1)
- {
- stack[sp - 1] = e1;
- break;
- }
- else
- {
- sp--;
- if (!sp)
- break;
- e = stack[sp - 1];
- }
- }
- }
- }
-
- free (stack);
-
-
- if (cgraph_dump_file)
- {
- fprintf (cgraph_dump_file, "Found inline predecesors of %s:",
- cgraph_node_name (node));
- for (i = 0; i < nfound; i++)
- {
- fprintf (cgraph_dump_file, " %s", cgraph_node_name (array[i]));
- if (INLINED_TIMES (array[i]) != 1)
- fprintf (cgraph_dump_file, " (%i times)",
- (int)INLINED_TIMES (array[i]));
- }
- fprintf (cgraph_dump_file, "\n");
- }
-
- return nfound;
-}
-
-/* Return list of nodes we decided to inline into NODE, set their output
- flag and compute INLINED_TIMES.
-
- This function is identical to cgraph_inlined_into with callers and callees
- nodes swapped. */
-
-static int
-cgraph_inlined_callees (struct cgraph_node *node, struct cgraph_node **array)
-{
- int nfound = 0;
- struct cgraph_edge **stack;
- struct cgraph_edge *e, *e1;
- int sp;
- int i;
-
- /* Fast path: since we traverse in mostly topological order, we will likely
- find no edges. */
- for (e = node->callees; e; e = e->next_callee)
- if (e->inline_call)
- break;
-
- if (!e)
- return 0;
-
- /* Allocate stack for back-tracking up callgraph. */
- stack = xmalloc ((cgraph_n_nodes + 1) * sizeof (struct cgraph_edge));
- sp = 0;
-
- /* Push the first edge on to the stack. */
- stack[sp++] = e;
-
- while (sp)
- {
- struct cgraph_node *callee;
-
- /* Look at the edge on the top of the stack. */
- e = stack[sp - 1];
- callee = e->callee;
-
- /* Check if the callee destination has been visited yet. */
- if (!callee->output)
- {
- array[nfound++] = e->callee;
- /* Mark that we have visited the destination. */
- callee->output = true;
- SET_INLINED_TIMES (callee, 0);
- }
- SET_INLINED_TIMES (callee, INLINED_TIMES (callee) + 1);
-
- for (e1 = callee->callees; e1; e1 = e1->next_callee)
- if (e1->inline_call)
- break;
- if (e1)
- stack[sp++] = e1;
- else
- {
- while (true)
- {
- for (e1 = e->next_callee; e1; e1 = e1->next_callee)
- if (e1->inline_call)
- break;
-
- if (e1)
- {
- stack[sp - 1] = e1;
- break;
- }
- else
- {
- sp--;
- if (!sp)
- break;
- e = stack[sp - 1];
- }
- }
- }
- }
-
- free (stack);
-
- if (cgraph_dump_file)
- {
- fprintf (cgraph_dump_file, "Found inline successors of %s:",
- cgraph_node_name (node));
- for (i = 0; i < nfound; i++)
- {
- fprintf (cgraph_dump_file, " %s", cgraph_node_name (array[i]));
- if (INLINED_TIMES (array[i]) != 1)
- fprintf (cgraph_dump_file, " (%i times)",
- (int)INLINED_TIMES (array[i]));
- }
- fprintf (cgraph_dump_file, "\n");
- }
-
- return nfound;
-}
-
-/* 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;
- int calls_saved = 0;
- int clones_added = 0;
- struct cgraph_edge *e;
-
- for (e = node->callers; e; e = e->next_caller)
- if (!e->inline_call)
- {
- growth += ((cgraph_estimate_size_after_inlining (1, e->caller, node)
- -
- e->caller->global.insns) *e->caller->global.cloned_times);
- calls_saved += e->caller->global.cloned_times;
- clones_added += e->caller->global.cloned_times;
- }
-
- /* ??? 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 && !node->origin && !DECL_EXTERNAL (node->decl))
- growth -= node->global.insns, clones_added--;
-
- if (!calls_saved)
- calls_saved = 1;
-
- return growth;
-}
-
-/* Update insn sizes after inlining WHAT into TO that is already inlined into
- all nodes in INLINED array. */
-
-static void
-cgraph_mark_inline (struct cgraph_node *to, struct cgraph_node *what,
- struct cgraph_node **inlined, int ninlined,
- struct cgraph_node **inlined_callees,
- int ninlined_callees)
-{
- int i;
- int times = 0;
- int clones = 0;
- struct cgraph_edge *e;
- bool called = false;
- int new_insns;
-
- for (e = what->callers; e; e = e->next_caller)
- {
- if (e->caller == to)
- {
- if (e->inline_call)
- abort ();
- e->inline_call = true;
- times++;
- clones += e->caller->global.cloned_times;
- }
- else if (!e->inline_call)
- called = true;
- }
- if (!times)
- abort ();
- ncalls_inlined += times;
-
- new_insns = cgraph_estimate_size_after_inlining (times, to, what);
- if (to->global.will_be_output)
- overall_insns += new_insns - to->global.insns;
- to->global.insns = new_insns;
-
- if (!called && !what->needed && !what->origin
- && !DECL_EXTERNAL (what->decl))
- {
- if (!what->global.will_be_output)
- abort ();
- clones--;
- nfunctions_inlined++;
- what->global.will_be_output = 0;
- overall_insns -= what->global.insns;
- }
- what->global.cloned_times += clones;
- for (i = 0; i < ninlined; i++)
- {
- new_insns =
- cgraph_estimate_size_after_inlining (INLINED_TIMES (inlined[i]) *
- times, inlined[i], what);
- if (inlined[i]->global.will_be_output)
- overall_insns += new_insns - inlined[i]->global.insns;
- inlined[i]->global.insns = new_insns;
- }
- for (i = 0; i < ninlined_callees; i++)
- {
- inlined_callees[i]->global.cloned_times +=
- INLINED_TIMES (inlined_callees[i]) * clones;
- }
-}
-
-/* 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,
- struct cgraph_node **inlined, int ninlined)
-{
- int i;
- int times = 0;
- struct cgraph_edge *e;
- int newsize;
- int limit;
-
- 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)
- return false;
- for (i = 0; i < ninlined; i++)
- {
- newsize =
- cgraph_estimate_size_after_inlining (INLINED_TIMES (inlined[i]) *
- times, inlined[i], what);
- if (newsize > PARAM_VALUE (PARAM_LARGE_FUNCTION_INSNS)
- && newsize >
- inlined[i]->local.self_insns *
- (100 + PARAM_VALUE (PARAM_LARGE_FUNCTION_GROWTH)) / 100)
- return false;
- }
- return true;
-}
-
-/* Return true when function N is small enought to be inlined. */
-
-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;
-}
-
-/* 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.
-
- INLINED and INLINED_CALEES are just pointers to arrays large enought
- to be passed to cgraph_inlined_into and cgraph_inlined_callees. */
-
-static void
-cgraph_decide_inlining_of_small_functions (struct cgraph_node **inlined,
- struct cgraph_node **inlined_callees)
-{
- int i;
- struct cgraph_node *node;
- fibheap_t heap = fibheap_new ();
- struct fibnode **heap_node =
- xcalloc (cgraph_max_uid, sizeof (struct fibnode *));
- int ninlined, ninlined_callees;
- int max_insns = ((HOST_WIDEST_INT) initial_insns
- * (100 + PARAM_VALUE (PARAM_INLINE_UNIT_GROWTH)) / 100);
-
- /* Put all inline candidates into the heap. */
-
- for (node = cgraph_nodes; node; node = node->next)
- {
- struct cgraph_edge *e;
-
- if (!node->local.inlinable || !node->callers
- || !cgraph_default_inline_p (node))
- continue;
-
- /* Rule out always_inline functions we dealt with earlier. */
- for (e = node->callers; e; e = e->next_caller)
- if (e->inline_call)
- break;
- if (e)
- continue;
- heap_node[node->uid] =
- fibheap_insert (heap, cgraph_estimate_growth (node), node);
- }
-
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file, "\n\nDeciding on inlining: ");
- while ((node = fibheap_extract_min (heap)) && overall_insns <= max_insns)
- {
- struct cgraph_edge *e;
- int old_insns = overall_insns;
-
- heap_node[node->uid] = NULL;
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file, "Considering %s %i insns, growth %i.\n",
- cgraph_node_name (node), node->global.insns,
- cgraph_estimate_growth (node));
- if (!cgraph_default_inline_p (node))
- {
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file, "Function too large.\n");
- continue;
- }
- ninlined_callees = cgraph_inlined_callees (node, inlined_callees);
- for (e = node->callers; e; e = e->next_caller)
- if (!e->inline_call && e->caller != node)
- {
- ninlined = cgraph_inlined_into (e->caller, inlined);
- if (e->callee->output
- || !cgraph_check_inline_limits (e->caller, node, inlined,
- ninlined))
- {
- for (i = 0; i < ninlined; i++)
- inlined[i]->output = 0, node->aux = 0;
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file, "Not inlining into %s\n",
- cgraph_node_name (e->caller));
- continue;
- }
- cgraph_mark_inline (e->caller, node, inlined, ninlined,
- inlined_callees, ninlined_callees);
- if (heap_node[e->caller->uid])
- fibheap_replace_key (heap, heap_node[e->caller->uid],
- cgraph_estimate_growth (e->caller));
-
- /* Size of the functions we updated into has changed, so update
- the keys. */
- for (i = 0; i < ninlined; i++)
- {
- inlined[i]->output = 0, node->aux = 0;
- if (heap_node[inlined[i]->uid])
- fibheap_replace_key (heap, heap_node[inlined[i]->uid],
- cgraph_estimate_growth (inlined[i]));
- }
- }
-
- /* Similarly all functions called by function we just inlined
- are now called more times; update keys. */
-
- for (e = node->callees; e; e = e->next_callee)
- if (!e->inline_call && heap_node[e->callee->uid])
- fibheap_replace_key (heap, heap_node[e->callee->uid],
- cgraph_estimate_growth (e->callee));
-
- for (i = 0; i < ninlined_callees; i++)
- {
- struct cgraph_edge *e;
-
- for (e = inlined_callees[i]->callees; e; e = e->next_callee)
- if (!e->inline_call && heap_node[e->callee->uid])
- fibheap_replace_key (heap, heap_node[e->callee->uid],
- cgraph_estimate_growth (e->callee));
-
- inlined_callees[i]->output = 0, node->aux = 0;
- }
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file,
- "Created %i clones, Num insns:%i (%+i), %.2f%%.\n\n",
- node->global.cloned_times - 1,
- overall_insns, overall_insns - old_insns,
- overall_insns * 100.0 / initial_insns);
- }
- if (cgraph_dump_file && !fibheap_empty (heap))
- fprintf (cgraph_dump_file, "inline-unit-growth limit reached.\n");
- fibheap_delete (heap);
- free (heap_node);
-}
-
-/* Decide on the inlining. We do so in the topological order to avoid
- expenses on updating datastructures. */
-
-static void
-cgraph_decide_inlining (void)
-{
- struct cgraph_node *node;
- int nnodes;
- struct cgraph_node **order =
- xcalloc (cgraph_n_nodes, sizeof (struct cgraph_node *));
- struct cgraph_node **inlined =
- xcalloc (cgraph_n_nodes, sizeof (struct cgraph_node *));
- struct cgraph_node **inlined_callees =
- xcalloc (cgraph_n_nodes, sizeof (struct cgraph_node *));
- int ninlined;
- int ninlined_callees;
- int i, y;
-
- for (node = cgraph_nodes; node; node = node->next)
- initial_insns += node->local.self_insns;
- overall_insns = initial_insns;
-
- nnodes = cgraph_postorder (order);
-
- for (node = cgraph_nodes; node; node = node->next)
- node->aux = 0;
-
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file, "\n\nDeciding on always_inline functions:\n");
-
- /* In the first pass mark all always_inline edges. Do this with a priority
- so no our decisions makes this impossible. */
- for (i = nnodes - 1; i >= 0; i--)
- {
- struct cgraph_edge *e;
-
- node = order[i];
-
- for (e = node->callees; e; e = e->next_callee)
- if (e->callee->local.disregard_inline_limits)
- break;
- if (!e)
- continue;
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file,
- "Considering %s %i insns (always inline)\n",
- cgraph_node_name (node), node->global.insns);
- ninlined = cgraph_inlined_into (order[i], inlined);
- for (; e; e = e->next_callee)
- {
- if (e->inline_call || !e->callee->local.disregard_inline_limits)
- continue;
- if (e->callee->output || e->callee == node)
- continue;
- ninlined_callees =
- cgraph_inlined_callees (e->callee, inlined_callees);
- cgraph_mark_inline (node, e->callee, inlined, ninlined,
- inlined_callees, ninlined_callees);
- for (y = 0; y < ninlined_callees; y++)
- inlined_callees[y]->output = 0, node->aux = 0;
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file, "Inlined %i times. Now %i insns\n\n",
- node->global.cloned_times, overall_insns);
- }
- for (y = 0; y < ninlined; y++)
- inlined[y]->output = 0, node->aux = 0;
- }
-
- cgraph_decide_inlining_of_small_functions (inlined, inlined_callees);
-
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file, "\n\nFunctions to inline once:\n");
-
- /* And finally decide what functions are called once. */
-
- for (i = nnodes - 1; i >= 0; i--)
- {
- node = order[i];
-
- if (node->callers && !node->callers->next_caller && !node->needed
- && node->local.inlinable && !node->callers->inline_call
- && !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_call
- && ok; node1 = node1->callers->caller)
- if (node1->callers->next_caller || node1->needed)
- ok = false;
- if (ok)
- {
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file,
- "Considering %s %i insns (called once)\n",
- cgraph_node_name (node), node->global.insns);
- ninlined = cgraph_inlined_into (node->callers->caller, inlined);
- if (cgraph_check_inline_limits
- (node->callers->caller, node, inlined, ninlined))
- {
- ninlined_callees =
- cgraph_inlined_callees (node, inlined_callees);
- cgraph_mark_inline (node->callers->caller, node, inlined,
- ninlined, inlined_callees,
- ninlined_callees);
- for (y = 0; y < ninlined_callees; y++)
- inlined_callees[y]->output = 0, node->aux = 0;
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file, "Inlined. Now %i insns\n\n", overall_insns);
- }
- for (y = 0; y < ninlined; y++)
- inlined[y]->output = 0, node->aux = 0;
- }
- }
- }
-
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file,
- "\nInlined %i calls, elliminated %i functions, %i insns turned to %i insns.\n",
- ncalls_inlined, nfunctions_inlined, initial_insns,
- overall_insns);
- free (order);
- free (inlined);
- free (inlined_callees);
-}
-
-/* Return true when CALLER_DECL should be inlined into CALLEE_DECL. */
-
-bool
-cgraph_inline_p (tree caller_decl, tree callee_decl)
-{
- struct cgraph_node *caller = cgraph_node (caller_decl);
- struct cgraph_node *callee = cgraph_node (callee_decl);
- struct cgraph_edge *e;
-
- for (e = caller->callees; e; e = e->next_callee)
- if (e->callee == callee)
- return e->inline_call;
- /* We do not record builtins in the callgraph. Perhaps it would make more
- sense to do so and then prune out those not overwritten by explicit
- function body. */
- return false;
-}
/* 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 it's callees (later we may choose to use a more
+ 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_functions (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;
+ int order_pos = 0, new_order_pos = 0;
int i;
- cgraph_mark_functions_to_output ();
-
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 = order_pos - 1; i >= 0; i--)
+ for (i = new_order_pos - 1; i >= 0; i--)
{
node = order[i];
if (node->output)
{
- if (!node->reachable)
- abort ();
+ gcc_assert (node->reachable);
node->output = 0;
cgraph_expand_function (node);
}
}
/* Mark all local functions.
-
- A local function is one whose calls can occur only in the
- current compilation unit, so we change its calling convention.
- We simply mark all static functions whose address is not taken
- as local. */
+
+ 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. */
static void
cgraph_mark_local_functions (void)
{
struct cgraph_node *node;
- if (cgraph_dump_file)
- fprintf (cgraph_dump_file, "Marking local functions:");
-
/* Figure out functions we want to assemble. */
for (node = cgraph_nodes; node; node = node->next)
{
node->local.local = (!node->needed
&& DECL_SAVED_TREE (node->decl)
&& !TREE_PUBLIC (node->decl));
- if (cgraph_dump_file && node->local.local)
- fprintf (cgraph_dump_file, " %s", cgraph_node_name (node));
}
+
if (cgraph_dump_file)
- fprintf (cgraph_dump_file, "\n");
+ {
+ fprintf (cgraph_dump_file, "\nMarking local functions:");
+ for (node = cgraph_nodes; node; node = node->next)
+ if (node->local.local)
+ fprintf (cgraph_dump_file, " %s", cgraph_node_name (node));
+ fprintf (cgraph_dump_file, "\n\n");
+ }
+}
+
+/* Return true when function body of DECL still needs to be kept around
+ for later re-use. */
+bool
+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);
+ /* Look if there is any clone around. */
+ for (node = cgraph_node (decl); node; node = node->next_clone)
+ if (node->global.inlined_to)
+ return true;
+ return false;
}
/* Perform simple optimizations based on callgraph. */
void
cgraph_optimize (void)
{
+#ifdef ENABLE_CHECKING
+ verify_cgraph ();
+#endif
if (!flag_unit_at_a_time)
- return;
+ {
+ cgraph_varpool_assemble_pending_decls ();
+ 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. */
+ cgraph_varpool_analyze_pending_decls ();
+
timevar_push (TV_CGRAPHOPT);
if (!quiet_flag)
fprintf (stderr, "Performing intraprocedural optimizations\n");
+
+ cgraph_mark_local_functions ();
if (cgraph_dump_file)
{
- fprintf (cgraph_dump_file, "Initial callgraph:");
+ fprintf (cgraph_dump_file, "Marked ");
dump_cgraph (cgraph_dump_file);
}
- cgraph_mark_local_functions ();
-
- cgraph_decide_inlining ();
-
+ ipa_passes ();
cgraph_global_info_ready = true;
if (cgraph_dump_file)
{
- fprintf (cgraph_dump_file, "Optimized callgraph:");
+ fprintf (cgraph_dump_file, "Optimized ");
dump_cgraph (cgraph_dump_file);
+ dump_varpool (cgraph_dump_file);
}
timevar_pop (TV_CGRAPHOPT);
- if (!quiet_flag)
- fprintf (stderr, "Assembling functions:");
/* Output everything. */
- cgraph_expand_functions ();
+ if (!quiet_flag)
+ fprintf (stderr, "Assembling functions:\n");
+#ifdef ENABLE_CHECKING
+ verify_cgraph ();
+#endif
+
+ cgraph_mark_functions_to_output ();
+ cgraph_expand_all_functions ();
+ cgraph_varpool_remove_unreferenced_decls ();
+
+ cgraph_varpool_assemble_pending_decls ();
+
if (cgraph_dump_file)
{
- fprintf (cgraph_dump_file, "Final callgraph:");
+ fprintf (cgraph_dump_file, "\nFinal ");
dump_cgraph (cgraph_dump_file);
}
+#ifdef ENABLE_CHECKING
+ verify_cgraph ();
+ /* 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;
+ bool error_found = false;
+
+ for (node = cgraph_nodes; node; node = node->next)
+ if (node->analyzed
+ && (node->global.inlined_to
+ || DECL_SAVED_TREE (node->decl)))
+ {
+ error_found = true;
+ dump_cgraph_node (stderr, node);
+ }
+ if (error_found)
+ internal_error ("Nodes with no released memory found.");
+ }
+#endif
+}
+
+/* Generate and emit a static constructor or destructor. WHICH must be
+ one of 'I' or 'D'. BODY should be a STATEMENT_LIST containing
+ GENERIC statements. */
+
+void
+cgraph_build_static_cdtor (char which, tree body, int priority)
+{
+ static int counter = 0;
+ char which_buf[16];
+ tree decl, name, resdecl;
+
+ sprintf (which_buf, "%c_%d", which, counter++);
+ name = get_file_function_name_long (which_buf);
+
+ decl = build_decl (FUNCTION_DECL, name,
+ build_function_type (void_type_node, void_list_node));
+ current_function_decl = decl;
+
+ resdecl = build_decl (RESULT_DECL, NULL_TREE, void_type_node);
+ DECL_ARTIFICIAL (resdecl) = 1;
+ DECL_IGNORED_P (resdecl) = 1;
+ DECL_RESULT (decl) = resdecl;
+
+ allocate_struct_function (decl);
+
+ TREE_STATIC (decl) = 1;
+ TREE_USED (decl) = 1;
+ DECL_ARTIFICIAL (decl) = 1;
+ DECL_IGNORED_P (decl) = 1;
+ DECL_NO_INSTRUMENT_FUNCTION_ENTRY_EXIT (decl) = 1;
+ DECL_SAVED_TREE (decl) = body;
+ TREE_PUBLIC (decl) = ! targetm.have_ctors_dtors;
+ DECL_UNINLINABLE (decl) = 1;
+
+ DECL_INITIAL (decl) = make_node (BLOCK);
+ TREE_USED (DECL_INITIAL (decl)) = 1;
+
+ DECL_SOURCE_LOCATION (decl) = input_location;
+ cfun->function_end_locus = input_location;
+
+ switch (which)
+ {
+ case 'I':
+ DECL_STATIC_CONSTRUCTOR (decl) = 1;
+ break;
+ case 'D':
+ DECL_STATIC_DESTRUCTOR (decl) = 1;
+ break;
+ default:
+ gcc_unreachable ();
+ }
+
+ gimplify_function_tree (decl);
+
+ /* ??? We will get called LATE in the compilation process. */
+ if (cgraph_global_info_ready)
+ {
+ tree_lowering_passes (decl);
+ tree_rest_of_compilation (decl);
+ }
+ else
+ cgraph_finalize_function (decl, 0);
+
+ if (targetm.have_ctors_dtors)
+ {
+ void (*fn) (rtx, int);
+
+ if (which == 'I')
+ fn = targetm.asm_out.constructor;
+ else
+ fn = targetm.asm_out.destructor;
+ fn (XEXP (DECL_RTL (decl), 0), priority);
+ }
+}
+
+void
+init_cgraph (void)
+{
+ cgraph_dump_file = dump_begin (TDI_cgraph, NULL);
}
OSDN Git Service
