5.12.3.2.1.2.1.4. 为产生代码准备
cp_parser_function_definition_after_declarator 最后调用的函数是 expand_or_defer_fn 。在适当的情况下,它将为该函数产生 RTL 代码;否则为 RTL 代码的产生做一些准备。
2973 void
2974 expand_or_defer_fn (tree fn) in semantics.c
2975 {
2976 /* When the parser calls us after finishing the body of a template
2977 function, we don't really want to expand the body. */
2978 if (processing_template_decl )
2979 {
2980 /* Normally, collection only occurs in rest_of_compilation. So,
2981 if we don't collect here, we never collect junk generated
2982 during the processing of templates until we hit a
2983 non-template function. */
2984 ggc_collect ();
2985 return ;
2986 }
2987
2988 /* Replace AGGR_INIT_EXPRs with appropriate CALL_EXPRs. */
2989 walk_tree_without_duplicates (&DECL_SAVED_TREE (fn),
2990 simplify_aggr_init_exprs_r,
2991 NULL);
2992
2993 /* If this is a constructor or destructor body, we have to clone
2994 it. */
2995 if (maybe_clone_body (fn))
2996 {
2997 /* We don't want to process FN again, so pretend we've written
2998 it out, even though we haven't. */
2999 TREE_ASM_WRITTEN (fn) = 1;
3000 return ;
3001 }
3002
3003 /* There's no reason to do any of the work here if we're only doing
3004 semantic analysis; this code just generates RTL. */
3005 if (flag_syntax_only )
3006 return ;
3007
3008 /* Compute the appropriate object-file linkage for inline functions. */
3009 if (DECL_DECLARED_INLINE_P (fn))
3010 import_export_decl (fn);
3011
3012 /* If this function is marked with the constructor attribute, add it
3013 to the list of functions to be called along with constructors
3014 from static duration objects. */
3015 if (DECL_STATIC_CONSTRUCTOR (fn))
3016 static_ctors = tree_cons (NULL_TREE, fn, static_ctors );
3017
3018 /* If this function is marked with the destructor attribute, add it
3019 to the list of functions to be called along with destructors from
3020 static duration objects. */
3021 if (DECL_STATIC_DESTRUCTOR (fn))
3022 static_dtors = tree_cons (NULL_TREE, fn, static_dtors );
3023
3024 function_depth ++;
3025
3026 /* Expand or defer, at the whim of the compilation unit manager. */
3027 cgraph_finalize_function (fn, function_depth > 1);
3028
3029 function_depth --;
3030 }
2989 行对函数定义树进行遍历,处理对经过具名返回值优化处理的函数的调用。我们将结合具名返回值优化来看。
上面判断 DECL_STATIC_CONSTRUCTOR 或 DECL_STATIC_DESTRUCTOR 成立,如果函数的声明中带有 __attribute__ ((constructor)) 或 __attribute__ ((destructor)) 。属性 constructor 使得该函数,在进入 main () 执行前,被自动调用。类似的,属性 destructor 使得该函数, main () 已经完成或 exit () 已经被调用后,被自动调用。具有这些属性的函数对于初始化那些在程序执行期间被隐含使用的数据很有用。
5.12.3.2.1.2.1.4.1. 更新函数调用图
前面的章节 函数调用关系图 中描述了数据结构 cgraph_node 及 cgraph_edge 。另外还学习了函数 cgraph_node ,这个函数为每个函数在第一次调用时构建唯一的 cgraph_node 节点,而在后面的调用中返回这个唯一的节点。这是在编译过程中,为由 decl 指定的函数第一次调用 cgraph_finalize_function 及 cgraph_node 。
157 void
158 cgraph_finalize_function (tree decl, bool nested) in cgraphunit.c
159 {
160 struct cgraph_node *node = cgraph_node (decl);
161
162 if (node->local.finalized)
163 {
164 /* As an GCC extension we allow redefinition of the function. The
165 semantics when both copies of bodies differ is not well defined.
166 We replace the old body with new body so in unit at a time mode
167 we always use new body, while in normal mode we may end up with
168 old body inlined into some functions and new body expanded and
169 inlined in others.
170
171 ??? It may make more sense to use one body for inlining and other
172 body for expanding the function but this is difficult to do. */
173
174 /* If node->output is set, then this is a unit-at-a-time compilation
175 and we have already begun whole-unit analysis. This is *not*
176 testing for whether we've already emitted the function. That
177 case can be sort-of legitimately seen with real function
178 redefinition errors. I would argue that the front end should
179 never present us with such a case, but don't enforce that for now. */
180 if (node->output)
181 abort ();
182
183 /* Reset our datastructures so we can analyze the function again. */
184 memset (&node->local, 0, sizeof (node->local));
185 memset (&node->global, 0, sizeof (node->global));
186 memset (&node->rtl, 0, sizeof (node->rtl));
187 node->analyzed = false;
188 node->local.redefined_extern_inline = true;
189 while (node->callees)
190 cgraph_remove_edge (node, node->callees->callee);
191
192 /* We may need to re-queue the node for assembling in case
193 we already proceeded it and ignored as not needed. */
194 if (node->reachable && !flag_unit_at_a_time )
195 {
196 struct cgraph_node *n;
197
198 for (n = cgraph_nodes_queue ; n; n = n->next_needed)
199 if (n == node)
200 break ;
201 if (!n)
202 node->reachable = 0;
203 }
204 }
205
206 notice_global_symbol (decl);
207 node->decl = decl;
208 node->local.finalized = true;
209
210 /* If not unit at a time, then we need to create the call graph
211 now, so that called functions can be queued and emitted now. */
212 if (!flag_unit_at_a_time )
213 {
214 cgraph_analyze_function (node);
215 cgraph_decide_inlining_incrementally (node);
216 }
在这个第一次调用中, node->local.finalized 一定是 false ,而它很快在 208 行被更新为 true 。在 206 行, notice_global_symbol 收集第一个可见的全局符号。因为全局符号指针这个编译单元是唯一的,这个信息对于编译器为匿名对象产生唯一名字是有用的,比如,在这里收集的符号将被用于产生匿名名字空间的内部名字。
1034 void
1035 notice_global_symbol (tree decl) in varasm.c
1036 {
1037 const char **type = &first_global_object_name ;
1038
1039 if (first_global_object_name
1040 || !TREE_PUBLIC (decl) || DECL_EXTERNAL (decl)
1041 || !DECL_NAME (decl)
1042 || (TREE_CODE (decl) != FUNCTION_DECL
1043 && (TREE_CODE (decl) != VAR_DECL
1044 || (DECL_COMMON (decl)
1045 && (DECL_INITIAL (decl) == 0
1046 || DECL_INITIAL (decl) == error_mark_node))))
1047 || GET_CODE (DECL_RTL (decl)) != MEM)
1048 return ;
1049
1050 /* We win when global object is found, but it is usefull to know about weak
1051 symbol as well so we can produce nicer unique names. */
1052 if (DECL_WEAK (decl) || DECL_ONE_ONLY (decl))
1053 type = &weak_global_object_name ;
1054
1055 if (!*type)
1056 {
1057 const char *p;
1058 char *name;
1059 rtx decl_rtl = DECL_RTL (decl);
1060
1061 p = (* targetm .strip_name_encoding) (XSTR (XEXP (decl_rtl, 0), 0));
1062 name = xstrdup (p);
1063
1064 *type = name;
1065 }
1066 }
在 cgraph_finalize_function 的 212 行,如果我们在编译指令中指定了高于 -O2 的优化选项, flag_unit_at_a_time 被设置为 1 。通常,我们倾向使用 –O3 选项来榨取编译器的所有优化能力。在这里假定 flag_unit_at_a_time 为 1 。
cgraph_finalize_function (continue)
218 if (decide_is_function_needed (node, decl))
219 cgraph_mark_needed_node (node);
220
221 /* If not unit at a time, go ahead and emit everything we've found
222 to be reachable at this time. */
223 if (!nested)
224 {
225 if (!cgraph_assemble_pending_functions ())
226 ggc_collect ();
227 }
228
229 /* If we've not yet emitted decl, tell the debug info about it. */
230 if (!TREE_ASM_WRITTEN (decl))
231 (*debug_hooks ->deferred_inline_function) (decl);
232
233 /* We will never really output the function body, clear the SAVED_INSNS array
234 early then. */
235 if (DECL_EXTERNAL (decl))
236 DECL_SAVED_INSNS (decl) = NULL;
237 }
函数 decide_is_function_needed 确定 decl 所引用的函数是否在编译单元以外,或者在系统配置的戏法中可见;或者(如果不是 unit-at-a-time —一个单元一起处理)为我们还没遇到的部分所可见。在 93 行,属性“ used ”表示该变量的代码必须被生成,就算看起来这个变量没有被引用(编译器会清除死代码)。
73 static bool
74 decide_is_function_needed (struct cgraph_node *node, tree decl) in cgraphunit.c
75 {
76 /* If we decided it was needed before, but at the time we didn't have
77 the body of the function available, then it's still needed. We have
78 to go back and re-check its dependencies now. */
79 if (node->needed)
80 return true;
81
82 /* Externally visible functions must be output. The exception is
83 COMDAT functions that must be output only when they are needed. */
84 if (TREE_PUBLIC (decl) && !DECL_COMDAT (decl) && !DECL_EXTERNAL (decl))
85 return true;
86
87 /* Constructors and destructors are reachable from the runtime by
88 some mechanism. */
89 if (DECL_STATIC_CONSTRUCTOR (decl) || DECL_STATIC_DESTRUCTOR (decl))
90 return true;
91
92 /* If the user told us it is used, then it must be so. */
93 if (lookup_attribute ("used", DECL_ATTRIBUTES (decl)))
94 return true;
95
96 /* ??? If the assembler name is set by hand, it is possible to assemble
97 the name later after finalizing the function and the fact is noticed
98 i n assemble_name then. This is arguably a bug. */
99 if (DECL_ASSEMBLER_NAME_SET_P (decl)
100 && TREE_SYMBOL_REFERENCED (DECL_ASSEMBLER_NAME (decl)))
101 return true;
102
103 if (flag_unit_at_a_time )
104 return false;
105
106 /* If not doing unit at a time, then we'll only defer this function
107 if its marked for inlining. Otherwise we want to emit it now. */
108
109 /* "extern inline" functions are never output locally. */
110 if (DECL_EXTERNAL (decl))
111 return false;
112 /* We want to emit COMDAT functions only when absolutely necessary. */
113 if (DECL_COMDAT (decl))
114 return false;
115 if (!DECL_INLINE (decl)
116 || (!node->local.disregard_inline_limits
117 /* When declared inline, defer even the uninlinable functions.
118 This allows them to be eliminated when unused. */
119 && !DECL_DECLARED_INLINE_P (decl)
120 && (!node->local.inlinable || !cgraph_default_inline_p (node))))
121 return true;
122
123 return false;
124 }
如果该声明是可见的, cgraph_mark_needed_node 需要把相关的 cgraph_node 节点设置为 needed 并且标记其中的子节点为可到达。
286 void
287 cgraph_mark_needed_node (struct cgraph_node *node) in cgraph.c
288 {
289 node->needed = 1;
290 cgraph_mark_reachable_node (node);
291 }
在 cgraph_finalize_function 中,域 local.finalized 已经被设置为 true 。在函数 cgraph_node 中,从 123 到 128 行,域 origin 指向包含这个函数的节点;域 nested 指向这个被包含函数的节点(即在其中所调用的函数),而域 next_nested 指向被包含的兄弟函数( sibling functions )。
260 void
261 cgraph_mark_reachable_node (struct cgraph_node *node) in cgraph.c
262 {
263 if (!node->reachable && node->local.finalized)
264 {
265 notice_global_symbol (node->decl);
266 node->reachable = 1;
267
268 node->next_needed = cgraph_nodes_queue ;
269 cgraph_nodes_queue = node;
270
271 /* At the moment frontend automatically emits all nested functions. */
272 if (node->nested)
273 {
274 struct cgraph_node *node2;
275
276 for (node2 = node->nested; node2; node2 = node2->next_nested)
277 if (!node2->reachable)
278 cgraph_mark_reachable_node (node2);
279 }
280 }
281 }
所有外部可见的函数都应该被串接起来。这个链表为 cgraph_nodes_queue 所指向。
回到 cgraph_finalize_function ,如果这不是一个嵌套函数(在函数作用域中或类定义中定义的内联函数),而且 flag_unit_at_a_time 是 true , cgraph_assemble_pending_functions 直接返回 false ,触发垃圾回收。