1 | /* Pass for parsing functions with multiple target attributes. |
2 | |
3 | Contributed by Evgeny Stupachenko <evstupac@gmail.com> |
4 | |
5 | Copyright (C) 2015-2024 Free Software Foundation, Inc. |
6 | |
7 | This file is part of GCC. |
8 | |
9 | GCC is free software; you can redistribute it and/or modify it under |
10 | the terms of the GNU General Public License as published by the Free |
11 | Software Foundation; either version 3, or (at your option) any later |
12 | version. |
13 | |
14 | GCC is distributed in the hope that it will be useful, but WITHOUT ANY |
15 | WARRANTY; without even the implied warranty of MERCHANTABILITY or |
16 | FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
17 | for more details. |
18 | |
19 | You should have received a copy of the GNU General Public License |
20 | along with GCC; see the file COPYING3. If not see |
21 | <http://www.gnu.org/licenses/>. */ |
22 | |
23 | #include "config.h" |
24 | #include "system.h" |
25 | #include "coretypes.h" |
26 | #include "backend.h" |
27 | #include "tree.h" |
28 | #include "stringpool.h" |
29 | #include "gimple.h" |
30 | #include "diagnostic-core.h" |
31 | #include "gimple-ssa.h" |
32 | #include "cgraph.h" |
33 | #include "tree-pass.h" |
34 | #include "target.h" |
35 | #include "attribs.h" |
36 | #include "pretty-print.h" |
37 | #include "gimple-iterator.h" |
38 | #include "gimple-walk.h" |
39 | #include "tree-inline.h" |
40 | #include "intl.h" |
41 | |
42 | /* Walker callback that replaces all FUNCTION_DECL of a function that's |
43 | going to be versioned. */ |
44 | |
45 | static tree |
46 | replace_function_decl (tree *op, int *walk_subtrees, void *data) |
47 | { |
48 | struct walk_stmt_info *wi = (struct walk_stmt_info *) data; |
49 | cgraph_function_version_info *info = (cgraph_function_version_info *)wi->info; |
50 | |
51 | if (TREE_CODE (*op) == FUNCTION_DECL |
52 | && info->this_node->decl == *op) |
53 | { |
54 | *op = info->dispatcher_resolver; |
55 | *walk_subtrees = 0; |
56 | } |
57 | |
58 | return NULL; |
59 | } |
60 | |
61 | /* If the call in NODE has multiple target attribute with multiple fields, |
62 | replace it with dispatcher call and create dispatcher (once). */ |
63 | |
64 | static void |
65 | create_dispatcher_calls (struct cgraph_node *node) |
66 | { |
67 | ipa_ref *ref; |
68 | |
69 | if (!targetm.has_ifunc_p ()) |
70 | { |
71 | error_at (DECL_SOURCE_LOCATION (node->decl), |
72 | "the call requires %<ifunc%>, which is not" |
73 | " supported by this target" ); |
74 | return; |
75 | } |
76 | else if (!targetm.get_function_versions_dispatcher) |
77 | { |
78 | error_at (DECL_SOURCE_LOCATION (node->decl), |
79 | "target does not support function version dispatcher" ); |
80 | return; |
81 | } |
82 | |
83 | tree idecl = targetm.get_function_versions_dispatcher (node->decl); |
84 | if (!idecl) |
85 | { |
86 | error_at (DECL_SOURCE_LOCATION (node->decl), |
87 | "default %<target_clones%> attribute was not set" ); |
88 | return; |
89 | } |
90 | |
91 | cgraph_node *inode = cgraph_node::get (decl: idecl); |
92 | gcc_assert (inode); |
93 | tree resolver_decl = targetm.generate_version_dispatcher_body (inode); |
94 | |
95 | /* Update aliases. */ |
96 | inode->alias = true; |
97 | inode->alias_target = resolver_decl; |
98 | if (!inode->analyzed) |
99 | inode->resolve_alias (target: cgraph_node::get (decl: resolver_decl)); |
100 | |
101 | auto_vec<cgraph_edge *> edges_to_redirect; |
102 | /* We need to capture the references by value rather than just pointers to them |
103 | and remove them right away, as removing them later would invalidate what |
104 | some other reference pointers point to. */ |
105 | auto_vec<ipa_ref> references_to_redirect; |
106 | |
107 | while (node->iterate_referring (i: 0, ref)) |
108 | { |
109 | references_to_redirect.safe_push (obj: *ref); |
110 | ref->remove_reference (); |
111 | } |
112 | |
113 | /* We need to remember NEXT_CALLER as it could be modified in the loop. */ |
114 | for (cgraph_edge *e = node->callers; e ; e = e->next_caller) |
115 | edges_to_redirect.safe_push (obj: e); |
116 | |
117 | if (!edges_to_redirect.is_empty () || !references_to_redirect.is_empty ()) |
118 | { |
119 | /* Redirect edges. */ |
120 | unsigned i; |
121 | cgraph_edge *e; |
122 | FOR_EACH_VEC_ELT (edges_to_redirect, i, e) |
123 | { |
124 | e->redirect_callee (n: inode); |
125 | cgraph_edge::redirect_call_stmt_to_callee (e); |
126 | } |
127 | |
128 | /* Redirect references. */ |
129 | FOR_EACH_VEC_ELT (references_to_redirect, i, ref) |
130 | { |
131 | if (ref->use == IPA_REF_ADDR) |
132 | { |
133 | struct walk_stmt_info wi; |
134 | memset (s: &wi, c: 0, n: sizeof (wi)); |
135 | wi.info = (void *)node->function_version (); |
136 | |
137 | if (dyn_cast<varpool_node *> (p: ref->referring)) |
138 | { |
139 | hash_set<tree> visited_nodes; |
140 | walk_tree (&DECL_INITIAL (ref->referring->decl), |
141 | replace_function_decl, &wi, &visited_nodes); |
142 | } |
143 | else |
144 | { |
145 | gimple_stmt_iterator it = gsi_for_stmt (ref->stmt); |
146 | if (ref->referring->decl != resolver_decl) |
147 | walk_gimple_stmt (&it, NULL, replace_function_decl, &wi); |
148 | } |
149 | |
150 | symtab_node *source = ref->referring; |
151 | source->create_reference (referred_node: inode, use_type: IPA_REF_ADDR); |
152 | } |
153 | else if (ref->use == IPA_REF_ALIAS) |
154 | { |
155 | symtab_node *source = ref->referring; |
156 | source->create_reference (referred_node: inode, use_type: IPA_REF_ALIAS); |
157 | if (inode->get_comdat_group ()) |
158 | source->add_to_same_comdat_group (old_node: inode); |
159 | } |
160 | else |
161 | gcc_unreachable (); |
162 | } |
163 | } |
164 | |
165 | tree fname = clone_function_name (decl: node->decl, suffix: "default" ); |
166 | symtab->change_decl_assembler_name (decl: node->decl, name: fname); |
167 | |
168 | if (node->definition) |
169 | { |
170 | /* FIXME: copy of cgraph_node::make_local that should be cleaned up |
171 | in next stage1. */ |
172 | node->make_decl_local (); |
173 | node->set_section (NULL); |
174 | node->set_comdat_group (NULL); |
175 | node->externally_visible = false; |
176 | node->forced_by_abi = false; |
177 | |
178 | DECL_ARTIFICIAL (node->decl) = 1; |
179 | node->force_output = true; |
180 | } |
181 | } |
182 | |
183 | /* Create string with attributes separated by comma. |
184 | Return number of attributes. */ |
185 | |
186 | static int |
187 | get_attr_str (tree arglist, char *attr_str) |
188 | { |
189 | tree arg; |
190 | size_t str_len_sum = 0; |
191 | int argnum = 0; |
192 | |
193 | for (arg = arglist; arg; arg = TREE_CHAIN (arg)) |
194 | { |
195 | const char *str = TREE_STRING_POINTER (TREE_VALUE (arg)); |
196 | size_t len = strlen (s: str); |
197 | for (const char *p = strchr (s: str, c: ','); p; p = strchr (s: p + 1, c: ',')) |
198 | argnum++; |
199 | memcpy (dest: attr_str + str_len_sum, src: str, n: len); |
200 | attr_str[str_len_sum + len] = TREE_CHAIN (arg) ? ',' : '\0'; |
201 | str_len_sum += len + 1; |
202 | argnum++; |
203 | } |
204 | return argnum; |
205 | } |
206 | |
207 | /* Return number of attributes separated by comma and put them into ARGS. |
208 | If there is no DEFAULT attribute return -1. |
209 | If there is an empty string in attribute return -2. |
210 | If there are multiple DEFAULT attributes return -3. |
211 | */ |
212 | |
213 | static int |
214 | separate_attrs (char *attr_str, char **attrs, int attrnum) |
215 | { |
216 | int i = 0; |
217 | int default_count = 0; |
218 | |
219 | for (char *attr = strtok (s: attr_str, delim: "," ); |
220 | attr != NULL; attr = strtok (NULL, delim: "," )) |
221 | { |
222 | if (strcmp (s1: attr, s2: "default" ) == 0) |
223 | { |
224 | default_count++; |
225 | continue; |
226 | } |
227 | attrs[i++] = attr; |
228 | } |
229 | if (default_count == 0) |
230 | return -1; |
231 | else if (default_count > 1) |
232 | return -3; |
233 | else if (i + default_count < attrnum) |
234 | return -2; |
235 | |
236 | return i; |
237 | } |
238 | |
239 | /* Return true if symbol is valid in assembler name. */ |
240 | |
241 | static bool |
242 | is_valid_asm_symbol (char c) |
243 | { |
244 | if ('a' <= c && c <= 'z') |
245 | return true; |
246 | if ('A' <= c && c <= 'Z') |
247 | return true; |
248 | if ('0' <= c && c <= '9') |
249 | return true; |
250 | if (c == '_') |
251 | return true; |
252 | return false; |
253 | } |
254 | |
255 | /* Replace all not valid assembler symbols with '_'. */ |
256 | |
257 | static void |
258 | create_new_asm_name (char *old_asm_name, char *new_asm_name) |
259 | { |
260 | int i; |
261 | int old_name_len = strlen (s: old_asm_name); |
262 | |
263 | /* Replace all not valid assembler symbols with '_'. */ |
264 | for (i = 0; i < old_name_len; i++) |
265 | if (!is_valid_asm_symbol (c: old_asm_name[i])) |
266 | new_asm_name[i] = '_'; |
267 | else |
268 | new_asm_name[i] = old_asm_name[i]; |
269 | new_asm_name[old_name_len] = '\0'; |
270 | } |
271 | |
272 | /* Creates target clone of NODE. */ |
273 | |
274 | static cgraph_node * |
275 | create_target_clone (cgraph_node *node, bool definition, char *name, |
276 | tree attributes) |
277 | { |
278 | cgraph_node *new_node; |
279 | |
280 | if (definition) |
281 | { |
282 | new_node |
283 | = node->create_version_clone_with_body (redirect_callers: vNULL, NULL, NULL, NULL, NULL, |
284 | clone_name: name, target_attributes: attributes, version_decl: false); |
285 | if (new_node == NULL) |
286 | return NULL; |
287 | new_node->force_output = true; |
288 | } |
289 | else |
290 | { |
291 | tree new_decl = copy_node (node->decl); |
292 | new_node = cgraph_node::get_create (new_decl); |
293 | DECL_ATTRIBUTES (new_decl) = attributes; |
294 | /* Generate a new name for the new version. */ |
295 | tree fname = clone_function_name (decl: node->decl, suffix: name); |
296 | symtab->change_decl_assembler_name (decl: new_node->decl, name: fname); |
297 | } |
298 | return new_node; |
299 | } |
300 | |
301 | /* If the function in NODE has multiple target attributes |
302 | create the appropriate clone for each valid target attribute. */ |
303 | |
304 | static bool |
305 | expand_target_clones (struct cgraph_node *node, bool definition) |
306 | { |
307 | int i; |
308 | /* Parsing target attributes separated by comma. */ |
309 | tree attr_target = lookup_attribute (attr_name: "target_clones" , |
310 | DECL_ATTRIBUTES (node->decl)); |
311 | /* No targets specified. */ |
312 | if (!attr_target) |
313 | return false; |
314 | |
315 | tree arglist = TREE_VALUE (attr_target); |
316 | int attr_len = get_target_clone_attr_len (arglist); |
317 | |
318 | /* No need to clone for 1 target attribute. */ |
319 | if (attr_len == -1) |
320 | { |
321 | warning_at (DECL_SOURCE_LOCATION (node->decl), |
322 | 0, "single %<target_clones%> attribute is ignored" ); |
323 | return false; |
324 | } |
325 | |
326 | if (node->definition |
327 | && (node->alias || !tree_versionable_function_p (node->decl))) |
328 | { |
329 | auto_diagnostic_group d; |
330 | error_at (DECL_SOURCE_LOCATION (node->decl), |
331 | "clones for %<target_clones%> attribute cannot be created" ); |
332 | const char *reason = NULL; |
333 | if (lookup_attribute (attr_name: "noclone" , DECL_ATTRIBUTES (node->decl))) |
334 | reason = G_("function %q+F can never be copied " |
335 | "because it has %<noclone%> attribute" ); |
336 | else if (node->alias) |
337 | reason |
338 | = "%<target_clones%> cannot be combined with %<alias%> attribute" ; |
339 | else |
340 | reason = copy_forbidden (DECL_STRUCT_FUNCTION (node->decl)); |
341 | if (reason) |
342 | inform (DECL_SOURCE_LOCATION (node->decl), reason, node->decl); |
343 | return false; |
344 | } |
345 | |
346 | char *attr_str = XNEWVEC (char, attr_len); |
347 | int attrnum = get_attr_str (arglist, attr_str); |
348 | char **attrs = XNEWVEC (char *, attrnum); |
349 | |
350 | attrnum = separate_attrs (attr_str, attrs, attrnum); |
351 | switch (attrnum) |
352 | { |
353 | case -1: |
354 | error_at (DECL_SOURCE_LOCATION (node->decl), |
355 | "%<default%> target was not set" ); |
356 | break; |
357 | case -2: |
358 | error_at (DECL_SOURCE_LOCATION (node->decl), |
359 | "an empty string cannot be in %<target_clones%> attribute" ); |
360 | break; |
361 | case -3: |
362 | error_at (DECL_SOURCE_LOCATION (node->decl), |
363 | "multiple %<default%> targets were set" ); |
364 | break; |
365 | default: |
366 | break; |
367 | } |
368 | |
369 | if (attrnum < 0) |
370 | { |
371 | XDELETEVEC (attrs); |
372 | XDELETEVEC (attr_str); |
373 | return false; |
374 | } |
375 | |
376 | const char *new_attr_name = (TARGET_HAS_FMV_TARGET_ATTRIBUTE |
377 | ? "target" : "target_version" ); |
378 | cgraph_function_version_info *decl1_v = NULL; |
379 | cgraph_function_version_info *decl2_v = NULL; |
380 | cgraph_function_version_info *before = NULL; |
381 | cgraph_function_version_info *after = NULL; |
382 | decl1_v = node->function_version (); |
383 | if (decl1_v == NULL) |
384 | decl1_v = node->insert_new_function_version (); |
385 | before = decl1_v; |
386 | DECL_FUNCTION_VERSIONED (node->decl) = 1; |
387 | |
388 | for (i = 0; i < attrnum; i++) |
389 | { |
390 | char *attr = attrs[i]; |
391 | |
392 | /* Create new target clone. */ |
393 | tree attributes = make_attribute (new_attr_name, attr, |
394 | DECL_ATTRIBUTES (node->decl)); |
395 | |
396 | char *suffix = XNEWVEC (char, strlen (attr) + 1); |
397 | create_new_asm_name (old_asm_name: attr, new_asm_name: suffix); |
398 | cgraph_node *new_node = create_target_clone (node, definition, name: suffix, |
399 | attributes); |
400 | XDELETEVEC (suffix); |
401 | if (new_node == NULL) |
402 | { |
403 | XDELETEVEC (attrs); |
404 | XDELETEVEC (attr_str); |
405 | return false; |
406 | } |
407 | new_node->local = false; |
408 | |
409 | decl2_v = new_node->function_version (); |
410 | if (decl2_v != NULL) |
411 | continue; |
412 | decl2_v = new_node->insert_new_function_version (); |
413 | |
414 | /* Chain decl2_v and decl1_v. All semantically identical versions |
415 | will be chained together. */ |
416 | after = decl2_v; |
417 | while (before->next != NULL) |
418 | before = before->next; |
419 | while (after->prev != NULL) |
420 | after = after->prev; |
421 | |
422 | before->next = after; |
423 | after->prev = before; |
424 | DECL_FUNCTION_VERSIONED (new_node->decl) = 1; |
425 | } |
426 | |
427 | XDELETEVEC (attrs); |
428 | XDELETEVEC (attr_str); |
429 | |
430 | /* Setting new attribute to initial function. */ |
431 | tree attributes = make_attribute (new_attr_name, "default" , |
432 | DECL_ATTRIBUTES (node->decl)); |
433 | DECL_ATTRIBUTES (node->decl) = attributes; |
434 | node->local = false; |
435 | return true; |
436 | } |
437 | |
438 | /* When NODE is a target clone, consider all callees and redirect |
439 | to a clone with equal target attributes. That prevents multiple |
440 | multi-versioning dispatches and a call-chain can be optimized. */ |
441 | |
442 | static void |
443 | redirect_to_specific_clone (cgraph_node *node) |
444 | { |
445 | cgraph_function_version_info *fv = node->function_version (); |
446 | if (fv == NULL) |
447 | return; |
448 | |
449 | tree attr_target = lookup_attribute (attr_name: "target" , DECL_ATTRIBUTES (node->decl)); |
450 | if (attr_target == NULL_TREE) |
451 | return; |
452 | |
453 | /* We need to remember NEXT_CALLER as it could be modified in the loop. */ |
454 | for (cgraph_edge *e = node->callees; e ; e = e->next_callee) |
455 | { |
456 | cgraph_function_version_info *fv2 = e->callee->function_version (); |
457 | if (!fv2) |
458 | continue; |
459 | |
460 | tree attr_target2 = lookup_attribute (attr_name: "target" , |
461 | DECL_ATTRIBUTES (e->callee->decl)); |
462 | |
463 | /* Function is not calling proper target clone. */ |
464 | if (attr_target2 == NULL_TREE |
465 | || !attribute_value_equal (attr_target, attr_target2)) |
466 | { |
467 | while (fv2->prev != NULL) |
468 | fv2 = fv2->prev; |
469 | |
470 | /* Try to find a clone with equal target attribute. */ |
471 | for (; fv2 != NULL; fv2 = fv2->next) |
472 | { |
473 | cgraph_node *callee = fv2->this_node; |
474 | attr_target2 = lookup_attribute (attr_name: "target" , |
475 | DECL_ATTRIBUTES (callee->decl)); |
476 | if (attr_target2 != NULL_TREE |
477 | && attribute_value_equal (attr_target, attr_target2)) |
478 | { |
479 | e->redirect_callee (n: callee); |
480 | cgraph_edge::redirect_call_stmt_to_callee (e); |
481 | break; |
482 | } |
483 | } |
484 | } |
485 | } |
486 | } |
487 | |
488 | static unsigned int |
489 | ipa_target_clone (void) |
490 | { |
491 | struct cgraph_node *node; |
492 | auto_vec<cgraph_node *> to_dispatch; |
493 | |
494 | FOR_EACH_FUNCTION (node) |
495 | if (expand_target_clones (node, definition: node->definition)) |
496 | to_dispatch.safe_push (obj: node); |
497 | |
498 | for (unsigned i = 0; i < to_dispatch.length (); i++) |
499 | create_dispatcher_calls (node: to_dispatch[i]); |
500 | |
501 | FOR_EACH_FUNCTION (node) |
502 | redirect_to_specific_clone (node); |
503 | |
504 | return 0; |
505 | } |
506 | |
507 | namespace { |
508 | |
509 | const pass_data pass_data_target_clone = |
510 | { |
511 | .type: SIMPLE_IPA_PASS, /* type */ |
512 | .name: "targetclone" , /* name */ |
513 | .optinfo_flags: OPTGROUP_NONE, /* optinfo_flags */ |
514 | .tv_id: TV_NONE, /* tv_id */ |
515 | .properties_required: ( PROP_ssa | PROP_cfg ), /* properties_required */ |
516 | .properties_provided: 0, /* properties_provided */ |
517 | .properties_destroyed: 0, /* properties_destroyed */ |
518 | .todo_flags_start: 0, /* todo_flags_start */ |
519 | TODO_update_ssa /* todo_flags_finish */ |
520 | }; |
521 | |
522 | class pass_target_clone : public simple_ipa_opt_pass |
523 | { |
524 | public: |
525 | pass_target_clone (gcc::context *ctxt) |
526 | : simple_ipa_opt_pass (pass_data_target_clone, ctxt) |
527 | {} |
528 | |
529 | /* opt_pass methods: */ |
530 | bool gate (function *) final override; |
531 | unsigned int execute (function *) final override |
532 | { |
533 | return ipa_target_clone (); |
534 | } |
535 | }; |
536 | |
537 | bool |
538 | pass_target_clone::gate (function *) |
539 | { |
540 | /* If there were any errors avoid pass property verification errors. */ |
541 | return !seen_error (); |
542 | } |
543 | |
544 | } // anon namespace |
545 | |
546 | simple_ipa_opt_pass * |
547 | make_pass_target_clone (gcc::context *ctxt) |
548 | { |
549 | return new pass_target_clone (ctxt); |
550 | } |
551 | |