|
0
|
1 /* The tracer pass for the GNU compiler.
|
|
|
2 Contributed by Jan Hubicka, SuSE Labs.
|
|
|
3 Adapted to work on GIMPLE instead of RTL by Robert Kidd, UIUC.
|
|
145
|
4 Copyright (C) 2001-2020 Free Software Foundation, Inc.
|
|
0
|
5
|
|
|
6 This file is part of GCC.
|
|
|
7
|
|
|
8 GCC is free software; you can redistribute it and/or modify it
|
|
|
9 under the terms of the GNU General Public License as published by
|
|
|
10 the Free Software Foundation; either version 3, or (at your option)
|
|
|
11 any later version.
|
|
|
12
|
|
|
13 GCC is distributed in the hope that it will be useful, but WITHOUT
|
|
|
14 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
|
15 or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
|
16 License for more details.
|
|
|
17
|
|
|
18 You should have received a copy of the GNU General Public License
|
|
|
19 along with GCC; see the file COPYING3. If not see
|
|
|
20 <http://www.gnu.org/licenses/>. */
|
|
|
21
|
|
|
22 /* This pass performs the tail duplication needed for superblock formation.
|
|
|
23 For more information see:
|
|
|
24
|
|
|
25 Design and Analysis of Profile-Based Optimization in Compaq's
|
|
|
26 Compilation Tools for Alpha; Journal of Instruction-Level
|
|
|
27 Parallelism 3 (2000) 1-25
|
|
|
28
|
|
|
29 Unlike Compaq's implementation we don't do the loop peeling as most
|
|
|
30 probably a better job can be done by a special pass and we don't
|
|
|
31 need to worry too much about the code size implications as the tail
|
|
|
32 duplicates are crossjumped again if optimizations are not
|
|
|
33 performed. */
|
|
|
34
|
|
|
35
|
|
|
36 #include "config.h"
|
|
|
37 #include "system.h"
|
|
|
38 #include "coretypes.h"
|
|
111
|
39 #include "backend.h"
|
|
0
|
40 #include "rtl.h"
|
|
111
|
41 #include "tree.h"
|
|
|
42 #include "gimple.h"
|
|
|
43 #include "cfghooks.h"
|
|
|
44 #include "tree-pass.h"
|
|
|
45 #include "profile.h"
|
|
|
46 #include "cfganal.h"
|
|
|
47 #include "gimple-iterator.h"
|
|
|
48 #include "tree-cfg.h"
|
|
|
49 #include "tree-ssa.h"
|
|
0
|
50 #include "tree-inline.h"
|
|
111
|
51 #include "cfgloop.h"
|
|
145
|
52 #include "alloc-pool.h"
|
|
111
|
53 #include "fibonacci_heap.h"
|
|
|
54 #include "tracer.h"
|
|
0
|
55
|
|
|
56 static int count_insns (basic_block);
|
|
|
57 static bool better_p (const_edge, const_edge);
|
|
|
58 static edge find_best_successor (basic_block);
|
|
|
59 static edge find_best_predecessor (basic_block);
|
|
|
60 static int find_trace (basic_block, basic_block *);
|
|
|
61
|
|
|
62 /* Minimal outgoing edge probability considered for superblock formation. */
|
|
|
63 static int probability_cutoff;
|
|
|
64 static int branch_ratio_cutoff;
|
|
|
65
|
|
|
66 /* A bit BB->index is set if BB has already been seen, i.e. it is
|
|
|
67 connected to some trace already. */
|
|
111
|
68 static sbitmap bb_seen;
|
|
0
|
69
|
|
|
70 static inline void
|
|
|
71 mark_bb_seen (basic_block bb)
|
|
|
72 {
|
|
111
|
73 unsigned int size = SBITMAP_SIZE (bb_seen);
|
|
0
|
74
|
|
|
75 if ((unsigned int)bb->index >= size)
|
|
|
76 bb_seen = sbitmap_resize (bb_seen, size * 2, 0);
|
|
|
77
|
|
111
|
78 bitmap_set_bit (bb_seen, bb->index);
|
|
0
|
79 }
|
|
|
80
|
|
|
81 static inline bool
|
|
|
82 bb_seen_p (basic_block bb)
|
|
|
83 {
|
|
111
|
84 return bitmap_bit_p (bb_seen, bb->index);
|
|
0
|
85 }
|
|
|
86
|
|
|
87 /* Return true if we should ignore the basic block for purposes of tracing. */
|
|
111
|
88 bool
|
|
0
|
89 ignore_bb_p (const_basic_block bb)
|
|
|
90 {
|
|
|
91 if (bb->index < NUM_FIXED_BLOCKS)
|
|
|
92 return true;
|
|
|
93 if (optimize_bb_for_size_p (bb))
|
|
|
94 return true;
|
|
111
|
95
|
|
|
96 if (gimple *g = last_stmt (CONST_CAST_BB (bb)))
|
|
|
97 {
|
|
|
98 /* A transaction is a single entry multiple exit region. It
|
|
|
99 must be duplicated in its entirety or not at all. */
|
|
|
100 if (gimple_code (g) == GIMPLE_TRANSACTION)
|
|
|
101 return true;
|
|
|
102
|
|
|
103 /* An IFN_UNIQUE call must be duplicated as part of its group,
|
|
|
104 or not at all. */
|
|
|
105 if (is_gimple_call (g)
|
|
|
106 && gimple_call_internal_p (g)
|
|
|
107 && gimple_call_internal_unique_p (g))
|
|
|
108 return true;
|
|
|
109 }
|
|
|
110
|
|
0
|
111 return false;
|
|
|
112 }
|
|
|
113
|
|
|
114 /* Return number of instructions in the block. */
|
|
|
115
|
|
|
116 static int
|
|
|
117 count_insns (basic_block bb)
|
|
|
118 {
|
|
|
119 gimple_stmt_iterator gsi;
|
|
111
|
120 gimple *stmt;
|
|
0
|
121 int n = 0;
|
|
|
122
|
|
|
123 for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi))
|
|
|
124 {
|
|
|
125 stmt = gsi_stmt (gsi);
|
|
|
126 n += estimate_num_insns (stmt, &eni_size_weights);
|
|
|
127 }
|
|
|
128 return n;
|
|
|
129 }
|
|
|
130
|
|
|
131 /* Return true if E1 is more frequent than E2. */
|
|
|
132 static bool
|
|
|
133 better_p (const_edge e1, const_edge e2)
|
|
|
134 {
|
|
131
|
135 if ((e1->count () > e2->count ()) || (e1->count () < e2->count ()))
|
|
111
|
136 return e1->count () > e2->count ();
|
|
0
|
137 /* This is needed to avoid changes in the decision after
|
|
|
138 CFG is modified. */
|
|
|
139 if (e1->src != e2->src)
|
|
|
140 return e1->src->index > e2->src->index;
|
|
|
141 return e1->dest->index > e2->dest->index;
|
|
|
142 }
|
|
|
143
|
|
|
144 /* Return most frequent successor of basic block BB. */
|
|
|
145
|
|
|
146 static edge
|
|
|
147 find_best_successor (basic_block bb)
|
|
|
148 {
|
|
|
149 edge e;
|
|
|
150 edge best = NULL;
|
|
|
151 edge_iterator ei;
|
|
|
152
|
|
|
153 FOR_EACH_EDGE (e, ei, bb->succs)
|
|
131
|
154 {
|
|
|
155 if (!e->count ().initialized_p ())
|
|
|
156 return NULL;
|
|
|
157 if (!best || better_p (e, best))
|
|
|
158 best = e;
|
|
|
159 }
|
|
0
|
160 if (!best || ignore_bb_p (best->dest))
|
|
|
161 return NULL;
|
|
131
|
162 if (!best->probability.initialized_p ()
|
|
|
163 || best->probability.to_reg_br_prob_base () <= probability_cutoff)
|
|
0
|
164 return NULL;
|
|
|
165 return best;
|
|
|
166 }
|
|
|
167
|
|
|
168 /* Return most frequent predecessor of basic block BB. */
|
|
|
169
|
|
|
170 static edge
|
|
|
171 find_best_predecessor (basic_block bb)
|
|
|
172 {
|
|
|
173 edge e;
|
|
|
174 edge best = NULL;
|
|
|
175 edge_iterator ei;
|
|
|
176
|
|
|
177 FOR_EACH_EDGE (e, ei, bb->preds)
|
|
131
|
178 {
|
|
|
179 if (!e->count ().initialized_p ())
|
|
|
180 return NULL;
|
|
|
181 if (!best || better_p (e, best))
|
|
|
182 best = e;
|
|
|
183 }
|
|
0
|
184 if (!best || ignore_bb_p (best->src))
|
|
|
185 return NULL;
|
|
131
|
186 if (bb->count.initialized_p ()
|
|
|
187 && (best->count ().to_frequency (cfun) * REG_BR_PROB_BASE
|
|
|
188 < bb->count.to_frequency (cfun) * branch_ratio_cutoff))
|
|
0
|
189 return NULL;
|
|
|
190 return best;
|
|
|
191 }
|
|
|
192
|
|
|
193 /* Find the trace using bb and record it in the TRACE array.
|
|
|
194 Return number of basic blocks recorded. */
|
|
|
195
|
|
|
196 static int
|
|
|
197 find_trace (basic_block bb, basic_block *trace)
|
|
|
198 {
|
|
|
199 int i = 0;
|
|
|
200 edge e;
|
|
|
201
|
|
|
202 if (dump_file)
|
|
131
|
203 fprintf (dump_file, "Trace seed %i [%i]", bb->index, bb->count.to_frequency (cfun));
|
|
0
|
204
|
|
|
205 while ((e = find_best_predecessor (bb)) != NULL)
|
|
|
206 {
|
|
|
207 basic_block bb2 = e->src;
|
|
|
208 if (bb_seen_p (bb2) || (e->flags & (EDGE_DFS_BACK | EDGE_COMPLEX))
|
|
|
209 || find_best_successor (bb2) != e)
|
|
|
210 break;
|
|
|
211 if (dump_file)
|
|
131
|
212 fprintf (dump_file, ",%i [%i]", bb->index, bb->count.to_frequency (cfun));
|
|
0
|
213 bb = bb2;
|
|
|
214 }
|
|
|
215 if (dump_file)
|
|
131
|
216 fprintf (dump_file, " forward %i [%i]", bb->index, bb->count.to_frequency (cfun));
|
|
0
|
217 trace[i++] = bb;
|
|
|
218
|
|
|
219 /* Follow the trace in forward direction. */
|
|
|
220 while ((e = find_best_successor (bb)) != NULL)
|
|
|
221 {
|
|
|
222 bb = e->dest;
|
|
|
223 if (bb_seen_p (bb) || (e->flags & (EDGE_DFS_BACK | EDGE_COMPLEX))
|
|
|
224 || find_best_predecessor (bb) != e)
|
|
|
225 break;
|
|
|
226 if (dump_file)
|
|
131
|
227 fprintf (dump_file, ",%i [%i]", bb->index, bb->count.to_frequency (cfun));
|
|
0
|
228 trace[i++] = bb;
|
|
|
229 }
|
|
|
230 if (dump_file)
|
|
|
231 fprintf (dump_file, "\n");
|
|
|
232 return i;
|
|
|
233 }
|
|
|
234
|
|
111
|
235 /* Duplicate block BB2, placing it after BB in the CFG. Return the
|
|
|
236 newly created block. */
|
|
|
237 basic_block
|
|
|
238 transform_duplicate (basic_block bb, basic_block bb2)
|
|
|
239 {
|
|
|
240 edge e;
|
|
|
241 basic_block copy;
|
|
|
242
|
|
|
243 e = find_edge (bb, bb2);
|
|
|
244
|
|
|
245 copy = duplicate_block (bb2, e, bb);
|
|
|
246 flush_pending_stmts (e);
|
|
|
247
|
|
|
248 add_phi_args_after_copy (©, 1, NULL);
|
|
|
249
|
|
|
250 return (copy);
|
|
|
251 }
|
|
|
252
|
|
0
|
253 /* Look for basic blocks in frequency order, construct traces and tail duplicate
|
|
|
254 if profitable. */
|
|
|
255
|
|
111
|
256 static bool
|
|
0
|
257 tail_duplicate (void)
|
|
|
258 {
|
|
111
|
259 auto_vec<fibonacci_node<long, basic_block_def>*> blocks;
|
|
|
260 blocks.safe_grow_cleared (last_basic_block_for_fn (cfun));
|
|
|
261
|
|
|
262 basic_block *trace = XNEWVEC (basic_block, n_basic_blocks_for_fn (cfun));
|
|
|
263 int *counts = XNEWVEC (int, last_basic_block_for_fn (cfun));
|
|
0
|
264 int ninsns = 0, nduplicated = 0;
|
|
|
265 gcov_type weighted_insns = 0, traced_insns = 0;
|
|
111
|
266 fibonacci_heap<long, basic_block_def> heap (LONG_MIN);
|
|
0
|
267 gcov_type cover_insns;
|
|
|
268 int max_dup_insns;
|
|
|
269 basic_block bb;
|
|
111
|
270 bool changed = false;
|
|
0
|
271
|
|
|
272 /* Create an oversized sbitmap to reduce the chance that we need to
|
|
|
273 resize it. */
|
|
111
|
274 bb_seen = sbitmap_alloc (last_basic_block_for_fn (cfun) * 2);
|
|
|
275 bitmap_clear (bb_seen);
|
|
0
|
276 initialize_original_copy_tables ();
|
|
|
277
|
|
111
|
278 if (profile_info && profile_status_for_fn (cfun) == PROFILE_READ)
|
|
145
|
279 probability_cutoff = param_tracer_min_branch_probability_feedback;
|
|
0
|
280 else
|
|
145
|
281 probability_cutoff = param_tracer_min_branch_probability;
|
|
0
|
282 probability_cutoff = REG_BR_PROB_BASE / 100 * probability_cutoff;
|
|
|
283
|
|
|
284 branch_ratio_cutoff =
|
|
145
|
285 (REG_BR_PROB_BASE / 100 * param_tracer_min_branch_ratio);
|
|
0
|
286
|
|
111
|
287 FOR_EACH_BB_FN (bb, cfun)
|
|
0
|
288 {
|
|
|
289 int n = count_insns (bb);
|
|
|
290 if (!ignore_bb_p (bb))
|
|
131
|
291 blocks[bb->index] = heap.insert (-bb->count.to_frequency (cfun), bb);
|
|
0
|
292
|
|
|
293 counts [bb->index] = n;
|
|
|
294 ninsns += n;
|
|
131
|
295 weighted_insns += n * bb->count.to_frequency (cfun);
|
|
0
|
296 }
|
|
|
297
|
|
111
|
298 if (profile_info && profile_status_for_fn (cfun) == PROFILE_READ)
|
|
145
|
299 cover_insns = param_tracer_dynamic_coverage_feedback;
|
|
0
|
300 else
|
|
145
|
301 cover_insns = param_tracer_dynamic_coverage;
|
|
0
|
302 cover_insns = (weighted_insns * cover_insns + 50) / 100;
|
|
145
|
303 max_dup_insns = (ninsns * param_tracer_max_code_growth + 50) / 100;
|
|
0
|
304
|
|
|
305 while (traced_insns < cover_insns && nduplicated < max_dup_insns
|
|
111
|
306 && !heap.empty ())
|
|
0
|
307 {
|
|
111
|
308 basic_block bb = heap.extract_min ();
|
|
0
|
309 int n, pos;
|
|
|
310
|
|
|
311 if (!bb)
|
|
|
312 break;
|
|
|
313
|
|
|
314 blocks[bb->index] = NULL;
|
|
|
315
|
|
|
316 if (ignore_bb_p (bb))
|
|
|
317 continue;
|
|
|
318 gcc_assert (!bb_seen_p (bb));
|
|
|
319
|
|
|
320 n = find_trace (bb, trace);
|
|
|
321
|
|
|
322 bb = trace[0];
|
|
131
|
323 traced_insns += bb->count.to_frequency (cfun) * counts [bb->index];
|
|
0
|
324 if (blocks[bb->index])
|
|
|
325 {
|
|
111
|
326 heap.delete_node (blocks[bb->index]);
|
|
0
|
327 blocks[bb->index] = NULL;
|
|
|
328 }
|
|
|
329
|
|
|
330 for (pos = 1; pos < n; pos++)
|
|
|
331 {
|
|
|
332 basic_block bb2 = trace[pos];
|
|
|
333
|
|
|
334 if (blocks[bb2->index])
|
|
|
335 {
|
|
111
|
336 heap.delete_node (blocks[bb2->index]);
|
|
0
|
337 blocks[bb2->index] = NULL;
|
|
|
338 }
|
|
131
|
339 traced_insns += bb2->count.to_frequency (cfun) * counts [bb2->index];
|
|
0
|
340 if (EDGE_COUNT (bb2->preds) > 1
|
|
111
|
341 && can_duplicate_block_p (bb2)
|
|
|
342 /* We have the tendency to duplicate the loop header
|
|
|
343 of all do { } while loops. Do not do that - it is
|
|
|
344 not profitable and it might create a loop with multiple
|
|
|
345 entries or at least rotate the loop. */
|
|
|
346 && bb2->loop_father->header != bb2)
|
|
0
|
347 {
|
|
|
348 nduplicated += counts [bb2->index];
|
|
111
|
349 basic_block copy = transform_duplicate (bb, bb2);
|
|
0
|
350
|
|
|
351 /* Reconsider the original copy of block we've duplicated.
|
|
|
352 Removing the most common predecessor may make it to be
|
|
|
353 head. */
|
|
131
|
354 blocks[bb2->index] = heap.insert (-bb2->count.to_frequency (cfun), bb2);
|
|
0
|
355
|
|
|
356 if (dump_file)
|
|
|
357 fprintf (dump_file, "Duplicated %i as %i [%i]\n",
|
|
131
|
358 bb2->index, copy->index, copy->count.to_frequency (cfun));
|
|
0
|
359
|
|
|
360 bb2 = copy;
|
|
111
|
361 changed = true;
|
|
0
|
362 }
|
|
|
363 mark_bb_seen (bb2);
|
|
|
364 bb = bb2;
|
|
|
365 /* In case the trace became infrequent, stop duplicating. */
|
|
|
366 if (ignore_bb_p (bb))
|
|
|
367 break;
|
|
|
368 }
|
|
|
369 if (dump_file)
|
|
|
370 fprintf (dump_file, " covered now %.1f\n\n",
|
|
|
371 traced_insns * 100.0 / weighted_insns);
|
|
|
372 }
|
|
|
373 if (dump_file)
|
|
|
374 fprintf (dump_file, "Duplicated %i insns (%i%%)\n", nduplicated,
|
|
|
375 nduplicated * 100 / ninsns);
|
|
|
376
|
|
|
377 free_original_copy_tables ();
|
|
|
378 sbitmap_free (bb_seen);
|
|
|
379 free (trace);
|
|
|
380 free (counts);
|
|
111
|
381
|
|
|
382 return changed;
|
|
0
|
383 }
|
|
111
|
384
|
|
|
385 namespace {
|
|
0
|
386
|
|
111
|
387 const pass_data pass_data_tracer =
|
|
|
388 {
|
|
|
389 GIMPLE_PASS, /* type */
|
|
|
390 "tracer", /* name */
|
|
|
391 OPTGROUP_NONE, /* optinfo_flags */
|
|
|
392 TV_TRACER, /* tv_id */
|
|
|
393 0, /* properties_required */
|
|
|
394 0, /* properties_provided */
|
|
|
395 0, /* properties_destroyed */
|
|
|
396 0, /* todo_flags_start */
|
|
|
397 TODO_update_ssa, /* todo_flags_finish */
|
|
|
398 };
|
|
0
|
399
|
|
111
|
400 class pass_tracer : public gimple_opt_pass
|
|
0
|
401 {
|
|
111
|
402 public:
|
|
|
403 pass_tracer (gcc::context *ctxt)
|
|
|
404 : gimple_opt_pass (pass_data_tracer, ctxt)
|
|
|
405 {}
|
|
|
406
|
|
|
407 /* opt_pass methods: */
|
|
|
408 virtual bool gate (function *)
|
|
|
409 {
|
|
|
410 return (optimize > 0 && flag_tracer && flag_reorder_blocks);
|
|
|
411 }
|
|
0
|
412
|
|
111
|
413 virtual unsigned int execute (function *);
|
|
|
414
|
|
|
415 }; // class pass_tracer
|
|
|
416
|
|
|
417 unsigned int
|
|
|
418 pass_tracer::execute (function *fun)
|
|
|
419 {
|
|
|
420 bool changed;
|
|
|
421
|
|
|
422 if (n_basic_blocks_for_fn (fun) <= NUM_FIXED_BLOCKS + 1)
|
|
0
|
423 return 0;
|
|
|
424
|
|
|
425 mark_dfs_back_edges ();
|
|
|
426 if (dump_file)
|
|
111
|
427 brief_dump_cfg (dump_file, dump_flags);
|
|
0
|
428
|
|
|
429 /* Trace formation is done on the fly inside tail_duplicate */
|
|
111
|
430 changed = tail_duplicate ();
|
|
|
431 if (changed)
|
|
|
432 {
|
|
|
433 free_dominance_info (CDI_DOMINATORS);
|
|
|
434 /* If we changed the CFG schedule loops for fixup by cleanup_cfg. */
|
|
|
435 loops_state_set (LOOPS_NEED_FIXUP);
|
|
|
436 }
|
|
0
|
437
|
|
111
|
438 if (dump_file)
|
|
|
439 brief_dump_cfg (dump_file, dump_flags);
|
|
|
440
|
|
|
441 return changed ? TODO_cleanup_cfg : 0;
|
|
0
|
442 }
|
|
111
|
443 } // anon namespace
|
|
0
|
444
|
|
111
|
445 gimple_opt_pass *
|
|
|
446 make_pass_tracer (gcc::context *ctxt)
|
|
0
|
447 {
|
|
111
|
448 return new pass_tracer (ctxt);
|
|
|
449 }
|
