]>
Commit | Line | Data |
---|---|---|
f3d9a16c | 1 | /* Loop header copying on trees. |
d353bf18 | 2 | Copyright (C) 2004-2015 Free Software Foundation, Inc. |
48e1416a | 3 | |
f3d9a16c | 4 | This file is part of GCC. |
48e1416a | 5 | |
f3d9a16c | 6 | GCC is free software; you can redistribute it and/or modify it |
7 | under the terms of the GNU General Public License as published by the | |
8c4c00c1 | 8 | Free Software Foundation; either version 3, or (at your option) any |
f3d9a16c | 9 | later version. |
48e1416a | 10 | |
f3d9a16c | 11 | GCC is distributed in the hope that it will be useful, but WITHOUT |
12 | ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
13 | FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
14 | for more details. | |
48e1416a | 15 | |
f3d9a16c | 16 | You should have received a copy of the GNU General Public License |
8c4c00c1 | 17 | along with GCC; see the file COPYING3. If not see |
18 | <http://www.gnu.org/licenses/>. */ | |
f3d9a16c | 19 | |
20 | #include "config.h" | |
21 | #include "system.h" | |
22 | #include "coretypes.h" | |
23 | #include "tm.h" | |
b20a8bb4 | 24 | #include "alias.h" |
25 | #include "symtab.h" | |
f3d9a16c | 26 | #include "tree.h" |
b20a8bb4 | 27 | #include "fold-const.h" |
f3d9a16c | 28 | #include "tm_p.h" |
94ea8568 | 29 | #include "predict.h" |
94ea8568 | 30 | #include "hard-reg-set.h" |
94ea8568 | 31 | #include "function.h" |
32 | #include "dominance.h" | |
33 | #include "cfg.h" | |
f3d9a16c | 34 | #include "basic-block.h" |
bc61cadb | 35 | #include "tree-ssa-alias.h" |
36 | #include "internal-fn.h" | |
37 | #include "gimple-expr.h" | |
073c1fd5 | 38 | #include "gimple.h" |
dcf1a1ec | 39 | #include "gimple-iterator.h" |
073c1fd5 | 40 | #include "gimple-ssa.h" |
41 | #include "tree-cfg.h" | |
42 | #include "tree-into-ssa.h" | |
f3d9a16c | 43 | #include "tree-pass.h" |
f3d9a16c | 44 | #include "cfgloop.h" |
45 | #include "tree-inline.h" | |
46 | #include "flags.h" | |
545372c5 | 47 | #include "tree-ssa-scopedtables.h" |
424a4a92 | 48 | #include "tree-ssa-threadedge.h" |
f3d9a16c | 49 | |
50 | /* Duplicates headers of loops if they are small enough, so that the statements | |
51 | in the loop body are always executed when the loop is entered. This | |
c26a6416 | 52 | increases effectiveness of code motion optimizations, and reduces the need |
f3d9a16c | 53 | for loop preconditioning. */ |
54 | ||
55 | /* Check whether we should duplicate HEADER of LOOP. At most *LIMIT | |
56 | instructions should be duplicated, limit is decreased by the actual | |
57 | amount. */ | |
58 | ||
59 | static bool | |
60 | should_duplicate_loop_header_p (basic_block header, struct loop *loop, | |
61 | int *limit) | |
62 | { | |
75a70cf9 | 63 | gimple_stmt_iterator bsi; |
64 | gimple last; | |
f3d9a16c | 65 | |
66 | /* Do not copy one block more than once (we do not really want to do | |
67 | loop peeling here). */ | |
68 | if (header->aux) | |
69 | return false; | |
70 | ||
94ba1cf1 | 71 | /* Loop header copying usually increases size of the code. This used not to |
72 | be true, since quite often it is possible to verify that the condition is | |
73 | satisfied in the first iteration and therefore to eliminate it. Jump | |
74 | threading handles these cases now. */ | |
75 | if (optimize_loop_for_size_p (loop)) | |
76 | return false; | |
77 | ||
cd665a06 | 78 | gcc_assert (EDGE_COUNT (header->succs) > 0); |
ea091dfd | 79 | if (single_succ_p (header)) |
f3d9a16c | 80 | return false; |
cd665a06 | 81 | if (flow_bb_inside_loop_p (loop, EDGE_SUCC (header, 0)->dest) |
82 | && flow_bb_inside_loop_p (loop, EDGE_SUCC (header, 1)->dest)) | |
f3d9a16c | 83 | return false; |
84 | ||
85 | /* If this is not the original loop header, we want it to have just | |
86 | one predecessor in order to match the && pattern. */ | |
ea091dfd | 87 | if (header != loop->header && !single_pred_p (header)) |
f3d9a16c | 88 | return false; |
89 | ||
90 | last = last_stmt (header); | |
75a70cf9 | 91 | if (gimple_code (last) != GIMPLE_COND) |
f3d9a16c | 92 | return false; |
93 | ||
94 | /* Approximately copy the conditions that used to be used in jump.c -- | |
95 | at most 20 insns and no calls. */ | |
75a70cf9 | 96 | for (bsi = gsi_start_bb (header); !gsi_end_p (bsi); gsi_next (&bsi)) |
f3d9a16c | 97 | { |
75a70cf9 | 98 | last = gsi_stmt (bsi); |
f3d9a16c | 99 | |
75a70cf9 | 100 | if (gimple_code (last) == GIMPLE_LABEL) |
f3d9a16c | 101 | continue; |
102 | ||
9845d120 | 103 | if (is_gimple_debug (last)) |
104 | continue; | |
105 | ||
75a70cf9 | 106 | if (is_gimple_call (last)) |
f3d9a16c | 107 | return false; |
108 | ||
bc8bb825 | 109 | *limit -= estimate_num_insns (last, &eni_size_weights); |
f3d9a16c | 110 | if (*limit < 0) |
111 | return false; | |
112 | } | |
113 | ||
114 | return true; | |
115 | } | |
116 | ||
f3d9a16c | 117 | /* Checks whether LOOP is a do-while style loop. */ |
118 | ||
f86b328b | 119 | static bool |
f3d9a16c | 120 | do_while_loop_p (struct loop *loop) |
121 | { | |
75a70cf9 | 122 | gimple stmt = last_stmt (loop->latch); |
f3d9a16c | 123 | |
124 | /* If the latch of the loop is not empty, it is not a do-while loop. */ | |
125 | if (stmt | |
75a70cf9 | 126 | && gimple_code (stmt) != GIMPLE_LABEL) |
f3d9a16c | 127 | return false; |
128 | ||
129 | /* If the header contains just a condition, it is not a do-while loop. */ | |
130 | stmt = last_and_only_stmt (loop->header); | |
131 | if (stmt | |
75a70cf9 | 132 | && gimple_code (stmt) == GIMPLE_COND) |
f3d9a16c | 133 | return false; |
134 | ||
135 | return true; | |
136 | } | |
137 | ||
138 | /* For all loops, copy the condition at the end of the loop body in front | |
139 | of the loop. This is beneficial since it increases efficiency of | |
140 | code motion optimizations. It also saves one jump on entry to the loop. */ | |
141 | ||
65b0537f | 142 | namespace { |
143 | ||
144 | const pass_data pass_data_ch = | |
145 | { | |
146 | GIMPLE_PASS, /* type */ | |
147 | "ch", /* name */ | |
148 | OPTGROUP_LOOP, /* optinfo_flags */ | |
65b0537f | 149 | TV_TREE_CH, /* tv_id */ |
150 | ( PROP_cfg | PROP_ssa ), /* properties_required */ | |
151 | 0, /* properties_provided */ | |
152 | 0, /* properties_destroyed */ | |
153 | 0, /* todo_flags_start */ | |
051d7389 | 154 | 0, /* todo_flags_finish */ |
65b0537f | 155 | }; |
156 | ||
157 | class pass_ch : public gimple_opt_pass | |
158 | { | |
159 | public: | |
160 | pass_ch (gcc::context *ctxt) | |
161 | : gimple_opt_pass (pass_data_ch, ctxt) | |
162 | {} | |
163 | ||
164 | /* opt_pass methods: */ | |
165 | virtual bool gate (function *) { return flag_tree_ch != 0; } | |
166 | virtual unsigned int execute (function *); | |
167 | ||
168 | }; // class pass_ch | |
169 | ||
170 | unsigned int | |
171 | pass_ch::execute (function *fun) | |
f3d9a16c | 172 | { |
f3d9a16c | 173 | struct loop *loop; |
174 | basic_block header; | |
4d6b11ab | 175 | edge exit, entry; |
176 | basic_block *bbs, *copied_bbs; | |
d8b5b4fe | 177 | unsigned n_bbs; |
4d6b11ab | 178 | unsigned bbs_size; |
051d7389 | 179 | bool changed = false; |
f3d9a16c | 180 | |
88e6f696 | 181 | loop_optimizer_init (LOOPS_HAVE_PREHEADERS |
182 | | LOOPS_HAVE_SIMPLE_LATCHES); | |
65b0537f | 183 | if (number_of_loops (fun) <= 1) |
7a3bf727 | 184 | { |
185 | loop_optimizer_finalize (); | |
186 | return 0; | |
187 | } | |
f3d9a16c | 188 | |
65b0537f | 189 | bbs = XNEWVEC (basic_block, n_basic_blocks_for_fn (fun)); |
190 | copied_bbs = XNEWVEC (basic_block, n_basic_blocks_for_fn (fun)); | |
191 | bbs_size = n_basic_blocks_for_fn (fun); | |
d8b5b4fe | 192 | |
f21d4d00 | 193 | FOR_EACH_LOOP (loop, 0) |
f3d9a16c | 194 | { |
195 | /* Copy at most 20 insns. */ | |
196 | int limit = 20; | |
197 | ||
d8b5b4fe | 198 | header = loop->header; |
f3d9a16c | 199 | |
200 | /* If the loop is already a do-while style one (either because it was | |
201 | written as such, or because jump threading transformed it into one), | |
202 | we might be in fact peeling the first iteration of the loop. This | |
203 | in general is not a good idea. */ | |
204 | if (do_while_loop_p (loop)) | |
205 | continue; | |
206 | ||
207 | /* Iterate the header copying up to limit; this takes care of the cases | |
208 | like while (a && b) {...}, where we want to have both of the conditions | |
209 | copied. TODO -- handle while (a || b) - like cases, by not requiring | |
210 | the header to have just a single successor and copying up to | |
d8b5b4fe | 211 | postdominator. */ |
212 | ||
213 | exit = NULL; | |
214 | n_bbs = 0; | |
f3d9a16c | 215 | while (should_duplicate_loop_header_p (header, loop, &limit)) |
216 | { | |
f3d9a16c | 217 | /* Find a successor of header that is inside a loop; i.e. the new |
218 | header after the condition is copied. */ | |
cd665a06 | 219 | if (flow_bb_inside_loop_p (loop, EDGE_SUCC (header, 0)->dest)) |
220 | exit = EDGE_SUCC (header, 0); | |
f3d9a16c | 221 | else |
cd665a06 | 222 | exit = EDGE_SUCC (header, 1); |
d8b5b4fe | 223 | bbs[n_bbs++] = header; |
4d6b11ab | 224 | gcc_assert (bbs_size > n_bbs); |
d8b5b4fe | 225 | header = exit->dest; |
f3d9a16c | 226 | } |
f3d9a16c | 227 | |
d8b5b4fe | 228 | if (!exit) |
229 | continue; | |
f3d9a16c | 230 | |
d8b5b4fe | 231 | if (dump_file && (dump_flags & TDF_DETAILS)) |
232 | fprintf (dump_file, | |
233 | "Duplicating header of the loop %d up to edge %d->%d.\n", | |
234 | loop->num, exit->src->index, exit->dest->index); | |
235 | ||
236 | /* Ensure that the header will have just the latch as a predecessor | |
237 | inside the loop. */ | |
ea091dfd | 238 | if (!single_pred_p (exit->dest)) |
88e6f696 | 239 | exit = single_pred_edge (split_edge (exit)); |
d8b5b4fe | 240 | |
4d6b11ab | 241 | entry = loop_preheader_edge (loop); |
4d6b11ab | 242 | |
80ed2d81 | 243 | propagate_threaded_block_debug_into (exit->dest, entry->dest); |
d99f53b2 | 244 | if (!gimple_duplicate_sese_region (entry, exit, bbs, n_bbs, copied_bbs, |
245 | true)) | |
d8b5b4fe | 246 | { |
247 | fprintf (dump_file, "Duplication failed.\n"); | |
248 | continue; | |
249 | } | |
250 | ||
c7addd8c | 251 | /* If the loop has the form "for (i = j; i < j + 10; i++)" then |
252 | this copying can introduce a case where we rely on undefined | |
253 | signed overflow to eliminate the preheader condition, because | |
254 | we assume that "j < j + 10" is true. We don't want to warn | |
255 | about that case for -Wstrict-overflow, because in general we | |
256 | don't warn about overflow involving loops. Prevent the | |
75a70cf9 | 257 | warning by setting the no_warning flag in the condition. */ |
c7addd8c | 258 | if (warn_strict_overflow > 0) |
259 | { | |
260 | unsigned int i; | |
261 | ||
262 | for (i = 0; i < n_bbs; ++i) | |
263 | { | |
75a70cf9 | 264 | gimple_stmt_iterator bsi; |
72c59a18 | 265 | |
75a70cf9 | 266 | for (bsi = gsi_start_bb (copied_bbs[i]); |
267 | !gsi_end_p (bsi); | |
268 | gsi_next (&bsi)) | |
72c59a18 | 269 | { |
75a70cf9 | 270 | gimple stmt = gsi_stmt (bsi); |
271 | if (gimple_code (stmt) == GIMPLE_COND) | |
272 | gimple_set_no_warning (stmt, true); | |
273 | else if (is_gimple_assign (stmt)) | |
72c59a18 | 274 | { |
75a70cf9 | 275 | enum tree_code rhs_code = gimple_assign_rhs_code (stmt); |
276 | if (TREE_CODE_CLASS (rhs_code) == tcc_comparison) | |
277 | gimple_set_no_warning (stmt, true); | |
72c59a18 | 278 | } |
279 | } | |
c7addd8c | 280 | } |
281 | } | |
282 | ||
d8b5b4fe | 283 | /* Ensure that the latch and the preheader is simple (we know that they |
284 | are not now, since there was the loop exit condition. */ | |
88e6f696 | 285 | split_edge (loop_preheader_edge (loop)); |
286 | split_edge (loop_latch_edge (loop)); | |
051d7389 | 287 | |
288 | changed = true; | |
f3d9a16c | 289 | } |
290 | ||
8cb0acff | 291 | update_ssa (TODO_update_ssa); |
d8b5b4fe | 292 | free (bbs); |
4d6b11ab | 293 | free (copied_bbs); |
d8b5b4fe | 294 | |
88e6f696 | 295 | loop_optimizer_finalize (); |
051d7389 | 296 | return changed ? TODO_cleanup_cfg : 0; |
f3d9a16c | 297 | } |
298 | ||
cbe8bda8 | 299 | } // anon namespace |
300 | ||
301 | gimple_opt_pass * | |
302 | make_pass_ch (gcc::context *ctxt) | |
303 | { | |
304 | return new pass_ch (ctxt); | |
305 | } |