|
1 /* |
|
2 * Copyright (c) 1998, 2013, Oracle and/or its affiliates. All rights reserved. |
|
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
|
4 * |
|
5 * This code is free software; you can redistribute it and/or modify it |
|
6 * under the terms of the GNU General Public License version 2 only, as |
|
7 * published by the Free Software Foundation. |
|
8 * |
|
9 * This code is distributed in the hope that it will be useful, but WITHOUT |
|
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
|
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
|
12 * version 2 for more details (a copy is included in the LICENSE file that |
|
13 * accompanied this code). |
|
14 * |
|
15 * You should have received a copy of the GNU General Public License version |
|
16 * 2 along with this work; if not, write to the Free Software Foundation, |
|
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
|
18 * |
|
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
|
20 * or visit www.oracle.com if you need additional information or have any |
|
21 * questions. |
|
22 * |
|
23 */ |
|
24 |
|
25 #ifndef SHARE_VM_OPTO_LOOPNODE_HPP |
|
26 #define SHARE_VM_OPTO_LOOPNODE_HPP |
|
27 |
|
28 #include "opto/cfgnode.hpp" |
|
29 #include "opto/multnode.hpp" |
|
30 #include "opto/phaseX.hpp" |
|
31 #include "opto/subnode.hpp" |
|
32 #include "opto/type.hpp" |
|
33 |
|
34 class CmpNode; |
|
35 class CountedLoopEndNode; |
|
36 class CountedLoopNode; |
|
37 class IdealLoopTree; |
|
38 class LoopNode; |
|
39 class Node; |
|
40 class PhaseIdealLoop; |
|
41 class VectorSet; |
|
42 class Invariance; |
|
43 struct small_cache; |
|
44 |
|
45 // |
|
46 // I D E A L I Z E D L O O P S |
|
47 // |
|
48 // Idealized loops are the set of loops I perform more interesting |
|
49 // transformations on, beyond simple hoisting. |
|
50 |
|
51 //------------------------------LoopNode--------------------------------------- |
|
52 // Simple loop header. Fall in path on left, loop-back path on right. |
|
53 class LoopNode : public RegionNode { |
|
54 // Size is bigger to hold the flags. However, the flags do not change |
|
55 // the semantics so it does not appear in the hash & cmp functions. |
|
56 virtual uint size_of() const { return sizeof(*this); } |
|
57 protected: |
|
58 short _loop_flags; |
|
59 // Names for flag bitfields |
|
60 enum { Normal=0, Pre=1, Main=2, Post=3, PreMainPostFlagsMask=3, |
|
61 MainHasNoPreLoop=4, |
|
62 HasExactTripCount=8, |
|
63 InnerLoop=16, |
|
64 PartialPeelLoop=32, |
|
65 PartialPeelFailed=64 }; |
|
66 char _unswitch_count; |
|
67 enum { _unswitch_max=3 }; |
|
68 |
|
69 public: |
|
70 // Names for edge indices |
|
71 enum { Self=0, EntryControl, LoopBackControl }; |
|
72 |
|
73 int is_inner_loop() const { return _loop_flags & InnerLoop; } |
|
74 void set_inner_loop() { _loop_flags |= InnerLoop; } |
|
75 |
|
76 int is_partial_peel_loop() const { return _loop_flags & PartialPeelLoop; } |
|
77 void set_partial_peel_loop() { _loop_flags |= PartialPeelLoop; } |
|
78 int partial_peel_has_failed() const { return _loop_flags & PartialPeelFailed; } |
|
79 void mark_partial_peel_failed() { _loop_flags |= PartialPeelFailed; } |
|
80 |
|
81 int unswitch_max() { return _unswitch_max; } |
|
82 int unswitch_count() { return _unswitch_count; } |
|
83 void set_unswitch_count(int val) { |
|
84 assert (val <= unswitch_max(), "too many unswitches"); |
|
85 _unswitch_count = val; |
|
86 } |
|
87 |
|
88 LoopNode( Node *entry, Node *backedge ) : RegionNode(3), _loop_flags(0), _unswitch_count(0) { |
|
89 init_class_id(Class_Loop); |
|
90 init_req(EntryControl, entry); |
|
91 init_req(LoopBackControl, backedge); |
|
92 } |
|
93 |
|
94 virtual Node *Ideal(PhaseGVN *phase, bool can_reshape); |
|
95 virtual int Opcode() const; |
|
96 bool can_be_counted_loop(PhaseTransform* phase) const { |
|
97 return req() == 3 && in(0) != NULL && |
|
98 in(1) != NULL && phase->type(in(1)) != Type::TOP && |
|
99 in(2) != NULL && phase->type(in(2)) != Type::TOP; |
|
100 } |
|
101 bool is_valid_counted_loop() const; |
|
102 #ifndef PRODUCT |
|
103 virtual void dump_spec(outputStream *st) const; |
|
104 #endif |
|
105 }; |
|
106 |
|
107 //------------------------------Counted Loops---------------------------------- |
|
108 // Counted loops are all trip-counted loops, with exactly 1 trip-counter exit |
|
109 // path (and maybe some other exit paths). The trip-counter exit is always |
|
110 // last in the loop. The trip-counter have to stride by a constant; |
|
111 // the exit value is also loop invariant. |
|
112 |
|
113 // CountedLoopNodes and CountedLoopEndNodes come in matched pairs. The |
|
114 // CountedLoopNode has the incoming loop control and the loop-back-control |
|
115 // which is always the IfTrue before the matching CountedLoopEndNode. The |
|
116 // CountedLoopEndNode has an incoming control (possibly not the |
|
117 // CountedLoopNode if there is control flow in the loop), the post-increment |
|
118 // trip-counter value, and the limit. The trip-counter value is always of |
|
119 // the form (Op old-trip-counter stride). The old-trip-counter is produced |
|
120 // by a Phi connected to the CountedLoopNode. The stride is constant. |
|
121 // The Op is any commutable opcode, including Add, Mul, Xor. The |
|
122 // CountedLoopEndNode also takes in the loop-invariant limit value. |
|
123 |
|
124 // From a CountedLoopNode I can reach the matching CountedLoopEndNode via the |
|
125 // loop-back control. From CountedLoopEndNodes I can reach CountedLoopNodes |
|
126 // via the old-trip-counter from the Op node. |
|
127 |
|
128 //------------------------------CountedLoopNode-------------------------------- |
|
129 // CountedLoopNodes head simple counted loops. CountedLoopNodes have as |
|
130 // inputs the incoming loop-start control and the loop-back control, so they |
|
131 // act like RegionNodes. They also take in the initial trip counter, the |
|
132 // loop-invariant stride and the loop-invariant limit value. CountedLoopNodes |
|
133 // produce a loop-body control and the trip counter value. Since |
|
134 // CountedLoopNodes behave like RegionNodes I still have a standard CFG model. |
|
135 |
|
136 class CountedLoopNode : public LoopNode { |
|
137 // Size is bigger to hold _main_idx. However, _main_idx does not change |
|
138 // the semantics so it does not appear in the hash & cmp functions. |
|
139 virtual uint size_of() const { return sizeof(*this); } |
|
140 |
|
141 // For Pre- and Post-loops during debugging ONLY, this holds the index of |
|
142 // the Main CountedLoop. Used to assert that we understand the graph shape. |
|
143 node_idx_t _main_idx; |
|
144 |
|
145 // Known trip count calculated by compute_exact_trip_count() |
|
146 uint _trip_count; |
|
147 |
|
148 // Expected trip count from profile data |
|
149 float _profile_trip_cnt; |
|
150 |
|
151 // Log2 of original loop bodies in unrolled loop |
|
152 int _unrolled_count_log2; |
|
153 |
|
154 // Node count prior to last unrolling - used to decide if |
|
155 // unroll,optimize,unroll,optimize,... is making progress |
|
156 int _node_count_before_unroll; |
|
157 |
|
158 public: |
|
159 CountedLoopNode( Node *entry, Node *backedge ) |
|
160 : LoopNode(entry, backedge), _main_idx(0), _trip_count(max_juint), |
|
161 _profile_trip_cnt(COUNT_UNKNOWN), _unrolled_count_log2(0), |
|
162 _node_count_before_unroll(0) { |
|
163 init_class_id(Class_CountedLoop); |
|
164 // Initialize _trip_count to the largest possible value. |
|
165 // Will be reset (lower) if the loop's trip count is known. |
|
166 } |
|
167 |
|
168 virtual int Opcode() const; |
|
169 virtual Node *Ideal(PhaseGVN *phase, bool can_reshape); |
|
170 |
|
171 Node *init_control() const { return in(EntryControl); } |
|
172 Node *back_control() const { return in(LoopBackControl); } |
|
173 CountedLoopEndNode *loopexit() const; |
|
174 Node *init_trip() const; |
|
175 Node *stride() const; |
|
176 int stride_con() const; |
|
177 bool stride_is_con() const; |
|
178 Node *limit() const; |
|
179 Node *incr() const; |
|
180 Node *phi() const; |
|
181 |
|
182 // Match increment with optional truncation |
|
183 static Node* match_incr_with_optional_truncation(Node* expr, Node** trunc1, Node** trunc2, const TypeInt** trunc_type); |
|
184 |
|
185 // A 'main' loop has a pre-loop and a post-loop. The 'main' loop |
|
186 // can run short a few iterations and may start a few iterations in. |
|
187 // It will be RCE'd and unrolled and aligned. |
|
188 |
|
189 // A following 'post' loop will run any remaining iterations. Used |
|
190 // during Range Check Elimination, the 'post' loop will do any final |
|
191 // iterations with full checks. Also used by Loop Unrolling, where |
|
192 // the 'post' loop will do any epilog iterations needed. Basically, |
|
193 // a 'post' loop can not profitably be further unrolled or RCE'd. |
|
194 |
|
195 // A preceding 'pre' loop will run at least 1 iteration (to do peeling), |
|
196 // it may do under-flow checks for RCE and may do alignment iterations |
|
197 // so the following main loop 'knows' that it is striding down cache |
|
198 // lines. |
|
199 |
|
200 // A 'main' loop that is ONLY unrolled or peeled, never RCE'd or |
|
201 // Aligned, may be missing it's pre-loop. |
|
202 int is_normal_loop() const { return (_loop_flags&PreMainPostFlagsMask) == Normal; } |
|
203 int is_pre_loop () const { return (_loop_flags&PreMainPostFlagsMask) == Pre; } |
|
204 int is_main_loop () const { return (_loop_flags&PreMainPostFlagsMask) == Main; } |
|
205 int is_post_loop () const { return (_loop_flags&PreMainPostFlagsMask) == Post; } |
|
206 int is_main_no_pre_loop() const { return _loop_flags & MainHasNoPreLoop; } |
|
207 void set_main_no_pre_loop() { _loop_flags |= MainHasNoPreLoop; } |
|
208 |
|
209 int main_idx() const { return _main_idx; } |
|
210 |
|
211 |
|
212 void set_pre_loop (CountedLoopNode *main) { assert(is_normal_loop(),""); _loop_flags |= Pre ; _main_idx = main->_idx; } |
|
213 void set_main_loop ( ) { assert(is_normal_loop(),""); _loop_flags |= Main; } |
|
214 void set_post_loop (CountedLoopNode *main) { assert(is_normal_loop(),""); _loop_flags |= Post; _main_idx = main->_idx; } |
|
215 void set_normal_loop( ) { _loop_flags &= ~PreMainPostFlagsMask; } |
|
216 |
|
217 void set_trip_count(uint tc) { _trip_count = tc; } |
|
218 uint trip_count() { return _trip_count; } |
|
219 |
|
220 bool has_exact_trip_count() const { return (_loop_flags & HasExactTripCount) != 0; } |
|
221 void set_exact_trip_count(uint tc) { |
|
222 _trip_count = tc; |
|
223 _loop_flags |= HasExactTripCount; |
|
224 } |
|
225 void set_nonexact_trip_count() { |
|
226 _loop_flags &= ~HasExactTripCount; |
|
227 } |
|
228 |
|
229 void set_profile_trip_cnt(float ptc) { _profile_trip_cnt = ptc; } |
|
230 float profile_trip_cnt() { return _profile_trip_cnt; } |
|
231 |
|
232 void double_unrolled_count() { _unrolled_count_log2++; } |
|
233 int unrolled_count() { return 1 << MIN2(_unrolled_count_log2, BitsPerInt-3); } |
|
234 |
|
235 void set_node_count_before_unroll(int ct) { _node_count_before_unroll = ct; } |
|
236 int node_count_before_unroll() { return _node_count_before_unroll; } |
|
237 |
|
238 #ifndef PRODUCT |
|
239 virtual void dump_spec(outputStream *st) const; |
|
240 #endif |
|
241 }; |
|
242 |
|
243 //------------------------------CountedLoopEndNode----------------------------- |
|
244 // CountedLoopEndNodes end simple trip counted loops. They act much like |
|
245 // IfNodes. |
|
246 class CountedLoopEndNode : public IfNode { |
|
247 public: |
|
248 enum { TestControl, TestValue }; |
|
249 |
|
250 CountedLoopEndNode( Node *control, Node *test, float prob, float cnt ) |
|
251 : IfNode( control, test, prob, cnt) { |
|
252 init_class_id(Class_CountedLoopEnd); |
|
253 } |
|
254 virtual int Opcode() const; |
|
255 |
|
256 Node *cmp_node() const { return (in(TestValue)->req() >=2) ? in(TestValue)->in(1) : NULL; } |
|
257 Node *incr() const { Node *tmp = cmp_node(); return (tmp && tmp->req()==3) ? tmp->in(1) : NULL; } |
|
258 Node *limit() const { Node *tmp = cmp_node(); return (tmp && tmp->req()==3) ? tmp->in(2) : NULL; } |
|
259 Node *stride() const { Node *tmp = incr (); return (tmp && tmp->req()==3) ? tmp->in(2) : NULL; } |
|
260 Node *phi() const { Node *tmp = incr (); return (tmp && tmp->req()==3) ? tmp->in(1) : NULL; } |
|
261 Node *init_trip() const { Node *tmp = phi (); return (tmp && tmp->req()==3) ? tmp->in(1) : NULL; } |
|
262 int stride_con() const; |
|
263 bool stride_is_con() const { Node *tmp = stride (); return (tmp != NULL && tmp->is_Con()); } |
|
264 BoolTest::mask test_trip() const { return in(TestValue)->as_Bool()->_test._test; } |
|
265 CountedLoopNode *loopnode() const { |
|
266 // The CountedLoopNode that goes with this CountedLoopEndNode may |
|
267 // have been optimized out by the IGVN so be cautious with the |
|
268 // pattern matching on the graph |
|
269 if (phi() == NULL) { |
|
270 return NULL; |
|
271 } |
|
272 Node *ln = phi()->in(0); |
|
273 if (ln->is_CountedLoop() && ln->as_CountedLoop()->loopexit() == this) { |
|
274 return (CountedLoopNode*)ln; |
|
275 } |
|
276 return NULL; |
|
277 } |
|
278 |
|
279 #ifndef PRODUCT |
|
280 virtual void dump_spec(outputStream *st) const; |
|
281 #endif |
|
282 }; |
|
283 |
|
284 |
|
285 inline CountedLoopEndNode *CountedLoopNode::loopexit() const { |
|
286 Node *bc = back_control(); |
|
287 if( bc == NULL ) return NULL; |
|
288 Node *le = bc->in(0); |
|
289 if( le->Opcode() != Op_CountedLoopEnd ) |
|
290 return NULL; |
|
291 return (CountedLoopEndNode*)le; |
|
292 } |
|
293 inline Node *CountedLoopNode::init_trip() const { return loopexit() ? loopexit()->init_trip() : NULL; } |
|
294 inline Node *CountedLoopNode::stride() const { return loopexit() ? loopexit()->stride() : NULL; } |
|
295 inline int CountedLoopNode::stride_con() const { return loopexit() ? loopexit()->stride_con() : 0; } |
|
296 inline bool CountedLoopNode::stride_is_con() const { return loopexit() && loopexit()->stride_is_con(); } |
|
297 inline Node *CountedLoopNode::limit() const { return loopexit() ? loopexit()->limit() : NULL; } |
|
298 inline Node *CountedLoopNode::incr() const { return loopexit() ? loopexit()->incr() : NULL; } |
|
299 inline Node *CountedLoopNode::phi() const { return loopexit() ? loopexit()->phi() : NULL; } |
|
300 |
|
301 //------------------------------LoopLimitNode----------------------------- |
|
302 // Counted Loop limit node which represents exact final iterator value: |
|
303 // trip_count = (limit - init_trip + stride - 1)/stride |
|
304 // final_value= trip_count * stride + init_trip. |
|
305 // Use HW instructions to calculate it when it can overflow in integer. |
|
306 // Note, final_value should fit into integer since counted loop has |
|
307 // limit check: limit <= max_int-stride. |
|
308 class LoopLimitNode : public Node { |
|
309 enum { Init=1, Limit=2, Stride=3 }; |
|
310 public: |
|
311 LoopLimitNode( Compile* C, Node *init, Node *limit, Node *stride ) : Node(0,init,limit,stride) { |
|
312 // Put it on the Macro nodes list to optimize during macro nodes expansion. |
|
313 init_flags(Flag_is_macro); |
|
314 C->add_macro_node(this); |
|
315 } |
|
316 virtual int Opcode() const; |
|
317 virtual const Type *bottom_type() const { return TypeInt::INT; } |
|
318 virtual uint ideal_reg() const { return Op_RegI; } |
|
319 virtual const Type *Value( PhaseTransform *phase ) const; |
|
320 virtual Node *Ideal(PhaseGVN *phase, bool can_reshape); |
|
321 virtual Node *Identity( PhaseTransform *phase ); |
|
322 }; |
|
323 |
|
324 // -----------------------------IdealLoopTree---------------------------------- |
|
325 class IdealLoopTree : public ResourceObj { |
|
326 public: |
|
327 IdealLoopTree *_parent; // Parent in loop tree |
|
328 IdealLoopTree *_next; // Next sibling in loop tree |
|
329 IdealLoopTree *_child; // First child in loop tree |
|
330 |
|
331 // The head-tail backedge defines the loop. |
|
332 // If tail is NULL then this loop has multiple backedges as part of the |
|
333 // same loop. During cleanup I'll peel off the multiple backedges; merge |
|
334 // them at the loop bottom and flow 1 real backedge into the loop. |
|
335 Node *_head; // Head of loop |
|
336 Node *_tail; // Tail of loop |
|
337 inline Node *tail(); // Handle lazy update of _tail field |
|
338 PhaseIdealLoop* _phase; |
|
339 |
|
340 Node_List _body; // Loop body for inner loops |
|
341 |
|
342 uint8 _nest; // Nesting depth |
|
343 uint8 _irreducible:1, // True if irreducible |
|
344 _has_call:1, // True if has call safepoint |
|
345 _has_sfpt:1, // True if has non-call safepoint |
|
346 _rce_candidate:1; // True if candidate for range check elimination |
|
347 |
|
348 Node_List* _safepts; // List of safepoints in this loop |
|
349 Node_List* _required_safept; // A inner loop cannot delete these safepts; |
|
350 bool _allow_optimizations; // Allow loop optimizations |
|
351 |
|
352 IdealLoopTree( PhaseIdealLoop* phase, Node *head, Node *tail ) |
|
353 : _parent(0), _next(0), _child(0), |
|
354 _head(head), _tail(tail), |
|
355 _phase(phase), |
|
356 _safepts(NULL), |
|
357 _required_safept(NULL), |
|
358 _allow_optimizations(true), |
|
359 _nest(0), _irreducible(0), _has_call(0), _has_sfpt(0), _rce_candidate(0) |
|
360 { } |
|
361 |
|
362 // Is 'l' a member of 'this'? |
|
363 int is_member( const IdealLoopTree *l ) const; // Test for nested membership |
|
364 |
|
365 // Set loop nesting depth. Accumulate has_call bits. |
|
366 int set_nest( uint depth ); |
|
367 |
|
368 // Split out multiple fall-in edges from the loop header. Move them to a |
|
369 // private RegionNode before the loop. This becomes the loop landing pad. |
|
370 void split_fall_in( PhaseIdealLoop *phase, int fall_in_cnt ); |
|
371 |
|
372 // Split out the outermost loop from this shared header. |
|
373 void split_outer_loop( PhaseIdealLoop *phase ); |
|
374 |
|
375 // Merge all the backedges from the shared header into a private Region. |
|
376 // Feed that region as the one backedge to this loop. |
|
377 void merge_many_backedges( PhaseIdealLoop *phase ); |
|
378 |
|
379 // Split shared headers and insert loop landing pads. |
|
380 // Insert a LoopNode to replace the RegionNode. |
|
381 // Returns TRUE if loop tree is structurally changed. |
|
382 bool beautify_loops( PhaseIdealLoop *phase ); |
|
383 |
|
384 // Perform optimization to use the loop predicates for null checks and range checks. |
|
385 // Applies to any loop level (not just the innermost one) |
|
386 bool loop_predication( PhaseIdealLoop *phase); |
|
387 |
|
388 // Perform iteration-splitting on inner loops. Split iterations to |
|
389 // avoid range checks or one-shot null checks. Returns false if the |
|
390 // current round of loop opts should stop. |
|
391 bool iteration_split( PhaseIdealLoop *phase, Node_List &old_new ); |
|
392 |
|
393 // Driver for various flavors of iteration splitting. Returns false |
|
394 // if the current round of loop opts should stop. |
|
395 bool iteration_split_impl( PhaseIdealLoop *phase, Node_List &old_new ); |
|
396 |
|
397 // Given dominators, try to find loops with calls that must always be |
|
398 // executed (call dominates loop tail). These loops do not need non-call |
|
399 // safepoints (ncsfpt). |
|
400 void check_safepts(VectorSet &visited, Node_List &stack); |
|
401 |
|
402 // Allpaths backwards scan from loop tail, terminating each path at first safepoint |
|
403 // encountered. |
|
404 void allpaths_check_safepts(VectorSet &visited, Node_List &stack); |
|
405 |
|
406 // Convert to counted loops where possible |
|
407 void counted_loop( PhaseIdealLoop *phase ); |
|
408 |
|
409 // Check for Node being a loop-breaking test |
|
410 Node *is_loop_exit(Node *iff) const; |
|
411 |
|
412 // Returns true if ctrl is executed on every complete iteration |
|
413 bool dominates_backedge(Node* ctrl); |
|
414 |
|
415 // Remove simplistic dead code from loop body |
|
416 void DCE_loop_body(); |
|
417 |
|
418 // Look for loop-exit tests with my 50/50 guesses from the Parsing stage. |
|
419 // Replace with a 1-in-10 exit guess. |
|
420 void adjust_loop_exit_prob( PhaseIdealLoop *phase ); |
|
421 |
|
422 // Return TRUE or FALSE if the loop should never be RCE'd or aligned. |
|
423 // Useful for unrolling loops with NO array accesses. |
|
424 bool policy_peel_only( PhaseIdealLoop *phase ) const; |
|
425 |
|
426 // Return TRUE or FALSE if the loop should be unswitched -- clone |
|
427 // loop with an invariant test |
|
428 bool policy_unswitching( PhaseIdealLoop *phase ) const; |
|
429 |
|
430 // Micro-benchmark spamming. Remove empty loops. |
|
431 bool policy_do_remove_empty_loop( PhaseIdealLoop *phase ); |
|
432 |
|
433 // Convert one iteration loop into normal code. |
|
434 bool policy_do_one_iteration_loop( PhaseIdealLoop *phase ); |
|
435 |
|
436 // Return TRUE or FALSE if the loop should be peeled or not. Peel if we can |
|
437 // make some loop-invariant test (usually a null-check) happen before the |
|
438 // loop. |
|
439 bool policy_peeling( PhaseIdealLoop *phase ) const; |
|
440 |
|
441 // Return TRUE or FALSE if the loop should be maximally unrolled. Stash any |
|
442 // known trip count in the counted loop node. |
|
443 bool policy_maximally_unroll( PhaseIdealLoop *phase ) const; |
|
444 |
|
445 // Return TRUE or FALSE if the loop should be unrolled or not. Unroll if |
|
446 // the loop is a CountedLoop and the body is small enough. |
|
447 bool policy_unroll( PhaseIdealLoop *phase ) const; |
|
448 |
|
449 // Return TRUE or FALSE if the loop should be range-check-eliminated. |
|
450 // Gather a list of IF tests that are dominated by iteration splitting; |
|
451 // also gather the end of the first split and the start of the 2nd split. |
|
452 bool policy_range_check( PhaseIdealLoop *phase ) const; |
|
453 |
|
454 // Return TRUE or FALSE if the loop should be cache-line aligned. |
|
455 // Gather the expression that does the alignment. Note that only |
|
456 // one array base can be aligned in a loop (unless the VM guarantees |
|
457 // mutual alignment). Note that if we vectorize short memory ops |
|
458 // into longer memory ops, we may want to increase alignment. |
|
459 bool policy_align( PhaseIdealLoop *phase ) const; |
|
460 |
|
461 // Return TRUE if "iff" is a range check. |
|
462 bool is_range_check_if(IfNode *iff, PhaseIdealLoop *phase, Invariance& invar) const; |
|
463 |
|
464 // Compute loop exact trip count if possible |
|
465 void compute_exact_trip_count( PhaseIdealLoop *phase ); |
|
466 |
|
467 // Compute loop trip count from profile data |
|
468 void compute_profile_trip_cnt( PhaseIdealLoop *phase ); |
|
469 |
|
470 // Reassociate invariant expressions. |
|
471 void reassociate_invariants(PhaseIdealLoop *phase); |
|
472 // Reassociate invariant add and subtract expressions. |
|
473 Node* reassociate_add_sub(Node* n1, PhaseIdealLoop *phase); |
|
474 // Return nonzero index of invariant operand if invariant and variant |
|
475 // are combined with an Add or Sub. Helper for reassociate_invariants. |
|
476 int is_invariant_addition(Node* n, PhaseIdealLoop *phase); |
|
477 |
|
478 // Return true if n is invariant |
|
479 bool is_invariant(Node* n) const; |
|
480 |
|
481 // Put loop body on igvn work list |
|
482 void record_for_igvn(); |
|
483 |
|
484 bool is_loop() { return !_irreducible && _tail && !_tail->is_top(); } |
|
485 bool is_inner() { return is_loop() && _child == NULL; } |
|
486 bool is_counted() { return is_loop() && _head != NULL && _head->is_CountedLoop(); } |
|
487 |
|
488 #ifndef PRODUCT |
|
489 void dump_head( ) const; // Dump loop head only |
|
490 void dump() const; // Dump this loop recursively |
|
491 void verify_tree(IdealLoopTree *loop, const IdealLoopTree *parent) const; |
|
492 #endif |
|
493 |
|
494 }; |
|
495 |
|
496 // -----------------------------PhaseIdealLoop--------------------------------- |
|
497 // Computes the mapping from Nodes to IdealLoopTrees. Organizes IdealLoopTrees into a |
|
498 // loop tree. Drives the loop-based transformations on the ideal graph. |
|
499 class PhaseIdealLoop : public PhaseTransform { |
|
500 friend class IdealLoopTree; |
|
501 friend class SuperWord; |
|
502 // Pre-computed def-use info |
|
503 PhaseIterGVN &_igvn; |
|
504 |
|
505 // Head of loop tree |
|
506 IdealLoopTree *_ltree_root; |
|
507 |
|
508 // Array of pre-order numbers, plus post-visited bit. |
|
509 // ZERO for not pre-visited. EVEN for pre-visited but not post-visited. |
|
510 // ODD for post-visited. Other bits are the pre-order number. |
|
511 uint *_preorders; |
|
512 uint _max_preorder; |
|
513 |
|
514 const PhaseIdealLoop* _verify_me; |
|
515 bool _verify_only; |
|
516 |
|
517 // Allocate _preorders[] array |
|
518 void allocate_preorders() { |
|
519 _max_preorder = C->unique()+8; |
|
520 _preorders = NEW_RESOURCE_ARRAY(uint, _max_preorder); |
|
521 memset(_preorders, 0, sizeof(uint) * _max_preorder); |
|
522 } |
|
523 |
|
524 // Allocate _preorders[] array |
|
525 void reallocate_preorders() { |
|
526 if ( _max_preorder < C->unique() ) { |
|
527 _preorders = REALLOC_RESOURCE_ARRAY(uint, _preorders, _max_preorder, C->unique()); |
|
528 _max_preorder = C->unique(); |
|
529 } |
|
530 memset(_preorders, 0, sizeof(uint) * _max_preorder); |
|
531 } |
|
532 |
|
533 // Check to grow _preorders[] array for the case when build_loop_tree_impl() |
|
534 // adds new nodes. |
|
535 void check_grow_preorders( ) { |
|
536 if ( _max_preorder < C->unique() ) { |
|
537 uint newsize = _max_preorder<<1; // double size of array |
|
538 _preorders = REALLOC_RESOURCE_ARRAY(uint, _preorders, _max_preorder, newsize); |
|
539 memset(&_preorders[_max_preorder],0,sizeof(uint)*(newsize-_max_preorder)); |
|
540 _max_preorder = newsize; |
|
541 } |
|
542 } |
|
543 // Check for pre-visited. Zero for NOT visited; non-zero for visited. |
|
544 int is_visited( Node *n ) const { return _preorders[n->_idx]; } |
|
545 // Pre-order numbers are written to the Nodes array as low-bit-set values. |
|
546 void set_preorder_visited( Node *n, int pre_order ) { |
|
547 assert( !is_visited( n ), "already set" ); |
|
548 _preorders[n->_idx] = (pre_order<<1); |
|
549 }; |
|
550 // Return pre-order number. |
|
551 int get_preorder( Node *n ) const { assert( is_visited(n), "" ); return _preorders[n->_idx]>>1; } |
|
552 |
|
553 // Check for being post-visited. |
|
554 // Should be previsited already (checked with assert(is_visited(n))). |
|
555 int is_postvisited( Node *n ) const { assert( is_visited(n), "" ); return _preorders[n->_idx]&1; } |
|
556 |
|
557 // Mark as post visited |
|
558 void set_postvisited( Node *n ) { assert( !is_postvisited( n ), "" ); _preorders[n->_idx] |= 1; } |
|
559 |
|
560 // Set/get control node out. Set lower bit to distinguish from IdealLoopTree |
|
561 // Returns true if "n" is a data node, false if it's a control node. |
|
562 bool has_ctrl( Node *n ) const { return ((intptr_t)_nodes[n->_idx]) & 1; } |
|
563 |
|
564 // clear out dead code after build_loop_late |
|
565 Node_List _deadlist; |
|
566 |
|
567 // Support for faster execution of get_late_ctrl()/dom_lca() |
|
568 // when a node has many uses and dominator depth is deep. |
|
569 Node_Array _dom_lca_tags; |
|
570 void init_dom_lca_tags(); |
|
571 void clear_dom_lca_tags(); |
|
572 |
|
573 // Helper for debugging bad dominance relationships |
|
574 bool verify_dominance(Node* n, Node* use, Node* LCA, Node* early); |
|
575 |
|
576 Node* compute_lca_of_uses(Node* n, Node* early, bool verify = false); |
|
577 |
|
578 // Inline wrapper for frequent cases: |
|
579 // 1) only one use |
|
580 // 2) a use is the same as the current LCA passed as 'n1' |
|
581 Node *dom_lca_for_get_late_ctrl( Node *lca, Node *n, Node *tag ) { |
|
582 assert( n->is_CFG(), "" ); |
|
583 // Fast-path NULL lca |
|
584 if( lca != NULL && lca != n ) { |
|
585 assert( lca->is_CFG(), "" ); |
|
586 // find LCA of all uses |
|
587 n = dom_lca_for_get_late_ctrl_internal( lca, n, tag ); |
|
588 } |
|
589 return find_non_split_ctrl(n); |
|
590 } |
|
591 Node *dom_lca_for_get_late_ctrl_internal( Node *lca, Node *n, Node *tag ); |
|
592 |
|
593 // Helper function for directing control inputs away from CFG split |
|
594 // points. |
|
595 Node *find_non_split_ctrl( Node *ctrl ) const { |
|
596 if (ctrl != NULL) { |
|
597 if (ctrl->is_MultiBranch()) { |
|
598 ctrl = ctrl->in(0); |
|
599 } |
|
600 assert(ctrl->is_CFG(), "CFG"); |
|
601 } |
|
602 return ctrl; |
|
603 } |
|
604 |
|
605 public: |
|
606 bool has_node( Node* n ) const { |
|
607 guarantee(n != NULL, "No Node."); |
|
608 return _nodes[n->_idx] != NULL; |
|
609 } |
|
610 // check if transform created new nodes that need _ctrl recorded |
|
611 Node *get_late_ctrl( Node *n, Node *early ); |
|
612 Node *get_early_ctrl( Node *n ); |
|
613 Node *get_early_ctrl_for_expensive(Node *n, Node* earliest); |
|
614 void set_early_ctrl( Node *n ); |
|
615 void set_subtree_ctrl( Node *root ); |
|
616 void set_ctrl( Node *n, Node *ctrl ) { |
|
617 assert( !has_node(n) || has_ctrl(n), "" ); |
|
618 assert( ctrl->in(0), "cannot set dead control node" ); |
|
619 assert( ctrl == find_non_split_ctrl(ctrl), "must set legal crtl" ); |
|
620 _nodes.map( n->_idx, (Node*)((intptr_t)ctrl + 1) ); |
|
621 } |
|
622 // Set control and update loop membership |
|
623 void set_ctrl_and_loop(Node* n, Node* ctrl) { |
|
624 IdealLoopTree* old_loop = get_loop(get_ctrl(n)); |
|
625 IdealLoopTree* new_loop = get_loop(ctrl); |
|
626 if (old_loop != new_loop) { |
|
627 if (old_loop->_child == NULL) old_loop->_body.yank(n); |
|
628 if (new_loop->_child == NULL) new_loop->_body.push(n); |
|
629 } |
|
630 set_ctrl(n, ctrl); |
|
631 } |
|
632 // Control nodes can be replaced or subsumed. During this pass they |
|
633 // get their replacement Node in slot 1. Instead of updating the block |
|
634 // location of all Nodes in the subsumed block, we lazily do it. As we |
|
635 // pull such a subsumed block out of the array, we write back the final |
|
636 // correct block. |
|
637 Node *get_ctrl( Node *i ) { |
|
638 assert(has_node(i), ""); |
|
639 Node *n = get_ctrl_no_update(i); |
|
640 _nodes.map( i->_idx, (Node*)((intptr_t)n + 1) ); |
|
641 assert(has_node(i) && has_ctrl(i), ""); |
|
642 assert(n == find_non_split_ctrl(n), "must return legal ctrl" ); |
|
643 return n; |
|
644 } |
|
645 // true if CFG node d dominates CFG node n |
|
646 bool is_dominator(Node *d, Node *n); |
|
647 // return get_ctrl for a data node and self(n) for a CFG node |
|
648 Node* ctrl_or_self(Node* n) { |
|
649 if (has_ctrl(n)) |
|
650 return get_ctrl(n); |
|
651 else { |
|
652 assert (n->is_CFG(), "must be a CFG node"); |
|
653 return n; |
|
654 } |
|
655 } |
|
656 |
|
657 private: |
|
658 Node *get_ctrl_no_update( Node *i ) const { |
|
659 assert( has_ctrl(i), "" ); |
|
660 Node *n = (Node*)(((intptr_t)_nodes[i->_idx]) & ~1); |
|
661 if (!n->in(0)) { |
|
662 // Skip dead CFG nodes |
|
663 do { |
|
664 n = (Node*)(((intptr_t)_nodes[n->_idx]) & ~1); |
|
665 } while (!n->in(0)); |
|
666 n = find_non_split_ctrl(n); |
|
667 } |
|
668 return n; |
|
669 } |
|
670 |
|
671 // Check for loop being set |
|
672 // "n" must be a control node. Returns true if "n" is known to be in a loop. |
|
673 bool has_loop( Node *n ) const { |
|
674 assert(!has_node(n) || !has_ctrl(n), ""); |
|
675 return has_node(n); |
|
676 } |
|
677 // Set loop |
|
678 void set_loop( Node *n, IdealLoopTree *loop ) { |
|
679 _nodes.map(n->_idx, (Node*)loop); |
|
680 } |
|
681 // Lazy-dazy update of 'get_ctrl' and 'idom_at' mechanisms. Replace |
|
682 // the 'old_node' with 'new_node'. Kill old-node. Add a reference |
|
683 // from old_node to new_node to support the lazy update. Reference |
|
684 // replaces loop reference, since that is not needed for dead node. |
|
685 public: |
|
686 void lazy_update( Node *old_node, Node *new_node ) { |
|
687 assert( old_node != new_node, "no cycles please" ); |
|
688 //old_node->set_req( 1, new_node /*NO DU INFO*/ ); |
|
689 // Nodes always have DU info now, so re-use the side array slot |
|
690 // for this node to provide the forwarding pointer. |
|
691 _nodes.map( old_node->_idx, (Node*)((intptr_t)new_node + 1) ); |
|
692 } |
|
693 void lazy_replace( Node *old_node, Node *new_node ) { |
|
694 _igvn.replace_node( old_node, new_node ); |
|
695 lazy_update( old_node, new_node ); |
|
696 } |
|
697 void lazy_replace_proj( Node *old_node, Node *new_node ) { |
|
698 assert( old_node->req() == 1, "use this for Projs" ); |
|
699 _igvn.hash_delete(old_node); // Must hash-delete before hacking edges |
|
700 old_node->add_req( NULL ); |
|
701 lazy_replace( old_node, new_node ); |
|
702 } |
|
703 |
|
704 private: |
|
705 |
|
706 // Place 'n' in some loop nest, where 'n' is a CFG node |
|
707 void build_loop_tree(); |
|
708 int build_loop_tree_impl( Node *n, int pre_order ); |
|
709 // Insert loop into the existing loop tree. 'innermost' is a leaf of the |
|
710 // loop tree, not the root. |
|
711 IdealLoopTree *sort( IdealLoopTree *loop, IdealLoopTree *innermost ); |
|
712 |
|
713 // Place Data nodes in some loop nest |
|
714 void build_loop_early( VectorSet &visited, Node_List &worklist, Node_Stack &nstack ); |
|
715 void build_loop_late ( VectorSet &visited, Node_List &worklist, Node_Stack &nstack ); |
|
716 void build_loop_late_post ( Node* n ); |
|
717 |
|
718 // Array of immediate dominance info for each CFG node indexed by node idx |
|
719 private: |
|
720 uint _idom_size; |
|
721 Node **_idom; // Array of immediate dominators |
|
722 uint *_dom_depth; // Used for fast LCA test |
|
723 GrowableArray<uint>* _dom_stk; // For recomputation of dom depth |
|
724 |
|
725 Node* idom_no_update(Node* d) const { |
|
726 assert(d->_idx < _idom_size, "oob"); |
|
727 Node* n = _idom[d->_idx]; |
|
728 assert(n != NULL,"Bad immediate dominator info."); |
|
729 while (n->in(0) == NULL) { // Skip dead CFG nodes |
|
730 //n = n->in(1); |
|
731 n = (Node*)(((intptr_t)_nodes[n->_idx]) & ~1); |
|
732 assert(n != NULL,"Bad immediate dominator info."); |
|
733 } |
|
734 return n; |
|
735 } |
|
736 Node *idom(Node* d) const { |
|
737 uint didx = d->_idx; |
|
738 Node *n = idom_no_update(d); |
|
739 _idom[didx] = n; // Lazily remove dead CFG nodes from table. |
|
740 return n; |
|
741 } |
|
742 uint dom_depth(Node* d) const { |
|
743 guarantee(d != NULL, "Null dominator info."); |
|
744 guarantee(d->_idx < _idom_size, ""); |
|
745 return _dom_depth[d->_idx]; |
|
746 } |
|
747 void set_idom(Node* d, Node* n, uint dom_depth); |
|
748 // Locally compute IDOM using dom_lca call |
|
749 Node *compute_idom( Node *region ) const; |
|
750 // Recompute dom_depth |
|
751 void recompute_dom_depth(); |
|
752 |
|
753 // Is safept not required by an outer loop? |
|
754 bool is_deleteable_safept(Node* sfpt); |
|
755 |
|
756 // Replace parallel induction variable (parallel to trip counter) |
|
757 void replace_parallel_iv(IdealLoopTree *loop); |
|
758 |
|
759 // Perform verification that the graph is valid. |
|
760 PhaseIdealLoop( PhaseIterGVN &igvn) : |
|
761 PhaseTransform(Ideal_Loop), |
|
762 _igvn(igvn), |
|
763 _dom_lca_tags(arena()), // Thread::resource_area |
|
764 _verify_me(NULL), |
|
765 _verify_only(true) { |
|
766 build_and_optimize(false, false); |
|
767 } |
|
768 |
|
769 // build the loop tree and perform any requested optimizations |
|
770 void build_and_optimize(bool do_split_if, bool skip_loop_opts); |
|
771 |
|
772 public: |
|
773 // Dominators for the sea of nodes |
|
774 void Dominators(); |
|
775 Node *dom_lca( Node *n1, Node *n2 ) const { |
|
776 return find_non_split_ctrl(dom_lca_internal(n1, n2)); |
|
777 } |
|
778 Node *dom_lca_internal( Node *n1, Node *n2 ) const; |
|
779 |
|
780 // Compute the Ideal Node to Loop mapping |
|
781 PhaseIdealLoop( PhaseIterGVN &igvn, bool do_split_ifs, bool skip_loop_opts = false) : |
|
782 PhaseTransform(Ideal_Loop), |
|
783 _igvn(igvn), |
|
784 _dom_lca_tags(arena()), // Thread::resource_area |
|
785 _verify_me(NULL), |
|
786 _verify_only(false) { |
|
787 build_and_optimize(do_split_ifs, skip_loop_opts); |
|
788 } |
|
789 |
|
790 // Verify that verify_me made the same decisions as a fresh run. |
|
791 PhaseIdealLoop( PhaseIterGVN &igvn, const PhaseIdealLoop *verify_me) : |
|
792 PhaseTransform(Ideal_Loop), |
|
793 _igvn(igvn), |
|
794 _dom_lca_tags(arena()), // Thread::resource_area |
|
795 _verify_me(verify_me), |
|
796 _verify_only(false) { |
|
797 build_and_optimize(false, false); |
|
798 } |
|
799 |
|
800 // Build and verify the loop tree without modifying the graph. This |
|
801 // is useful to verify that all inputs properly dominate their uses. |
|
802 static void verify(PhaseIterGVN& igvn) { |
|
803 #ifdef ASSERT |
|
804 PhaseIdealLoop v(igvn); |
|
805 #endif |
|
806 } |
|
807 |
|
808 // True if the method has at least 1 irreducible loop |
|
809 bool _has_irreducible_loops; |
|
810 |
|
811 // Per-Node transform |
|
812 virtual Node *transform( Node *a_node ) { return 0; } |
|
813 |
|
814 bool is_counted_loop( Node *x, IdealLoopTree *loop ); |
|
815 |
|
816 Node* exact_limit( IdealLoopTree *loop ); |
|
817 |
|
818 // Return a post-walked LoopNode |
|
819 IdealLoopTree *get_loop( Node *n ) const { |
|
820 // Dead nodes have no loop, so return the top level loop instead |
|
821 if (!has_node(n)) return _ltree_root; |
|
822 assert(!has_ctrl(n), ""); |
|
823 return (IdealLoopTree*)_nodes[n->_idx]; |
|
824 } |
|
825 |
|
826 // Is 'n' a (nested) member of 'loop'? |
|
827 int is_member( const IdealLoopTree *loop, Node *n ) const { |
|
828 return loop->is_member(get_loop(n)); } |
|
829 |
|
830 // This is the basic building block of the loop optimizations. It clones an |
|
831 // entire loop body. It makes an old_new loop body mapping; with this |
|
832 // mapping you can find the new-loop equivalent to an old-loop node. All |
|
833 // new-loop nodes are exactly equal to their old-loop counterparts, all |
|
834 // edges are the same. All exits from the old-loop now have a RegionNode |
|
835 // that merges the equivalent new-loop path. This is true even for the |
|
836 // normal "loop-exit" condition. All uses of loop-invariant old-loop values |
|
837 // now come from (one or more) Phis that merge their new-loop equivalents. |
|
838 // Parameter side_by_side_idom: |
|
839 // When side_by_size_idom is NULL, the dominator tree is constructed for |
|
840 // the clone loop to dominate the original. Used in construction of |
|
841 // pre-main-post loop sequence. |
|
842 // When nonnull, the clone and original are side-by-side, both are |
|
843 // dominated by the passed in side_by_side_idom node. Used in |
|
844 // construction of unswitched loops. |
|
845 void clone_loop( IdealLoopTree *loop, Node_List &old_new, int dom_depth, |
|
846 Node* side_by_side_idom = NULL); |
|
847 |
|
848 // If we got the effect of peeling, either by actually peeling or by |
|
849 // making a pre-loop which must execute at least once, we can remove |
|
850 // all loop-invariant dominated tests in the main body. |
|
851 void peeled_dom_test_elim( IdealLoopTree *loop, Node_List &old_new ); |
|
852 |
|
853 // Generate code to do a loop peel for the given loop (and body). |
|
854 // old_new is a temp array. |
|
855 void do_peeling( IdealLoopTree *loop, Node_List &old_new ); |
|
856 |
|
857 // Add pre and post loops around the given loop. These loops are used |
|
858 // during RCE, unrolling and aligning loops. |
|
859 void insert_pre_post_loops( IdealLoopTree *loop, Node_List &old_new, bool peel_only ); |
|
860 // If Node n lives in the back_ctrl block, we clone a private version of n |
|
861 // in preheader_ctrl block and return that, otherwise return n. |
|
862 Node *clone_up_backedge_goo( Node *back_ctrl, Node *preheader_ctrl, Node *n, VectorSet &visited, Node_Stack &clones ); |
|
863 |
|
864 // Take steps to maximally unroll the loop. Peel any odd iterations, then |
|
865 // unroll to do double iterations. The next round of major loop transforms |
|
866 // will repeat till the doubled loop body does all remaining iterations in 1 |
|
867 // pass. |
|
868 void do_maximally_unroll( IdealLoopTree *loop, Node_List &old_new ); |
|
869 |
|
870 // Unroll the loop body one step - make each trip do 2 iterations. |
|
871 void do_unroll( IdealLoopTree *loop, Node_List &old_new, bool adjust_min_trip ); |
|
872 |
|
873 // Return true if exp is a constant times an induction var |
|
874 bool is_scaled_iv(Node* exp, Node* iv, int* p_scale); |
|
875 |
|
876 // Return true if exp is a scaled induction var plus (or minus) constant |
|
877 bool is_scaled_iv_plus_offset(Node* exp, Node* iv, int* p_scale, Node** p_offset, int depth = 0); |
|
878 |
|
879 // Create a new if above the uncommon_trap_if_pattern for the predicate to be promoted |
|
880 ProjNode* create_new_if_for_predicate(ProjNode* cont_proj, Node* new_entry, |
|
881 Deoptimization::DeoptReason reason); |
|
882 void register_control(Node* n, IdealLoopTree *loop, Node* pred); |
|
883 |
|
884 // Clone loop predicates to cloned loops (peeled, unswitched) |
|
885 static ProjNode* clone_predicate(ProjNode* predicate_proj, Node* new_entry, |
|
886 Deoptimization::DeoptReason reason, |
|
887 PhaseIdealLoop* loop_phase, |
|
888 PhaseIterGVN* igvn); |
|
889 |
|
890 static Node* clone_loop_predicates(Node* old_entry, Node* new_entry, |
|
891 bool clone_limit_check, |
|
892 PhaseIdealLoop* loop_phase, |
|
893 PhaseIterGVN* igvn); |
|
894 Node* clone_loop_predicates(Node* old_entry, Node* new_entry, bool clone_limit_check); |
|
895 |
|
896 static Node* skip_loop_predicates(Node* entry); |
|
897 |
|
898 // Find a good location to insert a predicate |
|
899 static ProjNode* find_predicate_insertion_point(Node* start_c, Deoptimization::DeoptReason reason); |
|
900 // Find a predicate |
|
901 static Node* find_predicate(Node* entry); |
|
902 // Construct a range check for a predicate if |
|
903 BoolNode* rc_predicate(IdealLoopTree *loop, Node* ctrl, |
|
904 int scale, Node* offset, |
|
905 Node* init, Node* limit, Node* stride, |
|
906 Node* range, bool upper); |
|
907 |
|
908 // Implementation of the loop predication to promote checks outside the loop |
|
909 bool loop_predication_impl(IdealLoopTree *loop); |
|
910 |
|
911 // Helper function to collect predicate for eliminating the useless ones |
|
912 void collect_potentially_useful_predicates(IdealLoopTree *loop, Unique_Node_List &predicate_opaque1); |
|
913 void eliminate_useless_predicates(); |
|
914 |
|
915 // Change the control input of expensive nodes to allow commoning by |
|
916 // IGVN when it is guaranteed to not result in a more frequent |
|
917 // execution of the expensive node. Return true if progress. |
|
918 bool process_expensive_nodes(); |
|
919 |
|
920 // Check whether node has become unreachable |
|
921 bool is_node_unreachable(Node *n) const { |
|
922 return !has_node(n) || n->is_unreachable(_igvn); |
|
923 } |
|
924 |
|
925 // Eliminate range-checks and other trip-counter vs loop-invariant tests. |
|
926 void do_range_check( IdealLoopTree *loop, Node_List &old_new ); |
|
927 |
|
928 // Create a slow version of the loop by cloning the loop |
|
929 // and inserting an if to select fast-slow versions. |
|
930 ProjNode* create_slow_version_of_loop(IdealLoopTree *loop, |
|
931 Node_List &old_new); |
|
932 |
|
933 // Clone loop with an invariant test (that does not exit) and |
|
934 // insert a clone of the test that selects which version to |
|
935 // execute. |
|
936 void do_unswitching (IdealLoopTree *loop, Node_List &old_new); |
|
937 |
|
938 // Find candidate "if" for unswitching |
|
939 IfNode* find_unswitching_candidate(const IdealLoopTree *loop) const; |
|
940 |
|
941 // Range Check Elimination uses this function! |
|
942 // Constrain the main loop iterations so the affine function: |
|
943 // low_limit <= scale_con * I + offset < upper_limit |
|
944 // always holds true. That is, either increase the number of iterations in |
|
945 // the pre-loop or the post-loop until the condition holds true in the main |
|
946 // loop. Scale_con, offset and limit are all loop invariant. |
|
947 void add_constraint( int stride_con, int scale_con, Node *offset, Node *low_limit, Node *upper_limit, Node *pre_ctrl, Node **pre_limit, Node **main_limit ); |
|
948 // Helper function for add_constraint(). |
|
949 Node* adjust_limit( int stride_con, Node * scale, Node *offset, Node *rc_limit, Node *loop_limit, Node *pre_ctrl ); |
|
950 |
|
951 // Partially peel loop up through last_peel node. |
|
952 bool partial_peel( IdealLoopTree *loop, Node_List &old_new ); |
|
953 |
|
954 // Create a scheduled list of nodes control dependent on ctrl set. |
|
955 void scheduled_nodelist( IdealLoopTree *loop, VectorSet& ctrl, Node_List &sched ); |
|
956 // Has a use in the vector set |
|
957 bool has_use_in_set( Node* n, VectorSet& vset ); |
|
958 // Has use internal to the vector set (ie. not in a phi at the loop head) |
|
959 bool has_use_internal_to_set( Node* n, VectorSet& vset, IdealLoopTree *loop ); |
|
960 // clone "n" for uses that are outside of loop |
|
961 int clone_for_use_outside_loop( IdealLoopTree *loop, Node* n, Node_List& worklist ); |
|
962 // clone "n" for special uses that are in the not_peeled region |
|
963 void clone_for_special_use_inside_loop( IdealLoopTree *loop, Node* n, |
|
964 VectorSet& not_peel, Node_List& sink_list, Node_List& worklist ); |
|
965 // Insert phi(lp_entry_val, back_edge_val) at use->in(idx) for loop lp if phi does not already exist |
|
966 void insert_phi_for_loop( Node* use, uint idx, Node* lp_entry_val, Node* back_edge_val, LoopNode* lp ); |
|
967 #ifdef ASSERT |
|
968 // Validate the loop partition sets: peel and not_peel |
|
969 bool is_valid_loop_partition( IdealLoopTree *loop, VectorSet& peel, Node_List& peel_list, VectorSet& not_peel ); |
|
970 // Ensure that uses outside of loop are of the right form |
|
971 bool is_valid_clone_loop_form( IdealLoopTree *loop, Node_List& peel_list, |
|
972 uint orig_exit_idx, uint clone_exit_idx); |
|
973 bool is_valid_clone_loop_exit_use( IdealLoopTree *loop, Node* use, uint exit_idx); |
|
974 #endif |
|
975 |
|
976 // Returns nonzero constant stride if-node is a possible iv test (otherwise returns zero.) |
|
977 int stride_of_possible_iv( Node* iff ); |
|
978 bool is_possible_iv_test( Node* iff ) { return stride_of_possible_iv(iff) != 0; } |
|
979 // Return the (unique) control output node that's in the loop (if it exists.) |
|
980 Node* stay_in_loop( Node* n, IdealLoopTree *loop); |
|
981 // Insert a signed compare loop exit cloned from an unsigned compare. |
|
982 IfNode* insert_cmpi_loop_exit(IfNode* if_cmpu, IdealLoopTree *loop); |
|
983 void remove_cmpi_loop_exit(IfNode* if_cmp, IdealLoopTree *loop); |
|
984 // Utility to register node "n" with PhaseIdealLoop |
|
985 void register_node(Node* n, IdealLoopTree *loop, Node* pred, int ddepth); |
|
986 // Utility to create an if-projection |
|
987 ProjNode* proj_clone(ProjNode* p, IfNode* iff); |
|
988 // Force the iff control output to be the live_proj |
|
989 Node* short_circuit_if(IfNode* iff, ProjNode* live_proj); |
|
990 // Insert a region before an if projection |
|
991 RegionNode* insert_region_before_proj(ProjNode* proj); |
|
992 // Insert a new if before an if projection |
|
993 ProjNode* insert_if_before_proj(Node* left, bool Signed, BoolTest::mask relop, Node* right, ProjNode* proj); |
|
994 |
|
995 // Passed in a Phi merging (recursively) some nearly equivalent Bool/Cmps. |
|
996 // "Nearly" because all Nodes have been cloned from the original in the loop, |
|
997 // but the fall-in edges to the Cmp are different. Clone bool/Cmp pairs |
|
998 // through the Phi recursively, and return a Bool. |
|
999 BoolNode *clone_iff( PhiNode *phi, IdealLoopTree *loop ); |
|
1000 CmpNode *clone_bool( PhiNode *phi, IdealLoopTree *loop ); |
|
1001 |
|
1002 |
|
1003 // Rework addressing expressions to get the most loop-invariant stuff |
|
1004 // moved out. We'd like to do all associative operators, but it's especially |
|
1005 // important (common) to do address expressions. |
|
1006 Node *remix_address_expressions( Node *n ); |
|
1007 |
|
1008 // Attempt to use a conditional move instead of a phi/branch |
|
1009 Node *conditional_move( Node *n ); |
|
1010 |
|
1011 // Reorganize offset computations to lower register pressure. |
|
1012 // Mostly prevent loop-fallout uses of the pre-incremented trip counter |
|
1013 // (which are then alive with the post-incremented trip counter |
|
1014 // forcing an extra register move) |
|
1015 void reorg_offsets( IdealLoopTree *loop ); |
|
1016 |
|
1017 // Check for aggressive application of 'split-if' optimization, |
|
1018 // using basic block level info. |
|
1019 void split_if_with_blocks ( VectorSet &visited, Node_Stack &nstack ); |
|
1020 Node *split_if_with_blocks_pre ( Node *n ); |
|
1021 void split_if_with_blocks_post( Node *n ); |
|
1022 Node *has_local_phi_input( Node *n ); |
|
1023 // Mark an IfNode as being dominated by a prior test, |
|
1024 // without actually altering the CFG (and hence IDOM info). |
|
1025 void dominated_by( Node *prevdom, Node *iff, bool flip = false, bool exclude_loop_predicate = false ); |
|
1026 |
|
1027 // Split Node 'n' through merge point |
|
1028 Node *split_thru_region( Node *n, Node *region ); |
|
1029 // Split Node 'n' through merge point if there is enough win. |
|
1030 Node *split_thru_phi( Node *n, Node *region, int policy ); |
|
1031 // Found an If getting its condition-code input from a Phi in the |
|
1032 // same block. Split thru the Region. |
|
1033 void do_split_if( Node *iff ); |
|
1034 |
|
1035 // Conversion of fill/copy patterns into intrisic versions |
|
1036 bool do_intrinsify_fill(); |
|
1037 bool intrinsify_fill(IdealLoopTree* lpt); |
|
1038 bool match_fill_loop(IdealLoopTree* lpt, Node*& store, Node*& store_value, |
|
1039 Node*& shift, Node*& offset); |
|
1040 |
|
1041 private: |
|
1042 // Return a type based on condition control flow |
|
1043 const TypeInt* filtered_type( Node *n, Node* n_ctrl); |
|
1044 const TypeInt* filtered_type( Node *n ) { return filtered_type(n, NULL); } |
|
1045 // Helpers for filtered type |
|
1046 const TypeInt* filtered_type_from_dominators( Node* val, Node *val_ctrl); |
|
1047 |
|
1048 // Helper functions |
|
1049 Node *spinup( Node *iff, Node *new_false, Node *new_true, Node *region, Node *phi, small_cache *cache ); |
|
1050 Node *find_use_block( Node *use, Node *def, Node *old_false, Node *new_false, Node *old_true, Node *new_true ); |
|
1051 void handle_use( Node *use, Node *def, small_cache *cache, Node *region_dom, Node *new_false, Node *new_true, Node *old_false, Node *old_true ); |
|
1052 bool split_up( Node *n, Node *blk1, Node *blk2 ); |
|
1053 void sink_use( Node *use, Node *post_loop ); |
|
1054 Node *place_near_use( Node *useblock ) const; |
|
1055 |
|
1056 bool _created_loop_node; |
|
1057 public: |
|
1058 void set_created_loop_node() { _created_loop_node = true; } |
|
1059 bool created_loop_node() { return _created_loop_node; } |
|
1060 void register_new_node( Node *n, Node *blk ); |
|
1061 |
|
1062 #ifdef ASSERT |
|
1063 void dump_bad_graph(const char* msg, Node* n, Node* early, Node* LCA); |
|
1064 #endif |
|
1065 |
|
1066 #ifndef PRODUCT |
|
1067 void dump( ) const; |
|
1068 void dump( IdealLoopTree *loop, uint rpo_idx, Node_List &rpo_list ) const; |
|
1069 void rpo( Node *start, Node_Stack &stk, VectorSet &visited, Node_List &rpo_list ) const; |
|
1070 void verify() const; // Major slow :-) |
|
1071 void verify_compare( Node *n, const PhaseIdealLoop *loop_verify, VectorSet &visited ) const; |
|
1072 IdealLoopTree *get_loop_idx(Node* n) const { |
|
1073 // Dead nodes have no loop, so return the top level loop instead |
|
1074 return _nodes[n->_idx] ? (IdealLoopTree*)_nodes[n->_idx] : _ltree_root; |
|
1075 } |
|
1076 // Print some stats |
|
1077 static void print_statistics(); |
|
1078 static int _loop_invokes; // Count of PhaseIdealLoop invokes |
|
1079 static int _loop_work; // Sum of PhaseIdealLoop x _unique |
|
1080 #endif |
|
1081 }; |
|
1082 |
|
1083 inline Node* IdealLoopTree::tail() { |
|
1084 // Handle lazy update of _tail field |
|
1085 Node *n = _tail; |
|
1086 //while( !n->in(0) ) // Skip dead CFG nodes |
|
1087 //n = n->in(1); |
|
1088 if (n->in(0) == NULL) |
|
1089 n = _phase->get_ctrl(n); |
|
1090 _tail = n; |
|
1091 return n; |
|
1092 } |
|
1093 |
|
1094 |
|
1095 // Iterate over the loop tree using a preorder, left-to-right traversal. |
|
1096 // |
|
1097 // Example that visits all counted loops from within PhaseIdealLoop |
|
1098 // |
|
1099 // for (LoopTreeIterator iter(_ltree_root); !iter.done(); iter.next()) { |
|
1100 // IdealLoopTree* lpt = iter.current(); |
|
1101 // if (!lpt->is_counted()) continue; |
|
1102 // ... |
|
1103 class LoopTreeIterator : public StackObj { |
|
1104 private: |
|
1105 IdealLoopTree* _root; |
|
1106 IdealLoopTree* _curnt; |
|
1107 |
|
1108 public: |
|
1109 LoopTreeIterator(IdealLoopTree* root) : _root(root), _curnt(root) {} |
|
1110 |
|
1111 bool done() { return _curnt == NULL; } // Finished iterating? |
|
1112 |
|
1113 void next(); // Advance to next loop tree |
|
1114 |
|
1115 IdealLoopTree* current() { return _curnt; } // Return current value of iterator. |
|
1116 }; |
|
1117 |
|
1118 #endif // SHARE_VM_OPTO_LOOPNODE_HPP |