1 /*
2 * Copyright (c) 1999, 2023, 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 #include "precompiled.hpp"
26 #include "gc/shared/barrierSet.hpp"
27 #include "gc/shared/c2/barrierSetC2.hpp"
28 #include "memory/allocation.inline.hpp"
29 #include "memory/resourceArea.hpp"
30 #include "opto/addnode.hpp"
31 #include "opto/callnode.hpp"
32 #include "opto/castnode.hpp"
33 #include "opto/connode.hpp"
34 #include "opto/castnode.hpp"
35 #include "opto/divnode.hpp"
36 #include "opto/inlinetypenode.hpp"
37 #include "opto/loopnode.hpp"
38 #include "opto/matcher.hpp"
39 #include "opto/mulnode.hpp"
40 #include "opto/movenode.hpp"
41 #include "opto/opaquenode.hpp"
42 #include "opto/rootnode.hpp"
43 #include "opto/subnode.hpp"
44 #include "opto/subtypenode.hpp"
45 #include "utilities/macros.hpp"
46
47 //=============================================================================
48 //------------------------------split_thru_phi---------------------------------
49 // Split Node 'n' through merge point if there is enough win.
50 Node* PhaseIdealLoop::split_thru_phi(Node* n, Node* region, int policy) {
51 if (n->Opcode() == Op_ConvI2L && n->bottom_type() != TypeLong::LONG) {
52 // ConvI2L may have type information on it which is unsafe to push up
53 // so disable this for now
54 return NULL;
55 }
56
57 // Splitting range check CastIIs through a loop induction Phi can
58 // cause new Phis to be created that are left unrelated to the loop
59 // induction Phi and prevent optimizations (vectorization)
60 if (n->Opcode() == Op_CastII && region->is_CountedLoop() &&
61 n->in(1) == region->as_CountedLoop()->phi()) {
62 return NULL;
63 }
64
65 // Inline types should not be split through Phis because they cannot be merged
66 // through Phi nodes but each value input needs to be merged individually.
67 if (n->is_InlineType()) {
68 return NULL;
69 }
70
71 if (cannot_split_division(n, region)) {
72 return NULL;
73 }
74
75 int wins = 0;
76 assert(!n->is_CFG(), "");
77 assert(region->is_Region(), "");
78
79 const Type* type = n->bottom_type();
80 const TypeOopPtr* t_oop = _igvn.type(n)->isa_oopptr();
81 Node* phi;
82 if (t_oop != NULL && t_oop->is_known_instance_field()) {
83 int iid = t_oop->instance_id();
84 int index = C->get_alias_index(t_oop);
85 int offset = t_oop->offset();
86 phi = new PhiNode(region, type, NULL, iid, index, offset);
87 } else {
88 phi = PhiNode::make_blank(region, n);
89 }
90 uint old_unique = C->unique();
91 for (uint i = 1; i < region->req(); i++) {
92 Node* x;
93 Node* the_clone = NULL;
94 if (region->in(i) == C->top()) {
95 x = C->top(); // Dead path? Use a dead data op
96 } else {
97 x = n->clone(); // Else clone up the data op
98 the_clone = x; // Remember for possible deletion.
99 // Alter data node to use pre-phi inputs
100 if (n->in(0) == region)
101 x->set_req( 0, region->in(i) );
102 for (uint j = 1; j < n->req(); j++) {
103 Node* in = n->in(j);
104 if (in->is_Phi() && in->in(0) == region)
105 x->set_req(j, in->in(i)); // Use pre-Phi input for the clone
106 }
107 }
108 // Check for a 'win' on some paths
109 const Type* t = x->Value(&_igvn);
110
111 bool singleton = t->singleton();
112
113 // A TOP singleton indicates that there are no possible values incoming
114 // along a particular edge. In most cases, this is OK, and the Phi will
115 // be eliminated later in an Ideal call. However, we can't allow this to
116 // happen if the singleton occurs on loop entry, as the elimination of
117 // the PhiNode may cause the resulting node to migrate back to a previous
118 // loop iteration.
119 if (singleton && t == Type::TOP) {
120 // Is_Loop() == false does not confirm the absence of a loop (e.g., an
121 // irreducible loop may not be indicated by an affirmative is_Loop());
122 // therefore, the only top we can split thru a phi is on a backedge of
123 // a loop.
124 singleton &= region->is_Loop() && (i != LoopNode::EntryControl);
125 }
126
127 if (singleton) {
128 wins++;
129 x = ((PhaseGVN&)_igvn).makecon(t);
130 } else {
131 // We now call Identity to try to simplify the cloned node.
132 // Note that some Identity methods call phase->type(this).
133 // Make sure that the type array is big enough for
134 // our new node, even though we may throw the node away.
135 // (Note: This tweaking with igvn only works because x is a new node.)
136 _igvn.set_type(x, t);
137 // If x is a TypeNode, capture any more-precise type permanently into Node
138 // otherwise it will be not updated during igvn->transform since
139 // igvn->type(x) is set to x->Value() already.
140 x->raise_bottom_type(t);
141 Node* y = x->Identity(&_igvn);
142 if (y != x) {
143 wins++;
144 x = y;
145 } else {
146 y = _igvn.hash_find(x);
147 if (y) {
148 wins++;
149 x = y;
150 } else {
151 // Else x is a new node we are keeping
152 // We do not need register_new_node_with_optimizer
153 // because set_type has already been called.
154 _igvn._worklist.push(x);
155 }
156 }
157 }
158 if (x != the_clone && the_clone != NULL)
159 _igvn.remove_dead_node(the_clone);
160 phi->set_req( i, x );
161 }
162 // Too few wins?
163 if (wins <= policy) {
164 _igvn.remove_dead_node(phi);
165 return NULL;
166 }
167
168 // Record Phi
169 register_new_node( phi, region );
170
171 for (uint i2 = 1; i2 < phi->req(); i2++) {
172 Node *x = phi->in(i2);
173 // If we commoned up the cloned 'x' with another existing Node,
174 // the existing Node picks up a new use. We need to make the
175 // existing Node occur higher up so it dominates its uses.
176 Node *old_ctrl;
177 IdealLoopTree *old_loop;
178
179 if (x->is_Con()) {
180 // Constant's control is always root.
181 set_ctrl(x, C->root());
182 continue;
183 }
184 // The occasional new node
185 if (x->_idx >= old_unique) { // Found a new, unplaced node?
186 old_ctrl = NULL;
187 old_loop = NULL; // Not in any prior loop
188 } else {
189 old_ctrl = get_ctrl(x);
190 old_loop = get_loop(old_ctrl); // Get prior loop
191 }
192 // New late point must dominate new use
193 Node *new_ctrl = dom_lca(old_ctrl, region->in(i2));
194 if (new_ctrl == old_ctrl) // Nothing is changed
195 continue;
196
197 IdealLoopTree *new_loop = get_loop(new_ctrl);
198
199 // Don't move x into a loop if its uses are
200 // outside of loop. Otherwise x will be cloned
201 // for each use outside of this loop.
202 IdealLoopTree *use_loop = get_loop(region);
203 if (!new_loop->is_member(use_loop) &&
204 (old_loop == NULL || !new_loop->is_member(old_loop))) {
205 // Take early control, later control will be recalculated
206 // during next iteration of loop optimizations.
207 new_ctrl = get_early_ctrl(x);
208 new_loop = get_loop(new_ctrl);
209 }
210 // Set new location
211 set_ctrl(x, new_ctrl);
212 // If changing loop bodies, see if we need to collect into new body
213 if (old_loop != new_loop) {
214 if (old_loop && !old_loop->_child)
215 old_loop->_body.yank(x);
216 if (!new_loop->_child)
217 new_loop->_body.push(x); // Collect body info
218 }
219 }
220
221 return phi;
222 }
223
224 // Return true if 'n' is a Div or Mod node (without zero check If node which was removed earlier) with a loop phi divisor
225 // of a trip-counted (integer or long) loop with a backedge input that could be zero (include zero in its type range). In
226 // this case, we cannot split the division to the backedge as it could freely float above the loop exit check resulting in
227 // a division by zero. This situation is possible because the type of an increment node of an iv phi (trip-counter) could
228 // include zero while the iv phi does not (see PhiNode::Value() for trip-counted loops where we improve types of iv phis).
229 // We also need to check other loop phis as they could have been created in the same split-if pass when applying
230 // PhaseIdealLoop::split_thru_phi() to split nodes through an iv phi.
231 bool PhaseIdealLoop::cannot_split_division(const Node* n, const Node* region) const {
232 const Type* zero;
233 switch (n->Opcode()) {
234 case Op_DivI:
235 case Op_ModI:
236 zero = TypeInt::ZERO;
237 break;
238 case Op_DivL:
239 case Op_ModL:
240 zero = TypeLong::ZERO;
241 break;
242 default:
243 return false;
244 }
245
246 assert(n->in(0) == NULL, "divisions with zero check should already have bailed out earlier in split-if");
247 Node* divisor = n->in(2);
248 return is_divisor_counted_loop_phi(divisor, region) &&
249 loop_phi_backedge_type_contains_zero(divisor, zero);
250 }
251
252 bool PhaseIdealLoop::is_divisor_counted_loop_phi(const Node* divisor, const Node* loop) {
253 return loop->is_BaseCountedLoop() && divisor->is_Phi() && divisor->in(0) == loop;
254 }
255
256 bool PhaseIdealLoop::loop_phi_backedge_type_contains_zero(const Node* phi_divisor, const Type* zero) const {
257 return _igvn.type(phi_divisor->in(LoopNode::LoopBackControl))->filter_speculative(zero) != Type::TOP;
258 }
259
260 //------------------------------dominated_by------------------------------------
261 // Replace the dominated test with an obvious true or false. Place it on the
262 // IGVN worklist for later cleanup. Move control-dependent data Nodes on the
263 // live path up to the dominating control.
264 void PhaseIdealLoop::dominated_by(IfProjNode* prevdom, IfNode* iff, bool flip, bool exclude_loop_predicate) {
265 if (VerifyLoopOptimizations && PrintOpto) { tty->print_cr("dominating test"); }
266
267 // prevdom is the dominating projection of the dominating test.
268 assert(iff->Opcode() == Op_If ||
269 iff->Opcode() == Op_CountedLoopEnd ||
270 iff->Opcode() == Op_LongCountedLoopEnd ||
271 iff->Opcode() == Op_RangeCheck,
272 "Check this code when new subtype is added");
273
274 int pop = prevdom->Opcode();
275 assert( pop == Op_IfFalse || pop == Op_IfTrue, "" );
276 if (flip) {
277 if (pop == Op_IfTrue)
278 pop = Op_IfFalse;
279 else
280 pop = Op_IfTrue;
281 }
282 // 'con' is set to true or false to kill the dominated test.
283 Node *con = _igvn.makecon(pop == Op_IfTrue ? TypeInt::ONE : TypeInt::ZERO);
284 set_ctrl(con, C->root()); // Constant gets a new use
285 // Hack the dominated test
286 _igvn.replace_input_of(iff, 1, con);
287
288 // If I dont have a reachable TRUE and FALSE path following the IfNode then
289 // I can assume this path reaches an infinite loop. In this case it's not
290 // important to optimize the data Nodes - either the whole compilation will
291 // be tossed or this path (and all data Nodes) will go dead.
292 if (iff->outcnt() != 2) return;
293
294 // Make control-dependent data Nodes on the live path (path that will remain
295 // once the dominated IF is removed) become control-dependent on the
296 // dominating projection.
297 Node* dp = iff->proj_out_or_null(pop == Op_IfTrue);
298
299 // Loop predicates may have depending checks which should not
300 // be skipped. For example, range check predicate has two checks
301 // for lower and upper bounds.
302 if (dp == NULL)
303 return;
304
305 ProjNode* dp_proj = dp->as_Proj();
306 ProjNode* unc_proj = iff->proj_out(1 - dp_proj->_con)->as_Proj();
307 if (exclude_loop_predicate &&
308 (unc_proj->is_uncommon_trap_proj(Deoptimization::Reason_predicate) != NULL ||
309 unc_proj->is_uncommon_trap_proj(Deoptimization::Reason_profile_predicate) != NULL ||
310 unc_proj->is_uncommon_trap_proj(Deoptimization::Reason_range_check) != NULL)) {
311 // If this is a range check (IfNode::is_range_check), do not
312 // reorder because Compile::allow_range_check_smearing might have
313 // changed the check.
314 return; // Let IGVN transformation change control dependence.
315 }
316
317 IdealLoopTree* old_loop = get_loop(dp);
318
319 for (DUIterator_Fast imax, i = dp->fast_outs(imax); i < imax; i++) {
320 Node* cd = dp->fast_out(i); // Control-dependent node
321 // Do not rewire Div and Mod nodes which could have a zero divisor to avoid skipping their zero check.
322 if (cd->depends_only_on_test() && _igvn.no_dependent_zero_check(cd)) {
323 assert(cd->in(0) == dp, "");
324 _igvn.replace_input_of(cd, 0, prevdom);
325 set_early_ctrl(cd, false);
326 IdealLoopTree* new_loop = get_loop(get_ctrl(cd));
327 if (old_loop != new_loop) {
328 if (!old_loop->_child) {
329 old_loop->_body.yank(cd);
330 }
331 if (!new_loop->_child) {
332 new_loop->_body.push(cd);
333 }
334 }
335 --i;
336 --imax;
337 }
338 }
339 }
340
341 //------------------------------has_local_phi_input----------------------------
342 // Return TRUE if 'n' has Phi inputs from its local block and no other
343 // block-local inputs (all non-local-phi inputs come from earlier blocks)
344 Node *PhaseIdealLoop::has_local_phi_input( Node *n ) {
345 Node *n_ctrl = get_ctrl(n);
346 // See if some inputs come from a Phi in this block, or from before
347 // this block.
348 uint i;
349 for( i = 1; i < n->req(); i++ ) {
350 Node *phi = n->in(i);
351 if( phi->is_Phi() && phi->in(0) == n_ctrl )
352 break;
353 }
354 if( i >= n->req() )
355 return NULL; // No Phi inputs; nowhere to clone thru
356
357 // Check for inputs created between 'n' and the Phi input. These
358 // must split as well; they have already been given the chance
359 // (courtesy of a post-order visit) and since they did not we must
360 // recover the 'cost' of splitting them by being very profitable
361 // when splitting 'n'. Since this is unlikely we simply give up.
362 for( i = 1; i < n->req(); i++ ) {
363 Node *m = n->in(i);
364 if( get_ctrl(m) == n_ctrl && !m->is_Phi() ) {
365 // We allow the special case of AddP's with no local inputs.
366 // This allows us to split-up address expressions.
367 if (m->is_AddP() &&
368 get_ctrl(m->in(AddPNode::Base)) != n_ctrl &&
369 get_ctrl(m->in(AddPNode::Address)) != n_ctrl &&
370 get_ctrl(m->in(AddPNode::Offset)) != n_ctrl) {
371 // Move the AddP up to the dominating point. That's fine because control of m's inputs
372 // must dominate get_ctrl(m) == n_ctrl and we just checked that the input controls are != n_ctrl.
373 Node* c = find_non_split_ctrl(idom(n_ctrl));
374 if (c->is_OuterStripMinedLoop()) {
375 c->as_Loop()->verify_strip_mined(1);
376 c = c->in(LoopNode::EntryControl);
377 }
378 set_ctrl_and_loop(m, c);
379 continue;
380 }
381 return NULL;
382 }
383 assert(n->is_Phi() || m->is_Phi() || is_dominator(get_ctrl(m), n_ctrl), "m has strange control");
384 }
385
386 return n_ctrl;
387 }
388
389 // Replace expressions like ((V+I) << 2) with (V<<2 + I<<2).
390 Node* PhaseIdealLoop::remix_address_expressions_add_left_shift(Node* n, IdealLoopTree* n_loop, Node* n_ctrl, BasicType bt) {
391 assert(bt == T_INT || bt == T_LONG, "only for integers");
392 int n_op = n->Opcode();
393
394 if (n_op == Op_LShift(bt)) {
395 // Scale is loop invariant
396 Node* scale = n->in(2);
397 Node* scale_ctrl = get_ctrl(scale);
398 IdealLoopTree* scale_loop = get_loop(scale_ctrl);
399 if (n_loop == scale_loop || !scale_loop->is_member(n_loop)) {
400 return NULL;
401 }
402 const TypeInt* scale_t = scale->bottom_type()->isa_int();
403 if (scale_t != NULL && scale_t->is_con() && scale_t->get_con() >= 16) {
404 return NULL; // Dont bother with byte/short masking
405 }
406 // Add must vary with loop (else shift would be loop-invariant)
407 Node* add = n->in(1);
408 Node* add_ctrl = get_ctrl(add);
409 IdealLoopTree* add_loop = get_loop(add_ctrl);
410 if (n_loop != add_loop) {
411 return NULL; // happens w/ evil ZKM loops
412 }
413
414 // Convert I-V into I+ (0-V); same for V-I
415 if (add->Opcode() == Op_Sub(bt) &&
416 _igvn.type(add->in(1)) != TypeInteger::zero(bt)) {
417 assert(add->Opcode() == Op_SubI || add->Opcode() == Op_SubL, "");
418 Node* zero = _igvn.integercon(0, bt);
419 set_ctrl(zero, C->root());
420 Node* neg = SubNode::make(zero, add->in(2), bt);
421 register_new_node(neg, get_ctrl(add->in(2)));
422 add = AddNode::make(add->in(1), neg, bt);
423 register_new_node(add, add_ctrl);
424 }
425 if (add->Opcode() != Op_Add(bt)) return NULL;
426 assert(add->Opcode() == Op_AddI || add->Opcode() == Op_AddL, "");
427 // See if one add input is loop invariant
428 Node* add_var = add->in(1);
429 Node* add_var_ctrl = get_ctrl(add_var);
430 IdealLoopTree* add_var_loop = get_loop(add_var_ctrl);
431 Node* add_invar = add->in(2);
432 Node* add_invar_ctrl = get_ctrl(add_invar);
433 IdealLoopTree* add_invar_loop = get_loop(add_invar_ctrl);
434 if (add_invar_loop == n_loop) {
435 // Swap to find the invariant part
436 add_invar = add_var;
437 add_invar_ctrl = add_var_ctrl;
438 add_invar_loop = add_var_loop;
439 add_var = add->in(2);
440 } else if (add_var_loop != n_loop) { // Else neither input is loop invariant
441 return NULL;
442 }
443 if (n_loop == add_invar_loop || !add_invar_loop->is_member(n_loop)) {
444 return NULL; // No invariant part of the add?
445 }
446
447 // Yes! Reshape address expression!
448 Node* inv_scale = LShiftNode::make(add_invar, scale, bt);
449 Node* inv_scale_ctrl =
450 dom_depth(add_invar_ctrl) > dom_depth(scale_ctrl) ?
451 add_invar_ctrl : scale_ctrl;
452 register_new_node(inv_scale, inv_scale_ctrl);
453 Node* var_scale = LShiftNode::make(add_var, scale, bt);
454 register_new_node(var_scale, n_ctrl);
455 Node* var_add = AddNode::make(var_scale, inv_scale, bt);
456 register_new_node(var_add, n_ctrl);
457 _igvn.replace_node(n, var_add);
458 return var_add;
459 }
460 return NULL;
461 }
462
463 //------------------------------remix_address_expressions----------------------
464 // Rework addressing expressions to get the most loop-invariant stuff
465 // moved out. We'd like to do all associative operators, but it's especially
466 // important (common) to do address expressions.
467 Node* PhaseIdealLoop::remix_address_expressions(Node* n) {
468 if (!has_ctrl(n)) return NULL;
469 Node* n_ctrl = get_ctrl(n);
470 IdealLoopTree* n_loop = get_loop(n_ctrl);
471
472 // See if 'n' mixes loop-varying and loop-invariant inputs and
473 // itself is loop-varying.
474
475 // Only interested in binary ops (and AddP)
476 if (n->req() < 3 || n->req() > 4) return NULL;
477
478 Node* n1_ctrl = get_ctrl(n->in( 1));
479 Node* n2_ctrl = get_ctrl(n->in( 2));
480 Node* n3_ctrl = get_ctrl(n->in(n->req() == 3 ? 2 : 3));
481 IdealLoopTree* n1_loop = get_loop(n1_ctrl);
482 IdealLoopTree* n2_loop = get_loop(n2_ctrl);
483 IdealLoopTree* n3_loop = get_loop(n3_ctrl);
484
485 // Does one of my inputs spin in a tighter loop than self?
486 if ((n_loop->is_member(n1_loop) && n_loop != n1_loop) ||
487 (n_loop->is_member(n2_loop) && n_loop != n2_loop) ||
488 (n_loop->is_member(n3_loop) && n_loop != n3_loop)) {
489 return NULL; // Leave well enough alone
490 }
491
492 // Is at least one of my inputs loop-invariant?
493 if (n1_loop == n_loop &&
494 n2_loop == n_loop &&
495 n3_loop == n_loop) {
496 return NULL; // No loop-invariant inputs
497 }
498
499 Node* res = remix_address_expressions_add_left_shift(n, n_loop, n_ctrl, T_INT);
500 if (res != NULL) {
501 return res;
502 }
503 res = remix_address_expressions_add_left_shift(n, n_loop, n_ctrl, T_LONG);
504 if (res != NULL) {
505 return res;
506 }
507
508 int n_op = n->Opcode();
509 // Replace (I+V) with (V+I)
510 if (n_op == Op_AddI ||
511 n_op == Op_AddL ||
512 n_op == Op_AddF ||
513 n_op == Op_AddD ||
514 n_op == Op_MulI ||
515 n_op == Op_MulL ||
516 n_op == Op_MulF ||
517 n_op == Op_MulD) {
518 if (n2_loop == n_loop) {
519 assert(n1_loop != n_loop, "");
520 n->swap_edges(1, 2);
521 }
522 }
523
524 // Replace ((I1 +p V) +p I2) with ((I1 +p I2) +p V),
525 // but not if I2 is a constant.
526 if (n_op == Op_AddP) {
527 if (n2_loop == n_loop && n3_loop != n_loop) {
528 if (n->in(2)->Opcode() == Op_AddP && !n->in(3)->is_Con()) {
529 Node* n22_ctrl = get_ctrl(n->in(2)->in(2));
530 Node* n23_ctrl = get_ctrl(n->in(2)->in(3));
531 IdealLoopTree* n22loop = get_loop(n22_ctrl);
532 IdealLoopTree* n23_loop = get_loop(n23_ctrl);
533 if (n22loop != n_loop && n22loop->is_member(n_loop) &&
534 n23_loop == n_loop) {
535 Node* add1 = new AddPNode(n->in(1), n->in(2)->in(2), n->in(3));
536 // Stuff new AddP in the loop preheader
537 register_new_node(add1, n_loop->_head->in(LoopNode::EntryControl));
538 Node* add2 = new AddPNode(n->in(1), add1, n->in(2)->in(3));
539 register_new_node(add2, n_ctrl);
540 _igvn.replace_node(n, add2);
541 return add2;
542 }
543 }
544 }
545
546 // Replace (I1 +p (I2 + V)) with ((I1 +p I2) +p V)
547 if (n2_loop != n_loop && n3_loop == n_loop) {
548 if (n->in(3)->Opcode() == Op_AddX) {
549 Node* V = n->in(3)->in(1);
550 Node* I = n->in(3)->in(2);
551 if (is_member(n_loop,get_ctrl(V))) {
552 } else {
553 Node *tmp = V; V = I; I = tmp;
554 }
555 if (!is_member(n_loop,get_ctrl(I))) {
556 Node* add1 = new AddPNode(n->in(1), n->in(2), I);
557 // Stuff new AddP in the loop preheader
558 register_new_node(add1, n_loop->_head->in(LoopNode::EntryControl));
559 Node* add2 = new AddPNode(n->in(1), add1, V);
560 register_new_node(add2, n_ctrl);
561 _igvn.replace_node(n, add2);
562 return add2;
563 }
564 }
565 }
566 }
567
568 return NULL;
569 }
570
571 // Optimize ((in1[2*i] * in2[2*i]) + (in1[2*i+1] * in2[2*i+1]))
572 Node *PhaseIdealLoop::convert_add_to_muladd(Node* n) {
573 assert(n->Opcode() == Op_AddI, "sanity");
574 Node * nn = NULL;
575 Node * in1 = n->in(1);
576 Node * in2 = n->in(2);
577 if (in1->Opcode() == Op_MulI && in2->Opcode() == Op_MulI) {
578 IdealLoopTree* loop_n = get_loop(get_ctrl(n));
579 if (loop_n->is_counted() &&
580 loop_n->_head->as_Loop()->is_valid_counted_loop(T_INT) &&
581 Matcher::match_rule_supported(Op_MulAddVS2VI) &&
582 Matcher::match_rule_supported(Op_MulAddS2I)) {
583 Node* mul_in1 = in1->in(1);
584 Node* mul_in2 = in1->in(2);
585 Node* mul_in3 = in2->in(1);
586 Node* mul_in4 = in2->in(2);
587 if (mul_in1->Opcode() == Op_LoadS &&
588 mul_in2->Opcode() == Op_LoadS &&
589 mul_in3->Opcode() == Op_LoadS &&
590 mul_in4->Opcode() == Op_LoadS) {
591 IdealLoopTree* loop1 = get_loop(get_ctrl(mul_in1));
592 IdealLoopTree* loop2 = get_loop(get_ctrl(mul_in2));
593 IdealLoopTree* loop3 = get_loop(get_ctrl(mul_in3));
594 IdealLoopTree* loop4 = get_loop(get_ctrl(mul_in4));
595 IdealLoopTree* loop5 = get_loop(get_ctrl(in1));
596 IdealLoopTree* loop6 = get_loop(get_ctrl(in2));
597 // All nodes should be in the same counted loop.
598 if (loop_n == loop1 && loop_n == loop2 && loop_n == loop3 &&
599 loop_n == loop4 && loop_n == loop5 && loop_n == loop6) {
600 Node* adr1 = mul_in1->in(MemNode::Address);
601 Node* adr2 = mul_in2->in(MemNode::Address);
602 Node* adr3 = mul_in3->in(MemNode::Address);
603 Node* adr4 = mul_in4->in(MemNode::Address);
604 if (adr1->is_AddP() && adr2->is_AddP() && adr3->is_AddP() && adr4->is_AddP()) {
605 if ((adr1->in(AddPNode::Base) == adr3->in(AddPNode::Base)) &&
606 (adr2->in(AddPNode::Base) == adr4->in(AddPNode::Base))) {
607 nn = new MulAddS2INode(mul_in1, mul_in2, mul_in3, mul_in4);
608 register_new_node(nn, get_ctrl(n));
609 _igvn.replace_node(n, nn);
610 return nn;
611 } else if ((adr1->in(AddPNode::Base) == adr4->in(AddPNode::Base)) &&
612 (adr2->in(AddPNode::Base) == adr3->in(AddPNode::Base))) {
613 nn = new MulAddS2INode(mul_in1, mul_in2, mul_in4, mul_in3);
614 register_new_node(nn, get_ctrl(n));
615 _igvn.replace_node(n, nn);
616 return nn;
617 }
618 }
619 }
620 }
621 }
622 }
623 return nn;
624 }
625
626 //------------------------------conditional_move-------------------------------
627 // Attempt to replace a Phi with a conditional move. We have some pretty
628 // strict profitability requirements. All Phis at the merge point must
629 // be converted, so we can remove the control flow. We need to limit the
630 // number of c-moves to a small handful. All code that was in the side-arms
631 // of the CFG diamond is now speculatively executed. This code has to be
632 // "cheap enough". We are pretty much limited to CFG diamonds that merge
633 // 1 or 2 items with a total of 1 or 2 ops executed speculatively.
634 Node *PhaseIdealLoop::conditional_move( Node *region ) {
635
636 assert(region->is_Region(), "sanity check");
637 if (region->req() != 3) return NULL;
638
639 // Check for CFG diamond
640 Node *lp = region->in(1);
641 Node *rp = region->in(2);
642 if (!lp || !rp) return NULL;
643 Node *lp_c = lp->in(0);
644 if (lp_c == NULL || lp_c != rp->in(0) || !lp_c->is_If()) return NULL;
645 IfNode *iff = lp_c->as_If();
646
647 // Check for ops pinned in an arm of the diamond.
648 // Can't remove the control flow in this case
649 if (lp->outcnt() > 1) return NULL;
650 if (rp->outcnt() > 1) return NULL;
651
652 IdealLoopTree* r_loop = get_loop(region);
653 assert(r_loop == get_loop(iff), "sanity");
654 // Always convert to CMOVE if all results are used only outside this loop.
655 bool used_inside_loop = (r_loop == _ltree_root);
656
657 // Check profitability
658 int cost = 0;
659 int phis = 0;
660 for (DUIterator_Fast imax, i = region->fast_outs(imax); i < imax; i++) {
661 Node *out = region->fast_out(i);
662 if (!out->is_Phi()) continue; // Ignore other control edges, etc
663 phis++;
664 PhiNode* phi = out->as_Phi();
665 BasicType bt = phi->type()->basic_type();
666 switch (bt) {
667 case T_DOUBLE:
668 case T_FLOAT:
669 if (C->use_cmove()) {
670 continue; //TODO: maybe we want to add some cost
671 }
672 cost += Matcher::float_cmove_cost(); // Could be very expensive
673 break;
674 case T_LONG: {
675 cost += Matcher::long_cmove_cost(); // May encodes as 2 CMOV's
676 }
677 case T_INT: // These all CMOV fine
678 case T_ADDRESS: { // (RawPtr)
679 cost++;
680 break;
681 }
682 case T_NARROWOOP: // Fall through
683 case T_OBJECT: { // Base oops are OK, but not derived oops
684 const TypeOopPtr *tp = phi->type()->make_ptr()->isa_oopptr();
685 // Derived pointers are Bad (tm): what's the Base (for GC purposes) of a
686 // CMOVE'd derived pointer? It's a CMOVE'd derived base. Thus
687 // CMOVE'ing a derived pointer requires we also CMOVE the base. If we
688 // have a Phi for the base here that we convert to a CMOVE all is well
689 // and good. But if the base is dead, we'll not make a CMOVE. Later
690 // the allocator will have to produce a base by creating a CMOVE of the
691 // relevant bases. This puts the allocator in the business of
692 // manufacturing expensive instructions, generally a bad plan.
693 // Just Say No to Conditionally-Moved Derived Pointers.
694 if (tp && tp->offset() != 0)
695 return NULL;
696 cost++;
697 break;
698 }
699 default:
700 return NULL; // In particular, can't do memory or I/O
701 }
702 // Add in cost any speculative ops
703 for (uint j = 1; j < region->req(); j++) {
704 Node *proj = region->in(j);
705 Node *inp = phi->in(j);
706 if (get_ctrl(inp) == proj) { // Found local op
707 cost++;
708 // Check for a chain of dependent ops; these will all become
709 // speculative in a CMOV.
710 for (uint k = 1; k < inp->req(); k++)
711 if (get_ctrl(inp->in(k)) == proj)
712 cost += ConditionalMoveLimit; // Too much speculative goo
713 }
714 }
715 // See if the Phi is used by a Cmp or Narrow oop Decode/Encode.
716 // This will likely Split-If, a higher-payoff operation.
717 for (DUIterator_Fast kmax, k = phi->fast_outs(kmax); k < kmax; k++) {
718 Node* use = phi->fast_out(k);
719 if (use->is_Cmp() || use->is_DecodeNarrowPtr() || use->is_EncodeNarrowPtr())
720 cost += ConditionalMoveLimit;
721 // Is there a use inside the loop?
722 // Note: check only basic types since CMoveP is pinned.
723 if (!used_inside_loop && is_java_primitive(bt)) {
724 IdealLoopTree* u_loop = get_loop(has_ctrl(use) ? get_ctrl(use) : use);
725 if (r_loop == u_loop || r_loop->is_member(u_loop)) {
726 used_inside_loop = true;
727 }
728 }
729 }
730 }//for
731 Node* bol = iff->in(1);
732 if (bol->Opcode() == Op_Opaque4) {
733 return NULL; // Ignore loop predicate checks (the Opaque4 ensures they will go away)
734 }
735 assert(bol->Opcode() == Op_Bool, "Unexpected node");
736 int cmp_op = bol->in(1)->Opcode();
737 if (cmp_op == Op_SubTypeCheck) { // SubTypeCheck expansion expects an IfNode
738 return NULL;
739 }
740 // It is expensive to generate flags from a float compare.
741 // Avoid duplicated float compare.
742 if (phis > 1 && (cmp_op == Op_CmpF || cmp_op == Op_CmpD)) return NULL;
743
744 float infrequent_prob = PROB_UNLIKELY_MAG(3);
745 // Ignore cost and blocks frequency if CMOVE can be moved outside the loop.
746 if (used_inside_loop) {
747 if (cost >= ConditionalMoveLimit) return NULL; // Too much goo
748
749 // BlockLayoutByFrequency optimization moves infrequent branch
750 // from hot path. No point in CMOV'ing in such case (110 is used
751 // instead of 100 to take into account not exactness of float value).
752 if (BlockLayoutByFrequency) {
753 infrequent_prob = MAX2(infrequent_prob, (float)BlockLayoutMinDiamondPercentage/110.0f);
754 }
755 }
756 // Check for highly predictable branch. No point in CMOV'ing if
757 // we are going to predict accurately all the time.
758 if (C->use_cmove() && (cmp_op == Op_CmpF || cmp_op == Op_CmpD)) {
759 //keep going
760 } else if (iff->_prob < infrequent_prob ||
761 iff->_prob > (1.0f - infrequent_prob))
762 return NULL;
763
764 // --------------
765 // Now replace all Phis with CMOV's
766 Node *cmov_ctrl = iff->in(0);
767 uint flip = (lp->Opcode() == Op_IfTrue);
768 Node_List wq;
769 while (1) {
770 PhiNode* phi = NULL;
771 for (DUIterator_Fast imax, i = region->fast_outs(imax); i < imax; i++) {
772 Node *out = region->fast_out(i);
773 if (out->is_Phi()) {
774 phi = out->as_Phi();
775 break;
776 }
777 }
778 if (phi == NULL || _igvn.type(phi) == Type::TOP) {
779 break;
780 }
781 if (PrintOpto && VerifyLoopOptimizations) { tty->print_cr("CMOV"); }
782 // Move speculative ops
783 wq.push(phi);
784 while (wq.size() > 0) {
785 Node *n = wq.pop();
786 for (uint j = 1; j < n->req(); j++) {
787 Node* m = n->in(j);
788 if (m != NULL && !is_dominator(get_ctrl(m), cmov_ctrl)) {
789 #ifndef PRODUCT
790 if (PrintOpto && VerifyLoopOptimizations) {
791 tty->print(" speculate: ");
792 m->dump();
793 }
794 #endif
795 set_ctrl(m, cmov_ctrl);
796 wq.push(m);
797 }
798 }
799 }
800 Node *cmov = CMoveNode::make(cmov_ctrl, iff->in(1), phi->in(1+flip), phi->in(2-flip), _igvn.type(phi));
801 register_new_node( cmov, cmov_ctrl );
802 _igvn.replace_node( phi, cmov );
803 #ifndef PRODUCT
804 if (TraceLoopOpts) {
805 tty->print("CMOV ");
806 r_loop->dump_head();
807 if (Verbose) {
808 bol->in(1)->dump(1);
809 cmov->dump(1);
810 }
811 }
812 if (VerifyLoopOptimizations) verify();
813 #endif
814 }
815
816 // The useless CFG diamond will fold up later; see the optimization in
817 // RegionNode::Ideal.
818 _igvn._worklist.push(region);
819
820 return iff->in(1);
821 }
822
823 static void enqueue_cfg_uses(Node* m, Unique_Node_List& wq) {
824 for (DUIterator_Fast imax, i = m->fast_outs(imax); i < imax; i++) {
825 Node* u = m->fast_out(i);
826 if (u->is_CFG()) {
827 if (u->is_NeverBranch()) {
828 u = u->as_NeverBranch()->proj_out(0);
829 enqueue_cfg_uses(u, wq);
830 } else {
831 wq.push(u);
832 }
833 }
834 }
835 }
836
837 // Try moving a store out of a loop, right before the loop
838 Node* PhaseIdealLoop::try_move_store_before_loop(Node* n, Node *n_ctrl) {
839 // Store has to be first in the loop body
840 IdealLoopTree *n_loop = get_loop(n_ctrl);
841 if (n->is_Store() && n_loop != _ltree_root &&
842 n_loop->is_loop() && n_loop->_head->is_Loop() &&
843 n->in(0) != NULL) {
844 Node* address = n->in(MemNode::Address);
845 Node* value = n->in(MemNode::ValueIn);
846 Node* mem = n->in(MemNode::Memory);
847 IdealLoopTree* address_loop = get_loop(get_ctrl(address));
848 IdealLoopTree* value_loop = get_loop(get_ctrl(value));
849
850 // - address and value must be loop invariant
851 // - memory must be a memory Phi for the loop
852 // - Store must be the only store on this memory slice in the
853 // loop: if there's another store following this one then value
854 // written at iteration i by the second store could be overwritten
855 // at iteration i+n by the first store: it's not safe to move the
856 // first store out of the loop
857 // - nothing must observe the memory Phi: it guarantees no read
858 // before the store, we are also guaranteed the store post
859 // dominates the loop head (ignoring a possible early
860 // exit). Otherwise there would be extra Phi involved between the
861 // loop's Phi and the store.
862 // - there must be no early exit from the loop before the Store
863 // (such an exit most of the time would be an extra use of the
864 // memory Phi but sometimes is a bottom memory Phi that takes the
865 // store as input).
866
867 if (!n_loop->is_member(address_loop) &&
868 !n_loop->is_member(value_loop) &&
869 mem->is_Phi() && mem->in(0) == n_loop->_head &&
870 mem->outcnt() == 1 &&
871 mem->in(LoopNode::LoopBackControl) == n) {
872
873 assert(n_loop->_tail != NULL, "need a tail");
874 assert(is_dominator(n_ctrl, n_loop->_tail), "store control must not be in a branch in the loop");
875
876 // Verify that there's no early exit of the loop before the store.
877 bool ctrl_ok = false;
878 {
879 // Follow control from loop head until n, we exit the loop or
880 // we reach the tail
881 ResourceMark rm;
882 Unique_Node_List wq;
883 wq.push(n_loop->_head);
884
885 for (uint next = 0; next < wq.size(); ++next) {
886 Node *m = wq.at(next);
887 if (m == n->in(0)) {
888 ctrl_ok = true;
889 continue;
890 }
891 assert(!has_ctrl(m), "should be CFG");
892 if (!n_loop->is_member(get_loop(m)) || m == n_loop->_tail) {
893 ctrl_ok = false;
894 break;
895 }
896 enqueue_cfg_uses(m, wq);
897 if (wq.size() > 10) {
898 ctrl_ok = false;
899 break;
900 }
901 }
902 }
903 if (ctrl_ok) {
904 // move the Store
905 _igvn.replace_input_of(mem, LoopNode::LoopBackControl, mem);
906 _igvn.replace_input_of(n, 0, n_loop->_head->as_Loop()->skip_strip_mined()->in(LoopNode::EntryControl));
907 _igvn.replace_input_of(n, MemNode::Memory, mem->in(LoopNode::EntryControl));
908 // Disconnect the phi now. An empty phi can confuse other
909 // optimizations in this pass of loop opts.
910 _igvn.replace_node(mem, mem->in(LoopNode::EntryControl));
911 n_loop->_body.yank(mem);
912
913 set_ctrl_and_loop(n, n->in(0));
914
915 return n;
916 }
917 }
918 }
919 return NULL;
920 }
921
922 // Try moving a store out of a loop, right after the loop
923 void PhaseIdealLoop::try_move_store_after_loop(Node* n) {
924 if (n->is_Store() && n->in(0) != NULL) {
925 Node *n_ctrl = get_ctrl(n);
926 IdealLoopTree *n_loop = get_loop(n_ctrl);
927 // Store must be in a loop
928 if (n_loop != _ltree_root && !n_loop->_irreducible) {
929 Node* address = n->in(MemNode::Address);
930 Node* value = n->in(MemNode::ValueIn);
931 IdealLoopTree* address_loop = get_loop(get_ctrl(address));
932 // address must be loop invariant
933 if (!n_loop->is_member(address_loop)) {
934 // Store must be last on this memory slice in the loop and
935 // nothing in the loop must observe it
936 Node* phi = NULL;
937 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
938 Node* u = n->fast_out(i);
939 if (has_ctrl(u)) { // control use?
940 IdealLoopTree *u_loop = get_loop(get_ctrl(u));
941 if (!n_loop->is_member(u_loop)) {
942 continue;
943 }
944 if (u->is_Phi() && u->in(0) == n_loop->_head) {
945 assert(_igvn.type(u) == Type::MEMORY, "bad phi");
946 // multiple phis on the same slice are possible
947 if (phi != NULL) {
948 return;
949 }
950 phi = u;
951 continue;
952 }
953 }
954 return;
955 }
956 if (phi != NULL) {
957 // Nothing in the loop before the store (next iteration)
958 // must observe the stored value
959 bool mem_ok = true;
960 {
961 ResourceMark rm;
962 Unique_Node_List wq;
963 wq.push(phi);
964 for (uint next = 0; next < wq.size() && mem_ok; ++next) {
965 Node *m = wq.at(next);
966 for (DUIterator_Fast imax, i = m->fast_outs(imax); i < imax && mem_ok; i++) {
967 Node* u = m->fast_out(i);
968 if (u->is_Store() || u->is_Phi()) {
969 if (u != n) {
970 wq.push(u);
971 mem_ok = (wq.size() <= 10);
972 }
973 } else {
974 mem_ok = false;
975 break;
976 }
977 }
978 }
979 }
980 if (mem_ok) {
981 // Move the store out of the loop if the LCA of all
982 // users (except for the phi) is outside the loop.
983 Node* hook = new Node(1);
984 hook->init_req(0, n_ctrl); // Add an input to prevent hook from being dead
985 _igvn.rehash_node_delayed(phi);
986 int count = phi->replace_edge(n, hook, &_igvn);
987 assert(count > 0, "inconsistent phi");
988
989 // Compute latest point this store can go
990 Node* lca = get_late_ctrl(n, get_ctrl(n));
991 if (lca->is_OuterStripMinedLoop()) {
992 lca = lca->in(LoopNode::EntryControl);
993 }
994 if (n_loop->is_member(get_loop(lca))) {
995 // LCA is in the loop - bail out
996 _igvn.replace_node(hook, n);
997 return;
998 }
999 #ifdef ASSERT
1000 if (n_loop->_head->is_Loop() && n_loop->_head->as_Loop()->is_strip_mined()) {
1001 assert(n_loop->_head->Opcode() == Op_CountedLoop, "outer loop is a strip mined");
1002 n_loop->_head->as_Loop()->verify_strip_mined(1);
1003 Node* outer = n_loop->_head->as_CountedLoop()->outer_loop();
1004 IdealLoopTree* outer_loop = get_loop(outer);
1005 assert(n_loop->_parent == outer_loop, "broken loop tree");
1006 assert(get_loop(lca) == outer_loop, "safepoint in outer loop consume all memory state");
1007 }
1008 #endif
1009 lca = place_outside_loop(lca, n_loop);
1010 assert(!n_loop->is_member(get_loop(lca)), "control must not be back in the loop");
1011 assert(get_loop(lca)->_nest < n_loop->_nest || lca->in(0)->is_NeverBranch(), "must not be moved into inner loop");
1012
1013 // Move store out of the loop
1014 _igvn.replace_node(hook, n->in(MemNode::Memory));
1015 _igvn.replace_input_of(n, 0, lca);
1016 set_ctrl_and_loop(n, lca);
1017
1018 // Disconnect the phi now. An empty phi can confuse other
1019 // optimizations in this pass of loop opts..
1020 if (phi->in(LoopNode::LoopBackControl) == phi) {
1021 _igvn.replace_node(phi, phi->in(LoopNode::EntryControl));
1022 n_loop->_body.yank(phi);
1023 }
1024 }
1025 }
1026 }
1027 }
1028 }
1029 }
1030
1031 // If UseArrayMarkWordCheck is enabled, we can't use immutable memory for the flat array check
1032 // because we are loading the mark word which is mutable. Although the bits we are interested in
1033 // are immutable (we check for markWord::unlocked_value), we need to use raw memory to not break
1034 // anti dependency analysis. Below code will attempt to still move flat array checks out of loops,
1035 // mainly to enable loop unswitching.
1036 void PhaseIdealLoop::move_flat_array_check_out_of_loop(Node* n) {
1037 // Skip checks for more than one array
1038 if (n->req() > 3) {
1039 return;
1040 }
1041 Node* mem = n->in(FlatArrayCheckNode::Memory);
1042 Node* array = n->in(FlatArrayCheckNode::ArrayOrKlass)->uncast();
1043 IdealLoopTree* check_loop = get_loop(get_ctrl(n));
1044 IdealLoopTree* ary_loop = get_loop(get_ctrl(array));
1045
1046 // Check if array is loop invariant
1047 if (!check_loop->is_member(ary_loop)) {
1048 // Walk up memory graph from the check until we leave the loop
1049 VectorSet wq;
1050 wq.set(mem->_idx);
1051 while (check_loop->is_member(get_loop(ctrl_or_self(mem)))) {
1052 if (mem->is_Phi()) {
1053 mem = mem->in(1);
1054 } else if (mem->is_MergeMem()) {
1055 mem = mem->as_MergeMem()->memory_at(Compile::AliasIdxRaw);
1056 } else if (mem->is_Proj()) {
1057 mem = mem->in(0);
1058 } else if (mem->is_MemBar() || mem->is_SafePoint()) {
1059 mem = mem->in(TypeFunc::Memory);
1060 } else if (mem->is_Store() || mem->is_LoadStore() || mem->is_ClearArray()) {
1061 mem = mem->in(MemNode::Memory);
1062 } else {
1063 #ifdef ASSERT
1064 mem->dump();
1065 #endif
1066 ShouldNotReachHere();
1067 }
1068 if (wq.test_set(mem->_idx)) {
1069 return;
1070 }
1071 }
1072 // Replace memory input and re-compute ctrl to move the check out of the loop
1073 _igvn.replace_input_of(n, 1, mem);
1074 set_ctrl_and_loop(n, get_early_ctrl(n));
1075 Node* bol = n->unique_out();
1076 set_ctrl_and_loop(bol, get_early_ctrl(bol));
1077 }
1078 }
1079
1080 //------------------------------split_if_with_blocks_pre-----------------------
1081 // Do the real work in a non-recursive function. Data nodes want to be
1082 // cloned in the pre-order so they can feed each other nicely.
1083 Node *PhaseIdealLoop::split_if_with_blocks_pre( Node *n ) {
1084 // Cloning these guys is unlikely to win
1085 int n_op = n->Opcode();
1086 if (n_op == Op_MergeMem) {
1087 return n;
1088 }
1089 if (n->is_Proj()) {
1090 return n;
1091 }
1092
1093 if (UseArrayMarkWordCheck && n->isa_FlatArrayCheck()) {
1094 move_flat_array_check_out_of_loop(n);
1095 return n;
1096 }
1097
1098 // Do not clone-up CmpFXXX variations, as these are always
1099 // followed by a CmpI
1100 if (n->is_Cmp()) {
1101 return n;
1102 }
1103 // Attempt to use a conditional move instead of a phi/branch
1104 if (ConditionalMoveLimit > 0 && n_op == Op_Region) {
1105 Node *cmov = conditional_move( n );
1106 if (cmov) {
1107 return cmov;
1108 }
1109 }
1110 if (n->is_CFG() || n->is_LoadStore()) {
1111 return n;
1112 }
1113 if (n->is_Opaque1()) { // Opaque nodes cannot be mod'd
1114 if (!C->major_progress()) { // If chance of no more loop opts...
1115 _igvn._worklist.push(n); // maybe we'll remove them
1116 }
1117 return n;
1118 }
1119
1120 if (n->is_Con()) {
1121 return n; // No cloning for Con nodes
1122 }
1123
1124 Node *n_ctrl = get_ctrl(n);
1125 if (!n_ctrl) {
1126 return n; // Dead node
1127 }
1128
1129 Node* res = try_move_store_before_loop(n, n_ctrl);
1130 if (res != NULL) {
1131 return n;
1132 }
1133
1134 // Attempt to remix address expressions for loop invariants
1135 Node *m = remix_address_expressions( n );
1136 if( m ) return m;
1137
1138 if (n_op == Op_AddI) {
1139 Node *nn = convert_add_to_muladd( n );
1140 if ( nn ) return nn;
1141 }
1142
1143 if (n->is_ConstraintCast()) {
1144 Node* dom_cast = n->as_ConstraintCast()->dominating_cast(&_igvn, this);
1145 // ConstraintCastNode::dominating_cast() uses node control input to determine domination.
1146 // Node control inputs don't necessarily agree with loop control info (due to
1147 // transformations happened in between), thus additional dominance check is needed
1148 // to keep loop info valid.
1149 if (dom_cast != NULL && is_dominator(get_ctrl(dom_cast), get_ctrl(n))) {
1150 _igvn.replace_node(n, dom_cast);
1151 return dom_cast;
1152 }
1153 }
1154
1155 // Determine if the Node has inputs from some local Phi.
1156 // Returns the block to clone thru.
1157 Node *n_blk = has_local_phi_input( n );
1158 if( !n_blk ) return n;
1159
1160 // Do not clone the trip counter through on a CountedLoop
1161 // (messes up the canonical shape).
1162 if (((n_blk->is_CountedLoop() || (n_blk->is_Loop() && n_blk->as_Loop()->is_loop_nest_inner_loop())) && n->Opcode() == Op_AddI) ||
1163 (n_blk->is_LongCountedLoop() && n->Opcode() == Op_AddL)) {
1164 return n;
1165 }
1166 // Pushing a shift through the iv Phi can get in the way of addressing optimizations or range check elimination
1167 if (n_blk->is_BaseCountedLoop() && n->Opcode() == Op_LShift(n_blk->as_BaseCountedLoop()->bt()) &&
1168 n->in(1) == n_blk->as_BaseCountedLoop()->phi()) {
1169 return n;
1170 }
1171
1172 // Check for having no control input; not pinned. Allow
1173 // dominating control.
1174 if (n->in(0)) {
1175 Node *dom = idom(n_blk);
1176 if (dom_lca(n->in(0), dom) != n->in(0)) {
1177 return n;
1178 }
1179 }
1180 // Policy: when is it profitable. You must get more wins than
1181 // policy before it is considered profitable. Policy is usually 0,
1182 // so 1 win is considered profitable. Big merges will require big
1183 // cloning, so get a larger policy.
1184 int policy = n_blk->req() >> 2;
1185
1186 // If the loop is a candidate for range check elimination,
1187 // delay splitting through it's phi until a later loop optimization
1188 if (n_blk->is_BaseCountedLoop()) {
1189 IdealLoopTree *lp = get_loop(n_blk);
1190 if (lp && lp->_rce_candidate) {
1191 return n;
1192 }
1193 }
1194
1195 if (must_throttle_split_if()) return n;
1196
1197 // Split 'n' through the merge point if it is profitable
1198 Node *phi = split_thru_phi( n, n_blk, policy );
1199 if (!phi) return n;
1200
1201 // Found a Phi to split thru!
1202 // Replace 'n' with the new phi
1203 _igvn.replace_node( n, phi );
1204 // Moved a load around the loop, 'en-registering' something.
1205 if (n_blk->is_Loop() && n->is_Load() &&
1206 !phi->in(LoopNode::LoopBackControl)->is_Load())
1207 C->set_major_progress();
1208
1209 return phi;
1210 }
1211
1212 static bool merge_point_too_heavy(Compile* C, Node* region) {
1213 // Bail out if the region and its phis have too many users.
1214 int weight = 0;
1215 for (DUIterator_Fast imax, i = region->fast_outs(imax); i < imax; i++) {
1216 weight += region->fast_out(i)->outcnt();
1217 }
1218 int nodes_left = C->max_node_limit() - C->live_nodes();
1219 if (weight * 8 > nodes_left) {
1220 if (PrintOpto) {
1221 tty->print_cr("*** Split-if bails out: %d nodes, region weight %d", C->unique(), weight);
1222 }
1223 return true;
1224 } else {
1225 return false;
1226 }
1227 }
1228
1229 static bool merge_point_safe(Node* region) {
1230 // 4799512: Stop split_if_with_blocks from splitting a block with a ConvI2LNode
1231 // having a PhiNode input. This sidesteps the dangerous case where the split
1232 // ConvI2LNode may become TOP if the input Value() does not
1233 // overlap the ConvI2L range, leaving a node which may not dominate its
1234 // uses.
1235 // A better fix for this problem can be found in the BugTraq entry, but
1236 // expediency for Mantis demands this hack.
1237 #ifdef _LP64
1238 for (DUIterator_Fast imax, i = region->fast_outs(imax); i < imax; i++) {
1239 Node* n = region->fast_out(i);
1240 if (n->is_Phi()) {
1241 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
1242 Node* m = n->fast_out(j);
1243 if (m->Opcode() == Op_ConvI2L)
1244 return false;
1245 if (m->is_CastII()) {
1246 return false;
1247 }
1248 }
1249 }
1250 }
1251 #endif
1252 return true;
1253 }
1254
1255
1256 //------------------------------place_outside_loop---------------------------------
1257 // Place some computation outside of this loop on the path to the use passed as argument
1258 Node* PhaseIdealLoop::place_outside_loop(Node* useblock, IdealLoopTree* loop) const {
1259 Node* head = loop->_head;
1260 assert(!loop->is_member(get_loop(useblock)), "must be outside loop");
1261 if (head->is_Loop() && head->as_Loop()->is_strip_mined()) {
1262 loop = loop->_parent;
1263 assert(loop->_head->is_OuterStripMinedLoop(), "malformed strip mined loop");
1264 }
1265
1266 // Pick control right outside the loop
1267 for (;;) {
1268 Node* dom = idom(useblock);
1269 if (loop->is_member(get_loop(dom)) ||
1270 // NeverBranch nodes are not assigned to the loop when constructed
1271 (dom->is_NeverBranch() && loop->is_member(get_loop(dom->in(0))))) {
1272 break;
1273 }
1274 useblock = dom;
1275 }
1276 assert(find_non_split_ctrl(useblock) == useblock, "should be non split control");
1277 return useblock;
1278 }
1279
1280
1281 bool PhaseIdealLoop::identical_backtoback_ifs(Node *n) {
1282 if (!n->is_If() || n->is_BaseCountedLoopEnd()) {
1283 return false;
1284 }
1285 if (!n->in(0)->is_Region()) {
1286 return false;
1287 }
1288
1289 Node* region = n->in(0);
1290 Node* dom = idom(region);
1291 if (!dom->is_If() || dom->in(1) != n->in(1)) {
1292 return false;
1293 }
1294 IfNode* dom_if = dom->as_If();
1295 Node* proj_true = dom_if->proj_out(1);
1296 Node* proj_false = dom_if->proj_out(0);
1297
1298 for (uint i = 1; i < region->req(); i++) {
1299 if (is_dominator(proj_true, region->in(i))) {
1300 continue;
1301 }
1302 if (is_dominator(proj_false, region->in(i))) {
1303 continue;
1304 }
1305 return false;
1306 }
1307
1308 return true;
1309 }
1310
1311
1312 bool PhaseIdealLoop::can_split_if(Node* n_ctrl) {
1313 if (must_throttle_split_if()) {
1314 return false;
1315 }
1316
1317 // Do not do 'split-if' if irreducible loops are present.
1318 if (_has_irreducible_loops) {
1319 return false;
1320 }
1321
1322 if (merge_point_too_heavy(C, n_ctrl)) {
1323 return false;
1324 }
1325
1326 // Do not do 'split-if' if some paths are dead. First do dead code
1327 // elimination and then see if its still profitable.
1328 for (uint i = 1; i < n_ctrl->req(); i++) {
1329 if (n_ctrl->in(i) == C->top()) {
1330 return false;
1331 }
1332 }
1333
1334 // If trying to do a 'Split-If' at the loop head, it is only
1335 // profitable if the cmp folds up on BOTH paths. Otherwise we
1336 // risk peeling a loop forever.
1337
1338 // CNC - Disabled for now. Requires careful handling of loop
1339 // body selection for the cloned code. Also, make sure we check
1340 // for any input path not being in the same loop as n_ctrl. For
1341 // irreducible loops we cannot check for 'n_ctrl->is_Loop()'
1342 // because the alternative loop entry points won't be converted
1343 // into LoopNodes.
1344 IdealLoopTree *n_loop = get_loop(n_ctrl);
1345 for (uint j = 1; j < n_ctrl->req(); j++) {
1346 if (get_loop(n_ctrl->in(j)) != n_loop) {
1347 return false;
1348 }
1349 }
1350
1351 // Check for safety of the merge point.
1352 if (!merge_point_safe(n_ctrl)) {
1353 return false;
1354 }
1355
1356 return true;
1357 }
1358
1359 // Detect if the node is the inner strip-mined loop
1360 // Return: NULL if it's not the case, or the exit of outer strip-mined loop
1361 static Node* is_inner_of_stripmined_loop(const Node* out) {
1362 Node* out_le = NULL;
1363
1364 if (out->is_CountedLoopEnd()) {
1365 const CountedLoopNode* loop = out->as_CountedLoopEnd()->loopnode();
1366
1367 if (loop != NULL && loop->is_strip_mined()) {
1368 out_le = loop->in(LoopNode::EntryControl)->as_OuterStripMinedLoop()->outer_loop_exit();
1369 }
1370 }
1371
1372 return out_le;
1373 }
1374
1375 bool PhaseIdealLoop::flatten_array_element_type_check(Node *n) {
1376 // If the CmpP is a subtype check for a value that has just been
1377 // loaded from an array, the subtype check guarantees the value
1378 // can't be stored in a flattened array and the load of the value
1379 // happens with a flattened array check then: push the type check
1380 // through the phi of the flattened array check. This needs special
1381 // logic because the subtype check's input is not a phi but a
1382 // LoadKlass that must first be cloned through the phi.
1383 if (n->Opcode() != Op_CmpP) {
1384 return false;
1385 }
1386
1387 Node* klassptr = n->in(1);
1388 Node* klasscon = n->in(2);
1389
1390 if (klassptr->is_DecodeNarrowPtr()) {
1391 klassptr = klassptr->in(1);
1392 }
1393
1394 if (klassptr->Opcode() != Op_LoadKlass && klassptr->Opcode() != Op_LoadNKlass) {
1395 return false;
1396 }
1397
1398 if (!klasscon->is_Con()) {
1399 return false;
1400 }
1401
1402 Node* addr = klassptr->in(MemNode::Address);
1403
1404 if (!addr->is_AddP()) {
1405 return false;
1406 }
1407
1408 intptr_t offset;
1409 Node* obj = AddPNode::Ideal_base_and_offset(addr, &_igvn, offset);
1410
1411 if (obj == NULL) {
1412 return false;
1413 }
1414
1415 assert(obj != NULL && addr->in(AddPNode::Base) == addr->in(AddPNode::Address), "malformed AddP?");
1416 if (obj->Opcode() == Op_CastPP) {
1417 obj = obj->in(1);
1418 }
1419
1420 if (!obj->is_Phi()) {
1421 return false;
1422 }
1423
1424 Node* region = obj->in(0);
1425
1426 Node* phi = PhiNode::make_blank(region, n->in(1));
1427 for (uint i = 1; i < region->req(); i++) {
1428 Node* in = obj->in(i);
1429 Node* ctrl = region->in(i);
1430 if (addr->in(AddPNode::Base) != obj) {
1431 Node* cast = addr->in(AddPNode::Base);
1432 assert(cast->Opcode() == Op_CastPP && cast->in(0) != NULL, "inconsistent subgraph");
1433 Node* cast_clone = cast->clone();
1434 cast_clone->set_req(0, ctrl);
1435 cast_clone->set_req(1, in);
1436 register_new_node(cast_clone, ctrl);
1437 const Type* tcast = cast_clone->Value(&_igvn);
1438 _igvn.set_type(cast_clone, tcast);
1439 cast_clone->as_Type()->set_type(tcast);
1440 in = cast_clone;
1441 }
1442 Node* addr_clone = addr->clone();
1443 addr_clone->set_req(AddPNode::Base, in);
1444 addr_clone->set_req(AddPNode::Address, in);
1445 register_new_node(addr_clone, ctrl);
1446 _igvn.set_type(addr_clone, addr_clone->Value(&_igvn));
1447 Node* klassptr_clone = klassptr->clone();
1448 klassptr_clone->set_req(2, addr_clone);
1449 register_new_node(klassptr_clone, ctrl);
1450 _igvn.set_type(klassptr_clone, klassptr_clone->Value(&_igvn));
1451 if (klassptr != n->in(1)) {
1452 Node* decode = n->in(1);
1453 assert(decode->is_DecodeNarrowPtr(), "inconsistent subgraph");
1454 Node* decode_clone = decode->clone();
1455 decode_clone->set_req(1, klassptr_clone);
1456 register_new_node(decode_clone, ctrl);
1457 _igvn.set_type(decode_clone, decode_clone->Value(&_igvn));
1458 klassptr_clone = decode_clone;
1459 }
1460 phi->set_req(i, klassptr_clone);
1461 }
1462 register_new_node(phi, region);
1463 Node* orig = n->in(1);
1464 _igvn.replace_input_of(n, 1, phi);
1465 split_if_with_blocks_post(n);
1466 if (n->outcnt() != 0) {
1467 _igvn.replace_input_of(n, 1, orig);
1468 _igvn.remove_dead_node(phi);
1469 }
1470 return true;
1471 }
1472
1473 //------------------------------split_if_with_blocks_post----------------------
1474 // Do the real work in a non-recursive function. CFG hackery wants to be
1475 // in the post-order, so it can dirty the I-DOM info and not use the dirtied
1476 // info.
1477 void PhaseIdealLoop::split_if_with_blocks_post(Node *n) {
1478
1479 if (flatten_array_element_type_check(n)) {
1480 return;
1481 }
1482
1483 // Cloning Cmp through Phi's involves the split-if transform.
1484 // FastLock is not used by an If
1485 if (n->is_Cmp() && !n->is_FastLock()) {
1486 Node *n_ctrl = get_ctrl(n);
1487 // Determine if the Node has inputs from some local Phi.
1488 // Returns the block to clone thru.
1489 Node *n_blk = has_local_phi_input(n);
1490 if (n_blk != n_ctrl) {
1491 return;
1492 }
1493
1494 if (!can_split_if(n_ctrl)) {
1495 return;
1496 }
1497
1498 if (n->outcnt() != 1) {
1499 return; // Multiple bool's from 1 compare?
1500 }
1501 Node *bol = n->unique_out();
1502 assert(bol->is_Bool(), "expect a bool here");
1503 if (bol->outcnt() != 1) {
1504 return;// Multiple branches from 1 compare?
1505 }
1506 Node *iff = bol->unique_out();
1507
1508 // Check some safety conditions
1509 if (iff->is_If()) { // Classic split-if?
1510 if (iff->in(0) != n_ctrl) {
1511 return; // Compare must be in same blk as if
1512 }
1513 } else if (iff->is_CMove()) { // Trying to split-up a CMOVE
1514 // Can't split CMove with different control edge.
1515 if (iff->in(0) != NULL && iff->in(0) != n_ctrl ) {
1516 return;
1517 }
1518 if (get_ctrl(iff->in(2)) == n_ctrl ||
1519 get_ctrl(iff->in(3)) == n_ctrl) {
1520 return; // Inputs not yet split-up
1521 }
1522 if (get_loop(n_ctrl) != get_loop(get_ctrl(iff))) {
1523 return; // Loop-invar test gates loop-varying CMOVE
1524 }
1525 } else {
1526 return; // some other kind of node, such as an Allocate
1527 }
1528
1529 // When is split-if profitable? Every 'win' on means some control flow
1530 // goes dead, so it's almost always a win.
1531 int policy = 0;
1532 // Split compare 'n' through the merge point if it is profitable
1533 Node *phi = split_thru_phi( n, n_ctrl, policy);
1534 if (!phi) {
1535 return;
1536 }
1537
1538 // Found a Phi to split thru!
1539 // Replace 'n' with the new phi
1540 _igvn.replace_node(n, phi);
1541
1542 // Now split the bool up thru the phi
1543 Node *bolphi = split_thru_phi(bol, n_ctrl, -1);
1544 guarantee(bolphi != NULL, "null boolean phi node");
1545
1546 _igvn.replace_node(bol, bolphi);
1547 assert(iff->in(1) == bolphi, "");
1548
1549 if (bolphi->Value(&_igvn)->singleton()) {
1550 return;
1551 }
1552
1553 // Conditional-move? Must split up now
1554 if (!iff->is_If()) {
1555 Node *cmovphi = split_thru_phi(iff, n_ctrl, -1);
1556 _igvn.replace_node(iff, cmovphi);
1557 return;
1558 }
1559
1560 // Now split the IF
1561 do_split_if(iff);
1562 return;
1563 }
1564
1565 // Two identical ifs back to back can be merged
1566 if (try_merge_identical_ifs(n)) {
1567 return;
1568 }
1569
1570 // Check for an IF ready to split; one that has its
1571 // condition codes input coming from a Phi at the block start.
1572 int n_op = n->Opcode();
1573
1574 // Check for an IF being dominated by another IF same test
1575 if (n_op == Op_If ||
1576 n_op == Op_RangeCheck) {
1577 Node *bol = n->in(1);
1578 uint max = bol->outcnt();
1579 // Check for same test used more than once?
1580 if (max > 1 && bol->is_Bool()) {
1581 // Search up IDOMs to see if this IF is dominated.
1582 Node *cutoff = get_ctrl(bol);
1583
1584 // Now search up IDOMs till cutoff, looking for a dominating test
1585 Node *prevdom = n;
1586 Node *dom = idom(prevdom);
1587 while (dom != cutoff) {
1588 if (dom->req() > 1 && dom->in(1) == bol && prevdom->in(0) == dom &&
1589 safe_for_if_replacement(dom)) {
1590 // It's invalid to move control dependent data nodes in the inner
1591 // strip-mined loop, because:
1592 // 1) break validation of LoopNode::verify_strip_mined()
1593 // 2) move code with side-effect in strip-mined loop
1594 // Move to the exit of outer strip-mined loop in that case.
1595 Node* out_le = is_inner_of_stripmined_loop(dom);
1596 if (out_le != NULL) {
1597 prevdom = out_le;
1598 }
1599 // Replace the dominated test with an obvious true or false.
1600 // Place it on the IGVN worklist for later cleanup.
1601 C->set_major_progress();
1602 dominated_by(prevdom->as_IfProj(), n->as_If(), false, true);
1603 #ifndef PRODUCT
1604 if( VerifyLoopOptimizations ) verify();
1605 #endif
1606 return;
1607 }
1608 prevdom = dom;
1609 dom = idom(prevdom);
1610 }
1611 }
1612 }
1613
1614 try_sink_out_of_loop(n);
1615
1616 try_move_store_after_loop(n);
1617
1618 // Remove multiple allocations of the same inline type
1619 if (n->is_InlineType()) {
1620 n->as_InlineType()->remove_redundant_allocations(this);
1621 }
1622 }
1623
1624 // Transform:
1625 //
1626 // if (some_condition) {
1627 // // body 1
1628 // } else {
1629 // // body 2
1630 // }
1631 // if (some_condition) {
1632 // // body 3
1633 // } else {
1634 // // body 4
1635 // }
1636 //
1637 // into:
1638 //
1639 //
1640 // if (some_condition) {
1641 // // body 1
1642 // // body 3
1643 // } else {
1644 // // body 2
1645 // // body 4
1646 // }
1647 bool PhaseIdealLoop::try_merge_identical_ifs(Node* n) {
1648 if (identical_backtoback_ifs(n) && can_split_if(n->in(0))) {
1649 Node *n_ctrl = n->in(0);
1650 IfNode* dom_if = idom(n_ctrl)->as_If();
1651 ProjNode* dom_proj_true = dom_if->proj_out(1);
1652 ProjNode* dom_proj_false = dom_if->proj_out(0);
1653
1654 // Now split the IF
1655 RegionNode* new_false_region;
1656 RegionNode* new_true_region;
1657 do_split_if(n, &new_false_region, &new_true_region);
1658 assert(new_false_region->req() == new_true_region->req(), "");
1659 #ifdef ASSERT
1660 for (uint i = 1; i < new_false_region->req(); ++i) {
1661 assert(new_false_region->in(i)->in(0) == new_true_region->in(i)->in(0), "unexpected shape following split if");
1662 assert(i == new_false_region->req() - 1 || new_false_region->in(i)->in(0)->in(1) == new_false_region->in(i + 1)->in(0)->in(1), "unexpected shape following split if");
1663 }
1664 #endif
1665 assert(new_false_region->in(1)->in(0)->in(1) == dom_if->in(1), "dominating if and dominated if after split must share test");
1666
1667 // We now have:
1668 // if (some_condition) {
1669 // // body 1
1670 // if (some_condition) {
1671 // body3: // new_true_region
1672 // // body3
1673 // } else {
1674 // goto body4;
1675 // }
1676 // } else {
1677 // // body 2
1678 // if (some_condition) {
1679 // goto body3;
1680 // } else {
1681 // body4: // new_false_region
1682 // // body4;
1683 // }
1684 // }
1685 //
1686
1687 // clone pinned nodes thru the resulting regions
1688 push_pinned_nodes_thru_region(dom_if, new_true_region);
1689 push_pinned_nodes_thru_region(dom_if, new_false_region);
1690
1691 // Optimize out the cloned ifs. Because pinned nodes were cloned, this also allows a CastPP that would be dependent
1692 // on a projection of n to have the dom_if as a control dependency. We don't want the CastPP to end up with an
1693 // unrelated control dependency.
1694 for (uint i = 1; i < new_false_region->req(); i++) {
1695 if (is_dominator(dom_proj_true, new_false_region->in(i))) {
1696 dominated_by(dom_proj_true->as_IfProj(), new_false_region->in(i)->in(0)->as_If(), false, false);
1697 } else {
1698 assert(is_dominator(dom_proj_false, new_false_region->in(i)), "bad if");
1699 dominated_by(dom_proj_false->as_IfProj(), new_false_region->in(i)->in(0)->as_If(), false, false);
1700 }
1701 }
1702 return true;
1703 }
1704 return false;
1705 }
1706
1707 void PhaseIdealLoop::push_pinned_nodes_thru_region(IfNode* dom_if, Node* region) {
1708 for (DUIterator i = region->outs(); region->has_out(i); i++) {
1709 Node* u = region->out(i);
1710 if (!has_ctrl(u) || u->is_Phi() || !u->depends_only_on_test() || !_igvn.no_dependent_zero_check(u)) {
1711 continue;
1712 }
1713 assert(u->in(0) == region, "not a control dependent node?");
1714 uint j = 1;
1715 for (; j < u->req(); ++j) {
1716 Node* in = u->in(j);
1717 if (!is_dominator(ctrl_or_self(in), dom_if)) {
1718 break;
1719 }
1720 }
1721 if (j == u->req()) {
1722 Node *phi = PhiNode::make_blank(region, u);
1723 for (uint k = 1; k < region->req(); ++k) {
1724 Node* clone = u->clone();
1725 clone->set_req(0, region->in(k));
1726 register_new_node(clone, region->in(k));
1727 phi->init_req(k, clone);
1728 }
1729 register_new_node(phi, region);
1730 _igvn.replace_node(u, phi);
1731 --i;
1732 }
1733 }
1734 }
1735
1736 bool PhaseIdealLoop::safe_for_if_replacement(const Node* dom) const {
1737 if (!dom->is_CountedLoopEnd()) {
1738 return true;
1739 }
1740 CountedLoopEndNode* le = dom->as_CountedLoopEnd();
1741 CountedLoopNode* cl = le->loopnode();
1742 if (cl == NULL) {
1743 return true;
1744 }
1745 if (!cl->is_main_loop()) {
1746 return true;
1747 }
1748 if (cl->is_canonical_loop_entry() == NULL) {
1749 return true;
1750 }
1751 // Further unrolling is possible so loop exit condition might change
1752 return false;
1753 }
1754
1755 // See if a shared loop-varying computation has no loop-varying uses.
1756 // Happens if something is only used for JVM state in uncommon trap exits,
1757 // like various versions of induction variable+offset. Clone the
1758 // computation per usage to allow it to sink out of the loop.
1759 void PhaseIdealLoop::try_sink_out_of_loop(Node* n) {
1760 if (has_ctrl(n) &&
1761 !n->is_Phi() &&
1762 !n->is_Bool() &&
1763 !n->is_Proj() &&
1764 !n->is_MergeMem() &&
1765 !n->is_CMove() &&
1766 n->Opcode() != Op_Opaque4 &&
1767 !n->is_Type()) {
1768 Node *n_ctrl = get_ctrl(n);
1769 IdealLoopTree *n_loop = get_loop(n_ctrl);
1770
1771 if (n->in(0) != NULL) {
1772 IdealLoopTree* loop_ctrl = get_loop(n->in(0));
1773 if (n_loop != loop_ctrl && n_loop->is_member(loop_ctrl)) {
1774 // n has a control input inside a loop but get_ctrl() is member of an outer loop. This could happen, for example,
1775 // for Div nodes inside a loop (control input inside loop) without a use except for an UCT (outside the loop).
1776 // Rewire control of n to right outside of the loop, regardless if its input(s) are later sunk or not.
1777 _igvn.replace_input_of(n, 0, place_outside_loop(n_ctrl, loop_ctrl));
1778 }
1779 }
1780 if (n_loop != _ltree_root && n->outcnt() > 1) {
1781 // Compute early control: needed for anti-dependence analysis. It's also possible that as a result of
1782 // previous transformations in this loop opts round, the node can be hoisted now: early control will tell us.
1783 Node* early_ctrl = compute_early_ctrl(n, n_ctrl);
1784 if (n_loop->is_member(get_loop(early_ctrl)) && // check that this one can't be hoisted now
1785 ctrl_of_all_uses_out_of_loop(n, early_ctrl, n_loop)) { // All uses in outer loops!
1786 assert(!n->is_Store() && !n->is_LoadStore(), "no node with a side effect");
1787 Node* outer_loop_clone = NULL;
1788 for (DUIterator_Last jmin, j = n->last_outs(jmin); j >= jmin;) {
1789 Node* u = n->last_out(j); // Clone private computation per use
1790 _igvn.rehash_node_delayed(u);
1791 Node* x = n->clone(); // Clone computation
1792 Node* x_ctrl = NULL;
1793 if (u->is_Phi()) {
1794 // Replace all uses of normal nodes. Replace Phi uses
1795 // individually, so the separate Nodes can sink down
1796 // different paths.
1797 uint k = 1;
1798 while (u->in(k) != n) k++;
1799 u->set_req(k, x);
1800 // x goes next to Phi input path
1801 x_ctrl = u->in(0)->in(k);
1802 // Find control for 'x' next to use but not inside inner loops.
1803 x_ctrl = place_outside_loop(x_ctrl, n_loop);
1804 --j;
1805 } else { // Normal use
1806 if (has_ctrl(u)) {
1807 x_ctrl = get_ctrl(u);
1808 } else {
1809 x_ctrl = u->in(0);
1810 }
1811 // Find control for 'x' next to use but not inside inner loops.
1812 x_ctrl = place_outside_loop(x_ctrl, n_loop);
1813 // Replace all uses
1814 if (u->is_ConstraintCast() && u->bottom_type()->higher_equal(_igvn.type(n)) && u->in(0) == x_ctrl) {
1815 // If we're sinking a chain of data nodes, we might have inserted a cast to pin the use which is not necessary
1816 // anymore now that we're going to pin n as well
1817 _igvn.replace_node(u, x);
1818 --j;
1819 } else {
1820 int nb = u->replace_edge(n, x, &_igvn);
1821 j -= nb;
1822 }
1823 }
1824
1825 if (n->is_Load()) {
1826 // For loads, add a control edge to a CFG node outside of the loop
1827 // to force them to not combine and return back inside the loop
1828 // during GVN optimization (4641526).
1829 assert(x_ctrl == get_late_ctrl_with_anti_dep(x->as_Load(), early_ctrl, x_ctrl), "anti-dependences were already checked");
1830
1831 IdealLoopTree* x_loop = get_loop(x_ctrl);
1832 Node* x_head = x_loop->_head;
1833 if (x_head->is_Loop() && x_head->is_OuterStripMinedLoop()) {
1834 // Do not add duplicate LoadNodes to the outer strip mined loop
1835 if (outer_loop_clone != NULL) {
1836 _igvn.replace_node(x, outer_loop_clone);
1837 continue;
1838 }
1839 outer_loop_clone = x;
1840 }
1841 x->set_req(0, x_ctrl);
1842 } else if (n->in(0) != NULL){
1843 x->set_req(0, x_ctrl);
1844 }
1845 assert(dom_depth(n_ctrl) <= dom_depth(x_ctrl), "n is later than its clone");
1846 assert(!n_loop->is_member(get_loop(x_ctrl)), "should have moved out of loop");
1847 register_new_node(x, x_ctrl);
1848
1849 // Chain of AddP: (AddP base (AddP base )) must keep the same base after sinking so:
1850 // 1- We don't add a CastPP here when the first one is sunk so if the second one is not, their bases remain
1851 // the same.
1852 // (see 2- below)
1853 assert(!x->is_AddP() || !x->in(AddPNode::Address)->is_AddP() ||
1854 x->in(AddPNode::Address)->in(AddPNode::Base) == x->in(AddPNode::Base) ||
1855 !x->in(AddPNode::Address)->in(AddPNode::Base)->eqv_uncast(x->in(AddPNode::Base)), "unexpected AddP shape");
1856 if (x->in(0) == NULL && !x->is_DecodeNarrowPtr() &&
1857 !(x->is_AddP() && x->in(AddPNode::Address)->is_AddP() && x->in(AddPNode::Address)->in(AddPNode::Base) == x->in(AddPNode::Base))) {
1858 assert(!x->is_Load(), "load should be pinned");
1859 // Use a cast node to pin clone out of loop
1860 Node* cast = NULL;
1861 for (uint k = 0; k < x->req(); k++) {
1862 Node* in = x->in(k);
1863 if (in != NULL && n_loop->is_member(get_loop(get_ctrl(in)))) {
1864 const Type* in_t = _igvn.type(in);
1865 cast = ConstraintCastNode::make_cast_for_type(x_ctrl, in, in_t, ConstraintCastNode::UnconditionalDependency);
1866 }
1867 if (cast != NULL) {
1868 register_new_node(cast, x_ctrl);
1869 x->replace_edge(in, cast);
1870 // Chain of AddP:
1871 // 2- A CastPP of the base is only added now that both AddP nodes are sunk
1872 if (x->is_AddP() && k == AddPNode::Base) {
1873 for (DUIterator_Fast imax, i = x->fast_outs(imax); i < imax; i++) {
1874 Node* u = x->fast_out(i);
1875 if (u->is_AddP() && u->in(AddPNode::Base) == n->in(AddPNode::Base)) {
1876 _igvn.replace_input_of(u, AddPNode::Base, cast);
1877 assert(u->find_out_with(Op_AddP) == NULL, "more than 2 chained AddP nodes?");
1878 }
1879 }
1880 }
1881 break;
1882 }
1883 }
1884 assert(cast != NULL, "must have added a cast to pin the node");
1885 }
1886 }
1887 _igvn.remove_dead_node(n);
1888 }
1889 _dom_lca_tags_round = 0;
1890 }
1891 }
1892 }
1893
1894 // Compute the early control of a node by following its inputs until we reach
1895 // nodes that are pinned. Then compute the LCA of the control of all pinned nodes.
1896 Node* PhaseIdealLoop::compute_early_ctrl(Node* n, Node* n_ctrl) {
1897 Node* early_ctrl = NULL;
1898 ResourceMark rm;
1899 Unique_Node_List wq;
1900 wq.push(n);
1901 for (uint i = 0; i < wq.size(); i++) {
1902 Node* m = wq.at(i);
1903 Node* c = NULL;
1904 if (m->is_CFG()) {
1905 c = m;
1906 } else if (m->pinned()) {
1907 c = m->in(0);
1908 } else {
1909 for (uint j = 0; j < m->req(); j++) {
1910 Node* in = m->in(j);
1911 if (in != NULL) {
1912 wq.push(in);
1913 }
1914 }
1915 }
1916 if (c != NULL) {
1917 assert(is_dominator(c, n_ctrl), "control input must dominate current control");
1918 if (early_ctrl == NULL || is_dominator(early_ctrl, c)) {
1919 early_ctrl = c;
1920 }
1921 }
1922 }
1923 assert(is_dominator(early_ctrl, n_ctrl), "early control must dominate current control");
1924 return early_ctrl;
1925 }
1926
1927 bool PhaseIdealLoop::ctrl_of_all_uses_out_of_loop(const Node* n, Node* n_ctrl, IdealLoopTree* n_loop) {
1928 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
1929 Node* u = n->fast_out(i);
1930 if (u->is_Opaque1()) {
1931 return false; // Found loop limit, bugfix for 4677003
1932 }
1933 // We can't reuse tags in PhaseIdealLoop::dom_lca_for_get_late_ctrl_internal() so make sure calls to
1934 // get_late_ctrl_with_anti_dep() use their own tag
1935 _dom_lca_tags_round++;
1936 assert(_dom_lca_tags_round != 0, "shouldn't wrap around");
1937
1938 if (u->is_Phi()) {
1939 for (uint j = 1; j < u->req(); ++j) {
1940 if (u->in(j) == n && !ctrl_of_use_out_of_loop(n, n_ctrl, n_loop, u->in(0)->in(j))) {
1941 return false;
1942 }
1943 }
1944 } else {
1945 Node* ctrl = has_ctrl(u) ? get_ctrl(u) : u->in(0);
1946 if (!ctrl_of_use_out_of_loop(n, n_ctrl, n_loop, ctrl)) {
1947 return false;
1948 }
1949 }
1950 }
1951 return true;
1952 }
1953
1954 bool PhaseIdealLoop::ctrl_of_use_out_of_loop(const Node* n, Node* n_ctrl, IdealLoopTree* n_loop, Node* ctrl) {
1955 if (n->is_Load()) {
1956 ctrl = get_late_ctrl_with_anti_dep(n->as_Load(), n_ctrl, ctrl);
1957 }
1958 IdealLoopTree *u_loop = get_loop(ctrl);
1959 if (u_loop == n_loop) {
1960 return false; // Found loop-varying use
1961 }
1962 if (n_loop->is_member(u_loop)) {
1963 return false; // Found use in inner loop
1964 }
1965 return true;
1966 }
1967
1968 //------------------------------split_if_with_blocks---------------------------
1969 // Check for aggressive application of 'split-if' optimization,
1970 // using basic block level info.
1971 void PhaseIdealLoop::split_if_with_blocks(VectorSet &visited, Node_Stack &nstack) {
1972 Node* root = C->root();
1973 visited.set(root->_idx); // first, mark root as visited
1974 // Do pre-visit work for root
1975 Node* n = split_if_with_blocks_pre(root);
1976 uint cnt = n->outcnt();
1977 uint i = 0;
1978
1979 while (true) {
1980 // Visit all children
1981 if (i < cnt) {
1982 Node* use = n->raw_out(i);
1983 ++i;
1984 if (use->outcnt() != 0 && !visited.test_set(use->_idx)) {
1985 // Now do pre-visit work for this use
1986 use = split_if_with_blocks_pre(use);
1987 nstack.push(n, i); // Save parent and next use's index.
1988 n = use; // Process all children of current use.
1989 cnt = use->outcnt();
1990 i = 0;
1991 }
1992 }
1993 else {
1994 // All of n's children have been processed, complete post-processing.
1995 if (cnt != 0 && !n->is_Con()) {
1996 assert(has_node(n), "no dead nodes");
1997 split_if_with_blocks_post(n);
1998 }
1999 if (must_throttle_split_if()) {
2000 nstack.clear();
2001 }
2002 if (nstack.is_empty()) {
2003 // Finished all nodes on stack.
2004 break;
2005 }
2006 // Get saved parent node and next use's index. Visit the rest of uses.
2007 n = nstack.node();
2008 cnt = n->outcnt();
2009 i = nstack.index();
2010 nstack.pop();
2011 }
2012 }
2013 }
2014
2015
2016 //=============================================================================
2017 //
2018 // C L O N E A L O O P B O D Y
2019 //
2020
2021 //------------------------------clone_iff--------------------------------------
2022 // Passed in a Phi merging (recursively) some nearly equivalent Bool/Cmps.
2023 // "Nearly" because all Nodes have been cloned from the original in the loop,
2024 // but the fall-in edges to the Cmp are different. Clone bool/Cmp pairs
2025 // through the Phi recursively, and return a Bool.
2026 Node* PhaseIdealLoop::clone_iff(PhiNode* phi) {
2027
2028 // Convert this Phi into a Phi merging Bools
2029 uint i;
2030 for (i = 1; i < phi->req(); i++) {
2031 Node *b = phi->in(i);
2032 if (b->is_Phi()) {
2033 _igvn.replace_input_of(phi, i, clone_iff(b->as_Phi()));
2034 } else {
2035 assert(b->is_Bool() || b->Opcode() == Op_Opaque4, "");
2036 }
2037 }
2038
2039 Node* n = phi->in(1);
2040 Node* sample_opaque = NULL;
2041 Node *sample_bool = NULL;
2042 if (n->Opcode() == Op_Opaque4) {
2043 sample_opaque = n;
2044 sample_bool = n->in(1);
2045 assert(sample_bool->is_Bool(), "wrong type");
2046 } else {
2047 sample_bool = n;
2048 }
2049 Node* sample_cmp = sample_bool->in(1);
2050 const Type* t = Type::TOP;
2051 const TypePtr* at = NULL;
2052 if (sample_cmp->is_FlatArrayCheck()) {
2053 // Left input of a FlatArrayCheckNode is memory, set the (adr) type of the phi accordingly
2054 assert(sample_cmp->in(1)->bottom_type() == Type::MEMORY, "unexpected input type");
2055 t = Type::MEMORY;
2056 at = TypeRawPtr::BOTTOM;
2057 }
2058
2059 // Make Phis to merge the Cmp's inputs.
2060 PhiNode *phi1 = new PhiNode(phi->in(0), t, at);
2061 PhiNode *phi2 = new PhiNode(phi->in(0), Type::TOP);
2062 for (i = 1; i < phi->req(); i++) {
2063 Node *n1 = sample_opaque == NULL ? phi->in(i)->in(1)->in(1) : phi->in(i)->in(1)->in(1)->in(1);
2064 Node *n2 = sample_opaque == NULL ? phi->in(i)->in(1)->in(2) : phi->in(i)->in(1)->in(1)->in(2);
2065 phi1->set_req(i, n1);
2066 phi2->set_req(i, n2);
2067 phi1->set_type(phi1->type()->meet_speculative(n1->bottom_type()));
2068 phi2->set_type(phi2->type()->meet_speculative(n2->bottom_type()));
2069 }
2070 // See if these Phis have been made before.
2071 // Register with optimizer
2072 Node *hit1 = _igvn.hash_find_insert(phi1);
2073 if (hit1) { // Hit, toss just made Phi
2074 _igvn.remove_dead_node(phi1); // Remove new phi
2075 assert(hit1->is_Phi(), "" );
2076 phi1 = (PhiNode*)hit1; // Use existing phi
2077 } else { // Miss
2078 _igvn.register_new_node_with_optimizer(phi1);
2079 }
2080 Node *hit2 = _igvn.hash_find_insert(phi2);
2081 if (hit2) { // Hit, toss just made Phi
2082 _igvn.remove_dead_node(phi2); // Remove new phi
2083 assert(hit2->is_Phi(), "" );
2084 phi2 = (PhiNode*)hit2; // Use existing phi
2085 } else { // Miss
2086 _igvn.register_new_node_with_optimizer(phi2);
2087 }
2088 // Register Phis with loop/block info
2089 set_ctrl(phi1, phi->in(0));
2090 set_ctrl(phi2, phi->in(0));
2091 // Make a new Cmp
2092 Node *cmp = sample_cmp->clone();
2093 cmp->set_req(1, phi1);
2094 cmp->set_req(2, phi2);
2095 _igvn.register_new_node_with_optimizer(cmp);
2096 set_ctrl(cmp, phi->in(0));
2097
2098 // Make a new Bool
2099 Node *b = sample_bool->clone();
2100 b->set_req(1,cmp);
2101 _igvn.register_new_node_with_optimizer(b);
2102 set_ctrl(b, phi->in(0));
2103
2104 if (sample_opaque != NULL) {
2105 Node* opaque = sample_opaque->clone();
2106 opaque->set_req(1, b);
2107 _igvn.register_new_node_with_optimizer(opaque);
2108 set_ctrl(opaque, phi->in(0));
2109 return opaque;
2110 }
2111
2112 assert(b->is_Bool(), "");
2113 return b;
2114 }
2115
2116 //------------------------------clone_bool-------------------------------------
2117 // Passed in a Phi merging (recursively) some nearly equivalent Bool/Cmps.
2118 // "Nearly" because all Nodes have been cloned from the original in the loop,
2119 // but the fall-in edges to the Cmp are different. Clone bool/Cmp pairs
2120 // through the Phi recursively, and return a Bool.
2121 CmpNode*PhaseIdealLoop::clone_bool(PhiNode* phi) {
2122 uint i;
2123 // Convert this Phi into a Phi merging Bools
2124 for( i = 1; i < phi->req(); i++ ) {
2125 Node *b = phi->in(i);
2126 if( b->is_Phi() ) {
2127 _igvn.replace_input_of(phi, i, clone_bool(b->as_Phi()));
2128 } else {
2129 assert( b->is_Cmp() || b->is_top(), "inputs are all Cmp or TOP" );
2130 }
2131 }
2132
2133 Node *sample_cmp = phi->in(1);
2134
2135 // Make Phis to merge the Cmp's inputs.
2136 PhiNode *phi1 = new PhiNode( phi->in(0), Type::TOP );
2137 PhiNode *phi2 = new PhiNode( phi->in(0), Type::TOP );
2138 for( uint j = 1; j < phi->req(); j++ ) {
2139 Node *cmp_top = phi->in(j); // Inputs are all Cmp or TOP
2140 Node *n1, *n2;
2141 if( cmp_top->is_Cmp() ) {
2142 n1 = cmp_top->in(1);
2143 n2 = cmp_top->in(2);
2144 } else {
2145 n1 = n2 = cmp_top;
2146 }
2147 phi1->set_req( j, n1 );
2148 phi2->set_req( j, n2 );
2149 phi1->set_type(phi1->type()->meet_speculative(n1->bottom_type()));
2150 phi2->set_type(phi2->type()->meet_speculative(n2->bottom_type()));
2151 }
2152
2153 // See if these Phis have been made before.
2154 // Register with optimizer
2155 Node *hit1 = _igvn.hash_find_insert(phi1);
2156 if( hit1 ) { // Hit, toss just made Phi
2157 _igvn.remove_dead_node(phi1); // Remove new phi
2158 assert( hit1->is_Phi(), "" );
2159 phi1 = (PhiNode*)hit1; // Use existing phi
2160 } else { // Miss
2161 _igvn.register_new_node_with_optimizer(phi1);
2162 }
2163 Node *hit2 = _igvn.hash_find_insert(phi2);
2164 if( hit2 ) { // Hit, toss just made Phi
2165 _igvn.remove_dead_node(phi2); // Remove new phi
2166 assert( hit2->is_Phi(), "" );
2167 phi2 = (PhiNode*)hit2; // Use existing phi
2168 } else { // Miss
2169 _igvn.register_new_node_with_optimizer(phi2);
2170 }
2171 // Register Phis with loop/block info
2172 set_ctrl(phi1, phi->in(0));
2173 set_ctrl(phi2, phi->in(0));
2174 // Make a new Cmp
2175 Node *cmp = sample_cmp->clone();
2176 cmp->set_req( 1, phi1 );
2177 cmp->set_req( 2, phi2 );
2178 _igvn.register_new_node_with_optimizer(cmp);
2179 set_ctrl(cmp, phi->in(0));
2180
2181 assert( cmp->is_Cmp(), "" );
2182 return (CmpNode*)cmp;
2183 }
2184
2185 //------------------------------sink_use---------------------------------------
2186 // If 'use' was in the loop-exit block, it now needs to be sunk
2187 // below the post-loop merge point.
2188 void PhaseIdealLoop::sink_use( Node *use, Node *post_loop ) {
2189 if (!use->is_CFG() && get_ctrl(use) == post_loop->in(2)) {
2190 set_ctrl(use, post_loop);
2191 for (DUIterator j = use->outs(); use->has_out(j); j++)
2192 sink_use(use->out(j), post_loop);
2193 }
2194 }
2195
2196 void PhaseIdealLoop::clone_loop_handle_data_uses(Node* old, Node_List &old_new,
2197 IdealLoopTree* loop, IdealLoopTree* outer_loop,
2198 Node_List*& split_if_set, Node_List*& split_bool_set,
2199 Node_List*& split_cex_set, Node_List& worklist,
2200 uint new_counter, CloneLoopMode mode) {
2201 Node* nnn = old_new[old->_idx];
2202 // Copy uses to a worklist, so I can munge the def-use info
2203 // with impunity.
2204 for (DUIterator_Fast jmax, j = old->fast_outs(jmax); j < jmax; j++)
2205 worklist.push(old->fast_out(j));
2206
2207 while( worklist.size() ) {
2208 Node *use = worklist.pop();
2209 if (!has_node(use)) continue; // Ignore dead nodes
2210 if (use->in(0) == C->top()) continue;
2211 IdealLoopTree *use_loop = get_loop( has_ctrl(use) ? get_ctrl(use) : use );
2212 // Check for data-use outside of loop - at least one of OLD or USE
2213 // must not be a CFG node.
2214 #ifdef ASSERT
2215 if (loop->_head->as_Loop()->is_strip_mined() && outer_loop->is_member(use_loop) && !loop->is_member(use_loop) && old_new[use->_idx] == NULL) {
2216 Node* sfpt = loop->_head->as_CountedLoop()->outer_safepoint();
2217 assert(mode != IgnoreStripMined, "incorrect cloning mode");
2218 assert((mode == ControlAroundStripMined && use == sfpt) || !use->is_reachable_from_root(), "missed a node");
2219 }
2220 #endif
2221 if (!loop->is_member(use_loop) && !outer_loop->is_member(use_loop) && (!old->is_CFG() || !use->is_CFG())) {
2222
2223 // If the Data use is an IF, that means we have an IF outside of the
2224 // loop that is switching on a condition that is set inside of the
2225 // loop. Happens if people set a loop-exit flag; then test the flag
2226 // in the loop to break the loop, then test is again outside of the
2227 // loop to determine which way the loop exited.
2228 // Loop predicate If node connects to Bool node through Opaque1 node.
2229 //
2230 // If the use is an AllocateArray through its ValidLengthTest input,
2231 // make sure the Bool/Cmp input is cloned down to avoid a Phi between
2232 // the AllocateArray node and its ValidLengthTest input that could cause
2233 // split if to break.
2234 if (use->is_If() || use->is_CMove() || C->is_predicate_opaq(use) || use->Opcode() == Op_Opaque4 ||
2235 (use->Opcode() == Op_AllocateArray && use->in(AllocateNode::ValidLengthTest) == old)) {
2236 // Since this code is highly unlikely, we lazily build the worklist
2237 // of such Nodes to go split.
2238 if (!split_if_set) {
2239 split_if_set = new Node_List();
2240 }
2241 split_if_set->push(use);
2242 }
2243 if (use->is_Bool()) {
2244 if (!split_bool_set) {
2245 split_bool_set = new Node_List();
2246 }
2247 split_bool_set->push(use);
2248 }
2249 if (use->Opcode() == Op_CreateEx) {
2250 if (!split_cex_set) {
2251 split_cex_set = new Node_List();
2252 }
2253 split_cex_set->push(use);
2254 }
2255
2256
2257 // Get "block" use is in
2258 uint idx = 0;
2259 while( use->in(idx) != old ) idx++;
2260 Node *prev = use->is_CFG() ? use : get_ctrl(use);
2261 assert(!loop->is_member(get_loop(prev)) && !outer_loop->is_member(get_loop(prev)), "" );
2262 Node *cfg = prev->_idx >= new_counter
2263 ? prev->in(2)
2264 : idom(prev);
2265 if( use->is_Phi() ) // Phi use is in prior block
2266 cfg = prev->in(idx); // NOT in block of Phi itself
2267 if (cfg->is_top()) { // Use is dead?
2268 _igvn.replace_input_of(use, idx, C->top());
2269 continue;
2270 }
2271
2272 // If use is referenced through control edge... (idx == 0)
2273 if (mode == IgnoreStripMined && idx == 0) {
2274 LoopNode *head = loop->_head->as_Loop();
2275 if (head->is_strip_mined() && is_dominator(head->outer_loop_exit(), prev)) {
2276 // That node is outside the inner loop, leave it outside the
2277 // outer loop as well to not confuse verification code.
2278 assert(!loop->_parent->is_member(use_loop), "should be out of the outer loop");
2279 _igvn.replace_input_of(use, 0, head->outer_loop_exit());
2280 continue;
2281 }
2282 }
2283
2284 while(!outer_loop->is_member(get_loop(cfg))) {
2285 prev = cfg;
2286 cfg = cfg->_idx >= new_counter ? cfg->in(2) : idom(cfg);
2287 }
2288 // If the use occurs after merging several exits from the loop, then
2289 // old value must have dominated all those exits. Since the same old
2290 // value was used on all those exits we did not need a Phi at this
2291 // merge point. NOW we do need a Phi here. Each loop exit value
2292 // is now merged with the peeled body exit; each exit gets its own
2293 // private Phi and those Phis need to be merged here.
2294 Node *phi;
2295 if( prev->is_Region() ) {
2296 if( idx == 0 ) { // Updating control edge?
2297 phi = prev; // Just use existing control
2298 } else { // Else need a new Phi
2299 phi = PhiNode::make( prev, old );
2300 // Now recursively fix up the new uses of old!
2301 for( uint i = 1; i < prev->req(); i++ ) {
2302 worklist.push(phi); // Onto worklist once for each 'old' input
2303 }
2304 }
2305 } else {
2306 // Get new RegionNode merging old and new loop exits
2307 prev = old_new[prev->_idx];
2308 assert( prev, "just made this in step 7" );
2309 if( idx == 0) { // Updating control edge?
2310 phi = prev; // Just use existing control
2311 } else { // Else need a new Phi
2312 // Make a new Phi merging data values properly
2313 phi = PhiNode::make( prev, old );
2314 phi->set_req( 1, nnn );
2315 }
2316 }
2317 // If inserting a new Phi, check for prior hits
2318 if( idx != 0 ) {
2319 Node *hit = _igvn.hash_find_insert(phi);
2320 if( hit == NULL ) {
2321 _igvn.register_new_node_with_optimizer(phi); // Register new phi
2322 } else { // or
2323 // Remove the new phi from the graph and use the hit
2324 _igvn.remove_dead_node(phi);
2325 phi = hit; // Use existing phi
2326 }
2327 set_ctrl(phi, prev);
2328 }
2329 // Make 'use' use the Phi instead of the old loop body exit value
2330 _igvn.replace_input_of(use, idx, phi);
2331 if( use->_idx >= new_counter ) { // If updating new phis
2332 // Not needed for correctness, but prevents a weak assert
2333 // in AddPNode from tripping (when we end up with different
2334 // base & derived Phis that will become the same after
2335 // IGVN does CSE).
2336 Node *hit = _igvn.hash_find_insert(use);
2337 if( hit ) // Go ahead and re-hash for hits.
2338 _igvn.replace_node( use, hit );
2339 }
2340
2341 // If 'use' was in the loop-exit block, it now needs to be sunk
2342 // below the post-loop merge point.
2343 sink_use( use, prev );
2344 }
2345 }
2346 }
2347
2348 static void collect_nodes_in_outer_loop_not_reachable_from_sfpt(Node* n, const IdealLoopTree *loop, const IdealLoopTree* outer_loop,
2349 const Node_List &old_new, Unique_Node_List& wq, PhaseIdealLoop* phase,
2350 bool check_old_new) {
2351 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
2352 Node* u = n->fast_out(j);
2353 assert(check_old_new || old_new[u->_idx] == NULL, "shouldn't have been cloned");
2354 if (!u->is_CFG() && (!check_old_new || old_new[u->_idx] == NULL)) {
2355 Node* c = phase->get_ctrl(u);
2356 IdealLoopTree* u_loop = phase->get_loop(c);
2357 assert(!loop->is_member(u_loop) || !loop->_body.contains(u), "can be in outer loop or out of both loops only");
2358 if (!loop->is_member(u_loop)) {
2359 if (outer_loop->is_member(u_loop)) {
2360 wq.push(u);
2361 } else {
2362 // nodes pinned with control in the outer loop but not referenced from the safepoint must be moved out of
2363 // the outer loop too
2364 Node* u_c = u->in(0);
2365 if (u_c != NULL) {
2366 IdealLoopTree* u_c_loop = phase->get_loop(u_c);
2367 if (outer_loop->is_member(u_c_loop) && !loop->is_member(u_c_loop)) {
2368 wq.push(u);
2369 }
2370 }
2371 }
2372 }
2373 }
2374 }
2375 }
2376
2377 void PhaseIdealLoop::clone_outer_loop(LoopNode* head, CloneLoopMode mode, IdealLoopTree *loop,
2378 IdealLoopTree* outer_loop, int dd, Node_List &old_new,
2379 Node_List& extra_data_nodes) {
2380 if (head->is_strip_mined() && mode != IgnoreStripMined) {
2381 CountedLoopNode* cl = head->as_CountedLoop();
2382 Node* l = cl->outer_loop();
2383 Node* tail = cl->outer_loop_tail();
2384 IfNode* le = cl->outer_loop_end();
2385 Node* sfpt = cl->outer_safepoint();
2386 CountedLoopEndNode* cle = cl->loopexit();
2387 CountedLoopNode* new_cl = old_new[cl->_idx]->as_CountedLoop();
2388 CountedLoopEndNode* new_cle = new_cl->as_CountedLoop()->loopexit_or_null();
2389 Node* cle_out = cle->proj_out(false);
2390
2391 Node* new_sfpt = NULL;
2392 Node* new_cle_out = cle_out->clone();
2393 old_new.map(cle_out->_idx, new_cle_out);
2394 if (mode == CloneIncludesStripMined) {
2395 // clone outer loop body
2396 Node* new_l = l->clone();
2397 Node* new_tail = tail->clone();
2398 IfNode* new_le = le->clone()->as_If();
2399 new_sfpt = sfpt->clone();
2400
2401 set_loop(new_l, outer_loop->_parent);
2402 set_idom(new_l, new_l->in(LoopNode::EntryControl), dd);
2403 set_loop(new_cle_out, outer_loop->_parent);
2404 set_idom(new_cle_out, new_cle, dd);
2405 set_loop(new_sfpt, outer_loop->_parent);
2406 set_idom(new_sfpt, new_cle_out, dd);
2407 set_loop(new_le, outer_loop->_parent);
2408 set_idom(new_le, new_sfpt, dd);
2409 set_loop(new_tail, outer_loop->_parent);
2410 set_idom(new_tail, new_le, dd);
2411 set_idom(new_cl, new_l, dd);
2412
2413 old_new.map(l->_idx, new_l);
2414 old_new.map(tail->_idx, new_tail);
2415 old_new.map(le->_idx, new_le);
2416 old_new.map(sfpt->_idx, new_sfpt);
2417
2418 new_l->set_req(LoopNode::LoopBackControl, new_tail);
2419 new_l->set_req(0, new_l);
2420 new_tail->set_req(0, new_le);
2421 new_le->set_req(0, new_sfpt);
2422 new_sfpt->set_req(0, new_cle_out);
2423 new_cle_out->set_req(0, new_cle);
2424 new_cl->set_req(LoopNode::EntryControl, new_l);
2425
2426 _igvn.register_new_node_with_optimizer(new_l);
2427 _igvn.register_new_node_with_optimizer(new_tail);
2428 _igvn.register_new_node_with_optimizer(new_le);
2429 } else {
2430 Node *newhead = old_new[loop->_head->_idx];
2431 newhead->as_Loop()->clear_strip_mined();
2432 _igvn.replace_input_of(newhead, LoopNode::EntryControl, newhead->in(LoopNode::EntryControl)->in(LoopNode::EntryControl));
2433 set_idom(newhead, newhead->in(LoopNode::EntryControl), dd);
2434 }
2435 // Look at data node that were assigned a control in the outer
2436 // loop: they are kept in the outer loop by the safepoint so start
2437 // from the safepoint node's inputs.
2438 IdealLoopTree* outer_loop = get_loop(l);
2439 Node_Stack stack(2);
2440 stack.push(sfpt, 1);
2441 uint new_counter = C->unique();
2442 while (stack.size() > 0) {
2443 Node* n = stack.node();
2444 uint i = stack.index();
2445 while (i < n->req() &&
2446 (n->in(i) == NULL ||
2447 !has_ctrl(n->in(i)) ||
2448 get_loop(get_ctrl(n->in(i))) != outer_loop ||
2449 (old_new[n->in(i)->_idx] != NULL && old_new[n->in(i)->_idx]->_idx >= new_counter))) {
2450 i++;
2451 }
2452 if (i < n->req()) {
2453 stack.set_index(i+1);
2454 stack.push(n->in(i), 0);
2455 } else {
2456 assert(old_new[n->_idx] == NULL || n == sfpt || old_new[n->_idx]->_idx < new_counter, "no clone yet");
2457 Node* m = n == sfpt ? new_sfpt : n->clone();
2458 if (m != NULL) {
2459 for (uint i = 0; i < n->req(); i++) {
2460 if (m->in(i) != NULL && old_new[m->in(i)->_idx] != NULL) {
2461 m->set_req(i, old_new[m->in(i)->_idx]);
2462 }
2463 }
2464 } else {
2465 assert(n == sfpt && mode != CloneIncludesStripMined, "where's the safepoint clone?");
2466 }
2467 if (n != sfpt) {
2468 extra_data_nodes.push(n);
2469 _igvn.register_new_node_with_optimizer(m);
2470 assert(get_ctrl(n) == cle_out, "what other control?");
2471 set_ctrl(m, new_cle_out);
2472 old_new.map(n->_idx, m);
2473 }
2474 stack.pop();
2475 }
2476 }
2477 if (mode == CloneIncludesStripMined) {
2478 _igvn.register_new_node_with_optimizer(new_sfpt);
2479 _igvn.register_new_node_with_optimizer(new_cle_out);
2480 }
2481 // Some other transformation may have pessimistically assigned some
2482 // data nodes to the outer loop. Set their control so they are out
2483 // of the outer loop.
2484 ResourceMark rm;
2485 Unique_Node_List wq;
2486 for (uint i = 0; i < extra_data_nodes.size(); i++) {
2487 Node* old = extra_data_nodes.at(i);
2488 collect_nodes_in_outer_loop_not_reachable_from_sfpt(old, loop, outer_loop, old_new, wq, this, true);
2489 }
2490
2491 for (uint i = 0; i < loop->_body.size(); i++) {
2492 Node* old = loop->_body.at(i);
2493 collect_nodes_in_outer_loop_not_reachable_from_sfpt(old, loop, outer_loop, old_new, wq, this, true);
2494 }
2495
2496 Node* inner_out = sfpt->in(0);
2497 if (inner_out->outcnt() > 1) {
2498 collect_nodes_in_outer_loop_not_reachable_from_sfpt(inner_out, loop, outer_loop, old_new, wq, this, true);
2499 }
2500
2501 Node* new_ctrl = cl->outer_loop_exit();
2502 assert(get_loop(new_ctrl) != outer_loop, "must be out of the loop nest");
2503 for (uint i = 0; i < wq.size(); i++) {
2504 Node* n = wq.at(i);
2505 set_ctrl(n, new_ctrl);
2506 if (n->in(0) != NULL) {
2507 _igvn.replace_input_of(n, 0, new_ctrl);
2508 }
2509 collect_nodes_in_outer_loop_not_reachable_from_sfpt(n, loop, outer_loop, old_new, wq, this, false);
2510 }
2511 } else {
2512 Node *newhead = old_new[loop->_head->_idx];
2513 set_idom(newhead, newhead->in(LoopNode::EntryControl), dd);
2514 }
2515 }
2516
2517 //------------------------------clone_loop-------------------------------------
2518 //
2519 // C L O N E A L O O P B O D Y
2520 //
2521 // This is the basic building block of the loop optimizations. It clones an
2522 // entire loop body. It makes an old_new loop body mapping; with this mapping
2523 // you can find the new-loop equivalent to an old-loop node. All new-loop
2524 // nodes are exactly equal to their old-loop counterparts, all edges are the
2525 // same. All exits from the old-loop now have a RegionNode that merges the
2526 // equivalent new-loop path. This is true even for the normal "loop-exit"
2527 // condition. All uses of loop-invariant old-loop values now come from (one
2528 // or more) Phis that merge their new-loop equivalents.
2529 //
2530 // This operation leaves the graph in an illegal state: there are two valid
2531 // control edges coming from the loop pre-header to both loop bodies. I'll
2532 // definitely have to hack the graph after running this transform.
2533 //
2534 // From this building block I will further edit edges to perform loop peeling
2535 // or loop unrolling or iteration splitting (Range-Check-Elimination), etc.
2536 //
2537 // Parameter side_by_size_idom:
2538 // When side_by_size_idom is NULL, the dominator tree is constructed for
2539 // the clone loop to dominate the original. Used in construction of
2540 // pre-main-post loop sequence.
2541 // When nonnull, the clone and original are side-by-side, both are
2542 // dominated by the side_by_side_idom node. Used in construction of
2543 // unswitched loops.
2544 void PhaseIdealLoop::clone_loop( IdealLoopTree *loop, Node_List &old_new, int dd,
2545 CloneLoopMode mode, Node* side_by_side_idom) {
2546
2547 LoopNode* head = loop->_head->as_Loop();
2548 head->verify_strip_mined(1);
2549
2550 if (C->do_vector_loop() && PrintOpto) {
2551 const char* mname = C->method()->name()->as_quoted_ascii();
2552 if (mname != NULL) {
2553 tty->print("PhaseIdealLoop::clone_loop: for vectorize method %s\n", mname);
2554 }
2555 }
2556
2557 CloneMap& cm = C->clone_map();
2558 if (C->do_vector_loop()) {
2559 cm.set_clone_idx(cm.max_gen()+1);
2560 #ifndef PRODUCT
2561 if (PrintOpto) {
2562 tty->print_cr("PhaseIdealLoop::clone_loop: _clone_idx %d", cm.clone_idx());
2563 loop->dump_head();
2564 }
2565 #endif
2566 }
2567
2568 // Step 1: Clone the loop body. Make the old->new mapping.
2569 clone_loop_body(loop->_body, old_new, &cm);
2570
2571 IdealLoopTree* outer_loop = (head->is_strip_mined() && mode != IgnoreStripMined) ? get_loop(head->as_CountedLoop()->outer_loop()) : loop;
2572
2573 // Step 2: Fix the edges in the new body. If the old input is outside the
2574 // loop use it. If the old input is INside the loop, use the corresponding
2575 // new node instead.
2576 fix_body_edges(loop->_body, loop, old_new, dd, outer_loop->_parent, false);
2577
2578 Node_List extra_data_nodes; // data nodes in the outer strip mined loop
2579 clone_outer_loop(head, mode, loop, outer_loop, dd, old_new, extra_data_nodes);
2580
2581 // Step 3: Now fix control uses. Loop varying control uses have already
2582 // been fixed up (as part of all input edges in Step 2). Loop invariant
2583 // control uses must be either an IfFalse or an IfTrue. Make a merge
2584 // point to merge the old and new IfFalse/IfTrue nodes; make the use
2585 // refer to this.
2586 Node_List worklist;
2587 uint new_counter = C->unique();
2588 fix_ctrl_uses(loop->_body, loop, old_new, mode, side_by_side_idom, &cm, worklist);
2589
2590 // Step 4: If loop-invariant use is not control, it must be dominated by a
2591 // loop exit IfFalse/IfTrue. Find "proper" loop exit. Make a Region
2592 // there if needed. Make a Phi there merging old and new used values.
2593 Node_List *split_if_set = NULL;
2594 Node_List *split_bool_set = NULL;
2595 Node_List *split_cex_set = NULL;
2596 fix_data_uses(loop->_body, loop, mode, outer_loop, new_counter, old_new, worklist, split_if_set, split_bool_set, split_cex_set);
2597
2598 for (uint i = 0; i < extra_data_nodes.size(); i++) {
2599 Node* old = extra_data_nodes.at(i);
2600 clone_loop_handle_data_uses(old, old_new, loop, outer_loop, split_if_set,
2601 split_bool_set, split_cex_set, worklist, new_counter,
2602 mode);
2603 }
2604
2605 // Check for IFs that need splitting/cloning. Happens if an IF outside of
2606 // the loop uses a condition set in the loop. The original IF probably
2607 // takes control from one or more OLD Regions (which in turn get from NEW
2608 // Regions). In any case, there will be a set of Phis for each merge point
2609 // from the IF up to where the original BOOL def exists the loop.
2610 finish_clone_loop(split_if_set, split_bool_set, split_cex_set);
2611
2612 }
2613
2614 void PhaseIdealLoop::finish_clone_loop(Node_List* split_if_set, Node_List* split_bool_set, Node_List* split_cex_set) {
2615 if (split_if_set) {
2616 while (split_if_set->size()) {
2617 Node *iff = split_if_set->pop();
2618 uint input = iff->Opcode() == Op_AllocateArray ? AllocateNode::ValidLengthTest : 1;
2619 if (iff->in(input)->is_Phi()) {
2620 Node *b = clone_iff(iff->in(input)->as_Phi());
2621 _igvn.replace_input_of(iff, input, b);
2622 }
2623 }
2624 }
2625 if (split_bool_set) {
2626 while (split_bool_set->size()) {
2627 Node *b = split_bool_set->pop();
2628 Node *phi = b->in(1);
2629 assert(phi->is_Phi(), "");
2630 CmpNode *cmp = clone_bool((PhiNode*) phi);
2631 _igvn.replace_input_of(b, 1, cmp);
2632 }
2633 }
2634 if (split_cex_set) {
2635 while (split_cex_set->size()) {
2636 Node *b = split_cex_set->pop();
2637 assert(b->in(0)->is_Region(), "");
2638 assert(b->in(1)->is_Phi(), "");
2639 assert(b->in(0)->in(0) == b->in(1)->in(0), "");
2640 split_up(b, b->in(0), NULL);
2641 }
2642 }
2643 }
2644
2645 void PhaseIdealLoop::fix_data_uses(Node_List& body, IdealLoopTree* loop, CloneLoopMode mode, IdealLoopTree* outer_loop,
2646 uint new_counter, Node_List &old_new, Node_List &worklist, Node_List*& split_if_set,
2647 Node_List*& split_bool_set, Node_List*& split_cex_set) {
2648 for(uint i = 0; i < body.size(); i++ ) {
2649 Node* old = body.at(i);
2650 clone_loop_handle_data_uses(old, old_new, loop, outer_loop, split_if_set,
2651 split_bool_set, split_cex_set, worklist, new_counter,
2652 mode);
2653 }
2654 }
2655
2656 void PhaseIdealLoop::fix_ctrl_uses(const Node_List& body, const IdealLoopTree* loop, Node_List &old_new, CloneLoopMode mode,
2657 Node* side_by_side_idom, CloneMap* cm, Node_List &worklist) {
2658 LoopNode* head = loop->_head->as_Loop();
2659 for(uint i = 0; i < body.size(); i++ ) {
2660 Node* old = body.at(i);
2661 if( !old->is_CFG() ) continue;
2662
2663 // Copy uses to a worklist, so I can munge the def-use info
2664 // with impunity.
2665 for (DUIterator_Fast jmax, j = old->fast_outs(jmax); j < jmax; j++) {
2666 worklist.push(old->fast_out(j));
2667 }
2668
2669 while (worklist.size()) { // Visit all uses
2670 Node *use = worklist.pop();
2671 if (!has_node(use)) continue; // Ignore dead nodes
2672 IdealLoopTree *use_loop = get_loop(has_ctrl(use) ? get_ctrl(use) : use );
2673 if (!loop->is_member(use_loop) && use->is_CFG()) {
2674 // Both OLD and USE are CFG nodes here.
2675 assert(use->is_Proj(), "" );
2676 Node* nnn = old_new[old->_idx];
2677
2678 Node* newuse = NULL;
2679 if (head->is_strip_mined() && mode != IgnoreStripMined) {
2680 CountedLoopNode* cl = head->as_CountedLoop();
2681 CountedLoopEndNode* cle = cl->loopexit();
2682 Node* cle_out = cle->proj_out_or_null(false);
2683 if (use == cle_out) {
2684 IfNode* le = cl->outer_loop_end();
2685 use = le->proj_out(false);
2686 use_loop = get_loop(use);
2687 if (mode == CloneIncludesStripMined) {
2688 nnn = old_new[le->_idx];
2689 } else {
2690 newuse = old_new[cle_out->_idx];
2691 }
2692 }
2693 }
2694 if (newuse == NULL) {
2695 newuse = use->clone();
2696 }
2697
2698 // Clone the loop exit control projection
2699 if (C->do_vector_loop() && cm != NULL) {
2700 cm->verify_insert_and_clone(use, newuse, cm->clone_idx());
2701 }
2702 newuse->set_req(0,nnn);
2703 _igvn.register_new_node_with_optimizer(newuse);
2704 set_loop(newuse, use_loop);
2705 set_idom(newuse, nnn, dom_depth(nnn) + 1 );
2706
2707 // We need a Region to merge the exit from the peeled body and the
2708 // exit from the old loop body.
2709 RegionNode *r = new RegionNode(3);
2710 // Map the old use to the new merge point
2711 old_new.map( use->_idx, r );
2712 uint dd_r = MIN2(dom_depth(newuse), dom_depth(use));
2713 assert(dd_r >= dom_depth(dom_lca(newuse, use)), "" );
2714
2715 // The original user of 'use' uses 'r' instead.
2716 for (DUIterator_Last lmin, l = use->last_outs(lmin); l >= lmin;) {
2717 Node* useuse = use->last_out(l);
2718 _igvn.rehash_node_delayed(useuse);
2719 uint uses_found = 0;
2720 if (useuse->in(0) == use) {
2721 useuse->set_req(0, r);
2722 uses_found++;
2723 if (useuse->is_CFG()) {
2724 // This is not a dom_depth > dd_r because when new
2725 // control flow is constructed by a loop opt, a node and
2726 // its dominator can end up at the same dom_depth
2727 assert(dom_depth(useuse) >= dd_r, "");
2728 set_idom(useuse, r, dom_depth(useuse));
2729 }
2730 }
2731 for (uint k = 1; k < useuse->req(); k++) {
2732 if( useuse->in(k) == use ) {
2733 useuse->set_req(k, r);
2734 uses_found++;
2735 if (useuse->is_Loop() && k == LoopNode::EntryControl) {
2736 // This is not a dom_depth > dd_r because when new
2737 // control flow is constructed by a loop opt, a node
2738 // and its dominator can end up at the same dom_depth
2739 assert(dom_depth(useuse) >= dd_r , "");
2740 set_idom(useuse, r, dom_depth(useuse));
2741 }
2742 }
2743 }
2744 l -= uses_found; // we deleted 1 or more copies of this edge
2745 }
2746
2747 // Now finish up 'r'
2748 r->set_req(1, newuse);
2749 r->set_req(2, use);
2750 _igvn.register_new_node_with_optimizer(r);
2751 set_loop(r, use_loop);
2752 set_idom(r, (side_by_side_idom == NULL) ? newuse->in(0) : side_by_side_idom, dd_r);
2753 } // End of if a loop-exit test
2754 }
2755 }
2756 }
2757
2758 void PhaseIdealLoop::fix_body_edges(const Node_List &body, IdealLoopTree* loop, const Node_List &old_new, int dd,
2759 IdealLoopTree* parent, bool partial) {
2760 for(uint i = 0; i < body.size(); i++ ) {
2761 Node *old = body.at(i);
2762 Node *nnn = old_new[old->_idx];
2763 // Fix CFG/Loop controlling the new node
2764 if (has_ctrl(old)) {
2765 set_ctrl(nnn, old_new[get_ctrl(old)->_idx]);
2766 } else {
2767 set_loop(nnn, parent);
2768 if (old->outcnt() > 0) {
2769 Node* dom = idom(old);
2770 if (old_new[dom->_idx] != NULL) {
2771 dom = old_new[dom->_idx];
2772 set_idom(nnn, dom, dd );
2773 }
2774 }
2775 }
2776 // Correct edges to the new node
2777 for (uint j = 0; j < nnn->req(); j++) {
2778 Node *n = nnn->in(j);
2779 if (n != NULL) {
2780 IdealLoopTree *old_in_loop = get_loop(has_ctrl(n) ? get_ctrl(n) : n);
2781 if (loop->is_member(old_in_loop)) {
2782 if (old_new[n->_idx] != NULL) {
2783 nnn->set_req(j, old_new[n->_idx]);
2784 } else {
2785 assert(!body.contains(n), "");
2786 assert(partial, "node not cloned");
2787 }
2788 }
2789 }
2790 }
2791 _igvn.hash_find_insert(nnn);
2792 }
2793 }
2794
2795 void PhaseIdealLoop::clone_loop_body(const Node_List& body, Node_List &old_new, CloneMap* cm) {
2796 for (uint i = 0; i < body.size(); i++) {
2797 Node* old = body.at(i);
2798 Node* nnn = old->clone();
2799 old_new.map(old->_idx, nnn);
2800 if (old->is_reduction()) {
2801 // Reduction flag is not copied by default. Copy it here when cloning the entire loop body.
2802 nnn->add_flag(Node::Flag_is_reduction);
2803 }
2804 if (C->do_vector_loop() && cm != NULL) {
2805 cm->verify_insert_and_clone(old, nnn, cm->clone_idx());
2806 }
2807 _igvn.register_new_node_with_optimizer(nnn);
2808 }
2809 }
2810
2811
2812 //---------------------- stride_of_possible_iv -------------------------------------
2813 // Looks for an iff/bool/comp with one operand of the compare
2814 // being a cycle involving an add and a phi,
2815 // with an optional truncation (left-shift followed by a right-shift)
2816 // of the add. Returns zero if not an iv.
2817 int PhaseIdealLoop::stride_of_possible_iv(Node* iff) {
2818 Node* trunc1 = NULL;
2819 Node* trunc2 = NULL;
2820 const TypeInteger* ttype = NULL;
2821 if (!iff->is_If() || iff->in(1) == NULL || !iff->in(1)->is_Bool()) {
2822 return 0;
2823 }
2824 BoolNode* bl = iff->in(1)->as_Bool();
2825 Node* cmp = bl->in(1);
2826 if (!cmp || (cmp->Opcode() != Op_CmpI && cmp->Opcode() != Op_CmpU)) {
2827 return 0;
2828 }
2829 // Must have an invariant operand
2830 if (is_member(get_loop(iff), get_ctrl(cmp->in(2)))) {
2831 return 0;
2832 }
2833 Node* add2 = NULL;
2834 Node* cmp1 = cmp->in(1);
2835 if (cmp1->is_Phi()) {
2836 // (If (Bool (CmpX phi:(Phi ...(Optional-trunc(AddI phi add2))) )))
2837 Node* phi = cmp1;
2838 for (uint i = 1; i < phi->req(); i++) {
2839 Node* in = phi->in(i);
2840 Node* add = CountedLoopNode::match_incr_with_optional_truncation(in,
2841 &trunc1, &trunc2, &ttype, T_INT);
2842 if (add && add->in(1) == phi) {
2843 add2 = add->in(2);
2844 break;
2845 }
2846 }
2847 } else {
2848 // (If (Bool (CmpX addtrunc:(Optional-trunc((AddI (Phi ...addtrunc...) add2)) )))
2849 Node* addtrunc = cmp1;
2850 Node* add = CountedLoopNode::match_incr_with_optional_truncation(addtrunc,
2851 &trunc1, &trunc2, &ttype, T_INT);
2852 if (add && add->in(1)->is_Phi()) {
2853 Node* phi = add->in(1);
2854 for (uint i = 1; i < phi->req(); i++) {
2855 if (phi->in(i) == addtrunc) {
2856 add2 = add->in(2);
2857 break;
2858 }
2859 }
2860 }
2861 }
2862 if (add2 != NULL) {
2863 const TypeInt* add2t = _igvn.type(add2)->is_int();
2864 if (add2t->is_con()) {
2865 return add2t->get_con();
2866 }
2867 }
2868 return 0;
2869 }
2870
2871
2872 //---------------------- stay_in_loop -------------------------------------
2873 // Return the (unique) control output node that's in the loop (if it exists.)
2874 Node* PhaseIdealLoop::stay_in_loop( Node* n, IdealLoopTree *loop) {
2875 Node* unique = NULL;
2876 if (!n) return NULL;
2877 for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
2878 Node* use = n->fast_out(i);
2879 if (!has_ctrl(use) && loop->is_member(get_loop(use))) {
2880 if (unique != NULL) {
2881 return NULL;
2882 }
2883 unique = use;
2884 }
2885 }
2886 return unique;
2887 }
2888
2889 //------------------------------ register_node -------------------------------------
2890 // Utility to register node "n" with PhaseIdealLoop
2891 void PhaseIdealLoop::register_node(Node* n, IdealLoopTree *loop, Node* pred, int ddepth) {
2892 _igvn.register_new_node_with_optimizer(n);
2893 loop->_body.push(n);
2894 if (n->is_CFG()) {
2895 set_loop(n, loop);
2896 set_idom(n, pred, ddepth);
2897 } else {
2898 set_ctrl(n, pred);
2899 }
2900 }
2901
2902 //------------------------------ proj_clone -------------------------------------
2903 // Utility to create an if-projection
2904 ProjNode* PhaseIdealLoop::proj_clone(ProjNode* p, IfNode* iff) {
2905 ProjNode* c = p->clone()->as_Proj();
2906 c->set_req(0, iff);
2907 return c;
2908 }
2909
2910 //------------------------------ short_circuit_if -------------------------------------
2911 // Force the iff control output to be the live_proj
2912 Node* PhaseIdealLoop::short_circuit_if(IfNode* iff, ProjNode* live_proj) {
2913 guarantee(live_proj != NULL, "null projection");
2914 int proj_con = live_proj->_con;
2915 assert(proj_con == 0 || proj_con == 1, "false or true projection");
2916 Node *con = _igvn.intcon(proj_con);
2917 set_ctrl(con, C->root());
2918 if (iff) {
2919 iff->set_req(1, con);
2920 }
2921 return con;
2922 }
2923
2924 //------------------------------ insert_if_before_proj -------------------------------------
2925 // Insert a new if before an if projection (* - new node)
2926 //
2927 // before
2928 // if(test)
2929 // / \
2930 // v v
2931 // other-proj proj (arg)
2932 //
2933 // after
2934 // if(test)
2935 // / \
2936 // / v
2937 // | * proj-clone
2938 // v |
2939 // other-proj v
2940 // * new_if(relop(cmp[IU](left,right)))
2941 // / \
2942 // v v
2943 // * new-proj proj
2944 // (returned)
2945 //
2946 ProjNode* PhaseIdealLoop::insert_if_before_proj(Node* left, bool Signed, BoolTest::mask relop, Node* right, ProjNode* proj) {
2947 IfNode* iff = proj->in(0)->as_If();
2948 IdealLoopTree *loop = get_loop(proj);
2949 ProjNode *other_proj = iff->proj_out(!proj->is_IfTrue())->as_Proj();
2950 int ddepth = dom_depth(proj);
2951
2952 _igvn.rehash_node_delayed(iff);
2953 _igvn.rehash_node_delayed(proj);
2954
2955 proj->set_req(0, NULL); // temporary disconnect
2956 ProjNode* proj2 = proj_clone(proj, iff);
2957 register_node(proj2, loop, iff, ddepth);
2958
2959 Node* cmp = Signed ? (Node*) new CmpINode(left, right) : (Node*) new CmpUNode(left, right);
2960 register_node(cmp, loop, proj2, ddepth);
2961
2962 BoolNode* bol = new BoolNode(cmp, relop);
2963 register_node(bol, loop, proj2, ddepth);
2964
2965 int opcode = iff->Opcode();
2966 assert(opcode == Op_If || opcode == Op_RangeCheck, "unexpected opcode");
2967 IfNode* new_if = (opcode == Op_If) ? new IfNode(proj2, bol, iff->_prob, iff->_fcnt):
2968 new RangeCheckNode(proj2, bol, iff->_prob, iff->_fcnt);
2969 register_node(new_if, loop, proj2, ddepth);
2970
2971 proj->set_req(0, new_if); // reattach
2972 set_idom(proj, new_if, ddepth);
2973
2974 ProjNode* new_exit = proj_clone(other_proj, new_if)->as_Proj();
2975 guarantee(new_exit != NULL, "null exit node");
2976 register_node(new_exit, get_loop(other_proj), new_if, ddepth);
2977
2978 return new_exit;
2979 }
2980
2981 //------------------------------ insert_region_before_proj -------------------------------------
2982 // Insert a region before an if projection (* - new node)
2983 //
2984 // before
2985 // if(test)
2986 // / |
2987 // v |
2988 // proj v
2989 // other-proj
2990 //
2991 // after
2992 // if(test)
2993 // / |
2994 // v |
2995 // * proj-clone v
2996 // | other-proj
2997 // v
2998 // * new-region
2999 // |
3000 // v
3001 // * dum_if
3002 // / \
3003 // v \
3004 // * dum-proj v
3005 // proj
3006 //
3007 RegionNode* PhaseIdealLoop::insert_region_before_proj(ProjNode* proj) {
3008 IfNode* iff = proj->in(0)->as_If();
3009 IdealLoopTree *loop = get_loop(proj);
3010 ProjNode *other_proj = iff->proj_out(!proj->is_IfTrue())->as_Proj();
3011 int ddepth = dom_depth(proj);
3012
3013 _igvn.rehash_node_delayed(iff);
3014 _igvn.rehash_node_delayed(proj);
3015
3016 proj->set_req(0, NULL); // temporary disconnect
3017 ProjNode* proj2 = proj_clone(proj, iff);
3018 register_node(proj2, loop, iff, ddepth);
3019
3020 RegionNode* reg = new RegionNode(2);
3021 reg->set_req(1, proj2);
3022 register_node(reg, loop, iff, ddepth);
3023
3024 IfNode* dum_if = new IfNode(reg, short_circuit_if(NULL, proj), iff->_prob, iff->_fcnt);
3025 register_node(dum_if, loop, reg, ddepth);
3026
3027 proj->set_req(0, dum_if); // reattach
3028 set_idom(proj, dum_if, ddepth);
3029
3030 ProjNode* dum_proj = proj_clone(other_proj, dum_if);
3031 register_node(dum_proj, loop, dum_if, ddepth);
3032
3033 return reg;
3034 }
3035
3036 //------------------------------ insert_cmpi_loop_exit -------------------------------------
3037 // Clone a signed compare loop exit from an unsigned compare and
3038 // insert it before the unsigned cmp on the stay-in-loop path.
3039 // All new nodes inserted in the dominator tree between the original
3040 // if and it's projections. The original if test is replaced with
3041 // a constant to force the stay-in-loop path.
3042 //
3043 // This is done to make sure that the original if and it's projections
3044 // still dominate the same set of control nodes, that the ctrl() relation
3045 // from data nodes to them is preserved, and that their loop nesting is
3046 // preserved.
3047 //
3048 // before
3049 // if(i <u limit) unsigned compare loop exit
3050 // / |
3051 // v v
3052 // exit-proj stay-in-loop-proj
3053 //
3054 // after
3055 // if(stay-in-loop-const) original if
3056 // / |
3057 // / v
3058 // / if(i < limit) new signed test
3059 // / / |
3060 // / / v
3061 // / / if(i <u limit) new cloned unsigned test
3062 // / / / |
3063 // v v v |
3064 // region |
3065 // | |
3066 // dum-if |
3067 // / | |
3068 // ether | |
3069 // v v
3070 // exit-proj stay-in-loop-proj
3071 //
3072 IfNode* PhaseIdealLoop::insert_cmpi_loop_exit(IfNode* if_cmpu, IdealLoopTree *loop) {
3073 const bool Signed = true;
3074 const bool Unsigned = false;
3075
3076 BoolNode* bol = if_cmpu->in(1)->as_Bool();
3077 if (bol->_test._test != BoolTest::lt) return NULL;
3078 CmpNode* cmpu = bol->in(1)->as_Cmp();
3079 if (cmpu->Opcode() != Op_CmpU) return NULL;
3080 int stride = stride_of_possible_iv(if_cmpu);
3081 if (stride == 0) return NULL;
3082
3083 Node* lp_proj = stay_in_loop(if_cmpu, loop);
3084 guarantee(lp_proj != NULL, "null loop node");
3085
3086 ProjNode* lp_continue = lp_proj->as_Proj();
3087 ProjNode* lp_exit = if_cmpu->proj_out(!lp_continue->is_IfTrue())->as_Proj();
3088 if (!lp_exit->is_IfFalse()) {
3089 // The loop exit condition is (i <u limit) ==> (i >= 0 && i < limit).
3090 // We therefore can't add a single exit condition.
3091 return NULL;
3092 }
3093 // The loop exit condition is !(i <u limit) ==> (i < 0 || i >= limit).
3094 // Split out the exit condition (i < 0) for stride < 0 or (i >= limit) for stride > 0.
3095 Node* limit = NULL;
3096 if (stride > 0) {
3097 limit = cmpu->in(2);
3098 } else {
3099 limit = _igvn.makecon(TypeInt::ZERO);
3100 set_ctrl(limit, C->root());
3101 }
3102 // Create a new region on the exit path
3103 RegionNode* reg = insert_region_before_proj(lp_exit);
3104 guarantee(reg != NULL, "null region node");
3105
3106 // Clone the if-cmpu-true-false using a signed compare
3107 BoolTest::mask rel_i = stride > 0 ? bol->_test._test : BoolTest::ge;
3108 ProjNode* cmpi_exit = insert_if_before_proj(cmpu->in(1), Signed, rel_i, limit, lp_continue);
3109 reg->add_req(cmpi_exit);
3110
3111 // Clone the if-cmpu-true-false
3112 BoolTest::mask rel_u = bol->_test._test;
3113 ProjNode* cmpu_exit = insert_if_before_proj(cmpu->in(1), Unsigned, rel_u, cmpu->in(2), lp_continue);
3114 reg->add_req(cmpu_exit);
3115
3116 // Force original if to stay in loop.
3117 short_circuit_if(if_cmpu, lp_continue);
3118
3119 return cmpi_exit->in(0)->as_If();
3120 }
3121
3122 //------------------------------ remove_cmpi_loop_exit -------------------------------------
3123 // Remove a previously inserted signed compare loop exit.
3124 void PhaseIdealLoop::remove_cmpi_loop_exit(IfNode* if_cmp, IdealLoopTree *loop) {
3125 Node* lp_proj = stay_in_loop(if_cmp, loop);
3126 assert(if_cmp->in(1)->in(1)->Opcode() == Op_CmpI &&
3127 stay_in_loop(lp_proj, loop)->is_If() &&
3128 stay_in_loop(lp_proj, loop)->in(1)->in(1)->Opcode() == Op_CmpU, "inserted cmpi before cmpu");
3129 Node *con = _igvn.makecon(lp_proj->is_IfTrue() ? TypeInt::ONE : TypeInt::ZERO);
3130 set_ctrl(con, C->root());
3131 if_cmp->set_req(1, con);
3132 }
3133
3134 //------------------------------ scheduled_nodelist -------------------------------------
3135 // Create a post order schedule of nodes that are in the
3136 // "member" set. The list is returned in "sched".
3137 // The first node in "sched" is the loop head, followed by
3138 // nodes which have no inputs in the "member" set, and then
3139 // followed by the nodes that have an immediate input dependence
3140 // on a node in "sched".
3141 void PhaseIdealLoop::scheduled_nodelist( IdealLoopTree *loop, VectorSet& member, Node_List &sched ) {
3142
3143 assert(member.test(loop->_head->_idx), "loop head must be in member set");
3144 VectorSet visited;
3145 Node_Stack nstack(loop->_body.size());
3146
3147 Node* n = loop->_head; // top of stack is cached in "n"
3148 uint idx = 0;
3149 visited.set(n->_idx);
3150
3151 // Initially push all with no inputs from within member set
3152 for(uint i = 0; i < loop->_body.size(); i++ ) {
3153 Node *elt = loop->_body.at(i);
3154 if (member.test(elt->_idx)) {
3155 bool found = false;
3156 for (uint j = 0; j < elt->req(); j++) {
3157 Node* def = elt->in(j);
3158 if (def && member.test(def->_idx) && def != elt) {
3159 found = true;
3160 break;
3161 }
3162 }
3163 if (!found && elt != loop->_head) {
3164 nstack.push(n, idx);
3165 n = elt;
3166 assert(!visited.test(n->_idx), "not seen yet");
3167 visited.set(n->_idx);
3168 }
3169 }
3170 }
3171
3172 // traverse out's that are in the member set
3173 while (true) {
3174 if (idx < n->outcnt()) {
3175 Node* use = n->raw_out(idx);
3176 idx++;
3177 if (!visited.test_set(use->_idx)) {
3178 if (member.test(use->_idx)) {
3179 nstack.push(n, idx);
3180 n = use;
3181 idx = 0;
3182 }
3183 }
3184 } else {
3185 // All outputs processed
3186 sched.push(n);
3187 if (nstack.is_empty()) break;
3188 n = nstack.node();
3189 idx = nstack.index();
3190 nstack.pop();
3191 }
3192 }
3193 }
3194
3195
3196 //------------------------------ has_use_in_set -------------------------------------
3197 // Has a use in the vector set
3198 bool PhaseIdealLoop::has_use_in_set( Node* n, VectorSet& vset ) {
3199 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
3200 Node* use = n->fast_out(j);
3201 if (vset.test(use->_idx)) {
3202 return true;
3203 }
3204 }
3205 return false;
3206 }
3207
3208
3209 //------------------------------ has_use_internal_to_set -------------------------------------
3210 // Has use internal to the vector set (ie. not in a phi at the loop head)
3211 bool PhaseIdealLoop::has_use_internal_to_set( Node* n, VectorSet& vset, IdealLoopTree *loop ) {
3212 Node* head = loop->_head;
3213 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
3214 Node* use = n->fast_out(j);
3215 if (vset.test(use->_idx) && !(use->is_Phi() && use->in(0) == head)) {
3216 return true;
3217 }
3218 }
3219 return false;
3220 }
3221
3222
3223 //------------------------------ clone_for_use_outside_loop -------------------------------------
3224 // clone "n" for uses that are outside of loop
3225 int PhaseIdealLoop::clone_for_use_outside_loop( IdealLoopTree *loop, Node* n, Node_List& worklist ) {
3226 int cloned = 0;
3227 assert(worklist.size() == 0, "should be empty");
3228 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
3229 Node* use = n->fast_out(j);
3230 if( !loop->is_member(get_loop(has_ctrl(use) ? get_ctrl(use) : use)) ) {
3231 worklist.push(use);
3232 }
3233 }
3234
3235 if (C->check_node_count(worklist.size() + NodeLimitFudgeFactor,
3236 "Too many clones required in clone_for_use_outside_loop in partial peeling")) {
3237 return -1;
3238 }
3239
3240 while( worklist.size() ) {
3241 Node *use = worklist.pop();
3242 if (!has_node(use) || use->in(0) == C->top()) continue;
3243 uint j;
3244 for (j = 0; j < use->req(); j++) {
3245 if (use->in(j) == n) break;
3246 }
3247 assert(j < use->req(), "must be there");
3248
3249 // clone "n" and insert it between the inputs of "n" and the use outside the loop
3250 Node* n_clone = n->clone();
3251 _igvn.replace_input_of(use, j, n_clone);
3252 cloned++;
3253 Node* use_c;
3254 if (!use->is_Phi()) {
3255 use_c = has_ctrl(use) ? get_ctrl(use) : use->in(0);
3256 } else {
3257 // Use in a phi is considered a use in the associated predecessor block
3258 use_c = use->in(0)->in(j);
3259 }
3260 set_ctrl(n_clone, use_c);
3261 assert(!loop->is_member(get_loop(use_c)), "should be outside loop");
3262 get_loop(use_c)->_body.push(n_clone);
3263 _igvn.register_new_node_with_optimizer(n_clone);
3264 #ifndef PRODUCT
3265 if (TracePartialPeeling) {
3266 tty->print_cr("loop exit cloning old: %d new: %d newbb: %d", n->_idx, n_clone->_idx, get_ctrl(n_clone)->_idx);
3267 }
3268 #endif
3269 }
3270 return cloned;
3271 }
3272
3273
3274 //------------------------------ clone_for_special_use_inside_loop -------------------------------------
3275 // clone "n" for special uses that are in the not_peeled region.
3276 // If these def-uses occur in separate blocks, the code generator
3277 // marks the method as not compilable. For example, if a "BoolNode"
3278 // is in a different basic block than the "IfNode" that uses it, then
3279 // the compilation is aborted in the code generator.
3280 void PhaseIdealLoop::clone_for_special_use_inside_loop( IdealLoopTree *loop, Node* n,
3281 VectorSet& not_peel, Node_List& sink_list, Node_List& worklist ) {
3282 if (n->is_Phi() || n->is_Load()) {
3283 return;
3284 }
3285 assert(worklist.size() == 0, "should be empty");
3286 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
3287 Node* use = n->fast_out(j);
3288 if ( not_peel.test(use->_idx) &&
3289 (use->is_If() || use->is_CMove() || use->is_Bool()) &&
3290 use->in(1) == n) {
3291 worklist.push(use);
3292 }
3293 }
3294 if (worklist.size() > 0) {
3295 // clone "n" and insert it between inputs of "n" and the use
3296 Node* n_clone = n->clone();
3297 loop->_body.push(n_clone);
3298 _igvn.register_new_node_with_optimizer(n_clone);
3299 set_ctrl(n_clone, get_ctrl(n));
3300 sink_list.push(n_clone);
3301 not_peel.set(n_clone->_idx);
3302 #ifndef PRODUCT
3303 if (TracePartialPeeling) {
3304 tty->print_cr("special not_peeled cloning old: %d new: %d", n->_idx, n_clone->_idx);
3305 }
3306 #endif
3307 while( worklist.size() ) {
3308 Node *use = worklist.pop();
3309 _igvn.rehash_node_delayed(use);
3310 for (uint j = 1; j < use->req(); j++) {
3311 if (use->in(j) == n) {
3312 use->set_req(j, n_clone);
3313 }
3314 }
3315 }
3316 }
3317 }
3318
3319
3320 //------------------------------ insert_phi_for_loop -------------------------------------
3321 // Insert phi(lp_entry_val, back_edge_val) at use->in(idx) for loop lp if phi does not already exist
3322 void PhaseIdealLoop::insert_phi_for_loop( Node* use, uint idx, Node* lp_entry_val, Node* back_edge_val, LoopNode* lp ) {
3323 Node *phi = PhiNode::make(lp, back_edge_val);
3324 phi->set_req(LoopNode::EntryControl, lp_entry_val);
3325 // Use existing phi if it already exists
3326 Node *hit = _igvn.hash_find_insert(phi);
3327 if( hit == NULL ) {
3328 _igvn.register_new_node_with_optimizer(phi);
3329 set_ctrl(phi, lp);
3330 } else {
3331 // Remove the new phi from the graph and use the hit
3332 _igvn.remove_dead_node(phi);
3333 phi = hit;
3334 }
3335 _igvn.replace_input_of(use, idx, phi);
3336 }
3337
3338 #ifdef ASSERT
3339 //------------------------------ is_valid_loop_partition -------------------------------------
3340 // Validate the loop partition sets: peel and not_peel
3341 bool PhaseIdealLoop::is_valid_loop_partition( IdealLoopTree *loop, VectorSet& peel, Node_List& peel_list,
3342 VectorSet& not_peel ) {
3343 uint i;
3344 // Check that peel_list entries are in the peel set
3345 for (i = 0; i < peel_list.size(); i++) {
3346 if (!peel.test(peel_list.at(i)->_idx)) {
3347 return false;
3348 }
3349 }
3350 // Check at loop members are in one of peel set or not_peel set
3351 for (i = 0; i < loop->_body.size(); i++ ) {
3352 Node *def = loop->_body.at(i);
3353 uint di = def->_idx;
3354 // Check that peel set elements are in peel_list
3355 if (peel.test(di)) {
3356 if (not_peel.test(di)) {
3357 return false;
3358 }
3359 // Must be in peel_list also
3360 bool found = false;
3361 for (uint j = 0; j < peel_list.size(); j++) {
3362 if (peel_list.at(j)->_idx == di) {
3363 found = true;
3364 break;
3365 }
3366 }
3367 if (!found) {
3368 return false;
3369 }
3370 } else if (not_peel.test(di)) {
3371 if (peel.test(di)) {
3372 return false;
3373 }
3374 } else {
3375 return false;
3376 }
3377 }
3378 return true;
3379 }
3380
3381 //------------------------------ is_valid_clone_loop_exit_use -------------------------------------
3382 // Ensure a use outside of loop is of the right form
3383 bool PhaseIdealLoop::is_valid_clone_loop_exit_use( IdealLoopTree *loop, Node* use, uint exit_idx) {
3384 Node *use_c = has_ctrl(use) ? get_ctrl(use) : use;
3385 return (use->is_Phi() &&
3386 use_c->is_Region() && use_c->req() == 3 &&
3387 (use_c->in(exit_idx)->Opcode() == Op_IfTrue ||
3388 use_c->in(exit_idx)->Opcode() == Op_IfFalse ||
3389 use_c->in(exit_idx)->Opcode() == Op_JumpProj) &&
3390 loop->is_member( get_loop( use_c->in(exit_idx)->in(0) ) ) );
3391 }
3392
3393 //------------------------------ is_valid_clone_loop_form -------------------------------------
3394 // Ensure that all uses outside of loop are of the right form
3395 bool PhaseIdealLoop::is_valid_clone_loop_form( IdealLoopTree *loop, Node_List& peel_list,
3396 uint orig_exit_idx, uint clone_exit_idx) {
3397 uint len = peel_list.size();
3398 for (uint i = 0; i < len; i++) {
3399 Node *def = peel_list.at(i);
3400
3401 for (DUIterator_Fast jmax, j = def->fast_outs(jmax); j < jmax; j++) {
3402 Node *use = def->fast_out(j);
3403 Node *use_c = has_ctrl(use) ? get_ctrl(use) : use;
3404 if (!loop->is_member(get_loop(use_c))) {
3405 // use is not in the loop, check for correct structure
3406 if (use->in(0) == def) {
3407 // Okay
3408 } else if (!is_valid_clone_loop_exit_use(loop, use, orig_exit_idx)) {
3409 return false;
3410 }
3411 }
3412 }
3413 }
3414 return true;
3415 }
3416 #endif
3417
3418 //------------------------------ partial_peel -------------------------------------
3419 // Partially peel (aka loop rotation) the top portion of a loop (called
3420 // the peel section below) by cloning it and placing one copy just before
3421 // the new loop head and the other copy at the bottom of the new loop.
3422 //
3423 // before after where it came from
3424 //
3425 // stmt1 stmt1
3426 // loop: stmt2 clone
3427 // stmt2 if condA goto exitA clone
3428 // if condA goto exitA new_loop: new
3429 // stmt3 stmt3 clone
3430 // if !condB goto loop if condB goto exitB clone
3431 // exitB: stmt2 orig
3432 // stmt4 if !condA goto new_loop orig
3433 // exitA: goto exitA
3434 // exitB:
3435 // stmt4
3436 // exitA:
3437 //
3438 // Step 1: find the cut point: an exit test on probable
3439 // induction variable.
3440 // Step 2: schedule (with cloning) operations in the peel
3441 // section that can be executed after the cut into
3442 // the section that is not peeled. This may need
3443 // to clone operations into exit blocks. For
3444 // instance, a reference to A[i] in the not-peel
3445 // section and a reference to B[i] in an exit block
3446 // may cause a left-shift of i by 2 to be placed
3447 // in the peel block. This step will clone the left
3448 // shift into the exit block and sink the left shift
3449 // from the peel to the not-peel section.
3450 // Step 3: clone the loop, retarget the control, and insert
3451 // phis for values that are live across the new loop
3452 // head. This is very dependent on the graph structure
3453 // from clone_loop. It creates region nodes for
3454 // exit control and associated phi nodes for values
3455 // flow out of the loop through that exit. The region
3456 // node is dominated by the clone's control projection.
3457 // So the clone's peel section is placed before the
3458 // new loop head, and the clone's not-peel section is
3459 // forms the top part of the new loop. The original
3460 // peel section forms the tail of the new loop.
3461 // Step 4: update the dominator tree and recompute the
3462 // dominator depth.
3463 //
3464 // orig
3465 //
3466 // stmt1
3467 // |
3468 // v
3469 // loop predicate
3470 // |
3471 // v
3472 // loop<----+
3473 // | |
3474 // stmt2 |
3475 // | |
3476 // v |
3477 // ifA |
3478 // / | |
3479 // v v |
3480 // false true ^ <-- last_peel
3481 // / | |
3482 // / ===|==cut |
3483 // / stmt3 | <-- first_not_peel
3484 // / | |
3485 // | v |
3486 // v ifB |
3487 // exitA: / \ |
3488 // / \ |
3489 // v v |
3490 // false true |
3491 // / \ |
3492 // / ----+
3493 // |
3494 // v
3495 // exitB:
3496 // stmt4
3497 //
3498 //
3499 // after clone loop
3500 //
3501 // stmt1
3502 // |
3503 // v
3504 // loop predicate
3505 // / \
3506 // clone / \ orig
3507 // / \
3508 // / \
3509 // v v
3510 // +---->loop loop<----+
3511 // | | | |
3512 // | stmt2 stmt2 |
3513 // | | | |
3514 // | v v |
3515 // | ifA ifA |
3516 // | | \ / | |
3517 // | v v v v |
3518 // ^ true false false true ^ <-- last_peel
3519 // | | ^ \ / | |
3520 // | cut==|== \ \ / ===|==cut |
3521 // | stmt3 \ \ / stmt3 | <-- first_not_peel
3522 // | | dom | | | |
3523 // | v \ 1v v2 v |
3524 // | ifB regionA ifB |
3525 // | / \ | / \ |
3526 // | / \ v / \ |
3527 // | v v exitA: v v |
3528 // | true false false true |
3529 // | / ^ \ / \ |
3530 // +---- \ \ / ----+
3531 // dom \ /
3532 // \ 1v v2
3533 // regionB
3534 // |
3535 // v
3536 // exitB:
3537 // stmt4
3538 //
3539 //
3540 // after partial peel
3541 //
3542 // stmt1
3543 // |
3544 // v
3545 // loop predicate
3546 // /
3547 // clone / orig
3548 // / TOP
3549 // / \
3550 // v v
3551 // TOP->loop loop----+
3552 // | | |
3553 // stmt2 stmt2 |
3554 // | | |
3555 // v v |
3556 // ifA ifA |
3557 // | \ / | |
3558 // v v v v |
3559 // true false false true | <-- last_peel
3560 // | ^ \ / +------|---+
3561 // +->newloop \ \ / === ==cut | |
3562 // | stmt3 \ \ / TOP | |
3563 // | | dom | | stmt3 | | <-- first_not_peel
3564 // | v \ 1v v2 v | |
3565 // | ifB regionA ifB ^ v
3566 // | / \ | / \ | |
3567 // | / \ v / \ | |
3568 // | v v exitA: v v | |
3569 // | true false false true | |
3570 // | / ^ \ / \ | |
3571 // | | \ \ / v | |
3572 // | | dom \ / TOP | |
3573 // | | \ 1v v2 | |
3574 // ^ v regionB | |
3575 // | | | | |
3576 // | | v ^ v
3577 // | | exitB: | |
3578 // | | stmt4 | |
3579 // | +------------>-----------------+ |
3580 // | |
3581 // +-----------------<---------------------+
3582 //
3583 //
3584 // final graph
3585 //
3586 // stmt1
3587 // |
3588 // v
3589 // loop predicate
3590 // |
3591 // v
3592 // stmt2 clone
3593 // |
3594 // v
3595 // ........> ifA clone
3596 // : / |
3597 // dom / |
3598 // : v v
3599 // : false true
3600 // : | |
3601 // : | v
3602 // : | newloop<-----+
3603 // : | | |
3604 // : | stmt3 clone |
3605 // : | | |
3606 // : | v |
3607 // : | ifB |
3608 // : | / \ |
3609 // : | v v |
3610 // : | false true |
3611 // : | | | |
3612 // : | v stmt2 |
3613 // : | exitB: | |
3614 // : | stmt4 v |
3615 // : | ifA orig |
3616 // : | / \ |
3617 // : | / \ |
3618 // : | v v |
3619 // : | false true |
3620 // : | / \ |
3621 // : v v -----+
3622 // RegionA
3623 // |
3624 // v
3625 // exitA
3626 //
3627 bool PhaseIdealLoop::partial_peel( IdealLoopTree *loop, Node_List &old_new ) {
3628
3629 assert(!loop->_head->is_CountedLoop(), "Non-counted loop only");
3630 if (!loop->_head->is_Loop()) {
3631 return false;
3632 }
3633 LoopNode *head = loop->_head->as_Loop();
3634
3635 if (head->is_partial_peel_loop() || head->partial_peel_has_failed()) {
3636 return false;
3637 }
3638
3639 // Check for complex exit control
3640 for (uint ii = 0; ii < loop->_body.size(); ii++) {
3641 Node *n = loop->_body.at(ii);
3642 int opc = n->Opcode();
3643 if (n->is_Call() ||
3644 opc == Op_Catch ||
3645 opc == Op_CatchProj ||
3646 opc == Op_Jump ||
3647 opc == Op_JumpProj) {
3648 #ifndef PRODUCT
3649 if (TracePartialPeeling) {
3650 tty->print_cr("\nExit control too complex: lp: %d", head->_idx);
3651 }
3652 #endif
3653 return false;
3654 }
3655 }
3656
3657 int dd = dom_depth(head);
3658
3659 // Step 1: find cut point
3660
3661 // Walk up dominators to loop head looking for first loop exit
3662 // which is executed on every path thru loop.
3663 IfNode *peel_if = NULL;
3664 IfNode *peel_if_cmpu = NULL;
3665
3666 Node *iff = loop->tail();
3667 while (iff != head) {
3668 if (iff->is_If()) {
3669 Node *ctrl = get_ctrl(iff->in(1));
3670 if (ctrl->is_top()) return false; // Dead test on live IF.
3671 // If loop-varying exit-test, check for induction variable
3672 if (loop->is_member(get_loop(ctrl)) &&
3673 loop->is_loop_exit(iff) &&
3674 is_possible_iv_test(iff)) {
3675 Node* cmp = iff->in(1)->in(1);
3676 if (cmp->Opcode() == Op_CmpI) {
3677 peel_if = iff->as_If();
3678 } else {
3679 assert(cmp->Opcode() == Op_CmpU, "must be CmpI or CmpU");
3680 peel_if_cmpu = iff->as_If();
3681 }
3682 }
3683 }
3684 iff = idom(iff);
3685 }
3686
3687 // Prefer signed compare over unsigned compare.
3688 IfNode* new_peel_if = NULL;
3689 if (peel_if == NULL) {
3690 if (!PartialPeelAtUnsignedTests || peel_if_cmpu == NULL) {
3691 return false; // No peel point found
3692 }
3693 new_peel_if = insert_cmpi_loop_exit(peel_if_cmpu, loop);
3694 if (new_peel_if == NULL) {
3695 return false; // No peel point found
3696 }
3697 peel_if = new_peel_if;
3698 }
3699 Node* last_peel = stay_in_loop(peel_if, loop);
3700 Node* first_not_peeled = stay_in_loop(last_peel, loop);
3701 if (first_not_peeled == NULL || first_not_peeled == head) {
3702 return false;
3703 }
3704
3705 #ifndef PRODUCT
3706 if (TraceLoopOpts) {
3707 tty->print("PartialPeel ");
3708 loop->dump_head();
3709 }
3710
3711 if (TracePartialPeeling) {
3712 tty->print_cr("before partial peel one iteration");
3713 Node_List wl;
3714 Node* t = head->in(2);
3715 while (true) {
3716 wl.push(t);
3717 if (t == head) break;
3718 t = idom(t);
3719 }
3720 while (wl.size() > 0) {
3721 Node* tt = wl.pop();
3722 tt->dump();
3723 if (tt == last_peel) tty->print_cr("-- cut --");
3724 }
3725 }
3726 #endif
3727 VectorSet peel;
3728 VectorSet not_peel;
3729 Node_List peel_list;
3730 Node_List worklist;
3731 Node_List sink_list;
3732
3733 uint estimate = loop->est_loop_clone_sz(1);
3734 if (exceeding_node_budget(estimate)) {
3735 return false;
3736 }
3737
3738 // Set of cfg nodes to peel are those that are executable from
3739 // the head through last_peel.
3740 assert(worklist.size() == 0, "should be empty");
3741 worklist.push(head);
3742 peel.set(head->_idx);
3743 while (worklist.size() > 0) {
3744 Node *n = worklist.pop();
3745 if (n != last_peel) {
3746 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) {
3747 Node* use = n->fast_out(j);
3748 if (use->is_CFG() &&
3749 loop->is_member(get_loop(use)) &&
3750 !peel.test_set(use->_idx)) {
3751 worklist.push(use);
3752 }
3753 }
3754 }
3755 }
3756
3757 // Set of non-cfg nodes to peel are those that are control
3758 // dependent on the cfg nodes.
3759 for (uint i = 0; i < loop->_body.size(); i++) {
3760 Node *n = loop->_body.at(i);
3761 Node *n_c = has_ctrl(n) ? get_ctrl(n) : n;
3762 if (peel.test(n_c->_idx)) {
3763 peel.set(n->_idx);
3764 } else {
3765 not_peel.set(n->_idx);
3766 }
3767 }
3768
3769 // Step 2: move operations from the peeled section down into the
3770 // not-peeled section
3771
3772 // Get a post order schedule of nodes in the peel region
3773 // Result in right-most operand.
3774 scheduled_nodelist(loop, peel, peel_list);
3775
3776 assert(is_valid_loop_partition(loop, peel, peel_list, not_peel), "bad partition");
3777
3778 // For future check for too many new phis
3779 uint old_phi_cnt = 0;
3780 for (DUIterator_Fast jmax, j = head->fast_outs(jmax); j < jmax; j++) {
3781 Node* use = head->fast_out(j);
3782 if (use->is_Phi()) old_phi_cnt++;
3783 }
3784
3785 #ifndef PRODUCT
3786 if (TracePartialPeeling) {
3787 tty->print_cr("\npeeled list");
3788 }
3789 #endif
3790
3791 // Evacuate nodes in peel region into the not_peeled region if possible
3792 bool too_many_clones = false;
3793 uint new_phi_cnt = 0;
3794 uint cloned_for_outside_use = 0;
3795 for (uint i = 0; i < peel_list.size();) {
3796 Node* n = peel_list.at(i);
3797 #ifndef PRODUCT
3798 if (TracePartialPeeling) n->dump();
3799 #endif
3800 bool incr = true;
3801 if (!n->is_CFG()) {
3802 if (has_use_in_set(n, not_peel)) {
3803 // If not used internal to the peeled region,
3804 // move "n" from peeled to not_peeled region.
3805 if (!has_use_internal_to_set(n, peel, loop)) {
3806 // if not pinned and not a load (which maybe anti-dependent on a store)
3807 // and not a CMove (Matcher expects only bool->cmove).
3808 if (n->in(0) == NULL && !n->is_Load() && !n->is_CMove()) {
3809 int new_clones = clone_for_use_outside_loop(loop, n, worklist);
3810 if (new_clones == -1) {
3811 too_many_clones = true;
3812 break;
3813 }
3814 cloned_for_outside_use += new_clones;
3815 sink_list.push(n);
3816 peel.remove(n->_idx);
3817 not_peel.set(n->_idx);
3818 peel_list.remove(i);
3819 incr = false;
3820 #ifndef PRODUCT
3821 if (TracePartialPeeling) {
3822 tty->print_cr("sink to not_peeled region: %d newbb: %d",
3823 n->_idx, get_ctrl(n)->_idx);
3824 }
3825 #endif
3826 }
3827 } else {
3828 // Otherwise check for special def-use cases that span
3829 // the peel/not_peel boundary such as bool->if
3830 clone_for_special_use_inside_loop(loop, n, not_peel, sink_list, worklist);
3831 new_phi_cnt++;
3832 }
3833 }
3834 }
3835 if (incr) i++;
3836 }
3837
3838 estimate += cloned_for_outside_use + new_phi_cnt;
3839 bool exceed_node_budget = !may_require_nodes(estimate);
3840 bool exceed_phi_limit = new_phi_cnt > old_phi_cnt + PartialPeelNewPhiDelta;
3841
3842 if (too_many_clones || exceed_node_budget || exceed_phi_limit) {
3843 #ifndef PRODUCT
3844 if (TracePartialPeeling && exceed_phi_limit) {
3845 tty->print_cr("\nToo many new phis: %d old %d new cmpi: %c",
3846 new_phi_cnt, old_phi_cnt, new_peel_if != NULL?'T':'F');
3847 }
3848 #endif
3849 if (new_peel_if != NULL) {
3850 remove_cmpi_loop_exit(new_peel_if, loop);
3851 }
3852 // Inhibit more partial peeling on this loop
3853 assert(!head->is_partial_peel_loop(), "not partial peeled");
3854 head->mark_partial_peel_failed();
3855 if (cloned_for_outside_use > 0) {
3856 // Terminate this round of loop opts because
3857 // the graph outside this loop was changed.
3858 C->set_major_progress();
3859 return true;
3860 }
3861 return false;
3862 }
3863
3864 // Step 3: clone loop, retarget control, and insert new phis
3865
3866 // Create new loop head for new phis and to hang
3867 // the nodes being moved (sinked) from the peel region.
3868 LoopNode* new_head = new LoopNode(last_peel, last_peel);
3869 new_head->set_unswitch_count(head->unswitch_count()); // Preserve
3870 _igvn.register_new_node_with_optimizer(new_head);
3871 assert(first_not_peeled->in(0) == last_peel, "last_peel <- first_not_peeled");
3872 _igvn.replace_input_of(first_not_peeled, 0, new_head);
3873 set_loop(new_head, loop);
3874 loop->_body.push(new_head);
3875 not_peel.set(new_head->_idx);
3876 set_idom(new_head, last_peel, dom_depth(first_not_peeled));
3877 set_idom(first_not_peeled, new_head, dom_depth(first_not_peeled));
3878
3879 while (sink_list.size() > 0) {
3880 Node* n = sink_list.pop();
3881 set_ctrl(n, new_head);
3882 }
3883
3884 assert(is_valid_loop_partition(loop, peel, peel_list, not_peel), "bad partition");
3885
3886 clone_loop(loop, old_new, dd, IgnoreStripMined);
3887
3888 const uint clone_exit_idx = 1;
3889 const uint orig_exit_idx = 2;
3890 assert(is_valid_clone_loop_form(loop, peel_list, orig_exit_idx, clone_exit_idx), "bad clone loop");
3891
3892 Node* head_clone = old_new[head->_idx];
3893 LoopNode* new_head_clone = old_new[new_head->_idx]->as_Loop();
3894 Node* orig_tail_clone = head_clone->in(2);
3895
3896 // Add phi if "def" node is in peel set and "use" is not
3897
3898 for (uint i = 0; i < peel_list.size(); i++) {
3899 Node *def = peel_list.at(i);
3900 if (!def->is_CFG()) {
3901 for (DUIterator_Fast jmax, j = def->fast_outs(jmax); j < jmax; j++) {
3902 Node *use = def->fast_out(j);
3903 if (has_node(use) && use->in(0) != C->top() &&
3904 (!peel.test(use->_idx) ||
3905 (use->is_Phi() && use->in(0) == head)) ) {
3906 worklist.push(use);
3907 }
3908 }
3909 while( worklist.size() ) {
3910 Node *use = worklist.pop();
3911 for (uint j = 1; j < use->req(); j++) {
3912 Node* n = use->in(j);
3913 if (n == def) {
3914
3915 // "def" is in peel set, "use" is not in peel set
3916 // or "use" is in the entry boundary (a phi) of the peel set
3917
3918 Node* use_c = has_ctrl(use) ? get_ctrl(use) : use;
3919
3920 if ( loop->is_member(get_loop( use_c )) ) {
3921 // use is in loop
3922 if (old_new[use->_idx] != NULL) { // null for dead code
3923 Node* use_clone = old_new[use->_idx];
3924 _igvn.replace_input_of(use, j, C->top());
3925 insert_phi_for_loop( use_clone, j, old_new[def->_idx], def, new_head_clone );
3926 }
3927 } else {
3928 assert(is_valid_clone_loop_exit_use(loop, use, orig_exit_idx), "clone loop format");
3929 // use is not in the loop, check if the live range includes the cut
3930 Node* lp_if = use_c->in(orig_exit_idx)->in(0);
3931 if (not_peel.test(lp_if->_idx)) {
3932 assert(j == orig_exit_idx, "use from original loop");
3933 insert_phi_for_loop( use, clone_exit_idx, old_new[def->_idx], def, new_head_clone );
3934 }
3935 }
3936 }
3937 }
3938 }
3939 }
3940 }
3941
3942 // Step 3b: retarget control
3943
3944 // Redirect control to the new loop head if a cloned node in
3945 // the not_peeled region has control that points into the peeled region.
3946 // This necessary because the cloned peeled region will be outside
3947 // the loop.
3948 // from to
3949 // cloned-peeled <---+
3950 // new_head_clone: | <--+
3951 // cloned-not_peeled in(0) in(0)
3952 // orig-peeled
3953
3954 for (uint i = 0; i < loop->_body.size(); i++) {
3955 Node *n = loop->_body.at(i);
3956 if (!n->is_CFG() && n->in(0) != NULL &&
3957 not_peel.test(n->_idx) && peel.test(n->in(0)->_idx)) {
3958 Node* n_clone = old_new[n->_idx];
3959 _igvn.replace_input_of(n_clone, 0, new_head_clone);
3960 }
3961 }
3962
3963 // Backedge of the surviving new_head (the clone) is original last_peel
3964 _igvn.replace_input_of(new_head_clone, LoopNode::LoopBackControl, last_peel);
3965
3966 // Cut first node in original not_peel set
3967 _igvn.rehash_node_delayed(new_head); // Multiple edge updates:
3968 new_head->set_req(LoopNode::EntryControl, C->top()); // use rehash_node_delayed / set_req instead of
3969 new_head->set_req(LoopNode::LoopBackControl, C->top()); // multiple replace_input_of calls
3970
3971 // Copy head_clone back-branch info to original head
3972 // and remove original head's loop entry and
3973 // clone head's back-branch
3974 _igvn.rehash_node_delayed(head); // Multiple edge updates
3975 head->set_req(LoopNode::EntryControl, head_clone->in(LoopNode::LoopBackControl));
3976 head->set_req(LoopNode::LoopBackControl, C->top());
3977 _igvn.replace_input_of(head_clone, LoopNode::LoopBackControl, C->top());
3978
3979 // Similarly modify the phis
3980 for (DUIterator_Fast kmax, k = head->fast_outs(kmax); k < kmax; k++) {
3981 Node* use = head->fast_out(k);
3982 if (use->is_Phi() && use->outcnt() > 0) {
3983 Node* use_clone = old_new[use->_idx];
3984 _igvn.rehash_node_delayed(use); // Multiple edge updates
3985 use->set_req(LoopNode::EntryControl, use_clone->in(LoopNode::LoopBackControl));
3986 use->set_req(LoopNode::LoopBackControl, C->top());
3987 _igvn.replace_input_of(use_clone, LoopNode::LoopBackControl, C->top());
3988 }
3989 }
3990
3991 // Step 4: update dominator tree and dominator depth
3992
3993 set_idom(head, orig_tail_clone, dd);
3994 recompute_dom_depth();
3995
3996 // Inhibit more partial peeling on this loop
3997 new_head_clone->set_partial_peel_loop();
3998 C->set_major_progress();
3999 loop->record_for_igvn();
4000
4001 #ifndef PRODUCT
4002 if (TracePartialPeeling) {
4003 tty->print_cr("\nafter partial peel one iteration");
4004 Node_List wl;
4005 Node* t = last_peel;
4006 while (true) {
4007 wl.push(t);
4008 if (t == head_clone) break;
4009 t = idom(t);
4010 }
4011 while (wl.size() > 0) {
4012 Node* tt = wl.pop();
4013 if (tt == head) tty->print_cr("orig head");
4014 else if (tt == new_head_clone) tty->print_cr("new head");
4015 else if (tt == head_clone) tty->print_cr("clone head");
4016 tt->dump();
4017 }
4018 }
4019 #endif
4020 return true;
4021 }
4022
4023 // Transform:
4024 //
4025 // loop<-----------------+
4026 // | |
4027 // stmt1 stmt2 .. stmtn |
4028 // | | | |
4029 // \ | / |
4030 // v v v |
4031 // region |
4032 // | |
4033 // shared_stmt |
4034 // | |
4035 // v |
4036 // if |
4037 // / \ |
4038 // | -----------+
4039 // v
4040 //
4041 // into:
4042 //
4043 // loop<-------------------+
4044 // | |
4045 // v |
4046 // +->loop |
4047 // | | |
4048 // | stmt1 stmt2 .. stmtn |
4049 // | | | | |
4050 // | | \ / |
4051 // | | v v |
4052 // | | region1 |
4053 // | | | |
4054 // | shared_stmt shared_stmt |
4055 // | | | |
4056 // | v v |
4057 // | if if |
4058 // | /\ / \ |
4059 // +-- | | -------+
4060 // \ /
4061 // v v
4062 // region2
4063 //
4064 // (region2 is shown to merge mirrored projections of the loop exit
4065 // ifs to make the diagram clearer but they really merge the same
4066 // projection)
4067 //
4068 // Conditions for this transformation to trigger:
4069 // - the path through stmt1 is frequent enough
4070 // - the inner loop will be turned into a counted loop after transformation
4071 bool PhaseIdealLoop::duplicate_loop_backedge(IdealLoopTree *loop, Node_List &old_new) {
4072 if (!DuplicateBackedge) {
4073 return false;
4074 }
4075 assert(!loop->_head->is_CountedLoop() || StressDuplicateBackedge, "Non-counted loop only");
4076 if (!loop->_head->is_Loop()) {
4077 return false;
4078 }
4079
4080 uint estimate = loop->est_loop_clone_sz(1);
4081 if (exceeding_node_budget(estimate)) {
4082 return false;
4083 }
4084
4085 LoopNode *head = loop->_head->as_Loop();
4086
4087 Node* region = NULL;
4088 IfNode* exit_test = NULL;
4089 uint inner;
4090 float f;
4091 if (StressDuplicateBackedge) {
4092 if (head->is_strip_mined()) {
4093 return false;
4094 }
4095 Node* c = head->in(LoopNode::LoopBackControl);
4096
4097 while (c != head) {
4098 if (c->is_Region()) {
4099 region = c;
4100 }
4101 c = idom(c);
4102 }
4103
4104 if (region == NULL) {
4105 return false;
4106 }
4107
4108 inner = 1;
4109 } else {
4110 // Is the shape of the loop that of a counted loop...
4111 Node* back_control = loop_exit_control(head, loop);
4112 if (back_control == NULL) {
4113 return false;
4114 }
4115
4116 BoolTest::mask bt = BoolTest::illegal;
4117 float cl_prob = 0;
4118 Node* incr = NULL;
4119 Node* limit = NULL;
4120 Node* cmp = loop_exit_test(back_control, loop, incr, limit, bt, cl_prob);
4121 if (cmp == NULL || cmp->Opcode() != Op_CmpI) {
4122 return false;
4123 }
4124
4125 // With an extra phi for the candidate iv?
4126 // Or the region node is the loop head
4127 if (!incr->is_Phi() || incr->in(0) == head) {
4128 return false;
4129 }
4130
4131 PathFrequency pf(head, this);
4132 region = incr->in(0);
4133
4134 // Go over all paths for the extra phi's region and see if that
4135 // path is frequent enough and would match the expected iv shape
4136 // if the extra phi is removed
4137 inner = 0;
4138 for (uint i = 1; i < incr->req(); ++i) {
4139 Node* in = incr->in(i);
4140 Node* trunc1 = NULL;
4141 Node* trunc2 = NULL;
4142 const TypeInteger* iv_trunc_t = NULL;
4143 Node* orig_in = in;
4144 if (!(in = CountedLoopNode::match_incr_with_optional_truncation(in, &trunc1, &trunc2, &iv_trunc_t, T_INT))) {
4145 continue;
4146 }
4147 assert(in->Opcode() == Op_AddI, "wrong increment code");
4148 Node* xphi = NULL;
4149 Node* stride = loop_iv_stride(in, loop, xphi);
4150
4151 if (stride == NULL) {
4152 continue;
4153 }
4154
4155 PhiNode* phi = loop_iv_phi(xphi, NULL, head, loop);
4156 if (phi == NULL ||
4157 (trunc1 == NULL && phi->in(LoopNode::LoopBackControl) != incr) ||
4158 (trunc1 != NULL && phi->in(LoopNode::LoopBackControl) != trunc1)) {
4159 return false;
4160 }
4161
4162 f = pf.to(region->in(i));
4163 if (f > 0.5) {
4164 inner = i;
4165 break;
4166 }
4167 }
4168
4169 if (inner == 0) {
4170 return false;
4171 }
4172
4173 exit_test = back_control->in(0)->as_If();
4174 }
4175
4176 if (idom(region)->is_Catch()) {
4177 return false;
4178 }
4179
4180 // Collect all control nodes that need to be cloned (shared_stmt in the diagram)
4181 Unique_Node_List wq;
4182 wq.push(head->in(LoopNode::LoopBackControl));
4183 for (uint i = 0; i < wq.size(); i++) {
4184 Node* c = wq.at(i);
4185 assert(get_loop(c) == loop, "not in the right loop?");
4186 if (c->is_Region()) {
4187 if (c != region) {
4188 for (uint j = 1; j < c->req(); ++j) {
4189 wq.push(c->in(j));
4190 }
4191 }
4192 } else {
4193 wq.push(c->in(0));
4194 }
4195 assert(!is_dominator(c, region) || c == region, "shouldn't go above region");
4196 }
4197
4198 Node* region_dom = idom(region);
4199
4200 // Can't do the transformation if this would cause a membar pair to
4201 // be split
4202 for (uint i = 0; i < wq.size(); i++) {
4203 Node* c = wq.at(i);
4204 if (c->is_MemBar() && (c->as_MemBar()->trailing_store() || c->as_MemBar()->trailing_load_store())) {
4205 assert(c->as_MemBar()->leading_membar()->trailing_membar() == c, "bad membar pair");
4206 if (!wq.member(c->as_MemBar()->leading_membar())) {
4207 return false;
4208 }
4209 }
4210 }
4211
4212 // Collect data nodes that need to be clones as well
4213 int dd = dom_depth(head);
4214
4215 for (uint i = 0; i < loop->_body.size(); ++i) {
4216 Node* n = loop->_body.at(i);
4217 if (has_ctrl(n)) {
4218 Node* c = get_ctrl(n);
4219 if (wq.member(c)) {
4220 wq.push(n);
4221 }
4222 } else {
4223 set_idom(n, idom(n), dd);
4224 }
4225 }
4226
4227 // clone shared_stmt
4228 clone_loop_body(wq, old_new, NULL);
4229
4230 Node* region_clone = old_new[region->_idx];
4231 region_clone->set_req(inner, C->top());
4232 set_idom(region, region->in(inner), dd);
4233
4234 // Prepare the outer loop
4235 Node* outer_head = new LoopNode(head->in(LoopNode::EntryControl), old_new[head->in(LoopNode::LoopBackControl)->_idx]);
4236 register_control(outer_head, loop->_parent, outer_head->in(LoopNode::EntryControl));
4237 _igvn.replace_input_of(head, LoopNode::EntryControl, outer_head);
4238 set_idom(head, outer_head, dd);
4239
4240 fix_body_edges(wq, loop, old_new, dd, loop->_parent, true);
4241
4242 // Make one of the shared_stmt copies only reachable from stmt1, the
4243 // other only from stmt2..stmtn.
4244 Node* dom = NULL;
4245 for (uint i = 1; i < region->req(); ++i) {
4246 if (i != inner) {
4247 _igvn.replace_input_of(region, i, C->top());
4248 }
4249 Node* in = region_clone->in(i);
4250 if (in->is_top()) {
4251 continue;
4252 }
4253 if (dom == NULL) {
4254 dom = in;
4255 } else {
4256 dom = dom_lca(dom, in);
4257 }
4258 }
4259
4260 set_idom(region_clone, dom, dd);
4261
4262 // Set up the outer loop
4263 for (uint i = 0; i < head->outcnt(); i++) {
4264 Node* u = head->raw_out(i);
4265 if (u->is_Phi()) {
4266 Node* outer_phi = u->clone();
4267 outer_phi->set_req(0, outer_head);
4268 Node* backedge = old_new[u->in(LoopNode::LoopBackControl)->_idx];
4269 if (backedge == NULL) {
4270 backedge = u->in(LoopNode::LoopBackControl);
4271 }
4272 outer_phi->set_req(LoopNode::LoopBackControl, backedge);
4273 register_new_node(outer_phi, outer_head);
4274 _igvn.replace_input_of(u, LoopNode::EntryControl, outer_phi);
4275 }
4276 }
4277
4278 // create control and data nodes for out of loop uses (including region2)
4279 Node_List worklist;
4280 uint new_counter = C->unique();
4281 fix_ctrl_uses(wq, loop, old_new, ControlAroundStripMined, outer_head, NULL, worklist);
4282
4283 Node_List *split_if_set = NULL;
4284 Node_List *split_bool_set = NULL;
4285 Node_List *split_cex_set = NULL;
4286 fix_data_uses(wq, loop, ControlAroundStripMined, head->is_strip_mined() ? loop->_parent : loop, new_counter, old_new, worklist, split_if_set, split_bool_set, split_cex_set);
4287
4288 finish_clone_loop(split_if_set, split_bool_set, split_cex_set);
4289
4290 if (exit_test != NULL) {
4291 float cnt = exit_test->_fcnt;
4292 if (cnt != COUNT_UNKNOWN) {
4293 exit_test->_fcnt = cnt * f;
4294 old_new[exit_test->_idx]->as_If()->_fcnt = cnt * (1 - f);
4295 }
4296 }
4297
4298 C->set_major_progress();
4299
4300 return true;
4301 }